[PATCH] expand: Fix up expand_cond_expr_using_cmove [PR106030]

[PATCH] expand: Fix up expand_cond_expr_using_cmove [PR106030]

[Jakub Jelinek]

Hi!

If expand_cond_expr_using_cmove can't find a cmove optab for a particular
mode, it tries to promote the mode and perform the cmove in the promoted
mode.

The testcase in the patch ICEs on arm because in that case we pass temp which
has the promoted mode (SImode) as target to expand_operands where the
operands have the non-promoted mode (QImode).
Later on the function uses paradoxical subregs:
  if (GET_MODE (op1) != mode)
    op1 = gen_lowpart (mode, op1);

  if (GET_MODE (op2) != mode)
    op2 = gen_lowpart (mode, op2);
to change the operand modes.

The following patch fixes it by passing NULL_RTX as target if it has
promoted mode.

Bootstrapped/regtested on x86_64-linux and i686-linux and tested with cross
to arm on the testcase, ok for trunk?

2022-06-21  Jakub Jelinek  <jakub@redhat.com>

	PR middle-end/106030
	* expr.cc (expand_cond_expr_using_cmove): Pass NULL_RTX instead of
	temp to expand_operands if mode has been promoted.

	* gcc.c-torture/compile/pr106030.c: New test.

--- gcc/expr.cc.jj	2022-06-10 21:19:12.087803729 +0200
+++ gcc/expr.cc	2022-06-20 14:10:07.941681072 +0200
@@ -8832,7 +8832,8 @@ expand_cond_expr_using_cmove (tree treeo
   expanding_cond_expr_using_cmove = true;
   start_sequence ();
   expand_operands (treeop1, treeop2,
-		   temp, &op1, &op2, EXPAND_NORMAL);
+		   mode == orig_mode ? temp : NULL_RTX, &op1, &op2,
+		   EXPAND_NORMAL);
 
   if (TREE_CODE (treeop0) == SSA_NAME
       && (srcstmt = get_def_for_expr_class (treeop0, tcc_comparison)))
--- gcc/testsuite/gcc.c-torture/compile/pr106030.c.jj	2022-06-20 14:49:37.618142233 +0200
+++ gcc/testsuite/gcc.c-torture/compile/pr106030.c	2022-06-20 14:49:19.016382117 +0200
@@ -0,0 +1,16 @@
+/* PR middle-end/106030 */
+
+int a, b, c;
+
+char
+foo (int x, int y)
+{
+  return x * y;
+}
+
+void
+bar (void)
+{
+  char d = (foo <= b) * a;
+  c = foo (2 != bar, d);
+}

	Jakub

Re: [PATCH] expand: Fix up expand_cond_expr_using_cmove [PR106030]

[Richard Biener]

On Tue, 21 Jun 2022, Jakub Jelinek wrote:

> Hi!
> 
> If expand_cond_expr_using_cmove can't find a cmove optab for a particular
> mode, it tries to promote the mode and perform the cmove in the promoted
> mode.
> 
> The testcase in the patch ICEs on arm because in that case we pass temp which
> has the promoted mode (SImode) as target to expand_operands where the
> operands have the non-promoted mode (QImode).
> Later on the function uses paradoxical subregs:
>   if (GET_MODE (op1) != mode)
>     op1 = gen_lowpart (mode, op1);
> 
>   if (GET_MODE (op2) != mode)
>     op2 = gen_lowpart (mode, op2);
> to change the operand modes.
> 
> The following patch fixes it by passing NULL_RTX as target if it has
> promoted mode.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux and tested with cross
> to arm on the testcase, ok for trunk?

OK.

Thanks,
Richard.

> 2022-06-21  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR middle-end/106030
> 	* expr.cc (expand_cond_expr_using_cmove): Pass NULL_RTX instead of
> 	temp to expand_operands if mode has been promoted.
> 
> 	* gcc.c-torture/compile/pr106030.c: New test.
> 
> --- gcc/expr.cc.jj	2022-06-10 21:19:12.087803729 +0200
> +++ gcc/expr.cc	2022-06-20 14:10:07.941681072 +0200
> @@ -8832,7 +8832,8 @@ expand_cond_expr_using_cmove (tree treeo
>    expanding_cond_expr_using_cmove = true;
>    start_sequence ();
>    expand_operands (treeop1, treeop2,
> -		   temp, &op1, &op2, EXPAND_NORMAL);
> +		   mode == orig_mode ? temp : NULL_RTX, &op1, &op2,
> +		   EXPAND_NORMAL);
>  
>    if (TREE_CODE (treeop0) == SSA_NAME
>        && (srcstmt = get_def_for_expr_class (treeop0, tcc_comparison)))
> --- gcc/testsuite/gcc.c-torture/compile/pr106030.c.jj	2022-06-20 14:49:37.618142233 +0200
> +++ gcc/testsuite/gcc.c-torture/compile/pr106030.c	2022-06-20 14:49:19.016382117 +0200
> @@ -0,0 +1,16 @@
> +/* PR middle-end/106030 */
> +
> +int a, b, c;
> +
> +char
> +foo (int x, int y)
> +{
> +  return x * y;
> +}
> +
> +void
> +bar (void)
> +{
> +  char d = (foo <= b) * a;
> +  c = foo (2 != bar, d);
> +}
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstraße 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)