[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798
--- Comment #9 from Hongtao.liu ---
(In reply to Wojciech Mula from comment #8)
> Thank you for the answer. Thus my question is: is it possible to delay
> conversion from kmasks into ints? I'm not a language lawyer, but I guess a
> `x binop y` has to be treated as `(int)x binop (int)y`. If it's true, we
> will have to prove that `(int)(x avx512-binop y)` is equivalent to the
> latter expr.
It's quite tricky to teach RA to choose the best alternative(which is discussed
in pr101185).
alternatively, you can use intrinsic _kor_mask64 directly to avoid extra
movment from GPR to MASK.
uint64_t any_whitespace(__m512i string) {
return _kor_mask64 (_kor_mask64 (_mm512_cmpeq_epu8_mask(string,
_mm512_set1_epi8(' '))
,_mm512_cmpeq_epu8_mask(string, _mm512_set1_epi8('\n')))
,_mm512_cmpeq_epu8_mask(string, _mm512_set1_epi8('\r')));
}
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 --- Comment #8 from Wojciech Mula --- Thank you for the answer. Thus my question is: is it possible to delay conversion from kmasks into ints? I'm not a language lawyer, but I guess a `x binop y` has to be treated as `(int)x binop (int)y`. If it's true, we will have to prove that `(int)(x avx512-binop y)` is equivalent to the latter expr.
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798
--- Comment #7 from Hongtao.liu ---
(In reply to Wojciech Mula from comment #6)
> Hongtao, thank you for your patch and for pinging back! I checked the code
> from this issue against version 11.2.0 (Debian 11.2.0-14), but still, there
> are KMOVQs before performing any bit ops. Here is the output from `gcc -O3
> -march=icelake-server -S`
>
> vpcmpub $0, .LC0(%rip), %zmm0, %k0
> vpcmpub $0, .LC1(%rip), %zmm0, %k1
> vpcmpub $0, .LC2(%rip), %zmm0, %k2
> kmovq %k0, %rcx
> kmovq %k1, %rax
> orq %rcx, %rax
> kmovq %k2, %rdx
> orq %rdx, %rax
> ret
Oh, Yes, Because of pr101185, mask register is slightly disliked. mask bitwise
instructions are generated only if src and dest are both mask registers.
.i.e
#include
__m512i
foo_orq (__m512i a, __m512i b, __m512i c, __m512i d)
{
__mmask64 m1 = _mm512_cmpeq_epi8_mask (a, b);
__mmask64 m2 = _mm512_cmpeq_epi8_mask (c, d);
return _mm512_mask_add_epi8 (c, m1 | m2, a, d);
}
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 --- Comment #6 from Wojciech Mula --- Hongtao, thank you for your patch and for pinging back! I checked the code from this issue against version 11.2.0 (Debian 11.2.0-14), but still, there are KMOVQs before performing any bit ops. Here is the output from `gcc -O3 -march=icelake-server -S` vpcmpub $0, .LC0(%rip), %zmm0, %k0 vpcmpub $0, .LC1(%rip), %zmm0, %k1 vpcmpub $0, .LC2(%rip), %zmm0, %k2 kmovq %k0, %rcx kmovq %k1, %rax orq %rcx, %rax kmovq %k2, %rdx orq %rdx, %rax ret
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 Hongtao.liu changed: What|Removed |Added CC||crazylht at gmail dot com --- Comment #5 from Hongtao.liu --- Fixed in GCC11 by https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=388cb292a94f98a276548cd6ce01285cf36d17df
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 Richard Biener changed: What|Removed |Added Keywords||missed-optimization Target||x86_64-*-*, i?86-*-* Status|UNCONFIRMED |NEW Last reconfirmed||2019-01-14 Ever confirmed|0 |1 --- Comment #4 from Richard Biener --- The testcase still behaves the same on trunk, not sure if exactly a dup or not.
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 --- Comment #3 from Wojciech Mula --- Sorry, I didn't find that bug; I think you may close this one. BTW, I had checked the code on godbolt.org before submitting. I tested also with their "GCC (trunk)", but the generated code is the same as for 8.2. The trunk's version is "g++ (GCC-Explorer-Build) 9.0.0 20190109 (experimental)" -- seems it's a fresh version and should already include the fixes Andrew mentioned.
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 Jakub Jelinek changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #2 from Jakub Jelinek --- See PR88473 for more details. You can use _kor_mask64 if you want explicitly using the mask operations instead of GPR.
[Bug target/88798] AVX512BW code does not use bit-operations that work on mask registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88798 --- Comment #1 from Andrew Pinski --- Some if not all has been fixed on the trunk. There was just a few weeks ago a bug that asked for the similar thing.
