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

fix compile errors on cuda

Browse source
This commit is contained in:
Concedo 2023-05-21 21:13:17 +08:00
parent 33528f5b1d
commit 60e0c67874
2 changed files with 10 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

2
CMakeLists.txt
View file

@ -322,6 +322,6 @@ target_compile_features(${TARGET} PUBLIC cxx_std_11) # don't bump
set_target_properties(${TARGET} PROPERTIES PREFIX "")
set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME "koboldcpp")
set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_link_libraries(${TARGET} PUBLIC ggml ggml_v1 common2 gpttype_adapter ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PUBLIC ggml ggml_v1 ggml_v2 common2 gpttype_adapter ${CMAKE_THREAD_LIBS_INIT})
target_compile_features(${TARGET} PRIVATE cxx_std_11)

18
otherarch/ggml_v2.c
View file

@ -3893,7 +3893,7 @@ struct ggml_v2_context * ggml_v2_init(struct ggml_v2_init_params params) {
}
#if defined(GGML_USE_CUBLAS)
ggml_v2_init_cublas();
ggml_init_cublas();
#elif defined(GGML_USE_CLBLAST)
if(quants_unshuffled)
{
@ -9447,9 +9447,9 @@ static void ggml_v2_compute_forward_mul_mat_f32(
// compute by src0 rows
#if defined(GGML_USE_CUBLAS)
if (ggml_v2_cuda_can_mul_mat(src0, src1, dst)) {
if (ggml_cuda_can_mul_mat(src0, src1, dst)) {
if (params->ith == 0 && params->type == GGML_V2_TASK_COMPUTE) {
ggml_v2_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
ggml_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
}
return;
}
@ -9641,9 +9641,9 @@ static void ggml_v2_compute_forward_mul_mat_f16_f32(
// compute by src0 rows
#if defined(GGML_USE_CUBLAS)
if (ggml_v2_cuda_can_mul_mat(src0, src1, dst)) {
if (ggml_cuda_can_mul_mat(src0, src1, dst)) {
if (params->ith == 0 && params->type == GGML_V2_TASK_COMPUTE) {
ggml_v2_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
ggml_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
}
return;
}
@ -9880,9 +9880,9 @@ static void ggml_v2_compute_forward_mul_mat_q_f32(
// compute by src0 rows
#if defined(GGML_USE_CUBLAS)
if (ggml_v2_cuda_can_mul_mat(src0, src1, dst)) {
if (ggml_cuda_can_mul_mat(src0, src1, dst)) {
if (params->ith == 0 && params->type == GGML_V2_TASK_COMPUTE) {
ggml_v2_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
ggml_cuda_mul_mat(src0, src1, dst, params->wdata, params->wsize);
}
return;
}
@ -14060,10 +14060,10 @@ void ggml_v2_graph_compute(struct ggml_v2_context * ctx, struct ggml_v2_cgraph *
size_t cur = 0;
#if defined(GGML_USE_CUBLAS)
if (ggml_v2_cuda_can_mul_mat(node->src0, node->src1, node)) {
if (ggml_cuda_can_mul_mat(node->src0, node->src1, node)) {
node->n_tasks = 1; // TODO: this actually is doing nothing
// the threads are still spinning
cur = ggml_v2_cuda_mul_mat_get_wsize(node->src0, node->src1, node);
cur = ggml_cuda_mul_mat_get_wsize(node->src0, node->src1, node);
}
else
#elif defined(GGML_USE_CLBLAST)