[PATCH] Fix PR63266: Keep track of impact of sign extension in bswap

[Thomas Preud'homme]

> From: Christophe Lyon [mailto:christophe.lyon@linaro.org]
> Sent: Thursday, September 25, 2014 4:28 AM

> 
> Hi Thomas,

Hi Christophe,

> 
> Although I could notice the improvement:
> Pass disappears           [PASS =>     ]:
>   gcc.dg/optimize-bswapsi-1.c scan-tree-dump-times bswap "32 bit
> bswap
> implementation found at" 3
> New pass                  [     => PASS]:
>   gcc.dg/optimize-bswapsi-1.c scan-tree-dump-times bswap "32 bit
> bswap
> implementation found at" 4
> 
> for arm-*, armeb-* and aarch64-* targets, there is no change for
> aarch64_be: is this expected?

No, but neither is this:

@@ -1905,11 +1913,10 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
 
 	    /* Sign extension: result is dependent on the value.  */
 	    old_type_size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
-	    if (!TYPE_UNSIGNED (n->type)
-		&& type_size > old_type_size
-		&& n->n & ((uint64_t) 0xff << ((old_type_size - 1)
-					       * BITS_PER_MARKER)))
-	      return NULL;
+	    if (!TYPE_UNSIGNED (n->type) && type_size > old_type_size
+		&& HEAD_MARKER (n->n, old_type_size))
+	      for (i = 0; i < type_size - old_type_size; i++)
+		n->n |= MARKER_BYTE_UNKNOWN << (type_size - 1 - i);
 
 	    if (type_size < 64 / BITS_PER_MARKER)
 	      {

type_size - 1 - I gives a number of marker bytes to shift. I forgot to multiply by the number of bits in a marker. Can you do the change locally and tell me if the test now succeed for aarch64_be?

Best regards,

Thomas