xc-llm-ascend/vllm_ascend/compilation/passes/qknorm_rope_fusion_pass.py

#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
# This file is a part of the vllm-ascend project.
#
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import torch
from torch._inductor.pattern_matcher import PatternMatcherPass, PatternPrettyPrinter
from vllm.compilation.passes.vllm_inductor_pass import VllmInductorPass
from vllm.config import VllmConfig, get_layers_from_vllm_config
from vllm.config.compilation import Range
from vllm.logger import logger
from vllm.model_executor.layers.attention import Attention

from vllm_ascend.compilation.passes.base_pattern import BasePattern
from vllm_ascend.utils import get_rope_dim


class QKNormRopeFusionPattern(BasePattern):
    def __init__(self, vllm_config, head_dim, num_heads, num_kv_heads, eps=1e-6):
        super().__init__(vllm_config, eps)
        self.head_dim = head_dim
        self.num_heads = num_heads
        self.num_kv_heads = num_kv_heads
        self.q_size = self.num_heads * self.head_dim
        self.kv_size = self.num_kv_heads * self.head_dim
        self.device = vllm_config.device_config.device if vllm_config.device_config else None
        self.rope_dim = get_rope_dim(vllm_config)

    def get_inputs(self):
        T = 5
        max_position_embeddings = 16384
        qkv = torch.empty(T, self.q_size + 2 * self.kv_size, dtype=torch.bfloat16, device="npu")
        q_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        k_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        cos_sin_cache = torch.empty(max_position_embeddings, self.head_dim, dtype=torch.bfloat16, device="npu")
        positions = torch.ones(T, dtype=torch.int64, device="npu")
        return [qkv, q_weight, k_weight, cos_sin_cache, positions]

    def get_pattern(self):
        def pattern(
            qkv: torch.Tensor,
            q_weight: torch.Tensor,
            k_weight: torch.Tensor,
            cos_sin_cache: torch.Tensor,
            positions: torch.Tensor,
        ):
            q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)

            q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim, self.head_dim)
            q_norm_out, _ = torch.ops.npu.npu_rms_norm(q_by_head, q_weight, self.eps)

            k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim, self.head_dim)
            k_norm_out, _ = torch.ops.npu.npu_rms_norm(k_by_head, k_weight, self.eps)

            q_flat = q_norm_out.view(q.shape)
            k_flat = k_norm_out.view(k.shape)
            q_rope, k_rope = torch.ops.vllm.npu_rotary_embedding(
                positions, q_flat, k_flat, cos_sin_cache, self.head_dim, self.rope_dim, True
            )

            return q_rope, k_rope, v

        return pattern

    def get_replacement(self):
        def replacement(
            qkv: torch.Tensor,
            q_weight: torch.Tensor,
            k_weight: torch.Tensor,
            cos_sin_cache: torch.Tensor,
            positions: torch.Tensor,
        ):
            results = torch.ops.vllm.qkv_rmsnorm_rope(
                input=qkv,
                q_weight=q_weight,
                k_weight=k_weight,
                q_hidden_size=self.q_size,
                kv_hidden_size=self.kv_size,
                head_dim=self.head_dim,
                eps=self.eps,
                q_bias=None,
                k_bias=None,
                cos_sin_cache=cos_sin_cache,
                positions=positions,
            )

            return results

        return replacement


class QKNormRopeFusionPatternWithBias(BasePattern):
    def __init__(self, vllm_config, head_dim, num_heads, num_kv_heads, eps=1e-6):
        super().__init__(vllm_config, eps)
        self.head_dim = head_dim
        self.num_heads = num_heads
        self.num_kv_heads = num_kv_heads
        self.q_size = self.num_heads * self.head_dim
        self.kv_size = self.num_kv_heads * self.head_dim
        self.device = vllm_config.device_config.device if vllm_config.device_config else None
        self.rope_dim = get_rope_dim(vllm_config)

    def get_inputs(self):
        T = 5
        max_position_embeddings = 16384
        qkv = torch.empty(T, self.q_size + 2 * self.kv_size, dtype=torch.bfloat16, device="npu")
        q_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        k_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        q_bias = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        k_bias = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
        cos_sin_cache = torch.empty(max_position_embeddings, self.head_dim, dtype=torch.bfloat16, device="npu")
        positions = torch.ones(T, dtype=torch.int64, device="npu")

        return [qkv, q_weight, k_weight, q_bias, k_bias, cos_sin_cache, positions]

    def get_pattern(self):
        def pattern(
            qkv: torch.Tensor,
            q_weight: torch.Tensor,
            k_weight: torch.Tensor,
            q_bias: torch.Tensor,
            k_bias: torch.Tensor,
            cos_sin_cache: torch.Tensor,
            positions: torch.Tensor,
        ):
            q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)

            q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim, self.head_dim)
            q_norm_out, _ = torch.ops.npu.npu_rms_norm(q_by_head, q_weight, self.eps)
            q_normed = q_norm_out + q_bias

            k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim, self.head_dim)
            k_norm_out, _ = torch.ops.npu.npu_rms_norm(k_by_head, k_weight, self.eps)
            k_normed = k_norm_out + k_bias

            q_flat = q_normed.view(q.shape)
            k_flat = k_normed.view(k.shape)
            q_rope, k_rope = torch.ops.vllm.npu_rotary_embedding(
                positions, q_flat, k_flat, cos_sin_cache, self.head_dim, self.rope_dim, True
            )

            return q_rope, k_rope, v

        return pattern

    def get_replacement(self):
        def replacement(
            qkv: torch.Tensor,
            q_weight: torch.Tensor,
            k_weight: torch.Tensor,
            q_bias: torch.Tensor,
            k_bias: torch.Tensor,
            cos_sin_cache: torch.Tensor,
            positions: torch.Tensor,
        ):
            results = torch.ops.vllm.qkv_rmsnorm_rope(
                input=qkv,
                q_weight=q_weight,
                k_weight=k_weight,
                q_hidden_size=self.q_size,
                kv_hidden_size=self.kv_size,
                head_dim=self.head_dim,
                eps=self.eps,
                q_bias=q_bias,
                k_bias=k_bias,
                cos_sin_cache=cos_sin_cache,
                positions=positions,
            )
            return results

        return replacement


class QKNormRopeFusionPass(VllmInductorPass):
    """
    A pass for fusing QKV split and RMSNorm operations into a single qk_rmsnorm operator.
    """

    def __init__(self, vllm_config: VllmConfig):
        super().__init__(vllm_config)
        self.pattern_match_passes: PatternMatcherPass = PatternMatcherPass(pass_name="qknorm_rope_fusion_pass")

        dtype = vllm_config.model_config.dtype
        if dtype not in (torch.bfloat16,):
            logger.debug("QKNorm and Rope fusion not enabled: unsupported dtype %s", dtype)
            return

        # use one attn layer to get meta (such as head_dim) for QKNormRopeFusionPattern
        attn_layers: dict[str, Attention] = get_layers_from_vllm_config(vllm_config, Attention)
        if len(attn_layers) == 0:
            logger.debug("QKNorm and Rope fusion enabled, but no Attention layers were discovered.")
            return
        layer = next(iter(attn_layers.values()))
        for epsilon in [1e-6, 1e-5]:
            if layer.head_size != 128:
                logger.debug("QKNorm and Rope fusion not enabled: head_dim %d is not equal of 128", layer.head_size)
                continue
            QKNormRopeFusionPattern(
                vllm_config=vllm_config,
                head_dim=layer.head_size,
                num_heads=layer.num_heads,
                num_kv_heads=layer.num_kv_heads,
                eps=epsilon,
            ).register(self.pattern_match_passes)

            QKNormRopeFusionPatternWithBias(
                vllm_config=vllm_config,
                head_dim=layer.head_size,
                num_heads=layer.num_heads,
                num_kv_heads=layer.num_kv_heads,
                eps=epsilon,
            ).register(self.pattern_match_passes)

    def __call__(self, graph: torch.fx.Graph):
        self.begin()
        self.matched_count = self.pattern_match_passes.apply(graph)
        logger.debug("Fused %s QKNorm and Rope patterns", self.matched_count)
        logger.debug("Patterns registered for replacement:")
        pattern_idx = 0
        for pattern_entry in self.pattern_match_passes.patterns.values():
            for p in pattern_entry:
                p_str = PatternPrettyPrinter.run(p.pattern)
                logger.debug("Pattern %d: %s", pattern_idx, p_str)
                pattern_idx += 1
        self.end_and_log()

    def is_applicable_for_range(self, compile_range: Range) -> bool:
        """
        Check if the pass is applicable for the current configuration.
        """
        return True