NNPACK build error

[kaishijeng]

After python configure.py, I got the following error with ninja

ninja: error: build.ninja:271: unknown pool name

DO you know why?

[nicolasvasilache]

This is likely related to an older ninja version (pre 1.4.0 I think).
Try using the latest: https://ninja-build.org/, it worked for me.

[kaishijeng]

It works with the latest ninja.

Thanks

[kaishijeng]

How do I run benchmark after compile test since it is not mentioned in the
README?

On Fri, Mar 25, 2016 at 8:07 AM, Nicolas Vasilache <notifications@github.com

wrote:

This is likely related to an older ninja version (pre 1.4.0 I think).
Try using the latest: https://ninja-build.org/, it worked for me.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#3 (comment)

[Maratyszcza]

There are three ways to benchmark NNPACK:

1. Run bin/convolution-benchmark, bin/fully-connected-benchmark, or bin/pooling-benchmark benchmarks to benchmark layers with arbitrary parameters. E.g.

   bin/convolution-benchmark --mode output --batch 128 --input-channels 1024 --output-channels 1024 --input-size 40 40 --kernel-size 5 5 --input-padding 2 --algorithm ft16x16

2. Run ./benchmark.py script to benchmark layers of AlexNet, OverFeat-Fast, or VGG-A models. E.g.

   ./benchmark.py -l fully-connected -m inference -n vgg-a

3. Build Caffe-NNPACK and use CPU branch of convnet-benchmarks to benchmark NNPACK integrated into Caffe. You'd need to modify models' prototxt files to add engine: NNPACK inside convolution_param of the convolutional layers.

[kaishijeng]

Seem something is wrong when I tried

bin/convolution-benchmark --mode output --batch 128 --input-channels 1024
--output-channels 1024 --input-size 40 40 --kernel-size 5 5 --input-padding
2 --algorithm ft16x16

It gives the following error:

NNPACK initialization failed: error code 51

On Sat, Mar 26, 2016 at 8:00 PM, Marat Dukhan notifications@github.com
wrote:

There are three ways to benchmark NNPACK:

1. Run bin/convolution-benchmark, bin/fully-connected-benchmark, or
   bin/pooling-benchmark benchmarks to benchmark layers with arbitrary
   parameters. E.g.

bin/convolution-benchmark --mode output --batch 128 --input-channels 1024 --output-channels 1024 --input-size 40 40 --kernel-size 5 5 --input-padding 2 --algorithm ft16x16

1. Run ./benchmark.py script to benchmark layers of AlexNet,
   OverFeat-Fast, or VGG-A models. E.g.

./benchmark.py -l fully-connected -m inference -n vgg-a

1. Build Caffe-NNPACK https://github.com/Maratyszcza/caffe-nnpack
   and use CPU branch of convnet-benchmarks
   https://github.com/soumith/convnet-benchmarks/tree/cpu to benchmark
   NNPACK integrated into Caffe. You'd need to modify models' prototxt files
   to add engine: NNPACK inside convolution_param of the convolutional
   layers.

—
You are receiving this because you modified the open/close state.
Reply to this email directly or view it on GitHub
#3 (comment)

[Maratyszcza]

This error code means you are not running on a compatible hardware. NNPACK requires a processor with AVX2, i.e. Intel Haswell, Broadwell, or Skylake. It also checks that the processor has L1, L2, and L3 cache, but AFAIK also current Haswell/Broadwell/Skylake SKUs have it.

[kaishijeng]

Thanks

On Sat, Mar 26, 2016 at 8:55 PM, Marat Dukhan notifications@github.com
wrote:

This error code means you are not running on a compatible hardware. NNPACK
requires a processor with AVX2, i.e. Intel Haswell, Broadwell, or Skylake.
It also checks that the processor has L1, L2, and L3 cache, but AFAIK also
current Haswell/Broadwell/Skylake SKUs have it.

—
You are receiving this because you modified the open/close state.
Reply to this email directly or view it on GitHub
#3 (comment)