vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-27 00:14:49 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

sycl : Level Zero detection in ggml_sycl_init (#23097)

Browse source 
* [SYCL] Centralize Level Zero detection in ggml_sycl_init

* use the same wording

* get back the warning
This commit is contained in:
Alexey Kopytko 2026-05-22 21:49:45 +09:00 • committed by GitHub
parent 56f16f235c
commit bcfd1989e9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 10 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-sycl/common.hpp
View file

@ -238,6 +238,8 @@ struct ggml_sycl_device_info {
std::array<float, GGML_SYCL_MAX_DEVICES> default_tensor_split = {};
int max_work_group_sizes[GGML_SYCL_MAX_DEVICES] = {0};
bool ext_oneapi_level_zero = true; // sycl::backend::ext_oneapi_level_zero used by all enumerated GPU devices
};
const ggml_sycl_device_info & ggml_sycl_info();

26
ggml/src/ggml-sycl/ggml-sycl.cpp
View file

@ -98,7 +98,7 @@ static ggml_sycl_device_info ggml_sycl_init() {
for (int i = 0; i < info.device_count; ++i) {
info.devices[i].vmm = 0;
dpct::device_info prop;
sycl::device device = dpct::dev_mgr::instance().get_device(i);
auto & device = dpct::dev_mgr::instance().get_device(i);
SYCL_CHECK(CHECK_TRY_ERROR(dpct::get_device_info(
prop, device)));
@ -117,6 +117,12 @@ static ggml_sycl_device_info ggml_sycl_init() {
info.devices[i].max_wg_per_cu = info.max_work_group_sizes[i] / prop.get_max_compute_units();
info.devices[i].hw_info = get_device_hw_info(&device);
// Only check GPU devices; CPU devices use OpenCL and would otherwise
// disable Level Zero for the GPUs on systems without ONEAPI_DEVICE_SELECTOR set.
if (device.is_gpu() && device.default_queue().get_backend() != sycl::backend::ext_oneapi_level_zero) {
GGML_LOG_WARN("SYCL GPU device %d does not use Level Zero backend, disabling Level Zero memory API\n", i);
info.ext_oneapi_level_zero = false;
}
}
for (int id = 0; id < info.device_count; ++id) {
@ -230,26 +236,10 @@ static void ggml_check_sycl() try {
g_ggml_sycl_disable_dnn = get_sycl_env("GGML_SYCL_DISABLE_DNN", 0);
g_ggml_sycl_prioritize_dmmv = get_sycl_env("GGML_SYCL_PRIORITIZE_DMMV", 0);
#ifdef GGML_SYCL_SUPPORT_LEVEL_ZERO
g_ggml_sycl_enable_level_zero = get_sycl_env("GGML_SYCL_ENABLE_LEVEL_ZERO", 1);
g_ggml_sycl_enable_level_zero = get_sycl_env("GGML_SYCL_ENABLE_LEVEL_ZERO", ggml_sycl_info().ext_oneapi_level_zero);
#else
g_ggml_sycl_enable_level_zero = 0;
#endif
if (g_ggml_sycl_enable_level_zero) {
// Verify all GPU devices use the Level Zero backend before enabling L0 APIs.
// Only check GPU devices; CPU devices use OpenCL and would otherwise
// disable Level Zero for the GPUs on systems without ONEAPI_DEVICE_SELECTOR set.
for (unsigned int i = 0; i < dpct::dev_mgr::instance().device_count(); i++) {
auto & q = dpct::dev_mgr::instance().get_device(i).default_queue();
if (!q.get_device().is_gpu()) {
continue;
}
if (q.get_backend() != sycl::backend::ext_oneapi_level_zero) {
GGML_LOG_WARN("SYCL GPU device %d does not use Level Zero backend, disabling Level Zero memory API\n", i);
g_ggml_sycl_enable_level_zero = 0;
break;
}
}
}
#ifdef SYCL_FLASH_ATTN
g_ggml_sycl_enable_flash_attention = get_sycl_env("GGML_SYCL_ENABLE_FLASH_ATTN", 1);