[ROCm][Kernel] Enable permute_cols for ROCm (#44674)

Signed-off-by: charlifu <charlifu@amd.com>
This commit is contained in:
Charlie Fu
2026-06-07 09:50:03 +00:00
committed by GitHub
parent 3d3ba460a2
commit 228bcc436b
3 changed files with 3 additions and 8 deletions
+1 -3
View File
@@ -629,6 +629,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
"csrc/libtorch_stable/quantization/w8a8/fp8/common.cu"
"csrc/libtorch_stable/quantization/w8a8/fp8/per_token_group_quant.cu"
"csrc/libtorch_stable/quantization/w8a8/int8/per_token_group_quant.cu"
"csrc/libtorch_stable/permute_cols.cu"
"csrc/libtorch_stable/quantization/gptq/q_gemm.cu"
"csrc/libtorch_stable/quantization/gguf/gguf_kernel.cu"
"csrc/libtorch_stable/pos_encoding_kernels.cu"
@@ -657,9 +658,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
"csrc/libtorch_stable/quantization/w8a8/cutlass/scaled_mm_entry.cu"
"csrc/libtorch_stable/quantization/fp4/nvfp4_quant_entry.cu"
"csrc/libtorch_stable/quantization/fp4/nvfp4_scaled_mm_entry.cu"
"csrc/libtorch_stable/permute_cols.cu"
"csrc/libtorch_stable/quantization/w8a8/fp8/per_token_group_quant.cu"
"csrc/libtorch_stable/quantization/w8a8/int8/per_token_group_quant.cu"
"csrc/libtorch_stable/quantization/awq/gemm_kernels.cu"
"csrc/libtorch_stable/minimax_reduce_rms_kernel.cu")
+1 -1
View File
@@ -24,10 +24,10 @@ void per_token_group_quant_int8(const torch::stable::Tensor& input,
int64_t group_size, double eps, double int8_min,
double int8_max);
#ifndef USE_ROCM
torch::stable::Tensor permute_cols(torch::stable::Tensor const& A,
torch::stable::Tensor const& perm);
#ifndef USE_ROCM
bool cutlass_scaled_mm_supports_fp8(int64_t cuda_device_capability);
bool cutlass_scaled_mm_supports_block_fp8(int64_t cuda_device_capability);
bool cutlass_group_gemm_supported(int64_t cuda_device_capability);
+1 -4
View File
@@ -27,13 +27,12 @@ STABLE_TORCH_LIBRARY_FRAGMENT(_C, ops) {
"per_token_group_quant_int8(Tensor input, Tensor! output_q, Tensor! "
"output_s, int group_size, float eps, float int8_min, float int8_max) -> "
"()");
ops.def("permute_cols(Tensor A, Tensor perm) -> Tensor");
#ifndef USE_ROCM
// TODO: Remove this once ROCm upgrade to torch 2.11.
ops.def("get_cuda_view_from_cpu_tensor(Tensor cpu_tensor) -> Tensor");
ops.def("permute_cols(Tensor A, Tensor perm) -> Tensor");
#endif
#ifndef USE_ROCM
@@ -562,9 +561,7 @@ STABLE_TORCH_LIBRARY_IMPL(_C, CUDA, ops) {
ops.impl("per_token_group_quant_int8",
TORCH_BOX(&per_token_group_quant_int8));
#ifndef USE_ROCM
ops.impl("permute_cols", TORCH_BOX(&permute_cols));
#endif
#ifndef USE_ROCM
// CUTLASS scaled_mm ops