ailzhang

import torch
import torch.nn as nn
import time
import subprocess
import argparse
import numpy as np
from torch.autograd import Variable
import pdb
def linearforward(batchsize, dim_in, dim_out):
    data = np.random.random_sample([batchsize, dim_in])

ailzhang

Torch level 1

function	Symbolic_implemented
gather
equal
and
iand
or
ior
xor

ailzhang

Torch level 1: Uncategorized

function	Symbolic_implemented
pin_memory
size
stride
to
storage_offset
set_
is_contiguous

ailzhang

// This file contains all native_functions that can be registered to
// and the schema string that they should be registered with

Tensor __and__(const Tensor & self, Scalar other); // aten::__and__.Scalar(Tensor self, Scalar other) -> Tensor
Tensor __and__(const Tensor & self, const Tensor & other); // aten::__and__.Tensor(Tensor self, Tensor other) -> Tensor
Tensor & __iand__(Tensor & self, Scalar other); // aten::__iand__.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)
Tensor & __iand__(Tensor & self, const Tensor & other); // aten::__iand__.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!)
Tensor & __ilshift__(Tensor & self, Scalar other); // aten::__ilshift__.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)
Tensor & __ilshift__(Tensor & self, const Tensor & other); // aten::__ilshift__.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!)
Tensor & __ior__(Tensor & self, Scalar other); // aten::__ior__.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

ailzhang

import torch
import math
import torch.nn.functional as F
import torch_xla


def test_conv2d(batch, input_shape, kernel, Cin, Cout, stride, padding, dilation, groups, device):
    input = torch.ones(batch, Cin, input_shape, input_shape, dtype=torch.float, device=device).requires_grad_()
    weight = torch.ones(Cout, int(Cin / groups), kernel, kernel, dtype=torch.float, device=device).requires_grad_()
    out = F.conv2d(input, weight, stride=stride, padding=padding, dilation=dilation, groups=groups).to(device)

ailzhang

2019-09-04 20:04:55.629680: I tensorflow/compiler/xla/xla_client/xrt_computation_client.cc:196] XRT device (LOCAL) CPU:0 -> /job:localservice/replica:0/task:0/device:XLA_CPU:0
2019-09-04 20:04:55.629791: I tensorflow/compiler/xla/xla_client/xrt_computation_client.cc:200] Worker grpc://localhost:40934 for /job:localservice/replica:0/task:0
2019-09-04 20:04:55.629800: I tensorflow/compiler/xla/xla_client/xrt_computation_client.cc:204] XRT default device: CPU:0
2019-09-04 20:04:55.629828: I tensorflow/compiler/xla/xla_client/xrt_local_service.cc:40] Peer localservice 1 {localhost:40934}
2019-09-04 20:04:55.629959: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2019-09-04 20:04:55.655453: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2300000000 Hz
2019-09-04 20:04:55.661129: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x7f63e4000b50 executing computation

ailzhang

c10::optional<std::function<at::Tensor(const at::Tensor&)>> func=c10::nullopt;
if (!self.unsafeGetTensorImpl()->support_as_strided()) {
  func = [=](const at::Tensor& input_base) {
    return at::alias(input_base);
  };
}

c10::optional<std::function<at::Tensor(const at::Tensor&)>> func=c10::nullopt;
if (!self.unsafeGetTensorImpl()->support_as_strided()) {
  auto size_vec = size.vec();

ailzhang

0.Program arguments: /usr/lib/llvm-9/bin/clang -cc1 -triple x86_64-pc-linux-gnu -emit-obj -disable-free -disable-llvm-verifier -discard-value-names -main-file-name SobolEngineOps.cpp -mrelocation-model pic -pic-level 2 -mthread-mod
el posix -fno-trapping-math -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -momit-leaf-frame-pointer -coverage-notes-file /var/lib/jenkins/pytorch/bui
ld/caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/native/SobolEngineOps.cpp.gcno -resource-dir /usr/lib/llvm-9/lib/clang/9.0.0 -isystem /var/lib/jenkins/pytorch/build/third_party/gloo -isystem /var/lib/jenkins/pytorch/cmake/../th
ird_party/gloo -isystem /var/lib/jenkins/pytorch/cmake/../third_party/googletest/googlemock/include -isystem /var/lib/jenkins/pytorch/cmake/../third_party/googletest/googletest/include -isystem /var/lib/jenkins/pytorch/third_party/
protobuf/src -isystem /opt/conda/include -isystem /var/lib/jenkins/pytorch/third_party/gemmlowp

ailzhang

0.Program arguments: /usr/lib/llvm-9/bin/clang -cc1 -triple x86_64-pc-linux-gnu -emit-obj -disable-free -disable-llvm-verifier -discard-value-names -main-file-name SobolEngineOps.cpp -mrelocation-model pic -pic-level 2 -mthread-mod
el posix -fno-trapping-math -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -momit-leaf-frame-pointer -coverage-notes-file /var/lib/jenkins/pytorch/bui
ld/caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/native/SobolEngineOps.cpp.gcno -resource-dir /usr/lib/llvm-9/lib/clang/9.0.0 -isystem /var/lib/jenkins/pytorch/build/third_party/gloo -isystem /var/lib/jenkins/pytorch/cmake/../th
ird_party/gloo -isystem /var/lib/jenkins/pytorch/cmake/../third_party/googletest/googlemock/include -isystem /var/lib/jenkins/pytorch/cmake/../third_party/googletest/googletest/include -isystem /var/lib/jenkins/pytorch/third_party/
protobuf/src -isystem /opt/conda/include -isystem /var/lib/jenkins/pytorch/third_party/gemmlowp

ailzhang

	jit_premul_bias + master	jit_premul_bias + Reduce time per guard
fuser=none executor=simple	13.53	12.79
fuser=none executor=profiling	19.52	13.88
fuser=te executor=profiling	13.44	12.14