Skip to content

fix cuda kernel test for windows on cuda13 #75684

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

Closed
youge325 wants to merge 2 commits into from
Closed

fix cuda kernel test for windows on cuda13 #75684

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
7ec9f4d
fix cuda kernel test for windows on cuda13
youge325 Oct 8, 2025
30143c0
fix the length of error message
youge325 Oct 11, 2025
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
7 changes: 7 additions & 0 deletions paddle/phi/kernels/cpu/isfinite_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,5 +81,12 @@ INSTANTIATE_ISFINITE_KERNEL_Isinf(int, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(int64_t, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(phi::float16, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(phi::bfloat16, CPUContext);

INSTANTIATE_ISFINITE_KERNEL_Isfinite(float, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(double, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(int, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(int64_t, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(phi::float16, CPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(phi::bfloat16, CPUContext);
} // namespace phi
#endif
7 changes: 7 additions & 0 deletions paddle/phi/kernels/gpu/isfinite_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,5 +81,12 @@ INSTANTIATE_ISFINITE_KERNEL_Isinf(int, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(int64_t, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(phi::float16, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isinf(phi::bfloat16, GPUContext);

INSTANTIATE_ISFINITE_KERNEL_Isfinite(float, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(double, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(int, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(int64_t, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(phi::float16, GPUContext);
INSTANTIATE_ISFINITE_KERNEL_Isfinite(phi::bfloat16, GPUContext);
} // namespace phi
#endif
4 changes: 4 additions & 0 deletions paddle/phi/kernels/isfinite_kernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,5 +36,9 @@ DEFINE_ISFINITE_KERNEL(IsfiniteKernel)
#define INSTANTIATE_ISFINITE_KERNEL_Isnan(type, context) \
template PADDLE_API void IsnanKernel<type, context>( \
const context&, const DenseTensor&, DenseTensor*)

#define INSTANTIATE_ISFINITE_KERNEL_Isfinite(type, context) \
template PADDLE_API void IsfiniteKernel<type, context>( \
const context&, const DenseTensor&, DenseTensor*)
#endif
} // namespace phi
8 changes: 8 additions & 0 deletions test/cpp/fluid/platform/enforce_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -422,13 +422,21 @@ TEST(enforce, cuda_success) {
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_SETUP_FAILED, "CUFFT error"));
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_INVALID_SIZE, "CUFFT error"));
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_UNALIGNED_DATA, "CUFFT error"));

#ifdef CUFFT_INCOMPLETE_PARAMETER_LIST
EXPECT_TRUE(
CheckCudaStatusFailure(CUFFT_INCOMPLETE_PARAMETER_LIST, "CUFFT error"));
#endif
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_INVALID_DEVICE, "CUFFT error"));

#ifdef CUFFT_PARSE_ERROR
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_PARSE_ERROR, "CUFFT error"));
#endif
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_NO_WORKSPACE, "CUFFT error"));
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_NOT_IMPLEMENTED, "CUFFT error"));
#ifdef CUFFT_LICENSE_ERROR
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_LICENSE_ERROR, "CUFFT error"));
#endif
EXPECT_TRUE(CheckCudaStatusFailure(CUFFT_NOT_SUPPORTED, "CUFFT error"));

#if !defined(__APPLE__) && defined(PADDLE_WITH_NCCL)
Expand Down
93 changes: 82 additions & 11 deletions test/cpp/inference/tensorrt/test_tensorrt_engine_instruction.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,20 +85,79 @@ TEST(TensorRTEngineInstructionTest, test_tensorrt_engine_instruction) {
nvinfer1::DataType::kFLOAT, raw_bias, size);
auto *x = engine->DeclareInput(
"x", nvinfer1::DataType::kFLOAT, nvinfer1::Dims4{-1, 1, 1, 1});
auto *fc_layer = TRT_ENGINE_ADD_LAYER(
engine, FullyConnected, *x, size, weight.get(), bias.get());
PADDLE_ENFORCE_NOT_NULL(fc_layer,
common::errors::InvalidArgument(
"TRT fully connected layer building failed."));
auto *flatten_layer = engine->network()->addShuffle(*x);
PADDLE_ENFORCE_NOT_NULL(
flatten_layer,
common::errors::InvalidArgument(
"Unable to build the TensorRT shuffle layer for the input tensor "
"'x'. "
"This usually indicates the TensorRT network failed to allocate the "
"intermediate reshape layer."));
flatten_layer->setReshapeDimensions(nvinfer1::Dims2{-1, 1});

auto *weight_layer = TRT_ENGINE_ADD_LAYER(
engine, Constant, nvinfer1::Dims2{1, 1}, weight.get());
PADDLE_ENFORCE_NOT_NULL(
weight_layer,
common::errors::InvalidArgument("TensorRT failed to create the constant "
"layer for parameter 'weight'. "
"Please confirm the TensorRT builder "
"supports constant initialisation "
"for the provided weight shape."));

auto *bias_layer =
TRT_ENGINE_ADD_LAYER(engine, Constant, nvinfer1::Dims2{1, 1}, bias.get());
PADDLE_ENFORCE_NOT_NULL(
bias_layer,
common::errors::InvalidArgument(
"TensorRT failed to create the constant layer for parameter 'bias'. "
"Check whether the provided bias data matches the expected shape."));

auto *matmul_layer = TRT_ENGINE_ADD_LAYER(engine,
MatrixMultiply,
*flatten_layer->getOutput(0),
nvinfer1::MatrixOperation::kNONE,
*weight_layer->getOutput(0),
nvinfer1::MatrixOperation::kNONE);
PADDLE_ENFORCE_NOT_NULL(
matmul_layer,
common::errors::InvalidArgument(
"TensorRT returned a null matrix-multiply layer while fusing the "
"fully-connected op. Verify the network input ranks and TensorRT "
"version."));

auto *add_layer = TRT_ENGINE_ADD_LAYER(engine,
ElementWise,
*matmul_layer->getOutput(0),
*bias_layer->getOutput(0),
nvinfer1::ElementWiseOperation::kSUM);
PADDLE_ENFORCE_NOT_NULL(
add_layer,
common::errors::InvalidArgument(
"TensorRT could not construct the elementwise-add layer for bias "
"fusion. Ensure the bias tensor uses broadcastable dimensions."));

engine->DeclareOutput(fc_layer, 0, "y");
auto *reshape_layer = engine->network()->addShuffle(*add_layer->getOutput(0));
PADDLE_ENFORCE_NOT_NULL(
reshape_layer,
common::errors::InvalidArgument(
"TensorRT could not emit the final shuffle layer to restore the "
"output shape. Confirm the shape tensor and inferred dimensions are "
"valid."));
reshape_layer->setReshapeDimensions(nvinfer1::Dims4{-1, 1, 1, 1});

engine->DeclareOutput(reshape_layer, 0, "y");
std::vector<std::string> input_names = {"x", ""};
std::vector<std::string> output_names = {"y"};
std::vector<std::vector<int64_t>> outputs_shape = {{1}};
std::vector<phi::DataType> outputs_dtype = {phi::DataType::FLOAT32};
LOG(INFO) << "freeze network";
engine->FreezeNetwork();
#if IS_TRT_VERSION_GE(8600)
ASSERT_EQ(engine->engine()->getNbIOTensors(), 2);
#else
ASSERT_EQ(engine->engine()->getNbBindings(), 2);
#endif
nvinfer1::IHostMemory *serialized_engine_data = engine->Serialize();

std::ofstream outFile("engine_serialized_data.bin", std::ios::binary);
Expand Down Expand Up @@ -220,7 +279,10 @@ TEST(TensorRTEngineInstructionTest, test_tensorrt_engine_instruction_dynamic) {
layer->setInput(1, *shape);
PADDLE_ENFORCE_NOT_NULL(
layer,
common::errors::InvalidArgument("TRT shuffle layer building failed."));
common::errors::InvalidArgument(
"TensorRT failed to construct the dynamic shuffle layer that "
"consumes the runtime shape tensor. Please check the provided "
"shape binding."));
engine->DeclareOutput(layer, 0, "y");
engine->FreezeNetwork();

Expand Down Expand Up @@ -401,14 +463,19 @@ TEST(PluginTest, test_generic_plugin) {
creator->createPlugin("pir_generic_plugin", plugin_collection.get());
PADDLE_ENFORCE_NOT_NULL(
generic_plugin,
common::errors::InvalidArgument("TRT create generic plugin failed."));
common::errors::InvalidArgument(
"TensorRT plugin registry returned nullptr while creating "
"'pir_generic_plugin'. Verify the plugin has been registered before "
"building the engine."));
std::vector<nvinfer1::ITensor *> plugin_inputs;
plugin_inputs.emplace_back(x);
auto plugin_layer = engine->network()->addPluginV2(
plugin_inputs.data(), plugin_inputs.size(), *generic_plugin);
PADDLE_ENFORCE_NOT_NULL(plugin_layer,
common::errors::InvalidArgument(
"TRT generic plugin layer building failed."));
PADDLE_ENFORCE_NOT_NULL(
plugin_layer,
common::errors::InvalidArgument(
"TensorRT failed to add the generic plugin layer to the network. "
"Ensure the plugin inputs match the expected TensorRT types."));

engine->DeclareOutput(plugin_layer, 0, "y");
std::vector<std::string> input_names = {"x"};
Expand All @@ -417,7 +484,11 @@ TEST(PluginTest, test_generic_plugin) {
std::vector<phi::DataType> outputs_dtype = {phi::DataType::FLOAT32};
LOG(INFO) << "freeze network";
engine->FreezeNetwork();
#if IS_TRT_VERSION_GE(8600)
ASSERT_EQ(engine->engine()->getNbIOTensors(), 2);
#else
ASSERT_EQ(engine->engine()->getNbBindings(), 2);
#endif
nvinfer1::IHostMemory *serialized_engine_data = engine->Serialize();
std::ofstream outFile("engine_serialized_data.bin", std::ios::binary);
outFile.write(static_cast<const char *>(serialized_engine_data->data()),
Expand Down
Loading