CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-19 11:57:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: use mma PTX instructions for FlashAttention (#11583)

Browse Source 
* CUDA: use mma PTX instructions for FlashAttention

* __shfl_sync workaround for movmatrix

* add __shfl_sync to HIP

Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
Johannes Gäßler
2025-02-02 19:31:09 +01:00
committed by GitHub
parent 84ec8a58f7
commit 864a0b67a6
29 changed files with 2058 additions and 998 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -25,6 +25,7 @@
#define CU_MEM_LOCATION_TYPE_DEVICE hipMemLocationTypeDevice
#define CU_MEM_ACCESS_FLAGS_PROT_READWRITE hipMemAccessFlagsProtReadWrite
#define CU_CHECK(fn) {hipError_t err = fn; if(err != hipSuccess) { GGML_ABORT("HipVMM Failure: %s\n", hipGetErrorString(err)); }}
#define __shfl_sync(mask, var, laneMask, width) __shfl(var, laneMask, width)
#define __shfl_xor_sync(mask, var, laneMask, width) __shfl_xor(var, laneMask, width)
#define cublasComputeType_t hipblasDatatype_t //deprecated, new hipblasComputeType_t not in 5.6
#define cublasCreate hipblasCreate