root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 02:44:36 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Fix Vulkan crash on APUs with very little device memory (#5424)

Browse Source 
* Fix Vulkan crash on APUs with very little device memory

* Fix debug output function names
This commit is contained in:
0cc4m 2024-02-09 06:52:33 +01:00 committed by GitHub
parent 8e6a9d2de0
commit 44fbe34360
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
ggml-vulkan.cpp
View File

@ -744,6 +744,8 @@ static vk_buffer ggml_vk_create_buffer(ggml_backend_vk_context * ctx, size_t siz
}
if (memory_type_index >= mem_props.memoryTypeCount) {
ctx->device.lock()->device.destroyBuffer(buf->buffer);
buf->size = 0;
throw vk::OutOfDeviceMemoryError("No suitable memory type found");
}
@ -3875,7 +3877,7 @@ static ggml_tensor * ggml_vk_find_last_use(const ggml_tensor * node, ggml_cgraph
static void ggml_vk_preallocate_buffers_graph(ggml_backend_vk_context * ctx, ggml_tensor * node){
#ifdef GGML_VULKAN_DEBUG
std::cerr << "ggml_ctx->preallocate_buffers_graph(" << node << ")" << std::endl;
std::cerr << "ggml_vk_preallocate_buffers_graph(" << node << ")" << std::endl;
#endif
const bool any_on_device = node->backend == GGML_BACKEND_GPU
|| (node->src[0] != nullptr && (node->src[0]->backend == GGML_BACKEND_GPU || node->src[0]->backend == GGML_BACKEND_GPU_SPLIT))
@ -3994,8 +3996,7 @@ static void ggml_vk_preallocate_buffers(ggml_backend_vk_context * ctx) {
return;
}
#ifdef GGML_VULKAN_DEBUG
std::cerr << "ggml_ctx->preallocate_buffers()" << std::endl;
std::cerr << "qx_size: " << ctx->prealloc_size_qx << " qy_size: " << ctx->prealloc_size_qy << " x_size: " << ctx->prealloc_size_x << " y_size: " << ctx->prealloc_size_y << " split_k_size: " << ctx->prealloc_size_split_k << std::endl;
std::cerr << "ggml_vk_preallocate_buffers(qx_size: " << ctx->prealloc_size_qx << " qy_size: " << ctx->prealloc_size_qy << " x_size: " << ctx->prealloc_size_x << " y_size: " << ctx->prealloc_size_y << " split_k_size: " << ctx->prealloc_size_split_k << ")" << std::endl;
#endif
#if defined(GGML_VULKAN_RUN_TESTS)
ctx->staging = ggml_vk_create_buffer_check(ctx, 100ul * 1024ul * 1024ul, vk::MemoryPropertyFlagBits::eHostVisible | vk::MemoryPropertyFlagBits::eHostCoherent | vk::MemoryPropertyFlagBits::eHostCached);