[PATCH 7/7] ix86: don't recognize/derive Q suffix in the common case

H.J. Lu hjl.tools@gmail.com
Wed Aug 17 20:36:10 GMT 2022


On Tue, Aug 16, 2022 at 12:34 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> Have its use, except where actually legitimate, result in the same "only
> supported in 64-bit mode" diagnostic as emitted for other 64-bit only
> insns. Also suppress deriving of the suffix in Intel mode except in the
> legitimate cases. This in exchange allows dropping the respective code
> from match_template().
>
> Oddly enough despite gcc's preference towards FILDQ and FIST{,T}Q we

This is for inline assembly:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=39590

> had no testcase whatsoever for these. Therefore such tests are being
> added. Note that the removed line in the x86-64-lfence-load testcase
> was redundant with the exact same one a few lines up.
> ---
> With gcc's preference towards FILDQ / FIST{,T}Q I wonder whether the
> disassembler wouldn't better emit a Q suffix instead of the LL one.
>
> --- a/gas/config/tc-i386.c
> +++ b/gas/config/tc-i386.c
> @@ -4826,7 +4826,7 @@ void
>  md_assemble (char *line)
>  {
>    unsigned int j;
> -  char mnemonic[MAX_MNEM_SIZE], mnem_suffix, *copy;
> +  char mnemonic[MAX_MNEM_SIZE], mnem_suffix = 0, *copy;
>    const char *pass1_mnem = NULL;
>    enum i386_error pass1_err = 0;
>    const insn_template *t;
> @@ -4860,6 +4860,7 @@ md_assemble (char *line)
>             goto no_match;
>           /* No point in trying a 2nd pass - it'll only find the same suffix
>              again.  */
> +         mnem_suffix = i.suffix;
>           goto match_error;
>         }
>        free (copy);
> @@ -5010,9 +5011,15 @@ md_assemble (char *line)
>                   cpu_sub_arch_name ? cpu_sub_arch_name : "");
>           return;
>         case unsupported_64bit:
> -         as_bad (_("`%s' is %s supported in 64-bit mode"),
> -                 pass1_mnem ? pass1_mnem : current_templates->start->name,
> -                 flag_code == CODE_64BIT ? _("not") : _("only"));
> +         if (ISLOWER (mnem_suffix))
> +           as_bad (_("`%s%c' is %s supported in 64-bit mode"),
> +                   pass1_mnem ? pass1_mnem : current_templates->start->name,
> +                   mnem_suffix,
> +                   flag_code == CODE_64BIT ? _("not") : _("only"));
> +         else
> +           as_bad (_("`%s' is %s supported in 64-bit mode"),
> +                   pass1_mnem ? pass1_mnem : current_templates->start->name,
> +                   flag_code == CODE_64BIT ? _("not") : _("only"));
>           return;
>         case invalid_sib_address:
>           err_msg = _("invalid SIB address");
> @@ -5355,6 +5362,23 @@ md_assemble (char *line)
>      last_insn.kind = last_insn_other;
>  }
>
> +/* The Q suffix is generally valid only in 64-bit mode, with very few
> +   exceptions: fild, fistp, fisttp, and cmpxchg8b.  Note that for fild
> +   and fisttp only one of their two templates is matched below: That's
> +   sufficient since other relevant attributes are the same between both
> +   respective templates.  */
> +static INLINE bool q_suffix_allowed(const insn_template *t)
> +{
> +  return flag_code == CODE_64BIT
> +        || (t->opcode_modifier.opcodespace == SPACE_BASE
> +            && t->base_opcode == 0xdf
> +            && (t->extension_opcode & 1)) /* fild / fistp / fisttp */
> +        || (t->opcode_modifier.opcodespace == SPACE_0F
> +            && t->base_opcode == 0xc7
> +            && t->opcode_modifier.opcodeprefix == PREFIX_NONE
> +            && t->extension_opcode == 1) /* cmpxchg8b */;
> +}
> +
>  static char *
>  parse_insn (char *line, char *mnemonic)
>  {
> @@ -5626,6 +5650,10 @@ parse_insn (char *line, char *mnemonic)
>    for (t = current_templates->start; t < current_templates->end; ++t)
>      {
>        supported |= cpu_flags_match (t);
> +
> +      if (i.suffix == QWORD_MNEM_SUFFIX && !q_suffix_allowed (t))
> +       supported &= ~CPU_FLAGS_64BIT_MATCH;
> +
>        if (supported == CPU_FLAGS_PERFECT_MATCH)
>         return l;
>      }
> @@ -6661,20 +6689,12 @@ match_template (char mnem_suffix)
>        for (j = 0; j < MAX_OPERANDS; j++)
>         operand_types[j] = t->operand_types[j];
>
> -      /* In general, don't allow
> -        - 64-bit operands outside of 64-bit mode,
> -        - 32-bit operands on pre-386.  */
> +      /* In general, don't allow 32-bit operands on pre-386.  */
>        specific_error = progress (mnem_suffix ? invalid_instruction_suffix
>                                              : operand_size_mismatch);
>        j = i.imm_operands + (t->operands > i.imm_operands + 1);
> -      if (((i.suffix == QWORD_MNEM_SUFFIX
> -           && flag_code != CODE_64BIT
> -           && !(t->opcode_modifier.opcodespace == SPACE_0F
> -                && t->base_opcode == 0xc7
> -                && t->opcode_modifier.opcodeprefix == PREFIX_NONE
> -                && t->extension_opcode == 1) /* cmpxchg8b */)
> -          || (i.suffix == LONG_MNEM_SUFFIX
> -              && !cpu_arch_flags.bitfield.cpui386))
> +      if (i.suffix == LONG_MNEM_SUFFIX
> +         && !cpu_arch_flags.bitfield.cpui386
>           && (intel_syntax
>               ? (t->opcode_modifier.mnemonicsize != IGNORESIZE
>                  && !intel_float_operand (t->name))
> --- a/gas/config/tc-i386-intel.c
> +++ b/gas/config/tc-i386-intel.c
> @@ -824,7 +824,7 @@ i386_intel_operand (char *operand_string
>                     continue;
>                   break;
>                 case QWORD_MNEM_SUFFIX:
> -                 if (t->opcode_modifier.no_qsuf)
> +                 if (t->opcode_modifier.no_qsuf || !q_suffix_allowed (t))
>                     continue;
>                   break;
>                 case SHORT_MNEM_SUFFIX:
> --- a/gas/testsuite/gas/i386/opcode.d
> +++ b/gas/testsuite/gas/i386/opcode.d
> @@ -592,6 +592,10 @@ Disassembly of section .text:
>  [      ]*[a-f0-9]+:    0f 4b 90 90 90 90 90    cmovnp -0x6f6f6f70\(%eax\),%edx
>  [      ]*[a-f0-9]+:    66 0f 4a 90 90 90 90 90         cmovp  -0x6f6f6f70\(%eax\),%dx
>  [      ]*[a-f0-9]+:    66 0f 4b 90 90 90 90 90         cmovnp -0x6f6f6f70\(%eax\),%dx
> +[      ]*[a-f0-9]+:    df 28                   fildll \(%eax\)
> +[      ]*[a-f0-9]+:    df 28                   fildll \(%eax\)
> +[      ]*[a-f0-9]+:    df 38                   fistpll \(%eax\)
> +[      ]*[a-f0-9]+:    df 38                   fistpll \(%eax\)
>   +[a-f0-9]+:   82 c3 01                add    \$0x1,%bl
>   +[a-f0-9]+:   82 f3 01                xor    \$0x1,%bl
>   +[a-f0-9]+:   82 d3 01                adc    \$0x1,%bl
> --- a/gas/testsuite/gas/i386/opcode.s
> +++ b/gas/testsuite/gas/i386/opcode.s
> @@ -592,6 +592,11 @@ foo:
>   cmovpe  0x90909090(%eax),%dx
>   cmovpo 0x90909090(%eax),%dx
>
> + fildq  (%eax)
> + fildll (%eax)
> + fistpq (%eax)
> + fistpll (%eax)
> +
>         .byte 0x82, 0xc3, 0x01
>         .byte 0x82, 0xf3, 0x01
>         .byte 0x82, 0xd3, 0x01
> --- a/gas/testsuite/gas/i386/opcode-intel.d
> +++ b/gas/testsuite/gas/i386/opcode-intel.d
> @@ -593,6 +593,10 @@ Disassembly of section .text:
>  [      ]*[a-f0-9]+:    0f 4b 90 90 90 90 90    cmovnp edx,DWORD PTR \[eax-0x6f6f6f70\]
>  [      ]*[a-f0-9]+:    66 0f 4a 90 90 90 90 90         cmovp  dx,WORD PTR \[eax-0x6f6f6f70\]
>  [      ]*[a-f0-9]+:    66 0f 4b 90 90 90 90 90         cmovnp dx,WORD PTR \[eax-0x6f6f6f70\]
> +[      ]*[a-f0-9]+:    df 28                   fild   QWORD PTR \[eax\]
> +[      ]*[a-f0-9]+:    df 28                   fild   QWORD PTR \[eax\]
> +[      ]*[a-f0-9]+:    df 38                   fistp  QWORD PTR \[eax\]
> +[      ]*[a-f0-9]+:    df 38                   fistp  QWORD PTR \[eax\]
>   +[a-f0-9]+:   82 c3 01                add    bl,0x1
>   +[a-f0-9]+:   82 f3 01                xor    bl,0x1
>   +[a-f0-9]+:   82 d3 01                adc    bl,0x1
> --- a/gas/testsuite/gas/i386/opcode-suffix.d
> +++ b/gas/testsuite/gas/i386/opcode-suffix.d
> @@ -593,6 +593,10 @@ Disassembly of section .text:
>  [      ]*[a-f0-9]+:    0f 4b 90 90 90 90 90    cmovnpl -0x6f6f6f70\(%eax\),%edx
>  [      ]*[a-f0-9]+:    66 0f 4a 90 90 90 90 90         cmovpw -0x6f6f6f70\(%eax\),%dx
>  [      ]*[a-f0-9]+:    66 0f 4b 90 90 90 90 90         cmovnpw -0x6f6f6f70\(%eax\),%dx
> +[      ]*[a-f0-9]+:    df 28                   fildll \(%eax\)
> +[      ]*[a-f0-9]+:    df 28                   fildll \(%eax\)
> +[      ]*[a-f0-9]+:    df 38                   fistpll \(%eax\)
> +[      ]*[a-f0-9]+:    df 38                   fistpll \(%eax\)
>   +[a-f0-9]+:   82 c3 01                addb   \$0x1,%bl
>   +[a-f0-9]+:   82 f3 01                xorb   \$0x1,%bl
>   +[a-f0-9]+:   82 d3 01                adcb   \$0x1,%bl
> --- a/gas/testsuite/gas/i386/sse3.d
> +++ b/gas/testsuite/gas/i386/sse3.d
> @@ -13,29 +13,30 @@ Disassembly of section .text:
>    10:  df 88 90 90 90 90 [     ]*fisttps -0x6f6f6f70\(%eax\)
>    16:  db 88 90 90 90 90 [     ]*fisttpl -0x6f6f6f70\(%eax\)
>    1c:  dd 88 90 90 90 90 [     ]*fisttpll -0x6f6f6f70\(%eax\)
> -  22:  66 0f 7c 65 00 [        ]*haddpd 0x0\(%ebp\),%xmm4
> -  27:  66 0f 7c ee [   ]*haddpd %xmm6,%xmm5
> -  2b:  f2 0f 7c 37 [   ]*haddps \(%edi\),%xmm6
> -  2f:  f2 0f 7c f8 [   ]*haddps %xmm0,%xmm7
> -  33:  66 0f 7d c1 [   ]*hsubpd %xmm1,%xmm0
> -  37:  66 0f 7d 0a [   ]*hsubpd \(%edx\),%xmm1
> -  3b:  f2 0f 7d d2 [   ]*hsubps %xmm2,%xmm2
> -  3f:  f2 0f 7d 1c 24 [        ]*hsubps \(%esp\),%xmm3
> -  44:  f2 0f f0 2e [   ]*lddqu  \(%esi\),%xmm5
> -  48:  0f 01 c8 [      ]*monitor %eax,%ecx,%edx
> -  4b:  0f 01 c8 [      ]*monitor %eax,%ecx,%edx
> -  4e:  f2 0f 12 f7 [   ]*movddup %xmm7,%xmm6
> -  52:  f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
> -  56:  f3 0f 16 01 [   ]*movshdup \(%ecx\),%xmm0
> -  5a:  f3 0f 16 ca [   ]*movshdup %xmm2,%xmm1
> -  5e:  f3 0f 12 13 [   ]*movsldup \(%ebx\),%xmm2
> -  62:  f3 0f 12 dc [   ]*movsldup %xmm4,%xmm3
> -  66:  0f 01 c9 [      ]*mwait  %eax,%ecx
> -  69:  0f 01 c9 [      ]*mwait  %eax,%ecx
> -  6c:  67 0f 01 c8 [   ]*monitor %ax,%ecx,%edx
> -  70:  67 0f 01 c8 [   ]*monitor %ax,%ecx,%edx
> -  74:  f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
> -  78:  f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
> +[      ]*[0-9a-f]+:    dd 88 90 90 90 90 [     ]*fisttpll -0x6f6f6f70\(%eax\)
> +[      ]*[0-9a-f]+:    66 0f 7c 65 00 [        ]*haddpd 0x0\(%ebp\),%xmm4
> +[      ]*[0-9a-f]+:    66 0f 7c ee [   ]*haddpd %xmm6,%xmm5
> +[      ]*[0-9a-f]+:    f2 0f 7c 37 [   ]*haddps \(%edi\),%xmm6
> +[      ]*[0-9a-f]+:    f2 0f 7c f8 [   ]*haddps %xmm0,%xmm7
> +[      ]*[0-9a-f]+:    66 0f 7d c1 [   ]*hsubpd %xmm1,%xmm0
> +[      ]*[0-9a-f]+:    66 0f 7d 0a [   ]*hsubpd \(%edx\),%xmm1
> +[      ]*[0-9a-f]+:    f2 0f 7d d2 [   ]*hsubps %xmm2,%xmm2
> +[      ]*[0-9a-f]+:    f2 0f 7d 1c 24 [        ]*hsubps \(%esp\),%xmm3
> +[      ]*[0-9a-f]+:    f2 0f f0 2e [   ]*lddqu  \(%esi\),%xmm5
> +[      ]*[0-9a-f]+:    0f 01 c8 [      ]*monitor %eax,%ecx,%edx
> +[      ]*[0-9a-f]+:    0f 01 c8 [      ]*monitor %eax,%ecx,%edx
> +[      ]*[0-9a-f]+:    f2 0f 12 f7 [   ]*movddup %xmm7,%xmm6
> +[      ]*[0-9a-f]+:    f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
> +[      ]*[0-9a-f]+:    f3 0f 16 01 [   ]*movshdup \(%ecx\),%xmm0
> +[      ]*[0-9a-f]+:    f3 0f 16 ca [   ]*movshdup %xmm2,%xmm1
> +[      ]*[0-9a-f]+:    f3 0f 12 13 [   ]*movsldup \(%ebx\),%xmm2
> +[      ]*[0-9a-f]+:    f3 0f 12 dc [   ]*movsldup %xmm4,%xmm3
> +[      ]*[0-9a-f]+:    0f 01 c9 [      ]*mwait  %eax,%ecx
> +[      ]*[0-9a-f]+:    0f 01 c9 [      ]*mwait  %eax,%ecx
> +[      ]*[0-9a-f]+:    67 0f 01 c8 [   ]*monitor %ax,%ecx,%edx
> +[      ]*[0-9a-f]+:    67 0f 01 c8 [   ]*monitor %ax,%ecx,%edx
> +[      ]*[0-9a-f]+:    f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
> +[      ]*[0-9a-f]+:    f2 0f 12 38 [   ]*movddup \(%eax\),%xmm7
>  [      ]*[0-9a-f]+:    0f 01 c8[       ]+monitor %eax,%ecx,%edx
>  [      ]*[0-9a-f]+:    67 0f 01 c8[    ]+monitor %ax,%ecx,%edx
>  [      ]*[0-9a-f]+:    0f 01 c9[       ]+mwait  %eax,%ecx
> --- a/gas/testsuite/gas/i386/sse3.s
> +++ b/gas/testsuite/gas/i386/sse3.s
> @@ -8,6 +8,7 @@ foo:
>         addsubps        %xmm4,%xmm3
>         fisttps         0x90909090(%eax)
>         fisttpl         0x90909090(%eax)
> +       fisttpq         0x90909090(%eax)
>         fisttpll        0x90909090(%eax)
>         haddpd          0x0(%ebp),%xmm4
>         haddpd          %xmm6,%xmm5
> --- a/gas/testsuite/gas/i386/sse3-intel.d
> +++ b/gas/testsuite/gas/i386/sse3-intel.d
> @@ -14,6 +14,7 @@ Disassembly of section .text:
>  [      ]*[0-9a-f]+:    df 88 90 90 90 90[      ]+fisttp WORD PTR \[eax-0x6f6f6f70\]
>  [      ]*[0-9a-f]+:    db 88 90 90 90 90[      ]+fisttp DWORD PTR \[eax-0x6f6f6f70\]
>  [      ]*[0-9a-f]+:    dd 88 90 90 90 90[      ]+fisttp QWORD PTR \[eax-0x6f6f6f70\]
> +[      ]*[0-9a-f]+:    dd 88 90 90 90 90[      ]+fisttp QWORD PTR \[eax-0x6f6f6f70\]
>  [      ]*[0-9a-f]+:    66 0f 7c 65 00[         ]+haddpd xmm4,(XMMWORD PTR )?\[ebp(\+0x0)\]
>  [      ]*[0-9a-f]+:    66 0f 7c ee[    ]+haddpd xmm5,xmm6
>  [      ]*[0-9a-f]+:    f2 0f 7c 37[    ]+haddps xmm6,(XMMWORD PTR )?\[edi\]
> --- a/gas/testsuite/gas/i386/x86-64-lfence-load.d
> +++ b/gas/testsuite/gas/i386/x86-64-lfence-load.d
> @@ -44,16 +44,21 @@ Disassembly of section .text:
>   +[a-f0-9]+:   0f ae e8                lfence
>   +[a-f0-9]+:   db 55 00                fistl  0x0\(%rbp\)
>   +[a-f0-9]+:   df 55 00                fists  0x0\(%rbp\)
> + +[a-f0-9]+:   db 5d 00                fistpl 0x0\(%rbp\)
> + +[a-f0-9]+:   df 5d 00                fistps 0x0\(%rbp\)
> + +[a-f0-9]+:   df 7d 00                fistpll 0x0\(%rbp\)
>   +[a-f0-9]+:   db 45 00                fildl  0x0\(%rbp\)
>   +[a-f0-9]+:   0f ae e8                lfence
>   +[a-f0-9]+:   df 45 00                filds  0x0\(%rbp\)
>   +[a-f0-9]+:   0f ae e8                lfence
> + +[a-f0-9]+:   df 6d 00                fildll 0x0\(%rbp\)
> + +[a-f0-9]+:   0f ae e8                lfence
>   +[a-f0-9]+:   9b dd 75 00             fsave  0x0\(%rbp\)
>   +[a-f0-9]+:   dd 65 00                frstor 0x0\(%rbp\)
>   +[a-f0-9]+:   0f ae e8                lfence
> - +[a-f0-9]+:   df 45 00                filds  0x0\(%rbp\)
> - +[a-f0-9]+:   0f ae e8                lfence
> + +[a-f0-9]+:   db 4d 00                fisttpl 0x0\(%rbp\)
>   +[a-f0-9]+:   df 4d 00                fisttps 0x0\(%rbp\)
> + +[a-f0-9]+:   dd 4d 00                fisttpll 0x0\(%rbp\)
>   +[a-f0-9]+:   d9 65 00                fldenv 0x0\(%rbp\)
>   +[a-f0-9]+:   0f ae e8                lfence
>   +[a-f0-9]+:   9b d9 75 00             fstenv 0x0\(%rbp\)
> --- a/gas/testsuite/gas/i386/x86-64-lfence-load.s
> +++ b/gas/testsuite/gas/i386/x86-64-lfence-load.s
> @@ -27,12 +27,17 @@ _start:
>         flds (%rbp)
>         fistl (%rbp)
>         fists (%rbp)
> +       fistpl (%rbp)
> +       fistps (%rbp)
> +       fistpq (%rbp)
>         fildl (%rbp)
>         filds (%rbp)
> +       fildq (%rbp)
>         fsave (%rbp)
>         frstor (%rbp)
> -       filds (%rbp)
> +       fisttpl (%rbp)
>         fisttps (%rbp)
> +       fisttpq (%rbp)
>         fldenv (%rbp)
>         fstenv (%rbp)
>         fadds  (%rbp)
> --- a/gas/testsuite/gas/i386/x86-64-sse3.d
> +++ b/gas/testsuite/gas/i386/x86-64-sse3.d
> @@ -13,6 +13,7 @@ Disassembly of section .text:
>  [      ]*[a-f0-9]+:    df 88 90 90 90 00 [     ]*fisttps 0x909090\(%rax\)
>  [      ]*[a-f0-9]+:    db 88 90 90 90 00 [     ]*fisttpl 0x909090\(%rax\)
>  [      ]*[a-f0-9]+:    dd 88 90 90 90 00 [     ]*fisttpll 0x909090\(%rax\)
> +[      ]*[a-f0-9]+:    dd 88 90 90 90 00 [     ]*fisttpll 0x909090\(%rax\)
>  [      ]*[a-f0-9]+:    66 0f 7c 65 00 [        ]*haddpd 0x0\(%rbp\),%xmm4
>  [      ]*[a-f0-9]+:    66 0f 7c ee [   ]*haddpd %xmm6,%xmm5
>  [      ]*[a-f0-9]+:    f2 0f 7c 37 [   ]*haddps \(%rdi\),%xmm6
> --- a/gas/testsuite/gas/i386/x86-64-sse3.s
> +++ b/gas/testsuite/gas/i386/x86-64-sse3.s
> @@ -8,6 +8,7 @@ foo:
>         addsubps        %xmm4,%xmm3
>         fisttps         0x909090(%rax)
>         fisttpl         0x909090(%rax)
> +       fisttpq         0x909090(%rax)
>         fisttpll        0x909090(%rax)
>         haddpd          0x0(%rbp),%xmm4
>         haddpd          %xmm6,%xmm5
> --- a/gas/testsuite/gas/i386/x86-64-sse3-intel.d
> +++ b/gas/testsuite/gas/i386/x86-64-sse3-intel.d
> @@ -14,6 +14,7 @@ Disassembly of section .text:
>  [      ]*[a-f0-9]+:    df 88 90 90 90 00[      ]+fisttp WORD PTR \[rax\+0x909090\]
>  [      ]*[a-f0-9]+:    db 88 90 90 90 00[      ]+fisttp DWORD PTR \[rax\+0x909090\]
>  [      ]*[a-f0-9]+:    dd 88 90 90 90 00[      ]+fisttp QWORD PTR \[rax\+0x909090\]
> +[      ]*[a-f0-9]+:    dd 88 90 90 90 00[      ]+fisttp QWORD PTR \[rax\+0x909090\]
>  [      ]*[a-f0-9]+:    66 0f 7c 65 00[         ]+haddpd xmm4,(XMMWORD PTR )?\[rbp(\+0x0)\]
>  [      ]*[a-f0-9]+:    66 0f 7c ee[    ]+haddpd xmm5,xmm6
>  [      ]*[a-f0-9]+:    f2 0f 7c 37[    ]+haddps xmm6,(XMMWORD PTR )?\[rdi\]
>


-- 
H.J.


More information about the Binutils mailing list