[PATCH] hash: fix GFNI implementation build with GCC 12
Bruce Richardson
bruce.richardson at intel.com
Mon Jan 9 11:24:00 CET 2023
On Mon, Jan 09, 2023 at 11:03:37AM +0100, David Marchand wrote:
> On a system that has AVX512F and GFNI, compiling fails with:
>
> In file included from /usr/lib/gcc/x86_64-redhat-linux/12/include/immintrin.h:71,
> from /usr/lib/gcc/x86_64-redhat-linux/12/include/x86intrin.h:32,
> from ../../../git/pub/dpdk.org/main/lib/eal/x86/include/rte_vect.h:31,
> from ../../../git/pub/dpdk.org/main/lib/eal/x86/include/rte_memcpy.h:17,
> from ../../../git/pub/dpdk.org/main/lib/mempool/rte_mempool.h:48,
> from ../../../git/pub/dpdk.org/main/lib/mbuf/rte_mbuf.h:38,
> from ../../../git/pub/dpdk.org/main/lib/net/rte_ip.h:33,
> from ../../../git/pub/dpdk.org/main/lib/hash/rte_thash.h:25,
> from ../../../git/pub/dpdk.org/main/lib/hash/rte_thash.c:7:
> In function ‘_mm512_mask_permutexvar_epi8’,
> inlined from ‘__rte_thash_gfni’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:142:17,
> inlined from ‘rte_thash_gfni’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:182:20,
> inlined from ‘rte_thash_adjust_tuple’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash.c:784:11:
> /usr/lib/gcc/x86_64-redhat-linux/12/include/avx512vbmiintrin.h:97:20:
> error: ‘tuple_bytes’ may be used uninitialized [-Werror=maybe-uninitialized]
> 97 | return (__m512i) __builtin_ia32_permvarqi512_mask ((__v64qi) __B,
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 98 | (__v64qi) __A,
> | ~~~~~~~~~~~~~~
> 99 | (__v64qi) __W,
> | ~~~~~~~~~~~~~~
> 100 | (__mmask64) __M);
> | ~~~~~~~~~~~~~~~~
>
> And:
>
> In file included from ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_gfni.h:17,
> from ../../../git/pub/dpdk.org/main/lib/hash/rte_thash.h:27:
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:
> In function ‘rte_thash_adjust_tuple’:
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:91:33:
> note: ‘tuple_bytes’ was declared here
> 91 | __m512i vals, matrixes, tuple_bytes, tuple_bytes_2;
> | ^~~~~~~~~~~
> In function ‘_mm512_mask_permutexvar_epi8’,
> inlined from ‘__rte_thash_gfni’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:142:17,
> inlined from ‘rte_thash_gfni’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:182:20,
> inlined from ‘rte_thash_adjust_tuple’ at
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash.c:784:11:
> /usr/lib/gcc/x86_64-redhat-linux/12/include/avx512vbmiintrin.h:97:20:
> error: ‘permute_mask’ may be used uninitialized [-Werror=maybe-uninitialized]
> 97 | return (__m512i) __builtin_ia32_permvarqi512_mask ((__v64qi) __B,
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 98 | (__v64qi) __A,
> | ~~~~~~~~~~~~~~
> 99 | (__v64qi) __W,
> | ~~~~~~~~~~~~~~
> 100 | (__mmask64) __M);
> | ~~~~~~~~~~~~~~~~
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:
> In function ‘rte_thash_adjust_tuple’:
> ../../../git/pub/dpdk.org/main/lib/hash/rte_thash_x86_gfni.h:92:30:
> note: ‘permute_mask’ was declared here
> 92 | __mmask64 load_mask, permute_mask, permute_mask_2;
> | ^~~~~~~~~~~~
> cc1: all warnings being treated as errors
>
> Set those variables to 0.
>
> Fixes: 4fd8c4cb0de1 ("hash: add new Toeplitz hash implementation")
> Cc: stable at dpdk.org
>
> Signed-off-by: David Marchand <david.marchand at redhat.com>
> ---
> lib/hash/rte_thash_x86_gfni.h | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>
Hit this issue myself before, but somehow never got to pushing out a patch
for it. Thanks for doing so.
Acked-by: Bruce Richardson <bruce.richardson at intel.com>
More information about the stable
mailing list