zlib fails to build on s390x on oracular with gcc 14

[Frank Heimes]

Rebuilding zlib for s390x on oracular fails to build, probably due to gcc 14 usage (since it became more strict).

Full build log can be found here:
https://launchpadlibrarian.net/741961061/buildlog_ubuntu-oracular-s390x.zlib_1%3A1.3.dfsg-3.1ubuntu4_BUILDING.txt.gz

erroneous lines are:

D_LARGEFILE64_SOURCE=1 -DHAVE_HIDDEN -DHAVE_IFUNC -DHAVE_S390X_VX -march=z13 -mzarch -c -o crc32-vx.o contrib/s390/crc32-vx.c
contrib/s390/crc32-vx.c: In function ‘crc32_le_vgfm_16’:
contrib/s390/crc32-vx.c:91:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
   91 | v1 = (uv2di)vec_gfmsum_accum_128(r2r1, v1, part1);
      | ^~~~~
      | |
      | uv16qi {aka __vector(16) unsigned char}
contrib/s390/crc32-vx.c:91:52: note: expected ‘__int128 unsigned’ but argument is of type ‘uv16qi’ {aka ‘__vector(16) unsigned char’}
contrib/s390/crc32-vx.c:92:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
   92 | v2 = (uv2di)vec_gfmsum_accum_128(r2r1, v2, part2);
      | ^~~~~
      | |
      | uv16qi {aka __vector(16) unsigned char}
contrib/s390/crc32-vx.c:92:52: note: expected ‘__int128 unsigned’ but argument is of type ‘uv16qi’ {aka ‘__vector(16) unsigned char’}
contrib/s390/crc32-vx.c:93:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
   93 | v3 = (uv2di)vec_gfmsum_accum_128(r2r1, v3, part3);
      | ^~~~~
      | |
      | uv16qi {aka __vector(16) unsigned char}
contrib/s390/crc32-vx.c:93:52: note: expected ‘__int128 unsigned’ but argument is of type ‘uv16qi’ {aka ‘__vector(16) unsigned char’}
contrib/s390/crc32-vx.c:94:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
   94 | v4 = (uv2di)vec_gfmsum_accum_128(r2r1, v4, part4);
      | ^~~~~
      | |
      | uv16qi {aka __vector(16) unsigned char}
contrib/s390/crc32-vx.c:94:52: note: expected ‘__int128 unsigned’ but argument is of type ‘uv16qi’ {aka ‘__vector(16) unsigned char’}
contrib/s390/crc32-vx.c:105:48: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  105 | v1 = (uv2di)vec_gfmsum_accum_128(r4r3, v1, (uv16qi)v2);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:105:48: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
contrib/s390/crc32-vx.c:106:48: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  106 | v1 = (uv2di)vec_gfmsum_accum_128(r4r3, v1, (uv16qi)v3);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:106:48: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
contrib/s390/crc32-vx.c:107:48: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  107 | v1 = (uv2di)vec_gfmsum_accum_128(r4r3, v1, (uv16qi)v4);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:107:48: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
contrib/s390/crc32-vx.c:114:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  114 | v1 = (uv2di)vec_gfmsum_accum_128(r4r3, v1, (uv16qi)v2);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:114:52: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
contrib/s390/crc32-vx.c:161:46: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  161 | v1 = (uv2di)vec_gfmsum_accum_128(r5, v1, (uv16qi)v2);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:161:46: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
contrib/s390/crc32-vx.c:192:52: error: incompatible type for argument 3 of ‘__builtin_s390_vgfmag’
  192 | v2 = (uv2di)vec_gfmsum_accum_128(crc_poly, v2, (uv16qi)v1);
      | ^~~~~~~~~~
      | |
      | __vector(16) unsigned char
contrib/s390/crc32-vx.c:192:52: note: expected ‘__int128 unsigned’ but argument is of type ‘__vector(16) unsigned char’
make[1]: *** [Makefile:180: crc32-vx.o] Error 1

[Simon Chopin (schopin)]

Frank, can you confirm that you got in touch with the upstream author about it?

I probably could "fix" it by sprinkling casts all over the place, which could be wildly unsafe depending on the exact properties of __int128, or by actually using a local __int128 variable and memcpy it or something like that, which might destroy performance? (I recall something like that being a problem in some glibc routines on ARM where moving data from SIMD registers was *really* slow on the Pis)

As you see, I'm in much doubt so knowledge from the people who wrote the patch (and own the hardware) would be greatly appreciated!

[Frank Heimes (fheimes)]

Hi Simon, not yet, but I'm pushing for this ...

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-08-07 12:48 EDT-------
Thanks Ilya for your quick analysis, finding out that this is a compiler issue. Stefan will send a patch to the GCC mailing list.
We will post a link here once it's available.