[GitHub] [incubator-tvm] alexgl-github commented on pull request #6840: conv1d_transpose speedup

[GitBox <gi...@apache.org>]

alexgl-github commented on pull request #6840:
URL: https://github.com/apache/incubator-tvm/pull/6840#issuecomment-721412452


   Speedup transposed 1d convolution by eliminating unnecessary multiplications by data values which contain zeroes, for stride greater then 1. 
   
   Below are current vs proposed latency numbers for various transposed conv1d parameters:
   `
      current:                new:
   latency=0.06837 sec  latency=0.00049 sec   channels_out=   1 input_shape=(1, 257, 128) kernel_size=512 strides=128 padding=256
   latency=0.53093 sec  latency=0.10341 sec   channels_out= 257 input_shape=(1, 257, 128) kernel_size=512 strides=4 padding=256
   latency=0.00292 sec  latency=0.00307 sec   channels_out=   1 input_shape=(1, 257, 128) kernel_size=512 strides=1 padding=256
   latency=0.00474 sec  latency=0.00171 sec   channels_out=   1 input_shape=(1, 257, 128) kernel_size=512 strides=2 padding=256
   latency=0.00955 sec  latency=0.00056 sec   channels_out=   1 input_shape=(1, 257, 128) kernel_size=512 strides=16 padding=256
   latency=0.00054 sec  latency=0.00023 sec   channels_out=   1 input_shape=(1, 1, 16384) kernel_size=512 strides=2 padding=256
   latency=0.00385 sec  latency=0.00089 sec   channels_out=   4 input_shape=(1, 1, 16384) kernel_size=512 strides=4 padding=256
   latency=0.00013 sec  latency=0.00004 sec   channels_out=   1 input_shape=(1, 1, 1024) kernel_size=512 strides=5 padding=256
   latency=0.00002 sec  latency=0.00002 sec   channels_out=  32 input_shape=(1, 3, 224) kernel_size=5 strides=1 padding=0
   latency=0.00004 sec  latency=0.00003 sec   channels_out=  32 input_shape=(1, 3, 224) kernel_size=5 strides=2 padding=0
   latency=0.00006 sec  latency=0.00003 sec   channels_out= 128 input_shape=(1, 32, 32) kernel_size=5 strides=2 padding=0
   
   `
     


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org