Skip to content

clean TRT_PLUGIN_FP16_AVAILABLE define #76158

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
merged 1 commit into from
Nov 3, 2025
Merged

clean TRT_PLUGIN_FP16_AVAILABLE define #76158

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 0 additions & 4 deletions cmake/cuda.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -288,10 +288,6 @@ elseif(${CMAKE_CUDA_COMPILER_VERSION} LESS 13.0) # CUDA 12.0+
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
endif()

if(NOT ${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0)
add_definitions("-DTRT_PLUGIN_FP16_AVAILABLE")
endif()

add_definitions("-DCUDA_VERSION_MAJOR=\"${CUDA_VERSION_MAJOR}\"")
add_definitions("-DCUDA_VERSION_MINOR=\"${CUDA_VERSION_MINOR}\"")
add_definitions("-DCUDA_TOOLKIT_ROOT_DIR=\"${CUDA_TOOLKIT_ROOT_DIR}\"")
Expand Down
14 changes: 0 additions & 14 deletions paddle/fluid/inference/tensorrt/plugin/multihead_matmul_roformer_plugin.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,14 +84,9 @@ bool MultiheadMatmulRoformerPlugin::supportsFormatCombination(
const nvinfer1::PluginTensorDesc &in = in_out[pos];
if (pos == 0) {
if (with_fp16_) {
#ifdef TRT_PLUGIN_FP16_AVAILABLE
return (in.type == nvinfer1::DataType::kFLOAT ||
in.type == nvinfer1::DataType::kHALF) &&
(in.format == nvinfer1::TensorFormat::kLINEAR);
#else
return (in.type == nvinfer1::DataType::kFLOAT) &&
(in.format == nvinfer1::TensorFormat::kLINEAR);
#endif
} else {
return (in.type == nvinfer1::DataType::kFLOAT) &&
(in.format == nvinfer1::TensorFormat::kLINEAR);
Expand Down Expand Up @@ -270,7 +265,6 @@ int MultiheadMatmulRoformerPlugin::enqueue(
tptr, output, batch, seq_len, head_number_, head_size_);

} else if (input_type == nvinfer1::DataType::kHALF) {
#ifdef TRT_PLUGIN_FP16_AVAILABLE
VLOG(1) << "TRT Plugin DataType selected. QkvToContext-->fp16";
auto *multihead_temp_data =
multihead_temp_tensor.mutable_data<int16_t>( // NOLINT
Expand Down Expand Up @@ -353,14 +347,6 @@ int MultiheadMatmulRoformerPlugin::enqueue(
half *output = static_cast<half *>(outputs[0]);
transpose<half><<<grid, block, 0, stream>>>(
tptr, output, batch, seq_len, head_number_, head_size_);
#else
PADDLE_THROW(common::errors::Fatal(
"The Ernie(Bert) TensorRT Plugin should be "
"complied with CUDA version >= 10.0 when running with fp16. "
"Please recompile it or try to use fp32 by set "
"config.SetTRTDynamicShapeInfo(min_input_shape, "
"max_input_shape, opt_input_shape, true"));
#endif
} else {
PADDLE_THROW(common::errors::Fatal(
"The QKV TRT Plugin's input type should be float or half."));
Expand Down
3 changes: 0 additions & 3 deletions paddle/fluid/inference/tensorrt/plugin/trans_layernorm_op_plugin.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,6 @@ inline int getSMVersion() {
return prop.major * 10 + prop.minor;
}

#ifdef TRT_PLUGIN_FP16_AVAILABLE
#define FINAL_MASK 0xffffffff

template <int UNROLL_FACTOR>
Expand Down Expand Up @@ -105,8 +104,6 @@ __global__ void GeneralResidualLayerNormOpt2(half2 *normed_output,
half_n, \
eps);

#endif

int TransLayerNormPluginDynamic::initialize() TRT_NOEXCEPT {
if (!with_fp16_) {
cudaMalloc(&bias_gpu_, sizeof(float) * bias_.size());
Expand Down
Loading