This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH, i386][3/3] Update to match latest AVX512 spec.
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Ilya Tocar <tocarip dot intel at gmail dot com>
- Cc: Binutils <binutils at sourceware dot org>
- Date: Thu, 20 Mar 2014 10:30:45 -0700
- Subject: Re: [PATCH, i386][3/3] Update to match latest AVX512 spec.
- Authentication-results: sourceware.org; auth=none
- References: <20140220151936 dot GA1312 at msticlxl7 dot ims dot intel dot com> <CAMe9rOps3fYbJvx0v+6gm2jd=UrDT-0in8adyAUpfQk8AvNy2g at mail dot gmail dot com> <20140224132223 dot GA39232 at msticlxl7 dot ims dot intel dot com> <CAMe9rOqiBVCGAUY-q_HKfnxFVz2tspNCytPNz2mB1hVKrNxajQ at mail dot gmail dot com> <20140225094128 dot GB31394 at msticlxl7 dot ims dot intel dot com> <CAMe9rOrExVo8jbWf_jyefFxoUd0oYXOvdZK4YK5YkEv9Ma4jcQ at mail dot gmail dot com> <CAMe9rOrJb0qCEmznJ08vmc+BjCJwfQQ6i=Oq_rSZTfoo30fG2A at mail dot gmail dot com> <20140225162451 dot GB75072 at msticlxl7 dot ims dot intel dot com> <CAMe9rOqjHOoYhGUnYQri4cJfA+N4hYCyDYp_AWR7XvJf-YBevg at mail dot gmail dot com>
On Tue, Feb 25, 2014 at 8:53 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Tue, Feb 25, 2014 at 8:24 AM, Ilya Tocar <tocarip.intel@gmail.com> wrote:
>>> >>> vcvtps2ph supports memory destination. Please add AVX512F testcases
>>> >>> with memory destination.
>>> >>>
>>> >> We already have such tests. See e. g. on lines 6883-6890 in
>>> >> gas/testsuite/gas/i386/avx512f.s we have:
>>> >>
>>> >> vcvtps2ph $0xab, %zmm6, (%ecx) # AVX512F
>>> >> vcvtps2ph $0xab, %zmm6, (%ecx){%k7} # AVX512F
>>> >> vcvtps2ph $123, %zmm6, (%ecx) # AVX512F
>>> >> vcvtps2ph $123, %zmm6, -123456(%esp,%esi,8) # AVX512F
>>> >> vcvtps2ph $123, %zmm6, 4064(%edx) # AVX512F Disp8
>>> >> vcvtps2ph $123, %zmm6, 4096(%edx) # AVX512F
>>> >> vcvtps2ph $123, %zmm6, -4096(%edx) # AVX512F Disp8
>>> >>
>>> >> Do we need additional tests?
>>> >
>>> > Yes, please test with memory destination for:
>>> >
>>> > vcvtps2ph, 3, 0x661D, None, 1, CpuAVX512F,
>>> > Modrm|EVex=1|Masking=2|VexOpcode=2|VexW=1|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|SAE,
>>> > { Imm8, Imm8, RegZMM|RegMem }
>> This variant is removed in my patch. What exactly should I test?
>>>
>>> Also for
>>>
>>> vcvtps2ph, 4, 0x661D, None, 1, CpuAVX512F,
>>> Modrm|EVex=1|Masking=3|VexOpcode=2|VexW=1|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|SAE,
>>> { Imm8, Imm8, RegZMM, RegYMM|RegMem }
>> This is reg-reg version with SAE, which is already tested on e. g.
>> line 2602 in gas/testsuite/gas/i386/avx512f.s:
>>
>> vcvtps2ph $0xab, {sae}, %zmm5, %ymm6{%k7} # AVX512F
>>
>> Also instruction can't have memory operand and SAE at the same time.
>
> OK. I checked it in for you.
>
I backported it to binutils-2_24-branch.
--
H.J.