This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH] x86: Set Vex=1 on VEX.128 only vmovq
- From: "Jan Beulich" <jbeulich at suse dot com>
- To: <hjl dot tools at gmail dot com>,<binutils at sourceware dot org>
- Date: Sun, 16 Sep 2018 02:37:27 -0600
- Subject: Re: [PATCH] x86: Set Vex=1 on VEX.128 only vmovq
- References: <20180915215342.21963-1-hjl.tools@gmail.com>
>>> "H.J. Lu" <hjl.tools@gmail.com> 09/15/18 11:54 PM >>>
>AVX "VMOVQ xmm1, xmm2/m64" and "VMOVQ xmm1/m64, xmm2" can only be
>encoded with VEX.128. Set Vex=1 on VEX.128 only vmovq and update
>assembler tests.
Indeed, but ...
>--- a/opcodes/i386-opc.tbl
>+++ b/opcodes/i386-opc.tbl
>@@ -2030,8 +2030,8 @@ vmovntdq, 2, 0x66e7, None, 1, CpuAVX, Modrm|Vex|VexOpcode=0|VexW=3|CheckRegSize|
>vmovntdqa, 2, 0x662a, None, 1, CpuAVX, Modrm|Vex|VexOpcode=1|VexW=3|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Xmmword|Unspecified|BaseIndex, RegXMM }
>vmovntpd, 2, 0x662b, None, 1, CpuAVX, Modrm|Vex|VexOpcode=0|VexW=3|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegYMM, Xmmword|Ymmword|Unspecified|BaseIndex }
>vmovntps, 2, 0x2b, None, 1, CpuAVX, Modrm|Vex|VexOpcode=0|VexW=3|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegYMM, Xmmword|Ymmword|Unspecified|BaseIndex }
>-vmovq, 2, 0xf37e, None, 1, CpuAVX, Load|Modrm|Vex=3|VexOpcode=0|VexW=3|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|NoRex64, { Qword|Unspecified|BaseIndex|RegXMM, RegXMM }
>-vmovq, 2, 0x66d6, None, 1, CpuAVX, Modrm|Vex=3|VexOpcode=0|VexW=3|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|NoRex64, { RegXMM, Qword|Unspecified|BaseIndex|RegXMM }
>+vmovq, 2, 0xf37e, None, 1, CpuAVX, Load|Modrm|Vex=1|VexOpcode=0|VexW=3|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|NoRex64, { Qword|Unspecified|BaseIndex|RegXMM, RegXMM }
>+vmovq, 2, 0x66d6, None, 1, CpuAVX, Modrm|Vex=1|VexOpcode=0|VexW=3|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|NoRex64, { RegXMM, Qword|Unspecified|BaseIndex|RegXMM }
>vmovq, 2, 0x666e, None, 1, CpuAVX|Cpu64, D|Modrm|Vex=3|VexOpcode=0|VexW=2|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf|Size64, { Reg64|Unspecified|BaseIndex, RegXMM }
>vmovsd, 2, 0xf210, None, 1, CpuAVX, D|Modrm|Vex=3|VexOpcode=0|VexW=3|IgnoreSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Qword|Unspecified|BaseIndex, RegXMM }
>vmovsd, 3, 0xf210, None, 1, CpuAVX, D|Modrm|Vex=3|VexOpcode=0|VexVVVV=1|VexW=3|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegMem, RegXMM, RegXMM }
... you then also ought to change the respective SSE2AVX movq instances (one of the
undesirable side effects of these requiring separate templates). Furthermore their AVX512
counterparts appear to have the same issue, using Evex=4 currently instead of Evex=2.
And yet further, the other encodings (with the 32-bit variants named VMOVD) all look to
have the same issue as well.
Jan