public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug tree-optimization/64031] New: Vectorization of max/min is not robust enough
@ 2014-11-23 12:59 jacques-henri.jourdan at inria dot Fr
2014-11-23 14:08 ` [Bug tree-optimization/64031] " glisse at gcc dot gnu.org
2023-07-27 13:47 ` [Bug tree-optimization/64031] (un-)conditional execution state is not preserved by PRE/sink rguenth at gcc dot gnu.org
0 siblings, 2 replies; 3+ messages in thread
From: jacques-henri.jourdan at inria dot Fr @ 2014-11-23 12:59 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64031
Bug ID: 64031
Summary: Vectorization of max/min is not robust enough
Product: gcc
Version: 4.9.3
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jacques-henri.jourdan at inria dot Fr
The following code is sucessfully vectorized (using the minps instruction):
const int SIZE = 1<<15;
void test9(float * b)
{
unsigned i;
float *y =__builtin_assume_aligned(b, 16);
for (i = 0; i < SIZE; i++)
{
float f = y[i];
float f2 = f < f*f ? f : f*f;
y[i] = f2;
}
}
But not the following slightly modified version:
void test9(float * b)
{
unsigned i;
float *y =__builtin_assume_aligned(b, 16);
for (i = 0; i < SIZE; i++)
{
float f = y[i];
float f2 = f < f*f ? f : f*f;
y[i] = f2*f2;
}
}
Actually, it seems like vectorization of max/min operations fails as soon as
some computation is done with the result of this min/max operation.
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug tree-optimization/64031] Vectorization of max/min is not robust enough
2014-11-23 12:59 [Bug tree-optimization/64031] New: Vectorization of max/min is not robust enough jacques-henri.jourdan at inria dot Fr
@ 2014-11-23 14:08 ` glisse at gcc dot gnu.org
2023-07-27 13:47 ` [Bug tree-optimization/64031] (un-)conditional execution state is not preserved by PRE/sink rguenth at gcc dot gnu.org
1 sibling, 0 replies; 3+ messages in thread
From: glisse at gcc dot gnu.org @ 2014-11-23 14:08 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64031
--- Comment #1 from Marc Glisse <glisse at gcc dot gnu.org> ---
-ffast-math lets it vectorize (using min directly). Otherwise, for the first
test, ifcvt gives (a<b)?a:b which is vectorized as such, but for the second
test, it doesn't unless we pass -fno-trapping-math. That seems a bit
inconsistent.
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug tree-optimization/64031] (un-)conditional execution state is not preserved by PRE/sink
2014-11-23 12:59 [Bug tree-optimization/64031] New: Vectorization of max/min is not robust enough jacques-henri.jourdan at inria dot Fr
2014-11-23 14:08 ` [Bug tree-optimization/64031] " glisse at gcc dot gnu.org
@ 2023-07-27 13:47 ` rguenth at gcc dot gnu.org
1 sibling, 0 replies; 3+ messages in thread
From: rguenth at gcc dot gnu.org @ 2023-07-27 13:47 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64031
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |14.0
Status|NEW |RESOLVED
Resolution|--- |FIXED
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
This is now fixed in GCC 14.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2023-07-27 13:47 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-11-23 12:59 [Bug tree-optimization/64031] New: Vectorization of max/min is not robust enough jacques-henri.jourdan at inria dot Fr
2014-11-23 14:08 ` [Bug tree-optimization/64031] " glisse at gcc dot gnu.org
2023-07-27 13:47 ` [Bug tree-optimization/64031] (un-)conditional execution state is not preserved by PRE/sink rguenth at gcc dot gnu.org
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).