Skip to content

fix(rocm): avoid SIMT workgroup-count resonance #2012

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
sstamenk wants to merge 1 commit into
base: main
Choose a base branch
from
+8 −1
Open

fix(rocm): avoid SIMT workgroup-count resonance #2012

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
9 changes: 8 additions & 1 deletion csrc/gemm_4bit_simt.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -490,7 +490,14 @@ void launch_gemm_4bit_simt(
bnb_stream_t stream // CUDA/HIP stream
// clang-format on
) {
const int n_blocks = (N + WARPS_PER_BLOCK - 1) / WARPS_PER_BLOCK;
int n_blocks = (N + WARPS_PER_BLOCK - 1) / WARPS_PER_BLOCK;
#if BNB_HIP
// Large HIP SIMT dispatches can underutilize the GPU when the workgroup
// count is an exact multiple of 256. One additional bounds-checked block
// breaks the scheduler resonance without changing useful work.
if (n_blocks >= 512 && n_blocks % 256 == 0)
n_blocks++;
#endif

// M=1..8: M_BLOCK == M, so the inner m-loop fully unrolls at compile time.
// M>8: M_BLOCK=8, ceil(M/8) grid rows.
Expand Down