Skip to content

feat: support aten::extend evaluator #704

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
wants to merge 2 commits into from
Closed

feat: support aten::extend evaluator #704

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
0516eb0
feat: support aten::extend evaluator
ruoqianguo Nov 11, 2021
512ec83
chore: return self in aten::extend
ruoqianguo Nov 16, 2021
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
2 changes: 1 addition & 1 deletion core/conversion/conversion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -398,7 +398,7 @@ void ConvertBlockToNetDef(
EvaluateConditionalBlock(ctx, n);
} else if (to_eval) {
auto eval = EvaluateNode(ctx, n);
if (eval) {
if (eval && n->outputs().size() > 0) {
if (n->outputs().size() > 1) { // For ListUnpack scenario
if (eval.value().isTuple()) {
auto eval_list = eval.value().toTuple();
Expand Down
21 changes: 21 additions & 0 deletions core/conversion/evaluators/aten.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -285,6 +285,27 @@ auto aten_registrations TORCHTRT_UNUSED =
EvalOptions().validSchemas({
"aten::append.t(t[](a!) self, t(c -> *) el) -> (t[](a!))",
})})
.evaluator({c10::Symbol::fromQualString("aten::extend"),
[](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> {
if (args.at(n->input(0)).IValue()->isList() && args.at(n->input(1)).IValue()->isList()) {
auto self = args.at(n->input(0)).IValue()->to<c10::List<c10::IValue>>();
auto other = args.at(n->input(1)).IValue()->to<c10::List<c10::IValue>>();
const int64_t other_size = other.size();

for (int64_t i = 0; i < other_size; i++) {
self.push_back(other.get(i));
}
return self;
} else {
TORCHTRT_THROW_ERROR(
"Unimplemented data type for aten::extend.t evaluator: "
<< args.at(n->input(0)).IValue()->type()->str() << ", "
<< args.at(n->input(1)).IValue()->type()->str());
}
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The function doesnt return anything. While you might be modifying state correctly, we want to make transforms as functional as possible. Create a copy of self, modify it and return that

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @narendasan , i tried to return self, but the output of node aten::extend.t(t[](a!) self, t[] other) -> () is None. It will occurs an error in there like C++ exception with description "vector::_M_range_check: __n (which is 0) >= this->size() (which is 0)"

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The function doesnt return anything. While you might be modifying state correctly, we want to make transforms as functional as possible. Create a copy of self, modify it and return that

I return self in aten::extend evaluator and add a condition to estimate the number of ouputs in ConvertBlockToNetDef.

},
EvalOptions().validSchemas({
"aten::extend.t(t[](a!) self, t[] other) -> ()",
})})
.evaluator({c10::Symbol::fromQualString("aten::neg"),
[](const torch::jit::Node* n, kwargs& args) -> c10::optional<torch::jit::IValue> {
auto el = args.at(n->input(0)).unwrapToInt();
Expand Down
26 changes: 26 additions & 0 deletions tests/core/conversion/evaluators/test_aten_evaluators.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -302,6 +302,32 @@ TEST(Evaluators, FloorFloatIntEvaluatesCorrectly) {
ASSERT_TRUE(jit_results[0] == trt_results[0]);
}

TEST(Evaluators, ATenExtendEvaluatesCorrectly) {
const auto graph = R"IR(
graph(%0 : Tensor, %1 : Tensor):
%2 : int = prim::Constant[value=0]()
%3 : Tensor[] = prim::ListConstruct(%0)
%4 : Tensor[] = prim::ListConstruct(%1)
aten::extend(%3, %4)
%5 : Tensor = aten::cat(%3, %2)
return (%5))IR";

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

auto in0 = at::randint(1, 10, {3, 4}, {at::kCUDA});
auto in1 = at::randint(1, 10, {5, 4}, {at::kCUDA});

auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in0, in1});

params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto trt_results = torch_tensorrt::tests::util::RunGraphEngine(g, params, {in0, in1});

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

TEST(Evaluators, ATenAppendWithITensorEvaluatesCorrectly) {
const auto graph = R"IR(
graph(%0 : Tensor, %1 : Tensor):
Expand Down