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

CUDA: Improve flash decoding kernel GPU occupancy for BS=1 case (#12183)

Browse source 
- Find out active blocks per SM using cudaOccupancyMaxActiveBlocksPerMultiprocessor API. Use this value to determine the optimal parallel_blocks value.
- Prefer vector flash attention kernels over MMA kernel for BS=1

Fixes Issue: #12182
---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
This commit is contained in:
Gaurav Garg 2025-03-20 01:22:06 +05:30 • committed by GitHub
parent a9b59288e2
commit 517b5ddbf0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 214 additions and 266 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -129,6 +129,7 @@
#define cudaGraph_t hipGraph_t
#define cudaStream_t hipStream_t
#define cudaSuccess hipSuccess
#define cudaOccupancyMaxActiveBlocksPerMultiprocessor hipOccupancyMaxActiveBlocksPerMultiprocessor
#define __trap() do { abort(); __builtin_unreachable(); } while(0)
#define CUBLAS_STATUS_SUCCESS HIPBLAS_STATUS_SUCCESS
#define CUBLAS_STATUS_NOT_INITIALIZED HIPBLAS_STATUS_NOT_INITIALIZED

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

@ -134,5 +134,6 @@
#define cudaStreamCaptureModeRelaxed musaStreamCaptureModeRelaxed
#define cudaStreamBeginCapture musaStreamBeginCapture
#define cudaStreamEndCapture musaStreamEndCapture
#define cudaOccupancyMaxActiveBlocksPerMultiprocessor musaOccupancyMaxActiveBlocksPerMultiprocessor
typedef mt_bfloat16 nv_bfloat16;