Skip to content

cpu: skip NOPs to avoid barriers #17133

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
Nov 10, 2025
+21 −60
Merged

cpu: skip NOPs to avoid barriers #17133

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
011b2f9
cpu: skip NOPs to avoid barriers
max-krasnyansky Nov 10, 2025
3b5b5d3
cpu: use ggml_op_is_empty
max-krasnyansky Nov 10, 2025
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
37 changes: 21 additions & 16 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 @@ -1807,22 +1807,6 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
{
ggml_compute_forward_cont(params, tensor);
} break;
case GGML_OP_RESHAPE:
{
ggml_compute_forward_reshape(params, tensor);
} break;
case GGML_OP_VIEW:
{
ggml_compute_forward_view(params, tensor);
} break;
case GGML_OP_PERMUTE:
{
ggml_compute_forward_permute(params, tensor);
} break;
case GGML_OP_TRANSPOSE:
{
ggml_compute_forward_transpose(params, tensor);
} break;
case GGML_OP_GET_ROWS:
{
ggml_compute_forward_get_rows(params, tensor);
Expand Down Expand Up @@ -2042,6 +2026,22 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
{
// nop
} break;
case GGML_OP_RESHAPE:
{
// nop
} break;
case GGML_OP_PERMUTE:
{
// nop
} break;
case GGML_OP_VIEW:
{
// nop
} break;
case GGML_OP_TRANSPOSE:
{
// nop
} break;
case GGML_OP_COUNT:
{
GGML_ABORT("fatal error");
Expand Down Expand Up @@ -2884,6 +2884,11 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
for (int node_n = 0; node_n < cgraph->n_nodes && atomic_load_explicit(&tp->abort, memory_order_relaxed) != node_n; node_n++) {
struct ggml_tensor * node = cgraph->nodes[node_n];

if (ggml_op_is_empty(node->op)) {
// skip NOPs
continue;
}

ggml_compute_forward(&params, node);

if (state->ith == 0 && cplan->abort_callback &&
Expand Down
40 changes: 0 additions & 40 deletions ggml/src/ggml-cpu/ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4455,46 +4455,6 @@ void ggml_compute_forward_cont(
ggml_compute_forward_dup(params, dst);
}

// ggml_compute_forward_reshape

void ggml_compute_forward_reshape(
const ggml_compute_params * params,
ggml_tensor * dst) {
// NOP
GGML_UNUSED(params);
GGML_UNUSED(dst);
}

// ggml_compute_forward_view

void ggml_compute_forward_view(
const ggml_compute_params * params,
ggml_tensor * dst) {
// NOP
GGML_UNUSED(params);
GGML_UNUSED(dst);
}

// ggml_compute_forward_permute

void ggml_compute_forward_permute(
const ggml_compute_params * params,
ggml_tensor * dst) {
// NOP
GGML_UNUSED(params);
GGML_UNUSED(dst);
}

// ggml_compute_forward_transpose

void ggml_compute_forward_transpose(
const ggml_compute_params * params,
ggml_tensor * dst) {
// NOP
GGML_UNUSED(params);
GGML_UNUSED(dst);
}

// ggml_compute_forward_get_rows

static void ggml_compute_forward_get_rows_q(
Expand Down
4 changes: 0 additions & 4 deletions ggml/src/ggml-cpu/ops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,10 +51,6 @@ void ggml_compute_forward_scale(const struct ggml_compute_params * params, struc
void ggml_compute_forward_set(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_cpy(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_cont(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_reshape(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_view(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_permute(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_transpose(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_get_rows(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_get_rows_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_set_rows(const struct ggml_compute_params * params, struct ggml_tensor * dst);
Expand Down
Loading