Linking errors on Windows with msys2

[moneylobster]

After doing some replacements to adapt some parts of the repo to Windows (changing the foreign-library name in shared-object.lisp from libbmas.so to libbmas.dll and doing the same for make.sh) I got some errors during the linking stage with Windows 10 with msys2 on UCRT64.

Compilation Output

$ ./make.sh
In file included from simd/x86_64.h:11,
                 from bmas.c:11:
simd/x86_64/avx2.h: In function 'BMAS_vector_i64hindex':
simd/x86_64/avx2.h:1128:21: warning: left shift count >= width of type [-Wshift-count-overflow]
 1128 |   case 'a': m = 1UL << 63;  break;
      |                     ^~
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1ade7): undefined reference to `_ZGVdN8v_sinf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1aff9): undefined reference to `_ZGVdN8v_sinf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b19f): undefined reference to `_ZGVdN8v_sinf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b27b): undefined reference to `_ZGVdN8v_sinf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b397): undefined reference to `_ZGVdN8v_cosf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b5a9): undefined reference to `_ZGVdN8v_cosf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b74f): undefined reference to `_ZGVdN8v_cosf'
C:/msys64/ucrt64/bin/../lib/gcc/x86_64-w64-mingw32/14.1.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:\msys64\tmp\ccjCZ5xB.o:bmas.c:(.text+0x1b82b): undefined reference to `_ZGVdN8v_cosf'
collect2.exe: error: ld returned 1 exit status

The errors seem to have to do with the sinf and cosf functions, and indeed trying to compile with these lines commented out works:

bmas/bmas.c

Lines 186 to 187 in bf68699

	one_arg_fn_body(ssin, SIMD_SINGLE_STRIDE, float, BMAS_svec, float, BMAS_svec);
	one_arg_fn_body(scos, SIMD_SINGLE_STRIDE, float, BMAS_svec, float, BMAS_svec);

[digikar99]

The undefined reference must be stemming from here:

bmas/simd/x86_64/avx2.h

Lines 1680 to 1685 in bf68699

	/* BMAS_svec static inline BMAS_vector_ssin(BMAS_svec x){ return Sleef_sinf8_u10avx2(x);} */
	__m256 _ZGVdN8v_sinf(__m256 x);
	BMAS_svec static inline BMAS_vector_ssin(BMAS_svec x){ return _ZGVdN8v_sinf(x);}
	/* BMAS_svec static inline BMAS_vector_scos(BMAS_svec x){ return Sleef_cosf8_u10avx2(x);} */
	__m256 _ZGVdN8v_cosf(__m256 x);
	BMAS_svec static inline BMAS_vector_scos(BMAS_svec x){ return _ZGVdN8v_cosf(x);}

These are the SIMD sind and cos provided by gcc. Atleast on x86_64 linux, they are faster than SLEEF's versions. I'm not sure what the equivalent to this are on windows. As a starting point, you can use the the Sleef_{sin|cos}f8_u10avx2 and later look into whether these are slower than alternatives.

[moneylobster]

The SLEEF version compiled without a problem, thank you.