Skip to content

Commit

Permalink
[SYCL][CUDA][HIP] Add support for querying device architecture (#10573)
Browse files Browse the repository at this point in the history 
Implements `sycl_ext_oneapi_device_architecture` for AMD and NVIDIA.
The device architecture can be queried through

`dev.get_info<sycl::ext::oneapi::experimental::info::device::architecture>()`
( `dev` is a `sycl::device`)
  • Loading branch information
MartinWehking authored Aug 16, 2023
1 parent c043037 commit 1ad69e5
Show file tree
Hide file tree
Showing 3 changed files with 96 additions and 44 deletions.
6 changes: 0 additions & 6 deletions sycl/doc/extensions/experimental/sycl_ext_oneapi_device_architecture.asciidoc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -589,12 +589,6 @@ feature, the application must be compiled in ahead-of-time (AOT) mode using
description of the `-fsycl-targets` option. These are the target names of the
form "intel_gpu_*", "nvidia_gpu_*", or "amd_gpu_*".

The two APIs `device::ext_oneapi_architecture_is` and the
`ext::oneapi::experimental::info::device::architecture` device descriptor are
currently supported only for Intel devices (both GPU and CPU). There is no
support yet for Nvidia or AMD devices.


== Future direction

This experimental extension is still evolving. We expect that future versions
Expand Down
132 changes: 95 additions & 37 deletions sycl/source/detail/device_info.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -568,65 +568,123 @@ struct get_device_info_impl<range<Dimensions>,
}
};

// This macro is only for AMD and NVIDIA GPU architectures
#define NVIDIA_AMD_ARCHES(X) \
X("5.0", oneapi_exp_arch::nvidia_gpu_sm_50) \
X("5.2", oneapi_exp_arch::nvidia_gpu_sm_52) \
X("5.3", oneapi_exp_arch::nvidia_gpu_sm_53) \
X("6.0", oneapi_exp_arch::nvidia_gpu_sm_60) \
X("6.1", oneapi_exp_arch::nvidia_gpu_sm_61) \
X("6.2", oneapi_exp_arch::nvidia_gpu_sm_62) \
X("7.0", oneapi_exp_arch::nvidia_gpu_sm_70) \
X("7.2", oneapi_exp_arch::nvidia_gpu_sm_72) \
X("7.5", oneapi_exp_arch::nvidia_gpu_sm_75) \
X("8.0", oneapi_exp_arch::nvidia_gpu_sm_80) \
X("8.6", oneapi_exp_arch::nvidia_gpu_sm_86) \
X("8.7", oneapi_exp_arch::nvidia_gpu_sm_87) \
X("8.9", oneapi_exp_arch::nvidia_gpu_sm_89) \
X("9.0", oneapi_exp_arch::nvidia_gpu_sm_90) \
X("gfx701", oneapi_exp_arch::amd_gpu_gfx701) \
X("gfx702", oneapi_exp_arch::amd_gpu_gfx702) \
X("gfx801", oneapi_exp_arch::amd_gpu_gfx801) \
X("gfx802", oneapi_exp_arch::amd_gpu_gfx802) \
X("gfx803", oneapi_exp_arch::amd_gpu_gfx803) \
X("gfx805", oneapi_exp_arch::amd_gpu_gfx805) \
X("gfx810", oneapi_exp_arch::amd_gpu_gfx810) \
X("gfx900", oneapi_exp_arch::amd_gpu_gfx900) \
X("gfx902", oneapi_exp_arch::amd_gpu_gfx902) \
X("gfx904", oneapi_exp_arch::amd_gpu_gfx904) \
X("gfx906", oneapi_exp_arch::amd_gpu_gfx906) \
X("gfx908", oneapi_exp_arch::amd_gpu_gfx908) \
X("gfx90a", oneapi_exp_arch::amd_gpu_gfx90a) \
X("gfx1010", oneapi_exp_arch::amd_gpu_gfx1010) \
X("gfx1011", oneapi_exp_arch::amd_gpu_gfx1011) \
X("gfx1012", oneapi_exp_arch::amd_gpu_gfx1012) \
X("gfx1013", oneapi_exp_arch::amd_gpu_gfx1013) \
X("gfx1030", oneapi_exp_arch::amd_gpu_gfx1030) \
X("gfx1031", oneapi_exp_arch::amd_gpu_gfx1031) \
X("gfx1032", oneapi_exp_arch::amd_gpu_gfx1032) \
X("gfx1034", oneapi_exp_arch::amd_gpu_gfx1034)

// This macro is only for Intel GPU architectures
#define INTEL_ARCHES(X) \
X(0x02000000, oneapi_exp_arch::intel_gpu_bdw) \
X(0x02400009, oneapi_exp_arch::intel_gpu_skl) \
X(0x02404009, oneapi_exp_arch::intel_gpu_kbl) \
X(0x02408009, oneapi_exp_arch::intel_gpu_cfl) \
X(0x0240c000, oneapi_exp_arch::intel_gpu_apl) \
X(0x02410000, oneapi_exp_arch::intel_gpu_glk) \
X(0x02414000, oneapi_exp_arch::intel_gpu_whl) \
X(0x02418000, oneapi_exp_arch::intel_gpu_aml) \
X(0x0241c000, oneapi_exp_arch::intel_gpu_cml) \
X(0x02c00000, oneapi_exp_arch::intel_gpu_icllp) \
X(0x02c08000, oneapi_exp_arch::intel_gpu_ehl) \
X(0x03000000, oneapi_exp_arch::intel_gpu_tgllp) \
X(0x03004000, oneapi_exp_arch::intel_gpu_rkl) \
X(0x03008000, oneapi_exp_arch::intel_gpu_adl_s) \
X(0x0300c000, oneapi_exp_arch::intel_gpu_adl_p) \
X(0x03010000, oneapi_exp_arch::intel_gpu_adl_n) \
X(0x03028000, oneapi_exp_arch::intel_gpu_dg1) \
X(0x030dc008, oneapi_exp_arch::intel_gpu_acm_g10) \
X(0x030e0005, oneapi_exp_arch::intel_gpu_acm_g11) \
X(0x030e4000, oneapi_exp_arch::intel_gpu_acm_g12) \
X(0x030f0007, oneapi_exp_arch::intel_gpu_pvc)

#define CMP_NVIDIA_AMD(s, i) \
if (strcmp(s, arch) == 0) \
return i;

#define CMP_INTEL(p, i) \
if (p == arch) \
return i;

template <>
struct get_device_info_impl<
ext::oneapi::experimental::architecture,
ext::oneapi::experimental::info::device::architecture> {
static ext::oneapi::experimental::architecture get(const DeviceImplPtr &Dev) {
using oneapi_exp_arch = sycl::ext::oneapi::experimental::architecture;
auto ReturnHelper = [](auto MapDeviceIpToArch, auto DeviceIp) {
// TODO: use std::map::contains instead of try-catch when SYCL RT be moved
// to C++20
try {
oneapi_exp_arch Result = MapDeviceIpToArch.at(DeviceIp);
return Result;
} catch (std::out_of_range &) {
backend CurrentBackend = Dev->getBackend();
if (Dev->is_gpu() && (backend::ext_oneapi_level_zero == CurrentBackend ||
backend::opencl == CurrentBackend)) {
auto MapArchIDToArchName = [](const int arch) {
INTEL_ARCHES(CMP_INTEL);
throw sycl::exception(
make_error_code(errc::runtime),
"The current device architecture is not supported by "
"sycl_ext_oneapi_device_architecture.");
}
};
backend CurrentBackend = Dev->getBackend();
if (Dev->is_gpu() && (backend::ext_oneapi_level_zero == CurrentBackend ||
backend::opencl == CurrentBackend)) {
std::map<uint32_t, oneapi_exp_arch> MapDeviceIpToArch = {
{0x02000000, oneapi_exp_arch::intel_gpu_bdw},
{0x02400009, oneapi_exp_arch::intel_gpu_skl},
{0x02404009, oneapi_exp_arch::intel_gpu_kbl},
{0x02408009, oneapi_exp_arch::intel_gpu_cfl},
{0x0240c000, oneapi_exp_arch::intel_gpu_apl},
{0x02410000, oneapi_exp_arch::intel_gpu_glk},
{0x02414000, oneapi_exp_arch::intel_gpu_whl},
{0x02418000, oneapi_exp_arch::intel_gpu_aml},
{0x0241c000, oneapi_exp_arch::intel_gpu_cml},
{0x02c00000, oneapi_exp_arch::intel_gpu_icllp},
{0x02c08000, oneapi_exp_arch::intel_gpu_ehl},
{0x03000000, oneapi_exp_arch::intel_gpu_tgllp},
{0x03004000, oneapi_exp_arch::intel_gpu_rkl},
{0x03008000, oneapi_exp_arch::intel_gpu_adl_s},
{0x0300c000, oneapi_exp_arch::intel_gpu_adl_p},
{0x03010000, oneapi_exp_arch::intel_gpu_adl_n},
{0x03028000, oneapi_exp_arch::intel_gpu_dg1},
{0x030dc008, oneapi_exp_arch::intel_gpu_acm_g10},
{0x030e0005, oneapi_exp_arch::intel_gpu_acm_g11},
{0x030e4000, oneapi_exp_arch::intel_gpu_acm_g12},
{0x030f0007, oneapi_exp_arch::intel_gpu_pvc},
};
uint32_t DeviceIp;
Dev->getPlugin()->call<PiApiKind::piDeviceGetInfo>(
Dev->getHandleRef(),
PiInfoCode<
ext::oneapi::experimental::info::device::architecture>::value,
sizeof(DeviceIp), &DeviceIp, nullptr);
return ReturnHelper(MapDeviceIpToArch, DeviceIp);
return MapArchIDToArchName(DeviceIp);
} else if (Dev->is_gpu() && (backend::ext_oneapi_cuda == CurrentBackend ||
backend::ext_oneapi_hip == CurrentBackend)) {
auto MapArchIDToArchName = [](const char *arch) {
NVIDIA_AMD_ARCHES(CMP_NVIDIA_AMD);
throw sycl::exception(
make_error_code(errc::runtime),
"The current device architecture is not supported by "
"sycl_ext_oneapi_device_architecture.");
};
size_t ResultSize = 0;
Dev->getPlugin()->call<PiApiKind::piDeviceGetInfo>(
Dev->getHandleRef(), PiInfoCode<info::device::version>::value, 0,
nullptr, &ResultSize);
std::unique_ptr<char[]> DeviceArch(new char[ResultSize]);
Dev->getPlugin()->call<PiApiKind::piDeviceGetInfo>(
Dev->getHandleRef(), PiInfoCode<info::device::version>::value,
ResultSize, DeviceArch.get(), nullptr);
return MapArchIDToArchName(DeviceArch.get());
} else if (Dev->is_cpu() && backend::opencl == CurrentBackend) {
// TODO: add support of different CPU architectures to
// sycl_ext_oneapi_device_architecture
return sycl::ext::oneapi::experimental::architecture::x86_64;
} // else is not needed
// TODO: add support of other arhitectures by extending with else if

// TODO: add support of other architectures by extending with else if
// Generating a user-friendly error message
std::string DeviceStr;
if (Dev->is_gpu())
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/DeviceArchitecture/device_architecture_on_host.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// UNSUPPORTED: cuda, hip, esimd_emulator, accelerator
// UNSUPPORTED: esimd_emulator, accelerator

// RUN: %{build} -o %t.out
// RUN: %{run} %t.out
Expand Down

0 comments on commit 1ad69e5

Please sign in to comment.