happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

vulkan: disable async for older Intel devices (#17369) 

* vulkan: disable async for older Intel devices

* update detection logic

* use name string for detection
This commit is contained in:
Jeff Bolz 2025-11-21 02:58:17 -06:00 committed by GitHub
parent 2370665e56
commit f1ffbba68e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 19 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -513,6 +513,7 @@ struct vk_device_struct {
vk_queue compute_queue; vk_queue compute_queue;
vk_queue transfer_queue; vk_queue transfer_queue;
bool single_queue; bool single_queue;
bool support_async;
uint32_t subgroup_size; uint32_t subgroup_size;
uint32_t shader_core_count; uint32_t shader_core_count;
bool uma; bool uma;
@ -4273,6 +4274,16 @@ static vk_device ggml_vk_get_device(size_t idx) {
device->vendor_id = device->properties.vendorID; device->vendor_id = device->properties.vendorID;
device->driver_id = driver_props.driverID; device->driver_id = driver_props.driverID;
// Implementing the async backend interfaces seems broken on older Intel HW,
// see https://github.com/ggml-org/llama.cpp/issues/17302.
device->support_async = (device->vendor_id != VK_VENDOR_ID_INTEL ||
std::string(device->properties.deviceName.data()).find("(DG1)") == std::string::npos) &&
getenv("GGML_VK_DISABLE_ASYNC") == nullptr;
if (!device->support_async) {
GGML_LOG_DEBUG("ggml_vulkan: WARNING: Async execution disabled on certain Intel devices.\n");
}
const char* GGML_VK_FORCE_MAX_ALLOCATION_SIZE = getenv("GGML_VK_FORCE_MAX_ALLOCATION_SIZE"); const char* GGML_VK_FORCE_MAX_ALLOCATION_SIZE = getenv("GGML_VK_FORCE_MAX_ALLOCATION_SIZE");
if (GGML_VK_FORCE_MAX_ALLOCATION_SIZE != nullptr) { if (GGML_VK_FORCE_MAX_ALLOCATION_SIZE != nullptr) {
@ -13187,6 +13198,10 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
ctx->device->perf_logger->print_timings(); ctx->device->perf_logger->print_timings();
} }
if (!ctx->device->support_async) {
ggml_vk_synchronize(ctx);
}
return GGML_STATUS_SUCCESS; return GGML_STATUS_SUCCESS;
UNUSED(backend); UNUSED(backend);
@ -13480,6 +13495,10 @@ ggml_backend_t ggml_backend_vk_init(size_t dev_num) {
/* .context = */ ctx, /* .context = */ ctx,
}; };
if (!ctx->device->support_async) {
vk_backend->iface.get_tensor_async = nullptr;
}
return vk_backend; return vk_backend;
} }