Skip to content
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

Jump to bottom

ggml : fix row condition for i8mm kernels #10561

Merged
merged 1 commit into from
Nov 28, 2024
Merged

ggml : fix row condition for i8mm kernels #10561

Show file tree
Hide file tree
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
6 changes: 4 additions & 2 deletions ggml/src/ggml-cpu/ggml-cpu-quants.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1813,11 +1813,13 @@ void ggml_vec_dot_q4_0_q8_0(int n, float * restrict s, size_t bs, const void * r
sumv0 = vmlaq_f32(sumv0,(vcvtq_f32_s32(vmmlaq_s32((vmmlaq_s32((vmmlaq_s32((vmmlaq_s32(vdupq_n_s32(0), l0, r0)),
l1, r1)), l2, r2)), l3, r3))), scale);
}
float32x4_t sumv1 = vextq_f32(sumv0, sumv0, 2);

float32x4_t sumv1 = vextq_f32 (sumv0, sumv0, 2);
float32x4_t sumv2 = vzip1q_f32(sumv0, sumv1);

vst1_f32(s, vget_low_f32(sumv2));
vst1_f32(s, vget_low_f32 (sumv2));
vst1_f32(s + bs, vget_high_f32(sumv2));

return;
}
#endif
Expand Down
17 changes: 9 additions & 8 deletions ggml/src/ggml-cpu/ggml-cpu.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7560,14 +7560,6 @@ UseGgmlGemm2:;
// This is the size of the rest of the dimensions of the result
const int64_t nr1 = ne1 * ne2 * ne3;

// dot kernels can handle 1 row and col at a time, but mmla kernels can process 2 rows and cols
int64_t num_rows_per_vec_dot = vec_dot_num_rows;
// TODO: currently the mmla kernels support only even numbered rows/cols.
// this check can be removed once they are extended to support odd numbered rows/cols too
if ((nr0 % 2 != 0) || (ne11 % 2 != 0)) {
num_rows_per_vec_dot = 1;
}

// Now select a reasonable chunk size.
int chunk_size = 16;

Expand Down Expand Up @@ -7630,6 +7622,15 @@ UseGgmlGemm2:;
const int64_t ir1_start = dr1 * ith1;
const int64_t ir1_end = MIN(ir1_start + dr1, nr1);

// dot kernels can handle 1 row and col at a time, but mmla kernels can process 2 rows and cols
int64_t num_rows_per_vec_dot = vec_dot_num_rows;

// TODO: currently the mmla kernels support only even numbered rows/cols.
// this check can be removed once they are extended to support odd numbered rows/cols too
if ((nr0 % 2 != 0) || (ne11 % 2 != 0) || ((ir0_end - ir0_start) % 2 != 0) || ((ir1_end - ir1_start) % 2 != 0)) {
num_rows_per_vec_dot = 1;
}

ggml_compute_forward_mul_mat_one_chunk(params, dst, type, num_rows_per_vec_dot, ir0_start, ir0_end, ir1_start, ir1_end);

if (nth >= nchunk0 * nchunk1) {
Expand Down
Loading