Compiler errors for powerpc, arm, android, windows

[josopait]

I am getting compiler errors when cross-compiling for different platforms.

gcc-14/linux-arm64 (and the same for gcc-14/linux-powerpc64le):

In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754_longdouble.hpp:38, from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754.hpp:35, from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat_impl.hpp:24, from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat.hpp:63, from /home/goopax-build/goopax/lib/bits/vulkan/../gpu.h:18, from /home/goopax-build/goopax/lib/bits/vulkan/../kernelcore.h:5, from /home/goopax-build/goopax/lib/bits/vulkan/../env_implbase.h:8, from /home/goopax-build/goopax/lib/bits/vulkan/../device.h:7, from /home/goopax-build/goopax/lib/bits/vulkan/device.h:3, from /home/goopax-build/goopax/lib/bits/vulkan/device.cpp:1: /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp: In function 'std::string sw::universal::to_binary(long double, bool)': /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp:120:29: error: 'struct sw::universal::long_double_decoder::<unnamed>' has no member named 'bit63' 120 | s << (decoder.parts.bit63 ? '1' : '0'); | ^~~~~ /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp: In function 'std::string sw::universal::to_triple(long double)': /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp:154:29: error: 'struct sw::universal::long_double_decoder::<unnamed>' has no member named 'bit63' 154 | s << (decoder.parts.bit63 ? '1' : '0'); | ^~~~~ /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp: In function 'std::string sw::universal::color_print(long double)': /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp:198:40: error: 'struct sw::universal::long_double_decoder::<unnamed>' has no member named 'bit63' 198 | s << magenta << (decoder.parts.bit63 ? '1' : '0'); | ^~~~~

clang/android:

In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat.hpp:63: In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat_impl.hpp:24: In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754.hpp:35: In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754_longdouble.hpp:37: /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/clang_long_double.hpp:344:17: error: static assertion failed due to requirement 'sizeof(long double) == 8': This function only works when long double is cast to a 64 bit double 344 | static_assert(sizeof(long double) == 8, "This function only works when long double is cast to a 64 bit double"); | ^~~~~~~~~~~~~~~~~~~~~~~~ /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/clang_long_double.hpp:344:37: note: expression evaluates to '16 == 8' 344 | static_assert(sizeof(long double) == 8, "This function only works when long double is cast to a 64 bit double"); | ~~~~~~~~~~~~~~~~~~~~^~~~

gcc-14/windows:

In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754_longdouble.hpp:38, from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754.hpp:35, from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat_impl.hpp:24, from /home/goopax-build/goopax/ext/universal/include/sw/universal/number/cfloat/cfloat.hpp:63, from /home/goopax-build/goopax/lib/bits/vulkan/../gpu.h:18, from /home/goopax-build/goopax/lib/bits/vulkan/../kernelcore.h:5, from /home/goopax-build/goopax/lib/bits/vulkan/../env_implbase.h:8, from /home/goopax-build/goopax/lib/bits/vulkan/../device.h:7, from /home/goopax-build/goopax/lib/bits/vulkan/device.h:3, from /home/goopax-build/goopax/lib/bits/vulkan/device.cpp:1: /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/gcc_long_double.hpp:229:13: error: redefinition of 'void sw::universal::extract_fp_components(long double, bool&, int&, long double&, long long unsigned int&)' 229 | inline void extract_fp_components(long double fp, bool& _sign, int& _exponent, long double& _fr, unsigned long long& _fraction) { | ^~~~~~~~~~~~~~~~~~~~~ In file included from /home/goopax-build/goopax/ext/universal/include/sw/universal/native/ieee754_longdouble.hpp:33: /home/goopax-build/goopax/ext/universal/include/sw/universal/native/nonconstexpr/extract_fp_components.hpp:34:13: note: 'void sw::universal::extract_fp_components(long double, bool&, int&, long double&, uint64_t&)' previously defined here 34 | inline void extract_fp_components(long double fp, bool& _sign, int& _exponent, long double& _fr, std::uint64_t& _fraction) { | ^~~~~~~~~~~~~~~~~~~~~

[Ravenwater]

@josopait thanks for reporting that.

This is a gnarly piece of code that needs a bit of TLC to get right. The compiler, platform, and architecture all define some support for some kind of long double, and when I do not have access to a particular platform, I can't trace down what is not getting supported.

The comment you make about cross-compiling is intriguing, as I should be able to do that. Can you educate me how you generate these errors?

[josopait]

Yes, I noticed many platform-specific options. Would it not be easier to use other preprocessor defines such as __LDBL_MAX_EXP__?

I am using the crossdev package from gentoo linux to create cross-compiler toolchains, and then cmake with appropriate toolchain definitions.

I was looking for a way to emulate 8 and 16 bit floating point types. Your library seems ideal, if only I can get it working.

[Ravenwater]

@josopait Universal has 4, 8 and 16-bit floating-point types. It goes well beyond that and offers completely parameterized floating-point types so that you can ask for a 12-bit or a 6-bit floating-point just the same. Take a look at the cfloat number system.

The errors we are seeing is about the code that finds the fields in a long double. Because long doubles do not have a standard, each hardware and compiler does its own thing, and thus the code that is tripping us up in your cross-compilation, is caused by the combination of hardware, compiler, and platform not yielding the building blocks needed for the reflection of that type.

It is not rocket science, but it does require knowledge of how the include hierarchy works and how the compilation guards select code in that hierarchy.

I should be able to track down the gcc/x86/windows case as I do have that configuration.

Do you know if the crossdev package is available on Ubuntu?

[josopait]

I believe I saw some cross gcc packages in ubuntu that should work.