[PATCH] x86: correct VMOVSH attributes

[H.J. Lu]

On Mon, Jul 18, 2022 at 2:23 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 15.07.2022 19:49, H.J. Lu wrote:
> > On Fri, Jul 15, 2022 at 3:01 AM Jan Beulich <jbeulich@suse.com> wrote:
> >>
> >> Both forms were missing VexW0 (thus allowing Evex.W=1 to be encoded by
> >> suitable means, which would cause #UD). The memory operand form further
> >> was using the wrong Masking value, thus allowing zeroing-masking to be
> >> encoded for the store form (which would again cause #UD).
> >>
> >> --- a/gas/testsuite/gas/i386/evex-wig.s
> >> +++ b/gas/testsuite/gas/i386/evex-wig.s
> >> @@ -62,6 +62,18 @@ _start:
> >>         {evex} vpinsrw $0, %eax, %xmm0, %xmm0
> >>         {evex} vpinsrw $0, 2(%eax), %xmm0, %xmm0
> >>
> >> +       vmovss %xmm0, %xmm0, %xmm0{%k7}
> >> +       vmovss (%eax), %xmm0{%k7}
> >> +       vmovss %xmm0, (%eax){%k7}
> >> +
> >> +       vmovsd %xmm0, %xmm0, %xmm0{%k7}
> >> +       vmovsd (%eax), %xmm0{%k7}
> >> +       vmovsd %xmm0, (%eax){%k7}
> >> +
> >> +       vmovsh %xmm0, %xmm0, %xmm0{%k7}
> >> +       vmovsh (%eax), %xmm0{%k7}
> >> +       vmovsh %xmm0, (%eax){%k7}
> >> +
> >>         vpmovsxbd       %xmm5, %zmm6{%k7}        # AVX512
> >>         vpmovsxbd       %xmm5, %zmm6{%k7}{z}     # AVX512
> >>         vpmovsxbd       (%ecx), %zmm6{%k7}       # AVX512
> >> --- a/gas/testsuite/gas/i386/evex-wig1-intel.d
> >> +++ b/gas/testsuite/gas/i386/evex-wig1-intel.d
> >> @@ -45,6 +45,15 @@ Disassembly of section .text:
> >>  [      ]*[a-f0-9]+:    62 f3 fd 08 22 40 01 00         vpinsrd xmm0,xmm0,DWORD PTR \[eax\+0x4\],0x0
> >>  [      ]*[a-f0-9]+:    62 f1 fd 08 c4 c0 00    vpinsrw xmm0,xmm0,eax,0x0
> >>  [      ]*[a-f0-9]+:    62 f1 fd 08 c4 40 01 00         vpinsrw xmm0,xmm0,WORD PTR \[eax\+0x2\],0x0
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 10 c0       vmovss xmm0\{k7\},xmm0,xmm0
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 10 00       vmovss xmm0\{k7\},DWORD PTR \[eax\]
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 11 00       vmovss DWORD PTR \[eax\]\{k7\},xmm0
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 10 c0       vmovsd xmm0\{k7\},xmm0,xmm0
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 10 00       vmovsd xmm0\{k7\},QWORD PTR \[eax\]
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 11 00       vmovsd QWORD PTR \[eax\]\{k7\},xmm0
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 10 c0       vmovsh xmm0\{k7\},xmm0,xmm0
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 10 00       vmovsh xmm0\{k7\},WORD PTR \[eax\]
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 11 00       vmovsh WORD PTR \[eax\]\{k7\},xmm0
> >>  [      ]*[a-f0-9]+:    62 f2 fd 4f 21 f5       vpmovsxbd zmm6\{k7\},xmm5
> >>  [      ]*[a-f0-9]+:    62 f2 fd cf 21 f5       vpmovsxbd zmm6\{k7\}\{z\},xmm5
> >>  [      ]*[a-f0-9]+:    62 f2 fd 4f 21 31       vpmovsxbd zmm6\{k7\},XMMWORD PTR \[ecx\]
> >> --- a/gas/testsuite/gas/i386/evex-wig1.d
> >> +++ b/gas/testsuite/gas/i386/evex-wig1.d
> >> @@ -45,6 +45,15 @@ Disassembly of section .text:
> >>  [      ]*[a-f0-9]+:    62 f3 fd 08 22 40 01 00         vpinsrd \$0x0,0x4\(%eax\),%xmm0,%xmm0
> >>  [      ]*[a-f0-9]+:    62 f1 fd 08 c4 c0 00    vpinsrw \$0x0,%eax,%xmm0,%xmm0
> >>  [      ]*[a-f0-9]+:    62 f1 fd 08 c4 40 01 00         vpinsrw \$0x0,0x2\(%eax\),%xmm0,%xmm0
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 10 c0       vmovss %xmm0,%xmm0,%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 10 00       vmovss \(%eax\),%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f1 7e 0f 11 00       vmovss %xmm0,\(%eax\)\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 10 c0       vmovsd %xmm0,%xmm0,%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 10 00       vmovsd \(%eax\),%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f1 ff 0f 11 00       vmovsd %xmm0,\(%eax\)\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 10 c0       vmovsh %xmm0,%xmm0,%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 10 00       vmovsh \(%eax\),%xmm0\{%k7\}
> >> +[      ]*[a-f0-9]+:    62 f5 7e 0f 11 00       vmovsh %xmm0,\(%eax\)\{%k7\}
> >>  [      ]*[a-f0-9]+:    62 f2 fd 4f 21 f5       vpmovsxbd %xmm5,%zmm6\{%k7\}
> >>  [      ]*[a-f0-9]+:    62 f2 fd cf 21 f5       vpmovsxbd %xmm5,%zmm6\{%k7\}\{z\}
> >>  [      ]*[a-f0-9]+:    62 f2 fd 4f 21 31       vpmovsxbd \(%ecx\),%zmm6\{%k7\}
> >> --- a/opcodes/i386-opc.tbl
> >> +++ b/opcodes/i386-opc.tbl
> >> @@ -3684,8 +3684,8 @@ vmaxsh, 0xf35f, None, CpuAVX512_FP16, Mo
> >>  vminph, 0x5d, None, CpuAVX512_FP16, Modrm|VexVVVV|Masking=3|EVexMap5|VexW0|Broadcast|Disp8ShiftVL|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|SAE, { RegXMM|RegYMM|RegZMM|Word|Unspecified|BaseIndex, RegXMM|RegYMM|RegZMM, RegXMM|RegYMM|RegZMM }
> >>  vminsh, 0xf35d, None, CpuAVX512_FP16, Modrm|EVexLIG|Masking=3|EVexMap5|VexVVVV|VexW0|Disp8MemShift=1|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|SAE, { RegXMM|Word|Unspecified|BaseIndex, RegXMM, RegXMM }
> >>
> >> -vmovsh, 0xf310, None, CpuAVX512_FP16, D|Modrm|EVexLIG|Masking=3|EVexMap5|Disp8MemShift=1|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Word|Unspecified|BaseIndex, RegXMM }
> >> -vmovsh, 0xf310, None, CpuAVX512_FP16, D|Modrm|EVexLIG|Masking=3|EVexMap5|VexVVVV|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM, RegXMM, RegXMM }
> >> +vmovsh, 0xf310, None, CpuAVX512_FP16, D|Modrm|EVexLIG|MaskingMorZ|EVexMap5|VexW0|Disp8MemShift=1|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Word|Unspecified|BaseIndex, RegXMM }
> >> +vmovsh, 0xf310, None, CpuAVX512_FP16, D|Modrm|EVexLIG|Masking=3|EVexMap5|VexVVVV|VexW0|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM, RegXMM, RegXMM }
> >>
> >>  vmovw, 0x666e, None, CpuAVX512_FP16, D|Modrm|EVex128|VexWIG|EVexMap5|Disp8MemShift=1|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Word|Unspecified|BaseIndex, RegXMM }
> >>  vmovw, 0x667e, None, CpuAVX512_FP16, D|RegMem|EVex128|VexWIG|EVexMap5|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM, Reg32 }
> >
> > OK.
>
> Thanks. I wonder whether we don't also want this on the 2.39 branch.
>
> Jan

Please backport to 2.39.

Thanks.


-- 
H.J.