Skip to content

Instantly share code, notes, and snippets.

Avatar

Marcio M Pereira iviarcio

View GitHub Profile
@iviarcio
iviarcio / resnet50-16.lir
Last active Jul 4, 2021
resnet50 low-level intemediate representation compiled with int16 quantization
View resnet50-16.lir
function resnet50
declare {
%gpu_0_pred_b = WeightVar i16[S:0.000001907 O:0][-0.062,0.062]<1000> const // size: 2000 // Users: @in 205
%gpu_0_pred_w__2 = WeightVar i16[S:0.000030518 O:0][-1.000,1.000]<2048 x 1000> const // size: 4096000 // Users: @in 205
%gpu_0_conv1_bias_constfold = WeightVar i16[S:0.000244141 O:0][-8.000,8.000]<64> const // size: 128 // Users: @in 3
%gpu_0_res2_0_branch2c_bias_constfold = WeightVar i16[S:0.000122070 O:0][-4.000,4.000]<256> const // size: 512 // Users: @in 154
%gpu_0_res3_0_branch2a_bias_constfold = WeightVar i16[S:0.000122070 O:0][-4.000,4.000]<128> const // size: 256 // Users: @in 85
%gpu_0_res3_0_branch2c_bias_constfold = WeightVar i16[S:0.000122070 O:0][-4.000,4.000]<512> const // size: 1024 // Users: @in 192
%gpu_0_res4_0_branch2c_bias_constfold = WeightVar i16[S:0.000122070 O:0][-4.000,4.000]<1024> const // size: 2048 // Users: @in 157
%gpu_0_res5_0_branch2c_bias_constfold = WeightVar i16[S:0.000244141 O:0][-8.000,8.000]<2048> const // size: 4096 // Use
@iviarcio
iviarcio / resnet50.lir
Created Jun 26, 2021
resnet50 low level intermediate representation for NMP device
View resnet50.lir
function resnet50
declare {
%gpu_0_pred_b = WeightVar i8[S:0.000976562 O:0][-0.125,0.124]<1000> const // size: 1000 // Users: @in 205
%gpu_0_pred_w__2 = WeightVar i8[S:0.007812500 O:0][-1.000,0.992]<2048 x 1000> const // size: 2048000 // Users: @in 205
%gpu_0_conv1_bias_constfold = WeightVar i8[S:0.062500000 O:0][-8.000,7.938]<64> const // size: 64 // Users: @in 3
%gpu_0_res2_0_branch2c_bias_constfold = WeightVar i8[S:0.031250000 O:0][-4.000,3.969]<256> const // size: 256 // Users: @in 154
%gpu_0_res3_0_branch2a_bias_constfold = WeightVar i8[S:0.031250000 O:0][-4.000,3.969]<128> const // size: 128 // Users: @in 85
%gpu_0_res3_0_branch2c_bias_constfold = WeightVar i8[S:0.031250000 O:0][-4.000,3.969]<512> const // size: 512 // Users: @in 192
%gpu_0_res4_0_branch2c_bias_constfold = WeightVar i8[S:0.031250000 O:0][-4.000,3.969]<1024> const // size: 1024 // Users: @in 157
%gpu_0_res5_0_branch2c_bias_constfold = WeightVar i8[S:0.062500000 O:0][-8.000,7.938]<2048> const // size: 2048 // Users: @in 19