Skip to content

aten::layer_norm perf improvement #2289

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

aten::layer_norm perf improvement #2289

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
eccca5f
Rewrite layer norm converter for perf
mfeliz-cruise Sep 5, 2023
5bf7711
Improve performance of converted layer_norm ops
mfeliz-cruise Sep 5, 2023
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
59 changes: 12 additions & 47 deletions core/conversion/converters/impl/layer_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,60 +88,25 @@ auto layer_norm_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns()
return true;
}

// Remove batch dimension from input shape for expand_size, which will
// be used to create weights for addScaleNd later.
auto expand_size = shape;
expand_size.erase(expand_size.begin(), expand_size.begin() + 1);

// Set up gamma_weights and beta_weights from gamma_expand and
// beta_expand.
auto gamma_weights = Weights(ctx, at::ones(expand_size));
auto beta_weights = Weights(ctx, at::zeros(expand_size));
auto normalized = div_out;

// gamma
if (args[2].IValue()->isTensor()) {
torch::Tensor gamma;
gamma = args[2].unwrapToTensor();
auto gamma_expand = gamma.expand(expand_size);
gamma_weights = Weights(ctx, gamma_expand);
} else {
gamma_weights = Weights(ctx, at::ones(expand_size));
auto gamma = args[2].ITensorOrFreeze(ctx);
auto gamma_prod = add_elementwise(
ctx, nvinfer1::ElementWiseOperation::kPROD, normalized, gamma, (util::node_info(n) + "_gamma").c_str());
normalized = gamma_prod->getOutput(0);
}

// beta
if (args[3].IValue()->isTensor()) {
torch::Tensor beta;
beta = args[3].unwrapToTensor();
auto beta_expand = beta.expand(expand_size);
beta_weights = Weights(ctx, beta_expand);
} else {
beta_weights = Weights(ctx, at::zeros(expand_size));
auto beta = args[3].ITensorOrFreeze(ctx);
auto beta_sum = add_elementwise(
ctx, nvinfer1::ElementWiseOperation::kSUM, normalized, beta, (util::node_info(n) + "_beta").c_str());
normalized = beta_sum->getOutput(0);
}

auto power = Weights(ctx, at::ones(expand_size));

auto gamma_tensor = ctx->net->addConstant(gamma_weights.shape, gamma_weights.data)->getOutput(0);
auto scale_l = add_elementwise(
ctx, nvinfer1::ElementWiseOperation::kPROD, div_out, gamma_tensor, (util::node_info(n) + "_scale").c_str());

auto beta_tensor = ctx->net->addConstant(beta_weights.shape, beta_weights.data)->getOutput(0);
auto shift_l = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kSUM,
scale_l->getOutput(0),
beta_tensor,
(util::node_info(n) + "_shift").c_str());

auto power_tensor = ctx->net->addConstant(power.shape, power.data)->getOutput(0);
auto power_l = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPOW,
shift_l->getOutput(0),
power_tensor,
(util::node_info(n) + "_power").c_str());

power_l->setName((util::node_info(n) + "_scale_nd").c_str());
auto power_l_out = power_l->getOutput(0);

ctx->AssociateValueAndTensor(n->outputs()[0], power_l_out);
ctx->AssociateValueAndTensor(n->outputs()[0], normalized);
return true;
}});

Expand Down