Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

[SYCL] Updated SYCL device filtering (#8901)

Browse Source 
* Updated device filter to depend on default_selector (fixes non-intel device issues)
* Small related update to example/sycl Readme
This commit is contained in:
Ouadie EL FAROUKI 2024-08-07 11:25:36 +01:00 committed by GitHub
parent a8dbc6f753
commit 0478174d59
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 25 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
examples/sycl/README.md
View File

@ -12,9 +12,9 @@ This example program provides the tools for llama.cpp for SYCL on Intel GPU.
List all SYCL devices with ID, compute capability, max work group size, ect. List all SYCL devices with ID, compute capability, max work group size, ect.
1. Build the llama.cpp for SYCL for all targets. 1. Build the llama.cpp for SYCL for the specified target *(using GGML_SYCL_TARGET)*.
2. Enable oneAPI running environment 2. Enable oneAPI running environment *(if GGML_SYCL_TARGET is set to INTEL -default-)*
``` ```
source /opt/intel/oneapi/setvars.sh source /opt/intel/oneapi/setvars.sh
@ -29,19 +29,13 @@ source /opt/intel/oneapi/setvars.sh
Check the ID in startup log, like: Check the ID in startup log, like:
``` ```
found 4 SYCL devices: found 2 SYCL devices:
Device 0: Intel(R) Arc(TM) A770 Graphics, compute capability 1.3, | | | | |Max | |Max |Global | |
max compute_units 512, max work group size 1024, max sub group size 32, global mem size 16225243136 | | | | |compute|Max work|sub |mem | |
Device 1: Intel(R) FPGA Emulation Device, compute capability 1.2, |ID| Device Type| Name|Version|units |group |group|size | Driver version|
max compute_units 24, max work group size 67108864, max sub group size 64, global mem size 67065057280 |--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|
Device 2: 13th Gen Intel(R) Core(TM) i7-13700K, compute capability 3.0, | 0| [level_zero:gpu:0]| Intel Arc A770 Graphics| 1.3| 512| 1024| 32| 16225M| 1.3.29138|
max compute_units 24, max work group size 8192, max sub group size 64, global mem size 67065057280 | 1| [level_zero:gpu:1]| Intel UHD Graphics 750| 1.3| 32| 512| 32| 62631M| 1.3.29138|
Device 3: Intel(R) Arc(TM) A770 Graphics, compute capability 3.0,
max compute_units 512, max work group size 1024, max sub group size 32, global mem size 16225243136
``` ```
|Attribute|Note|
|-|-|
|compute capability 1.3|Level-zero running time, recommended |
|compute capability 3.0|OpenCL running time, slower than level-zero in most cases|

19
ggml/src/ggml-sycl/dpct/helper.hpp
View File

@ -874,7 +874,7 @@ namespace dpct
inline std::string get_preferred_gpu_platform_name() { inline std::string get_preferred_gpu_platform_name() {
std::string result; std::string result;
std::string filter = "level-zero"; std::string filter = "";
char* env = getenv("ONEAPI_DEVICE_SELECTOR"); char* env = getenv("ONEAPI_DEVICE_SELECTOR");
if (env) { if (env) {
if (std::strstr(env, "level_zero")) { if (std::strstr(env, "level_zero")) {
@ -892,11 +892,24 @@ namespace dpct
else { else {
throw std::runtime_error("invalid device filter: " + std::string(env)); throw std::runtime_error("invalid device filter: " + std::string(env));
} }
} else {
auto default_device = sycl::device(sycl::default_selector_v);
auto default_platform_name = default_device.get_platform().get_info<sycl::info::platform::name>();
if (std::strstr(default_platform_name.c_str(), "Level-Zero") || default_device.is_cpu()) {
filter = "level-zero";
}
else if (std::strstr(default_platform_name.c_str(), "CUDA")) {
filter = "cuda";
}
else if (std::strstr(default_platform_name.c_str(), "HIP")) {
filter = "hip";
}
} }
auto plaform_list = sycl::platform::get_platforms(); auto platform_list = sycl::platform::get_platforms();
for (const auto& platform : plaform_list) { for (const auto& platform : platform_list) {
auto devices = platform.get_devices(); auto devices = platform.get_devices();
auto gpu_dev = std::find_if(devices.begin(), devices.end(), [](const sycl::device& d) { auto gpu_dev = std::find_if(devices.begin(), devices.end(), [](const sycl::device& d) {
return d.is_gpu(); return d.is_gpu();