Skip to content

ggml: Re-enable CUDA graphs in presence of CONT and DUP nodes #12970

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 1 commit into from
Apr 17, 2025
Merged

ggml: Re-enable CUDA graphs in presence of CONT and DUP nodes #12970

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
9 changes: 5 additions & 4 deletions ggml/src/ggml-cuda/cpy.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -551,7 +551,7 @@ static void ggml_cpy_f16_f16_cuda(
(cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, cdst_indirect, graph_cpynode_index++);
}

void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, ggml_tensor * src1) {
void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, ggml_tensor * src1, bool disable_indirection_for_this_node) {
const int64_t ne = ggml_nelements(src0);
GGML_ASSERT(ne == ggml_nelements(src1));

Expand Down Expand Up @@ -588,7 +588,7 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg
char ** dest_ptrs_d = nullptr;
int graph_cpynode_index = -1;
#if defined(GGML_CUDA_USE_GRAPHS) || defined(GGML_HIP_GRAPHS)
if(ctx.cuda_graph->use_cpy_indirection) {
if(ctx.cuda_graph->use_cpy_indirection && !disable_indirection_for_this_node) {
dest_ptrs_d = ctx.cuda_graph->dest_ptrs_d;
graph_cpynode_index = ctx.cuda_graph->graph_cpynode_index;
}
Expand Down Expand Up @@ -636,7 +636,7 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg
ggml_type_name(src0->type), ggml_type_name(src1->type));
}
#if defined(GGML_CUDA_USE_GRAPHS) || defined(GGML_HIP_GRAPHS)
if(ctx.cuda_graph->use_cpy_indirection) {
if(ctx.cuda_graph->use_cpy_indirection && !disable_indirection_for_this_node) {
ctx.cuda_graph->graph_cpynode_index = graph_cpynode_index;
}
#endif
Expand All @@ -645,7 +645,8 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg

void ggml_cuda_dup(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
const ggml_tensor * src0 = dst->src[0];
ggml_cuda_cpy(ctx, src0, dst);
bool disable_indirection = true;
ggml_cuda_cpy(ctx, src0, dst, disable_indirection);
}

void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1) {
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-cuda/cpy.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@

#define CUDA_CPY_BLOCK_SIZE 64

void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, ggml_tensor * src1);
void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, ggml_tensor * src1, bool disable_indirection = false);

void ggml_cuda_dup(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-cuda/ggml-cuda.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2489,7 +2489,7 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
#endif
}

if (node->op == GGML_OP_MUL_MAT_ID || node->op == GGML_OP_CONT || node->op == GGML_OP_DUP) {
if (node->op == GGML_OP_MUL_MAT_ID) {
use_cuda_graph = false; // This node type is not supported by CUDA graph capture
#ifndef NDEBUG
GGML_LOG_DEBUG("%s: disabling CUDA graphs due to unsupported node type\n", __func__);
Expand Down
Loading