]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
HIP: implement FlashAttention via rocWMMA for CDNA and RDNA3+ (llama/12032)
authorDavid Huang <redacted>
Mon, 3 Mar 2025 21:10:54 +0000 (05:10 +0800)
committerGeorgi Gerganov <redacted>
Tue, 4 Mar 2025 19:24:42 +0000 (21:24 +0200)
commita0a3dd6b989ca1a3b839760241da41654a2501c6
tree728d718774dda2e82c70e89c6a329fab2c97ed74
parent1aeda84192d694672b2580913b6ec7b744a38844
HIP: implement FlashAttention via rocWMMA for CDNA and RDNA3+ (llama/12032)

Adds GGML_HIP_ROCWMMA_FATTN and rocwmma header check
Adds rocWMMA support to fattn-wmma-f16
CMakeLists.txt
src/ggml-cuda/common.cuh
src/ggml-cuda/fattn-common.cuh
src/ggml-cuda/fattn-wmma-f16.cu
src/ggml-cuda/fattn.cu
src/ggml-hip/CMakeLists.txt