Patchwork Set FPU defaults for -ffast-math on MIPS.

login
register
mail settings
Submitter Kazu Hirata
Date June 9, 2010, 5:07 p.m.
Message ID <20100609170717.B3F305664EF@henry1.codesourcery.com>
Download mbox | patch
Permalink /patch/55113/
State New
Headers show

Comments

Kazu Hirata - June 9, 2010, 5:07 p.m.
Hi,

Attached is a patch to set FPU defaults for -ffast-math on MIPS.

When -ffast-math is given, the FPU default are set similarly to how
the IA32 backend does.

Tested on mips-linux-gnu.  OK to apply?

Kazu Hirata

gcc/
2010-06-09  Kazu Hirata  <kazu@codesourcery.com>

	* config.gcc (mips64*-*-linux*, mipsisa64*-*-linux*,
	mips*-*-linux*): Add mips/t-crtfm to tmake_file.
	* config/mips/crtfastmath.c: New.
	* config/mips/linux.h (ENDFILE_SPEC): New.
	* config/mips/linux64.h (ENDFILE_SPEC): New.
	* config/mips/t-crtfm: New.

libgcc/
2010-06-09  Kazu Hirata  <kazu@codesourcery.com>

	* config.host (mips64*-*-linux*, mips*-*-linux*): Add mips/t-crtfm
	to tmake_file.  Add crtfastmath.o to extra_parts.
	* config/mips/t-crtfm: New.
Andrew Pinski - June 9, 2010, 5:52 p.m.
On Wed, Jun 9, 2010 at 10:07 AM, Kazu Hirata <kazu@codesourcery.com> wrote:
> Hi,
>
> Attached is a patch to set FPU defaults for -ffast-math on MIPS.
>
> When -ffast-math is given, the FPU default are set similarly to how
> the IA32 backend does.
>
> Tested on mips-linux-gnu.  OK to apply?

How does this interact with processors which don't have a FPU?  Like the Octeon.

Thanks,
Andrew Pinski
Nathan Froyd - June 9, 2010, 5:58 p.m.
On Wed, Jun 09, 2010 at 10:52:47AM -0700, Andrew Pinski wrote:
> On Wed, Jun 9, 2010 at 10:07 AM, Kazu Hirata <kazu@codesourcery.com> wrote:
> > Attached is a patch to set FPU defaults for -ffast-math on MIPS.
> 
> How does this interact with processors which don't have a FPU?  Like the Octeon.

The new crtfastmath.c bits are protected with #ifdef __mips_hard_float.

-Nathan
Richard Sandiford - June 9, 2010, 8:06 p.m.
kazu@codesourcery.com (Kazu Hirata) writes:
> Hi,
>
> Attached is a patch to set FPU defaults for -ffast-math on MIPS.
>
> When -ffast-math is given, the FPU default are set similarly to how
> the IA32 backend does.
>
> Tested on mips-linux-gnu.  OK to apply?

Could you give it a spin 64-bit GNU/Linux too?  I'd have expected...

> 	* config/mips/linux.h (ENDFILE_SPEC): New.
> 	* config/mips/linux64.h (ENDFILE_SPEC): New.

...this to trigger a redefinition error, since linux.h is included for
64-bit as well as 32-bit targets.  It would certainly be nice to only
need the linux.h definition.

It's probably a daft question, but why do both t-crtfms need a make rule
for crtfastmath.o?

Richard
Richard Sandiford - June 10, 2010, 8:03 a.m.
> Could you give it a spin 64-bit GNU/Linux too?  I'd have expected...
>
>>       * config/mips/linux.h (ENDFILE_SPEC): New.
>>       * config/mips/linux64.h (ENDFILE_SPEC): New.
>
> ...this to trigger a redefinition error, since linux.h is included for
> 64-bit as well as 32-bit targets.  It would certainly be nice to only
> need the linux.h definition.

Duh, sorry, don't know how I missed the #undef.  But the point
about having only one definition still stands.

Richard
Kazu Hirata - June 11, 2010, 9:12 p.m.
Hi Richard,

Thank you for a review.

> Could you give it a spin 64-bit GNU/Linux too?  I'd have expected...
>
>> 	* config/mips/linux.h (ENDFILE_SPEC): New.
>> 	* config/mips/linux64.h (ENDFILE_SPEC): New.
>
> ...this to trigger a redefinition error, since linux.h is included for
> 64-bit as well as 32-bit targets.  It would certainly be nice to only
> need the linux.h definition.

I've removed the changes to linux64.h.  As you point out, one definition 
is enough.

> It's probably a daft question, but why do both t-crtfms need a make rule
> for crtfastmath.o?

Actually we don't need the duplicate make rules, so I've removed t-crtfm 
for the next iteration.

However, we still need to add crtfastmath.o to EXTRA_PARTS in t-crtfm or 
to extra_parts in config.gcc because libgcc/Makefile.in checks to see if 
EXTRA_PARTS from libgcc and EXTRA_PARTS from gcc match.

Anyway, I'll post the next iteration shortly.

Thanks,

Kazu Hirata

Patch

Index: gcc/config.gcc
===================================================================
--- gcc/config.gcc	(revision 160484)
+++ gcc/config.gcc	(working copy)
@@ -1750,6 +1750,7 @@  mips64*-*-linux* | mipsisa64*-*-linux*)
 			tm_defines="${tm_defines} MIPS_ISA_DEFAULT=65"
 			;;
 	esac
+	tmake_file="$tmake_file mips/t-crtfm"
 	gnu_ld=yes
 	gas=yes
 	test x$with_llsc != x || with_llsc=yes
@@ -1765,6 +1766,7 @@  mips*-*-linux*)				# Linux MIPS, either 
 		tm_defines="${tm_defines} MIPS_ISA_DEFAULT=32"
         esac
 	test x$with_llsc != x || with_llsc=yes
+	tmake_file="$tmake_file mips/t-crtfm"
 	;;
 mips*-*-openbsd*)
 	tm_defines="${tm_defines} OBSD_HAS_DECLARE_FUNCTION_NAME OBSD_HAS_DECLARE_OBJECT OBSD_HAS_CORRECT_SPECS"
Index: gcc/config/mips/linux.h
===================================================================
--- gcc/config/mips/linux.h	(revision 160484)
+++ gcc/config/mips/linux.h	(working copy)
@@ -147,3 +147,9 @@  extern const char *host_detect_local_cpu
 #define DRIVER_SELF_SPECS \
   BASE_DRIVER_SELF_SPECS, \
   LINUX_DRIVER_SELF_SPECS
+
+/* Similar to standard Linux, but adding -ffast-math support.  */
+#undef  ENDFILE_SPEC
+#define ENDFILE_SPEC \
+  "%{ffast-math|funsafe-math-optimizations:crtfastmath.o%s} \
+   %{shared|pie:crtendS.o%s;:crtend.o%s} crtn.o%s"
Index: gcc/config/mips/linux64.h
===================================================================
--- gcc/config/mips/linux64.h	(revision 160484)
+++ gcc/config/mips/linux64.h	(working copy)
@@ -69,3 +69,9 @@  along with GCC; see the file COPYING3.  
    ieee_quad_format is the default, but let's put this here to make
    sure nobody thinks we just forgot to set it to something else.  */
 #define MIPS_TFMODE_FORMAT mips_quad_format
+
+/* Similar to standard Linux, but adding -ffast-math support.  */
+#undef  ENDFILE_SPEC
+#define ENDFILE_SPEC \
+  "%{ffast-math|funsafe-math-optimizations:crtfastmath.o%s} \
+   %{shared|pie:crtendS.o%s;:crtend.o%s} crtn.o%s"
Index: gcc/config/mips/crtfastmath.c
===================================================================
--- gcc/config/mips/crtfastmath.c	(revision 0)
+++ gcc/config/mips/crtfastmath.c	(revision 0)
@@ -0,0 +1,53 @@ 
+/* Copyright (C) 2010 Free Software Foundation, Inc. 
+
+   This file is part of GCC.
+
+   GCC is free software; you can redistribute it and/or modify it
+   under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3, or (at your option)
+   any later version.
+
+   GCC is distributed in the hope that it will be useful, but WITHOUT
+   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
+   License for more details.
+
+   Under Section 7 of GPL version 3, you are granted additional
+   permissions described in the GCC Runtime Library Exception, version
+   3.1, as published by the Free Software Foundation.
+
+   You should have received a copy of the GNU General Public License
+   and a copy of the GCC Runtime Library Exception along with this
+   program; see the files COPYING3 and COPYING.RUNTIME respectively.
+   If not, see <http://www.gnu.org/licenses/>.  */
+
+#ifdef __mips_hard_float
+
+/* Flush denormalized numbers to zero.  */
+#define _FPU_FLUSH_TZ   0x1000000
+
+/* Rounding control.  */
+#define _FPU_RC_NEAREST 0x0     /* RECOMMENDED */
+#define _FPU_RC_ZERO    0x1
+#define _FPU_RC_UP      0x2
+#define _FPU_RC_DOWN    0x3
+
+/* Enable interrupts for IEEE exceptions.  */
+#define _FPU_IEEE     0x00000F80
+
+/* Macros for accessing the hardware control word.  */
+#define _FPU_GETCW(cw) __asm__ ("cfc1 %0,$31" : "=r" (cw))
+#define _FPU_SETCW(cw) __asm__ ("ctc1 %0,$31" : : "r" (cw))
+
+static void __attribute__((constructor))
+set_fast_math (void)
+{
+  unsigned int fcr;
+
+  /* Flush to zero, round to nearest, IEEE exceptions disabled.  */
+  fcr = _FPU_FLUSH_TZ | _FPU_RC_NEAREST;
+
+  _FPU_SETCW(fcr);
+}
+
+#endif /* __mips_hard_float */
Index: gcc/config/mips/t-crtfm
===================================================================
--- gcc/config/mips/t-crtfm	(revision 0)
+++ gcc/config/mips/t-crtfm	(revision 0)
@@ -0,0 +1,9 @@ 
+
+EXTRA_MULTILIB_PARTS += crtfastmath.o
+
+EXTRA_PARTS += crtfastmath.o
+
+$(T)crtfastmath.o: $(srcdir)/config/mips/crtfastmath.c $(GCC_PASSES)
+	$(GCC_FOR_TARGET) $(GCC_CFLAGS) $(MULTILIB_CFLAGS) $(INCLUDES) \
+	-c -o $(T)crtfastmath.o $(srcdir)/config/mips/crtfastmath.c
+
Index: libgcc/config.host
===================================================================
--- libgcc/config.host	(revision 160484)
+++ libgcc/config.host	(working copy)
@@ -400,8 +400,12 @@  mips-sgi-irix[56]*)
 mips*-*-netbsd*)			# NetBSD/mips, either endian.
 	;;
 mips64*-*-linux*)
+	extra_parts="$extra_parts crtfastmath.o"
+	tmake_file="{$tmake_file} mips/t-crtfm"
 	;;
 mips*-*-linux*)				# Linux MIPS, either endian.
+	extra_parts="$extra_parts crtfastmath.o"
+	tmake_file="{$tmake_file} mips/t-crtfm"
 	;;
 mips*-*-openbsd*)
 	;;
Index: libgcc/config/mips/t-crtfm
===================================================================
--- libgcc/config/mips/t-crtfm	(revision 0)
+++ libgcc/config/mips/t-crtfm	(revision 0)
@@ -0,0 +1,3 @@ 
+crtfastmath.o: $(gcc_srcdir)/config/mips/crtfastmath.c
+	$(gcc_compile) -c $(gcc_srcdir)/config/mips/crtfastmath.c
+