CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-16 11:27:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: replace GGML_CUDA_F16 with CUDA arch checks (#15433)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-08-20 16:58:49 +02:00
committed by GitHub
parent fec9519802
commit 7a6e91ad26
12 changed files with 32 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
ggml/src/ggml-cuda/common.cuh
View File

@@ -206,14 +206,6 @@ static const char * cu_get_error_str(CUresult err) {
#define GGML_CUDA_ASSUME(x)
#endif // CUDART_VERSION >= 11010
#ifdef GGML_CUDA_F16
typedef half dfloat; // dequantize float
typedef half2 dfloat2;
#else
typedef float dfloat; // dequantize float
typedef float2 dfloat2;
#endif // GGML_CUDA_F16
#if (!defined(GGML_USE_HIP) && !defined(GGML_CUDA_NO_VMM)) || (defined(GGML_USE_HIP) && !defined(GGML_HIP_NO_VMM))
#define GGML_USE_VMM
#endif // (!defined(GGML_USE_HIP) && !defined(GGML_CUDA_NO_VMM)) || (defined(GGML_USE_HIP) && !defined(GGML_HIP_NO_VMM))
@@ -559,7 +551,7 @@ static __device__ __forceinline__ float ggml_cuda_e8m0_to_fp32(uint8_t x) {
#endif // CUDART_VERSION >= 12050
}
typedef void (*dequantize_kernel_t)(const void * vx, const int64_t ib, const int iqs, dfloat2 & v);
typedef void (*dequantize_kernel_t)(const void * vx, const int64_t ib, const int iqs, float2 & v);
static __device__ __forceinline__ float get_alibi_slope(
const float max_bias, const uint32_t h, const uint32_t n_head_log2, const float m0, const float m1