Skip to content

ggml-opencl : store GPU buffer in ggml_tensor::extra #2994

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
Sep 4, 2023
Merged

ggml-opencl : store GPU buffer in ggml_tensor::extra #2994

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
12 changes: 6 additions & 6 deletions ggml-opencl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1334,7 +1334,7 @@ void ggml_cl_free_data(const struct ggml_tensor* tensor) {
return;
}

cl_mem mem = (cl_mem)tensor->data;
cl_mem mem = (cl_mem)tensor->extra;
clReleaseMemObject(mem);
}

Expand Down Expand Up @@ -1393,7 +1393,7 @@ static void ggml_cl_mul_f32(const ggml_tensor * src0, const ggml_tensor * src1,
size_t d_size;

cl_mem d_X = ggml_cl_pool_malloc(ne0 * sizeof(float), &x_size); // src0
cl_mem d_Y = (cl_mem) src1->data; // src1 is already on device, broadcasted.
cl_mem d_Y = (cl_mem) src1->extra; // src1 is already on device, broadcasted.
cl_mem d_D = ggml_cl_pool_malloc(ne0 * sizeof(float), &d_size); // dst


Expand Down Expand Up @@ -1491,7 +1491,7 @@ static void ggml_cl_mul_mat_f32(const ggml_tensor * src0, const ggml_tensor * sr
size_t d_size;
cl_mem d_X;
if (src0->backend == GGML_BACKEND_GPU) { // NOLINT
d_X = (cl_mem) src0->data;
d_X = (cl_mem) src0->extra;
} else {
d_X = ggml_cl_pool_malloc(sizeof(float) * x_ne, &x_size);
}
Expand Down Expand Up @@ -1567,7 +1567,7 @@ static void ggml_cl_mul_mat_f16(const ggml_tensor * src0, const ggml_tensor * sr
size_t d_size;
cl_mem d_X;
if (src0->backend == GGML_BACKEND_GPU) { // NOLINT
d_X = (cl_mem) src0->data;
d_X = (cl_mem) src0->extra;
} else {
d_X = ggml_cl_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &x_size);
}
Expand Down Expand Up @@ -1697,7 +1697,7 @@ static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
events.emplace_back();
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_Q, 0, src0, i03, i02, events.data() + ev_idx++));
} else if (src0->backend == GGML_BACKEND_GPU) {
d_Q = (cl_mem) src0->data;
d_Q = (cl_mem) src0->extra;
} else {
GGML_ASSERT(false);
}
Expand Down Expand Up @@ -1860,6 +1860,6 @@ void ggml_cl_transform_tensor(void * data, ggml_tensor * tensor) {

CL_CHECK(clFinish(queue));

tensor->data = dst;
tensor->extra = dst;
GGML_ASSERT(tensor->backend == GGML_BACKEND_GPU);
}