[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 --- Comment #7 from CVS Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:eb032893675afea4b01cc6ad06a3e0dcfe9b51cd commit r12-4472-geb032893675afea4b01cc6ad06a3e0dcfe9b51cd Author: Richard Biener Date: Mon Oct 18 10:31:19 2021 +0200 tree-optimization/102788 - avoid spurious bool pattern fails Bool pattern recog is required for correctness since vectorized compares otherwise produce -1 for true so any context where bool is used as value and not as condition or mask needs to be replaced with CMP ? 1 : 0. When we fail to find a vector type for the result of such use we may not simply elide such transform since a new bool result can emerge when for example the cast_forwprop pattern is applied. So the following avoids failing of the bool pattern recog process and instead not assign a vector type for the stmt. 2021-10-18 Richard Biener PR tree-optimization/102788 * tree-vect-patterns.c (vect_init_pattern_stmt): Allow a NULL vectype. (vect_pattern_recog_1): Likewise. (vect_recog_bool_pattern): Continue matching the pattern even if we do not have a vector type for a conversion result. * g++.dg/vect/pr102788.cc: New testcase.
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 --- Comment #6 from Richard Biener --- Created attachment 51621 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51621=edit patch I am testing
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788
Richard Biener changed:
What|Removed |Added
CC||rguenth at gcc dot gnu.org,
||rsandifo at gcc dot gnu.org
--- Comment #5 from Richard Biener ---
The interesting thing is that the difference starts with pattern recog (for the
epilogue loop) where the main loop recognizes a bool pattern but the epilogue
one
does not. That's because 'long long unsigned int' doesn't have a vector type
with V4QImode.
But then vect_recog_cast_forwprop_pattern makes this cast irrelevant but we
do not re-visit the bool pattern as we only (re-)process the pattern def
sequence stmts but not the stmt itself:
/* If this statement has already been replaced with pattern statements,
leave the original statement alone, since the first match wins.
Instead try to match against the definition statements that feed
the main pattern statement. */
if (STMT_VINFO_IN_PATTERN_P (stmt_info))
{
gimple_stmt_iterator gsi;
for (gsi = gsi_start (STMT_VINFO_PATTERN_DEF_SEQ (stmt_info));
!gsi_end_p (gsi); gsi_next ())
vect_pattern_recog_1 (vinfo, recog_func,
vinfo->lookup_stmt (gsi_stmt (gsi)));
return;
}
and we're also not expecting to do this:
void
vect_mark_pattern_stmts (vec_info *vinfo,
stmt_vec_info orig_stmt_info, gimple *pattern_stmt,
tree pattern_vectype)
{
...
/* We shouldn't be replacing the main pattern statement. */
gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info)->stmt
!= orig_pattern_stmt);
but bool pattern recog is required for correctness, we are not going to
fail vectorization later.
Richard - you added most of the re-processing of patterns (and also the
forwprop pattern that triggers the failure situation). Can you share
insights and maybe fix this?
It's also really a latent issue. One could use __int128_t to trigger it
for V8QI vectorization, but that's likely it - the trigger is the FAIL
of the bool pattern recog (which should probably a fatal vectorization
FAIL in case the stmt is ending up relevant).
What "works" (for the testcase) is to not terminate vect_recog_bool_pattern
when we have no vector type for the LHS but allow NULL to be propagated
through the pattern machinery (nothing will in the end use that type
for this testcase). As said, since bool patterns are required for
correctness we cannot really early-out here but need to communicate
failure upthread. I'm going to test a patch with this route.
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 Richard Biener changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot gnu.org --- Comment #4 from Richard Biener --- I will have a look.
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 --- Comment #3 from Andrew Pinski --- (In reply to Jakub Jelinek from comment #1) > Started with r12-1075-g28484d00c45b7bf094a22a4fddf9ffdc7482c7e1 I just think that exposed the latent bug in the vectorizer as far as I can tell.
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 --- Comment #2 from Andrew Pinski --- Confirmed. ... mask__43.21_62 = vect_cst__60 != vect_cst__61; _43 = var_12.0_1 != 0; _3 = (long long unsigned int) _43; vect_patt_34.22_63 = VIEW_CONVERT_EXPR(mask__43.21_62); _26 = (unsigned char) _3; Somehow the vectorizer missed the conversion but has the conversion there. /app/example.cpp:19:23: note: add new stmt: mask__43.21_62 = vect_cst__60 != vect_cst__61; /app/example.cpp:19:23: note: -->vectorizing statement: _3 = (long long unsigned int) _43; /app/example.cpp:19:23: note: -->vectorizing statement: patt_34 = (unsigned char) _43; /app/example.cpp:19:23: note: transform statement. /app/example.cpp:19:23: note: vect_is_simple_use: operand var_12.0_1 != 0, type of def: internal /app/example.cpp:19:23: note: vect_is_simple_use: vectype vector(4) /app/example.cpp:19:23: note: transform assignment. /app/example.cpp:19:23: note: vect_get_vec_defs_for_operand: _43 /app/example.cpp:19:23: note: vect_is_simple_use: operand var_12.0_1 != 0, type of def: internal /app/example.cpp:19:23: note:def_stmt = _43 = var_12.0_1 != 0; /app/example.cpp:19:23: note: add new stmt: vect_patt_34.22_63 = VIEW_CONVERT_EXPR(mask__43.21_62); /app/example.cpp:19:23: note: extracting lane for live stmt _26 = (unsigned char) _3;
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 Jakub Jelinek changed: What|Removed |Added Priority|P3 |P1 Status|UNCONFIRMED |NEW Ever confirmed|0 |1 Last reconfirmed||2021-10-15 CC||jakub at gcc dot gnu.org, ||uros at gcc dot gnu.org --- Comment #1 from Jakub Jelinek --- Started with r12-1075-g28484d00c45b7bf094a22a4fddf9ffdc7482c7e1
[Bug tree-optimization/102788] [12 Regression] Wrong code with -O3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102788 Andrew Pinski changed: What|Removed |Added Target Milestone|--- |12.0 Summary|Wrong code with -O3 |[12 Regression] Wrong code ||with -O3
