frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Compile] Fix Compile Warning for w4a8_mm_entry.cu (#23660)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2025-09-02 23:45:52 -04:00
committed by GitHub
parent 1bd007f234
commit c4ed78b14f
1 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
csrc/quantization/cutlass_w4a8/w4a8_mm_entry.cu
View File

@ -11,6 +11,7 @@
#include "core/registration.h"
#include "cutlass/cutlass.h"
#include <limits>
#include "cute/tensor.hpp"
#include "cutlass/gemm/collective/collective_builder.hpp"
@ -169,6 +170,11 @@ struct W4A8GemmKernel {
int k = A.size(1);
int n = B.size(1);
// safely cast group_size to int
TORCH_CHECK(group_size > 0 && group_size <= std::numeric_limits<int>::max(),
"group_size out of supported range for int: ", group_size);
int const group_size_int = static_cast<int>(group_size);
// Allocate output
const at::cuda::OptionalCUDAGuard device_guard(device_of(A));
auto device = A.device();
@ -192,7 +198,7 @@ struct W4A8GemmKernel {
cute::tile_to_shape(LayoutAtomQuant{}, shape_B);
// strides
int const scale_k = cutlass::ceil_div(k, group_size);
int const scale_k = cutlass::ceil_div(k, group_size_int);
StrideA stride_A =
cutlass::make_cute_packed_stride(StrideA{}, cute::make_shape(m, k, 1));
// Reverse stride here due to swap and transpose
@ -211,8 +217,8 @@ struct W4A8GemmKernel {
using EpilogueArguments = typename GemmKernelShuffled::EpilogueArguments;
MainloopArguments mainloop_arguments{
B_ptr, layout_B_reordered, A_ptr, stride_A,
S_ptr, stride_S, group_size};
B_ptr, layout_B_reordered, A_ptr, stride_A,
S_ptr, stride_S, group_size_int};
EpilogueArguments epilogue_arguments{
ChTokScalesEpilogue::prepare_args(channel_scales, token_scales),