Skip to content

HIP: use upstream rows_per_warp() on RDNA3.5 #106

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
liangliangchang wants to merge 1 commit into
base: gfx11
Choose a base branch
from
+0 −14
Open

HIP: use upstream rows_per_warp() on RDNA3.5 #106

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 0 additions & 14 deletions ggml/src/ggml-cuda/mmq.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -183,21 +183,7 @@ struct ggml_cuda_mmq_config {

constexpr __device__ int rows_per_warp() const {
#if defined(AMD_MFMA_AVAILABLE) || defined(AMD_WMMA_AVAILABLE)
#if defined(RDNA3_5)
if ((type == GGML_TYPE_Q4_K || type == GGML_TYPE_Q4_0 || type == GGML_TYPE_Q8_0 ||
type == GGML_TYPE_Q5_K) && J == 128) {
return 16;
}
// Block quants only have a batched WMMA vec_dot at J=128; at the other two widths
// that would take ntx=2 (J=64 and J=96, the only J>=64 multiples of 32) they fall
// back to the generic schedule, which is several times slower there. Keep ntx=1.
if ((type == GGML_TYPE_Q4_0 || type == GGML_TYPE_Q8_0) && (J == 64 || J == 96)) {
return 16;
}
return J >= 64 && J % 32 == 0 ? 32 : 16;
#else
return 16;
#endif // defined(RDNA3_5)
#else
return J >= 48 && J % 16 == 0 ? 32 : 16;
#endif // defined(AMD_MFMA_AVAILABLE) || defined(AMD_WMMA_AVAILABLE)
Expand Down
Loading