Skip to content

ggml-cuda: fix a mountain of VGPR spills on Q2_K #29910

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

Closed
thelittlefireman wants to merge 4 commits into from
Closed

ggml-cuda: fix a mountain of VGPR spills on Q2_K #29910

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a13695b
ggml-cuda: HIP: fix Q2_K spill on CDNA, RDNA3+
thelittlefireman Oct 3, 2026
edfa142
ggml-cuda: fix Q2_K spill on AMD: GCN, RDNA1, RDNA2 and NVIDIA Pascal…
thelittlefireman Oct 3, 2026
d00c584
ggml-cuda: Optimize and reduce spill on Q2_K by remove temporary y_df…
thelittlefireman Oct 3, 2026
f2179ef
ggml-cuda: rollback unroll
thelittlefireman Oct 12, 2026
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
13 changes: 3 additions & 10 deletions ggml/src/ggml-cuda/mmq-vec-dot.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -619,14 +619,6 @@ template <ggml_type type, int J, bool fallback> static __device__ __forceinline_
const int * y_qs = (const int *) y + 4;
const half2 * y_ds = (const half2 *) y;

float2 y_df[J/nwarps];
#pragma unroll
for (int j0 = 0; j0 < J; j0 += nwarps) {
const int j = j0 + threadIdx.y;

y_df[j0/nwarps] = __half22float2(y_ds[j*MMQ_TILE_Y_K]);
}

#pragma unroll
for (int k01 = 0; k01 < MMQ_TILE_NE_K/2; k01 += QR2_K*VDR_Q2_K_Q8_1_MMQ) {
const int k0 = k00 + k01;
Expand All @@ -642,7 +634,7 @@ template <ggml_type type, int J, bool fallback> static __device__ __forceinline_
constexpr int ns = 2;
sum[j0/nwarps*I/warp_size + i0/warp_size] += vec_dot_q2_K_q8_1_impl_mmq<ns>(
&x_qs[i*(2*MMQ_TILE_NE_K + 1) + k0], &y_qs[j*MMQ_TILE_Y_K + k01],
&x_dm[i*(MMQ_TILE_NE_K + 1) + k0/4], k01 < MMQ_TILE_NE_K/2 ? y_df[j0/nwarps].x : y_df[j0/nwarps].y,
&x_dm[i*(MMQ_TILE_NE_K + 1) + k0/4], __half22float2(y_ds[j*MMQ_TILE_Y_K]).x,
&y_ds[j*MMQ_TILE_Y_K + (1 + k01/QI8_1)]);
}
}
Expand All @@ -665,7 +657,7 @@ template <ggml_type type, int J, bool fallback> static __device__ __forceinline_
constexpr int ns = 1;
sum[j0/nwarps*I/warp_size + i0/warp_size] += vec_dot_q2_K_q8_1_impl_mmq<ns>(
&x_qs[i*(2*MMQ_TILE_NE_K + 1) + k0], &y_qs[j*MMQ_TILE_Y_K + k01],
&x_dm[i*(MMQ_TILE_NE_K + 1) + k0/4], k01 < MMQ_TILE_NE_K/2 ? y_df[j0/nwarps].x : y_df[j0/nwarps].y,
&x_dm[i*(MMQ_TILE_NE_K + 1) + k0/4], __half22float2(y_ds[j*MMQ_TILE_Y_K]).y,
&y_ds[j*MMQ_TILE_Y_K + (1 + k01/QI8_1)]);
}
}
Expand Down Expand Up @@ -693,6 +685,7 @@ template <ggml_type type, int J, bool fallback> static __device__ __forceinline_

const int i0 = (threadIdx.y / ntx) * rows_per_warp;

#pragma unroll
for (int k01 = 0; k01 < MMQ_TILE_NE_K; k01 += 4) {
const int k0 = k00 + k01;

Expand Down