tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

test

Browse Source
This commit is contained in:
Georgi Gerganov
2025-04-07 13:18:07 +03:00
parent 6232ceec72
commit 5ef588ba58
3 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cuda/convert.cu
View File

@ -580,7 +580,11 @@ static __global__ void convert_unary(const void * __restrict__ vx, dst_t * __res
const src_t * x = (const src_t *) vx; const src_t * x = (const src_t *) vx;
if constexpr (std::is_same_v<src_t, nv_bfloat16>) { if constexpr (std::is_same_v<src_t, nv_bfloat16>) {
#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)
y[i] = (float)x[i];
#else
y[i] = __bfloat162float(x[i]); y[i] = __bfloat162float(x[i]);
#endif
} else if constexpr (std::is_same_v<dst_t, nv_bfloat16> && std::is_same_v<src_t, half>) { } else if constexpr (std::is_same_v<dst_t, nv_bfloat16> && std::is_same_v<src_t, half>) {
y[i] = (float)x[i]; y[i] = (float)x[i];
} else { } else {

1
ggml/src/ggml-cuda/vendors/hip.h vendored
View File

@ -20,6 +20,7 @@
#define CUBLAS_STATUS_SUCCESS HIPBLAS_STATUS_SUCCESS #define CUBLAS_STATUS_SUCCESS HIPBLAS_STATUS_SUCCESS
#define CUBLAS_TF32_TENSOR_OP_MATH 0 #define CUBLAS_TF32_TENSOR_OP_MATH 0
#define CUDA_R_16F HIPBLAS_R_16F #define CUDA_R_16F HIPBLAS_R_16F
#define CUDA_R_16BF HIPBLAS_R_16BF
#define CUDA_R_32F HIPBLAS_R_32F #define CUDA_R_32F HIPBLAS_R_32F
#define CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED hipDeviceAttributeVirtualMemoryManagementSupported #define CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED hipDeviceAttributeVirtualMemoryManagementSupported
#define CU_MEM_ALLOC_GRANULARITY_RECOMMENDED hipMemAllocationGranularityRecommended #define CU_MEM_ALLOC_GRANULARITY_RECOMMENDED hipMemAllocationGranularityRecommended

1
ggml/src/ggml-cuda/vendors/musa.h vendored
View File

@ -15,6 +15,7 @@
#define CUBLAS_STATUS_SUCCESS MUBLAS_STATUS_SUCCESS #define CUBLAS_STATUS_SUCCESS MUBLAS_STATUS_SUCCESS
#define CUBLAS_TF32_TENSOR_OP_MATH MUBLAS_MATH_MODE_DEFAULT #define CUBLAS_TF32_TENSOR_OP_MATH MUBLAS_MATH_MODE_DEFAULT
#define CUDA_R_16F MUSA_R_16F #define CUDA_R_16F MUSA_R_16F
#define CUDA_R_16BF MUSA_R_16BF
#define CUDA_R_32F MUSA_R_32F #define CUDA_R_32F MUSA_R_32F
#define cublasComputeType_t cudaDataType_t #define cublasComputeType_t cudaDataType_t
#define cublasCreate mublasCreate #define cublasCreate mublasCreate