public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-10-20 17:23:00 +02:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: fix amd iGPU slow checkpoint read (llama/30049)

Browse Source
This commit is contained in:
Ruben Ortlam
2026-10-19 09:47:10 +02:00
committed by Daniel Bevenius
parent 770ce1ad21
commit 372eda18ab
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/ggml-vulkan-buffers.cpp
+4 -1
View File
@@ -623,7 +623,10 @@ void ggml_vk_buffer_read_2d(vk_buffer& src, size_t offset, void * dst, size_t sp
// If the device is not an UMA device the memory is host-accessible through rebar. While writing
// through PCIe is sufficient fast reading back data from PCIe is slower than going through
// the HW device to host copy path.
if(src->memory_property_flags & vk::MemoryPropertyFlagBits::eHostVisible && src->device->uma) {
// AMD UMA: uncached host-visible memory is write-combined, CPU reads are slow
const bool slow_host_read = src->device->vendor_id == VK_VENDOR_ID_AMD &&
!(src->memory_property_flags & vk::MemoryPropertyFlagBits::eHostCached);
if(src->memory_property_flags & vk::MemoryPropertyFlagBits::eHostVisible && src->device->uma && !slow_host_read) {
GGML_ASSERT(src->memory_property_flags & vk::MemoryPropertyFlagBits::eHostCoherent);
std::lock_guard<std::recursive_mutex> guard(src->device->mutex);