[PATCH] x86: don't mistake ordinary immediates for SAE / rounding control

[PATCH] x86: don't mistake ordinary immediates for SAE / rounding control

[Jan Beulich]

The way SAE templates are constructed was always puzzling me (including
the need for separate templates in the first place), and expressing the
extzra attribute via Imm8 actually has a bad effect: Ordinary immediates
would also be accepted, leading to an extra byte being added after the
instruction (i.e. generating bad code). Before re-working this (in
particular to accept proper Intel syntax there), fix the immediate issue
by adding the so far missing check.

--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -6285,6 +6285,11 @@ check_VecOperands (const insn_template *
 	  return 1;
 	}
     }
+  else if (t->opcode_modifier.sae)
+    {
+	i.error = unsupported_syntax;
+	return 1;
+    }
 
   /* Check the special Imm4 cases; must be the first operand.  */
   if (t->cpu_flags.bitfield.cpuxop && t->operands == 5)
--- a/gas/testsuite/gas/i386/inval-avx512f.l
+++ b/gas/testsuite/gas/i386/inval-avx512f.l
@@ -218,6 +218,8 @@
 .*:313: Error: .*unsupported broadcast for `vcvtneps2bf16'
 .*:316: Error: .*unsupported broadcast for `vcvtneps2bf16'
 .*:319: Error: .*unsupported broadcast for `vcvtneps2bf16'
+.*:321: Error: .*vaddps.*
+.*:322: Error: .*vcmpss.*
 GAS LISTING .*
 
 
@@ -569,6 +571,7 @@ GAS LISTING .*
 [ 	]*318 \?\?\?\? 62F27E38[ 	]+vcvtneps2bf16 \(%eax\)\{1to8\}, %xmm1
 [ 	]*318[ 	]+7208
 [ 	]*319[ 	]+vcvtneps2bf16 \(%eax\)\{1to16\}, %xmm1
-[ 	]*320[ 	]+\?\?\?\? 8DB42600 		\.p2align 4
-[ 	]*320[ 	]+0000008D 
-[ 	]*320[ 	]+7600
+[ 	]*320[ 	]*
+[ 	]*321[ 	]+vaddps \$0xcc, %zmm0, %zmm0, %zmm0
+[ 	]*322[ 	]+vcmpss \$0, \$0xcc, %xmm0, %xmm0, %k0
+#pass
--- a/gas/testsuite/gas/i386/inval-avx512f.s
+++ b/gas/testsuite/gas/i386/inval-avx512f.s
@@ -317,4 +317,8 @@ _start:
 	vcvtneps2bf16 (%eax){1to4}, %xmm1
 	vcvtneps2bf16 (%eax){1to8}, %xmm1
 	vcvtneps2bf16 (%eax){1to16}, %xmm1
+
+	vaddps $0xcc, %zmm0, %zmm0, %zmm0
+	vcmpss $0, $0xcc, %xmm0, %xmm0, %k0
+
 	.p2align 4

Re: [PATCH] x86: don't mistake ordinary immediates for SAE / rounding control

[H.J. Lu]

On Thu, Apr 14, 2022 at 7:12 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> The way SAE templates are constructed was always puzzling me (including
> the need for separate templates in the first place), and expressing the
> extzra attribute via Imm8 actually has a bad effect: Ordinary immediates
> would also be accepted, leading to an extra byte being added after the
> instruction (i.e. generating bad code). Before re-working this (in
> particular to accept proper Intel syntax there), fix the immediate issue
> by adding the so far missing check.
>
> --- a/gas/config/tc-i386.c
> +++ b/gas/config/tc-i386.c
> @@ -6285,6 +6285,11 @@ check_VecOperands (const insn_template *
>           return 1;
>         }
>      }
> +  else if (t->opcode_modifier.sae)
> +    {
> +       i.error = unsupported_syntax;
> +       return 1;
> +    }
>
>    /* Check the special Imm4 cases; must be the first operand.  */
>    if (t->cpu_flags.bitfield.cpuxop && t->operands == 5)
> --- a/gas/testsuite/gas/i386/inval-avx512f.l
> +++ b/gas/testsuite/gas/i386/inval-avx512f.l
> @@ -218,6 +218,8 @@
>  .*:313: Error: .*unsupported broadcast for `vcvtneps2bf16'
>  .*:316: Error: .*unsupported broadcast for `vcvtneps2bf16'
>  .*:319: Error: .*unsupported broadcast for `vcvtneps2bf16'
> +.*:321: Error: .*vaddps.*
> +.*:322: Error: .*vcmpss.*
>  GAS LISTING .*
>
>
> @@ -569,6 +571,7 @@ GAS LISTING .*
>  [      ]*318 \?\?\?\? 62F27E38[        ]+vcvtneps2bf16 \(%eax\)\{1to8\}, %xmm1
>  [      ]*318[  ]+7208
>  [      ]*319[  ]+vcvtneps2bf16 \(%eax\)\{1to16\}, %xmm1
> -[      ]*320[  ]+\?\?\?\? 8DB42600             \.p2align 4
> -[      ]*320[  ]+0000008D
> -[      ]*320[  ]+7600
> +[      ]*320[  ]*
> +[      ]*321[  ]+vaddps \$0xcc, %zmm0, %zmm0, %zmm0
> +[      ]*322[  ]+vcmpss \$0, \$0xcc, %xmm0, %xmm0, %k0
> +#pass
> --- a/gas/testsuite/gas/i386/inval-avx512f.s
> +++ b/gas/testsuite/gas/i386/inval-avx512f.s
> @@ -317,4 +317,8 @@ _start:
>         vcvtneps2bf16 (%eax){1to4}, %xmm1
>         vcvtneps2bf16 (%eax){1to8}, %xmm1
>         vcvtneps2bf16 (%eax){1to16}, %xmm1
> +
> +       vaddps $0xcc, %zmm0, %zmm0, %zmm0
> +       vcmpss $0, $0xcc, %xmm0, %xmm0, %k0
> +
>         .p2align 4
>

OK.

Thanks.

-- 
H.J.