Skip to content

Modify fusion_emitter_large_test to work on ROCm. #568

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
i-chaochen merged 2 commits into from
Jan 27, 2026
Merged

Modify fusion_emitter_large_test to work on ROCm. #568

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
dd9bc22
Modify fusion_emitter_large_test to work on ROCm.
zoranjovanovic-ns Jan 23, 2026
bad7ccb
Fix fall-through warning in support_legacy.cc
zoranjovanovic-ns Jan 23, 2026
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
25 changes: 17 additions & 8 deletions xla/backends/gpu/codegen/triton/fusion_emitter_large_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,13 +37,6 @@ class TritonGemmTest : public GpuCodegenTest {
.gpu_compute_capability();
}

void SetUp() override {
if (std::holds_alternative<se::RocmComputeCapability>(
GetGpuComputeCapability())) {
GTEST_SKIP() << "Not supported on ROCm until Triton is re-enabled.";
}
}

DebugOptions GetDebugOptionsForTest() const override {
DebugOptions debug_options = HloTestBase::GetDebugOptionsForTest();
debug_options.set_xla_gpu_cublas_fallback(false);
Expand All @@ -52,6 +45,10 @@ class TritonGemmTest : public GpuCodegenTest {
};

TEST_F(TritonGemmTest, IndexUsing64Bits) {
if (std::holds_alternative<se::RocmComputeCapability>(
GetGpuComputeCapability())) {
GTEST_SKIP() << "Not enough memory on ROCm.";
}
const char* kHloTextRef = R"(
HloModule r

Expand Down Expand Up @@ -135,10 +132,22 @@ ENTRY e {
EXPECT_TRUE(RunAndCompare(kHloText, ErrorSpec{/*aabs=*/1e-3, /*arel=*/1e-3}));
}

using TritonNormalizationTest = GpuCodegenTest;
class TritonNormalizationTest : public GpuCodegenTest {
public:
se::GpuComputeCapability GetGpuComputeCapability() {
return backend()
.default_stream_executor()
->GetDeviceDescription()
.gpu_compute_capability();
}
};

TEST_F(TritonNormalizationTest,
CanEmitDiamondWithInputNumberOfElementsLargerThanInt32Max) {
if (std::holds_alternative<se::RocmComputeCapability>(
GetGpuComputeCapability())) {
GTEST_SKIP() << "Not enough memory on ROCm.";
}
const std::string hlo_text = R"(
HloModule softmax

Expand Down
2 changes: 2 additions & 0 deletions xla/backends/gpu/codegen/triton/support_legacy.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,6 +247,7 @@ bool IsDotAlgorithmSupportedByTriton(
if (rocm_compute_capability) {
return rocm_compute_capability->has_bf16_dtype_support();
}
return false;
case PrecisionConfig::ALG_DOT_BF16_BF16_F32_X6:
case PrecisionConfig::ALG_DOT_BF16_BF16_F32_X9:
if (cuda_compute_capability) {
Expand All @@ -260,6 +261,7 @@ bool IsDotAlgorithmSupportedByTriton(
if (rocm_compute_capability) {
return true;
}
return false;
default:
return false;
}
Expand Down