vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'occam/opencl-dev' into concedo_experimental

Browse source
This commit is contained in:
Concedo 2023-05-31 17:53:14 +08:00
commit 50c85bea4c
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml-opencl.cpp
View file

@ -701,7 +701,7 @@ static void ggml_cl_mul_f32(const ggml_tensor * src0, const ggml_tensor * src1,
size_t d_size; size_t d_size;
cl_mem d_X = ggml_cl_pool_malloc(ne0 * sizeof(float), &x_size, CL_MEM_READ_ONLY); // src0 cl_mem d_X = ggml_cl_pool_malloc(ne0 * sizeof(float), &x_size, CL_MEM_READ_ONLY); // src0
cl_mem d_Y = *(cl_mem*) src1->data; // src1 is already on device, broadcasted. cl_mem d_Y = (cl_mem) src1->data; // src1 is already on device, broadcasted.
cl_mem d_D = ggml_cl_pool_malloc(ne0 * sizeof(float), &d_size, CL_MEM_WRITE_ONLY); // dst cl_mem d_D = ggml_cl_pool_malloc(ne0 * sizeof(float), &d_size, CL_MEM_WRITE_ONLY); // dst
for (int64_t i03 = 0; i03 < ne03; i03++) { for (int64_t i03 = 0; i03 < ne03; i03++) {
@ -1174,8 +1174,8 @@ void ggml_cl_load_data(const char * fname, struct ggml_tensor * tensor, const si
const size_t size = ggml_nbytes(tensor); const size_t size = ggml_nbytes(tensor);
cl_mem* dst = (cl_mem*) malloc(sizeof(cl_mem)); cl_mem dst;
CL_CHECK((*dst = clCreateBuffer(context, CL_MEM_READ_ONLY, size, nullptr, &err), err)); CL_CHECK((dst = clCreateBuffer(context, CL_MEM_READ_ONLY, size, nullptr, &err), err));
void * buf_host = malloc(size); void * buf_host = malloc(size);
#ifdef _WIN32 #ifdef _WIN32
@ -1191,7 +1191,7 @@ void ggml_cl_load_data(const char * fname, struct ggml_tensor * tensor, const si
exit(1); exit(1);
} }
clEnqueueWriteBuffer(queue, *dst, CL_TRUE, 0, size, buf_host, 0, nullptr, nullptr); clEnqueueWriteBuffer(queue, dst, CL_TRUE, 0, size, buf_host, 0, nullptr, nullptr);
tensor->data = dst; tensor->data = dst;
free(buf_host); free(buf_host);