[PATCH] lra: Fix up debug_p handling in lra_substitute_pseudo [PR104778]

[PATCH] lra: Fix up debug_p handling in lra_substitute_pseudo [PR104778]

[Jakub Jelinek]

Hi!

The following testcase ICEs on powerpc-linux, because lra_substitute_pseudo
substitutes (const_int 1) into a subreg operand.  First a subreg of subreg
of a reg appears in a debug insn (which surely is invalid outside of
debug insns, but in debug insns we allow even what is normally invalid in
RTL like subregs which the target doesn't like, because either dwarf2out
is able to handle it, or we just throw away the location expression,
making some var <optimized out>.

lra_substitute_pseudo already has some code to deal with specifically
SUBREG of REG with the REG being substituted for VOIDmode constant,
but that doesn't cover this case, so the following patch extends
lra_substitute_pseudo for debug_p mode to treat stuff like e.g.
combiner's subst function to ensure we don't lose mode which is essential
for the IL.

Bootstrapped/regtested on {powerpc64{,le},x86_64,i686}-linux, ok for trunk?

2022-03-12  Jakub Jelinek  <jakub@redhat.com>

	PR debug/104778
	* lra.cc (lra_substitute_pseudo): For debug_p mode, simplify
	SUBREG, ZERO_EXTEND, SIGN_EXTEND, FLOAT or UNSIGNED_FLOAT if recursive
	call simplified the first operand into VOIDmode constant.

	* gcc.target/powerpc/pr104778.c: New test.

--- gcc/lra.cc.jj	2022-02-04 14:36:55.375600131 +0100
+++ gcc/lra.cc	2022-03-11 18:47:15.555025540 +0100
@@ -2015,8 +2015,39 @@ lra_substitute_pseudo (rtx *loc, int old
     {
       if (fmt[i] == 'e')
 	{
-	  if (lra_substitute_pseudo (&XEXP (x, i), old_regno,
-				     new_reg, subreg_p, debug_p))
+	  if (debug_p
+	      && i == 0
+	      && (code == SUBREG
+		  || code == ZERO_EXTEND
+		  || code == SIGN_EXTEND
+		  || code == FLOAT
+		  || code == UNSIGNED_FLOAT))
+	    {
+	      rtx y = XEXP (x, 0);
+	      if (lra_substitute_pseudo (&y, old_regno,
+					 new_reg, subreg_p, debug_p))
+		{
+		  result = true;
+		  if (CONST_SCALAR_INT_P (y))
+		    {
+		      if (code == SUBREG)
+			y = simplify_subreg (GET_MODE (x), y,
+					     GET_MODE (SUBREG_REG (x)),
+					     SUBREG_BYTE (x));
+		      else
+			y = simplify_unary_operation (code, GET_MODE (x), y,
+						      GET_MODE (XEXP (x, 0)));
+		      if (y)
+			*loc = y;
+		      else
+			*loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
+		    }
+		  else
+		    XEXP (x, 0) = y;
+		}
+	    }
+	  else if (lra_substitute_pseudo (&XEXP (x, i), old_regno,
+					  new_reg, subreg_p, debug_p))
 	    result = true;
 	}
       else if (fmt[i] == 'E')
--- gcc/testsuite/gcc.target/powerpc/pr104778.c.jj	2022-03-11 18:54:09.455264514 +0100
+++ gcc/testsuite/gcc.target/powerpc/pr104778.c	2022-03-11 18:52:57.216269904 +0100
@@ -0,0 +1,51 @@
+/* PR debug/104778 */
+/* { dg-do compile } */
+/* { dg-options "-mcmpb -Og -g" } */
+/* { dg-additional-options "-fpie" { target pie } } */
+
+unsigned long long int p;
+short int m, n;
+
+void
+foo (double u, int v, int x, int y, int z)
+{
+  long long int a = v;
+  short int b = v;
+  int c = 0, d = m, e = u;
+
+  if (n)
+    {
+      int q = b;
+
+      while (p / 1.0)
+        c = 0;
+
+      if (n * n == (d + 1) / (1LL << x))
+        a = 1;
+
+      b = u;
+      while (d)
+        {
+          u = m + 1ULL;
+          b = a - (unsigned long long int) u + a + (char) (u + 1.0);
+          d = (v - 1LL) * n / d + q + x;
+          q = m;
+        }
+    }
+
+  while (c < 1)
+    {
+      int r;
+
+      if (m == y)
+        m = e * z;
+
+      e = !a;
+
+      while (!r)
+        ;
+
+      if (b)
+        m = d;
+    }
+}

	Jakub

Re: [PATCH] lra: Fix up debug_p handling in lra_substitute_pseudo [PR104778]

[Vladimir Makarov]

On 2022-03-12 14:37, Jakub Jelinek wrote:
> Hi!
>
> The following testcase ICEs on powerpc-linux, because lra_substitute_pseudo
> substitutes (const_int 1) into a subreg operand.  First a subreg of subreg
> of a reg appears in a debug insn (which surely is invalid outside of
> debug insns, but in debug insns we allow even what is normally invalid in
> RTL like subregs which the target doesn't like, because either dwarf2out
> is able to handle it, or we just throw away the location expression,
> making some var <optimized out>.
>
> lra_substitute_pseudo already has some code to deal with specifically
> SUBREG of REG with the REG being substituted for VOIDmode constant,
> but that doesn't cover this case, so the following patch extends
> lra_substitute_pseudo for debug_p mode to treat stuff like e.g.
> combiner's subst function to ensure we don't lose mode which is essential
> for the IL.
>
> Bootstrapped/regtested on {powerpc64{,le},x86_64,i686}-linux, ok for trunk?


Sure.  Thank you for working on this PR, Jakub.


> 2022-03-12  Jakub Jelinek  <jakub@redhat.com>
>
> 	PR debug/104778
> 	* lra.cc (lra_substitute_pseudo): For debug_p mode, simplify
> 	SUBREG, ZERO_EXTEND, SIGN_EXTEND, FLOAT or UNSIGNED_FLOAT if recursive
> 	call simplified the first operand into VOIDmode constant.
>
> 	* gcc.target/powerpc/pr104778.c: New test.
>