https://gcc.gnu.org/g:166a776b590eeed1282e65ce0b4b489d780ee0be

commit r16-6811-g166a776b590eeed1282e65ce0b4b489d780ee0be
Author: Jeff Law <[email protected]>
Date:   Thu Jan 15 12:49:55 2026 -0700

    Fix RISC-V test after recent vectorizer changes
    
    Richi's patch to fix costing of permuted contiguous loads changed the code
    generation for pr122445.c in the testsuite.  After reviewing the code, it 
looks
    clearly better to me -- essentially covering 2X as many elements per vector 
op
    with fewer total vector ops in the end (without resulting to LMUL > 1).
    
    I've verified this fixes the overnight regressions seen on riscv{32,64}-elf.
    The bootstrap and regression tests are in flight, but won't finish for many
    more hours.
    
    gcc/testsuite
            * gcc.target/riscv/rvv/autovec/pr122445.c: Adjust expected output.

Diff:
---
 gcc/testsuite/gcc.target/riscv/rvv/autovec/pr122445.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr122445.c 
b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr122445.c
index 47368684faa1..cd046647656a 100644
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr122445.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr122445.c
@@ -22,5 +22,5 @@ x264_add4x4_idct (uint8_t *p_dst, int16_t d[16])
     }
 }
 
-/* { dg-final { scan-rtl-dump "Propagating AVL: \\(const_int 4" "avlprop" } } 
*/
+/* { dg-final { scan-rtl-dump "Propagating AVL: \\(const_int 8" "avlprop" } } 
*/
 /* { dg-final { scan-rtl-dump-not "Propagating AVL: \\(const_int 1" "avlprop" 
} } */

Reply via email to