fix: fix the bug that introduces kLong Tensor in prim::NumToTensor #972
Conversation
This file contains hidden or 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: Bo Wang <[email protected]>
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/partitioning/partitioning.cpp b/tmp/changes.txt
index d171ae1..c69c32e 100644
--- a/workspace/core/partitioning/partitioning.cpp
+++ b/tmp/changes.txt
@@ -231,8 +231,8 @@ std::unordered_map<torch::jit::Value*, usage_info> getInputUsageCounts(
return usage_counts;
}
-std::unordered_map<size_t, std::list<SegmentedBlock>::iterator>
-getIdxtoIterMap(std::list<SegmentedBlock> &segmented_blocks_list) {
+std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> getIdxtoIterMap(
+ std::list<SegmentedBlock>& segmented_blocks_list) {
std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> idx_to_iter;
auto iter = segmented_blocks_list.begin();
for (int i = 0; i < segmented_blocks_list.size(); ++i, ++iter) {
@@ -283,9 +283,10 @@ void resolveNonTensorInputBlocks(PartitionedGraph& segmented_blocks) {
}
void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
- // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which produces/contains it.
- auto usage_counts = getInputUsageCounts(
- segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
+ // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which
+ // produces/contains it.
+ auto usage_counts =
+ getInputUsageCounts(segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
// Get idx of the segblock to its iterator mapping
std::list<SegmentedBlock> segmented_blocks_list(segmented_blocks.cbegin(), segmented_blocks.cend());
@@ -293,12 +294,13 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
std::unordered_set<int> updated_segments;
// we need to re-segment TensorRT segments whose inputs are TensorLists
- for (auto &use : usage_counts) {
+ for (auto& use : usage_counts) {
auto use_info = use.second;
// For a particular tensorlist input, traverse through all ids of segmented blocks whose target is TensorRT
for (auto i : use_info.tensorrt_use_id) {
if (!updated_segments.count(i)) {
- // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this tensorlist input}
+ // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this
+ // tensorlist input}
std::unordered_map<torch::jit::Value*, SegmentedBlock> tensorlistinput_to_segblock;
for (auto input : segmented_blocks[i].raw_inputs()) {
if (isTensorList(input)) {
@@ -308,18 +310,20 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
// For each tensorlist input in tensorlistinput_to_segblock, get the node which actually uses this input.
// Once we retrieve the node, we remove it from the current TensorRT segmented_blocks[i]. This node should be
- // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first place.
+ // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first
+ // place.
auto seg_blocks = segmentBlocksWithTensorListInputs(segmented_blocks[i], tensorlistinput_to_segblock);
auto append_blocks = seg_blocks.first;
auto trt_block = seg_blocks.second;
- // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that uses tensorlist input removed.
+ // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that
+ // uses tensorlist input removed.
auto next_iter = segmented_blocks_list.erase(idx_to_iter[i]);
if (trt_block.raw_nodes().size() > 0) {
segmented_blocks_list.insert(next_iter, trt_block);
}
// append blocks' nodes to the producer seg_block
- for (auto append_block: append_blocks) {
+ for (auto append_block : append_blocks) {
auto input = append_block.first; // corresponds to the tensorlist input
auto block = append_block.second;
// append nodes to segmented_blocks_list
diff --git a/workspace/core/partitioning/shape_analysis.cpp b/tmp/changes.txt
index 96b1312..e773b8f 100644
--- a/workspace/core/partitioning/shape_analysis.cpp
+++ b/tmp/changes.txt
@@ -1,5 +1,5 @@
-#include <ATen/ATen.h>
#include "core/partitioning/shape_analysis.h"
+#include <ATen/ATen.h>
#include "core/util/prelude.h"
#include "torch/csrc/jit/api/module.h"
#include "torch/csrc/jit/passes/constant_pooling.h"
diff --git a/workspace/core/lowering/passes/reduce_gelu.cpp b/tmp/changes.txt
index 15315ba..946df75 100644
--- a/workspace/core/lowering/passes/reduce_gelu.cpp
+++ b/tmp/changes.txt
@@ -12,8 +12,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
%out : Tensor = aten::gelu(%x)
return (%out))IR";
- // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions use
- // an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
+ // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions
+ // use an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
std::string gelu_approximate_pattern = R"IR(
graph(%x : Tensor, %approx):
%out : Tensor = aten::gelu(%x, %approx)
@@ -64,7 +64,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
map_gelu_to_pointwise_ops.runOnGraph(graph);
torch::jit::SubgraphRewriter map_gelu_approximate_to_pointwise_ops;
- map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
+ map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(
+ gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
map_gelu_approximate_to_pointwise_ops.runOnGraph(graph);
LOG_GRAPH("Post lowering of [aten::gelu] -> " << *graph);
diff --git a/workspace/core/lowering/passes/linear_to_addmm.cpp b/tmp/changes.txt
index c3160a8..e0e9ca3 100644
--- a/workspace/core/lowering/passes/linear_to_addmm.cpp
+++ b/tmp/changes.txt
@@ -1,15 +1,15 @@
#include <torch/csrc/jit/runtime/operator.h>
+#include "core/util/prelude.h"
+#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/ir/alias_analysis.h"
#include "torch/csrc/jit/jit_log.h"
#include "torch/csrc/jit/passes/constant_propagation.h"
#include "torch/csrc/jit/passes/dead_code_elimination.h"
#include "torch/csrc/jit/passes/guard_elimination.h"
#include "torch/csrc/jit/passes/peephole.h"
-#include "torch/csrc/jit/runtime/graph_executor.h"
-#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/passes/subgraph_rewrite.h"
-#include "core/util/prelude.h"
+#include "torch/csrc/jit/runtime/graph_executor.h"
namespace torch_tensorrt {
namespace core {
@@ -34,7 +34,8 @@ void replaceLinearWithBiasNonePattern(std::shared_ptr<torch::jit::Graph> graph)
continue;
} else {
torch::jit::WithInsertPoint guard(*it);
- std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();;
+ std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();
+ ;
torch::jit::Value* new_output = insertGraph(*it->owningGraph(), *d_graph, it->inputs()).at(0);
new_output->setType(it->output()->type());
it->output()->replaceAllUsesWith(new_output);
diff --git a/workspace/tests/util/util.h b/tmp/changes.txt
index 38ba81e..b795667 100644
--- a/workspace/tests/util/util.h
+++ b/tmp/changes.txt
@@ -1,8 +1,8 @@
#pragma once
+#include <ATen/ATen.h>
#include <string>
#include <vector>
-#include <ATen/ATen.h>
#include "ATen/Tensor.h"
#include "core/ir/ir.h"
#include "core/util/prelude.h"
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/partitioning/partitioning.cpp b/tmp/changes.txt
index d171ae1..c69c32e 100644
--- a/workspace/core/partitioning/partitioning.cpp
+++ b/tmp/changes.txt
@@ -231,8 +231,8 @@ std::unordered_map<torch::jit::Value*, usage_info> getInputUsageCounts(
return usage_counts;
}
-std::unordered_map<size_t, std::list<SegmentedBlock>::iterator>
-getIdxtoIterMap(std::list<SegmentedBlock> &segmented_blocks_list) {
+std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> getIdxtoIterMap(
+ std::list<SegmentedBlock>& segmented_blocks_list) {
std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> idx_to_iter;
auto iter = segmented_blocks_list.begin();
for (int i = 0; i < segmented_blocks_list.size(); ++i, ++iter) {
@@ -283,9 +283,10 @@ void resolveNonTensorInputBlocks(PartitionedGraph& segmented_blocks) {
}
void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
- // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which produces/contains it.
- auto usage_counts = getInputUsageCounts(
- segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
+ // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which
+ // produces/contains it.
+ auto usage_counts =
+ getInputUsageCounts(segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
// Get idx of the segblock to its iterator mapping
std::list<SegmentedBlock> segmented_blocks_list(segmented_blocks.cbegin(), segmented_blocks.cend());
@@ -293,12 +294,13 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
std::unordered_set<int> updated_segments;
// we need to re-segment TensorRT segments whose inputs are TensorLists
- for (auto &use : usage_counts) {
+ for (auto& use : usage_counts) {
auto use_info = use.second;
// For a particular tensorlist input, traverse through all ids of segmented blocks whose target is TensorRT
for (auto i : use_info.tensorrt_use_id) {
if (!updated_segments.count(i)) {
- // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this tensorlist input}
+ // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this
+ // tensorlist input}
std::unordered_map<torch::jit::Value*, SegmentedBlock> tensorlistinput_to_segblock;
for (auto input : segmented_blocks[i].raw_inputs()) {
if (isTensorList(input)) {
@@ -308,18 +310,20 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
// For each tensorlist input in tensorlistinput_to_segblock, get the node which actually uses this input.
// Once we retrieve the node, we remove it from the current TensorRT segmented_blocks[i]. This node should be
- // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first place.
+ // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first
+ // place.
auto seg_blocks = segmentBlocksWithTensorListInputs(segmented_blocks[i], tensorlistinput_to_segblock);
auto append_blocks = seg_blocks.first;
auto trt_block = seg_blocks.second;
- // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that uses tensorlist input removed.
+ // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that
+ // uses tensorlist input removed.
auto next_iter = segmented_blocks_list.erase(idx_to_iter[i]);
if (trt_block.raw_nodes().size() > 0) {
segmented_blocks_list.insert(next_iter, trt_block);
}
// append blocks' nodes to the producer seg_block
- for (auto append_block: append_blocks) {
+ for (auto append_block : append_blocks) {
auto input = append_block.first; // corresponds to the tensorlist input
auto block = append_block.second;
// append nodes to segmented_blocks_list
diff --git a/workspace/core/partitioning/shape_analysis.cpp b/tmp/changes.txt
index 96b1312..e773b8f 100644
--- a/workspace/core/partitioning/shape_analysis.cpp
+++ b/tmp/changes.txt
@@ -1,5 +1,5 @@
-#include <ATen/ATen.h>
#include "core/partitioning/shape_analysis.h"
+#include <ATen/ATen.h>
#include "core/util/prelude.h"
#include "torch/csrc/jit/api/module.h"
#include "torch/csrc/jit/passes/constant_pooling.h"
diff --git a/workspace/core/lowering/passes/reduce_gelu.cpp b/tmp/changes.txt
index 15315ba..946df75 100644
--- a/workspace/core/lowering/passes/reduce_gelu.cpp
+++ b/tmp/changes.txt
@@ -12,8 +12,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
%out : Tensor = aten::gelu(%x)
return (%out))IR";
- // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions use
- // an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
+ // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions
+ // use an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
std::string gelu_approximate_pattern = R"IR(
graph(%x : Tensor, %approx):
%out : Tensor = aten::gelu(%x, %approx)
@@ -64,7 +64,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
map_gelu_to_pointwise_ops.runOnGraph(graph);
torch::jit::SubgraphRewriter map_gelu_approximate_to_pointwise_ops;
- map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
+ map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(
+ gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
map_gelu_approximate_to_pointwise_ops.runOnGraph(graph);
LOG_GRAPH("Post lowering of [aten::gelu] -> " << *graph);
diff --git a/workspace/core/lowering/passes/linear_to_addmm.cpp b/tmp/changes.txt
index c3160a8..e0e9ca3 100644
--- a/workspace/core/lowering/passes/linear_to_addmm.cpp
+++ b/tmp/changes.txt
@@ -1,15 +1,15 @@
#include <torch/csrc/jit/runtime/operator.h>
+#include "core/util/prelude.h"
+#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/ir/alias_analysis.h"
#include "torch/csrc/jit/jit_log.h"
#include "torch/csrc/jit/passes/constant_propagation.h"
#include "torch/csrc/jit/passes/dead_code_elimination.h"
#include "torch/csrc/jit/passes/guard_elimination.h"
#include "torch/csrc/jit/passes/peephole.h"
-#include "torch/csrc/jit/runtime/graph_executor.h"
-#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/passes/subgraph_rewrite.h"
-#include "core/util/prelude.h"
+#include "torch/csrc/jit/runtime/graph_executor.h"
namespace torch_tensorrt {
namespace core {
@@ -34,7 +34,8 @@ void replaceLinearWithBiasNonePattern(std::shared_ptr<torch::jit::Graph> graph)
continue;
} else {
torch::jit::WithInsertPoint guard(*it);
- std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();;
+ std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();
+ ;
torch::jit::Value* new_output = insertGraph(*it->owningGraph(), *d_graph, it->inputs()).at(0);
new_output->setType(it->output()->type());
it->output()->replaceAllUsesWith(new_output);
diff --git a/workspace/tests/util/util.h b/tmp/changes.txt
index 38ba81e..b795667 100644
--- a/workspace/tests/util/util.h
+++ b/tmp/changes.txt
@@ -1,8 +1,8 @@
#pragma once
+#include <ATen/ATen.h>
#include <string>
#include <vector>
-#include <ATen/ATen.h>
#include "ATen/Tensor.h"
#include "core/ir/ir.h"
#include "core/util/prelude.h"
ERROR: Some files do not conform to style guidelines
|
Hi @narendasan, I didn't change anything other than the 3 files that in the commit, why i'm still getting these linting issues? |
narendasan
reviewed
Apr 12, 2022
| // won't be upgraded to kDouble or kLong since we don't support these 2 types in conversion | ||
| if (device == at::kCPU) { | ||
| if (s.isFloatingPoint()) { | ||
| LOG_WARNING("Unable to process input type of at::kDouble, truncate type to at::kInt in scalar_to_tensor_util "); |
There was a problem hiding this comment.
Can you add a comment here that is something like
// TODO: Conditionally enable truncation based on user setting?
| // won't be upgraded to kDouble or kLong since we don't support these 2 types in conversion | ||
| if (device == at::kCPU) { | ||
| if (s.isFloatingPoint()) { | ||
| LOG_WARNING("Unable to process input type of at::kDouble, truncate type to at::kInt in scalar_to_tensor_util "); |
There was a problem hiding this comment.
truncate type to at::kInt at::kFloat
| } | ||
| } | ||
| if (s.isFloatingPoint()) { | ||
| LOG_WARNING("Unable to process input type of at::kDouble, truncate type to at::kInt in scalar_to_tensor_util "); |
There was a problem hiding this comment.
truncate type to at::kInt at::kFloat
| @@ -119,6 +119,38 @@ void checkSequenceSize(int64_t n, int64_t dim, int64_t seq_size) { | |||
| } | |||
| } | |||
|
|
|||
| at::Tensor scalar_to_tensor_util(const at::Scalar& s, const at::Device device = at::kCPU) { | |||
There was a problem hiding this comment.
We can just call this scale_to_tensor since its in our namespace
core/conversion/evaluators/prim.cpp
Outdated
| @@ -31,7 +31,7 @@ auto prim_registrations = | |||
| }}) | |||
| .evaluator({torch::jit::prim::NumToTensor, | |||
| [](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> { | |||
| return at::scalar_to_tensor(args.at(n->input(0)).IValue()->toScalar()); | |||
| return scalar_to_tensor_util(args.at(n->input(0)).IValue()->toScalar()); | |||
There was a problem hiding this comment.
just change this to evaluators::scalar_to_tensor
Signed-off-by: Bo Wang <[email protected]>
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/partitioning/partitioning.cpp b/tmp/changes.txt
index d171ae1..c69c32e 100644
--- a/workspace/core/partitioning/partitioning.cpp
+++ b/tmp/changes.txt
@@ -231,8 +231,8 @@ std::unordered_map<torch::jit::Value*, usage_info> getInputUsageCounts(
return usage_counts;
}
-std::unordered_map<size_t, std::list<SegmentedBlock>::iterator>
-getIdxtoIterMap(std::list<SegmentedBlock> &segmented_blocks_list) {
+std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> getIdxtoIterMap(
+ std::list<SegmentedBlock>& segmented_blocks_list) {
std::unordered_map<size_t, std::list<SegmentedBlock>::iterator> idx_to_iter;
auto iter = segmented_blocks_list.begin();
for (int i = 0; i < segmented_blocks_list.size(); ++i, ++iter) {
@@ -283,9 +283,10 @@ void resolveNonTensorInputBlocks(PartitionedGraph& segmented_blocks) {
}
void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
- // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which produces/contains it.
- auto usage_counts = getInputUsageCounts(
- segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
+ // usage_counts is a map with key as non-tensor/tensorlist inputs and value as the idx of segmented block which
+ // produces/contains it.
+ auto usage_counts =
+ getInputUsageCounts(segmented_blocks, [](torch::jit::Value* input) -> bool { return isTensorList(input); });
// Get idx of the segblock to its iterator mapping
std::list<SegmentedBlock> segmented_blocks_list(segmented_blocks.cbegin(), segmented_blocks.cend());
@@ -293,12 +294,13 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
std::unordered_set<int> updated_segments;
// we need to re-segment TensorRT segments whose inputs are TensorLists
- for (auto &use : usage_counts) {
+ for (auto& use : usage_counts) {
auto use_info = use.second;
// For a particular tensorlist input, traverse through all ids of segmented blocks whose target is TensorRT
for (auto i : use_info.tensorrt_use_id) {
if (!updated_segments.count(i)) {
- // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this tensorlist input}
+ // tensorlistinput_to_segblock is a mapping from {tensorlist input : segmented block which produced this
+ // tensorlist input}
std::unordered_map<torch::jit::Value*, SegmentedBlock> tensorlistinput_to_segblock;
for (auto input : segmented_blocks[i].raw_inputs()) {
if (isTensorList(input)) {
@@ -308,18 +310,20 @@ void resolveTensorListInputBlocks(PartitionedGraph& segmented_blocks) {
// For each tensorlist input in tensorlistinput_to_segblock, get the node which actually uses this input.
// Once we retrieve the node, we remove it from the current TensorRT segmented_blocks[i]. This node should be
- // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first place.
+ // added to block that generated/produced (can be obtained via produce_id) this tensorlist input in the first
+ // place.
auto seg_blocks = segmentBlocksWithTensorListInputs(segmented_blocks[i], tensorlistinput_to_segblock);
auto append_blocks = seg_blocks.first;
auto trt_block = seg_blocks.second;
- // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that uses tensorlist input removed.
+ // Remove the current TensorRT seg_block and replace it with new TRT block (non empty) which has the node that
+ // uses tensorlist input removed.
auto next_iter = segmented_blocks_list.erase(idx_to_iter[i]);
if (trt_block.raw_nodes().size() > 0) {
segmented_blocks_list.insert(next_iter, trt_block);
}
// append blocks' nodes to the producer seg_block
- for (auto append_block: append_blocks) {
+ for (auto append_block : append_blocks) {
auto input = append_block.first; // corresponds to the tensorlist input
auto block = append_block.second;
// append nodes to segmented_blocks_list
diff --git a/workspace/core/partitioning/shape_analysis.cpp b/tmp/changes.txt
index 96b1312..e773b8f 100644
--- a/workspace/core/partitioning/shape_analysis.cpp
+++ b/tmp/changes.txt
@@ -1,5 +1,5 @@
-#include <ATen/ATen.h>
#include "core/partitioning/shape_analysis.h"
+#include <ATen/ATen.h>
#include "core/util/prelude.h"
#include "torch/csrc/jit/api/module.h"
#include "torch/csrc/jit/passes/constant_pooling.h"
diff --git a/workspace/core/conversion/evaluators/eval_util.cpp b/tmp/changes.txt
old mode 100755
new mode 100644
diff --git a/workspace/core/conversion/evaluators/prim.cpp b/tmp/changes.txt
old mode 100755
new mode 100644
diff --git a/workspace/core/lowering/passes/reduce_gelu.cpp b/tmp/changes.txt
index 15315ba..946df75 100644
--- a/workspace/core/lowering/passes/reduce_gelu.cpp
+++ b/tmp/changes.txt
@@ -12,8 +12,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
%out : Tensor = aten::gelu(%x)
return (%out))IR";
- // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions use
- // an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
+ // This gelu_approximate_pattern schema exists in 21.11, 21.12, 22.01 containers of pytorch. These container versions
+ // use an unmerged PR in pytorch : https://github.com/pytorch/pytorch/pull/61439. We reduce this to regular Gelu.
std::string gelu_approximate_pattern = R"IR(
graph(%x : Tensor, %approx):
%out : Tensor = aten::gelu(%x, %approx)
@@ -64,7 +64,8 @@ void ReduceGelu(std::shared_ptr<torch::jit::Graph>& graph) {
map_gelu_to_pointwise_ops.runOnGraph(graph);
torch::jit::SubgraphRewriter map_gelu_approximate_to_pointwise_ops;
- map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
+ map_gelu_approximate_to_pointwise_ops.RegisterRewritePattern(
+ gelu_approximate_pattern, gelu_reduce_multi_input_pattern);
map_gelu_approximate_to_pointwise_ops.runOnGraph(graph);
LOG_GRAPH("Post lowering of [aten::gelu] -> " << *graph);
diff --git a/workspace/core/lowering/passes/linear_to_addmm.cpp b/tmp/changes.txt
index c3160a8..e0e9ca3 100644
--- a/workspace/core/lowering/passes/linear_to_addmm.cpp
+++ b/tmp/changes.txt
@@ -1,15 +1,15 @@
#include <torch/csrc/jit/runtime/operator.h>
+#include "core/util/prelude.h"
+#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/ir/alias_analysis.h"
#include "torch/csrc/jit/jit_log.h"
#include "torch/csrc/jit/passes/constant_propagation.h"
#include "torch/csrc/jit/passes/dead_code_elimination.h"
#include "torch/csrc/jit/passes/guard_elimination.h"
#include "torch/csrc/jit/passes/peephole.h"
-#include "torch/csrc/jit/runtime/graph_executor.h"
-#include "torch/csrc/jit/api/function_impl.h"
#include "torch/csrc/jit/passes/subgraph_rewrite.h"
-#include "core/util/prelude.h"
+#include "torch/csrc/jit/runtime/graph_executor.h"
namespace torch_tensorrt {
namespace core {
@@ -34,7 +34,8 @@ void replaceLinearWithBiasNonePattern(std::shared_ptr<torch::jit::Graph> graph)
continue;
} else {
torch::jit::WithInsertPoint guard(*it);
- std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();;
+ std::shared_ptr<torch::jit::Graph> d_graph = toGraphFunction(decompose_funcs.get_function("linear")).graph();
+ ;
torch::jit::Value* new_output = insertGraph(*it->owningGraph(), *d_graph, it->inputs()).at(0);
new_output->setType(it->output()->type());
it->output()->replaceAllUsesWith(new_output);
diff --git a/workspace/core/conversion/evaluators/eval_util.h b/tmp/changes.txt
old mode 100755
new mode 100644
diff --git a/workspace/tests/util/util.h b/tmp/changes.txt
index 38ba81e..b795667 100644
--- a/workspace/tests/util/util.h
+++ b/tmp/changes.txt
@@ -1,8 +1,8 @@
#pragma once
+#include <ATen/ATen.h>
#include <string>
#include <vector>
-#include <ATen/ATen.h>
#include "ATen/Tensor.h"
#include "core/ir/ir.h"
#include "core/util/prelude.h"
ERROR: Some files do not conform to style guidelines
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.
Signed-off-by: Bo Wang [email protected]
Description
In our evaluation pass the prim::NumToTensor will use the function here https://github.com/pytorch/pytorch/blob/8a7c9a5e01a24d465126210234aa9d3775b25032/aten/src/ATen/ScalarOps.h#L29 to convert a number to a Tensor. However, this will convert int to Tensor kLong, this will incur the conversion phase break down since we don't support kLong types in TRT.
We can either add a truncate_long_and_double in conversion phase right after evaluation or change the native function from pytorch to fix this issue.
Fixes #956
Type of change
Checklist: