Add configurable precision support of print().

[AtomicVar]

This PR aims to implement what #495 needs. I mimic NumPy's numpy.set_printoptions() - it allows users to set a precision. So I added a similar function set_print_precision() into tensor_utils.h.

Example usage:

cudaStream_t stream;
cudaStreamCreate(&stream);

tensor_t<float, 2> x({2, 3});
(x = random<float>({2, 3}, NORMAL)).run(stream);
cudaStreamSynchronize(stream);

printf("x:\n");
print(x);

printf("Current precision: %u.\n", get_print_precision());
set_print_precision(6);
printf("Current precision: %u.\n", get_print_precision());

printf("x:\n");
print(x);

Output:

x:
Tensor{float} Rank: 2, Sizes:[2, 3], Strides:[3,1]
000000: -9.2466e-01 -4.2534e-01 -2.6438e+00 
000001: 1.4518e-01 -1.2087e-01 -5.7973e-01 
Current precision: 4.
Current precision: 6.
x:
Tensor{float} Rank: 2, Sizes:[2, 3], Strides:[3,1]
000000: -9.246624e-01 -4.253442e-01 -2.643846e+00 
000001: 1.451839e-01 -1.208664e-01 -5.797257e-01

[cliffburdick]

/blossom-ci

[cliffburdick]

Thanks for the PR @AtomicVar! It looks like it failed the build test:

[ 77%] Linking CUDA executable matx_bench
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/fft.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/fft.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/conv.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/conv.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/cub.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/cub.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/einsum.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/einsum.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/svd_power.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/svd_power.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/qr.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_desc.h:289: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_transform/qr.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_desc.h:289: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_operators/operators.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_operators/operators.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_operators/reduction.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/00_operators/reduction.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/01_radar/SingleChanSimplePipeline.cu.o: in function `matx::set_print_precision(unsigned int)':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: multiple definition of `matx::set_print_precision(unsigned int)'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:923: first defined here
/usr/bin/ld: CMakeFiles/matx_bench.dir/01_radar/SingleChanSimplePipeline.cu.o: in function `matx::get_print_precision()':
/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: multiple definition of `matx::get_print_precision()'; CMakeFiles/matx_bench.dir/00_transform/matmul.cu.o:/home/jenkins/workspace/unit-tests/include/matx/core/tensor_utils.h:932: first defined here

My guess is you're missing MATX_INLINE.

[cliffburdick]

/blossom-ci

[cliffburdick]

Thanks @AtomicVar !