cf1b701afe
唯一改动: computility-run.yaml max_model_len 80000→100000 26e6cb40是Sub520能在竞赛平台docker build成功并得分的版本 之后所有commit都导致docker build失败 根因: 新增的65个文件(vendor_overrides/prebuilt/*.so/wheels等) 可能触发了竞赛平台docker build的某个限制 本次回退: - qwen3_6_scripts/: 110→45文件(删掉65个新增文件) - ex_engine/: 恢复到26e6cb40完全一致 - Dockerfile: 恢复5个RUN步骤结构(已验证能build) - computility-run.yaml: max_model_len=100000(避免replay 400拒绝)
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 32
|
|
#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
|