* [PATCH] x86: simplify VexVVVV_SRC2 handling for the XOP case
@ 2024-05-17 12:57 Jan Beulich
2024-05-20 7:58 ` Cui, Lili
0 siblings, 1 reply; 2+ messages in thread
From: Jan Beulich @ 2024-05-17 12:57 UTC (permalink / raw)
To: Binutils; +Cc: Lili Cui, H.J. Lu
As already suggested during review, rather than having an extra
conditional in build_modrm_byte() (a code path used for quite a few
more insns, including even certain GPR ones), adjust the attribute in
the installed template to properly describe things with operands
swapped.
---
I think we could even omit the conditional in match_template(), as all
other insns where XOP.W / VEX.W controls operand order are Src1VVVV
anyway. Opinions?
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -9255,6 +9255,10 @@ match_template (char mnem_suffix)
flipping VEX.W. */
i.tm.opcode_modifier.vexw ^= VEXW0 ^ VEXW1;
+ /* In 3-operand insns XOP.W changes which operand goes into XOP.vvvv. */
+ if (i.tm.opcode_modifier.vexvvvv == VexVVVV_SRC2)
+ i.tm.opcode_modifier.vexvvvv = VexVVVV_SRC1;
+
swap_first_2:
j = i.tm.operand_types[0].bitfield.imm8;
i.tm.operand_types[j] = operand_types[j + 1];
@@ -10441,15 +10445,8 @@ build_modrm_byte (void)
{
/* VEX.vvvv encodes the last source register operand. */
case VexVVVV_SRC2:
- if (source != op)
- {
- v = source++;
- break;
- }
- /* For vprot*, vpshl*, and vpsha*, XOP.W controls the swapping of src1
- and src2, and it requires fall through when the operands are swapped.
- */
- /* Fall through. */
+ v = source++;
+ break;
/* VEX.vvvv encodes the first source register operand. */
case VexVVVV_SRC1:
v = dest - 1;
^ permalink raw reply [flat|nested] 2+ messages in thread
* RE: [PATCH] x86: simplify VexVVVV_SRC2 handling for the XOP case
2024-05-17 12:57 [PATCH] x86: simplify VexVVVV_SRC2 handling for the XOP case Jan Beulich
@ 2024-05-20 7:58 ` Cui, Lili
0 siblings, 0 replies; 2+ messages in thread
From: Cui, Lili @ 2024-05-20 7:58 UTC (permalink / raw)
To: Beulich, Jan, Binutils; +Cc: H.J. Lu
> As already suggested during review, rather than having an extra conditional in
> build_modrm_byte() (a code path used for quite a few more insns, including
> even certain GPR ones), adjust the attribute in the installed template to
> properly describe things with operands swapped.
Agreed, putting special handlings together makes it easier to understand.
> ---
> I think we could even omit the conditional in match_template(), as all other
> insns where XOP.W / VEX.W controls operand order are Src1VVVV anyway.
> Opinions?
>
Yes, since we change i.tm.opcode_modifier.vexw, i.tm.opcode_modifier.vexvvvv should also be changed together without any conditions.
Lili.
> --- a/gas/config/tc-i386.c
> +++ b/gas/config/tc-i386.c
> @@ -9255,6 +9255,10 @@ match_template (char mnem_suffix)
> flipping VEX.W. */
> i.tm.opcode_modifier.vexw ^= VEXW0 ^ VEXW1;
>
> + /* In 3-operand insns XOP.W changes which operand goes into XOP.vvvv.
> */
> + if (i.tm.opcode_modifier.vexvvvv == VexVVVV_SRC2)
> + i.tm.opcode_modifier.vexvvvv = VexVVVV_SRC1;
> +
> swap_first_2:
> j = i.tm.operand_types[0].bitfield.imm8;
> i.tm.operand_types[j] = operand_types[j + 1]; @@ -10441,15 +10445,8
> @@ build_modrm_byte (void)
> {
> /* VEX.vvvv encodes the last source register operand. */
> case VexVVVV_SRC2:
> - if (source != op)
> - {
> - v = source++;
> - break;
> - }
> - /* For vprot*, vpshl*, and vpsha*, XOP.W controls the swapping of src1
> - and src2, and it requires fall through when the operands are swapped.
> - */
> - /* Fall through. */
> + v = source++;
> + break;
> /* VEX.vvvv encodes the first source register operand. */
> case VexVVVV_SRC1:
> v = dest - 1;
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-05-20 7:58 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-05-17 12:57 [PATCH] x86: simplify VexVVVV_SRC2 handling for the XOP case Jan Beulich
2024-05-20 7:58 ` Cui, Lili
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).