[PATCH] bswap: Improve perform_symbolic_merge [PR103376]

Jakub Jelinek jakub@redhat.com
Thu Nov 25 09:44:13 GMT 2021


On Thu, Nov 25, 2021 at 09:21:37AM +0100, Richard Biener wrote:
> OK if you can add a testcase that exercises this "feature".

Sure, that is easy.

Here is what I've committed.  f2 tests the x | x = x handling in it,
f3 tests x | y = unknown instead of punting, f4 tests x ^ x = 0
and f5 tests x ^ y = unknown.  Without the patch only f2 is optimized
to __builtin_bswap32, with the patch all of them.

2021-11-25  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/103376
	* gimple-ssa-store-merging.c (perform_symbolic_merge): For
	BIT_IOR_EXPR, if masked1 && masked2 && masked1 != masked2, don't
	punt, but set the corresponding result byte to MARKER_BYTE_UNKNOWN.
	For BIT_XOR_EXPR similarly and if masked1 == masked2 and the
	byte isn't MARKER_BYTE_UNKNOWN, set the corresponding result byte to
	0.

	* gcc.dg/optimize-bswapsi-7.c: New test.

--- gcc/gimple-ssa-store-merging.c.jj	2021-11-24 09:54:37.684365460 +0100
+++ gcc/gimple-ssa-store-merging.c	2021-11-24 11:18:54.422226266 +0100
@@ -556,6 +556,7 @@ perform_symbolic_merge (gimple *source_s
   n->bytepos = n_start->bytepos;
   n->type = n_start->type;
   size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
+  uint64_t res_n = n1->n | n2->n;
 
   for (i = 0, mask = MARKER_MASK; i < size; i++, mask <<= BITS_PER_MARKER)
     {
@@ -563,12 +564,33 @@ perform_symbolic_merge (gimple *source_s
 
       masked1 = n1->n & mask;
       masked2 = n2->n & mask;
-      /* For BIT_XOR_EXPR or PLUS_EXPR, at least one of masked1 and masked2
-	 has to be 0, for BIT_IOR_EXPR x | x is still x.  */
-      if (masked1 && masked2 && (code != BIT_IOR_EXPR || masked1 != masked2))
-	return NULL;
+      /* If at least one byte is 0, all of 0 | x == 0 ^ x == 0 + x == x.  */
+      if (masked1 && masked2)
+	{
+	  /* + can carry into upper bits, just punt.  */
+	  if (code == PLUS_EXPR)
+	    return NULL;
+	  /* x | x is still x.  */
+	  if (code == BIT_IOR_EXPR && masked1 == masked2)
+	    continue;
+	  if (code == BIT_XOR_EXPR)
+	    {
+	      /* x ^ x is 0, but MARKER_BYTE_UNKNOWN stands for
+		 unknown values and unknown ^ unknown is unknown.  */
+	      if (masked1 == masked2
+		  && masked1 != ((uint64_t) MARKER_BYTE_UNKNOWN
+				 << i * BITS_PER_MARKER))
+		{
+		  res_n &= ~mask;
+		  continue;
+		}
+	    }
+	  /* Otherwise set the byte to unknown, it might still be
+	     later masked off.  */
+	  res_n |= mask;
+	}
     }
-  n->n = n1->n | n2->n;
+  n->n = res_n;
   n->n_ops = n1->n_ops + n2->n_ops;
 
   return source_stmt;
--- gcc/testsuite/gcc.dg/optimize-bswapsi-7.c.jj	2021-11-25 10:36:03.847529686 +0100
+++ gcc/testsuite/gcc.dg/optimize-bswapsi-7.c	2021-11-25 10:35:46.522778192 +0100
@@ -0,0 +1,37 @@
+/* PR tree-optimization/103376 */
+/* { dg-do compile } */
+/* { dg-require-effective-target bswap } */
+/* { dg-options "-O2 -fno-tree-vectorize -fdump-tree-optimized" } */
+/* { dg-additional-options "-march=z900" { target s390-*-* } } */
+
+static unsigned int
+f1 (unsigned int x)
+{
+  return (x << 24) | (x >> 8);
+}
+
+unsigned int
+f2 (unsigned *p)
+{
+  return ((f1 (p[0]) | (p[0] >> 8)) & 0xff000000U) | (p[0] >> 24) | ((p[0] & 0xff00U) << 8) | ((p[0] & 0xff0000U) >> 8);
+}
+
+unsigned int
+f3 (unsigned *p)
+{
+  return ((f1 (p[0]) | (p[0] & 0x00ff00ffU)) & 0xff00ff00U) | (f1 (f1 (f1 (p[0]))) & 0x00ff00ffU);
+}
+
+unsigned int
+f4 (unsigned *p)
+{
+  return (f1 (p[0]) ^ (p[0] >> 8)) ^ (p[0] >> 24) ^ ((p[0] & 0xff00U) << 8) ^ ((p[0] & 0xff0000U) >> 8);
+}
+
+unsigned int
+f5 (unsigned *p)
+{
+  return (((f1 (p[0]) | (p[0] >> 16)) ^ (p[0] >> 8)) & 0xffff0000U) ^ (p[0] >> 24) ^ ((p[0] & 0xff00U) << 8) ^ ((p[0] & 0xff0000U) >> 8);
+}
+
+/* { dg-final { scan-tree-dump-times "= __builtin_bswap32 \\\(" 4 "optimized" } } */


	Jakub



More information about the Gcc-patches mailing list