I wonder if it is possible to "strip" a general-use .exe/.dll/.so into a arch-specific version.

Accelerated Computing CUDA CUDA Programming and Performance
1

For me, it tooks about 10s to execute

import mxnet as mx

in my python 3.6.6. The slow import procedure may due to the large mxnet.dll file, which is generated by

	KNOWN_CUDA_ARCHS := 30 35 50 52 60 61 70 75
	# Run nvcc on a zero-length file to check architecture-level support.
	# Create args to include SASS in the fat binary for supported levels.
	CUDA_ARCH := $(foreach arch,$(KNOWN_CUDA_ARCHS), \
				$(shell $(NVCC) -arch=sm_$(arch) -E --x cu /dev/null >/dev/null 2>&1 && \
						echo -gencode arch=compute_$(arch),code=sm_$(arch)))
	# Convert a trailing "code=sm_NN" to "code=[sm_NN,compute_NN]" to also
	# include the PTX of the most recent arch in the fat-binaries for
	# forward compatibility with newer GPUs.
	CUDA_ARCH := $(shell echo $(CUDA_ARCH) | sed 's/sm_\([0-9]*\)$$/[sm_\1,compute_\1]/')
	# Add fat binary compression if supported by nvcc.
	COMPRESS := --fatbin-options -compress-all
	CUDA_ARCH += $(shell $(NVCC) -cuda $(COMPRESS) --x cu /dev/null -o /dev/null >/dev/null 2>&1 && \
						 echo $(COMPRESS))

These terrible instructions came up with a REALLY HUGE .dll file, mxnet.dll
Now, the question is, can we strip it for a specific arch, e.g., sm_61,compute_61 for my GTX 1060?

2

KNOWN_CUDA_ARCHS := 61

3

Thanks for your reply. You are absolutely right, it may works if the compile routing is fine.
But… compiling from the source code is always not the best choice for Windows users(-_-)
I want to know if there is some tools, such as strip in linux, could delete useless information and reduce the size of .so/.dll/.exe files.

4

[url]CUDA Binary Utilities :: CUDA Toolkit Documentation

5

Thank you!!!
This program is exactly what I need
…Although libmxnet is not relocatable

I had commited an issue in github.
Perhaps we will recieve a thin libmxnet in the future.

…
I found that I have both libmxnet.dll without symbols, and a libmxnet.lib that contains symbols according to the .dll file.
The question is that nvprune could only accept a single input file, and I don’t know how to combine them together.

Any sugestion?
Thanks.