ref(upstream): FULL TREE — Deep-Spark xllm (1470) + ds_vllm csrc/models (703)

Replaces cherry-picked upstream_ref with complete source trees.

xllm/ — Iluvatar official C++ inference engine (15MB, 1470 files)
  Complete: kernels → layers → models → runtime → scheduler → api
  Excluded: .git, binary images, third_party submodule checkouts

ds_vllm/ — Iluvatar official vllm fork (8MB, 703 files)
  Included: csrc/ (ALL CUDA kernels), fused_moe/, qwen3_5 model, _custom_ops
  Excluded: tests, benchmarks, docs, examples (not needed for reference)

Critical call chains now fully traceable:
  MoE: moe_topk_softmax_kernels.cuh → ixformer.h → fused_moe.cpp → layer
  GDN: qwen3_gated_delta_net_base.cpp → qwen3_5_gated_delta_net.cpp
  Attention: ixformer.h → xllm_paged_attention → attention.cpp
This commit is contained in:
EX Engine
2026-08-10 02:53:54 +00:00
parent 9e4fb3712f
commit 002f9879b2
2179 changed files with 494021 additions and 79 deletions

View File

@@ -0,0 +1,36 @@
# PpMatmul 算子优化
## 背景
针对大模型推理中矩阵乘法占比高、耗时长的问题,优化了矩阵乘法算子的实现。
## 功能介绍
PpMatmul 算子使用 Tiling 切分策略,将矩阵乘法分解为多个小的矩阵乘法任务。然而当 tile 数量较小时任务无法被均匀分配到所有 npu 核心上,导致 tail effect 问题,影响计算效率。我们通过预取内存或重新划分任务的方式,优化 PpMatmul 算子的性能。
## 用户接口
### 算子直调 API
```cpp
aclnnStatus aclnnPpMatmulOptGetWorkspaceSize(
const aclTensor *a,
const aclTensor *b,
const aclTensor *out,
uint64_t *workspaceSize,
aclOpExecutor **executor);
aclnnStatus aclnnPpMatmulOpt(
void *workspace,
uint64_t workspaceSize,
aclOpExecutor *executor,
aclrtStream stream);
```
- `a`: 输入矩阵 A。
- `b`: 输入矩阵 B。
- `out`: 输出矩阵,存储计算结果。
## 性能效果
对于 tile 数量较小的情况(例如 M 较小,对应于 batch size 较小的情况TP=4算子较优化前有 **18%** 的性能提升。