Skip to content

Commit

Permalink
feat: Add converter files for reflection pad 1d and 2d
Browse files Browse the repository at this point in the history 
Signed-off-by: Dheeraj Peri <peri.dheeraj@gmail.com>
  • Loading branch information
@peri044
peri044 committed Feb 24, 2022
1 parent 8e89578 commit 406d860
Show file tree
Hide file tree
Showing 2 changed files with 201 additions and 0 deletions.
145 changes: 145 additions & 0 deletions core/conversion/converters/impl/reflection_pad.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,145 @@
#include <ATen/ATen.h>
#include <vector>
#include "NvInfer.h"
#include "core/conversion/converters/converters.h"
#include "core/util/prelude.h"
#include "torch/torch.h"

namespace torch_tensorrt {
namespace core {
namespace conversion {
namespace converters {
namespace impl {
namespace {

auto reflection_padXd TORCHTRT_UNUSED =
RegisterNodeConversionPatterns()
.pattern({"aten::reflection_pad2d(Tensor self, int[4] padding) -> (Tensor)",
[](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
auto in = args[0].ITensor();
auto inDims = in->getDimensions();
int64_t inRank = inDims.nbDims;
auto padding = args[1].unwrapToIntList().vec();
if (padding.size() == 1) {
for (int64_t i = 0; i < 3; i++)
padding.push_back(padding[0]);
}
if (inRank == 4) {
TORCHTRT_CHECK(padding.size() == 4, "4D tensors expect 4 values for padding");
} else {
TORCHTRT_THROW_ERROR("Only 4D padding are supported for now");
}

std::vector<nvinfer1::ITensor*> tensors_vec;
// 2d padding: (padding_left, padding_right, padding_top, padding_bottom)

for (int64_t i = 0; i < int(padding.size() / 2); i++) {
int64_t axis = inRank - (i + 1); // axis = {inRank - 1, inRank - 2}
int64_t padding_index = i * 2;

if (padding[padding_index] > 0) { // left/top padding value
tensors_vec.clear();

for (int i = 0; i < padding[padding_index]; i++) {
at::Tensor left_indices = torch::tensor({padding[padding_index] - i}, torch::kInt32);
auto indicesTensor = tensor_to_const(ctx, left_indices);
auto left_gather_layer = ctx->net->addGather(*in, *indicesTensor, axis);
auto left_gather_out = left_gather_layer->getOutput(0);
tensors_vec.push_back(left_gather_out);
}
tensors_vec.push_back(in);
auto concat_layer = ctx->net->addConcatenation(tensors_vec.data(), tensors_vec.size());
concat_layer->setAxis(axis);
in = concat_layer->getOutput(0);
inDims = in->getDimensions();
}

if (padding[padding_index + 1] > 0) { // right/bottom padding value
tensors_vec.clear();
tensors_vec.push_back(in);

for (int i = 0; i < padding[padding_index + 1]; i++) {
nvinfer1::ITensor* indicesTensor = NULL;
auto indices = torch::tensor({inDims.d[axis] - 1 - (i + 1)}, torch::kInt32);
indicesTensor = tensor_to_const(ctx, indices);
auto right_gather_layer = ctx->net->addGather(*in, *indicesTensor, axis);
auto right_gather_out = right_gather_layer->getOutput(0);
tensors_vec.push_back(right_gather_out);
}

auto concat_layer = ctx->net->addConcatenation(tensors_vec.data(), tensors_vec.size());
concat_layer->setAxis(axis);
in = concat_layer->getOutput(0);
inDims = in->getDimensions();
}
}
auto out = ctx->AssociateValueAndTensor(n->outputs()[0], in);
LOG_DEBUG("Output tensor shape: " << out->getDimensions());

return true;
}})
.pattern({"aten::reflection_pad1d(Tensor self, int[2] padding) -> (Tensor)",
[](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
auto in = args[0].ITensor();
auto inDims = in->getDimensions();
int64_t inRank = inDims.nbDims;
auto padding = args[1].unwrapToIntList().vec();
if (padding.size() == 1) {
for (int64_t i = 0; i < 1; i++)
padding.push_back(padding[0]);
}

std::vector<nvinfer1::ITensor*> tensors_vec;
// 1d padding: (padding_left, padding_right)

int64_t axis = inRank - 1;
int64_t padding_index = 0;

if (padding[padding_index] > 0) { // left padding value
tensors_vec.clear();

for (int i = 0; i < padding[padding_index]; i++) {
at::Tensor left_indices = torch::tensor({padding[padding_index] - i}, torch::kInt32);
auto indicesTensor = tensor_to_const(ctx, left_indices);
auto left_gather_layer = ctx->net->addGather(*in, *indicesTensor, axis);
auto left_gather_out = left_gather_layer->getOutput(0);
tensors_vec.push_back(left_gather_out);
}
tensors_vec.push_back(in);
auto concat_layer = ctx->net->addConcatenation(tensors_vec.data(), tensors_vec.size());
concat_layer->setAxis(axis);
in = concat_layer->getOutput(0);
inDims = in->getDimensions();
}

if (padding[padding_index + 1] > 0) { // right padding value
tensors_vec.clear();
tensors_vec.push_back(in);

for (int i = 0; i < padding[padding_index + 1]; i++) {
nvinfer1::ITensor* indicesTensor = NULL;
auto indices = torch::tensor({inDims.d[axis] - 1 - (i + 1)}, torch::kInt32);
indicesTensor = tensor_to_const(ctx, indices);
auto right_gather_layer = ctx->net->addGather(*in, *indicesTensor, axis);
auto right_gather_out = right_gather_layer->getOutput(0);
tensors_vec.push_back(right_gather_out);
}

auto concat_layer = ctx->net->addConcatenation(tensors_vec.data(), tensors_vec.size());
concat_layer->setAxis(axis);
in = concat_layer->getOutput(0);
inDims = in->getDimensions();
}

auto out = ctx->AssociateValueAndTensor(n->outputs()[0], in);
LOG_DEBUG("Output tensor shape: " << out->getDimensions());

return true;
}});

} // namespace
} // namespace impl
} // namespace converters
} // namespace conversion
} // namespace core
} // namespace torch_tensorrt
56 changes: 56 additions & 0 deletions tests/core/conversion/converters/test_reflection_pad.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,56 @@
#include <iostream>
#include <string>
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"

TEST(Converters, ATenReflection_pad2dTensorConvertsCorrectly) {
const auto graph = R"IR(
graph(%0 : Tensor):
%1 : int = prim::Constant[value=1]()
%2 : int = prim::Constant[value=1]()
%3 : int = prim::Constant[value=2]()
%4 : int = prim::Constant[value=0]()
%5 : int[] = prim::ListConstruct(%1, %2, %3, %4)
%6 : Tensor = aten::reflection_pad2d(%0, %5)
return (%6))IR";

auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());

auto in1 = at::randint(1, 10, {1, 3, 5, 5}, {at::kCUDA});

auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in1});

params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto trt_results = torch_tensorrt::tests::util::RunGraphEngine(g, params, {in1});

ASSERT_TRUE(
torch_tensorrt::tests::util::almostEqual(jit_results[0], trt_results[0].reshape_as(jit_results[0]), 2e-6));
}

TEST(Converters, ATenReflection_pad1dTensorConvertsCorrectly) {
const auto graph = R"IR(
graph(%0 : Tensor):
%1 : int = prim::Constant[value=1]()
%2 : int = prim::Constant[value=2]()
%3 : int[] = prim::ListConstruct(%1, %2)
%4 : Tensor = aten::reflection_pad1d(%0, %3)
return (%4))IR";

auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());

auto in1 = at::randint(1, 10, {1, 2, 4}, {at::kCUDA});

auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in1});

params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto trt_results = torch_tensorrt::tests::util::RunGraphEngine(g, params, {in1});

ASSERT_TRUE(
torch_tensorrt::tests::util::almostEqual(jit_results[0], trt_results[0].reshape_as(jit_results[0]), 2e-6));
}

0 comments on commit 406d860

Please sign in to comment.