Skip to content

ggml-cpu: support K tails in power10 Q8/Q4 MMA matmul #24753

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
ggerganov merged 3 commits into from
Jun 19, 2026
Merged

ggml-cpu: support K tails in power10 Q8/Q4 MMA matmul #24753

Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e38fe54
ggml-cpu: support K tails in Power10 MMA Q8/Q4 matmul
shalinib-ibm Jun 18, 2026
02e86cf
Merge pull request #27 from shalinib-ibm/shalinib-ibm-patch-1
shalinib-ibm Jun 18, 2026
2e8b42c
Apply suggestion from @taronaeo
shalinib-ibm Jun 18, 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
11 changes: 6 additions & 5 deletions ggml/src/ggml-cpu/llamafile/sgemm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2345,7 +2345,7 @@ class tinyBLAS_Q0_PPC {
else if (n_aligned % 16 == 0) nc = 16;
else nc = 8;
}
bool can_use_tiled = n_aligned > 0 && (m % mc == 0) && (k % kc == 0);
bool can_use_tiled = n_aligned > 0 && (m % mc == 0);
if (can_use_tiled) {
matmul_tiled(m, n_aligned, mc, nc, kc);
if (n > n_aligned) {
Expand Down Expand Up @@ -3063,13 +3063,14 @@ class tinyBLAS_Q0_PPC {
int64_t ii = (job / xtiles) * mc;
int64_t jj = (job % xtiles) * nc;
for (int64_t kk = 0; kk < k; kk += kc) {
int64_t k_cur = MIN(kc, k - kk);
if constexpr(is_Ablock_q4) {
packNormal_q4_fp16(A + ii * lda + kk, lda, mc, kc, (uint8_t *)A_pack);
packNormal_q4_fp16(A + ii * lda + kk, lda, mc, k_cur, (uint8_t *)A_pack);
} else {
packNormal_q8_fp16(A + ii * lda + kk, lda, mc, kc, (uint8_t *)A_pack);
packNormal_q8_fp16(A + ii * lda + kk, lda, mc, k_cur, (uint8_t *)A_pack);
}
packNormal_q8_fp16(B + jj * ldb + kk, ldb, nc, kc, (uint8_t *)B_pack);
KERNEL_Q0(ii, jj, mc, nc, kc, kk, A_pack, B_pack);
packNormal_q8_fp16(B + jj * ldb + kk, ldb, nc, k-cur, (uint8_t *)B_pack);
Comment thread
shalinib-ibm marked this conversation as resolved.
Outdated
KERNEL_Q0(ii, jj, mc, nc, k_cur, kk, A_pack, B_pack);
}
}
}
Expand Down
Loading