ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

cuda : fix HIP and MUSA BF16 (#0)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-04-07 13:18:07 +03:00
parent a4e46e28f9
commit 1a1ab7e7a4
3 changed files with 3 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -20,6 +20,7 @@
#define CUBLAS_STATUS_SUCCESS HIPBLAS_STATUS_SUCCESS
#define CUBLAS_TF32_TENSOR_OP_MATH 0
#define CUDA_R_16F HIPBLAS_R_16F
#define CUDA_R_16BF HIPBLAS_R_16B
#define CUDA_R_32F HIPBLAS_R_32F
#define CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED hipDeviceAttributeVirtualMemoryManagementSupported
#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_TF32_TENSOR_OP_MATH MUBLAS_MATH_MODE_DEFAULT
#define CUDA_R_16F MUSA_R_16F
#define CUDA_R_16BF MUSA_R_16BF
#define CUDA_R_32F MUSA_R_32F
#define cublasComputeType_t cudaDataType_t
#define cublasCreate mublasCreate