tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

HIP: implement FlashAttention via rocWMMA for CDNA and RDNA3+ (#12032)

Browse Source 
Adds GGML_HIP_ROCWMMA_FATTN and rocwmma header check
Adds rocWMMA support to fattn-wmma-f16

---

Signed-off-by: Carl Klemm <carl@uvos.xyz>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Ben Jackson <ben@ben.com>
This commit is contained in:
David Huang
2025-03-04 05:10:54 +08:00
committed by GitHub
parent dfd6b2c0be
commit becade5de7
6 changed files with 145 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/CMakeLists.txt
View File

@ -162,6 +162,7 @@ set_property(CACHE GGML_CUDA_COMPRESSION_MODE PROPERTY STRINGS "none;speed;balan
option(GGML_HIP "ggml: use HIP" OFF)
option(GGML_HIP_GRAPHS "ggml: use HIP graph, experimental, slow" OFF)
option(GGML_HIP_NO_VMM "ggml: do not try to use HIP VMM" ON)
option(GGML_HIP_ROCWMMA_FATTN "ggml: enable rocWMMA for FlashAttention" OFF)
option(GGML_HIP_UMA "ggml: use HIP unified memory architecture" OFF)
option(GGML_VULKAN "ggml: use Vulkan" OFF)
option(GGML_VULKAN_CHECK_RESULTS "ggml: run Vulkan op checks" OFF)