Skip to content

ROCm: fix 8-bit affine QMV miscompile from uint4 weight load #7

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

Merged
Geramy merged 1 commit into from
May 20, 2026
Merged

ROCm: fix 8-bit affine QMV miscompile from uint4 weight load #7

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
16 changes: 11 additions & 5 deletions mlx/backend/rocm/quantized/qdequant.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,11 +94,17 @@ __device__ __forceinline__ void load_weight_vec(
out[0] = v.x;
out[1] = v.y;
} else if constexpr (PPT == 4) {
uint4 v = *reinterpret_cast<const uint4*>(ptr);
out[0] = v.x;
out[1] = v.y;
out[2] = v.z;
out[3] = v.w;
// Two uint2 loads instead of one uint4. The single-uint4 load
// (global_load_b128) miscomputes in the 8-bit affine QMV/gather paths
// (root cause: HIP_vector_type<unsigned int,4> codegen on RDNA 3.5 with
// hipcc 7.13 / LLVM 23). Two paired global_load_b64 ops yield the same
// throughput on RDNA 3.5 without the miscompile.
uint2 v0 = *reinterpret_cast<const uint2*>(ptr);
uint2 v1 = *reinterpret_cast<const uint2*>(ptr + 2);
out[0] = v0.x;
out[1] = v0.y;
out[2] = v1.x;
out[3] = v1.y;
} else {
#pragma unroll
for (int p = 0; p < PPT; p++) {
Expand Down
Loading