xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-08 03:29:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm] warpSize is being made non constexpr in ROCm 7.0 (#20330)

Browse Source 
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
This commit is contained in:
Gregory Shtrasberg 2025-07-15 14:01:44 -04:00 committed by GitHub
parent b637e9dcb8
commit ed10f3cea1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 6 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
csrc/attention/attention_kernels.cuh
View File

@ -24,6 +24,7 @@
#include "attention_dtypes.h" #include "attention_dtypes.h"
#include "attention_utils.cuh" #include "attention_utils.cuh"
#include "cuda_compat.h"
#ifdef USE_ROCM #ifdef USE_ROCM
#include <hip/hip_bf16.h> #include <hip/hip_bf16.h>
@ -33,12 +34,6 @@ typedef __hip_bfloat16 __nv_bfloat16;
#include "../quantization/fp8/nvidia/quant_utils.cuh" #include "../quantization/fp8/nvidia/quant_utils.cuh"
#endif #endif
#ifndef USE_ROCM
#define WARP_SIZE 32
#else
#define WARP_SIZE warpSize
#endif
#define MAX(a, b) ((a) > (b) ? (a) : (b)) #define MAX(a, b) ((a) > (b) ? (a) : (b))
#define MIN(a, b) ((a) < (b) ? (a) : (b)) #define MIN(a, b) ((a) < (b) ? (a) : (b))
#define DIVIDE_ROUND_UP(a, b) (((a) + (b) - 1) / (b)) #define DIVIDE_ROUND_UP(a, b) (((a) + (b) - 1) / (b))
@ -670,7 +665,6 @@ __global__ void paged_attention_v2_reduce_kernel(
} // namespace vllm } // namespace vllm
#undef WARP_SIZE
#undef MAX #undef MAX
#undef MIN #undef MIN
#undef DIVIDE_ROUND_UP #undef DIVIDE_ROUND_UP

8
csrc/attention/paged_attention_v1.cu
View File

@ -18,12 +18,7 @@
*/ */
#include "attention_kernels.cuh" #include "attention_kernels.cuh"
#include "cuda_compat.h"
#ifndef USE_ROCM
#define WARP_SIZE 32
#else
#define WARP_SIZE warpSize
#endif
#define MAX(a, b) ((a) > (b) ? (a) : (b)) #define MAX(a, b) ((a) > (b) ? (a) : (b))
#define MIN(a, b) ((a) < (b) ? (a) : (b)) #define MIN(a, b) ((a) < (b) ? (a) : (b))
@ -187,7 +182,6 @@ void paged_attention_v1(
CALL_V1_LAUNCHER_BLOCK_SIZE) CALL_V1_LAUNCHER_BLOCK_SIZE)
} }
#undef WARP_SIZE
#undef MAX #undef MAX
#undef MIN #undef MIN
#undef DIVIDE_ROUND_UP #undef DIVIDE_ROUND_UP

8
csrc/attention/paged_attention_v2.cu
View File

@ -18,12 +18,7 @@
*/ */
#include "attention_kernels.cuh" #include "attention_kernels.cuh"
#include "cuda_compat.h"
#ifndef USE_ROCM
#define WARP_SIZE 32
#else
#define WARP_SIZE warpSize
#endif
#define MAX(a, b) ((a) > (b) ? (a) : (b)) #define MAX(a, b) ((a) > (b) ? (a) : (b))
#define MIN(a, b) ((a) < (b) ? (a) : (b)) #define MIN(a, b) ((a) < (b) ? (a) : (b))
@ -197,7 +192,6 @@ void paged_attention_v2(
CALL_V2_LAUNCHER_BLOCK_SIZE) CALL_V2_LAUNCHER_BLOCK_SIZE)
} }
#undef WARP_SIZE
#undef MAX #undef MAX
#undef MIN #undef MIN
#undef DIVIDE_ROUND_UP #undef DIVIDE_ROUND_UP

6
csrc/cuda_compat.h
View File

@ -4,10 +4,10 @@
#include <hip/hip_runtime.h> #include <hip/hip_runtime.h>
#endif #endif
#ifndef USE_ROCM #if defined(USE_ROCM) && defined(__GFX9__)
#define WARP_SIZE 32 #define WARP_SIZE 64
#else #else
#define WARP_SIZE warpSize #define WARP_SIZE 32
#endif #endif
#ifndef USE_ROCM #ifndef USE_ROCM