root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-29 04:44:34 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

vulkan : sync with "migrate to dynamic graphs"

Browse Source
This commit is contained in:
Jared Van Bortel 2023-11-23 12:20:07 -05:00
parent 23f6d51f68
commit a4bb9c5ced
1 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
ggml-vulkan.cpp
View File

@ -1350,6 +1350,15 @@ void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml_cgraph
struct ggml_tensor * dst = gf->nodes[i];
GGML_ASSERT(dst->data != nullptr);
switch (dst->op) {
case GGML_OP_NONE:
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
continue; // noop -> next node
}
const int32_t ne00 = src0 ? src0->ne[0] : 0;
const int32_t ne01 = src0 ? src0->ne[1] : 0;
const int32_t ne02 = src0 ? src0->ne[2] : 0;
@ -1393,13 +1402,6 @@ void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml_cgraph
const std::shared_ptr<kp::Tensor>& id_dst = dst ? ggml_vk_get_tensor(ctx, dst, &off_dst) : nullTensor;
switch (dst->op) {
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
{
// noop
} break;
case GGML_OP_ADD:
{
if (ggml_nelements(src1) == ne10 && ne00 % 4 == 0) {