As reported in PR117079, commit ab18785840d7b8 broke the test pr105493.c.
When looking at the generated code, we can see that the generated code
is vectorized differently, resulting in a reduction from 225 instructions
down to 109. On the performance side, no changes were measured on a 5950X.
This patch adjusts the test condition to fit how the function gets
vectorized after ab18785840d7b8 (and probably further related changes).
Signed-off-by: Christoph Müllner <christoph.muell...@vrull.eu>
PR target/117079
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr105493.c: Fix expected vectorization
---
gcc/testsuite/gcc.target/i386/pr105493.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gcc/testsuite/gcc.target/i386/pr105493.c
b/gcc/testsuite/gcc.target/i386/pr105493.c
index c6fd16753cd9..fb0bf8aa0af7 100644
--- a/gcc/testsuite/gcc.target/i386/pr105493.c
+++ b/gcc/testsuite/gcc.target/i386/pr105493.c
@@ -48,4 +48,4 @@ foo ( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
/* The first loop should be vectorized, which will eliminate redundant stores
and loads. */
-/* { dg-final { scan-tree-dump-times " MEM <vector\\\(4\\\) unsigned int>
\\\[\[\^\]\]\*\\\] = " 4 "slp1" } } */
+/* { dg-final { scan-tree-dump "= MEM <vector\\\(8\\\) unsigned char>
\\\[\[\^\]\]\*\\\]" "slp1" } } */
--
2.47.1
