Skip to content

metal: somewhat faster f16 x f32 matrix multiply kernel #2951

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
merged 2 commits into from
Sep 1, 2023
Merged

metal: somewhat faster f16 x f32 matrix multiply kernel #2951

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
af226bd
Somewhat faster f16 x f32 matrix multiply kernel
Sep 1, 2023
cad50d1
Better use 32 thread groups for f16 x f32
Sep 1, 2023
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
2 changes: 1 addition & 1 deletion ggml-metal.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -840,7 +840,7 @@ void ggml_metal_graph_compute(
switch (src0t) {
case GGML_TYPE_F16:
{
nth0 = 64;
nth0 = 32;
nth1 = 1;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32];
} break;
Expand Down
38 changes: 28 additions & 10 deletions ggml-metal.metal
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -528,24 +528,42 @@ kernel void kernel_mul_mat_f16_f32(
device const half * x = (device const half *) (src0 + r0*nb01 + im/(ne12/ne02)*nb02);
device const float * y = (device const float *) (src1 + r1*nb11 + im*nb12);

sum[tpitg.x] = 0.0f;
uint ith = tpitg.x;
uint nth = tptg.x;

for (int i = tpitg.x; i < ne00; i += tptg.x) {
sum[tpitg.x] += (float) x[i] * (float) y[i];
sum[ith] = 0.0f;

for (int i = ith; i < ne00; i += nth) {
sum[ith] += (float) x[i] * (float) y[i];
}

// accumulate the sum from all threads in the threadgroup
threadgroup_barrier(mem_flags::mem_threadgroup);
for (uint i = tptg.x/2; i > 0; i /= 2) {
if (tpitg.x < i) {
sum[tpitg.x] += sum[tpitg.x + i];
}
threadgroup_barrier(mem_flags::mem_threadgroup);
if (ith%4 == 0) {
for (int i = 1; i < 4; ++i) sum[ith] += sum[ith + i];
}

if (tpitg.x == 0) {
threadgroup_barrier(mem_flags::mem_threadgroup);
if (ith%16 == 0) {
for (int i = 4; i < 16; i += 4) sum[ith] += sum[ith + i];
}
threadgroup_barrier(mem_flags::mem_threadgroup);
if (ith == 0) {
for (int i = 16; i < nth; i += 16) sum[0] += sum[i];
dst[im*ne1*ne0 + r1*ne0 + r0] = sum[0];
}

// Original implementation. Left behind commented out for now
//threadgroup_barrier(mem_flags::mem_threadgroup);
//for (uint i = tptg.x/2; i > 0; i /= 2) {
// if (tpitg.x < i) {
// sum[tpitg.x] += sum[tpitg.x + i];
// }
// threadgroup_barrier(mem_flags::mem_threadgroup);
//}
//
//if (tpitg.x == 0) {
// dst[im*ne1*ne0 + r1*ne0 + r0] = sum[0];
//}
}

kernel void kernel_alibi_f32(
Expand Down