fix(//core/converters): Handle dtype mismatch in elementwise ops #1238
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Dheeraj Peri <peri.dheeraj@gmail.com>
Signed-off-by: Dheeraj Peri <peri.dheeraj@gmail.com>
github-actions
bot
added
component: conversion
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index 4d8795f..c8c2c00 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -184,7 +184,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/converters/converter_util.cpp b/tmp/changes.txt
index ee8a589..bc6abdd 100644
--- a/workspace/core/conversion/converters/converter_util.cpp
+++ b/tmp/changes.txt
@@ -118,9 +118,11 @@ nvinfer1::ILayer* add_elementwise(
}
// Two types are compatible if they are the same type or are both in the set {kFLOAT, kHALF}
- auto fp32_and_fp16 = (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
- auto fp16_and_fp32 = (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
- if (!fp32_and_fp16 && !fp16_and_fp32){
+ auto fp32_and_fp16 =
+ (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
+ auto fp16_and_fp32 =
+ (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
+ if (!fp32_and_fp16 && !fp16_and_fp32) {
if (self->getType() > other->getType()) {
self = castITensor(ctx, self, other->getType());
} else if (self->getType() < other->getType()) {
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 994fb25..540fa12 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -27,8 +27,8 @@ void pointwise_test_helper(
if (!singleInput) {
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
}
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index 4d8795f..c8c2c00 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -184,7 +184,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/converters/converter_util.cpp b/tmp/changes.txt
index ee8a589..bc6abdd 100644
--- a/workspace/core/conversion/converters/converter_util.cpp
+++ b/tmp/changes.txt
@@ -118,9 +118,11 @@ nvinfer1::ILayer* add_elementwise(
}
// Two types are compatible if they are the same type or are both in the set {kFLOAT, kHALF}
- auto fp32_and_fp16 = (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
- auto fp16_and_fp32 = (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
- if (!fp32_and_fp16 && !fp16_and_fp32){
+ auto fp32_and_fp16 =
+ (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
+ auto fp16_and_fp32 =
+ (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
+ if (!fp32_and_fp16 && !fp16_and_fp32) {
if (self->getType() > other->getType()) {
self = castITensor(ctx, self, other->getType());
} else if (self->getType() < other->getType()) {
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 994fb25..540fa12 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -27,8 +27,8 @@ void pointwise_test_helper(
if (!singleInput) {
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
}
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
ERROR: Some files do not conform to style guidelines
|
@peri044 I have this fix in the collections branch with extra tests, take a look |
14 tasks
7 tasks
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/py/torch_tensorrt/csrc/register_tensorrt_classes.cpp b/tmp/changes.txt
index 9db567c..ba07250 100644
--- a/workspace/py/torch_tensorrt/csrc/register_tensorrt_classes.cpp
+++ b/tmp/changes.txt
@@ -26,11 +26,11 @@ void RegisterTRTCompileSpec() {
static auto TORCHTRT_UNUSED TRTInputSignatureTSRegistration =
torch::class_<torch_tensorrt::pyapi::InputSignature>("tensorrt", "_InputSignature")
.def(torch::init<>())
- .def("_set_signature_ivalue_torchbind",
- [](const c10::intrusive_ptr<torch_tensorrt::pyapi::InputSignature>& self,
- torch::jit::IValue ival) {
- self->signature_ivalue = ival;
- })
+ .def(
+ "_set_signature_ivalue_torchbind",
+ [](const c10::intrusive_ptr<torch_tensorrt::pyapi::InputSignature>& self, torch::jit::IValue ival) {
+ self->signature_ivalue = ival;
+ })
.def("__str__", &torch_tensorrt::pyapi::InputSignature::to_str);
ADD_FIELD_GET_SET_REGISTRATION(
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index ca9ff4d..aa58dc7 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -176,7 +176,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/evaluators/eval_util.cpp b/tmp/changes.txt
index 742a4f4..c14f9a6 100644
--- a/workspace/core/conversion/evaluators/eval_util.cpp
+++ b/tmp/changes.txt
@@ -20,7 +20,6 @@ int64_t normalizeIndex(int64_t idx, int64_t list_size) {
return idx;
}
-
// TODO: Switch back to PyTorch canonical implimentation
c10::optional<torch::jit::IValue> toIValue(const torch::jit::Value* v) {
if (v->node()->kind() != torch::jit::prim::Constant || v->type()->cast<c10::FunctionType>()) {
diff --git a/workspace/core/conversion/evaluators/prim.cpp b/tmp/changes.txt
index 338c427..244c3e8 100644
--- a/workspace/core/conversion/evaluators/prim.cpp
+++ b/tmp/changes.txt
@@ -260,40 +260,39 @@ auto prim_registrations =
},
EvalOptions().validSchemas({"prim::shape(Tensor a) -> (int[])"})})
.evaluator({torch::jit::prim::TupleConstruct,
- [](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> {
- auto num_inputs = n->inputs().size();
- c10::IValue tuple = c10::ivalue::Tuple::create();
- switch (num_inputs) {
- case 0:
- tuple = c10::ivalue::Tuple::create();
- break;
- case 1:
- tuple = c10::ivalue::Tuple::create(std::move((*args.at(n->input(0)).IValue())));
- break;
- case 2: {
- tuple = c10::ivalue::Tuple::create(
- std::move(*(args.at(n->input(0)).IValue())),
- std::move(*(args.at(n->input(1)).IValue())));
- break;
- }
- case 3: {
- tuple = c10::ivalue::Tuple::create(
- std::move(*(args.at(n->input(0)).IValue())),
- std::move(*(args.at(n->input(1)).IValue())),
- std::move(*(args.at(n->input(2)).IValue())));
- break;
- }
- default: {
- std::vector<c10::IValue> elems;
- for (size_t i = 0; i < num_inputs; i++) {
- elems.push_back(*(args.at(n->input(i)).IValue()));
+ [](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> {
+ auto num_inputs = n->inputs().size();
+ c10::IValue tuple = c10::ivalue::Tuple::create();
+ switch (num_inputs) {
+ case 0:
+ tuple = c10::ivalue::Tuple::create();
+ break;
+ case 1:
+ tuple = c10::ivalue::Tuple::create(std::move((*args.at(n->input(0)).IValue())));
+ break;
+ case 2: {
+ tuple = c10::ivalue::Tuple::create(
+ std::move(*(args.at(n->input(0)).IValue())), std::move(*(args.at(n->input(1)).IValue())));
+ break;
+ }
+ case 3: {
+ tuple = c10::ivalue::Tuple::create(
+ std::move(*(args.at(n->input(0)).IValue())),
+ std::move(*(args.at(n->input(1)).IValue())),
+ std::move(*(args.at(n->input(2)).IValue())));
+ break;
+ }
+ default: {
+ std::vector<c10::IValue> elems;
+ for (size_t i = 0; i < num_inputs; i++) {
+ elems.push_back(*(args.at(n->input(i)).IValue()));
+ }
+ tuple = c10::ivalue::Tuple::create(std::move(elems));
+ break;
}
- tuple = c10::ivalue::Tuple::create(std::move(elems));
- break;
}
- }
- return c10::optional<torch::jit::IValue>(std::move(tuple));
- }})
+ return c10::optional<torch::jit::IValue>(std::move(tuple));
+ }})
.evaluator({torch::jit::prim::TupleIndex,
[](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> {
// Outputs is an IValue which has list of tensors which can be found in ctx->evaluated_value_map
diff --git a/workspace/core/conversion/converters/converter_util.cpp b/tmp/changes.txt
index 7bca1a9..9ef39c0 100644
--- a/workspace/core/conversion/converters/converter_util.cpp
+++ b/tmp/changes.txt
@@ -125,9 +125,11 @@ nvinfer1::ILayer* add_elementwise(
}
// Two types are compatible if they are the same type or are both in the set {kFLOAT, kHALF}
- auto fp32_and_fp16 = (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
- auto fp16_and_fp32 = (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
- if (!fp32_and_fp16 && !fp16_and_fp32){
+ auto fp32_and_fp16 =
+ (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
+ auto fp16_and_fp32 =
+ (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
+ if (!fp32_and_fp16 && !fp16_and_fp32) {
if (self->getType() > other->getType()) {
self = castITensor(ctx, self, other->getType());
} else if (self->getType() < other->getType()) {
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 939c9b7..db60896 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -30,14 +30,15 @@ void pointwise_test_helper(
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- TORCHTRT_CHECK(!((int_tensors && (float_int_tensors || int_float_tensors)) || (float_int_tensors && int_float_tensors)),
- "Invalid test configuration, only one of int_tensors, float_int_tensors, int_float_tensors can be true");
+ TORCHTRT_CHECK(
+ !((int_tensors && (float_int_tensors || int_float_tensors)) || (float_int_tensors && int_float_tensors)),
+ "Invalid test configuration, only one of int_tensors, float_int_tensors, int_float_tensors can be true");
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
- } else if(float_int_tensors) {
+ } else if (float_int_tensors) {
TORCHTRT_CHECK(!singleInput, "float_int_tensors tests require two inputs");
torch_inputs[0] = torch_inputs[0].to(at::kFloat);
torch_inputs[1] = torch_inputs[1].to(at::kInt);
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
diff --git a/workspace/tests/cpp/test_collections.cpp b/tmp/changes.txt
index 31495a4..d01665a 100644
--- a/workspace/tests/cpp/test_collections.cpp
+++ b/tmp/changes.txt
@@ -135,7 +135,7 @@ TEST(CppAPITests, TestCollectionListInput) {
auto compile_settings = torch_tensorrt::ts::CompileSpec(complex_input_shape2);
compile_settings.min_block_size = 1;
- //compile_settings.torch_executed_ops.push_back("aten::__getitem__");
+ // compile_settings.torch_executed_ops.push_back("aten::__getitem__");
// // FP16 execution
compile_settings.enabled_precisions = {torch::kHalf};
diff --git a/workspace/cpp/src/compile_spec.cpp b/tmp/changes.txt
index 8c7bb8b..cfbc228 100644
--- a/workspace/cpp/src/compile_spec.cpp
+++ b/tmp/changes.txt
@@ -69,22 +69,24 @@ torchtrt::core::CompileSpec init_compile_spec(CompileSpec& external) {
return internal;
} else {
torch::jit::IValue converted_input_signature;
- LOG_WARNING( "Input signature parsing is an experimental feature, behavior and APIs may change");
+ LOG_WARNING("Input signature parsing is an experimental feature, behavior and APIs may change");
to_internal_input_signature(external.graph_inputs.input_signature, converted_input_signature);
torchtrt::core::CompileSpec internal(converted_input_signature);
- TORCHTRT_CHECK(!external.require_full_compilation, \
- "Grouped inputs currently requires partial compilation to be enabled, \
+ TORCHTRT_CHECK(
+ !external.require_full_compilation,
+ "Grouped inputs currently requires partial compilation to be enabled, \
this restriction will be relaxed in a future release");
LOG_DEBUG("Grouped inputs currently requires additional settings to enable the feature");
- LOG_DEBUG("Adding the following ops to torch_executed_ops:" \
- << std::endl << " - aten::__getitem__" \
- << std::endl << " - prim::ListConstruct" \
- << std::endl << " - prim::ListUnpack" \
- << std::endl << " - prim::TupleIndex" \
- << std::endl << " - prim::TupleConstruct" \
- << std::endl << " - prim::TupleUnpack");
+ LOG_DEBUG(
+ "Adding the following ops to torch_executed_ops:" << std::endl
+ << " - aten::__getitem__" << std::endl
+ << " - prim::ListConstruct" << std::endl
+ << " - prim::ListUnpack" << std::endl
+ << " - prim::TupleIndex" << std::endl
+ << " - prim::TupleConstruct" << std::endl
+ << " - prim::TupleUnpack");
external.torch_executed_ops.push_back("aten::__getitem__");
external.torch_executed_ops.push_back("prim::ListConstruct");
external.torch_executed_ops.push_back("prim::ListUnpack");
ERROR: Some files do not conform to style guidelinesSigned-off-by: Dheeraj Peri <peri.dheeraj@gmail.com>
|
@narendasan Rebased with master and made the modifications as discussed in the collections PR |
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/converters/converter_util.cpp b/tmp/changes.txt
index 68bfa93..fc02dcc 100644
--- a/workspace/core/conversion/converters/converter_util.cpp
+++ b/tmp/changes.txt
@@ -118,9 +118,11 @@ nvinfer1::ILayer* add_elementwise(
}
// Two types are compatible if they are the same type or are both in the set {kFLOAT, kHALF}
- auto fp32_and_fp16 = (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
- auto fp16_and_fp32 = (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
- if (!fp32_and_fp16 && !fp16_and_fp32){
+ auto fp32_and_fp16 =
+ (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF);
+ auto fp16_and_fp32 =
+ (other->getType() == nvinfer1::DataType::kFLOAT) && (self->getType() == nvinfer1::DataType::kHALF);
+ if (!fp32_and_fp16 && !fp16_and_fp32) {
if (self->getType() > other->getType()) {
LOG_DEBUG("Type mismatch in node : " << name << ", casting self to " << other->getType());
self = castITensor(ctx, self, other->getType());
ERROR: Some files do not conform to style guidelines
narendasan
reviewed
Aug 9, 2022
| @@ -123,6 +116,19 @@ nvinfer1::ILayer* add_elementwise( | |||
| std::swap(self, other); | |||
| swapSelfOther = false; | |||
| } | |||
|
|
|||
| // Two types are compatible if they are the same type or are both in the set {kFLOAT, kHALF} | |||
| auto fp32_and_fp16 = (self->getType() == nvinfer1::DataType::kFLOAT) && (other->getType() == nvinfer1::DataType::kHALF); | |||
There was a problem hiding this comment.
This does not handle INT32 which was causing tests to fail earlier
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Handle dtype mismatch in elementwise ops
Type of change
Checklist: