Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions sycl/test-e2e/AOT/fallback.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -7,12 +7,12 @@
//===----------------------------------------------------------------------===//

// REQUIRES: ocloc, gpu, target-spir
// UNSUPPORTED: gpu-intel-gen12
// UNSUPPORTED: gpu-intel-dg2
// UNSUPPORTED-INTENDED: Remove support for platform used as compile target
// since AOT image should be not applicable.

// AOT-compiled image for absent gen platform, run on GPU.
// RUN: %clangxx -fsycl -fsycl-targets=spir64,intel_gpu_tgl %S/Inputs/aot.cpp -o %t_spv_gpu.out
// RUN: %clangxx -fsycl -fsycl-targets=spir64,intel_gpu_acm_g10 %S/Inputs/aot.cpp -o %t_spv_gpu.out
// RUN: env ONEAPI_DEVICE_SELECTOR="*:gpu" SYCL_UR_TRACE=2 %{run-unfiltered-devices} %t_spv_gpu.out | FileCheck %s

// CHECK: ---> urProgramCreateWithIL
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/AOT/half.cpp
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
// This test ensures that a program that has a kernel
// using fp16 can be compiled AOT.

// REQUIRES: ocloc, opencl-aot, any-device-is-cpu
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_tgllp -o %t.tgllp.out %s
// REQUIRES: ocloc, opencl-aot, any-device-is-cpu, (!gpu-intel-gen12 || linux)
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_acm_g10 -o %t.dg2.out %s

// CPU AOT targets host isa, so we compile on the run system instead.
// RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64 -o %t.x86.out %s
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/AOT/reqd-sg-size.cpp
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
// This test ensures that a program that has a kernel
// using various required sub-group sizes can be compiled AOT.

// REQUIRES: ocloc, opencl-aot, any-device-is-cpu, opencl-cpu-rt
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_tgllp -o %t.tgllp.out %s
// REQUIRES: ocloc, opencl-aot, any-device-is-cpu, opencl-cpu-rt, (!gpu-intel-gen12 || linux)
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_acm_g10 -o %t.dg2.out %s
// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64 -o %t.x86.out %s

// ocloc on windows does not have support for PVC, so this command will
Expand Down
6 changes: 3 additions & 3 deletions sycl/test-e2e/BFloat16/bfloat16_example_aot.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -3,16 +3,16 @@
/// both GPU and CPU.
///

// REQUIRES: opencl-aot, ocloc, gpu-intel-gen12, any-device-is-cpu
// REQUIRES: opencl-aot, ocloc, any-device-is-cpu, (!gpu-intel-gen12)

// RUN: %clangxx -fsycl -fsycl-targets=spir64 %s -o %t.out
// RUN: %{run} %t.out

// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_gen -Xsycl-target-backend=spir64_gen "-device gen12lp" %s -o %t.out
// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" %s -o %t.out
// RUN: %{run} %t.out

// CPU AOT targets host isa, so we compile on the run system instead.
// RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device gen12lp" %s -o %t.out
// RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" %s -o %t.out
// RUN: %{run} %t.out

#include "bfloat16_example.hpp"
Expand Down
6 changes: 3 additions & 3 deletions sycl/test-e2e/BFloat16/bfloat16_example_aot_cpu.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -3,13 +3,13 @@
/// CPU.
///

// REQUIRES: opencl-aot, ocloc, gpu-intel-gen12, any-device-is-cpu
// REQUIRES: opencl-aot, ocloc, gpu-intel-dg2, any-device-is-cpu

// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg1" %s -o %t.out
// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" %s -o %t.out
// RUN: %if cpu %{ %{run} %t.out %}

// CPU AOT targets host isa, so we compile on the run system instead.
// RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg1" %s -o %t.out
// RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" %s -o %t.out
// RUN: %if cpu %{ %{run} %t.out %}

#include "bfloat16_example.hpp"
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/BFloat16/bfloat16_example_aot_gpu.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -3,9 +3,9 @@
/// GPU.
///

// REQUIRES: opencl-aot, ocloc, gpu-intel-gen12, any-device-is-gpu
// REQUIRES: opencl-aot, ocloc, gpu-intel-dg2, any-device-is-gpu

// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend "-device gen12lp" %s -o %t.out
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend "-device dg2" %s -o %t.out
// RUN: %if gpu %{%{run} %t.out %}

// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend "-device *" %s -o %t.out
Expand Down
6 changes: 3 additions & 3 deletions sycl/test-e2e/Basic/build_log.cpp
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
// REQUIRES: opencl || level_zero, gpu, ocloc
// UNSUPPORTED: arch-intel_gpu_dg1
//
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device dg1" %s -o %t.out
// UNSUPPORTED: arch-intel_gpu_dg2

// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" %s -o %t.out
// RUN: env SYCL_RT_WARNING_LEVEL=2 %{run} %t.out 2>&1 | FileCheck %s

#include <sycl/detail/core.hpp>
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/DeviceCodeSplit/aot-gpu.cpp
Original file line number Diff line number Diff line change
@@ -1,9 +1,9 @@
// REQUIRES: ocloc, gpu, target-spir
// REQUIRES: ocloc, gpu, target-spir, !gpu-intel-gen12
//
// RUN: %clangxx -fsycl -fsycl-device-code-split=per_source \
// RUN: -fsycl-targets=spir64_gen \
// RUN: -Xsycl-target-backend=spir64_gen \
// RUN: "-device tgllp" -I %S/Inputs -o %t.out \
// RUN: "-device dg2" -I %S/Inputs -o %t.out \
// RUN: %S/split-per-source-main.cpp \
// RUN: %S/Inputs/split-per-source-second-file.cpp \
// RUN: -fsycl-dead-args-optimization
Expand Down
6 changes: 3 additions & 3 deletions sycl/test-e2e/ESIMD/aot_mixed.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,10 @@
//
//===----------------------------------------------------------------------===//
// TODO: Enable on other GPUs once internal ticket is fixed
// REQUIRES: ocloc && gpu-intel-gen12
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device tgllp" -o %t.sycl.out -DENABLE_SYCL=0 %s
// REQUIRES: ocloc && gpu-intel-dg2
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" -o %t.sycl.out -DENABLE_SYCL=0 %s
// RUN: %{run} %t.sycl.out
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device tgllp" -o %t.out %s
// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device dg2" -o %t.out %s
// RUN: %{run} %t.out

// This test checks the following ESIMD ahead-of-time compilation scenarios:
Expand Down
18 changes: 9 additions & 9 deletions sycl/test-e2e/ESIMD/hardware_dispatch.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,8 @@
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// REQUIRES: ocloc && arch-intel_gpu_tgllp
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_tgllp %s -o %t.out
// REQUIRES: ocloc && gpu-intel-dg2
// RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_acm_g10 %s -o %t.out
// RUN: %{run-unfiltered-devices} %t.out

// This is basic test to test hardware dispatch functionality with ESIMD.
Expand Down Expand Up @@ -36,30 +36,30 @@ int main() {

// test if_architecture_is
sycl::ext::oneapi::experimental::if_architecture_is<
sycl::ext::oneapi::experimental::architecture::intel_gpu_tgllp>(
sycl::ext::oneapi::experimental::architecture::intel_gpu_acm_g10>(
[&]() { result[0] = 1; })
.otherwise([&]() { result[0] = 0; });

// test else_if_architecture_is
sycl::ext::oneapi::experimental::if_architecture_is<
sycl::ext::oneapi::experimental::architecture::intel_gpu_dg1>(
sycl::ext::oneapi::experimental::architecture::intel_gpu_pvc>(
[&]() { result[1] = 0; })
.else_if_architecture_is<
sycl::ext::oneapi::experimental::architecture::intel_gpu_tgllp>(
.else_if_architecture_is<sycl::ext::oneapi::experimental::
architecture::intel_gpu_acm_g10>(
[&]() { result[1] = 2; })
.otherwise([&]() { result[1] = 0; });

// test otherwise
sycl::ext::oneapi::experimental::if_architecture_is<
sycl::ext::oneapi::experimental::architecture::intel_gpu_dg1>(
sycl::ext::oneapi::experimental::architecture::intel_gpu_pvc>(
[&]() { result[2] = 0; })
.otherwise([&]() { result[2] = 3; });

// test more than one architecture template parameter is passed to
// if_architecture_is
sycl::ext::oneapi::experimental::if_architecture_is<
sycl::ext::oneapi::experimental::architecture::intel_gpu_dg1,
sycl::ext::oneapi::experimental::architecture::intel_gpu_tgllp>(
sycl::ext::oneapi::experimental::architecture::intel_gpu_pvc,
sycl::ext::oneapi::experimental::architecture::intel_gpu_acm_g10>(
[&]() { result[3] = 4; })
.otherwise([&]() { result[3] = 0; });
result.copy_to(output_ptr);
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/NewOffloadDriver/aot-gpu.cpp
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
// REQUIRES: ocloc, gpu, target-spir
// REQUIRES: ocloc, gpu, target-spir, !gpu-intel-gen12
// Test with `--offload-new-driver`
//
// RUN: %clangxx -fsycl -fsycl-device-code-split=per_source \
// RUN: -fsycl-targets=spir64_gen \
// RUN: -Xsycl-target-backend=spir64_gen \
// RUN: "-device tgllp" -I %S/Inputs -o %t.out \
// RUN: "-device dg2" -I %S/Inputs -o %t.out \
// RUN: %S/split-per-source-main.cpp \
// RUN: %S/Inputs/split-per-source-second-file.cpp \
// RUN: -fsycl-dead-args-optimization --offload-new-driver
Expand Down