procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[vect]Account for epilogue's peeling for gaps when checking if we have enough

Browse source 
niters for epilogue

gcc/ChangeLog:
2019-11-11  Andre Vieira  <andre.simoesdiasvieira@arm.com>

	* tree-vect-loop-manip.c (vect_do_peeling): Take epilogue gaps into
	account when checking if there are enough iterations to vectorize
	epilogue.

gcc/testsuite/ChangeLog:
2019-11-11  Andre Vieira  <andre.simoesdiasvieira@arm.com>

	* gcc.dg/vect/vect-reduc-epilogue-gaps.c: New test.

From-SVN: r278049
This commit is contained in:
Andre Vieira 2019-11-11 12:07:04 +00:00 • committed by Andre Vieira
parent 3f246567a4
commit 87b4725192
4 changed files with 59 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
gcc/ChangeLog
View file

@ -1,3 +1,9 @@
2019-11-11 Andre Vieira <andre.simoesdiasvieira@arm.com>
* tree-vect-loop-manip.c (vect_do_peeling): Take epilogue gaps into
account when checking if there are enough iterations to vectorize
epilogue.
2019-11-11 Tobias Burnus <tobias@codesourcery.com>
Kwok Cheung Yeung <kcy@codesourcery.com>

4
gcc/testsuite/ChangeLog
View file

@ -1,3 +1,7 @@
2019-11-11 Andre Vieira <andre.simoesdiasvieira@arm.com>
* gcc.dg/vect/vect-reduc-epilogue-gaps.c: New test.
2019-11-11 José Rui Faustino de Sousa <jrfsousa@gmail.com>
PR fortran/92142

45
gcc/testsuite/gcc.dg/vect/vect-reduc-epilogue-gaps.c Normal file
View file

@ -0,0 +1,45 @@
/* { dg-options "-O3 -fno-vect-cost-model" } */
struct {
float real;
float img;
} g[11];
float __attribute__ ((noclone))
foo_11 (void)
{
float sum = 0.0;
for (int i = 0; i < 11; ++i)
sum += g[i].real;
return sum;
}
float __attribute__ ((noclone))
foo_10 (void)
{
float sum = 0.0;
for (int i = 0; i < 10; ++i)
sum += g[i].real;
return sum;
}
int main (void)
{
float check_10 = 0.0;
float check_11 = 0.0;
for (int i = 0; i < 11; ++i)
{
asm volatile ("" : : : "memory");
g[i].real = (float) i;
g[i].img = (float) -i;
if (i < 10)
check_10 += (float) i;
check_11 += (float) i;
}
if (foo_10 () != check_10)
__builtin_abort ();
if (foo_11 () != check_11)
__builtin_abort ();
return 0;
}

5
gcc/tree-vect-loop-manip.c
View file

@ -2530,9 +2530,11 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1,
= eiters % lowest_vf + LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo);
unsigned int ratio;
unsigned int epilogue_gaps
= LOOP_VINFO_PEELING_FOR_GAPS (epilogue_vinfo);
while (!(constant_multiple_p (loop_vinfo->vector_size,
epilogue_vinfo->vector_size, &ratio)
&& eiters >= lowest_vf / ratio))
&& eiters >= lowest_vf / ratio + epilogue_gaps))
{
delete epilogue_vinfo;
epilogue_vinfo = NULL;
@ -2543,6 +2545,7 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1,
}
epilogue_vinfo = loop_vinfo->epilogue_vinfos[0];
loop_vinfo->epilogue_vinfos.ordered_remove (0);
epilogue_gaps = LOOP_VINFO_PEELING_FOR_GAPS (epilogue_vinfo);
}
}
/* Prolog loop may be skipped. */