Skip to content

metal : fix ACC op #19427

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 1 commit into from
Feb 14, 2026
Merged

metal : fix ACC op #19427

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
2 changes: 1 addition & 1 deletion ggml/src/ggml-metal/ggml-metal-device.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1058,8 +1058,8 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
case GGML_OP_MUL:
case GGML_OP_DIV:
case GGML_OP_ADD_ID:
return ggml_is_contiguous_rows(op->src[0]) && ggml_is_contiguous_rows(op->src[1]) && op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_ACC:
return ggml_is_contiguous_rows(op->src[0]) && ggml_is_contiguous_rows(op->src[1]) && op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_REPEAT:
case GGML_OP_SCALE:
case GGML_OP_FILL:
Expand Down
28 changes: 17 additions & 11 deletions ggml/src/ggml-metal/ggml-metal-ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -628,8 +628,8 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
GGML_ASSERT(op->src[1]->type == GGML_TYPE_F32);
GGML_ASSERT(op->type == GGML_TYPE_F32);

GGML_ASSERT(ggml_is_contiguous(op->src[0]));
GGML_ASSERT(ggml_is_contiguous(op->src[1]));
GGML_ASSERT(ggml_is_contiguous_rows(op->src[0]));
GGML_ASSERT(ggml_is_contiguous_rows(op->src[1]));

const size_t pnb1 = ((const int32_t *) op->op_params)[0];
const size_t pnb2 = ((const int32_t *) op->op_params)[1];
Expand Down Expand Up @@ -679,10 +679,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
}

ggml_metal_kargs_bin args = {
/*.ne00 =*/ ne00,
/*.ne01 =*/ ne01,
/*.ne02 =*/ ne02,
/*.ne03 =*/ ne03,
/*.ne00 =*/ ne10,
/*.ne01 =*/ ne11,
/*.ne02 =*/ ne12,
/*.ne03 =*/ ne13,
/*.nb00 =*/ nb00,
/*.nb01 =*/ pnb1,
/*.nb02 =*/ pnb2,
Expand All @@ -695,10 +695,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
/*.nb11 =*/ nb11,
/*.nb12 =*/ nb12,
/*.nb13 =*/ nb13,
/*.ne0 =*/ ne0,
/*.ne1 =*/ ne1,
/*.ne2 =*/ ne2,
/*.ne3 =*/ ne3,
/*.ne0 =*/ ne10,
/*.ne1 =*/ ne11,
/*.ne2 =*/ ne12,
/*.ne3 =*/ ne13,
/*.nb0 =*/ nb0,
/*.nb1 =*/ pnb1,
/*.nb2 =*/ pnb2,
Expand All @@ -715,7 +715,13 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
ggml_metal_encoder_set_buffer (enc, ggml_metal_get_buffer_id(op->src[1]), 2);
ggml_metal_encoder_set_buffer (enc, ggml_metal_get_buffer_id(op), 3);

const int nth = std::min(ggml_metal_pipeline_max_theads_per_threadgroup(pipeline), ne00);
const int nth_max = MIN(256, ggml_metal_pipeline_max_theads_per_threadgroup(pipeline));

int nth = 1;

while (2*nth < args.ne0 && nth < nth_max) {
nth *= 2;
}

ggml_metal_encoder_dispatch_threadgroups(enc, ne11, ne12, ne13, nth, 1, 1);

Expand Down
Loading