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

Generate avx_vzeroupper after reload is completed.

Browse source 
gcc/

2010-11-03  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/46285
	* config/i386/i386.c (ix86_expand_call): Conditionally decorate
	calls with UNSPEC_CALL_NEEDS_VZEROUPPER unspec before reload
	is completed.

gcc/testsuite/

2010-11-03  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/46285
	* gcc.target/i386/pr46285.c.

From-SVN: r166240
This commit is contained in:
H.J. Lu 2010-11-03 12:02:06 +00:00 • committed by H.J. Lu
parent 02ff830b58
commit 72fdf9e9c4
4 changed files with 27 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
gcc/ChangeLog
View file

@ -1,3 +1,10 @@
2010-11-03 H.J. Lu <hongjiu.lu@intel.com>
PR target/46285
* config/i386/i386.c (ix86_expand_call): Conditionally decorate
calls with UNSPEC_CALL_NEEDS_VZEROUPPER unspec before reload
is completed.
2010-11-03 Richard Guenther <rguenther@suse.de>
PR tree-optimization/46217

14
gcc/config/i386/i386.c
View file

@ -21540,10 +21540,16 @@ ix86_expand_call (rtx retval, rtx fnaddr, rtx callarg1,
else
avx256 = call_no_avx256;
unspec = gen_rtx_UNSPEC (VOIDmode,
gen_rtvec (1, GEN_INT (avx256)),
UNSPEC_CALL_NEEDS_VZEROUPPER);
call = gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, call, unspec));
if (reload_completed)
emit_insn (gen_avx_vzeroupper (GEN_INT (avx256)));
else
{
unspec = gen_rtx_UNSPEC (VOIDmode,
gen_rtvec (1, GEN_INT (avx256)),
UNSPEC_CALL_NEEDS_VZEROUPPER);
call = gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (2, call, unspec));
}
}
call = emit_call_insn (call);

5
gcc/testsuite/ChangeLog
View file

@ -1,3 +1,8 @@
2010-11-03 H.J. Lu <hongjiu.lu@intel.com>
PR target/46285
* gcc.target/i386/pr46285.c.
2010-11-03 Richard Guenther <rguenther@suse.de>
PR tree-optimization/46217

5
gcc/testsuite/gcc.target/i386/pr46285.c Normal file
View file

@ -0,0 +1,5 @@
/* { dg-do compile } */
/* { dg-options "-mavx -fsplit-stack -mtune=generic" } */
typedef char __m256 __attribute__ ((__vector_size__ (32)));
void foo (__m256 x) {}