https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79622
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |spop at gcc dot gnu.org
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
we fail to represent the PHI defining g5_23:
<bb 4> [83.33%] [count: INV]:
# l9_22 = PHI <0(3), l9_16(5)>
# g5_23 = PHI <-1(3), 0(5)>
# ivtmp_20 = PHI <5(3), ivtmp_13(5)>
dc[l9_22] = g5_23;
l9_16 = l9_22 + 1;
ivtmp_13 = ivtmp_20 - 1;
if (ivtmp_13 != 0)
goto <bb 5>; [80.00%] [count: INV]
else
goto <bb 6>; [20.00%] [count: INV]
<bb 5> [66.66%] [count: INV]:
goto <bb 4>; [100.00%] [count: INV]
[scheduler] original ast:
for (int c0 = 0; c0 <= 1; c0 += 1)
for (int c1 = 0; c1 <= 4; c1 += 1)
S_4(c0, c1);
not sure if this is a case of "cross-BB-scalar-vars" but that doesn't handle
non-SSA uses. So a similar case would be the following:
# _1 = PHI <_2(entry), 0(latch)>
The following patch fixes this by always marking non-SCEV analyzable
PHI defs as "writes".
Index: gcc/graphite-scop-detection.c
===================================================================
--- gcc/graphite-scop-detection.c (revision 252062)
+++ gcc/graphite-scop-detection.c (working copy)
@@ -1744,7 +1744,9 @@ build_cross_bb_scalars_def (scop_p scop,
gimple *use_stmt;
imm_use_iterator imm_iter;
FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, def)
- if (def_bb != gimple_bb (use_stmt) && !is_gimple_debug (use_stmt))
+ if ((def_bb != gimple_bb (use_stmt) && !is_gimple_debug (use_stmt))
+ /* ??? PR79622. */
+ || gimple_code (SSA_NAME_DEF_STMT (def)) == GIMPLE_PHI)
{
writes->safe_push (def);
DEBUG_PRINT (dp << "Adding scalar write: ";
passes the graphite.exp testsuites, otherwise untested.
Sebastian?
