bench_approx.cpp File Reference

#include <benchmark/benchmark.h>
#include <torch/csrc/jit/tensorexpr/ir_simplifier.h>
#include <torch/csrc/jit/tensorexpr/loopnest.h>
#include <torch/csrc/jit/tensorexpr/tensor.h>
#include <torch/csrc/jit/tensorexpr/llvm_codegen.h>
#include <torch/torch.h>
#include "caffe2/operators/tanh_op.h"
#include "caffe2/operators/logit_op.h"

Include dependency graph for bench_approx.cpp:

Go to the source code of this file.

Functions
void	optimizeSleef (tensorexpr::LoopNest *ln, tensorexpr::Tensor *target)
void	optimizePointwise (tensorexpr::LoopNest *ln, tensorexpr::Tensor *target)
static void	log_nnc_sleef (benchmark::State &state)
static void	log_nnc_fast (benchmark::State &state)
static void	log_aten (benchmark::State &state)
static void	logit_nnc_sleef (benchmark::State &state)
static void	logit_nnc_fast (benchmark::State &state)
static void	logit_aten (benchmark::State &state)
template<typename T >
void	logit_caffe2_impl (int size, const T *X, T *Y, float eps_=1e-6f)
static void	logit_caffe2 (benchmark::State &state)
static void	tanh_nnc_fast (benchmark::State &state)
static void	tanh_aten (benchmark::State &state)
static void	tanh_caffe2 (benchmark::State &state)
	Args ({2<< 5}) -> Args({2<< 8}) ->Args({2<< 12}) ->Args({2<< 14})

Function Documentation

Args()

Args

(

{2<< 5}

)

-> Args({2<< 8}) ->Args({2<< 12}) ->Args({2<< 14})

Referenced by torch::autograd::Function< T >::apply(), c10::impl::boxArgs(), c10::TypedOperatorHandle< Return(Args...)>::call(), c10::KernelFunction::call(), c10::Dispatcher::call(), c10::impl::BoxedKernelWrapper< Result(Args...), std::enable_if_t< can_box_all< Args... >::value &&is_tuple_of_mutable_tensor_refs< Result >::value &&!std::is_same< Result, guts::typelist::to_tuple_t< guts::typelist::take_t< guts::typelist::typelist< Args... >, sizeof...(Args) >=std::tuple_size< Result >::value ? std::tuple_size< Result >::value :sizeof...(Args) > > >::value, void >::call(), c10::impl::BoxedKernelWrapper< Result(Args...), std::enable_if_t< can_box_all< Args... >::value &&is_tuple_of_mutable_tensor_refs< Result >::value &&std::is_same< Result, guts::typelist::to_tuple_t< guts::typelist::take_t< guts::typelist::typelist< Args... >, sizeof...(Args) >=std::tuple_size< Result >::value ? std::tuple_size< Result >::value :sizeof...(Args) > > >::value, void >::call(), c10::callUnboxedKernelFunction(), c10::Dispatcher::callWithDispatchKey(), c10::TypedOperatorHandle< Return(Args...)>::callWithDispatchKey(), c10::Dispatcher::callWithDispatchKeySlowPath(), c10::optional< T >::emplace(), c10::SmallVectorImpl< T >::emplace_back(), torch::jit::TuplePacker< remaining, Args >::execute(), c10::guts::filter_map(), c10::generic_to(), at::native::metal::mpscnn::if(), torch::jit::pack(), and c10::Dispatcher::redispatch().

log_aten()

static void log_aten ( benchmark::State &  state )

static

Definition at line 90 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, c10::prim::abs(), at::native::log_out(), at::native::randn(), and state.

log_nnc_fast()

static void log_nnc_fast ( benchmark::State &  state )

static

Definition at line 60 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, caffe2::A, c10::prim::abs(), at::native::allclose(), compare-fastrnn-results::args, B, torch::jit::tensorexpr::Compute(), at::Tensor::data_ptr(), torch::jit::tensorexpr::fast_log(), torch::jit::i, c10::kFloat, c10::kInt, at::log(), N, optimizePointwise(), at::native::randn(), operator_benchmark.common.repeat_benchmark::s, torch::jit::tensorexpr::IRSimplifier::simplify(), state, and TORCH_CHECK.

log_nnc_sleef()

static void log_nnc_sleef ( benchmark::State &  state )

static

Definition at line 30 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, caffe2::A, c10::prim::abs(), at::native::allclose(), compare-fastrnn-results::args, B, torch::jit::tensorexpr::Compute(), at::Tensor::data_ptr(), torch::jit::i, c10::kFloat, c10::kInt, at::log(), caffe2.contrib.playground.AnyExp::log, N, optimizeSleef(), torch::jit::tensorexpr::LoopNest::prepareForCodegen(), at::native::randn(), operator_benchmark.common.repeat_benchmark::s, torch::jit::tensorexpr::IRSimplifier::simplify(), state, and TORCH_CHECK.

logit_aten()

static void logit_aten ( benchmark::State &  state )

static

Definition at line 174 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, c10::prim::abs(), c10::aten::clamp(), at::native::logit_out(), at::native::randn(), and state.

logit_caffe2()

static void logit_caffe2 ( benchmark::State &  state )

static

Definition at line 195 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, c10::prim::abs(), at::native::allclose(), c10::aten::clamp(), at::Tensor::data_ptr(), logit_caffe2_impl(), at::native::logit_out(), N, at::native::nan_to_num(), at::native::randn(), state, TORCH_CHECK, X, and at::native::metal::mpscnn::Y.

logit_caffe2_impl()

template<typename T >

void logit_caffe2_impl	(	int	size,
		const T *	X,
		T *	Y,
		float	eps_ = 1e-6f
	)

Definition at line 186 of file bench_approx.cpp.

References size, X, and at::native::metal::mpscnn::Y.

Referenced by logit_caffe2().

logit_nnc_fast()

static void logit_nnc_fast ( benchmark::State &  state )

static

Definition at line 137 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, caffe2::A, c10::prim::abs(), at::native::allclose(), compare-fastrnn-results::args, B, c10::aten::clamp(), torch::jit::tensorexpr::Compute(), at::Tensor::data_ptr(), torch::jit::tensorexpr::fast_log(), torch::jit::i, c10::kFloat, torch::jit::tensorexpr::kGT, c10::kInt, torch::jit::tensorexpr::kLT, at::native::logit(), max, min, N, at::native::nan_to_num(), optimizePointwise(), torch::jit::tensorexpr::LoopNest::prepareForCodegen(), at::native::randn(), operator_benchmark.common.repeat_benchmark::s, torch::jit::tensorexpr::IRSimplifier::simplify(), state, and TORCH_CHECK.

logit_nnc_sleef()

static void logit_nnc_sleef ( benchmark::State &  state )

static

Definition at line 100 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, caffe2::A, c10::prim::abs(), at::native::allclose(), compare-fastrnn-results::args, B, c10::aten::clamp(), torch::jit::tensorexpr::Compute(), at::Tensor::data_ptr(), torch::jit::i, c10::kFloat, torch::jit::tensorexpr::kGT, c10::kInt, torch::jit::tensorexpr::kLT, caffe2.contrib.playground.AnyExp::log, at::native::logit(), max, min, N, at::native::nan_to_num(), optimizePointwise(), torch::jit::tensorexpr::LoopNest::prepareForCodegen(), at::native::randn(), operator_benchmark.common.repeat_benchmark::s, torch::jit::tensorexpr::IRSimplifier::simplify(), state, and TORCH_CHECK.

optimizePointwise()

void optimizePointwise	(	tensorexpr::LoopNest *	ln,
		tensorexpr::Tensor *	target
	)

Definition at line 20 of file bench_approx.cpp.

References torch::jit::tensorexpr::LoopNest::getLoopStmtsFor(), at::native::inner(), loops, c10::aten::outer(), torch::jit::tensorexpr::LoopNest::splitWithTail(), caffe2::tail, caffe2::target(), caffe2.perfkernels.hp_emblookup_codegen::unroll(), and torch::jit::tensorexpr::LoopNest::vectorize().

Referenced by log_nnc_fast(), logit_nnc_fast(), logit_nnc_sleef(), and tanh_nnc_fast().

optimizeSleef()

void optimizeSleef	(	tensorexpr::LoopNest *	ln,
		tensorexpr::Tensor *	target
	)

Definition at line 13 of file bench_approx.cpp.

References torch::jit::tensorexpr::LoopNest::getLoopStmtsFor(), at::native::inner(), loops, c10::aten::outer(), torch::jit::tensorexpr::LoopNest::splitWithTail(), caffe2::tail, caffe2::target(), and torch::jit::tensorexpr::LoopNest::vectorize().

Referenced by log_nnc_sleef().

tanh_aten()

static void tanh_aten ( benchmark::State &  state )

static

Definition at line 245 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, c10::prim::abs(), at::native::randn(), state, and at::native::tanh_out().

tanh_caffe2()

static void tanh_caffe2 ( benchmark::State &  state )

static

Definition at line 255 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, c10::prim::abs(), at::native::allclose(), c, at::Tensor::data_ptr(), N, at::native::randn(), state, at::native::tanh(), at::native::tanh_out(), TORCH_CHECK, X, and at::native::metal::mpscnn::Y.

tanh_nnc_fast()

static void tanh_nnc_fast ( benchmark::State &  state )

static

Definition at line 215 of file bench_approx.cpp.

References caffe2.contrib.aten.gen_op::_, caffe2::A, c10::prim::abs(), at::native::allclose(), compare-fastrnn-results::args, B, torch::jit::tensorexpr::Compute(), at::Tensor::data_ptr(), torch::jit::tensorexpr::fast_tanh(), torch::jit::i, c10::kFloat, c10::kInt, N, optimizePointwise(), at::native::randn(), operator_benchmark.common.repeat_benchmark::s, torch::jit::tensorexpr::IRSimplifier::simplify(), state, at::native::tanh(), and TORCH_CHECK.