init src 0.9.2
This commit is contained in:
0
vllm/model_executor/layers/__init__.py
Normal file
0
vllm/model_executor/layers/__init__.py
Normal file
430
vllm/model_executor/layers/activation.py
Normal file
430
vllm/model_executor/layers/activation.py
Normal file
@@ -0,0 +1,430 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
"""Custom activation functions."""
|
||||
import math
|
||||
from typing import Optional
|
||||
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
from vllm.distributed import (divide, get_tensor_model_parallel_rank,
|
||||
get_tensor_model_parallel_world_size)
|
||||
from vllm.model_executor.custom_op import CustomOp
|
||||
from vllm.model_executor.utils import set_weight_attrs
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils import LazyDict
|
||||
import vllm.envs as envs
|
||||
|
||||
|
||||
@CustomOp.register("fatrelu_and_mul")
|
||||
class FatreluAndMul(CustomOp):
|
||||
"""An activation function for FATReLU.
|
||||
|
||||
The function computes x -> FATReLU(x[:d]) * x[d:] where
|
||||
d = x.shape[-1] // 2.
|
||||
This is used in openbmb/MiniCPM-S-1B-sft.
|
||||
|
||||
Shapes:
|
||||
x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
|
||||
return: (num_tokens, d) or (batch_size, seq_len, d)
|
||||
"""
|
||||
|
||||
def __init__(self, threshold: float = 0.):
|
||||
super().__init__()
|
||||
self.threshold = threshold
|
||||
if current_platform.is_cuda_alike():
|
||||
self.op = torch.ops._C.fatrelu_and_mul
|
||||
elif current_platform.is_cpu():
|
||||
self._forward_method = self.forward_native
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
x1 = x[..., :d]
|
||||
x2 = x[..., d:]
|
||||
x1 = F.threshold(x1, self.threshold, 0.0)
|
||||
return x1 * x2
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
self.op(out, x, self.threshold)
|
||||
return out
|
||||
|
||||
|
||||
@CustomOp.register("silu_and_mul")
|
||||
class SiluAndMul(CustomOp):
|
||||
"""An activation function for SwiGLU.
|
||||
|
||||
The function computes x -> silu(x[:d]) * x[d:] where d = x.shape[-1] // 2.
|
||||
|
||||
Shapes:
|
||||
x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
|
||||
return: (num_tokens, d) or (batch_size, seq_len, d)
|
||||
"""
|
||||
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
if current_platform.is_cuda_alike() or current_platform.is_cpu():
|
||||
self.op = torch.ops._C.silu_and_mul
|
||||
self.op_opt = torch.ops._C.silu_and_mul_opt
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
self.op = ipex_ops.silu_and_mul
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
d = x.shape[-1] // 2
|
||||
return F.silu(x[..., :d]) * x[..., d:]
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
if envs.VLLM_USE_OPT_OP:
|
||||
self.op_opt(out, x)
|
||||
else:
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_neuron(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
x_reshaped = x.view(-1, x.shape[-1])
|
||||
s = x_reshaped[:, :d] * F.sigmoid(x_reshaped[:, :d])
|
||||
result = s * x_reshaped[:, d:]
|
||||
return result.view(*x.shape[:-1], d)
|
||||
|
||||
|
||||
@CustomOp.register("mul_and_silu")
|
||||
class MulAndSilu(CustomOp):
|
||||
"""An activation function for SwiGLU.
|
||||
|
||||
The function computes x -> x[:d] * silu(x[d:]) where d = x.shape[-1] // 2.
|
||||
|
||||
Shapes:
|
||||
x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
|
||||
return: (num_tokens, d) or (batch_size, seq_len, d)
|
||||
"""
|
||||
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
if current_platform.is_cuda_alike():
|
||||
self.op = torch.ops._C.mul_and_silu
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
self.op = ipex_ops.silu_and_mul
|
||||
elif current_platform.is_cpu():
|
||||
self._forward_method = self.forward_native
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
d = x.shape[-1] // 2
|
||||
return x[..., :d] * F.silu(x[..., d:])
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
# TODO implement forward_xpu for MulAndSilu
|
||||
# def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
|
||||
|
||||
@CustomOp.register("gelu_and_mul_sparse")
|
||||
class GeluAndMulSparse(CustomOp):
|
||||
"""An activation function for GeluAndMulSparse.
|
||||
This activation function is used in Gemma3n. It computes:
|
||||
up_proj = self.up_proj(x)
|
||||
gate_proj = self.gate_proj(x)
|
||||
gate_proj = self._gaussian_topk(gate_proj) # sparsity
|
||||
activations = self.act_fn(gate_proj) # gelu
|
||||
down_proj = self.down_proj(activations * up_proj)
|
||||
Shapes:
|
||||
x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
|
||||
return: (num_tokens, d) or (batch_size, seq_len, d)
|
||||
"""
|
||||
|
||||
def __init__(self, activation_sparsity: float, approximate: str = "none"):
|
||||
super().__init__()
|
||||
# Gelu.
|
||||
self.approximate = approximate
|
||||
if approximate not in ("none", "tanh"):
|
||||
raise ValueError(f"Unknown approximate mode: {approximate}")
|
||||
|
||||
# Sparsity.
|
||||
if activation_sparsity == 0.0:
|
||||
raise ValueError(
|
||||
"activation_sparsity is 0.0. Please use GeluAndMul.")
|
||||
target_sparsity_tensor = torch.tensor(activation_sparsity,
|
||||
dtype=torch.float32)
|
||||
normal_dist = torch.distributions.normal.Normal(0, 1)
|
||||
self.std_multiplier = normal_dist.icdf(target_sparsity_tensor)
|
||||
|
||||
def _gaussian_topk(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""Get % sparse percentile of the Gaussian distribution."""
|
||||
# NOTE(rob): for TP>1, we could all-gather to get the means/std.
|
||||
# But we do not do this because in expectation they are the same
|
||||
# and in practice the eval scores are good without gathering.
|
||||
mean = torch.mean(x, dim=-1, keepdim=True)
|
||||
std = torch.std(x, dim=-1, keepdim=True, unbiased=False)
|
||||
cutoff_x = mean + std * self.std_multiplier
|
||||
return nn.functional.relu(x - cutoff_x)
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
d = x.shape[-1] // 2
|
||||
out = self._gaussian_topk(x[..., :d])
|
||||
out = F.gelu(out, approximate=self.approximate)
|
||||
return out * x[..., d:]
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return self.forward_native(x)
|
||||
|
||||
|
||||
@CustomOp.register("gelu_and_mul")
|
||||
class GeluAndMul(CustomOp):
|
||||
"""An activation function for GeGLU.
|
||||
|
||||
The function computes x -> GELU(x[:d]) * x[d:] where d = x.shape[-1] // 2.
|
||||
|
||||
Shapes:
|
||||
x: (batch_size, seq_len, 2 * d) or (num_tokens, 2 * d)
|
||||
return: (batch_size, seq_len, d) or (num_tokens, d)
|
||||
"""
|
||||
|
||||
def __init__(self, approximate: str = "none"):
|
||||
super().__init__()
|
||||
self.approximate = approximate
|
||||
if approximate not in ("none", "tanh"):
|
||||
raise ValueError(f"Unknown approximate mode: {approximate}")
|
||||
if current_platform.is_cuda_alike() or current_platform.is_cpu():
|
||||
if approximate == "none":
|
||||
self.op = torch.ops._C.gelu_and_mul
|
||||
self.op_opt = torch.ops._C.gelu_and_mul_opt
|
||||
elif approximate == "tanh":
|
||||
self.op = torch.ops._C.gelu_tanh_and_mul
|
||||
self.op_opt = torch.ops._C.gelu_tanh_and_mul_opt
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
if approximate == "none":
|
||||
self.op = ipex_ops.gelu_and_mul
|
||||
else:
|
||||
self.op = ipex_ops.gelu_tanh_and_mul
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
d = x.shape[-1] // 2
|
||||
return F.gelu(x[..., :d], approximate=self.approximate) * x[..., d:]
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
if envs.VLLM_USE_OPT_OP:
|
||||
self.op_opt(out, x)
|
||||
else:
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
d = x.shape[-1] // 2
|
||||
output_shape = (x.shape[:-1] + (d, ))
|
||||
out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def extra_repr(self) -> str:
|
||||
return f'approximate={repr(self.approximate)}'
|
||||
|
||||
|
||||
@CustomOp.register("gelu_new")
|
||||
class NewGELU(CustomOp):
|
||||
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
if current_platform.is_cuda_alike() or current_platform.is_cpu():
|
||||
self.op = torch.ops._C.gelu_new
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
self.op = ipex_ops.gelu_new
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
c = math.sqrt(2.0 / math.pi)
|
||||
return 0.5 * x * (1.0 + torch.tanh(c *
|
||||
(x + 0.044715 * torch.pow(x, 3.0))))
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
out = torch.empty_like(x)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return self.op(x)
|
||||
|
||||
|
||||
@CustomOp.register("gelu_fast")
|
||||
class FastGELU(CustomOp):
|
||||
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
if current_platform.is_cuda_alike() or current_platform.is_cpu():
|
||||
self.op = torch.ops._C.gelu_fast
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
self.op = ipex_ops.gelu_fast
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
return 0.5 * x * (1.0 + torch.tanh(x * 0.7978845608 *
|
||||
(1.0 + 0.044715 * x * x)))
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
out = torch.empty_like(x)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return self.op(x)
|
||||
|
||||
|
||||
@CustomOp.register("quick_gelu")
|
||||
class QuickGELU(CustomOp):
|
||||
# https://github.com/huggingface/transformers/blob/main/src/transformers/activations.py#L90
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
if current_platform.is_cuda_alike() or current_platform.is_cpu():
|
||||
self.op = torch.ops._C.gelu_quick
|
||||
elif current_platform.is_xpu():
|
||||
from vllm._ipex_ops import ipex_ops
|
||||
self.op = ipex_ops.gelu_quick
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
return x * torch.sigmoid(1.702 * x)
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
out = torch.empty_like(x)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
out = torch.empty_like(x)
|
||||
self.op(out, x)
|
||||
return out
|
||||
|
||||
# TODO implement forward_xpu for QuickGELU
|
||||
# def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
|
||||
|
||||
|
||||
@CustomOp.register("relu2")
|
||||
class ReLUSquaredActivation(CustomOp):
|
||||
"""
|
||||
Applies the relu^2 activation introduced in https://arxiv.org/abs/2109.08668v2
|
||||
"""
|
||||
|
||||
def forward_native(self, x: torch.Tensor) -> torch.Tensor:
|
||||
"""PyTorch-native implementation equivalent to forward()."""
|
||||
return torch.square(F.relu(x))
|
||||
|
||||
def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return self.forward_native(x)
|
||||
|
||||
|
||||
class ScaledActivation(nn.Module):
|
||||
"""An activation function with post-scale parameters.
|
||||
|
||||
This is used for some quantization methods like AWQ.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
act_module: nn.Module,
|
||||
intermediate_size: int,
|
||||
input_is_parallel: bool = True,
|
||||
params_dtype: Optional[torch.dtype] = None,
|
||||
):
|
||||
super().__init__()
|
||||
self.act = act_module
|
||||
self.input_is_parallel = input_is_parallel
|
||||
if input_is_parallel:
|
||||
tp_size = get_tensor_model_parallel_world_size()
|
||||
intermediate_size_per_partition = divide(intermediate_size,
|
||||
tp_size)
|
||||
else:
|
||||
intermediate_size_per_partition = intermediate_size
|
||||
if params_dtype is None:
|
||||
params_dtype = torch.get_default_dtype()
|
||||
self.scales = nn.Parameter(
|
||||
torch.empty(intermediate_size_per_partition, dtype=params_dtype))
|
||||
set_weight_attrs(self.scales, {"weight_loader": self.weight_loader})
|
||||
|
||||
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return self.act(x) / self.scales
|
||||
|
||||
def weight_loader(self, param: nn.Parameter, loaded_weight: torch.Tensor):
|
||||
param_data = param.data
|
||||
if self.input_is_parallel:
|
||||
tp_rank = get_tensor_model_parallel_rank()
|
||||
shard_size = param_data.shape[0]
|
||||
start_idx = tp_rank * shard_size
|
||||
loaded_weight = loaded_weight.narrow(0, start_idx, shard_size)
|
||||
assert param_data.shape == loaded_weight.shape
|
||||
param_data.copy_(loaded_weight)
|
||||
|
||||
|
||||
_ACTIVATION_REGISTRY = LazyDict({
|
||||
"gelu":
|
||||
lambda: nn.GELU(),
|
||||
"gelu_fast":
|
||||
lambda: FastGELU(),
|
||||
"gelu_new":
|
||||
lambda: NewGELU(),
|
||||
"gelu_pytorch_tanh":
|
||||
lambda: nn.GELU(approximate="tanh"),
|
||||
"relu":
|
||||
lambda: nn.ReLU(),
|
||||
"relu2":
|
||||
lambda: ReLUSquaredActivation(),
|
||||
"silu":
|
||||
lambda: nn.SiLU(),
|
||||
"quick_gelu":
|
||||
lambda: QuickGELU(),
|
||||
})
|
||||
|
||||
|
||||
def get_act_fn(act_fn_name: str) -> nn.Module:
|
||||
"""Get an activation function by name."""
|
||||
act_fn_name = act_fn_name.lower()
|
||||
if act_fn_name not in _ACTIVATION_REGISTRY:
|
||||
raise ValueError(
|
||||
f"Activation function {act_fn_name!r} is not supported.")
|
||||
|
||||
return _ACTIVATION_REGISTRY[act_fn_name]
|
||||
|
||||
|
||||
_ACTIVATION_AND_MUL_REGISTRY = LazyDict({
|
||||
"gelu": lambda: GeluAndMul(),
|
||||
"silu": lambda: SiluAndMul(),
|
||||
"geglu": lambda: GeluAndMul(),
|
||||
})
|
||||
|
||||
|
||||
def get_act_and_mul_fn(act_fn_name: str) -> nn.Module:
|
||||
"""Get an activation-and-mul (i.e. SiluAndMul) function by name."""
|
||||
act_fn_name = act_fn_name.lower()
|
||||
if act_fn_name not in _ACTIVATION_AND_MUL_REGISTRY:
|
||||
raise ValueError(
|
||||
f"Activation function {act_fn_name!r} is not supported.")
|
||||
|
||||
return _ACTIVATION_AND_MUL_REGISTRY[act_fn_name]
|
||||
78
vllm/model_executor/layers/fused_moe/__init__.py
Normal file
78
vllm/model_executor/layers/fused_moe/__init__.py
Normal file
@@ -0,0 +1,78 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
|
||||
from contextlib import contextmanager
|
||||
from typing import Any, Optional
|
||||
|
||||
from vllm.model_executor.layers.fused_moe.config import FusedMoEConfig
|
||||
from vllm.model_executor.layers.fused_moe.layer import (
|
||||
FusedMoE, FusedMoEMethodBase, FusedMoeWeightScaleSupported)
|
||||
from vllm.model_executor.layers.fused_moe.modular_kernel import (
|
||||
FusedMoEActivationFormat, FusedMoEPermuteExpertsUnpermute,
|
||||
FusedMoEPrepareAndFinalize)
|
||||
from vllm.triton_utils import HAS_TRITON
|
||||
|
||||
_config: Optional[dict[str, Any]] = None
|
||||
|
||||
|
||||
@contextmanager
|
||||
def override_config(config):
|
||||
global _config
|
||||
old_config = _config
|
||||
_config = config
|
||||
yield
|
||||
_config = old_config
|
||||
|
||||
|
||||
def get_config() -> Optional[dict[str, Any]]:
|
||||
return _config
|
||||
|
||||
|
||||
__all__ = [
|
||||
"FusedMoE",
|
||||
"FusedMoEConfig",
|
||||
"FusedMoEMethodBase",
|
||||
"FusedMoeWeightScaleSupported",
|
||||
"FusedMoEPermuteExpertsUnpermute",
|
||||
"FusedMoEActivationFormat",
|
||||
"FusedMoEPrepareAndFinalize",
|
||||
"override_config",
|
||||
"get_config",
|
||||
]
|
||||
|
||||
if HAS_TRITON:
|
||||
# import to register the custom ops
|
||||
import vllm.model_executor.layers.fused_moe.fused_marlin_moe # noqa
|
||||
import vllm.model_executor.layers.fused_moe.fused_moe # noqa
|
||||
from vllm.model_executor.layers.fused_moe.batched_deep_gemm_moe import (
|
||||
BatchedDeepGemmExperts)
|
||||
from vllm.model_executor.layers.fused_moe.batched_triton_or_deep_gemm_moe import ( # noqa: E501
|
||||
BatchedTritonOrDeepGemmExperts)
|
||||
from vllm.model_executor.layers.fused_moe.cutlass_moe import (
|
||||
CutlassExpertsFp8, cutlass_moe_fp4, cutlass_moe_fp8)
|
||||
from vllm.model_executor.layers.fused_moe.deep_gemm_moe import (
|
||||
DeepGemmExperts)
|
||||
from vllm.model_executor.layers.fused_moe.fused_batched_moe import (
|
||||
BatchedTritonExperts)
|
||||
from vllm.model_executor.layers.fused_moe.fused_moe import (
|
||||
TritonExperts, fused_experts, fused_moe, fused_topk,
|
||||
get_config_file_name, grouped_topk)
|
||||
from vllm.model_executor.layers.fused_moe.triton_deep_gemm_moe import (
|
||||
TritonOrDeepGemmExperts)
|
||||
|
||||
__all__ += [
|
||||
"fused_moe",
|
||||
"fused_topk",
|
||||
"fused_experts",
|
||||
"get_config_file_name",
|
||||
"grouped_topk",
|
||||
"cutlass_moe_fp8",
|
||||
"cutlass_moe_fp4",
|
||||
"CutlassExpertsFp8",
|
||||
"TritonExperts",
|
||||
"BatchedTritonExperts",
|
||||
"DeepGemmExperts",
|
||||
"BatchedDeepGemmExperts",
|
||||
"TritonOrDeepGemmExperts",
|
||||
"BatchedTritonOrDeepGemmExperts",
|
||||
]
|
||||
298
vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py
Normal file
298
vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py
Normal file
@@ -0,0 +1,298 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
from typing import Optional
|
||||
|
||||
import torch
|
||||
|
||||
import vllm.model_executor.layers.fused_moe.modular_kernel as mk
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
|
||||
from vllm.model_executor.layers.fused_moe.utils import _resize_cache
|
||||
from vllm.triton_utils import tl, triton
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
|
||||
@triton.jit
|
||||
def _silu_mul_fp8_quant_deep_gemm(
|
||||
# Pointers ------------------------------------------------------------
|
||||
input_ptr, # 16-bit activations (E, T, 2*H)
|
||||
y_q_ptr, # fp8 quantized activations (E, T, H)
|
||||
y_s_ptr, # 16-bit scales (E, T, G)
|
||||
counts_ptr, # int32 num tokens per expert (E)
|
||||
|
||||
# Sizes ---------------------------------------------------------------
|
||||
H: tl.constexpr, # hidden dimension (per output)
|
||||
GROUP_SIZE: tl.constexpr, # elements per group (usually 128)
|
||||
|
||||
# Strides for input (elements) ---------------------------------------
|
||||
stride_i_e,
|
||||
stride_i_t,
|
||||
stride_i_h,
|
||||
|
||||
# Strides for y_q (elements) -----------------------------------------
|
||||
stride_yq_e,
|
||||
stride_yq_t,
|
||||
stride_yq_h,
|
||||
|
||||
# Strides for y_s (elements) -----------------------------------------
|
||||
stride_ys_e,
|
||||
stride_ys_t,
|
||||
stride_ys_g,
|
||||
|
||||
# Stride for counts (elements)
|
||||
stride_counts_e,
|
||||
|
||||
# Numeric params ------------------------------------------------------
|
||||
eps: tl.constexpr,
|
||||
fp8_min: tl.constexpr,
|
||||
fp8_max: tl.constexpr,
|
||||
|
||||
# Meta ---------------------------------------------------------------
|
||||
BLOCK: tl.constexpr,
|
||||
):
|
||||
G = H // GROUP_SIZE
|
||||
|
||||
# map program id -> (e, g)
|
||||
pid = tl.program_id(0)
|
||||
e = pid // G
|
||||
g = pid % G
|
||||
|
||||
e = e.to(tl.int64)
|
||||
g = g.to(tl.int64)
|
||||
|
||||
# number of valid tokens for this expert
|
||||
n_tokens = tl.load(counts_ptr + e * stride_counts_e).to(tl.int64)
|
||||
|
||||
cols = tl.arange(0, BLOCK)
|
||||
cols = cols.to(tl.int64)
|
||||
mask_h = cols < BLOCK
|
||||
|
||||
t = tl.zeros([], tl.int64)
|
||||
while t < n_tokens:
|
||||
base_i_offset = (e * stride_i_e + t * stride_i_t +
|
||||
g * GROUP_SIZE * stride_i_h)
|
||||
base_yq_offset = (e * stride_yq_e + t * stride_yq_t +
|
||||
g * GROUP_SIZE * stride_yq_h)
|
||||
base_ys_offset = e * stride_ys_e + t * stride_ys_t + g * stride_ys_g
|
||||
|
||||
mask = mask_h
|
||||
x = tl.load(input_ptr + base_i_offset + cols * stride_i_h,
|
||||
mask=mask,
|
||||
other=0.0).to(tl.float32)
|
||||
y2 = tl.load(input_ptr + base_i_offset + H * stride_i_h +
|
||||
cols * stride_i_h,
|
||||
mask=mask,
|
||||
other=0.0).to(tl.float32)
|
||||
|
||||
x = x * (1.0 / (1.0 + tl.exp(-x)))
|
||||
y = x * y2
|
||||
|
||||
_absmax = tl.maximum(tl.max(tl.abs(y)), eps)
|
||||
y_s = _absmax / fp8_max
|
||||
y_q = tl.clamp(y / y_s, fp8_min, fp8_max).to(y_q_ptr.dtype.element_ty)
|
||||
|
||||
tl.store(y_q_ptr + base_yq_offset + cols * stride_yq_h, y_q, mask=mask)
|
||||
tl.store(y_s_ptr + base_ys_offset, y_s)
|
||||
|
||||
t += 1
|
||||
|
||||
|
||||
def silu_mul_fp8_quant_deep_gemm(
|
||||
y: torch.Tensor, # (E, T, 2*H) float32
|
||||
tokens_per_expert: torch.Tensor, # (E,) number of valid tokens per expert
|
||||
group_size: int = 128,
|
||||
eps: float = 1e-10,
|
||||
):
|
||||
"""Quantize silu(y[..., :H]) * y[..., H:] to FP8 with group per-token scales
|
||||
|
||||
y has shape (E, T, 2*H). The first half of the last dimension is
|
||||
silu-activated, multiplied by the second half, then quantized into FP8.
|
||||
|
||||
Returns `(y_q, y_s)` where
|
||||
* `y_q` is the FP8 tensor of shape `(E, T, H)`, same layout as `y[..., :H]`.
|
||||
* `y_s` has shape `(E, T, H // group_size)` and strides `(T*G, 1, T)`
|
||||
"""
|
||||
assert y.ndim == 3, "y must be (E, T, 2*H)"
|
||||
E, T, H2 = y.shape
|
||||
assert H2 % 2 == 0, "last dim of y must be even (2*H)"
|
||||
H = H2 // 2
|
||||
G = H // group_size
|
||||
assert H % group_size == 0, "H must be divisible by group_size"
|
||||
assert tokens_per_expert.ndim == 1 and tokens_per_expert.shape[0] == E, \
|
||||
"tokens_per_expert must be shape (E,)"
|
||||
tokens_per_expert = tokens_per_expert.to(device=y.device,
|
||||
dtype=torch.int32)
|
||||
|
||||
# allocate outputs
|
||||
fp8_dtype = torch.float8_e4m3fn
|
||||
y_q = torch.empty((E, T, H), dtype=fp8_dtype, device=y.device)
|
||||
|
||||
# strides (elements)
|
||||
stride_i_e, stride_i_t, stride_i_h = y.stride()
|
||||
stride_yq_e, stride_yq_t, stride_yq_h = y_q.stride()
|
||||
|
||||
# desired scale strides (elements): (T*G, 1, T)
|
||||
stride_ys_e = T * G
|
||||
stride_ys_t = 1
|
||||
stride_ys_g = T
|
||||
y_s = torch.empty_strided((E, T, G),
|
||||
(stride_ys_e, stride_ys_t, stride_ys_g),
|
||||
dtype=torch.float32,
|
||||
device=y.device)
|
||||
|
||||
stride_cnt_e = tokens_per_expert.stride()[0]
|
||||
|
||||
# static grid over experts and H-groups.
|
||||
# A loop inside the kernel handles the token dim
|
||||
grid = (E * G, )
|
||||
|
||||
f_info = torch.finfo(fp8_dtype)
|
||||
fp8_max = f_info.max
|
||||
fp8_min = f_info.min
|
||||
|
||||
_silu_mul_fp8_quant_deep_gemm[grid](
|
||||
y,
|
||||
y_q,
|
||||
y_s,
|
||||
tokens_per_expert,
|
||||
H,
|
||||
group_size,
|
||||
stride_i_e,
|
||||
stride_i_t,
|
||||
stride_i_h,
|
||||
stride_yq_e,
|
||||
stride_yq_t,
|
||||
stride_yq_h,
|
||||
stride_ys_e,
|
||||
stride_ys_t,
|
||||
stride_ys_g,
|
||||
stride_cnt_e,
|
||||
eps,
|
||||
fp8_min,
|
||||
fp8_max,
|
||||
BLOCK=group_size,
|
||||
num_warps=4,
|
||||
)
|
||||
|
||||
return y_q, y_s
|
||||
|
||||
|
||||
class BatchedDeepGemmExperts(mk.FusedMoEPermuteExpertsUnpermute):
|
||||
|
||||
# The Deep Gemm kernels only support block size of 128
|
||||
DEEPGEMM_BLOCK_SHAPE: list[int] = [128, 128]
|
||||
|
||||
def __init__(self,
|
||||
max_num_tokens: int,
|
||||
num_dispatchers: int,
|
||||
block_shape: list[int],
|
||||
per_act_token_quant=False):
|
||||
"""
|
||||
max_num_tokens: Maximum number of tokens from a DP Rank
|
||||
num_dispatchers: The number of DP dispatchers.
|
||||
block_shape: Block quantization block shape.
|
||||
per_act_token_quant: Per activation token quantization flag.
|
||||
"""
|
||||
super().__init__(
|
||||
FusedMoEQuantConfig(
|
||||
quant_dtype=torch.float8_e4m3fn,
|
||||
per_act_token_quant=per_act_token_quant,
|
||||
block_shape=block_shape,
|
||||
))
|
||||
assert self.block_shape == self.DEEPGEMM_BLOCK_SHAPE
|
||||
self.max_num_tokens = max_num_tokens
|
||||
self.num_dispatchers = num_dispatchers
|
||||
|
||||
@property
|
||||
def activation_formats(
|
||||
self
|
||||
) -> tuple[mk.FusedMoEActivationFormat, mk.FusedMoEActivationFormat]:
|
||||
return (mk.FusedMoEActivationFormat.BatchedExperts,
|
||||
mk.FusedMoEActivationFormat.BatchedExperts)
|
||||
|
||||
def supports_chunking(self) -> bool:
|
||||
return False
|
||||
|
||||
def supports_expert_map(self) -> bool:
|
||||
return False
|
||||
|
||||
def workspace_shapes(
|
||||
self,
|
||||
a: torch.Tensor,
|
||||
aq: torch.Tensor,
|
||||
M: int,
|
||||
N: int,
|
||||
K: int,
|
||||
topk: int,
|
||||
global_num_experts: int,
|
||||
local_num_experts: int,
|
||||
) -> tuple[tuple[int, ...], tuple[int, ...], tuple[int, ...], torch.dtype]:
|
||||
assert a.dim() == 2
|
||||
# FIXME (varun): We should be able to dispatch only from the leader
|
||||
# DP ranks in the case of TP > 1. At the moment, all the Ranks
|
||||
# end up sending their tokens. This needs to be fixed.
|
||||
num_dispatchers = self.num_dispatchers
|
||||
num_experts = local_num_experts
|
||||
max_num_tokens = a.size(
|
||||
0) if self.max_num_tokens is None else self.max_num_tokens
|
||||
workspace13 = (num_experts, max_num_tokens * num_dispatchers,
|
||||
max(K, N))
|
||||
workspace2 = (num_experts, max_num_tokens * num_dispatchers, (N // 2))
|
||||
output = (num_experts, max_num_tokens * num_dispatchers, K)
|
||||
return (workspace13, workspace2, output, a.dtype)
|
||||
|
||||
def apply(
|
||||
self,
|
||||
output: torch.Tensor,
|
||||
hidden_states: torch.Tensor,
|
||||
w1: torch.Tensor,
|
||||
w2: torch.Tensor,
|
||||
topk_ids: torch.Tensor,
|
||||
activation: str,
|
||||
global_num_experts: int,
|
||||
expert_map: Optional[torch.Tensor],
|
||||
w1_scale: Optional[torch.Tensor],
|
||||
w2_scale: Optional[torch.Tensor],
|
||||
w1_zp: Optional[torch.Tensor],
|
||||
w2_zp: Optional[torch.Tensor],
|
||||
a1q_scale: Optional[torch.Tensor],
|
||||
a2_scale: Optional[torch.Tensor],
|
||||
workspace13: torch.Tensor,
|
||||
workspace2: torch.Tensor,
|
||||
expert_num_tokens: Optional[torch.Tensor],
|
||||
):
|
||||
import deep_gemm as dg
|
||||
assert hidden_states.ndim == 3
|
||||
assert self.block_shape is not None
|
||||
|
||||
a1q = hidden_states
|
||||
_, N, K = w1.size()
|
||||
|
||||
assert w2.size(1) == K
|
||||
|
||||
E, max_num_tokens, N, K, top_k_num = mk._moe_problem_size(
|
||||
hidden_states, w1, w2, topk_ids)
|
||||
|
||||
workspace1 = _resize_cache(workspace13, (E, max_num_tokens, N))
|
||||
|
||||
# (from deepgemm docs) : A value hint (which is a value on CPU)
|
||||
# for the M expectation of each batch, correctly setting this value
|
||||
# may lead to better performance.
|
||||
expected_m = max_num_tokens
|
||||
|
||||
dg.m_grouped_gemm_fp8_fp8_bf16_nt_masked((a1q, a1q_scale),
|
||||
(w1, w1_scale),
|
||||
out=workspace1,
|
||||
masked_m=expert_num_tokens,
|
||||
expected_m=expected_m)
|
||||
|
||||
assert expert_num_tokens is not None
|
||||
a2q, a2q_scale = silu_mul_fp8_quant_deep_gemm(workspace1,
|
||||
expert_num_tokens)
|
||||
|
||||
dg.m_grouped_gemm_fp8_fp8_bf16_nt_masked((a2q, a2q_scale),
|
||||
(w2, w2_scale),
|
||||
out=output,
|
||||
masked_m=expert_num_tokens,
|
||||
expected_m=expected_m)
|
||||
@@ -0,0 +1,140 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
from typing import Optional
|
||||
|
||||
import torch
|
||||
|
||||
import vllm.model_executor.layers.fused_moe.modular_kernel as mk
|
||||
from vllm.model_executor.layers.fused_moe.batched_deep_gemm_moe import (
|
||||
BatchedDeepGemmExperts)
|
||||
from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
|
||||
from vllm.model_executor.layers.fused_moe.fused_batched_moe import (
|
||||
BatchedTritonExperts)
|
||||
|
||||
|
||||
class BatchedTritonOrDeepGemmExperts(mk.FusedMoEPermuteExpertsUnpermute):
|
||||
|
||||
def __init__(self,
|
||||
max_num_tokens: int,
|
||||
num_dispatchers: int,
|
||||
use_fp8_w8a8: bool = False,
|
||||
use_int8_w8a8: bool = False,
|
||||
use_int8_w8a16: bool = False,
|
||||
use_int4_w4a16: bool = False,
|
||||
block_shape: Optional[list[int]] = None,
|
||||
per_act_token_quant: bool = False,
|
||||
allow_deep_gemm: bool = False):
|
||||
assert not use_int8_w8a8, "NYI"
|
||||
assert not use_int8_w8a16, "NYI"
|
||||
assert not use_int4_w4a16, "NYI"
|
||||
|
||||
super().__init__(
|
||||
FusedMoEQuantConfig.make(
|
||||
use_fp8_w8a8=use_fp8_w8a8,
|
||||
use_int8_w8a8=use_int8_w8a8,
|
||||
use_int8_w8a16=use_int8_w8a16,
|
||||
use_int4_w4a16=use_int4_w4a16,
|
||||
block_shape=block_shape,
|
||||
per_act_token_quant=per_act_token_quant,
|
||||
))
|
||||
self.allow_deep_gemm = allow_deep_gemm
|
||||
|
||||
self.batched_triton_experts = BatchedTritonExperts(
|
||||
max_num_tokens=max_num_tokens,
|
||||
num_dispatchers=num_dispatchers,
|
||||
use_fp8_w8a8=use_fp8_w8a8,
|
||||
use_int8_w8a8=use_int8_w8a8,
|
||||
use_int8_w8a16=use_int8_w8a16,
|
||||
use_int4_w4a16=use_int4_w4a16,
|
||||
per_act_token_quant=self.per_act_token_quant,
|
||||
block_shape=self.block_shape,
|
||||
)
|
||||
|
||||
self.allow_deep_gemm = (allow_deep_gemm and use_fp8_w8a8
|
||||
and self.block_shape
|
||||
== BatchedDeepGemmExperts.DEEPGEMM_BLOCK_SHAPE)
|
||||
|
||||
self.batched_deep_gemm_experts = BatchedDeepGemmExperts(
|
||||
max_num_tokens=max_num_tokens,
|
||||
num_dispatchers=num_dispatchers,
|
||||
block_shape=self.block_shape, # type: ignore[arg-type]
|
||||
) if self.allow_deep_gemm else None
|
||||
|
||||
assert (self.batched_deep_gemm_experts is not None
|
||||
or self.batched_triton_experts is not None)
|
||||
|
||||
@property
|
||||
def activation_formats(
|
||||
self
|
||||
) -> tuple[mk.FusedMoEActivationFormat, mk.FusedMoEActivationFormat]:
|
||||
if self.batched_triton_experts is not None:
|
||||
assert (self.batched_deep_gemm_experts is None
|
||||
or self.batched_deep_gemm_experts.activation_formats
|
||||
== self.batched_triton_experts.activation_formats)
|
||||
return self.batched_triton_experts.activation_formats
|
||||
else:
|
||||
assert self.batched_deep_gemm_experts is not None
|
||||
return self.batched_deep_gemm_experts.activation_formats
|
||||
|
||||
def supports_chunking(self) -> bool:
|
||||
bdge = self.batched_deep_gemm_experts
|
||||
bte = self.batched_triton_experts
|
||||
return ((bdge is None or bdge.supports_chunking())
|
||||
and (bte is None or bte.supports_chunking()))
|
||||
|
||||
def supports_expert_map(self) -> bool:
|
||||
bdge = self.batched_deep_gemm_experts
|
||||
bte = self.batched_triton_experts
|
||||
return ((bdge is None or bdge.supports_expert_map())
|
||||
and (bte is None or bte.supports_expert_map()))
|
||||
|
||||
def workspace_shapes(
|
||||
self,
|
||||
a: torch.Tensor,
|
||||
aq: torch.Tensor,
|
||||
M: int,
|
||||
N: int,
|
||||
K: int,
|
||||
topk: int,
|
||||
global_num_experts: int,
|
||||
local_num_experts: int,
|
||||
) -> tuple[tuple[int, ...], tuple[int, ...], tuple[int, ...], torch.dtype]:
|
||||
# Note: the deep gemm workspaces are strictly larger than the triton
|
||||
# workspaces so we can be pessimistic here and allocate for DeepGemm
|
||||
# even if we fall back to triton later, e.g. if expert maps are set.
|
||||
if self.allow_deep_gemm:
|
||||
assert self.batched_deep_gemm_experts is not None
|
||||
return self.batched_deep_gemm_experts.workspace_shapes(
|
||||
a, aq, M, N, K, topk, global_num_experts, local_num_experts)
|
||||
else:
|
||||
assert self.batched_triton_experts is not None
|
||||
return self.batched_triton_experts.workspace_shapes(
|
||||
a, aq, M, N, K, topk, global_num_experts, local_num_experts)
|
||||
|
||||
def apply(
|
||||
self,
|
||||
output: torch.Tensor,
|
||||
hidden_states: torch.Tensor,
|
||||
w1: torch.Tensor,
|
||||
w2: torch.Tensor,
|
||||
topk_ids: torch.Tensor,
|
||||
activation: str,
|
||||
global_num_experts: int,
|
||||
expert_map: Optional[torch.Tensor],
|
||||
w1_scale: Optional[torch.Tensor],
|
||||
w2_scale: Optional[torch.Tensor],
|
||||
w1_zp: Optional[torch.Tensor],
|
||||
w2_zp: Optional[torch.Tensor],
|
||||
a1q_scale: Optional[torch.Tensor],
|
||||
a2_scale: Optional[torch.Tensor],
|
||||
workspace13: torch.Tensor,
|
||||
workspace2: torch.Tensor,
|
||||
expert_num_tokens: Optional[torch.Tensor],
|
||||
):
|
||||
experts = (self.batched_deep_gemm_experts
|
||||
if self.allow_deep_gemm else self.batched_triton_experts)
|
||||
assert experts is not None
|
||||
experts.apply(output, hidden_states, w1, w2, topk_ids, activation,
|
||||
global_num_experts, expert_map, w1_scale, w2_scale,
|
||||
w1_zp, w2_zp, a1q_scale, a2_scale, workspace13,
|
||||
workspace2, expert_num_tokens)
|
||||
460
vllm/model_executor/layers/fused_moe/config.py
Normal file
460
vllm/model_executor/layers/fused_moe/config.py
Normal file
@@ -0,0 +1,460 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
from dataclasses import dataclass
|
||||
from typing import Optional, Union
|
||||
|
||||
import torch
|
||||
from compressed_tensors.quantization import (QuantizationArgs,
|
||||
QuantizationStrategy,
|
||||
QuantizationType)
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.config import ParallelConfig
|
||||
from vllm.distributed import get_dp_group, get_tensor_model_parallel_rank
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.quantization.base_config import (
|
||||
QuantizationConfig)
|
||||
from vllm.utils import cdiv
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
|
||||
def _get_quant_config_quantization_args(
|
||||
quant_config: Optional[QuantizationConfig],
|
||||
prop_name: str,
|
||||
) -> Optional[QuantizationArgs]:
|
||||
if (quant_config is not None and hasattr(quant_config, 'target_scheme_map')
|
||||
and "Linear" in quant_config.target_scheme_map and
|
||||
"input_activations" in quant_config.target_scheme_map["Linear"]):
|
||||
return quant_config.target_scheme_map["Linear"].get(prop_name)
|
||||
else:
|
||||
return None
|
||||
|
||||
|
||||
def get_quant_config_input_quant(
|
||||
quant_config: Optional[QuantizationConfig]
|
||||
) -> Optional[QuantizationArgs]:
|
||||
return _get_quant_config_quantization_args(quant_config,
|
||||
"input_activations")
|
||||
|
||||
|
||||
def get_quant_config_weight_quant(
|
||||
quant_config: Optional[QuantizationConfig]
|
||||
) -> Optional[QuantizationArgs]:
|
||||
return _get_quant_config_quantization_args(quant_config, "weights")
|
||||
|
||||
|
||||
# TODO (bnell): use scalar_type instead of bools?
|
||||
def get_config_quant_dtype(
|
||||
use_fp8_w8a8: bool,
|
||||
use_int8_w8a8: bool,
|
||||
use_int8_w8a16: bool,
|
||||
use_int4_w4a16: bool,
|
||||
use_int4_w4a8: bool,
|
||||
) -> Optional[torch.dtype]:
|
||||
if use_fp8_w8a8:
|
||||
return torch.float8_e4m3fn
|
||||
elif use_int8_w8a8:
|
||||
return torch.int8
|
||||
return None
|
||||
|
||||
|
||||
@dataclass
|
||||
class FusedMoEQuantConfig:
|
||||
# The post quantization activation type.
|
||||
quant_dtype: Optional[torch.dtype] = None
|
||||
per_act_token_quant: bool = False
|
||||
per_out_ch_quant: bool = False
|
||||
block_shape: Optional[list[int]] = None
|
||||
|
||||
# TODO: add col major flag?
|
||||
# add detailed quant info for input, intermediates, weights, etc?
|
||||
|
||||
def __post_init__(self):
|
||||
assert (not self.per_act_token_quant
|
||||
or self.block_shape is None), "illegal quantization"
|
||||
|
||||
@property
|
||||
def is_quantized(self) -> bool:
|
||||
return self.quant_dtype is not None
|
||||
|
||||
@property
|
||||
def is_per_act_token(self) -> bool:
|
||||
return self.per_act_token_quant
|
||||
|
||||
@property
|
||||
def is_block_quantized(self) -> bool:
|
||||
return self.block_shape is not None
|
||||
|
||||
@property
|
||||
def is_per_tensor(self) -> bool:
|
||||
return not self.per_act_token_quant and self.block_shape is None
|
||||
|
||||
def scale_shape(
|
||||
self,
|
||||
max_tokens: int,
|
||||
hidden_dim: int,
|
||||
) -> Optional[tuple[int, int]]:
|
||||
if self.is_quantized:
|
||||
if self.is_block_quantized:
|
||||
assert self.block_shape is not None
|
||||
_, block_k = self.block_shape
|
||||
k_tiles = cdiv(hidden_dim, block_k)
|
||||
return (max_tokens, k_tiles)
|
||||
elif self.is_per_act_token:
|
||||
return (max_tokens, 1)
|
||||
else:
|
||||
return (1, 1)
|
||||
else:
|
||||
return None
|
||||
|
||||
def batched_scale_shape(
|
||||
self,
|
||||
num_experts: int,
|
||||
max_tokens: int,
|
||||
hidden_dim: int,
|
||||
) -> Optional[tuple[int, int, int]]:
|
||||
if self.is_quantized:
|
||||
scale_shape = self.scale_shape(max_tokens, hidden_dim)
|
||||
assert scale_shape is not None
|
||||
return (num_experts, *scale_shape)
|
||||
else:
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
def make(
|
||||
use_fp8_w8a8: bool = False,
|
||||
use_int8_w8a8: bool = False,
|
||||
use_int8_w8a16: bool = False,
|
||||
use_int4_w4a16: bool = False,
|
||||
use_int4_w4a8: bool = False,
|
||||
per_act_token_quant: bool = False,
|
||||
per_out_ch_quant: bool = False,
|
||||
block_shape: Optional[list[int]] = None,
|
||||
) -> "FusedMoEQuantConfig":
|
||||
assert sum([
|
||||
int(flag) for flag in [
|
||||
use_fp8_w8a8,
|
||||
use_int8_w8a8,
|
||||
use_int8_w8a16,
|
||||
use_int4_w4a16,
|
||||
use_int4_w4a8,
|
||||
]
|
||||
]) <= 1, "Quantization flags are mutually exclusive."
|
||||
|
||||
quant_dtype = get_config_quant_dtype(
|
||||
use_fp8_w8a8=use_fp8_w8a8,
|
||||
use_int8_w8a8=use_int8_w8a8,
|
||||
use_int8_w8a16=use_int8_w8a16,
|
||||
use_int4_w4a16=use_int4_w4a16,
|
||||
use_int4_w4a8=use_int4_w4a8,
|
||||
)
|
||||
return FusedMoEQuantConfig(
|
||||
quant_dtype,
|
||||
per_act_token_quant,
|
||||
per_out_ch_quant,
|
||||
block_shape,
|
||||
)
|
||||
|
||||
|
||||
@dataclass
|
||||
class FusedMoEParallelConfig:
|
||||
tp_size: int
|
||||
dp_size: int
|
||||
ep_size: int
|
||||
tp_rank: int
|
||||
dp_rank: int
|
||||
ep_rank: int
|
||||
|
||||
use_ep: bool # whether to use EP or not
|
||||
|
||||
@property
|
||||
def use_all2all_kernels(self):
|
||||
return self.dp_size > 1 and self.use_ep
|
||||
|
||||
@property
|
||||
def use_pplx_kernels(self):
|
||||
return (self.use_all2all_kernels
|
||||
and envs.VLLM_ALL2ALL_BACKEND == "pplx")
|
||||
|
||||
@property
|
||||
def use_deepep_ht_kernels(self):
|
||||
return (self.use_all2all_kernels
|
||||
and envs.VLLM_ALL2ALL_BACKEND == "deepep_high_throughput")
|
||||
|
||||
@property
|
||||
def use_deepep_ll_kernels(self):
|
||||
return (self.use_all2all_kernels
|
||||
and envs.VLLM_ALL2ALL_BACKEND == "deepep_low_latency")
|
||||
|
||||
@staticmethod
|
||||
def make(tp_size_: int, dp_size_: int,
|
||||
vllm_parallel_config: ParallelConfig) -> "FusedMoEParallelConfig":
|
||||
"""
|
||||
Determine MoE parallel configuration. Based on the input tp_size_,
|
||||
dp_size_, ep_size_ and vllm's parallel config, determine what
|
||||
level's of parallelism to use in the fused moe layer.
|
||||
|
||||
Args:
|
||||
tp_size_ (int): tp_size passed into the FusedMoE constructor.
|
||||
dp_size_ (int): dp_size passed into the FusedMoE constructor.
|
||||
ep_size_ (int): ep_size passed into the FusedMoE constructor.
|
||||
vllm_parallel_config (ParallelConfig): vllm's parallel config
|
||||
object.
|
||||
|
||||
Examples:
|
||||
When there is no parallelism requested, i.e. tp_size_ = dp_size_ = 1,
|
||||
we simply return the sizes unaltered and the ranks set to 0.
|
||||
|
||||
Expert Parallelism is considered only when either dp_size_ or tp_size_
|
||||
is non trivial.
|
||||
|
||||
When TP = 2, DP = 1 and EP = False, the configuration on different
|
||||
devices,
|
||||
- device 0 : TP = {2, 0} DP = {1, 0} EP = {1, 0} //
|
||||
legend : {size, rank}
|
||||
- device 1 : TP = {2, 1} DP = {1, 0} EP = {1, 0}
|
||||
- Comment : Tensors are sharded across 2 devices.
|
||||
|
||||
When TP = 1, DP = 2 and EP = False, the configuration on different
|
||||
devices,
|
||||
- device 0 : TP = {2, 0} DP = {2, 0} EP = {1, 0}
|
||||
- device 1 : TP = {2, 1} DP = {2, 1} EP = {1, 0}
|
||||
- Comment: There are 2 engine instances and the tensors are sharded
|
||||
across 2 decvices.
|
||||
|
||||
When TP = 2, DP = 2 and EP = False, the configuration on different
|
||||
devices,
|
||||
- device 0: TP = {4, 0} DP = {2, 0} EP = {1, 0}
|
||||
- device 1: TP = {4, 1} DP = {2, 0} EP = {1, 0}
|
||||
- device 2: TP = {4, 2} DP = {2, 1} EP = {1, 0}
|
||||
- device 3: TP = {4, 3} DP = {2, 1} EP = {1, 0}
|
||||
- Comment: There are 2 engine instances and the tensors are sharded
|
||||
across 4 devices.
|
||||
|
||||
When, TP = 2, DP = 1 and EP = True, the configuration on different
|
||||
devices,
|
||||
- device 0: TP = {1, 0} DP = {1, 0} EP = {2, 0}
|
||||
- device 1: TP = {1, 0} DP = {1, 0} EP = {2, 1}
|
||||
- Comment: The experts are split between the 2 devices.
|
||||
|
||||
When, TP = 1, DP = 2 and EP = True, the configuration on different
|
||||
devices,
|
||||
- device 0: TP = {1, 0} DP = {2, 0} EP = {2, 0}
|
||||
- device 1: TP = {1, 0} DP = {2, 1} EP = {2, 1}
|
||||
- Comment: There are 2 engine instances and the experts are split
|
||||
between the 2 devices.
|
||||
|
||||
When TP = 2, DP = 2 and EP = True, the configuration on different
|
||||
devices,
|
||||
- device 0: TP = {1, 0} DP = {2, 0} EP = {4, 0}
|
||||
- device 1: TP = {1, 0} DP = {2, 0} EP = {4, 1}
|
||||
- device 2: TP = {1, 0} DP = {2, 1} EP = {4, 2}
|
||||
- device 3: TP = {1, 0} DP = {2, 1} EP = {4, 3}
|
||||
- Comment: There are 2 engine instances and the experts are split
|
||||
between the 4 devices.
|
||||
"""
|
||||
|
||||
def flatten_tp_across_dp(dp_rank: int):
|
||||
tp_rank = 0 if tp_size_ == 1 else get_tensor_model_parallel_rank()
|
||||
# There are actually dp_size_ * tp_size_ devices. Update tp_size
|
||||
# and tp_rank so we shard across all devices.
|
||||
tp_size = dp_size_ * tp_size_
|
||||
tp_rank = dp_rank * tp_size_ + tp_rank
|
||||
return tp_size, tp_rank
|
||||
|
||||
use_ep = (dp_size_ * tp_size_ > 1
|
||||
and vllm_parallel_config.enable_expert_parallel)
|
||||
|
||||
dp_size = dp_size_
|
||||
dp_rank = get_dp_group().rank_in_group if dp_size > 1 else 0
|
||||
tp_size, tp_rank = flatten_tp_across_dp(dp_rank)
|
||||
|
||||
if not use_ep:
|
||||
return FusedMoEParallelConfig(tp_size=tp_size,
|
||||
tp_rank=tp_rank,
|
||||
dp_size=dp_size,
|
||||
dp_rank=dp_rank,
|
||||
ep_size=1,
|
||||
ep_rank=0,
|
||||
use_ep=False)
|
||||
# DP + EP / TP + EP / DP + TP + EP
|
||||
assert use_ep
|
||||
# In EP, each device owns a set of experts fully. There is no tensor
|
||||
# parallel update tp_size, tp_rank, ep_size and ep_rank to reflect that.
|
||||
ep_size = tp_size
|
||||
ep_rank = tp_rank
|
||||
return FusedMoEParallelConfig(tp_size=1,
|
||||
tp_rank=0,
|
||||
dp_size=dp_size,
|
||||
dp_rank=dp_rank,
|
||||
ep_size=ep_size,
|
||||
ep_rank=ep_rank,
|
||||
use_ep=True)
|
||||
|
||||
|
||||
# Adapted from pplx-kernels tests/all_to_all_utils.py
|
||||
@dataclass
|
||||
class FusedMoEConfig:
|
||||
num_experts: int
|
||||
experts_per_token: int
|
||||
hidden_dim: int
|
||||
|
||||
num_local_experts: int
|
||||
moe_parallel_config: FusedMoEParallelConfig
|
||||
|
||||
# The activation type.
|
||||
in_dtype: torch.dtype
|
||||
|
||||
quant_config: Optional[FusedMoEQuantConfig] = None
|
||||
|
||||
max_num_tokens: int = envs.VLLM_MOE_DP_CHUNK_SIZE
|
||||
|
||||
def __post_init__(self):
|
||||
if self.dp_size > 1:
|
||||
logger.debug("Using FusedMoEConfig::max_num_tokens=%d",
|
||||
self.max_num_tokens)
|
||||
|
||||
assert self.max_num_tokens > 0
|
||||
|
||||
@property
|
||||
def quant_dtype(self) -> Optional[torch.dtype]:
|
||||
if self.quant_config is not None:
|
||||
return self.quant_config.quant_dtype
|
||||
else:
|
||||
return None
|
||||
|
||||
@property
|
||||
def block_shape(self) -> Optional[list[int]]:
|
||||
if self.quant_config is not None:
|
||||
return self.quant_config.block_shape
|
||||
else:
|
||||
return None
|
||||
|
||||
@property
|
||||
def per_act_token_quant(self) -> bool:
|
||||
if self.quant_config is not None:
|
||||
return self.quant_config.per_act_token_quant
|
||||
else:
|
||||
return False
|
||||
|
||||
@property
|
||||
def per_out_ch_quant(self) -> bool:
|
||||
if self.quant_config is not None:
|
||||
return self.quant_config.per_out_ch_quant
|
||||
else:
|
||||
return False
|
||||
|
||||
@property
|
||||
def tp_size(self):
|
||||
return self.moe_parallel_config.tp_size
|
||||
|
||||
@property
|
||||
def dp_size(self):
|
||||
return self.moe_parallel_config.dp_size
|
||||
|
||||
@property
|
||||
def ep_size(self):
|
||||
return self.moe_parallel_config.ep_size
|
||||
|
||||
@property
|
||||
def tp_rank(self):
|
||||
return self.moe_parallel_config.tp_rank
|
||||
|
||||
@property
|
||||
def dp_rank(self):
|
||||
return self.moe_parallel_config.dp_rank
|
||||
|
||||
@property
|
||||
def ep_rank(self):
|
||||
return self.moe_parallel_config.ep_rank
|
||||
|
||||
@property
|
||||
def use_ep(self):
|
||||
return self.moe_parallel_config.use_ep
|
||||
|
||||
@property
|
||||
def use_pplx_kernels(self):
|
||||
return self.moe_parallel_config.use_pplx_kernels
|
||||
|
||||
@property
|
||||
def use_deepep_ht_kernels(self):
|
||||
return self.moe_parallel_config.use_deepep_ht_kernels
|
||||
|
||||
@property
|
||||
def use_deepep_ll_kernels(self):
|
||||
return self.moe_parallel_config.use_deepep_ll_kernels
|
||||
|
||||
@staticmethod
|
||||
def make(
|
||||
num_experts: int,
|
||||
experts_per_token: int,
|
||||
hidden_dim: int,
|
||||
num_local_experts: int,
|
||||
moe_parallel_config: FusedMoEParallelConfig,
|
||||
in_dtype: torch.dtype,
|
||||
max_num_tokens: int = envs.VLLM_MOE_DP_CHUNK_SIZE,
|
||||
quant_config: Optional[Union[FusedMoEQuantConfig,
|
||||
QuantizationConfig]] = None
|
||||
) -> "FusedMoEConfig":
|
||||
|
||||
_quant_config: Optional[FusedMoEQuantConfig] = None
|
||||
|
||||
if quant_config is not None and isinstance(quant_config,
|
||||
QuantizationConfig):
|
||||
if hasattr(quant_config, 'weight_block_size'):
|
||||
block_shape = quant_config.weight_block_size
|
||||
else:
|
||||
block_shape = None
|
||||
per_act_token_quant = False
|
||||
per_out_ch_quant = False
|
||||
quant_dtype: Optional[torch.dtype] = None
|
||||
|
||||
input_quant = get_quant_config_input_quant(quant_config)
|
||||
weight_quant = get_quant_config_weight_quant(quant_config)
|
||||
|
||||
if input_quant is not None:
|
||||
per_act_token_quant = (input_quant.strategy
|
||||
== QuantizationStrategy.TOKEN
|
||||
if input_quant is not None else False)
|
||||
|
||||
if input_quant.num_bits == 8:
|
||||
if input_quant.type == QuantizationType.FLOAT:
|
||||
quant_dtype = torch.float8_e4m3fn
|
||||
elif input_quant.type == QuantizationType.INT:
|
||||
quant_dtype = torch.int8
|
||||
|
||||
from vllm.model_executor.layers.quantization.fp8 import Fp8Config
|
||||
if quant_dtype is None and isinstance(quant_config, Fp8Config):
|
||||
quant_dtype = torch.float8_e4m3fn
|
||||
|
||||
if weight_quant is not None:
|
||||
per_out_ch_quant = (
|
||||
weight_quant.strategy == QuantizationStrategy.CHANNEL)
|
||||
|
||||
if quant_dtype is not None:
|
||||
_quant_config = FusedMoEQuantConfig(
|
||||
quant_dtype=quant_dtype,
|
||||
per_act_token_quant=per_act_token_quant,
|
||||
per_out_ch_quant=per_out_ch_quant,
|
||||
block_shape=block_shape,
|
||||
)
|
||||
else:
|
||||
_quant_config = FusedMoEQuantConfig()
|
||||
logger.warning_once("MoE DP setup unable to determine "
|
||||
"quantization scheme or unsupported "
|
||||
"quantization type. This model will "
|
||||
"not run with DP enabled.")
|
||||
else:
|
||||
_quant_config = quant_config
|
||||
|
||||
return FusedMoEConfig(
|
||||
num_experts=num_experts,
|
||||
experts_per_token=experts_per_token,
|
||||
hidden_dim=hidden_dim,
|
||||
num_local_experts=num_local_experts,
|
||||
moe_parallel_config=moe_parallel_config,
|
||||
in_dtype=in_dtype,
|
||||
quant_config=_quant_config,
|
||||
max_num_tokens=max_num_tokens,
|
||||
)
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,182 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"6144": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,147 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,130 @@
|
||||
{
|
||||
"3328": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"768": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2560": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2816": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3584": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3840": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1280": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2304": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,218 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"5120": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"9216": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"13312": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"17408": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"25600": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"33792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"41984": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"50176": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"58368": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,130 @@
|
||||
{
|
||||
"3840": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3584": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2816": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1280": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"768": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3328": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2560": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2304": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 5
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 256,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,130 @@
|
||||
{
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1792": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3328": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2560": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"768": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2816": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2304": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1280": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"3840": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3584": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 1
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,173 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,182 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"6144": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,182 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"6144": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 32,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 1,
|
||||
"num_ldmatrixes": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,164 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 16,
|
||||
"BLOCK_SIZE_K": 256,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 1
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 4,
|
||||
"num_warps": 1,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 2,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 8,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2,
|
||||
"waves_per_eu": 0,
|
||||
"matrix_instr_nonkdim": 16,
|
||||
"kpack": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 32,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 2
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 128,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 8,
|
||||
"num_stages": 2
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"1": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 64,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"2": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"4": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"8": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 8,
|
||||
"num_stages": 4
|
||||
},
|
||||
"16": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 5
|
||||
},
|
||||
"24": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 64,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"32": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"48": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"64": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"96": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 128,
|
||||
"BLOCK_SIZE_K": 64,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 4
|
||||
},
|
||||
"512": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 32,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"1536": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 1,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"3072": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"BLOCK_SIZE_N": 256,
|
||||
"BLOCK_SIZE_K": 128,
|
||||
"GROUP_SIZE_M": 16,
|
||||
"num_warps": 4,
|
||||
"num_stages": 3
|
||||
}
|
||||
}
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user