https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126366
--- Comment #2 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
Yes masked peeling with PEELED loops are still broken
diff --git a/gcc/tree-vect-data-refs.cc b/gcc/tree-vect-data-refs.cc
index 0e0754769ae..5e9bc869241 100644
--- a/gcc/tree-vect-data-refs.cc
+++ b/gcc/tree-vect-data-refs.cc
@@ -2645,11 +2645,7 @@ vect_enhance_data_refs_alignment (loop_vec_info
loop_vinfo)
if (!vect_can_advance_ivs_p (loop_vinfo)
|| !slpeel_can_duplicate_loop_p (loop, LOOP_VINFO_MAIN_EXIT
(loop_vinfo),
loop_preheader_edge (loop))
- || loop->inner
- /* We don't currently maintain the LCSSA for prologue peeled inversed
- loops. */
- || (LOOP_VINFO_EARLY_BREAKS_VECT_PEELED (loop_vinfo)
- && !LOOP_VINFO_NITERS_UNCOUNTED_P (loop_vinfo)))
+ || loop->inner)
do_peeling = false;
struct _vect_peel_extended_info peel_for_known_alignment;
due to a junk threading of the induction value
;; basic block 24, loop depth 0, count 65751655 (estimated locally, freq
0.4096), maybe hot
;; prev block 6, next block 18, flags: (NEW)
;; pred: 6 [10.0% (guessed)] count:65751655 (estimated locally, freq
0.4096) (FALSE_VALUE)
# RANGE [irange] unsigned int [0, 99] MASK 0x7f VALUE 0x0
# d_98 = PHI <d_11(6)>
_62 = niters.4_52 - POLY_INT_CST [4, 4];
;; succ: 18 [always] count:65751655 (estimated locally, freq 0.4096)
(FALLTHRU)
;; basic block 18, loop depth 0, count 77309413 (estimated locally, freq
0.4816), maybe hot
;; Invalid sum of incoming counts 143061068 (estimated locally, freq 0.8912),
should be 77309413 (estimated locally, freq 0.4816)
;; prev block 24, next block 22, flags: (NEW)
;; pred: 4 [10.0% (guessed)] count:77309413 (estimated locally, freq
0.4816) (FALSE_VALUE,EXECUTABLE)
;; 24 [always] count:65751655 (estimated locally, freq 0.4096)
(FALLTHRU)
# d_59 = PHI <d_11(4), d_11(24)>
;; succ: 22 [always] count:77309413 (estimated locally, freq 0.4816)
(FALLTHRU)
;; basic block 22, loop depth 0, count 128419524 (estimated locally, freq
0.8000), maybe hot
;; Invalid sum of incoming counts 90151366 (estimated locally, freq 0.5616),
should be 128419524 (estimated locally, freq 0.8000)
;; prev block 18, next block 19, flags: (NEW)
;; pred: 18 [always] count:77309413 (estimated locally, freq 0.4816)
(FALLTHRU)
;; 13 [10.0% (guessed)] count:12841953 (estimated locally, freq
0.0800) (TRUE_VALUE)
# d_61 = PHI <_63(18), 0(13)>
;; succ: 19 [always] count:128419524 (estimated locally, freq 0.8000)
(FALLTHRU)
;; basic block 19, loop depth 1, count 858993463 (estimated locally, freq
5.3512), maybe hot
;; prev block 22, next block 20, flags: (NEW, REACHABLE, VISITED)
;; pred: 22 [always] count:128419524 (estimated locally, freq 0.8000)
(FALLTHRU)
;; 20 [always] count:730573940 (estimated locally, freq 4.5512)
(FALLTHRU,DFS_BACK,EXECUTABLE)
_63 isn't even in the same subgraph.
Will fix.
