biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[fix] CPUDNNLGEMMHandler pointer baked into inductor artifact (#32913)

Browse Source 
Signed-off-by: dolpm <34420038+dolpm@users.noreply.github.com>
This commit is contained in:
dolpm
2026-01-26 13:59:44 -08:00
committed by GitHub
parent 6ee7f18f33
commit 58a05b0ca1
4 changed files with 38 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
csrc/cpu/torch_bindings.cpp
View File

@@ -19,13 +19,14 @@ void onednn_scaled_mm(torch::Tensor& c, const torch::Tensor& a,
const std::optional<torch::Tensor>& azp,
const std::optional<torch::Tensor>& azp_adj,
const std::optional<torch::Tensor>& bias,
int64_t handler);
const torch::Tensor& handler_tensor);
int64_t create_onednn_mm_handler(const torch::Tensor& b,
int64_t primitive_cache_size);
void onednn_mm(torch::Tensor& c, const torch::Tensor& a,
const std::optional<torch::Tensor>& bias, int64_t handler);
const std::optional<torch::Tensor>& bias,
const torch::Tensor& handler_tensor);
bool is_onednn_acl_supported();
@@ -196,7 +197,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
// oneDNN GEMM
ops.def(
"onednn_mm(Tensor! c, Tensor a, Tensor? bias, "
"int handler) -> ()");
"Tensor handler_tensor) -> ()");
ops.impl("onednn_mm", torch::kCPU, &onednn_mm);
// Check if oneDNN was built with ACL backend
@@ -212,7 +213,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
// oneDNN scaled_mm for W8A8 with static per-tensor activation quantization
ops.def(
"onednn_scaled_mm(Tensor! c, Tensor a, Tensor a_scales, Tensor? azp, "
"Tensor? azp_adj, Tensor? bias, int handler) -> ()");
"Tensor? azp_adj, Tensor? bias, Tensor handler_tensor) -> ()");
ops.impl("onednn_scaled_mm", torch::kCPU, &onednn_scaled_mm);
// Compute int8 quantized tensor for given scaling factor.