Skip to content

Fix: wait on the PDL dependency in the PTQ1_0 mat-vec kernel (garbage output on Hopper/Blackwell) #1

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
LamplighterPaul wants to merge 1 commit into
base: pr-ptq1-mmv
Choose a base branch
from
Open

Fix: wait on the PDL dependency in the PTQ1_0 mat-vec kernel (garbage output on Hopper/Blackwell) #1

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
4 changes: 4 additions & 0 deletions ggml/src/ggml-cuda/mmvq-ptq1_0.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -298,6 +298,10 @@ static __global__ void mul_mat_vec_ptq1_0_pt(
const void * GGML_CUDA_RESTRICT vx = vx_ptr;
const void * GGML_CUDA_RESTRICT vy = vy_ptr;
float * GGML_CUDA_RESTRICT dst = dst_ptr;
// launched through ggml_cuda_kernel_launch, which opts into PDL on Hopper and newer: wait for the kernel
// that wrote vy (the q8_1 activation quantization) before reading it
ggml_cuda_pdl_sync();

extern __shared__ float partials_dyn[];
float * partials = partials_dyn; // [ncols][rows_per_cta][bpr]
[[maybe_unused]] float * partials_gate = partials_dyn + ncols*rows_per_cta*(ncols_x / QK_PTQ1_0);
Expand Down