On Tue, Jan 12, 2016 at 5:12 AM, Kirill Yukhin <kirill.yuk...@gmail.com> wrote:
> Hello Jakub
> On 08 Jan 21:20, Jakub Jelinek wrote:
>> Hi!
>>
>> This patch fixes
>> FAIL: gcc.target/i386/avx512vl-vmovapd-1.c scan-assembler-times vmovapd[
>> \\\\t]+[^{\\n]*%xmm[0-9]+[^\\n]*\\\\){%k[1-7]}(?:\\n|[ \\\\t]+#) 1
>> FAIL: gcc.target/i386/avx512vl-vmovapd-1.c scan-assembler-times vmovapd[
>> \\\\t]+[^{\\n]*%ymm[0-9]+[^\\n]*\\\\){%k[1-7]}(?:\\n|[ \\\\t]+#) 1
>> FAIL: gcc.target/i386/avx512vl-vmovaps-1.c scan-assembler-times vmovaps[
>> \\\\t]+[^{\\n]*%xmm[0-9]+[^\\n]*\\\\){%k[1-7]}(?:\\n|[ \\\\t]+#) 1
>> FAIL: gcc.target/i386/avx512vl-vmovaps-1.c scan-assembler-times vmovaps[
>> \\\\t]+[^{\\n]*%ymm[0-9]+[^\\n]*\\\\){%k[1-7]}(?:\\n|[ \\\\t]+#) 1
>> regressions that were introduced recently by fixing up the masked store
>> check for misalignment.
>> The problem is that for v2df/v4df/v4sf/v8sf masked stores
>> ix86_expand_special_args_builtin
>> failed to set aligned_mem and thus didn't set correct memory alignment.
>>
>> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> Followed you discussion w/ HJ.
> I think that metioned intrinsics should assume proper alignement and this
> agrees with SDM.
>
> So, your patch is ok for main trunk.
>
> --
> Thanks, K
>
>
>>
>> 2016-01-08 Jakub Jelinek <ja...@redhat.com>
>>
>> PR target/69198
>> * config/i386/i386.c (ix86_expand_special_args_builtin): Ensure
>> aligned_mem is properly set for AVX512-VL floating point masked
>> stores.
>>
>> --- gcc/config/i386/i386.c.jj 2016-01-08 07:31:11.000000000 +0100
>> +++ gcc/config/i386/i386.c 2016-01-08 18:16:21.030354042 +0100
>> @@ -39776,7 +39776,11 @@ ix86_expand_special_args_builtin (const
>> memory = 0;
>> break;
>> case VOID_FTYPE_PV8DF_V8DF_UQI:
>> + case VOID_FTYPE_PV4DF_V4DF_UQI:
>> + case VOID_FTYPE_PV2DF_V2DF_UQI:
>> case VOID_FTYPE_PV16SF_V16SF_UHI:
>> + case VOID_FTYPE_PV8SF_V8SF_UQI:
>> + case VOID_FTYPE_PV4SF_V4SF_UQI:
>> case VOID_FTYPE_PV8DI_V8DI_UQI:
>> case VOID_FTYPE_PV4DI_V4DI_UQI:
>> case VOID_FTYPE_PV2DI_V2DI_UQI:
>> @@ -39834,10 +39838,6 @@ ix86_expand_special_args_builtin (const
>> case VOID_FTYPE_PV16QI_V16QI_UHI:
>> case VOID_FTYPE_PV32QI_V32QI_USI:
>> case VOID_FTYPE_PV64QI_V64QI_UDI:
>> - case VOID_FTYPE_PV4DF_V4DF_UQI:
>> - case VOID_FTYPE_PV2DF_V2DF_UQI:
>> - case VOID_FTYPE_PV8SF_V8SF_UQI:
>> - case VOID_FTYPE_PV4SF_V4SF_UQI:
>> nargs = 2;
>> klass = store;
>> /* Reserve memory operand for target. */
>>
>> Jakub
GCC 5 has the same issue. This patch should be backported to GCC 5
with
https://gcc.gnu.org/ml/gcc-patches/2016-01/msg00528.html
which supersedes:
https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=231269
OK to backport Jakub's and my patch for GCC 5?
--
H.J.
