Hi!
On Sat, Nov 05, 2011 at 10:50:44AM +0100, Jakub Jelinek wrote:
> On the following testcase with -m64 -O3 -mavx2 (but it is just an example,
> you can replace the loop there with any code that doesn't touch the
> stack or frame pointer at all), only f3 is shrink wrapped and in that case
> it on the other side doesn't add vzeroupper before leaving the AVX using
> code that it IMNSHO should. But I wonder why we can't shrink-wrap also
Here is a quick hack that deals with the missing vzeroupper issue.
Probably it would be nicer to create a helper in i386.c for that though,
because call_no_avx256 is an enum private to i386.c.
--- gcc/config/i386/i386.md.jj 2011-11-04 07:49:41.000000000 +0100
+++ gcc/config/i386/i386.md 2011-11-05 14:00:32.000000000 +0100
@@ -11725,6 +11725,12 @@ (define_expand "return"
[(simple_return)]
"ix86_can_use_return_insn_p ()"
{
+ /* Emit vzeroupper if needed. */
+ if (TARGET_VZEROUPPER
+ && !TREE_THIS_VOLATILE (cfun->decl)
+ && !cfun->machine->caller_return_avx256_p)
+ emit_insn (gen_avx_vzeroupper (const2_rtx));
+
if (crtl->args.pops_args)
{
rtx popc = GEN_INT (crtl->args.pops_args);
@@ -11741,6 +11747,12 @@ (define_expand "simple_return"
[(simple_return)]
"!TARGET_SEH"
{
+ /* Emit vzeroupper if needed. */
+ if (TARGET_VZEROUPPER
+ && !TREE_THIS_VOLATILE (cfun->decl)
+ && !cfun->machine->caller_return_avx256_p)
+ emit_insn (gen_avx_vzeroupper (const2_rtx));
+
if (crtl->args.pops_args)
{
rtx popc = GEN_INT (crtl->args.pops_args);
Jakub
