Re: [PATCH] Fix PR81175, make gather builtins pure

[Jakub Jelinek <jakub@redhat.com>]

On Tue, Jun 27, 2017 at 12:08:59PM +0200, Uros Bizjak wrote:
> > Ok for trunk if it passes bootstrap/regtest?
> >
> > 2017-06-27  Jakub Jelinek  <jakub@redhat.com>
> >
> >         PR target/81175
> >         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Use def_builtin
> >         rather than def_builtin_pure for __builtin_ia32_gatherpf*.
> 
> Looks obvious to me, especially after the above explanation.

Thanks.  Apparently the same commit regressed another testcase, where we
want to check the expansion of the gather builtin without the right ISA
enabled, but that doesn't trigger anymore because it is optimized away.

Fixed thusly, ok for trunk?  Perhaps we should add another testcase to check
similarly gatherpf builtin without the lhs, but we'd need different options.

2017-06-27  Jakub Jelinek  <jakub@redhat.com>

	PR target/81175
	* gcc.target/i386/pr69255-2.c (foo): Use the return value of the
	gather.

--- gcc/testsuite/gcc.target/i386/pr69255-2.c.jj	2017-05-05 09:19:48.000000000 +0200
+++ gcc/testsuite/gcc.target/i386/pr69255-2.c	2017-06-27 12:20:31.697944761 +0200
@@ -12,7 +12,8 @@ __attribute__ ((__vector_size__ (16))) i
 void
 foo (const long long *p)
 {
-  __builtin_ia32_gather3siv4di (a, p, b, 1, 1);		/* { dg-error "needs isa option -m32 -mavx512vl" } */
+  volatile __attribute__ ((__vector_size__ (32))) long long c;
+  c = __builtin_ia32_gather3siv4di (a, p, b, 1, 1);		/* { dg-error "needs isa option -m32 -mavx512vl" } */
   /* { dg-warning "AVX vector return without AVX enabled changes the ABI" "" { target *-*-* } .-1 } */
   /* { dg-warning "AVX vector argument without AVX enabled changes the ABI" "" { target *-*-* } .-2 } */
 }


	Jakub