Skip to content

feat: Added a variant for aten::fake_quant_per_tensor #2107

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
Aug 2, 2023
Merged

feat: Added a variant for aten::fake_quant_per_tensor #2107

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
36 changes: 25 additions & 11 deletions core/conversion/converters/impl/quantization.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,22 @@ namespace {

#if NV_TENSORRT_MAJOR > 7
// clang-format off

bool add_qdq(ConversionCtx *ctx, const torch::jit::Node* n, nvinfer1::ITensor* input, nvinfer1::ITensor* scale, std::string& opName) {
nvinfer1::IQuantizeLayer* quantize_layer = ctx->net->addQuantize(*input, *scale);
TORCHTRT_CHECK(quantize_layer, "Unable to create QuantizeLayer from node: " << *n);
quantize_layer->setAxis(0);

nvinfer1::IDequantizeLayer* dequantize_layer = ctx->net->addDequantize(*quantize_layer->getOutput(0), *scale);
TORCHTRT_CHECK(dequantize_layer, "Unable to create DequantizeLayer from node: " << *n);
dequantize_layer->setAxis(0);

auto qdq_out = ctx->AssociateValueAndTensor(n->outputs()[0], dequantize_layer->getOutput(0));
LOG_DEBUG("[" << opName << "]"<< " Output tensor shape: " << qdq_out->getDimensions());

return true;
}

auto quantization_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns()
.pattern({"aten::fake_quantize_per_tensor_affine(Tensor self, float scale, int zero_point, int quant_min, int quant_max) -> (Tensor)",
[](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
Expand All @@ -20,18 +36,16 @@ auto quantization_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns
auto scale = args[1].unwrapToScalar().to<float>();
auto scaleTensor = tensor_to_const(ctx, torch::tensor({scale}));
// Add and configure a QuantizeLayer.
nvinfer1::IQuantizeLayer* quantize_layer = ctx->net->addQuantize(*input, *scaleTensor);
quantize_layer->setAxis(0);

// Add and configure DequantizeLayer following a QuantizeLayer
nvinfer1::IDequantizeLayer* dequantize_layer = ctx->net->addDequantize(*quantize_layer->getOutput(0), *scaleTensor);
dequantize_layer->setAxis(0);

auto qdq_out = ctx->AssociateValueAndTensor(n->outputs()[0], dequantize_layer->getOutput(0));
LOG_DEBUG("[fake_quantize_per_tensor_affine] Output tensor shape: " << qdq_out->getDimensions());

return true;
std::string opName("aten::fake_quantize_per_tensor_affine");
return add_qdq(ctx, n, input, scaleTensor, opName);
}})
.pattern({"aten::fake_quantize_per_tensor_affine.tensor_qparams(Tensor self, Tensor scale, Tensor zero_point, int quant_min, int quant_max) -> (Tensor)",
[](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
auto input = args[0].ITensorOrFreeze(ctx);
auto scale = args[1].ITensorOrFreeze(ctx);
std::string opName("aten::fake_quantize_per_tensor_affine.tensor_qparams");
return add_qdq(ctx, n, input, scale, opName);
}})
.pattern({"aten::fake_quantize_per_channel_affine(Tensor self, Tensor scale, Tensor zero_point, int axis, int quant_min, int quant_max) -> (Tensor)",
[](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
// This aten operator is generated from torch.fake_quantize_per_channel_affine op in Pytorch python API.
Expand Down
34 changes: 34 additions & 0 deletions tests/core/conversion/converters/test_quantization.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,40 @@ TEST(Converters, ATenFakeQuantizePerTensorConvertsCorrectly) {
torch_tensorrt::tests::util::almostEqual(jit_results[0], trt_results[0].reshape_as(jit_results[0]), 2e-6));
}

TEST(Converters, ATenFakeQuantizePerTensorWithParamsConvertsCorrectly) {
const auto graph = R"IR(
graph(%x.1 : Tensor):
%22 : int = prim::Constant[value=-128]()
%14 : int = prim::Constant[value=4]()
%9 : None = prim::Constant()
%35 : Device = prim::Constant[value="cuda:0"]()
%6 : int = prim::Constant[value=6]()
%7 : int = prim::Constant[value=3]()
%3 : int = prim::Constant[value=1]()
%5 : float = prim::Constant[value=3.5]()
%13 : int = prim::Constant[value=1]()
%23 : int = prim::Constant[value=127]()
%4 : int[] = prim::ListConstruct(%3)
%11 : Tensor = aten::full(%4, %5, %6, %9, %35, %9)
%12 : int[] = prim::ListConstruct(%3)
%19 : Tensor = aten::full(%12, %13, %7, %9, %35, %9)
%quant_input.1 : Tensor = aten::fake_quantize_per_tensor_affine(%x.1, %11, %19, %22, %23)
return (%quant_input.1))IR";

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

auto in = at::randint(1, 10, {1, 5, 5, 5}, {at::kCUDA}).to(at::kFloat);

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

params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
auto trt_results = torch_tensorrt::tests::util::RunGraphEngine(g, params, {in}, nvinfer1::DataType::kINT8);

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

TEST(Converters, ATenFakeQuantizePerChannelConvertsCorrectly) {
const auto graph = R"IR(
graph(%x.1 : Tensor):
Expand Down