masahi

	.text
	.file	"fused_nn_contrib_conv2d_NCHWc_add_cast_multiply_add_right_shift_add_cast_clip_ca_10134105474844747540__1"
	.globl	fused_nn_contrib_conv2d_NCHWc_add_cast_multiply_add_right_shift_add_cast_clip_ca_10134105474844747540__1
	.p2align	4, 0x90
	.type	fused_nn_contrib_conv2d_NCHWc_add_cast_multiply_add_right_shift_add_cast_clip_ca_10134105474844747540__1,@function
fused_nn_contrib_conv2d_NCHWc_add_cast_multiply_add_right_shift_add_cast_clip_ca_10134105474844747540__1:
.Lfunc_begin0:
	.cfi_startproc
	pushq	%rbp
	.cfi_def_cfa_offset 16

masahi

import numpy as np
import torch
import tvm
from tvm import relay
from torchvision import models
from tvm.relay.frontend.pytorch import from_pytorch, get_graph_input_names


class SegmentationModelWrapper(torch.nn.Module):
    def __init__(self, model):

masahi

graph(%self : __torch__.torch.nn.modules.module.Module,
      %input : Tensor):
  %2 : int[] = prim::Constant[value=[1, 1]]()
  %3 : bool = prim::Constant[value=0]() # /home/masa/projects/deep/pytorch/torch/nn/modules/conv.py:345:0
  %4 : int[] = prim::Constant[value=[0, 0]]()
  %5 : int = prim::Constant[value=16]() # /home/masa/projects/deep/pytorch/torch/nn/modules/conv.py:345:0
  %6 : bool = prim::Constant[value=1]() # /home/masa/projects/deep/pytorch/torch/nn/modules/conv.py:345:0
  %7 : int = prim::BailoutTemplate_0()
  %8 : Float(1, 16, 224, 224) = prim::BailOut[index=0](%7, %input, %self)
  %9 : Tensor = prim::GetAttr[name="bias"](%self)

masahi

import numpy as np
import torch
import tvm
from tvm import relay
from torchvision import models

from torch_frontend import parse_script_module


class SegmentationModelWrapper(torch.nn.Module):

masahi

import torch
import torchvision
import numpy as np


def do_script(model, in_size=100):
    model_script = torch.jit.script(model)
    model_script.eval()
    return model_script

masahi

fn (%X: Tensor[(10, 20), float32]) -> Tensor[(10, 20), float32] {
  %0 = full(0 /* ty=int32 */, shape=[10, 20], dtype="float32") /* ty=Tensor[(10, 20), float32] */;
  %4 = (
    let %while_loop: fn (int32, Tensor[(10, 20), float32]) -> (int32, Tensor[(10, 20), float32]) = fn (%i: int32, %a.6: Tensor[(10, 20), float32]) -> (int32, Tensor[(10, 20), float32]) {
      %1 = less(%i, 10 /* ty=int32 */) /* ty=bool */;
      if (%1) {
        %2 = add(%i, 1 /* ty=int32 */) /* ty=int32 */;
        %3 = add(%a.6, %X) /* ty=Tensor[(10, 20), float32] */;
        %while_loop(%2, %3) /* ty=(int32, Tensor[(10, 20), float32]) */
      } else {

masahi

before: v0.0.4
def @main(%x: Tensor[(2), float32]) -> Tensor[(2), float32] {
  let %while_loop: fn (int32, Tensor[(2), float32]) -> Tensor[(2), float32] = fn (%i: int32, %s: Tensor[(2), float32]) -> Tensor[(2), float32] {
    %0 = less(%i, 10 /* ty=int32 */) /* ty=bool */;
    if (%0) {
      %1 = add(%i, 1 /* ty=int32 */) /* ty=int32 */;
      %2 = add(%s, %x) /* ty=Tensor[(2), float32] */;
      %while_loop(%1, %2) /* ty=Tensor[(2), float32] */
    } else {
      %s

masahi

import os
from packaging import version
import numpy as np
import tvm
from tvm import relay
import torch
from torch import nn
from torch.quantization.observer import MovingAverageMinMaxObserver, default_weight_observer
from torchvision.models.quantization import resnet as qresnet
from torchvision.models.quantization import mobilenet as qmobilenet

masahi

import os
import numpy as np
import tvm
from tvm import relay
import torch
from torch.quantization.observer import MovingAverageMinMaxObserver, default_weight_observer
# from torchvision.models.quantization import shufflenetv2 as qshufflenetv2
from torch import nn
from torch.quantization import QuantStub, DeQuantStub, fuse_modules
from torch_frontend import parse_script_module

masahi

import os
import numpy as np
import tvm
from tvm import relay
import torch


from torch_frontend import parse_script_module