0478628f17
Tested on real machine (cc-b2042074, BI-V100, IX-ML 3.2.3): _moe_C.topk_softmax() → SUCCESS, correct output Two fixes proven on hardware: 1. cuda_compat.h: WARP_SIZE=64 (BI-V100 warp is 64, not 32) 2. topk_softmax_kernels.cu: cub/block/block_reduce.cuh instead of cub/cub.cuh (cub.cuh pulls radix_sort which has WARP_SIZE conflict) Key finding: ixformer SDK on this base image does NOT have topk_softmax. The ixformer::infer namespace from xllm's ixformer.h is for newer SDK. We MUST compile our own _moe_C kernel — which now works. Build flags (clang 16, ivcore10): CUDA: -O3 -cl-fast-relaxed-math (NOT --use_fast_math) C++: -O2 -std=c++17 Dockerfile simplified: 3 steps (was 6) _custom_ops.py: _moe_C as Priority 0, in-place vllm API
50 lines
1.4 KiB
C
50 lines
1.4 KiB
C
#pragma once
|
|
|
|
#ifdef USE_ROCM
|
|
#include <hip/hip_runtime.h>
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define WARP_SIZE 64
|
|
#else
|
|
#define WARP_SIZE warpSize
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define VLLM_LDG(arg) __ldg(arg)
|
|
#else
|
|
#define VLLM_LDG(arg) *(arg)
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define VLLM_SHFL_XOR_SYNC(var, lane_mask) \
|
|
__shfl_xor_sync(uint32_t(-1), var, lane_mask)
|
|
#define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
|
|
__shfl_xor_sync(uint32_t(-1), var, lane_mask, width)
|
|
#else
|
|
#define VLLM_SHFL_XOR_SYNC(var, lane_mask) __shfl_xor(var, lane_mask)
|
|
#define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
|
|
__shfl_xor(var, lane_mask, width)
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define VLLM_SHFL_SYNC(var, src_lane) __shfl_sync(uint32_t(-1), var, src_lane)
|
|
#else
|
|
#define VLLM_SHFL_SYNC(var, src_lane) __shfl(var, src_lane)
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define VLLM_SHFL_DOWN_SYNC(var, lane_delta) \
|
|
__shfl_down_sync(uint32_t(-1), var, lane_delta)
|
|
#else
|
|
#define VLLM_SHFL_DOWN_SYNC(var, lane_delta) __shfl_down(var, lane_delta)
|
|
#endif
|
|
|
|
#ifndef USE_ROCM
|
|
#define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
|
|
cudaFuncSetAttribute(FUNC, cudaFuncAttributeMaxDynamicSharedMemorySize, VAL)
|
|
#else
|
|
#define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
|
|
hipFuncSetAttribute(FUNC, hipFuncAttributeMaxDynamicSharedMemorySize, VAL)
|
|
#endif
|