[Bug string/31675] New: glibc.cpu.hwcaps doesn't work as documented

[Bug string/31675] New: glibc.cpu.hwcaps doesn't work as documented

[mikulas at artax dot karlin.mff.cuni.cz]

https://sourceware.org/bugzilla/show_bug.cgi?id=31675

            Bug ID: 31675
           Summary: glibc.cpu.hwcaps doesn't work as documented
           Product: glibc
           Version: 2.37
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: string
          Assignee: unassigned at sourceware dot org
          Reporter: mikulas at artax dot karlin.mff.cuni.cz
  Target Milestone: ---

If I disable AVX512 and AVX2 with
"export GLIBC_TUNABLES=glibc.cpu.hwcaps=-AVX512F,-AVX2,-AVX512VL",
glibc uses __memcpy_avx_unaligned_erms version of memcpy that uses AVX2
instructions.

We need to add "AVX_Fast_Unaligned_Load" to suppress AVX2. If I set:
"export
GLIBC_TUNABLES=glibc.cpu.hwcaps=-AVX512F,-AVX2,-AVX512VL,-AVX_Fast_Unaligned_Load",
then glibc will use the SSE3 version of memcpy.

The need to set "AVX_Fast_Unaligned_Load" should be either documented or fixed.

The documentation says "The glibc.cpu.hwcaps=-xxx,yyy,-zzz... tunable allows
the user to enable CPU/ARCH feature yyy, disable CPU/ARCH feature xxx and zzz
where the feature name is case-sensitive and has to match the ones in
sysdeps/x86/include/cpu-features.h.", but the file "cpu-features.h" doesn't
contain "AVX_Fast_Unaligned_Load".

-- 
You are receiving this mail because:
You are on the CC list for the bug.