cuda : performance optimizations (#1530)
* xor hack * block y dim * loop unrolling * Fixed cmake LLAMA_CUDA_BY option * Removed hipblas compatibility code * Define GGML_CUDA_DMMV_BLOCK_Y if not defined * Fewer iters, more ops per iter * Renamed DMMV X/Y compilation options
J
Johannes Gäßler committed
1fcdcc28b119a6608774d52de905931bd5f8a43d
Parent: ac7876a
Committed by GitHub <noreply@github.com>
on 5/25/2023, 9:07:29 PM