a8acfbbb8f
26e6cb40也无法通过竞赛平台build，说明平台环境已变化。 去掉所有 | tee（可能在某些shell配置下传播错误码）， 每个RUN命令直接用 || true 结尾，绝对不可能返回非零。
74 lines
2.3 KiB
C++
74 lines
2.3 KiB
C++
/* Copyright 2025 The xLLM Authors. All Rights Reserved.
|
|
|
|
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
|
|
|
|
https://github.com/jd-opensource/xllm/blob/main/LICENSE
|
|
|
|
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.
|
|
==============================================================================*/
|
|
|
|
#include "ilu_ops_api.h"
|
|
|
|
|
|
namespace xllm::kernel::ilu {
|
|
|
|
bool gemv_conditions(const torch::Tensor& input,
|
|
const torch::Tensor& weight,
|
|
const torch::Tensor& bias,
|
|
int64_t gemv_max_batch) {
|
|
// gemv input:[m,k] weight:[n,k]
|
|
// 1. m <= gemv_max_batch
|
|
// 2. k % 32 == 0 && n % 2 == 0
|
|
// 3. bias is None
|
|
|
|
torch::Tensor input_view = input.view({-1, input.size(-1)});
|
|
torch::Tensor weight_view = weight.view({-1, weight.size(-1)});
|
|
|
|
int64_t m = input_view.size(0);
|
|
int64_t k = input_view.size(1);
|
|
int64_t n = weight_view.size(0);
|
|
|
|
if (bias.defined() == false && m <= gemv_max_batch && k % 32 == 0 &&
|
|
n % 2 == 0) {
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
torch::Tensor matmul(torch::Tensor a,
|
|
torch::Tensor b,
|
|
c10::optional<torch::Tensor> bias) {
|
|
int64_t act_type = -1;
|
|
bool persistent = false;
|
|
std::vector<int64_t> output_shape = a.sizes().vec();
|
|
if (!output_shape.empty()) {
|
|
output_shape[output_shape.size() - 1] = b.size(0);
|
|
}
|
|
torch::Tensor output = a.new_empty(output_shape);
|
|
|
|
bool use_gemv = true;
|
|
const int64_t gemv_max_batch = 1;
|
|
const bool disable_infer_gemm_ex =
|
|
std::getenv("DISABLE_INFER_GEMM_EX") != nullptr;
|
|
|
|
use_gemv =
|
|
use_gemv &&
|
|
gemv_conditions(a, b, bias.value_or(at::Tensor()), gemv_max_batch) &&
|
|
!disable_infer_gemm_ex && (act_type == -1);
|
|
|
|
if (use_gemv) {
|
|
output = infer::ixformer_linear_ex(a, b, bias, output);
|
|
} else {
|
|
output = infer::ixformer_linear(a, b, act_type, bias, output, persistent);
|
|
}
|
|
return output;
|
|
}
|
|
|
|
} // namespace xllm::kernel::ilu
|