hptt
hptt copied to clipboard
springer13
High-Performance Tensor Transpose library
The C interface does not compile with recent versions of Clang. This PR provides a fix for that.
FreeBSD uses powerpc64le name for 64-bit POWER on little endian.
``` g++10 -O3 -std=c++11 -I../src/ -c ../benchmark/reference.cpp -o ../benchmark/reference.o ../benchmark/reference.cpp: In instantiation of 'void transpose_ref(uint32_t*, uint32_t*, int, const floatType*, floatType, floatType*, floatType, bool) [with floatType = float; uint32_t = unsigned...
[here](https://github.com/springer13/hptt/blob/master/CMakeLists.txt#L40) If ```STATIC``` is removed - cmake's variable [BUILD_SHARED_LIBS](https://cmake.org/cmake/help/latest/variable/BUILD_SHARED_LIBS.html) changeable by the user would define the library type.
Build fails with clang-13: ``` ===> Building for hptt-1.0.5.18 [ 20% 4/5] /usr/bin/c++ -I/disk-samsung/freebsd-ports/math/hptt/work/hptt-1.0.5-18-g9425386/include -O2 -pipe -fno-omit-frame-pointer -fstack-protector-strong -fno-strict-aliasing -fno-omit-frame-pointer -O2 -pipe -fno-omit-frame-pointer -fstack-protector-strong -fno-strict-aliasing -fno-omit-frame-pointer -fopenmp -march=native -std=gnu++11 -MD...
Hello, I tried to do a tensor transpose from a 3x3x3 tensor into a 3x3x3 sub-tensor of a 5x5x5 tensor, but the result is unexpected. The following code snippet is...
When I compile the benchmark and reference files by operating `make` in the `/benchmark` folder, I get the following error: `reference.cpp:60:30: error: cannot convert 'std::complex' to 'float' in assignment 60...
Hello, I compiled the benchmark using the Makefile, but then i got an error, when I tried to run the exe. "Error while loading shared libraries: libhptt.so: cannot open shared...
when something like dimension is not correct, hptt exit(-1) directly, it is not good for debugging, it would be better if change them to throw error.
