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

Allow functions for AVX target inline functions for lower target.

Browse source 
Allow functions for AVX target inline functions for lower target.  This bug was
reported here: http://gcc.gnu.org/ml/gcc-patches/2013-06/msg01620.html

2013-07-02  Sriraman Tallam  <tmsriram@google.com>

	* config/i386/i386.c (gate_insert_vzeroupper): Check if target
	ISA is AVX.
	(ix86_option_override_internal):Turn on all -mavx target flags by
	default as they are dependent on AVX anyway.

	* gcc.target/i386/avx-inline.c: New test.

From-SVN: r200627
This commit is contained in:
Sriraman Tallam 2013-07-03 00:03:06 +00:00 • committed by Sriraman Tallam
parent 4b04355371
commit 786fb9b6c4
4 changed files with 52 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
gcc/ChangeLog
View file

@ -1,3 +1,10 @@
2013-07-02 Sriraman Tallam <tmsriram@google.com>
* config/i386/i386.c (gate_insert_vzeroupper): Check if target
ISA is AVX.
(ix86_option_override_internal):Turn on all -mavx target flags by
default as they are dependent on AVX anyway.
2013-07-02 Cary Coutant <ccoutant@google.com> 2013-07-02 Cary Coutant <ccoutant@google.com>
* dwarf2out.c (loc_checksum): Call hash_loc_operands for a * dwarf2out.c (loc_checksum): Call hash_loc_operands for a

50
gcc/config/i386/i386.c
View file

@ -2578,7 +2578,7 @@ static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =
static bool static bool
gate_insert_vzeroupper (void) gate_insert_vzeroupper (void)
{ {
return TARGET_VZEROUPPER; return TARGET_AVX && TARGET_VZEROUPPER;
} }
static unsigned int static unsigned int
@ -3936,34 +3936,26 @@ ix86_option_override_internal (bool main_args_p)
#endif #endif
} }
if (TARGET_AVX) /* When not optimize for size, enable vzeroupper optimization for
{ TARGET_AVX with -fexpensive-optimizations and split 32-byte
/* When not optimize for size, enable vzeroupper optimization for AVX unaligned load/store. */
TARGET_AVX with -fexpensive-optimizations and split 32-byte if (!optimize_size)
AVX unaligned load/store. */ {
if (!optimize_size) if (flag_expensive_optimizations
{ && !(target_flags_explicit & MASK_VZEROUPPER))
if (flag_expensive_optimizations target_flags |= MASK_VZEROUPPER;
&& !(target_flags_explicit & MASK_VZEROUPPER)) if ((x86_avx256_split_unaligned_load & ix86_tune_mask)
target_flags |= MASK_VZEROUPPER; && !(target_flags_explicit & MASK_AVX256_SPLIT_UNALIGNED_LOAD))
if ((x86_avx256_split_unaligned_load & ix86_tune_mask) target_flags |= MASK_AVX256_SPLIT_UNALIGNED_LOAD;
&& !(target_flags_explicit & MASK_AVX256_SPLIT_UNALIGNED_LOAD)) if ((x86_avx256_split_unaligned_store & ix86_tune_mask)
target_flags |= MASK_AVX256_SPLIT_UNALIGNED_LOAD; && !(target_flags_explicit & MASK_AVX256_SPLIT_UNALIGNED_STORE))
if ((x86_avx256_split_unaligned_store & ix86_tune_mask) target_flags |= MASK_AVX256_SPLIT_UNALIGNED_STORE;
&& !(target_flags_explicit & MASK_AVX256_SPLIT_UNALIGNED_STORE)) /* Enable 128-bit AVX instruction generation
target_flags |= MASK_AVX256_SPLIT_UNALIGNED_STORE; for the auto-vectorizer. */
/* Enable 128-bit AVX instruction generation if (TARGET_AVX128_OPTIMAL
for the auto-vectorizer. */ && !(target_flags_explicit & MASK_PREFER_AVX128))
if (TARGET_AVX128_OPTIMAL target_flags |= MASK_PREFER_AVX128;
&& !(target_flags_explicit & MASK_PREFER_AVX128)) }
target_flags |= MASK_PREFER_AVX128;
}
}
else
{
/* Disable vzeroupper pass if TARGET_AVX is disabled. */
target_flags &= ~MASK_VZEROUPPER;
}
if (ix86_recip_name) if (ix86_recip_name)
{ {

4
gcc/testsuite/ChangeLog
View file

@ -1,3 +1,7 @@
2013-07-02 Sriraman Tallam <tmsriram@google.com>
* gcc.target/i386/avx-inline.c: New test.
2013-07-02 Maciej W. Rozycki <macro@codesourcery.com> 2013-07-02 Maciej W. Rozycki <macro@codesourcery.com>
* gcc.target/mips/call-1.c: Accept JALRS and JALR. * gcc.target/mips/call-1.c: Accept JALRS and JALR.

20
gcc/testsuite/gcc.target/i386/avx-inline.c Normal file
View file

@ -0,0 +1,20 @@
/* Check if avx target functions can inline lower target functions. */
/* { dg-do compile } */
/* { dg-options "-O0 -mno-avx -mno-sse3" } */
__attribute__((always_inline,target("sse3")))
inline int callee ()
{
return 0;
}
__attribute__((target("avx")))
inline int caller ()
{
return callee ();
}
int main ()
{
return caller ();
}