Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
cuda_compat.h 1.98 KB
Newer Older
raojy's avatar
raw_vllm  
raojy committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14 
#pragma once

#ifdef USE_ROCM
  #include <hip/hip_runtime.h>
#endif

#ifdef USE_ROCM
struct Utils {
  static __host__ int get_warp_size() {
    static bool is_cached = false;
    static int result;

    if (!is_cached) {
      int device_id;
raojy's avatar
nemotron_enable  
raojy committed
15
16
17
18
19
20 
      // cudaDeviceProp deviceProp;
      // cudaGetDevice(&device_id);
      // cudaGetDeviceProperties(&deviceProp, device_id);
      hipDeviceProp_t deviceProp;
      hipGetDevice(&device_id);
      hipGetDeviceProperties(&deviceProp, device_id);
raojy's avatar
raw_vllm  
raojy committed
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79 

      result = deviceProp.warpSize;
      is_cached = true;
    }

    return result;
  }

  static __device__ constexpr int get_warp_size() {
  #ifdef __GFX9__
    return 64;
  #else
    return 32;
  #endif
  }
};

  #define WARP_SIZE Utils::get_warp_size()
#else
  #define WARP_SIZE 32
#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