happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

metal : fix ACC op (#19427)

This commit is contained in:
Georgi Gerganov 2026-02-14 09:54:03 +02:00 committed by GitHub
parent c7db95f106
commit 6e473fb384
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 18 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-metal/ggml-metal-device.m
View File

@ -1067,8 +1067,8 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
case GGML_OP_MUL: case GGML_OP_MUL:
case GGML_OP_DIV: case GGML_OP_DIV:
case GGML_OP_ADD_ID: 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: 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_REPEAT:
case GGML_OP_CONV_TRANSPOSE_1D: case GGML_OP_CONV_TRANSPOSE_1D:
return true; return true;

28
ggml/src/ggml-metal/ggml-metal-ops.cpp
View File

@ -620,8 +620,8 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
GGML_ASSERT(op->src[1]->type == GGML_TYPE_F32); GGML_ASSERT(op->src[1]->type == GGML_TYPE_F32);
GGML_ASSERT(op->type == GGML_TYPE_F32); GGML_ASSERT(op->type == GGML_TYPE_F32);
GGML_ASSERT(ggml_is_contiguous(op->src[0])); GGML_ASSERT(ggml_is_contiguous_rows(op->src[0]));
GGML_ASSERT(ggml_is_contiguous(op->src[1])); GGML_ASSERT(ggml_is_contiguous_rows(op->src[1]));
const size_t pnb1 = ((const int32_t *) op->op_params)[0]; const size_t pnb1 = ((const int32_t *) op->op_params)[0];
const size_t pnb2 = ((const int32_t *) op->op_params)[1]; const size_t pnb2 = ((const int32_t *) op->op_params)[1];
@ -671,10 +671,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
} }
ggml_metal_kargs_bin args = { ggml_metal_kargs_bin args = {
/*.ne00 =*/ ne00, /*.ne00 =*/ ne10,
/*.ne01 =*/ ne01, /*.ne01 =*/ ne11,
/*.ne02 =*/ ne02, /*.ne02 =*/ ne12,
/*.ne03 =*/ ne03, /*.ne03 =*/ ne13,
/*.nb00 =*/ nb00, /*.nb00 =*/ nb00,
/*.nb01 =*/ pnb1, /*.nb01 =*/ pnb1,
/*.nb02 =*/ pnb2, /*.nb02 =*/ pnb2,
@ -687,10 +687,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
/*.nb11 =*/ nb11, /*.nb11 =*/ nb11,
/*.nb12 =*/ nb12, /*.nb12 =*/ nb12,
/*.nb13 =*/ nb13, /*.nb13 =*/ nb13,
/*.ne0 =*/ ne0, /*.ne0 =*/ ne10,
/*.ne1 =*/ ne1, /*.ne1 =*/ ne11,
/*.ne2 =*/ ne2, /*.ne2 =*/ ne12,
/*.ne3 =*/ ne3, /*.ne3 =*/ ne13,
/*.nb0 =*/ nb0, /*.nb0 =*/ nb0,
/*.nb1 =*/ pnb1, /*.nb1 =*/ pnb1,
/*.nb2 =*/ pnb2, /*.nb2 =*/ pnb2,
@ -707,7 +707,13 @@ 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->src[1]), 2);
ggml_metal_encoder_set_buffer (enc, ggml_metal_get_buffer_id(op), 3); 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); ggml_metal_encoder_dispatch_threadgroups(enc, ne11, ne12, ne13, nth, 1, 1);