config.host (arc*-*-elf*, [...]): New configurations.

2013-10-01  Joern Rennecke  <joern.rennecke@embecosm.com>
            Brendan Kehoe  <brendan@zen.org>
            Simon Cook  <simon.cook@embecosm.com>

        * config.host (arc*-*-elf*, arc*-*-linux-uclibc*): New configurations.
        * config/arc: New directory.
        * longlong.h [__arc__] (umul_ppmm): Remove.
        [__arc__] (__umulsidi3): Define.
        [__arc__ && __ARC_NORM__] (count_leading_zeroes): Define.
        [__arc__ && __ARC_NORM__] (COUNT_LEADING_ZEROS_0): Likewise.

Co-Authored-By: Brendan Kehoe <brendan@zen.org>
Co-Authored-By: Simon Cook <simon.cook@embecosm.com>

From-SVN: r203073

libgcc/ChangeLog

+2013-10-01  Joern Rennecke  <joern.rennecke@embecosm.com>
+	    Brendan Kehoe  <brendan@zen.org>
+	    Simon Cook  <simon.cook@embecosm.com>
+
+	* config.host (arc*-*-elf*, arc*-*-linux-uclibc*): New configurations.
+	* config/arc: New directory.
+	* longlong.h [__arc__] (umul_ppmm): Remove.
+	[__arc__] (__umulsidi3): Define.
+	[__arc__ && __ARC_NORM__] (count_leading_zeroes): Define.
+	[__arc__ && __ARC_NORM__] (COUNT_LEADING_ZEROS_0): Likewise.
+
 2013-09-17  Jacek Caban  <jacek@codeweavers.com>
 
 	* config/i386/gthr-win32.c: CreateSemaphoreW instead of

libgcc/config.host

@@ -91,6 +91,9 @@ alpha*-*-*)
 am33_2.0-*-linux*)
 	cpu_type=mn10300
 	;;
+arc*-*-*)
+	cpu_type=arc
+	;;
 arm*-*-*)
 	cpu_type=arm
 	;;
@@ -315,6 +318,14 @@ alpha*-dec-*vms*)
 	extra_parts="$extra_parts vms-dwarf2.o vms-dwarf2eh.o"
 	md_unwind_header=alpha/vms-unwind.h
 	;;
+arc*-*-elf*)
+	tmake_file="arc/t-arc-newlib arc/t-arc"
+	extra_parts="crti.o crtn.o crtend.o crtbegin.o crtendS.o crtbeginS.o libgmon.a crtg.o crtgend.o"
+	;;
+arc*-*-linux-uclibc*)
+	tmake_file="${tmake_file} t-slibgcc-libgcc t-slibgcc-nolc-override arc/t-arc700-uClibc arc/t-arc"
+	extra_parts="crti.o crtn.o crtend.o crtbegin.o crtendS.o crtbeginS.o libgmon.a crtg.o crtgend.o"
+	;;
 arm-wrs-vxworks)
 	tmake_file="$tmake_file arm/t-arm arm/t-vxworks t-softfp-sfdf t-softfp-excl arm/t-softfp t-softfp"
 	extra_parts="$extra_parts crti.o crtn.o"

libgcc/config/arc/asm.h

+/* Assembler macros for the Synopsys DesignWare ARC CPU.
+
+   Copyright (C) 1994, 1995, 1997, 2004, 2007-2012
+   Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#define FUNC(X)         .type X,@function
+#define ENDFUNC(X)      .size X, .-X

libgcc/config/arc/crtg.S

+/* Code to start and stop profiling for the Synopsys DesignWare ARC CPU.
+
+   Copyright (C) 1994, 1995, 1997, 2004, 2007-2012
+   Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+	.section .init
+	.global _init
+	.global _fini
+	.global __monstartup
+	mov_s	r0,_init
+	mov_s	r1,_fini
+	jl	__monstartup
+
+	.section .__arc_profile_desc, "a"
+	.global __arc_profile_desc_secstart
+	.balign	4
+__arc_profile_desc_secstart:
+	.section .__arc_profile_forward, "a"
+	.global __arc_profile_forward_secstart
+	.balign 4
+__arc_profile_forward_secstart:
+	.section .__arc_profile_counters, "aw"
+	.global __arc_profile_counters_secstart
+	.balign	4
+__arc_profile_counters_secstart:
+
+	.section .fini
+	.global _mcleanup
+	jl	_mcleanup

libgcc/config/arc/crtgend.S

+/* Code to start and stop profiling for the Synopsys DesignWare ARC CPU.
+
+   Copyright (C) 1994, 1995, 1997, 2004, 2007-2012
+   Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+	.section .__arc_profile_desc, "a"
+	.global __arc_profile_desc_secend
+__arc_profile_desc_secend:
+	.section .__arc_profile_forward, "a"
+	.global __arc_profile_forward_secend
+__arc_profile_forward_secend:

libgcc/config/arc/crti.S

+/* .fini/.init stack frame setup for the Synopsys DesignWare ARC CPU.
+
+   Copyright (C) 1994, 1995, 1997, 2004, 2007-2012
+   Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#  This file contains the stack frame setup for contents of the .fini and
+# .init sections.
+
+	.section .init
+	.global _init
+	.word 0
+_init:
+	push_s	blink
+
+	.section .fini
+	.global _fini
+	.word 0
+_fini:
+	push_s	blink

libgcc/config/arc/crtn.S

+/* Ensure .fini/.init return for the Synopsys DesignWare ARC CPU.
+
+   Copyright (C) 1994, 1995, 1997, 2004, 2007-2012
+   Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+                on behalf of Synopsys 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/>.  */
+
+# This file just makes sure that the .fini and .init sections do in
+# fact return. This file is the last thing linked into any executable.
+
+	.section .init
+	pop_s	blink
+	j_s	[blink]
+
+
+	.section .fini
+	pop_s	blink
+	j_s	[blink]

libgcc/config/arc/divtab-arc700.c

+/* Copyright (C) 2004, 2006, 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys Inc.
+
+This file 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.
+
+In addition to the permissions in the GNU General Public License, the
+Free Software Foundation gives you unlimited permission to link the
+compiled version of this file into combinations with other programs,
+and to distribute those combinations without any restriction coming
+from the use of this file.  (The General Public License restrictions
+do apply in other respects; for example, they cover modification of
+the file, and distribution when not linked into a combine
+executable.)
+
+This file 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.
+
+You should have received a copy of the GNU General Public License
+along with this program; see the file COPYING3.  If not see
+<http://www.gnu.org/licenses/>.  */
+
+/* Calculate division table for ARC700 integer division
+   Contributed by Joern Rennecke
+   joern.rennecke@arc.com  */
+
+#include <stdio.h>
+#include <math.h>
+
+int
+main ()
+{
+  int i, j;
+  unsigned x;
+  double q, r, err, max_err = -1;
+
+  puts("/* This table has been generated by divtab-arc700.c.  */");
+  puts("\
+/* 1/512 .. 1/256, normalized.  There is a leading 1 in bit 31.\n\
+   For powers of two, we list unnormalized numbers instead.  The values\n\
+   for powers of 2 are loaded, but not used.  The value for 1 is actually\n\
+   the first instruction after .Lmuldiv.  */\n\
+	.balign 4");
+  puts (".Ldivtab:\n");
+  for (i = 256; i >= 2; --i)
+    {
+      j = i < 0 ? -i : i;
+      if (j & (j-1))
+	while (j < 128)
+	  j += j;
+      else
+	/* Power of two. */
+	j *= 128;
+      q = 4.*(1<<30)*128/j;
+      r = ceil (q);
+      printf ("\t.long\t0x%X\n", (unsigned) r);
+      err = r - q;
+      if (err > max_err)
+	max_err = err;
+    }
+#if 0
+  printf ("\t/* maximum error: %f */\n", max_err);
+#endif
+  exit (0);
+}

libgcc/config/arc/dp-hack.h

+/* Copyright (C) 2007-2013 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/* This file selects the double-precision parts of fp-bit.c that are
+   still needed for some ARC hardware variants; it also renames functions
+   that duplicate asm-coded functionality so that their results can be
+   used to compare with the optimized versions for debugging.  */
+
+#define FINE_GRAINED_LIBRARIES
+#define ARC_DP_DEBUG 1
+#if !defined (__ARC_NORM__) || ARC_DP_DEBUG
+#define L_pack_df
+#define L_unpack_df
+#define L_make_df
+#define L_thenan_df
+#define L_sf_to_df
+#endif
+#ifndef __ARC_NORM__
+#define L_addsub_df
+#elif ARC_DP_DEBUG
+#define L_addsub_df
+#define __adddf3 __adddf3_c
+#define __subdf3 __subdf3_c
+#endif
+#ifndef __ARC_NORM__
+#define L_mul_df
+#define L_div_df
+#elif (!defined (__ARC700__) && !defined (__ARC_MUL64__) \
+       && !defined(__ARC_MUL32BY16__))
+#define L_mul_df
+#define L_div_df
+#undef QUIET_NAN
+#define QUIET_NAN 0xfffffffffffffLL
+#elif ARC_DP_DEBUG
+#define L_mul_df
+#define __muldf3 __muldf3_c
+#define L_div_df
+#define __divdf3 __divdf3_c
+#endif
+#ifndef __ARC_NORM__
+#define L_df_to_sf
+#define L_si_to_df
+#define L_df_to_si
+#define L_tf_to_usi /* need to defined this instead of df_to_usi */
+#define L_usi_to_df
+#elif ARC_DP_DEBUG
+#define L_df_to_sf
+#define __truncdfsf2 __truncdfsf2_c
+#define L_si_to_df
+#define __floatsidf __floatsidf_c
+#define L_df_to_si
+#define __fixdfsi __fixdfsi_c
+#define L_tf_to_usi
+#define __fixunsdfsi __fixunsdfsi_c
+#define L_usi_to_df
+#define __floatunsidf __floatunsidf_c
+#endif
+#ifndef __ARC_NORM__
+#define L_fpcmp_parts_df
+#define L_compare_df
+#define L_eq_df
+#define L_ne_df
+#define L_gt_df
+#define L_ge_df
+#define L_lt_df
+#define L_le_df
+#define L_unord_df
+#define L_negate_df
+#elif ARC_DP_DEBUG
+#define L_fpcmp_parts_df
+#define L_eq_df
+#define __eqdf2 __eqdf2_c
+#define L_gt_df
+#define __gtdf2 __gtdf2_c
+#define L_ge_df
+#define __gedf2 __gedf2_c
+#define L_unord_df
+#define __unorddf2 __unorddf2_c
+#endif

libgcc/config/arc/fp-hack.h

+/* Copyright (C) 2007-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/* This file selects the single-precision parts of fp-bit.c that are
+   still needed for some ARC hardware variants; it also renames functions
+   that duplicate asm-coded functionality so that their results can be
+   used to compare with the optimized versions for debugging.  */
+
+#define ARC_FP_DEBUG 1
+#define FINE_GRAINED_LIBRARIES
+#if !defined (__ARC_NORM__) || ARC_FP_DEBUG
+#define L_pack_sf
+#define L_unpack_sf
+#define L_make_sf
+#define L_thenan_sf
+#endif
+#ifndef __ARC_NORM__
+#define L_addsub_sf
+#define L_mul_sf
+#define L_div_sf
+#define L_sf_to_df
+#define L_si_to_sf
+#define L_sf_to_si
+#define L_usi_to_sf
+#elif ARC_FP_DEBUG
+#define L_addsub_sf
+#define __addsf3 __addsf3_c
+#define __subsf3 __subsf3_c
+#define L_mul_sf
+#define __mulsf3 __mulsf3_c
+#define L_div_sf
+#define __divsf3 __divsf3_c
+#define L_sf_to_df
+#define __extendsfdf2 __extendsfdf2_c
+#define L_si_to_sf
+#define __floatsisf __floatsisf_c
+#define L_sf_to_si
+#define __fixsfsi __fixsfsi_c
+#define L_usi_to_sf
+#define __floatunsisf __floatunsisf_c
+#endif
+#ifndef __ARC_NORM__
+#define L_fpcmp_parts_sf
+#define L_compare_sf
+#define L_eq_sf
+#define L_ne_sf
+#define L_gt_sf
+#define L_ge_sf
+#define L_lt_sf
+#define L_le_sf
+#define L_unord_sf
+#define L_negate_sf
+#elif ARC_FP_DEBUG
+#define L_fpcmp_parts_sf
+#define L_eq_sf
+#define __eqsf2 __eqsf2_c
+#define L_gt_sf
+#define __gtsf2 __gtsf2_c
+#define L_ge_sf
+#define __gesf2 __gesf2_c
+#define L_unord_sf
+#define __unordsf2 __unordsf2_c
+#endif

libgcc/config/arc/gmon/atomic.h

+/* Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/* File deliberately left blank. */

libgcc/config/arc/gmon/auxreg.h

+/* Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#define	LP_START	0x02
+#define	LP_END		0x03
+#define IDENTITY	0x04
+#define STATUS32	0x0a
+#define	COUNT0		0x21 /* Timer 0 count */
+#define	CONTROL0	0x22 /* Timer 0 control */
+#define	LIMIT0		0x23 /* Timer 0 limit */
+#define INT_VECTOR_BASE	0x25
+#define D_CACHE_BUILD	0x72
+#define DC_FLDL		0x4c

libgcc/config/arc/gmon/dcache_linesz.S

+/* This file contains code to do profiling.
+
+   Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../asm.h"
+#include "auxreg.h"
+/*  This file contains code to do profiling.  */
+	.weak	__profile_timer_cycles
+	.global __profile_timer_cycles
+	.set	__profile_timer_cycles, 200
+	.text
+	; For Arctangent-A5, if no data cache is present, a read of the
+	; cache build register returns the ID register.  For ARC600 and
+	; later, the version field will be zero.
+	.global	__dcache_linesz
+	.balign	4
+__dcache_linesz:
+	lr	r12,[D_CACHE_BUILD]
+	extb_s	r0,r12
+	breq_s	r0,0,.Lsz_nocache
+	brge	r0,0x20,.Lsz_havecache
+	lr	r0,[IDENTITY]
+	breq	r12,r0,.Lsz_nocache
+.Lsz_havecache:
+	lsr_s	r12,r12,16
+	mov_s	r0,16
+	bmsk_s	r12,r12,3
+	asl_s	r0,r0,r12
+	j_s	[blink]
+.Lsz_nocache:
+	mov_s	r0,1
+	j_s	[blink]

libgcc/config/arc/gmon/machine-gmon.h

+/* Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#ifndef MACHINE_GMON_H
+#define MACHINE_GMON_H
+
+/* We can't fake out own <sys/types.h> header because the newlib / uclibc
+   headers in GCC_FOR_TARGET take precedence.  */
+
+#define __BEGIN_DECLS
+#define __END_DECLS
+
+#define __THROW
+
+extern int __dcache_linesz (void);
+
+#define _MCOUNT_DECL(countp, selfpc) \
+  static inline void _mcount_internal (void *countp, u_long selfpc)
+
+extern void _mcount (void);
+extern void _mcount_call (void);
+
+/* N.B.: the calling point might be a sibcall, thus blink does not necessarily
+  hold the caller's address.  r8 doesn't hold the caller's address, either,
+  but rather a pointer to the counter data structure associated with the
+  caller.
+  This function must be compiled with optimization turned on in order to
+  enable a sibcall for the final call to selfpc; this is important when trying
+  to profile a program with deep tail-recursion that would get a stack
+  overflow otherwise.  */
+#define MCOUNT \
+void \
+_mcount_call (void) \
+{ \
+  register void *countp __asm("r8"); \
+  register u_long selfpc __asm("r9"); \
+  _mcount_internal (countp, selfpc); \
+  ((void (*)(void)) selfpc) (); \
+}
+
+extern int __profil (u_short *,size_t, size_t, u_int);
+
+#endif /* MACHINE_GMON_H */

libgcc/config/arc/gmon/mcount.c

+/*-
+ * Copyright (c) 1983, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * Copyright (C) 2007-2012 Free Software Foundation, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#if !defined(lint) && !defined(KERNEL) && defined(LIBC_SCCS)
+static char sccsid[] = "@(#)mcount.c	8.1 (Berkeley) 6/4/93";
+#endif
+
+#if 0
+#include <unistd.h>
+#include <sys/param.h>
+#endif
+#include <sys/gmon.h>
+
+/* This file provides the machine-dependent definitions of the _MCOUNT_DECL
+   and MCOUNT macros.  */
+#include <machine-gmon.h>
+
+#include <atomic.h>
+
+/*
+ * mcount is called on entry to each function compiled with the profiling
+ * switch set.  _mcount(), which is declared in a machine-dependent way
+ * with _MCOUNT_DECL, does the actual work and is either inlined into a
+ * C routine or called by an assembly stub.  In any case, this magic is
+ * taken care of by the MCOUNT definition in <machine/profile.h>.
+ *
+ * _mcount updates data structures that represent traversals of the
+ * program's call graph edges.  frompc and selfpc are the return
+ * address and function address that represents the given call graph edge.
+ *
+ * Note: the original BSD code used the same variable (frompcindex) for
+ * both frompcindex and frompc.  Any reasonable, modern compiler will
+ * perform this optimization.
+ */
+_MCOUNT_DECL(count_ptr, selfpc)	/* _mcount; may be static, inline, etc */
+{
+	register ARCINDEX *frompcindex;
+	register struct tostruct *top, *prevtop;
+	register struct gmonparam *p;
+	register ARCINDEX toindex;
+
+	/* Check for nested function trampoline.  */
+	if (selfpc & 2)
+	  selfpc = *(u_long *) (selfpc + 10);
+
+	p = &_gmonparam;
+	/*
+	 * check that we are profiling
+	 * and that we aren't recursively invoked.
+	 */
+#if 0
+	if (catomic_compare_and_exchange_bool_acq (&p->state, GMON_PROF_BUSY,
+						   GMON_PROF_ON))
+	  return;
+#elif defined (__ARC700__)
+/* ??? This could temporrarily loose the ERROR / OFF condition in a race,
+   but doing an actual compare_and_exchange would be too costly.  It would
+   be better if we had a semaphore independent of the 'sticky' state, but
+   then we could run into ABI compatibility problems with the size of struct
+   gmonparam.  */
+	{
+	  u_long old_state;
+
+	  __asm ("ex %0,%1": "=r" (old_state), "+m" (p->state)
+		 : "0" (GMON_PROF_BUSY));
+	  if (old_state != GMON_PROF_ON)
+	    {
+	      switch (old_state)
+		{
+		case GMON_PROF_OFF:
+		  __asm ("ex %0,%1": "+r" (old_state), "+m" (p->state));
+		  if (old_state == GMON_PROF_BUSY
+		      /* Switching off while we say we are busy while profiling
+			 was actually already switched off is all right.  */
+		      || old_state == GMON_PROF_OFF)
+		    break;
+		  /* It is not clear if we should allow switching on
+		     profiling at this point, and how to handle further races.
+		     For now, record an error in this case.  */
+		  /* Fall through.  */
+		default: /* We expect here only GMON_PROF_ERROR.  */
+		  p->state = GMON_PROF_ERROR;
+		  break;
+		case GMON_PROF_BUSY: break;
+		}
+	      return;
+	    }
+	}
+#else /* ??? No semaphore primitives available.  */
+	if (p->state != GMON_PROF_ON)
+	  return;
+	p->state = GMON_PROF_BUSY;
+#endif
+
+	frompcindex = count_ptr;
+	toindex = *frompcindex;
+	if (toindex == 0) {
+		/*
+		 *	first time traversing this arc
+		 */
+		toindex = ++p->tos[0].link;
+		if (toindex >= (ARCINDEX) p->tolimit)
+			/* halt further profiling */
+			goto overflow;
+
+		*frompcindex = toindex;
+		top = &p->tos[toindex];
+		top->selfpc = selfpc;
+		top->count = 1;
+		top->link = 0;
+		goto done;
+	}
+	top = &p->tos[toindex];
+	if (top->selfpc == selfpc) {
+		/*
+		 * arc at front of chain; usual case.
+		 */
+		top->count++;
+		goto done;
+	}
+	/*
+	 * have to go looking down chain for it.
+	 * top points to what we are looking at,
+	 * prevtop points to previous top.
+	 * we know it is not at the head of the chain.
+	 */
+	for (; /* goto done */; ) {
+		if (top->link == 0) {
+			/*
+			 * top is end of the chain and none of the chain
+			 * had top->selfpc == selfpc.
+			 * so we allocate a new tostruct
+			 * and link it to the head of the chain.
+			 */
+			toindex = ++p->tos[0].link;
+			if (toindex >= (ARCINDEX) p->tolimit)
+				goto overflow;
+
+			top = &p->tos[toindex];
+			top->selfpc = selfpc;
+			top->count = 1;
+			top->link = *frompcindex;
+			*frompcindex = toindex;
+			goto done;
+		}
+		/*
+		 * otherwise, check the next arc on the chain.
+		 */
+		prevtop = top;
+		top = &p->tos[top->link];
+		if (top->selfpc == selfpc) {
+			/*
+			 * there it is.
+			 * increment its count
+			 * move it to the head of the chain.
+			 */
+			top->count++;
+			toindex = prevtop->link;
+			prevtop->link = top->link;
+			top->link = *frompcindex;
+			*frompcindex = toindex;
+			goto done;
+		}
+
+	}
+done:
+	p->state = GMON_PROF_ON;
+	return;
+overflow:
+	p->state = GMON_PROF_ERROR;
+	return;
+}
+
+/*
+ * Actual definition of mcount function.  Defined in <machine/profile.h>,
+ * which is included by <sys/gmon.h>.
+ */
+MCOUNT

libgcc/config/arc/gmon/prof-freq-stub.S

+/* This file contains code to do profiling.
+
+   Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../asm.h"
+/*  This file contains code to do profiling.  */
+	.weak	__profile_frequency_value
+	.global __profile_frequency_value
+	.set	__profile_frequency_value, 1000
+	.text
+	.balign	4
+	.global __profile_frequency
+	FUNC(__profile_frequency)
+__profile_frequency:
+	mov_s	r0,__profile_frequency_value
+	j_s	[blink]
+	ENDFUNC(__profile_frequency)

libgcc/config/arc/gmon/prof-freq.c

+/* Return frequency of ticks reported by profil.  Generic version. */
+/*-
+ * Copyright (c) 1983, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * Copyright (C) 2007-2012 Free Software Foundation, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+
+#include <sys/types.h>
+#include <sys/time.h>
+#if 0
+#include <libc-internal.h>
+#else
+#include "sys/gmon.h"
+#endif
+
+int
+__profile_frequency (void)
+{
+  /*
+   * Discover the tick frequency of the machine if something goes wrong,
+   * we return 0, an impossible hertz.
+   */
+  struct itimerval tim;
+
+  tim.it_interval.tv_sec = 0;
+  tim.it_interval.tv_usec = 1;
+  tim.it_value.tv_sec = 0;
+  tim.it_value.tv_usec = 0;
+  setitimer(ITIMER_REAL, &tim, 0);
+  setitimer(ITIMER_REAL, 0, &tim);
+  if (tim.it_interval.tv_usec < 2)
+    return 0;
+  return (1000000 / tim.it_interval.tv_usec);
+}

libgcc/config/arc/gmon/profil.S

+/* This file contains code to do profiling.
+
+   Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../asm.h"
+#include "auxreg.h"
+/*  This file contains code to do profiling.  */
+	.weak	__profile_timer_cycles
+	.global __profile_timer_cycles
+	.set	__profile_timer_cycles, 200
+
+        .section .bss
+        .global __profil_offset
+        .align 4
+        .type	__profil_offset, @object
+        .size	__profil_offset, 4
+__profil_offset:
+        .zero   4
+
+	.text
+	.global	__dcache_linesz
+	.global __profil
+	FUNC(__profil)
+.Lstop_profiling:
+	sr	r0,[CONTROL0]
+	j_s	[blink]
+	.balign	4
+__profil:
+.Lprofil:
+	breq_s	r0,0,.Lstop_profiling
+	; r0: buf r1: bufsiz r2: offset r3: scale
+	bxor.f	r3,r3,15; scale must be 0x8000, i.e. 1/2; generate 0.
+	push_s	blink
+	lsr_s	r2,r2,1
+	mov_s	r8,r0
+	flag.ne	1	; halt if wrong scale
+	sub_s	r0,r0,r2
+	st	r0,[__profil_offset]
+	bl	__dcache_linesz
+	pop_s	blink
+	bbit1.d	r0,0,nocache
+	mov_s	r0,r8
+#ifdef __ARC700__
+	add_s	r1,r1,31
+	lsr.f	lp_count,r1,5
+	lpne	2f
+	sr	r0,[DC_FLDL]
+	add_s	r0,r0,32
+#else /* !__ARC700__ */
+# FIX ME: set up loop according to cache line size
+	lr	r12,[D_CACHE_BUILD]
+	sub_s	r0,r0,16
+	sub_s	r1,r1,1
+	lsr_s	r12,r12,16
+	asr_s	r1,r1,4
+	bmsk_s	r12,r12,3
+	asr_s	r1,r1,r12
+	add.f	lp_count,r1,1
+	mov_s	r1,16
+	asl_s	r1,r1,r12
+	lpne	2f
+	add	r0,r0,r1
+	sr	r0,[DC_FLDL]
+#endif /* __ARC700__ */
+2:	b_s	.Lcounters_cleared
+nocache:
+.Lcounters_cleared:
+	lr	r1,[INT_VECTOR_BASE] ; disable timer0 interrupts
+	sr	r3,[CONTROL0]
+	sr	r3,[COUNT0]
+0:	ld_s	r0,[pcl,1f-0b+((0b-.Lprofil) & 2)] ; 1f@GOTOFF
+0:	ld_s	r12,[pcl,1f+4-0b+((0b-.Lprofil) & 2)] ; 1f@GOTOFF + 4
+	st_s	r0,[r1,24]; timer0 uses vector3
+	st_s	r12,[r1,24+4]; timer0 uses vector3
+	;sr	10000,[LIMIT0]
+	sr	__profile_timer_cycles,[LIMIT0]
+	mov_s	r12,3	; enable timer interrupts; count only when not halted.
+	sr	r12,[CONTROL0]
+	lr	r12,[STATUS32]
+	bset_s	r12,r12,1 ; allow level 1 interrupts
+	flag	r12
+	mov_s	r0,0
+	j_s	[blink]
+	.balign	4
+1:	j	__profil_irq
+	ENDFUNC(__profil)
+
+	FUNC(__profil_irq)
+	.balign 4	; make final jump unaligned to avoid delay penalty
+	.balign 32,0,12	; make sure the code spans no more that two cache lines
+	nop_s
+__profil_irq:
+	push_s	r0
+	ld	r0,[__profil_offset]
+	push_s	r1
+	lsr	r1,ilink1,2
+	push_s	r2
+	ldw.as.di r2,[r0,r1]
+	add1	r0,r0,r1
+	ld_s	r1,[sp,4]
+	add_s	r2,r2,1
+	bbit1	r2,16,nostore
+	stw.di	r2,[r0]
+nostore:ld.ab	r2,[sp,8]
+	pop_s	r0
+	j.f	[ilink1]
+	ENDFUNC(__profil_irq)
+
+; could save one cycle if the counters were allocated at link time and
+; the contents of __profil_offset were pre-computed at link time, like this:
+#if 0
+; __profil_offset needs to be PROVIDEd as __profile_base-text/4
+	.global	__profil_offset
+	.balign 4
+__profil_irq:
+	push_s	r0
+	lsr	r0,ilink1,2
+	add1	r0,__profil_offset,r0
+	push_s	r1
+	ldw.di	r1,[r0]
+
+
+	add_s	r1,r1,1
+	bbit1	r1,16,nostore
+	stw.di	r1,[r0]
+nostore:pop_s	r1
+	pop_s	r0
+	j	[ilink1]
+#endif /* 0 */

libgcc/config/arc/gmon/sys/gmon.h

+/*-
+ * Copyright (c) 1982, 1986, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ * Copyright (C) 2007-2012 Free Software Foundation, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *	@(#)gmon.h	8.2 (Berkeley) 1/4/94
+ */
+
+#ifndef	_SYS_GMON_H
+#define	_SYS_GMON_H	1
+
+#if 0
+#include <features.h>
+#include <sys/types.h>
+#else
+#include <sys/types.h>
+#include "machine-gmon.h"
+#define attribute_hidden __attribute__ ((visibility("hidden")))
+#endif
+
+#include <stdint.h>
+
+/*
+ * See gmon_out.h for gmon.out format.
+ */
+
+/* structure emitted by "gcc -a".  This must match struct bb in
+   gcc/libgcc2.c.  It is OK for gcc to declare a longer structure as
+   long as the members below are present.  */
+struct __bb
+{
+  long			zero_word;
+  const char		*filename;
+  long			*counts;
+  long			ncounts;
+  struct __bb		*next;
+  const unsigned long	*addresses;
+};
+
+extern struct __bb *__bb_head;
+
+/*
+ * histogram counters are unsigned shorts (according to the kernel).
+ */
+#define	HISTCOUNTER	unsigned short
+
+/*
+ * fraction of text space to allocate for histogram counters here, 1/2
+ */
+#define	HISTFRACTION	2
+
+/*
+ * Fraction of text space to allocate for from hash buckets.
+ * The value of HASHFRACTION is based on the minimum number of bytes
+ * of separation between two subroutine call points in the object code.
+ * Given MIN_SUBR_SEPARATION bytes of separation the value of
+ * HASHFRACTION is calculated as:
+ *
+ *	HASHFRACTION = MIN_SUBR_SEPARATION / (2 * sizeof(short) - 1);
+ *
+ * For example, on the VAX, the shortest two call sequence is:
+ *
+ *	calls	$0,(r0)
+ *	calls	$0,(r0)
+ *
+ * which is separated by only three bytes, thus HASHFRACTION is
+ * calculated as:
+ *
+ *	HASHFRACTION = 3 / (2 * 2 - 1) = 1
+ *
+ * Note that the division above rounds down, thus if MIN_SUBR_FRACTION
+ * is less than three, this algorithm will not work!
+ *
+ * In practice, however, call instructions are rarely at a minimal
+ * distance.  Hence, we will define HASHFRACTION to be 2 across all
+ * architectures.  This saves a reasonable amount of space for
+ * profiling data structures without (in practice) sacrificing
+ * any granularity.
+ */
+#define	HASHFRACTION	2
+
+/*
+ * Percent of text space to allocate for tostructs.
+ * This is a heuristic; we will fail with a warning when profiling programs
+ * with a very large number of very small functions, but that's
+ * normally OK.
+ * 2 is probably still a good value for normal programs.
+ * Profiling a test case with 64000 small functions will work if
+ * you raise this value to 3 and link statically (which bloats the
+ * text size, thus raising the number of arcs expected by the heuristic).
+ */
+#define ARCDENSITY	3
+
+/*
+ * Always allocate at least this many tostructs.  This
+ * hides the inadequacy of the ARCDENSITY heuristic, at least
+ * for small programs.
+ */
+#define MINARCS		50
+
+/*
+ * The type used to represent indices into gmonparam.tos[].
+ */
+#define	ARCINDEX	u_long
+
+/* 
+ * Maximum number of arcs we want to allow.
+ * Used to be max representable value of ARCINDEX minus 2, but now 
+ * that ARCINDEX is a long, that's too large; we don't really want 
+ * to allow a 48 gigabyte table.
+ * The old value of 1<<16 wasn't high enough in practice for large C++
+ * programs; will 1<<20 be adequate for long?  FIXME
+ */
+#define MAXARCS		(1 << 20)
+
+struct tostruct {
+	u_long		selfpc;
+	long		count;
+	ARCINDEX	link;
+};
+
+/*
+ * a raw arc, with pointers to the calling site and
+ * the called site and a count.
+ */
+struct rawarc {
+	u_long	raw_frompc;
+	u_long	raw_selfpc;
+	long	raw_count;
+};
+
+/*
+ * general rounding functions.
+ */
+#define ROUNDDOWN(x,y)	(((x)/(y))*(y))
+#define ROUNDUP(x,y)	((((x)+(y)-1)/(y))*(y))
+
+/*
+ * The profiling data structures are housed in this structure.
+ */
+struct gmonparam {
+	long int	state;
+	u_short		*kcount;
+	u_long		kcountsize;
+	ARCINDEX	*froms;
+	u_long		fromssize;
+	struct tostruct	*tos;
+	u_long		tossize;
+	long		tolimit;
+	u_long		lowpc;
+	u_long		highpc;
+	u_long		textsize;
+	u_long		hashfraction;
+	long		log_hashfraction;
+};
+extern struct gmonparam _gmonparam;
+
+/*
+ * Possible states of profiling.
+ */
+#define	GMON_PROF_ON	0
+#define	GMON_PROF_BUSY	1
+#define	GMON_PROF_ERROR	2
+#define	GMON_PROF_OFF	3
+
+/*
+ * Sysctl definitions for extracting profiling information from the kernel.
+ */
+#define	GPROF_STATE	0	/* int: profiling enabling variable */
+#define	GPROF_COUNT	1	/* struct: profile tick count buffer */
+#define	GPROF_FROMS	2	/* struct: from location hash bucket */
+#define	GPROF_TOS	3	/* struct: destination/count structure */
+#define	GPROF_GMONPARAM	4	/* struct: profiling parameters (see above) */
+
+__BEGIN_DECLS
+
+/* Set up data structures and start profiling.  */
+extern void __monstartup (u_long __lowpc, u_long __highpc) __THROW;
+extern void monstartup (u_long __lowpc, u_long __highpc) __THROW;
+
+/* Clean up profiling and write out gmon.out.  */
+extern void _mcleanup (void) __THROW;
+
+extern void __write_profiling (void);
+extern int attribute_hidden __profile_frequency (void);
+
+extern u_long __arc_profile_desc_secstart[], __arc_profile_desc_secend[];
+extern u_long __arc_profile_forward_secstart[], __arc_profile_forward_secend[];
+extern u_long __arc_profile_counters_secstart[];
+
+__END_DECLS
+
+#endif /* sys/gmon.h */

libgcc/config/arc/gmon/sys/gmon_out.h

+/* Copyright (C) 2007-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#define GMON_TAG_TIME_HIST 0
+#define GMON_TAG_CG_ARC 1
+#define GMON_TAG_BB_COUNT 2
+
+#define GMON_MAGIC "gmon"
+#define GMON_VERSION 1
+
+struct arc_gmon_hist_hdr
+{
+  char low_pc[4];
+  char high_pc[4];
+  char hist_size[4];
+  char prof_rate[4];
+  char dimen[15];
+  char dimen_abbrev;
+};
+
+struct gmon_cg_arc_record
+{
+  char afrompc[4];
+  char selfpc[4];
+  char count[4];
+};
+
+struct gmon_hdr
+{
+  char cookie[4];
+  char version[4];
+  char c[12];
+};

libgcc/config/arc/ieee-754/addsf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+#if 0 /* DEBUG */
+	.global __addsf3
+	FUNC(__addsf3)
+	.balign 4
+__addsf3:
+	push_s blink
+	push_s r1
+	bl.d __addsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __addsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__addsf3)
+	.global __subsf3
+	FUNC(__subsf3)
+	.balign 4
+__subsf3:
+	push_s blink
+	push_s r1
+	bl.d __subsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __subsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__subsf3)
+#define __addsf3 __addsf3_asm
+#define __subsf3 __subsf3_asm
+#endif /* DEBUG */
+/* N.B. This is optimized for ARC700.
+  ARC600 has very different scheduling / instruction selection criteria.  */
+
+/* inputs: r0, r1
+   output: r0
+   clobber: r1-r10, r12, flags  */
+
+	.balign 4
+	.global __addsf3
+	.global __subsf3
+	FUNC(__addsf3)
+	FUNC(__subsf3)
+	.long 0x7f800000 ; exponent mask
+__subsf3:
+	bxor_l r1,r1,31
+__addsf3:
+	ld r9,[pcl,-8]
+	bmsk r4,r0,30
+	xor r10,r0,r1
+	and r6,r1,r9
+	sub.f r12,r4,r6
+	asr_s r12,r12,23
+	blo .Ldbl1_gt
+	brhs r4,r9,.Linf_nan
+	brne r12,0,.Lsmall_shift
+	brge r10,0,.Ladd_same_exp ; r12 == 0
+/* After subtracting, we need to normalize; when shifting to place the
+  leading 1 into position for the implicit 1 and adding that to DBL0,
+  we increment the exponent.  Thus, we have to subtract one more than
+  the shift count from the exponent beforehand.  Iff the exponent drops thus
+  below zero (before adding in the fraction with the leading one), we have
+  generated a denormal number.  Denormal handling is basicallly reducing the
+  shift count so that we produce a zero exponent instead; FWIW, this way
+  the shift count can become zero (if we started out with exponent 1).
+  On the plus side, we don't need to check for denorm input, the result
+  of subtracing these looks just the same as denormals generated during
+  subtraction.  */
+	bmsk r7,r1,30
+	breq	r4,r7,.Lret0
+	sub.f r5,r4,r7
+	lsr r12,r4,23
+	neg.cs r5,r5
+	norm r3,r5
+	bmsk r2,r0,22
+	sub_s r3,r3,6
+	min r12,r12,r3
+	bic r1,r0,r2
+	sub_s r3,r12,1
+	asl_s r12,r12,23
+	asl r2,r5,r3
+	sub_s r1,r1,r12
+	add_s r0,r1,r2
+	j_s.d [blink]
+	bxor.cs r0,r0,31
+	.balign 4
+.Linf_nan:
+	; If both inputs are inf, but with different signs, the result is NaN.
+	asr r12,r10,31
+	or_s r1,r1,r12
+	j_s.d [blink]
+	or.eq r0,r0,r1
+	.balign 4
+.Ladd_same_exp:
+	/* This is a special case because we can't test for need to shift
+	   down by checking if bit 23 of DBL0 changes.  OTOH, here we know
+	   that we always need to shift down.  */
+	; adding the two floating point numbers together makes the sign
+	; cancel out and apear as carry; the exponent is doubled, and the
+	; fraction also in need of shifting left by one. The two implicit
+	; ones of the sources make an implicit 1 of the result, again
+	; non-existent in a place shifted by one.
+	add.f	r0,r0,r1
+	btst_s	r0,1
+	breq	r6,0,.Ldenorm_add
+	add.ne	r0,r0,1 ; round to even.
+	rrc	r0,r0
+	bmsk	r1,r9,23
+	add	r0,r0,r1 ; increment exponent
+	bic.f	0,r9,r0; check for overflow -> infinity.
+	jne_l	[blink]
+	mov_s	r0,r9
+	j_s.d	[blink]
+	bset.cs	r0,r0,31
+
+.Ldenorm_add:
+	j_s.d [blink]
+	add r0,r4,r1
+
+.Lret_dbl0:
+        j_s [blink]
+
+	.balign 4
+.Lsmall_shift:
+	brhi r12,25,.Lret_dbl0
+	breq.d r6,0,.Ldenorm_small_shift
+	bmsk_s r1,r1,22
+	bset_s r1,r1,23
+.Lfixed_denorm_small_shift:
+	neg r8,r12
+	asl r5,r1,r8
+	brge.d r10,0,.Ladd
+	lsr_l r1,r1,r12
+/* subtract, abs(DBL0) > abs(DBL1) */
+/* DBL0: original values
+   DBL1: fraction with explicit leading 1, shifted into place
+   r4:  orig. DBL0 & 0x7fffffff
+   r6:  orig. DBL1 & 0x7f800000
+   r9:  0x7f800000
+   r10: orig. DBL0H ^ DBL1H
+   r5 : guard bits */
+	.balign 4
+.Lsub:
+	neg.f r12,r5
+	bmsk r3,r0,22
+	bset r5,r3,23
+	sbc.f r4,r5,r1
+	beq.d .Large_cancel_sub
+	bic r7,r0,r3
+	norm r3,r4
+	bmsk r6,r7,30
+.Lsub_done:
+	sub_s r3,r3,6
+	breq r3,1,.Lsub_done_noshift
+	asl r5,r3,23
+	sub_l r3,r3,1
+	brlo r6,r5,.Ldenorm_sub
+	sub r0,r7,r5
+	neg_s r1,r3
+	lsr.f r2,r12,r1
+	asl_s r12,r12,r3
+	btst_s	r2,0
+	bmsk.eq.f r12,r12,30
+	asl r5,r4,r3
+	add_s r0,r0,r2
+	adc.ne r0,r0,0
+	j_s.d [blink]
+	add_l r0,r0,r5
+
+.Lret0:
+	j_s.d	[blink]
+	mov_l	r0,0
+
+	.balign 4
+.Ldenorm_small_shift:
+	brne.d	r12,1,.Lfixed_denorm_small_shift
+	sub_s	r12,r12,1
+	brlt.d	r10,0,.Lsub
+	mov_s	r5,r12 ; zero r5, and align following code
+.Ladd: ; Both bit 23 of DBL1 and bit 0 of r5 are clear.
+	bmsk	r2,r0,22
+	add_s	r2,r2,r1
+	bbit0.d	r2,23,.Lno_shiftdown
+	add_s	r0,r0,r1
+	bic.f	0,r9,r0; check for overflow -> infinity; eq : infinity
+	bmsk	r1,r2,22
+	lsr.ne.f r2,r2,2; cc: even ; hi: might round down
+	lsr.ne	r1,r1,1
+	rcmp.hi	r5,1; hi : round down
+	bclr.hi	r0,r0,0
+	j_l.d	[blink]
+	sub_s	r0,r0,r1
+
+/* r4: DBL0H & 0x7fffffff
+   r6: DBL1H & 0x7f800000
+   r9: 0x7f800000
+   r10: sign difference
+   r12: shift count (negative) */
+	.balign 4
+.Ldbl1_gt:
+	brhs r6,r9,.Lret_dbl1 ; inf or NaN
+	neg r8,r12
+	brhi r8,25,.Lret_dbl1
+.Lsmall_shift_dbl0:
+	breq.d r6,0,.Ldenorm_small_shift_dbl0
+	bmsk_s r0,r0,22
+	bset_s r0,r0,23
+.Lfixed_denorm_small_shift_dbl0:
+	asl r5,r0,r12
+	brge.d r10,0,.Ladd_dbl1_gt
+	lsr r0,r0,r8
+/* subtract, abs(DBL0) < abs(DBL1) */
+/* DBL0: fraction with explicit leading 1, shifted into place
+   DBL1: original value
+   r6:  orig. DBL1 & 0x7f800000
+   r9:  0x7f800000
+   r5: guard bits */
+	.balign 4
+.Lrsub:
+	neg.f r12,r5
+	bmsk r5,r1,22
+	bic r7,r1,r5
+	bset r5,r5,23
+	sbc.f r4,r5,r0
+	bne.d .Lsub_done ; note: r6 is already set up.
+	norm r3,r4
+	/* Fall through */
+
+/* r4:r12 : unnormalized result fraction
+   r7: result sign and exponent         */
+/* When seeing large cancellation, only the topmost guard bit might be set.  */
+	.balign 4
+.Large_cancel_sub:
+	breq_s	r12,0,.Lret0
+	sub	r0,r7,24<<23
+	xor.f	0,r0,r7 ; test if exponent is negative
+	tst.pl	r9,r0  ; test if exponent is zero
+	jpnz	[blink] ; return if non-denormal result
+	bmsk	r6,r7,30
+	lsr	r3,r6,23
+	xor	r0,r6,r7
+	sub_s	r3,r3,24-22
+	j_s.d	[blink]
+	bset	r0,r0,r3
+
+	; If a denorm is produced, we have an exact result -
+	; no need for rounding.
+	.balign 4
+.Ldenorm_sub:
+	sub r3,r6,1
+	lsr.f r3,r3,23
+	xor r0,r6,r7
+	neg_s r1,r3
+	asl.ne r4,r4,r3
+	lsr_s r12,r12,r1
+	add_s r0,r0,r4
+	j_s.d [blink]
+	add.ne r0,r0,r12
+
+	.balign 4
+.Lsub_done_noshift:
+	add.f 0,r12,r12
+	btst.eq r4,0
+	bclr r4,r4,23
+	add r0,r7,r4
+	j_s.d [blink]
+	adc.ne r0,r0,0
+
+	.balign 4
+.Lno_shiftdown:
+	add.f 0,r5,r5
+	btst.eq r0,0
+	cmp.eq r5,r5
+	j_s.d [blink]
+	add.cs r0,r0,1
+
+.Lret_dbl1:
+	j_s.d [blink]
+	mov_l r0,r1
+	.balign 4
+.Ldenorm_small_shift_dbl0:
+	sub.f r8,r8,1
+	bne.d .Lfixed_denorm_small_shift_dbl0
+	add_s r12,r12,1
+	brlt.d r10,0,.Lrsub
+	mov r5,0
+.Ladd_dbl1_gt: ; both bit 23 of DBL0 and bit 0 of r5 are clear.
+	bmsk	r2,r1,22
+	add_s	r2,r2,r0
+	bbit0.d	r2,23,.Lno_shiftdown_dbl1_gt
+	add_s	r0,r1,r0
+	bic.f	0,r9,r0; check for overflow -> infinity; eq : infinity
+	bmsk	r1,r2,22
+	lsr.ne.f r2,r2,2; cc: even ; hi: might round down
+	lsr.ne	r1,r1,1
+	rcmp.hi	r5,1; hi : round down
+	bclr.hi	r0,r0,0
+	j_l.d	[blink]
+	sub_s	r0,r0,r1
+
+	.balign	4
+.Lno_shiftdown_dbl1_gt:
+	add.f	0,r5,r5
+	btst.eq	r0,0
+	cmp.eq	r5,r5
+	j_s.d	[blink]
+	add.cs	r0,r0,1
+	ENDFUNC(__addsf3)
+	ENDFUNC(__subsf3)

libgcc/config/arc/ieee-754/arc-ieee-754.h

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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 __LITTLE_ENDIAN__
+#define DBL0L r0
+#define DBL0H r1
+#define DBL1L r2
+#define DBL1H r3
+#else
+#define DBL0L r1
+#define DBL0H r0
+#define DBL1L r3
+#define DBL1H r2
+#endif
+#define add_l add
+#define asr_l asr
+#define j_l j
+#define jne_l jne
+#define jeq_l jeq
+#define or_l or
+#define mov_l mov
+#define b_l b
+#define beq_l beq
+#define bne_l bne
+#define brne_l brne
+#define bset_l bset
+#define sub_l sub
+#define sub1_l sub1
+#define lsr_l lsr
+#define xor_l xor
+#define bic_l bic
+#define bmsk_l bmsk
+#define bxor_l bxor
+#define bcs_s blo_s

libgcc/config/arc/ieee-754/arc600-dsp/divsf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/*
+   - calculate 15..18 bit inverse using a table of approximating polynoms.
+     precision is higher for polynoms used to evaluate input with larger
+     value.
+   - do one newton-raphson iteration step to double the precision,
+     then multiply this with the divisor
+	-> more time to decide if dividend is subnormal
+     - the worst error propagation is on the side of the value range
+       with the least initial defect, thus giving us about 30 bits precision.
+ */
+#include "../arc-ieee-754.h"
+#define mlo acc2
+#define mhi acc1
+#define mul64(b,c) mullw 0,b,c` machlw 0,b,c
+#define mulu64(b,c) mululw 0,b,c` machulw 0,b,c
+
+#if 0 /* DEBUG */
+	.global __divsf3
+	FUNC(__divsf3)
+	.balign 4
+__divsf3:
+	push_s blink
+	push_s r1
+	bl.d __divsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __divsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+#if 1
+	bne abort
+	jeq_s [blink]
+	b abort
+#else
+	bne abort
+	j_s [blink]
+#endif
+	ENDFUNC(__divsf3)
+#define __divsf3 __divsf3_asm
+#endif /* DEBUG */
+
+	FUNC(__divsf3)
+	.balign 4
+.Ldivtab:
+	.long 0xfc0ffff0
+	.long 0xf46ffefd
+	.long 0xed1ffd2a
+	.long 0xe627fa8e
+	.long 0xdf7ff73b
+	.long 0xd917f33b
+	.long 0xd2f7eea3
+	.long 0xcd1fe986
+	.long 0xc77fe3e7
+	.long 0xc21fdddb
+	.long 0xbcefd760
+	.long 0xb7f7d08c
+	.long 0xb32fc960
+	.long 0xae97c1ea
+	.long 0xaa27ba26
+	.long 0xa5e7b22e
+	.long 0xa1cfa9fe
+	.long 0x9ddfa1a0
+	.long 0x9a0f990c
+	.long 0x9667905d
+	.long 0x92df878a
+	.long 0x8f6f7e84
+	.long 0x8c27757e
+	.long 0x88f76c54
+	.long 0x85df630c
+	.long 0x82e759c5
+	.long 0x8007506d
+	.long 0x7d3f470a
+	.long 0x7a8f3da2
+	.long 0x77ef341e
+	.long 0x756f2abe
+	.long 0x72f7212d
+	.long 0x709717ad
+	.long 0x6e4f0e44
+	.long 0x6c1704d6
+	.long 0x69e6fb44
+	.long 0x67cef1d7
+	.long 0x65c6e872
+	.long 0x63cedf18
+	.long 0x61e6d5cd
+	.long 0x6006cc6d
+	.long 0x5e36c323
+	.long 0x5c76b9f3
+	.long 0x5abeb0b7
+	.long 0x5916a79b
+	.long 0x57769e77
+	.long 0x55de954d
+	.long 0x54568c4e
+	.long 0x52d6834d
+	.long 0x51667a7f
+	.long 0x4ffe71b5
+	.long 0x4e9e68f1
+	.long 0x4d466035
+	.long 0x4bf65784
+	.long 0x4aae4ede
+	.long 0x496e4646
+	.long 0x48363dbd
+	.long 0x47063547
+	.long 0x45de2ce5
+	.long 0x44be2498
+	.long 0x43a61c64
+	.long 0x4296144a
+	.long 0x41860c0e
+	.long 0x407e03ee
+.L7f800000:
+	.long 0x7f800000
+	.balign 4
+	.global __divsf3_support
+__divsf3_support:
+.Linf_NaN:
+	bclr.f 0,r0,31 ; 0/0 -> NaN
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	bic_s r0,r0,r1
+	sub.eq r0,r0,1
+	j_s.d [blink]
+	or r0,r0,r9
+.Lret0:
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	j_s.d [blink]
+	bic_s r0,r0,r1
+/* N.B. the spacing between divtab and the sub3 to get its address must
+   be a multiple of 8.  */
+__divsf3:
+	ld.as r9,[pcl,-9]; [pcl,(-((.-.L7f800000) >> 2))] ; 0x7f800000
+	sub3 r3,pcl,37;(.-.Ldivtab) >> 3
+	lsr r2,r1,17
+	and.f r11,r1,r9
+	bmsk r5,r2,5
+	beq.d .Ldenorm_fp1
+	asl r6,r1,8
+	and.f r2,r0,r9
+	ld.as r5,[r3,r5]
+	asl r4,r1,9
+	bset r6,r6,31
+	breq.d r11,r9,.Linf_nan_fp1
+.Lpast_denorm_fp1:
+	mululw 0,r5,r4
+	machulw r8,r5,r4
+	breq.d r2,r9,.Linf_nan_fp0
+	asl r5,r5,13
+	sub r7,r5,r8
+	mululw 0,r7,r6
+	machulw r8,r7,r6
+	beq.d .Ldenorm_fp0
+	asl r12,r0,8
+	mulu64 (r8,r7)
+	bset r3,r12,31
+.Lpast_denorm_fp0:
+	cmp_s r3,r6
+	lsr.cc r3,r3,1
+	add_s r2,r2, /* wait for immediate */ \
+		0x3f000000
+	sub r7,r7,mhi ; u1.31 inverse, about 30 bit
+	mulu64 (r3,r7)
+	sbc r2,r2,r11
+	xor.f 0,r0,r1
+	and r0,r2,r9
+	bclr r3,r9,23 ; 0x7f000000
+	brhs.d r2,r3,.Linf_denorm
+	bxor.mi r0,r0,31
+.Lpast_denorm:
+	add r3,mhi,0x22 ; round to nearest or higher
+	tst r3,0x3c ; check if rounding was unsafe
+	lsr r3,r3,6
+	jne.d [blink] ; return if rounding was safe.
+	add_s r0,r0,r3
+        /* work out exact rounding if we fall through here.  */
+        /* We know that the exact result cannot be represented in single
+           precision.  Find the mid-point between the two nearest
+           representable values, multiply with the divisor, and check if
+           the result is larger than the dividend.  */
+        add_s r3,r3,r3
+        sub_s r3,r3,1
+        mulu64 (r3,r6)
+	asr.f 0,r0,1 ; for round-to-even in case this is a denorm
+	rsub r2,r9,25
+        asl_s r12,r12,r2
+        sub.f 0,r12,mlo
+        j_s.d [blink]
+        sub.mi r0,r0,1
+.Linf_nan_fp1:
+	lsr_s r0,r0,31
+	bmsk.f 0,r1,22
+	asl_s r0,r0,31
+	bne_s 0f ; inf/inf -> nan
+	brne r2,r9,.Lsigned0 ; x/inf -> 0, but x/nan -> nan
+0:	j_s.d [blink]
+	mov r0,-1
+.Lsigned0:
+.Linf_nan_fp0:
+	tst_s r1,r1
+	j_s.d [blink]
+	bxor.mi r0,r0,31
+	.balign 4
+	.global __divsf3
+/* For denormal results, it is possible that an exact result needs
+   rounding, and thus the round-to-even rule has to come into play.  */
+.Linf_denorm:
+	brlo r2,0xc0000000,.Linf
+.Ldenorm:
+	asr_s r2,r2,23
+	bic r0,r0,r9
+	neg r9,r2
+	brlo.d r9,25,.Lpast_denorm
+	lsr r3,mlo,r9
+	/* Fall through: return +- 0 */
+	j_s [blink]
+.Linf:
+	j_s.d [blink]
+	or r0,r0,r9
+	.balign 4
+.Ldenorm_fp1:
+	norm.f r12,r6 ; flag for x/0 -> Inf check
+	add r6,r6,r6
+	rsub r5,r12,16
+	ror r5,r1,r5
+	bmsk r5,r5,5
+	bic.ne.f 0, \
+		0x60000000,r0 ; large number / denorm -> Inf
+	ld.as r5,[r3,r5]
+	asl r6,r6,r12
+	beq.d .Linf_NaN
+	and.f r2,r0,r9
+	add r4,r6,r6
+	asl_s r12,r12,23
+	bne.d .Lpast_denorm_fp1
+	add_s r2,r2,r12
+.Ldenorm_fp0:
+	mulu64 (r8,r7)
+	bclr r12,r12,31
+	norm.f r3,r12 ; flag for 0/x -> 0 check
+	bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
+	beq_s .Lret0
+	asl_s r12,r12,r3
+	asl_s r3,r3,23
+	add_s r12,r12,r12
+	add r11,r11,r3
+	b.d .Lpast_denorm_fp0
+	mov_s r3,r12
+	ENDFUNC(__divsf3)

libgcc/config/arc/ieee-754/arc600-dsp/muldf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __muldf3
+	.balign 4
+__muldf3:
+	push_s blink
+	push_s r2
+	push_s r3
+	push_s r0
+	bl.d __muldf3_c
+	push_s r1
+	ld_s r2,[sp,12]
+	ld_s r3,[sp,8]
+	st_s r0,[sp,12]
+	st_s r1,[sp,8]
+	pop_s r1
+	bl.d __muldf3_asm
+	pop_s r0
+	pop_s r3
+	pop_s r2
+	pop_s blink
+	cmp r0,r2
+	cmp.eq r1,r3
+	jeq_s [blink]
+	b abort
+#define __muldf3 __muldf3_asm
+#endif /* DEBUG */
+
+__muldf3_support: /* This label makes debugger output saner.  */
+	.balign 4
+	FUNC(__muldf3)
+.Ldenorm_2:
+	breq.d DBL1L,0,.Lret0_2 ; 0 input -> 0 output
+	norm.f r12,DBL1L
+	mov.mi r12,21
+	add.pl r12,r12,22
+	neg r11,r12
+	asl_s r12,r12,20
+	lsr.f DBL1H,DBL1L,r11
+	ror DBL1L,DBL1L,r11
+	sub_s DBL0H,DBL0H,r12
+	mov.eq DBL1H,DBL1L
+	sub_l DBL1L,DBL1L,DBL1H
+	/* Fall through.  */
+	.global __muldf3
+	.balign 4
+__muldf3:
+	mululw 0,DBL0L,DBL1L
+	machulw r4,DBL0L,DBL1L
+	ld.as r9,[pcl,0x67] ; ((.L7ff00000-.+2)/4)]
+	bmsk r6,DBL0H,19
+	bset r6,r6,20
+	mov r8,acc2
+	mululw 0,r4,1
+	and r11,DBL0H,r9
+	breq.d r11,0,.Ldenorm_dbl0
+	and r12,DBL1H,r9
+	breq.d r12,0,.Ldenorm_dbl1
+	maclw 0,r6,DBL1L
+	machulw 0,r6,DBL1L
+	breq.d r11,r9,.Linf_nan
+	bmsk r10,DBL1H,19
+	breq.d r12,r9,.Linf_nan
+	bset r10,r10,20
+	maclw 0,r10,DBL0L
+	machulw r5,r10,DBL0L
+	add_s r12,r12,r11 ; add exponents
+	mov r4,acc2
+	mululw 0,r5,1
+	maclw 0,r6,r10
+	machulw r7,r6,r10 ; fraction product in r7:acc2:r4:r8
+	tst r8,r8
+	bclr r8,r9,30 ; 0x3ff00000
+	bset.ne r4,r4,0 ; put least significant word into sticky bit
+	bclr r6,r9,20 ; 0x7fe00000
+	lsr.f r10,r7,9
+	rsub.eq r8,r8,r9 ; 0x40000000
+	sub r12,r12,r8 ; subtract bias + implicit 1
+	brhs.d r12,r6,.Linf_denorm
+	rsub r10,r10,12
+.Lshift_frac:
+	neg r8,r10
+	asl r6,r4,r10
+	lsr DBL0L,r4,r8
+	add.f 0,r6,r6
+	btst.eq DBL0L,0
+	cmp.eq r4,r4 ; round to nearest / round to even
+	asl r4,acc2,r10
+	lsr r5,acc2,r8
+	adc.f DBL0L,DBL0L,r4
+	xor.f 0,DBL0H,DBL1H
+	asl r7,r7,r10
+	add_s r12,r12,r5
+	adc DBL0H,r12,r7
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+
+/* N.B. This is optimized for ARC700.
+  ARC600 has very different scheduling / instruction selection criteria.  */
+
+/* If one number is denormal, subtract some from the exponent of the other
+   one (if the other exponent is too small, return 0), and normalize the
+   denormal.  Then re-run the computation.  */
+.Lret0_2:
+	lsr_s DBL0H,DBL0H,31
+	asl_s DBL0H,DBL0H,31
+	j_s.d [blink]
+	mov_s DBL0L,0
+	.balign 4
+.Ldenorm_dbl0:
+	mov_s r12,DBL0L
+	mov_s DBL0L,DBL1L
+	mov_s DBL1L,r12
+	mov_s r12,DBL0H
+	mov_s DBL0H,DBL1H
+	mov_s DBL1H,r12
+	and r11,DBL0H,r9
+.Ldenorm_dbl1:
+	brhs r11,r9,.Linf_nan
+	brhs 0x3ca00001,r11,.Lret0
+	sub_s DBL0H,DBL0H,DBL1H
+	bmsk.f DBL1H,DBL1H,30
+	add_s DBL0H,DBL0H,DBL1H
+	beq.d .Ldenorm_2
+	norm r12,DBL1H
+	sub_s r12,r12,10
+	asl r5,r12,20
+	asl_s DBL1H,DBL1H,r12
+	sub DBL0H,DBL0H,r5
+	neg r5,r12
+	lsr r6,DBL1L,r5
+	asl_s DBL1L,DBL1L,r12
+	b.d __muldf3
+	add_s DBL1H,DBL1H,r6
+
+.Lret0:	xor_s DBL0H,DBL0H,DBL1H
+	bclr DBL1H,DBL0H,31
+	xor_s DBL0H,DBL0H,DBL1H
+	j_s.d [blink]
+	mov_s DBL0L,0
+
+	.balign 4
+.Linf_nan:
+	bclr r12,DBL1H,31
+	xor_s DBL1H,DBL1H,DBL0H
+	bclr_s DBL0H,DBL0H,31
+	max r8,DBL0H,r12 ; either NaN -> NaN ; otherwise inf
+	or.f 0,DBL0H,DBL0L
+	mov_s DBL0L,0
+	or.ne.f DBL1L,DBL1L,r12
+	not_s DBL0H,DBL0L ; inf * 0 -> NaN
+	mov.ne DBL0H,r8
+	tst_s DBL1H,DBL1H
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+
+/* We have checked for infinitey / NaN input before, and transformed
+   denormalized inputs into normalized inputs.  Thus, the worst case
+   exponent overflows are:
+       1 +     1 - 0x400 == 0xc02 : maximum underflow
+   0x7fe + 0x7fe - 0x3ff == 0xbfd ; maximum overflow
+   N.B. 0x7e and 0x7f are also values for overflow.
+
+   If (r12 <= -54), we have an underflow to zero.  */
+	.balign 4
+.Linf_denorm:
+	lsr r6,r12,28
+	brlo.d r6,0xc,.Linf
+	asr r6,r12,20
+	add.f r10,r10,r6
+	brgt.d r10,0,.Lshift_frac
+	mov_s r12,0
+	beq.d .Lround_frac
+	add r10,r10,32
+.Lshift32_frac:
+	tst r4,r4
+	mov r4,acc2
+	bset.ne r4,r4,1
+	mululw 0,r7,1
+	brge.d r10,1,.Lshift_frac
+	mov r7,0
+	breq.d r10,0,.Lround_frac
+	add r10,r10,32
+	brgt r10,21,.Lshift32_frac
+	b_s .Lret0
+
+.Lround_frac:
+	add.f 0,r4,r4
+	btst.eq acc2,0
+	mov_s DBL0L,acc2
+	mov_s DBL0H,r7
+	adc.eq.f DBL0L,DBL0L,0
+	j_s.d [blink]
+	adc.eq DBL0H,DBL0H,0
+
+.Linf:	mov_s DBL0L,0
+	xor.f DBL1H,DBL1H,DBL0H
+	mov_s DBL0H,r9
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+	ENDFUNC(__muldf3)
+
+	.balign 4
+.L7ff00000:
+	.long 0x7ff00000

libgcc/config/arc/ieee-754/arc600-dsp/mulsf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __mulsf3
+	FUNC(__mulsf3)
+	.balign 4
+__mulsf3:
+	push_s blink
+	push_s r1
+	bl.d __mulsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __mulsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	and r12,r0,r1
+	bic.f 0,0x7f800000,r12
+	bne 0f
+	bmsk.f 0,r0,22
+	bmsk.ne.f r1,r1,22
+	jne_s [blink] ; both NaN -> OK
+0:	bl abort
+	ENDFUNC(__mulsf3)
+#define __mulsf3 __mulsf3_asm
+#endif /* DEBUG */
+
+	.balign	4
+	.global	__mulsf3
+	FUNC(__mulsf3)
+__mulsf3:
+	ld.as	r9,[pcl,80]; [pcl,((.L7f800000-.+2)/4)]
+	bmsk	r4,r1,22
+	bset	r2,r0,23
+	asl_s	r2,r2,8
+	bset	r3,r4,23
+	and	r11,r0,r9
+	breq.d	r11,0,.Ldenorm_dbl0
+	and	r12,r1,r9
+	breq.d	r12,0,.Ldenorm_dbl1
+	xor_s	r0,r0,r1
+	mululw	0,r2,r3
+	machulw	r6,r2,r3
+	breq.d	r11,r9,.Linf_nan_dbl0
+	ld.as	r4,[pcl,69]; [pcl,((.L7fffffff-.+2)/4)]
+	breq.d	r12,r9,.Linf_nan_dbl1
+.Lpast_denorm:
+	asl.f	0,r6,8
+	mov	r7,acc2
+	add.pl	r6,r6,r6
+	bclr.pl	r6,r6,23
+	add.pl.f r7,r7,r7
+	add.cs	r6,r6,1
+	lsr.f	0,r6,1
+	add_s	r12,r12,r11
+	adc.f	0,r7,r4
+	add_s	r12,r12, \
+		-0x3f800000
+	adc.f	r8,r6,r12
+	tst.pl	r8,r9
+	bic	r0,r0,r4
+	min	r3,r8,r9
+	jpnz.d	[blink]
+	add.pnz	r0,r0,r3
+; infinity or denormal number
+	add.ne.f r3,r3,r3
+	asr_s	r3,r3,23+1
+	bset	r6,r6,23
+	bpnz.d	.Linfinity
+	sub_s	r3,r3,1
+	neg_s	r2,r3
+	brhi.d	r2,24,.Lret_r0 ; right shift shift > 24 -> return +-0
+	lsr	r2,r6,r2
+	asl	r9,r6,r3
+	lsr.f	0,r2,1
+	tst	r7,r7
+	add_s	r0,r0,r2
+	bset.ne	r9,r9,0
+	adc.f	0,r9,r4
+	j_s.d	[blink]
+	add.cs	r0,r0,1
+.Linfinity:
+	j_s.d	[blink]
+	add_s	r0,r0,r9
+
+.Lret_r0: j_s [blink]
+
+	.balign	4
+.Ldenorm_dbl0:
+	bclr_s	r2,r2,31
+	norm.f	r4,r2
+	add_s	r2,r2,r2
+	asl	r2,r2,r4
+	breq.d	r12,r9,.Ldenorm_dbl0_inf_nan_dbl1
+	asl	r4,r4,23
+	mululw	0,r2,r3
+	machulw	r6,r2,r3
+	sub.ne.f r12,r12,r4
+	ld.as	r4,[pcl,28]; [pcl,((.L7fffffff-.+2)/4)]
+	bhi.d	.Lpast_denorm
+	xor_s	r0,r0,r1
+	bmsk	r1,r0,30
+	j_s.d	[blink]
+	bic_s	r0,r0,r1
+
+	.balign	4
+.Ldenorm_dbl0_inf_nan_dbl1:
+	bmsk.f	0,r0,30
+	mov.eq	r1,-1
+.Linf_nan_dbl1:
+	xor_s	r1,r1,r0
+.Linf_nan_dbl0:
+	bclr_s	r1,r1,31
+	j_s.d	[blink]
+	xor_s	r0,r0,r1
+
+	.balign	4
+.Ldenorm_dbl1:
+	breq.d	r11,r9,.Linf_nan_dbl0_2
+	norm.f	r3,r4
+	sub_s	r3,r3,7
+	asl	r4,r4,r3
+	mululw	0,r2,r4
+	machulw	r6,r2,r4
+	sub_s	r3,r3,1
+	asl_s	r3,r3,23
+	sub.ne.f r11,r11,r3
+	ld.as	r4,[pcl,11]; [pcl,((.L7fffffff-.+2)/4)]
+	bhi.d	.Lpast_denorm
+	bmsk	r8,r0,30
+	j_s.d	[blink]
+	bic	r0,r0,r8
+
+	.balign	4
+.Linf_nan_dbl0_2:
+	bclr_s	r1,r1,31
+	xor_s	r0,r0,r1
+	sub.eq	r1,r1,1 ; inf/nan * 0 -> nan
+	bic.f	0,r9,r1
+	j_s.d	[blink]
+	or.eq	r0,r0,r1 ; r1 nan -> result nan
+
+	.balign	4
+.L7f800000:
+	.long	0x7f800000
+.L7fffffff:
+	.long	0x7fffffff
+	ENDFUNC(__mulsf3)

libgcc/config/arc/ieee-754/arc600-mul64/divsf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/*
+   - calculate 15..18 bit inverse using a table of approximating polynoms.
+     precision is higher for polynoms used to evaluate input with larger
+     value.
+   - do one newton-raphson iteration step to double the precision,
+     then multiply this with the divisor
+	-> more time to decide if dividend is subnormal
+     - the worst error propagation is on the side of the value range
+       with the least initial defect, thus giving us about 30 bits precision.
+ */
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __divsf3
+	FUNC(__divsf3)
+	.balign 4
+__divsf3:
+	push_s blink
+	push_s r1
+	bl.d __divsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __divsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+#if 1
+	bne abort
+	jeq_s [blink]
+	b abort
+#else
+	bne abort
+	j_s [blink]
+#endif
+	ENDFUNC(__divsf3)
+#define __divsf3 __divsf3_asm
+#endif /* DEBUG */
+
+	FUNC(__divsf3)
+	.balign 4
+.Ldivtab:
+	.long 0xfc0ffff0
+	.long 0xf46ffefd
+	.long 0xed1ffd2a
+	.long 0xe627fa8e
+	.long 0xdf7ff73b
+	.long 0xd917f33b
+	.long 0xd2f7eea3
+	.long 0xcd1fe986
+	.long 0xc77fe3e7
+	.long 0xc21fdddb
+	.long 0xbcefd760
+	.long 0xb7f7d08c
+	.long 0xb32fc960
+	.long 0xae97c1ea
+	.long 0xaa27ba26
+	.long 0xa5e7b22e
+	.long 0xa1cfa9fe
+	.long 0x9ddfa1a0
+	.long 0x9a0f990c
+	.long 0x9667905d
+	.long 0x92df878a
+	.long 0x8f6f7e84
+	.long 0x8c27757e
+	.long 0x88f76c54
+	.long 0x85df630c
+	.long 0x82e759c5
+	.long 0x8007506d
+	.long 0x7d3f470a
+	.long 0x7a8f3da2
+	.long 0x77ef341e
+	.long 0x756f2abe
+	.long 0x72f7212d
+	.long 0x709717ad
+	.long 0x6e4f0e44
+	.long 0x6c1704d6
+	.long 0x69e6fb44
+	.long 0x67cef1d7
+	.long 0x65c6e872
+	.long 0x63cedf18
+	.long 0x61e6d5cd
+	.long 0x6006cc6d
+	.long 0x5e36c323
+	.long 0x5c76b9f3
+	.long 0x5abeb0b7
+	.long 0x5916a79b
+	.long 0x57769e77
+	.long 0x55de954d
+	.long 0x54568c4e
+	.long 0x52d6834d
+	.long 0x51667a7f
+	.long 0x4ffe71b5
+	.long 0x4e9e68f1
+	.long 0x4d466035
+	.long 0x4bf65784
+	.long 0x4aae4ede
+	.long 0x496e4646
+	.long 0x48363dbd
+	.long 0x47063547
+	.long 0x45de2ce5
+	.long 0x44be2498
+	.long 0x43a61c64
+	.long 0x4296144a
+	.long 0x41860c0e
+	.long 0x407e03ee
+.L7f800000:
+	.long 0x7f800000
+	.balign 4
+	.global __divsf3_support
+__divsf3_support:
+.Linf_NaN:
+	bclr.f 0,r0,31 ; 0/0 -> NaN
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	bic_s r0,r0,r1
+	sub.eq r0,r0,1
+	j_s.d [blink]
+	or r0,r0,r9
+.Lret0:
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	j_s.d [blink]
+	bic_s r0,r0,r1
+/* N.B. the spacing between divtab and the sub3 to get its address must
+   be a multiple of 8.  */
+__divsf3:
+	lsr r2,r1,17
+	sub3 r3,pcl,37 ; (.-.Ldivtab) >> 3
+	bmsk_s r2,r2,5
+	ld.as r5,[r3,r2]
+	asl r4,r1,9
+	ld.as r9,[pcl,-13]; [pcl,(-((.-.L7f800000) >> 2))] ; 0x7f800000
+	mulu64 r5,r4
+	and.f r11,r1,r9
+	asl r6,r1,8
+	bset r6,r6,31
+	beq.d .Ldenorm_fp1
+	asl r5,r5,13
+	breq.d r11,r9,.Linf_nan_fp1
+	and.f r2,r0,r9
+	sub r7,r5,mhi
+	mulu64 r7,r6
+	beq.d .Ldenorm_fp0
+	asl r12,r0,8
+	breq.d r2,r9,.Linf_nan_fp0
+	mulu64 mhi,r7
+.Lpast_denorm_fp1:
+	bset r3,r12,31
+.Lpast_denorm_fp0:
+	cmp_s r3,r6
+	lsr.cc r3,r3,1
+	add_s r2,r2, /* wait for immediate */ \
+		0x3f000000
+	sub r7,r7,mhi ; u1.31 inverse, about 30 bit
+	mulu64 r3,r7
+	sbc r2,r2,r11
+	xor.f 0,r0,r1
+	and r0,r2,r9
+	bclr r3,r9,23 ; 0x7f000000
+	brhs.d r2,r3,.Linf_denorm
+	bxor.mi r0,r0,31
+.Lpast_denorm:
+	add r3,mhi,0x22 ; round to nearest or higher
+	tst r3,0x3c ; check if rounding was unsafe
+	lsr r3,r3,6
+	jne.d [blink] ; return if rounding was safe.
+	add_s r0,r0,r3
+        /* work out exact rounding if we fall through here.  */
+        /* We know that the exact result cannot be represented in single
+           precision.  Find the mid-point between the two nearest
+           representable values, multiply with the divisor, and check if
+           the result is larger than the dividend.  */
+        add_s r3,r3,r3
+        sub_s r3,r3,1
+        mulu64 r3,r6
+	asr.f 0,r0,1 ; for round-to-even in case this is a denorm
+	rsub r2,r9,25
+        asl_s r12,r12,r2
+        sub.f 0,r12,mlo
+        j_s.d [blink]
+        sub.mi r0,r0,1
+.Linf_nan_fp1:
+	lsr_s r0,r0,31
+	bmsk.f 0,r1,22
+	asl_s r0,r0,31
+	bne_s 0f ; inf/inf -> nan
+	brne r2,r9,.Lsigned0 ; x/inf -> 0, but x/nan -> nan
+0:	j_s.d [blink]
+	mov r0,-1
+.Lsigned0:
+.Linf_nan_fp0:
+	tst_s r1,r1
+	j_s.d [blink]
+	bxor.mi r0,r0,31
+	.balign 4
+	.global __divsf3
+/* For denormal results, it is possible that an exact result needs
+   rounding, and thus the round-to-even rule has to come into play.  */
+.Linf_denorm:
+	brlo r2,0xc0000000,.Linf
+.Ldenorm:
+	asr_s r2,r2,23
+	bic r0,r0,r9
+	neg r9,r2
+	brlo.d r9,25,.Lpast_denorm
+	lsr r3,mlo,r9
+	/* Fall through: return +- 0 */
+	j_s [blink]
+.Linf:
+	j_s.d [blink]
+	or r0,r0,r9
+	.balign 4
+.Ldenorm_fp1:
+	bclr r6,r6,31
+	norm.f r12,r6 ; flag for x/0 -> Inf check
+	add r6,r6,r6
+	rsub r5,r12,16
+	ror r5,r1,r5
+	asl r6,r6,r12
+	bmsk r5,r5,5
+	ld.as r5,[r3,r5]
+	add r4,r6,r6
+	; load latency
+	mulu64 r5,r4
+	bic.ne.f 0, \
+		0x60000000,r0 ; large number / denorm -> Inf
+	asl r5,r5,13
+	sub r7,r5,mhi
+	beq.d .Linf_NaN
+	mulu64 r7,r6
+	asl_s r12,r12,23
+	and.f r2,r0,r9
+	add_s r2,r2,r12
+	asl r12,r0,8
+	bne.d .Lpast_denorm_fp1
+.Ldenorm_fp0: mulu64 mhi,r7
+	bclr r12,r12,31
+	norm.f r3,r12 ; flag for 0/x -> 0 check
+	bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
+	beq_s .Lret0
+	asl_s r12,r12,r3
+	asl_s r3,r3,23
+	add_s r12,r12,r12
+	add r11,r11,r3
+	b.d .Lpast_denorm_fp0
+	mov_s r3,r12
+	ENDFUNC(__divsf3)

libgcc/config/arc/ieee-754/arc600-mul64/muldf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __muldf3
+	.balign 4
+__muldf3:
+	push_s blink
+	push_s r2
+	push_s r3
+	push_s r0
+	bl.d __muldf3_c
+	push_s r1
+	ld_s r2,[sp,12]
+	ld_s r3,[sp,8]
+	st_s r0,[sp,12]
+	st_s r1,[sp,8]
+	pop_s r1
+	bl.d __muldf3_asm
+	pop_s r0
+	pop_s r3
+	pop_s r2
+	pop_s blink
+	cmp r0,r2
+	cmp.eq r1,r3
+	jeq_s [blink]
+	and r12,DBL0H,DBL1H
+	bic.f 0,0x7ff80000,r12 ; both NaN -> OK
+	jeq_s [blink]
+	b abort
+#define __muldf3 __muldf3_asm
+#endif /* DEBUG */
+
+__muldf3_support: /* This label makes debugger output saner.  */
+	.balign 4
+	FUNC(__muldf3)
+.Ldenorm_2:
+	breq.d DBL1L,0,.Lret0_2 ; 0 input -> 0 output
+	norm.f r12,DBL1L
+	mov.mi r12,21
+	add.pl r12,r12,22
+	neg r11,r12
+	asl_s r12,r12,20
+	lsr.f DBL1H,DBL1L,r11
+	ror DBL1L,DBL1L,r11
+	sub_s DBL0H,DBL0H,r12
+	mov.eq DBL1H,DBL1L
+	sub_l DBL1L,DBL1L,DBL1H
+	/* Fall through.  */
+	.global __muldf3
+	.balign 4
+__muldf3:
+	mulu64 DBL0L,DBL1L
+	ld.as r9,[pcl,0x68] ; ((.L7ff00000-.+2)/4)]
+	bmsk r6,DBL0H,19
+	bset r6,r6,20
+	and r11,DBL0H,r9
+	breq.d r11,0,.Ldenorm_dbl0
+	and r12,DBL1H,r9
+	breq.d r12,0,.Ldenorm_dbl1
+	mov r8,mlo
+	mov r4,mhi
+	mulu64 r6,DBL1L
+	breq.d r11,r9,.Linf_nan
+	bmsk r10,DBL1H,19
+	breq.d r12,r9,.Linf_nan
+	bset r10,r10,20
+	add.f r4,r4,mlo
+	adc r5,mhi,0
+	mulu64 r10,DBL0L
+	add_s r12,r12,r11 ; add exponents
+	add.f r4,r4,mlo
+	adc r5,r5,mhi
+	mulu64 r6,r10
+	tst r8,r8
+	bclr r8,r9,30 ; 0x3ff00000
+	bset.ne r4,r4,0 ; put least significant word into sticky bit
+	bclr r6,r9,20 ; 0x7fe00000
+	add.f r5,r5,mlo
+	adc r7,mhi,0 ; fraction product in r7:r5:r4
+	lsr.f r10,r7,9
+	rsub.eq r8,r8,r9 ; 0x40000000
+	sub r12,r12,r8 ; subtract bias + implicit 1
+	brhs.d r12,r6,.Linf_denorm
+	rsub r10,r10,12
+.Lshift_frac:
+	neg r8,r10
+	asl r6,r4,r10
+	lsr DBL0L,r4,r8
+	add.f 0,r6,r6
+	btst.eq DBL0L,0
+	cmp.eq r4,r4 ; round to nearest / round to even
+	asl r4,r5,r10
+	lsr r5,r5,r8
+	adc.f DBL0L,DBL0L,r4
+	xor.f 0,DBL0H,DBL1H
+	asl r7,r7,r10
+	add_s r12,r12,r5
+	adc DBL0H,r12,r7
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+
+/* N.B. This is optimized for ARC700.
+  ARC600 has very different scheduling / instruction selection criteria.  */
+
+/* If one number is denormal, subtract some from the exponent of the other
+   one (if the other exponent is too small, return 0), and normalize the
+   denormal.  Then re-run the computation.  */
+.Lret0_2:
+	lsr_s DBL0H,DBL0H,31
+	asl_s DBL0H,DBL0H,31
+	j_s.d [blink]
+	mov_s DBL0L,0
+	.balign 4
+.Ldenorm_dbl0:
+	mov_s r12,DBL0L
+	mov_s DBL0L,DBL1L
+	mov_s DBL1L,r12
+	mov_s r12,DBL0H
+	mov_s DBL0H,DBL1H
+	mov_s DBL1H,r12
+	and r11,DBL0H,r9
+.Ldenorm_dbl1:
+	brhs r11,r9,.Linf_nan
+	brhs 0x3ca00001,r11,.Lret0
+	sub_s DBL0H,DBL0H,DBL1H
+	bmsk.f DBL1H,DBL1H,30
+	add_s DBL0H,DBL0H,DBL1H
+	beq.d .Ldenorm_2
+	norm r12,DBL1H
+	sub_s r12,r12,10
+	asl r5,r12,20
+	asl_s DBL1H,DBL1H,r12
+	sub DBL0H,DBL0H,r5
+	neg r5,r12
+	lsr r6,DBL1L,r5
+	asl_s DBL1L,DBL1L,r12
+	b.d __muldf3
+	add_s DBL1H,DBL1H,r6
+
+.Lret0:	xor_s DBL0H,DBL0H,DBL1H
+	bclr DBL1H,DBL0H,31
+	xor_s DBL0H,DBL0H,DBL1H
+	j_s.d [blink]
+	mov_s DBL0L,0
+
+	.balign 4
+.Linf_nan:
+	bclr r12,DBL1H,31
+	xor_s DBL1H,DBL1H,DBL0H
+	bclr_s DBL0H,DBL0H,31
+	max r8,DBL0H,r12 ; either NaN -> NaN ; otherwise inf
+	or.f 0,DBL0H,DBL0L
+	mov_s DBL0L,0
+	or.ne.f DBL1L,DBL1L,r12
+	not_s DBL0H,DBL0L ; inf * 0 -> NaN
+	mov.ne DBL0H,r8
+	tst_s DBL1H,DBL1H
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+
+/* We have checked for infinitey / NaN input before, and transformed
+   denormalized inputs into normalized inputs.  Thus, the worst case
+   exponent overflows are:
+       1 +     1 - 0x400 == 0xc02 : maximum underflow
+   0x7fe + 0x7fe - 0x3ff == 0xbfd ; maximum overflow
+   N.B. 0x7e and 0x7f are also values for overflow.
+
+   If (r12 <= -54), we have an underflow to zero.  */
+	.balign 4
+.Linf_denorm:
+	lsr r6,r12,28
+	brlo.d r6,0xc,.Linf
+	asr r6,r12,20
+	add.f r10,r10,r6
+	brgt.d r10,0,.Lshift_frac
+	mov_s r12,0
+	beq.d .Lround_frac
+	add r10,r10,32
+.Lshift32_frac:
+	tst r4,r4
+	mov r4,r5
+	bset.ne r4,r4,1
+	mov r5,r7
+	brge.d r10,1,.Lshift_frac
+	mov r7,0
+	breq.d r10,0,.Lround_frac
+	add r10,r10,32
+	brgt r10,21,.Lshift32_frac
+	b_s .Lret0
+
+.Lround_frac:
+	add.f 0,r4,r4
+	btst.eq r5,0
+	mov_s DBL0L,r5
+	mov_s DBL0H,r7
+	adc.eq.f DBL0L,DBL0L,0
+	j_s.d [blink]
+	adc.eq DBL0H,DBL0H,0
+
+.Linf:	mov_s DBL0L,0
+	xor.f DBL1H,DBL1H,DBL0H
+	mov_s DBL0H,r9
+	j_s.d [blink]
+	bset.mi DBL0H,DBL0H,31
+	ENDFUNC(__muldf3)
+
+	.balign 4
+.L7ff00000:
+	.long 0x7ff00000

libgcc/config/arc/ieee-754/arc600-mul64/mulsf3.S

+/* Copyright (C) 2008-2013 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __mulsf3
+	FUNC(__mulsf3)
+	.balign 4
+__mulsf3:
+	push_s blink
+	push_s r1
+	bl.d __mulsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __mulsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	and r12,r0,r1
+	bic.f 0,0x7f800000,r12
+	bne 0f
+	bmsk.f 0,r0,22
+	bmsk.ne.f r1,r1,22
+	jne_s [blink] ; both NaN -> OK
+0:	bl abort
+	ENDFUNC(__mulsf3)
+#define __mulsf3 __mulsf3_asm
+#endif /* DEBUG */
+
+	.balign	4
+	.global	__mulsf3
+	FUNC(__mulsf3)
+__mulsf3:
+	ld.as	r9,[pcl,80]; [pcl,((.L7f800000-.+2)/4)]
+	bmsk	r4,r1,22
+	bset	r2,r0,23
+	asl_s	r2,r2,8
+	bset	r3,r4,23
+	mulu64	r2,r3
+	and	r11,r0,r9
+	breq.d	r11,0,.Ldenorm_dbl0
+	and	r12,r1,r9
+	breq.d	r12,0,.Ldenorm_dbl1
+	xor_s	r0,r0,r1
+	breq.d	r11,r9,.Linf_nan_dbl0
+	ld.as	r4,[pcl,70]; [pcl,((.L7fffffff-.+2)/4)]
+	breq.d	r12,r9,.Linf_nan_dbl1
+.Lpast_denorm:
+	asl.f	0,mhi,8
+	mov	r6,mhi
+	mov	r7,mlo
+	add.pl	r6,r6,r6
+	bclr.pl	r6,r6,23
+	add.pl.f r7,r7,r7
+	add.cs	r6,r6,1
+	lsr.f	0,r6,1
+	add_s	r12,r12,r11
+	adc.f	0,r7,r4
+	add_s	r12,r12, \
+		-0x3f800000
+	adc.f	r8,r6,r12
+	tst.pl	r8,r9
+	bic	r0,r0,r4
+	min	r3,r8,r9
+	jpnz.d	[blink]
+	add.pnz	r0,r0,r3
+; infinity or denormal number
+	add.ne.f r3,r3,r3
+	asr_s	r3,r3,23+1
+	bset	r6,r6,23
+	bpnz.d	.Linfinity
+	sub_s	r3,r3,1
+	neg_s	r2,r3
+	brhi.d	r2,24,.Lret_r0 ; right shift shift > 24 -> return +-0
+	lsr	r2,r6,r2
+	asl	r9,r6,r3
+	lsr.f	0,r2,1
+	tst	r7,r7
+	add_s	r0,r0,r2
+	bset.ne	r9,r9,0
+	adc.f	0,r9,r4
+	j_s.d	[blink]
+	add.cs	r0,r0,1
+.Linfinity:
+	j_s.d	[blink]
+	add_s	r0,r0,r9
+
+.Lret_r0: j_s [blink]
+
+	.balign	4
+.Ldenorm_dbl0:
+	bclr_s	r2,r2,31
+	norm.f	r4,r2
+	add_s	r2,r2,r2
+	asl	r2,r2,r4
+	mulu64	r2,r3
+	breq.d	r12,r9,.Ldenorm_dbl0_inf_nan_dbl1
+	asl	r4,r4,23
+	sub.ne.f r12,r12,r4
+	ld.as	r4,[pcl,29]; [pcl,((.L7fffffff-.+2)/4)]
+	bhi.d	.Lpast_denorm
+	xor_s	r0,r0,r1
+	bmsk	r1,r0,30
+	j_s.d	[blink]
+	bic_s	r0,r0,r1
+
+	.balign	4
+.Ldenorm_dbl0_inf_nan_dbl1:
+	bmsk.f	0,r0,30
+	beq_s	.Lretnan
+	xor_s	r0,r0,r1
+.Linf_nan_dbl1:
+	xor_s	r1,r1,r0
+.Linf_nan_dbl0:
+	bclr_s	r1,r1,31
+	cmp_s	r1,r9
+	jls.d	[blink]
+	xor_s	r0,r0,r1
+; r1 NaN -> result NaN
+.Lretnan:
+	j_s.d [blink]
+	mov r0,-1
+
+	.balign	4
+.Ldenorm_dbl1:
+	breq.d	r11,r9,.Linf_nan_dbl0_2
+	norm.f	r3,r4
+	sub_s	r3,r3,7
+	asl	r4,r4,r3
+	mulu64	r2,r4
+	sub_s	r3,r3,1
+	asl_s	r3,r3,23
+	sub.ne.f r11,r11,r3
+	ld.as	r4,[pcl,11]; [pcl,((.L7fffffff-.+2)/4)]
+	bhi.d	.Lpast_denorm
+	bmsk	r8,r0,30
+	j_s.d	[blink]
+	bic	r0,r0,r8
+
+	.balign	4
+.Linf_nan_dbl0_2:
+	bclr_s	r1,r1,31
+	xor_s	r0,r0,r1
+	sub.eq	r1,r1,1 ; inf/nan * 0 -> nan
+	bic.f	0,r9,r1
+	j_s.d	[blink]
+	or.eq	r0,r0,r1 ; r1 nan -> result nan
+
+	.balign	4
+.L7f800000:
+	.long	0x7f800000
+.L7fffffff:
+	.long	0x7fffffff
+	ENDFUNC(__mulsf3)

libgcc/config/arc/ieee-754/arc600/divsf3.S

+/* Copyright (C) 2008-2013 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __divsf3
+	FUNC(__divsf3)
+	.balign 4
+__divsf3:
+	push_s blink
+	push_s r1
+	bl.d __divsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __divsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+        jeq_s [blink]
+        and r12,r0,r1
+        bic.f 0,0x7f800000,r12 ; both NaN -> OK
+        jeq_s [blink]
+        bl abort
+	ENDFUNC(__divsf3)
+#define __divsf3 __divsf3_asm
+#endif /* DEBUG */
+
+	.balign 4
+__divdf3_support: /* This label makes debugger output saner.  */
+	FUNC(__divsf3)
+.Ldenorm_fp0:
+	norm.f r12,r2 ; flag for 0/x -> 0 check
+	bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
+	beq_s .Lret0_NaN
+	tst r1,r9
+	add_s r2,r2,r2
+	sub_s r12,r12,8
+	asl_s r2,r2,r12
+	asl_l r12,r12,23
+	bne.d .Lpast_denorm_fp0
+	add r5,r5,r12
+/* r0 is subnormal, r1 is subnormal or 0.  */
+
+	.balign 4
+.Ldenorm_fp1:
+	norm.f r12,r3 ; flag for  x/0 -> Inf check
+	bic.ne.f 0,0x60000000,r0 ; large number/denorm -> Inf
+	beq_s .Linf
+	add_s r3,r3,r3
+	sub_s r12,r12,8
+	asl_s r3,r3,r12
+	asl_s r12,r12,23
+	b.d .Lpast_denorm_fp1
+	add r4,r4,r12
+
+.Lret0_NaN:
+	bclr.f 0,r1,31 ; 0/0 -> NaN
+	bic r0,r10,r9
+	j_s.d [blink]
+	sub.eq r0,r0,1
+
+	.balign 4
+.Linf_nan_fp0:
+	bic.f 0,r9,r1 ; fp1 Inf -> result NaN
+	bic r1,r5,r9  ; fp1 sign
+	sub.eq r1,r1,1
+	j_s.d [blink]
+	xor_s r0,r0,r1
+.Linf_nan_fp1:
+	bic r0,r4,r9 ; fp0 sign
+	bmsk.f 0,r1,22 ; x/inf -> 0, x/nan -> nan
+	xor.eq r1,r1,r9
+	j_s.d [blink]
+	xor_s r0,r0,r1
+
+	.global __divsf3
+	.balign 4
+        .long 0x7f800000 ; exponent mask
+__divsf3:
+	ld r9,[pcl,-4]
+	bmsk r2,r0,22
+	xor r4,r0,r2
+	bmsk r3,r1,22
+	xor r5,r1,r3
+	and r11,r0,r9
+	breq.d r11,0,.Ldenorm_fp0
+	xor r10,r4,r5
+	breq r11,r9,.Linf_nan_fp0
+	bset_s r2,r2,23
+	and r11,r1,r9
+	breq r11,0,.Ldenorm_fp1
+	breq r11,r9,.Linf_nan_fp1
+.Lpast_denorm_fp0:
+	bset_s r3,r3,23
+.Lpast_denorm_fp1:
+	cmp r2,r3
+	asl_s r2,r2,6+1
+	asl_s r3,r3,7
+	add.lo r2,r2,r2
+	bclr r8,r9,30 ; exponent bias
+	bclr.lo r8,r8,23 ; reduce exp by one if fraction is shifted
+	sub r4,r4,r5
+	add r4,r4,r8
+	xor.f 0,r10,r4
+	bmi .Linf_denorm
+	and.f r12,r4,r9
+	beq .Ldenorm
+	sub_s r2,r2,r3 ; discard implicit 1
+	rsub r3,r3,1 ; prime r3 for two-insn divide-step use
+.Ldiv_23bit:
+	.rep 6
+	add1.f r2,r3,r2
+	sub.cc r2,r2,r3
+	.endr
+	breq r12,r9,.Linf
+	bmsk r0,r2,6
+	xor_s r2,r2,r0
+.Ldiv_17bit:
+	.rep 7
+	add1.f r2,r3,r2
+	sub.cc r2,r2,r3
+	.endr
+	asl_s r0,r0,7
+	bmsk r1,r2,6
+	xor_s r2,r2,r1
+	or_s r0,r0,r1
+.Ldiv_10bit:
+	.rep 7
+	add1.f r2,r3,r2
+	sub.cc r2,r2,r3
+	.endr
+	asl_s r0,r0,7
+	bmsk r1,r2,6
+	xor_s r2,r2,r1
+	or_s r0,r0,r1
+.Ldiv_3bit:
+	.rep 3
+	add1.f r2,r3,r2
+	sub.cc r2,r2,r3
+	.endr
+	asl_s r0,r0,3
+.Ldiv_0bit:
+	add1.f r1,r3,r2
+	sub.cc r1,r1,r3
+	bmsk_s r2,r2,2
+	tst r1,-0x7e ; 0xffffff82, test for rest or odd
+	bmsk_s r1,r1,0
+	add_s r0,r0,r2 ; assemble fraction
+	add_s r0,r0,r4 ; add in sign & exponent
+	j_s.d [blink]
+	add.ne r0,r0,r1 ; round to nearest / even
+
+	.balign 4
+.Linf:
+	j_s.d [blink]
+	or r0,r10,r9
+
+.Lret_r4:
+	j_s.d [blink]
+	mov_s r0,r4
+	.balign 4
+.Linf_denorm:
+	add.f r12,r4,r4
+	asr_l r12,r12,24
+	bpl .Linf
+	max r12,r12,-24
+.Ldenorm:
+	rsub r3,r3,1
+	add r1,pcl,68; .Ldenorm_tab-.
+	ldw.as r12,[r1,r12]
+	mov_s r0,0
+	lsr_s r2,r2
+	sub_s r1,r1,r12
+	j_s.d [r1]
+	bic r4,r10,r9
+	.short .Ldenorm_tab-.Lret_r4
+	.short .Ldenorm_tab-.Ldiv_0bit
+	.short .Ldenorm_tab-.Ldiv_3bit-2*8
+	.short .Ldenorm_tab-.Ldiv_3bit-1*8
+	.short .Ldenorm_tab-.Ldiv_3bit
+	.short .Ldenorm_tab-.Ldiv_10bit-6*8
+	.short .Ldenorm_tab-.Ldiv_10bit-5*8
+	.short .Ldenorm_tab-.Ldiv_10bit-3*8
+	.short .Ldenorm_tab-.Ldiv_10bit-3*8
+	.short .Ldenorm_tab-.Ldiv_10bit-2*8
+	.short .Ldenorm_tab-.Ldiv_10bit-1*8
+	.short .Ldenorm_tab-.Ldiv_10bit
+	.short .Ldenorm_tab-.Ldiv_17bit-6*8
+	.short .Ldenorm_tab-.Ldiv_17bit-5*8
+	.short .Ldenorm_tab-.Ldiv_17bit-4*8
+	.short .Ldenorm_tab-.Ldiv_17bit-3*8
+	.short .Ldenorm_tab-.Ldiv_17bit-2*8
+	.short .Ldenorm_tab-.Ldiv_17bit-1*8
+	.short .Ldenorm_tab-.Ldiv_17bit
+	.short .Ldenorm_tab-.Ldiv_23bit-5*8
+	.short .Ldenorm_tab-.Ldiv_23bit-4*8
+	.short .Ldenorm_tab-.Ldiv_23bit-3*8
+	.short .Ldenorm_tab-.Ldiv_23bit-2*8
+	.short .Ldenorm_tab-.Ldiv_23bit-1*8
+.Ldenorm_tab:
+	.short .Ldenorm_tab-.Ldiv_23bit
+	ENDFUNC(__divsf3)

libgcc/config/arc/ieee-754/arc600/mulsf3.S

+/* Copyright (C) 2008-2013 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "../arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __mulsf3
+	FUNC(__mulsf3)
+	.balign 4
+__mulsf3:
+	push_s blink
+	push_s r1
+	bl.d __mulsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __mulsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	and r12,r0,r1
+	bic.f 0,0x7f800000,r12
+	bne 0f
+	bmsk.f 0,r0,22
+	bmsk.ne.f r1,r1,22
+	jne_s [blink] ; both NaN -> OK
+0:	bl abort
+	ENDFUNC(__mulsf3)
+#define __mulsf3 __mulsf3_asm
+#endif /* DEBUG */
+
+	.balign	4
+	.global	__mulsf3
+	FUNC(__mulsf3)
+__mulsf3:
+	ld.as	r9,[pcl,76]; [pcl,((.L7f800000-.+2)/4)]
+	bmsk	r4,r1,22
+	bset	r3,r4,23
+	bmsk	r2,r0,22
+	and	r11,r0,r9
+	breq.d	r11,0,.Ldenorm_dbl0
+	and	r12,r1,r9
+	xor_s	r0,r0,r1
+	breq.d	r11,r9,.Linf_nan_dbl0
+	bset_s	r2,r2,23
+	breq	r12,0,.Ldenorm_dbl1
+	breq	r12,r9,.Linf_nan_dbl1
+.Lpast_denorm:
+	mov r6,0
+	lsr.f r7,r2
+; We could so this a bit faster here with a 32 bit shift register and
+; inserting the r2 factor / retrieving the low result a byte at a time,
+; but that'd increase code size.
+	mov lp_count,24
+	.balign 4
+	lp 0f
+	add.cs r6,r6,r3
+	lsr.f r6,r6
+	rrc.f r7,r7
+0:
+	ld.as	r4,[pcl,59]; [pcl,((.L7fffffff-.+2)/4)]
+	asl.f 0,r6,8
+	add.pl	r6,r6,r6
+	bclr.pl	r6,r6,23
+	add.pl.f r7,r7,r7
+	add.cs	r6,r6,1
+	lsr.f	0,r6,1
+	add_s	r12,r12,r11
+	adc.f	0,r7,r4
+	add_s	r12,r12, \
+		-0x3f800000
+	adc.f	r8,r6,r12
+	tst.pl	r8,r9
+	bic	r0,r0,r4
+	min	r3,r8,r9
+	jpnz.d	[blink]
+	add.pnz	r0,r0,r3
+; infinity or denormal number
+	add.ne.f r3,r3,r3
+	asr_s	r3,r3,23+1
+	bset	r6,r6,23
+	bpnz.d	.Linfinity
+	sub_s	r3,r3,1
+	neg_s	r2,r3
+	brhi.d	r2,24,.Lret_r0 ; right shift shift > 24 -> return +-0
+	lsr	r2,r6,r2
+	asl	r9,r6,r3
+	lsr.f	0,r2,1
+	tst	r7,r7
+	add_s	r0,r0,r2
+	bset.ne	r9,r9,0
+	adc.f	0,r9,r4
+	j_s.d	[blink]
+	add.cs	r0,r0,1
+.Linfinity:
+	j_s.d	[blink]
+	add_s	r0,r0,r9
+
+.Lret_r0: j_s [blink]
+
+	.balign	4
+.Ldenorm_dbl0:
+	asl_s	r2,r2,8
+	norm.f	r4,r2
+	lsr_s	r2,r2,7
+	asl	r2,r2,r4
+	breq.d	r12,r9,.Ldenorm_dbl0_inf_nan_dbl1
+	asl	r4,r4,23
+	sub.ne.f r12,r12,r4
+	bhi.d	.Lpast_denorm
+	xor_s	r0,r0,r1
+	bmsk	r1,r0,30
+	j_s.d	[blink]
+	bic_s	r0,r0,r1
+
+	.balign	4
+.Ldenorm_dbl0_inf_nan_dbl1:
+	bmsk.f	0,r0,30
+	beq_s   .Lretnan
+	xor_s	r0,r0,r1
+.Linf_nan_dbl1:
+	xor_s	r1,r1,r0
+	bclr_s	r1,r1,31
+	j_s.d	[blink]
+	xor_s	r0,r0,r1
+.Linf_nan_dbl0:
+	sub_s   r2,r1,1 ; inf/nan * 0 -> nan; inf * nan -> nan (use |r2| >= inf)
+	bic.f   0,r9,r2
+	xor_s   r0,r0,r1
+	bclr_s  r1,r1,31
+	xor_s   r0,r0,r1
+	jne_s   [blink]
+.Lretnan:
+	j_s.d   [blink]
+	mov     r0,-1
+	.balign	4
+.Ldenorm_dbl1:
+	norm.f	r3,r4
+	sub_s	r3,r3,7
+	asl	r4,r4,r3
+	sub_s	r3,r3,1
+	asl_s	r3,r3,23
+	sub.ne.f r11,r11,r3
+	bhi.d	.Lpast_denorm
+	mov_s	r3,r4
+	bmsk	r3,r0,30
+	j_s.d	[blink]
+	bic_s	r0,r0,r3
+
+	.balign	4
+.L7f800000:
+	.long	0x7f800000
+.L7fffffff:
+	.long	0x7fffffff
+	ENDFUNC(__mulsf3)

libgcc/config/arc/ieee-754/divsf3-stdmul.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/*
+   - calculate 15..18 bit inverse using a table of approximating polynoms.
+     precision is higher for polynoms used to evaluate input with larger
+     value.
+   - do one newton-raphson iteration step to double the precision,
+     then multiply this with the divisor
+	-> more time to decide if dividend is subnormal
+     - the worst error propagation is on the side of the value range
+       with the least initial defect, thus giving us about 30 bits precision.
+ */
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __divsf3
+	FUNC(__divsf3)
+	.balign 4
+__divsf3:
+	push_s blink
+	push_s r1
+	bl.d __divsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __divsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+#if 1
+	bne abort
+	jeq_s [blink]
+	b abort
+#else
+	bne abort
+	j_s [blink]
+#endif
+	ENDFUNC(__divsf3)
+#define __divsf3 __divsf3_asm
+#endif /* DEBUG */
+
+	FUNC(__divsf3)
+	.balign 4
+.L7f800000:
+	.long 0x7f800000
+.Ldivtab:
+	.long 0xfc0ffff0
+	.long 0xf46ffefd
+	.long 0xed1ffd2a
+	.long 0xe627fa8e
+	.long 0xdf7ff73b
+	.long 0xd917f33b
+	.long 0xd2f7eea3
+	.long 0xcd1fe986
+	.long 0xc77fe3e7
+	.long 0xc21fdddb
+	.long 0xbcefd760
+	.long 0xb7f7d08c
+	.long 0xb32fc960
+	.long 0xae97c1ea
+	.long 0xaa27ba26
+	.long 0xa5e7b22e
+	.long 0xa1cfa9fe
+	.long 0x9ddfa1a0
+	.long 0x9a0f990c
+	.long 0x9667905d
+	.long 0x92df878a
+	.long 0x8f6f7e84
+	.long 0x8c27757e
+	.long 0x88f76c54
+	.long 0x85df630c
+	.long 0x82e759c5
+	.long 0x8007506d
+	.long 0x7d3f470a
+	.long 0x7a8f3da2
+	.long 0x77ef341e
+	.long 0x756f2abe
+	.long 0x72f7212d
+	.long 0x709717ad
+	.long 0x6e4f0e44
+	.long 0x6c1704d6
+	.long 0x69e6fb44
+	.long 0x67cef1d7
+	.long 0x65c6e872
+	.long 0x63cedf18
+	.long 0x61e6d5cd
+	.long 0x6006cc6d
+	.long 0x5e36c323
+	.long 0x5c76b9f3
+	.long 0x5abeb0b7
+	.long 0x5916a79b
+	.long 0x57769e77
+	.long 0x55de954d
+	.long 0x54568c4e
+	.long 0x52d6834d
+	.long 0x51667a7f
+	.long 0x4ffe71b5
+	.long 0x4e9e68f1
+	.long 0x4d466035
+	.long 0x4bf65784
+	.long 0x4aae4ede
+	.long 0x496e4646
+	.long 0x48363dbd
+	.long 0x47063547
+	.long 0x45de2ce5
+	.long 0x44be2498
+	.long 0x43a61c64
+	.long 0x4296144a
+	.long 0x41860c0e
+	.long 0x407e03ee
+__divsf3_support: /* This label makes debugger output saner.  */
+.Ldenorm_fp1:
+	bclr r6,r6,31
+	norm.f r12,r6 ; flag for x/0 -> Inf check
+	add r6,r6,r6
+	rsub r5,r12,16
+	ror r5,r1,r5
+	asl r6,r6,r12
+	bmsk r5,r5,5
+	ld.as r5,[r3,r5]
+	add r4,r6,r6
+	; load latency
+	mpyhu r7,r5,r4
+	bic.ne.f 0, \
+		0x60000000,r0 ; large number / denorm -> Inf
+	beq_s .Linf_NaN
+	asl r5,r5,13
+	; wb stall
+	; slow track
+	sub r7,r5,r7
+	mpyhu r8,r7,r6
+	asl_s r12,r12,23
+	and.f r2,r0,r9
+	add r2,r2,r12
+	asl r12,r0,8
+	; wb stall
+	bne.d .Lpast_denorm_fp1
+.Ldenorm_fp0:
+	mpyhu r8,r8,r7
+	bclr r12,r12,31
+	norm.f r3,r12 ; flag for 0/x -> 0 check
+	bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
+	beq_s .Lret0
+	asl_s r12,r12,r3
+	asl_s r3,r3,23
+	add_s r12,r12,r12
+	add r11,r11,r3
+	b.d .Lpast_denorm_fp0
+	mov_s r3,r12
+	.balign 4
+.Linf_NaN:
+	bclr.f 0,r0,31 ; 0/0 -> NaN
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	bic_s r0,r0,r1
+	sub.eq r0,r0,1
+	j_s.d [blink]
+	or r0,r0,r9
+.Lret0:
+	xor_s r0,r0,r1
+	bmsk r1,r0,30
+	j_s.d [blink]
+	bic_s r0,r0,r1
+.Linf_nan_fp1:
+	lsr_s r0,r0,31
+	bmsk.f 0,r1,22
+	asl_s r0,r0,31
+	bne_s 0f ; inf/inf -> nan
+	brne r2,r9,.Lsigned0 ; x/inf -> 0, but x/nan -> nan
+0:	j_s.d [blink]
+	mov r0,-1
+.Lsigned0:
+.Linf_nan_fp0:
+	tst_s r1,r1
+	j_s.d [blink]
+	bxor.mi r0,r0,31
+	.balign 4
+	.global __divsf3
+/* N.B. the spacing between divtab and the sub3 to get its address must
+   be a multiple of 8.  */
+__divsf3:
+	lsr r2,r1,17
+	sub3 r3,pcl,55;(.-.Ldivtab) >> 3
+	bmsk_s r2,r2,5
+	ld.as r5,[r3,r2]
+	asl r4,r1,9
+	ld.as r9,[pcl,-114]; [pcl,(-((.-.L7f800000) >> 2))] ; 0x7f800000
+	mpyhu r7,r5,r4
+	asl r6,r1,8
+	and.f r11,r1,r9
+	bset r6,r6,31
+	asl r5,r5,13
+	; wb stall
+	beq .Ldenorm_fp1
+	sub r7,r5,r7
+	mpyhu r8,r7,r6
+	breq.d r11,r9,.Linf_nan_fp1
+	and.f r2,r0,r9
+	beq.d .Ldenorm_fp0
+	asl r12,r0,8
+	; wb stall
+	breq r2,r9,.Linf_nan_fp0
+	mpyhu r8,r8,r7
+.Lpast_denorm_fp1:
+	bset r3,r12,31
+.Lpast_denorm_fp0:
+	cmp_s r3,r6
+	lsr.cc r3,r3,1
+	add_s r2,r2, /* wait for immediate */ \
+	/* wb stall */ \
+		0x3f000000
+	sub r7,r7,r8 ; u1.31 inverse, about 30 bit
+	mpyhu r3,r3,r7
+	sbc r2,r2,r11
+	xor.f 0,r0,r1
+	and r0,r2,r9
+	bxor.mi r0,r0,31
+	brhs r2, /* wb stall / wait for immediate */ \
+		0x7f000000,.Linf_denorm
+.Lpast_denorm:
+	add_s r3,r3,0x22 ; round to nearest or higher
+	tst r3,0x3c ; check if rounding was unsafe
+	lsr r3,r3,6
+	jne.d [blink] ; return if rounding was safe.
+	add_s r0,r0,r3
+        /* work out exact rounding if we fall through here.  */
+        /* We know that the exact result cannot be represented in single
+           precision.  Find the mid-point between the two nearest
+           representable values, multiply with the divisor, and check if
+           the result is larger than the dividend.  */
+        add_s r3,r3,r3
+        sub_s r3,r3,1
+        mpyu r3,r3,r6
+	asr.f 0,r0,1 ; for round-to-even in case this is a denorm
+	rsub r2,r9,25
+        asl_s r12,r12,r2
+	; wb stall
+	; slow track
+        sub.f 0,r12,r3
+        j_s.d [blink]
+        sub.mi r0,r0,1
+/* For denormal results, it is possible that an exact result needs
+   rounding, and thus the round-to-even rule has to come into play.  */
+.Linf_denorm:
+	brlo r2,0xc0000000,.Linf
+.Ldenorm:
+	asr_s r2,r2,23
+	bic r0,r0,r9
+	neg r9,r2
+	brlo.d r9,25,.Lpast_denorm
+	lsr r3,r3,r9
+	/* Fall through: return +- 0 */
+	j_s [blink]
+.Linf:
+	j_s.d [blink]
+	or r0,r0,r9
+	ENDFUNC(__divsf3)

libgcc/config/arc/ieee-754/divsf3.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __divsf3
+	FUNC(__divsf3)
+	.balign 4
+__divsf3:
+	push_s blink
+	push_s r1
+	bl.d __divsf3_c
+	push_s r0
+	ld_s r1,[sp,4]
+	st_s r0,[sp,4]
+	bl.d __divsf3_asm
+	pop_s r0
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+#if 1
+	bne abort
+	jeq_s [blink]
+	b abort
+#else
+	bne abort
+	j_s [blink]
+#endif
+	ENDFUNC(__divsf3)
+#define __divsf3 __divsf3_asm
+#endif /* DEBUG */
+
+	.balign 4
+__divdf3_support: /* This label makes debugger output saner.  */
+	FUNC(__divsf3)
+.Ldenorm_fp0:
+	norm.f r12,r2 ; flag for 0/x -> 0 check
+	bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
+	beq_s .Lret0_NaN
+	tst r1,r9
+	add_s r2,r2,r2
+	sub_s r12,r12,8
+	asl_s r2,r2,r12
+	asl_l r12,r12,23
+	bne.d .Lpast_denorm_fp0
+	add r5,r5,r12
+/* r0 is subnormal, r1 is subnormal or 0.  */
+
+	.balign 4
+.Ldenorm_fp1:
+	norm.f r12,r3 ; flag for  x/0 -> Inf check
+	bic.ne.f 0,0x60000000,r0 ; large number/denorm -> Inf
+	beq_s .Linf
+	add_s r3,r3,r3
+	sub_s r12,r12,8
+	asl_s r3,r3,r12
+	asl_s r12,r12,23
+	b.d .Lpast_denorm_fp1
+	add r4,r4,r12
+
+.Lret0_NaN:
+	bclr.f 0,r1,31 ; 0/0 -> NaN
+	bic r0,r10,r9
+	j_s.d [blink]
+	sub.eq r0,r0,1
+
+	.global __divsf3
+	.balign 4
+        .long 0x7f800000 ; exponent mask
+__divsf3:
+	ld r9,[pcl,-4]
+	bmsk r2,r0,22
+	xor r4,r0,r2
+	bmsk r3,r1,22
+	xor r5,r1,r3
+	and r11,r0,r9
+	breq.d r11,0,.Ldenorm_fp0
+	xor r10,r4,r5
+	breq r11,r9,.Linf_nan_fp0
+	bset_s r2,r2,23
+	and r11,r1,r9
+	breq r11,0,.Ldenorm_fp1
+	breq r11,r9,.Linf_nan_fp1
+.Lpast_denorm_fp0:
+	bset_s r3,r3,23
+.Lpast_denorm_fp1:
+	cmp r2,r3
+	asl_s r2,r2,6+1
+	asl_s r3,r3,7
+	add.lo r2,r2,r2
+	bclr r8,r9,30 ; exponent bias
+	bclr.lo r8,r8,23 ; reduce exp by one if fraction is shifted
+	sub r4,r4,r5
+	add r4,r4,r8
+	xor.f 0,r10,r4
+	bmi .Linf_denorm
+	and r12,r4,r9
+	breq r12,0,.Ldenorm
+	sub_s r2,r2,r3 ; discard implicit 1
+.Ldiv_23bit:
+	.rep 6
+	divaw r2,r2,r3
+	.endr
+	breq r12,r9,.Linf
+	bmsk r0,r2,6
+	xor_s r2,r2,r0
+.Ldiv_17bit:
+	.rep 7
+	divaw r2,r2,r3
+	.endr
+	asl_s r0,r0,7
+	bmsk r1,r2,6
+	xor_s r2,r2,r1
+	or_s r0,r0,r1
+.Ldiv_10bit:
+	.rep 7
+	divaw r2,r2,r3
+	.endr
+	asl_s r0,r0,7
+	bmsk r1,r2,6
+	xor_s r2,r2,r1
+	or_s r0,r0,r1
+.Ldiv_3bit:
+	.rep 3
+	divaw r2,r2,r3
+	.endr
+	asl_s r0,r0,3
+.Ldiv_0bit:
+	divaw r1,r2,r3
+	bmsk_s r2,r2,2
+	tst r1,-0x7e ; 0xffffff82, test for rest or odd
+	bmsk_s r1,r1,0
+	add_s r0,r0,r2 ; assemble fraction
+	add_s r0,r0,r4 ; add in sign & exponent
+	j_s.d [blink]
+	add.ne r0,r0,r1 ; round to nearest / even
+
+	.balign 4
+.Linf_nan_fp0:
+	bic.f 0,r9,r1 ; fp1 Inf -> result NaN
+	bic r1,r5,r9  ; fp1 sign
+	sub.eq r1,r1,1
+	j_s.d [blink]
+	xor_s r0,r0,r1
+.Linf_nan_fp1:
+	bic r0,r4,r9 ; fp0 sign
+	bmsk.f 0,r1,22 ; x/inf -> 0, x/nan -> nan
+	xor.eq r1,r1,r9
+	j_s.d [blink]
+	xor_s r0,r0,r1
+.Linf:
+	j_s.d [blink]
+	or r0,r10,r9
+
+.Lret_r4:
+	j_s.d [blink]
+	mov_s r0,r4
+	.balign 4
+.Linf_denorm:
+	add.f r12,r4,r4
+	asr_l r12,r12,24
+	bpl .Linf
+	max r12,r12,-24
+.Ldenorm:
+	add r1,pcl,42; .Ldenorm_tab-.
+	ldb_s r12,[r12,r1]
+	mov_s r0,0
+	lsr_s r2,r2
+	sub_s r1,r1,r12
+	j_s.d [r1]
+	bic r4,r10,r9
+	.byte .Ldenorm_tab-.Lret_r4
+	.byte .Ldenorm_tab-.Ldiv_0bit
+	.byte .Ldenorm_tab-.Ldiv_3bit-8
+	.byte .Ldenorm_tab-.Ldiv_3bit-4
+	.byte .Ldenorm_tab-.Ldiv_3bit
+	.byte .Ldenorm_tab-.Ldiv_10bit-24
+	.byte .Ldenorm_tab-.Ldiv_10bit-20
+	.byte .Ldenorm_tab-.Ldiv_10bit-16
+	.byte .Ldenorm_tab-.Ldiv_10bit-12
+	.byte .Ldenorm_tab-.Ldiv_10bit-8
+	.byte .Ldenorm_tab-.Ldiv_10bit-4
+	.byte .Ldenorm_tab-.Ldiv_10bit
+	.byte .Ldenorm_tab-.Ldiv_17bit-24
+	.byte .Ldenorm_tab-.Ldiv_17bit-20
+	.byte .Ldenorm_tab-.Ldiv_17bit-16
+	.byte .Ldenorm_tab-.Ldiv_17bit-12
+	.byte .Ldenorm_tab-.Ldiv_17bit-8
+	.byte .Ldenorm_tab-.Ldiv_17bit-4
+	.byte .Ldenorm_tab-.Ldiv_17bit
+	.byte .Ldenorm_tab-.Ldiv_23bit-20
+	.byte .Ldenorm_tab-.Ldiv_23bit-16
+	.byte .Ldenorm_tab-.Ldiv_23bit-12
+	.byte .Ldenorm_tab-.Ldiv_23bit-8
+	.byte .Ldenorm_tab-.Ldiv_23bit-4
+.Ldenorm_tab:
+	.byte .Ldenorm_tab-.Ldiv_23bit
+	ENDFUNC(__divsf3)

libgcc/config/arc/ieee-754/divtab-arc-df.c

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/* We use a polynom similar to a Tchebycheff polynom to get an initial
+   seed, and then use a newton-raphson iteration step to get an
+   approximate result
+   If this result can't be rounded to the exact result with confidence, we
+   round to the value between the two closest representable values, and
+   test if the correctly rounded value is above or below this value.
+ 
+   Because of the Newton-raphson iteration step, an error in the seed at X
+   is amplified by X.  Therefore, we don't want a Tchebycheff polynom
+   or a polynom that is close to optimal according to the maximum norm
+   on the errro of the seed value; we want one that is close to optimal
+   according to the maximum norm on the error of the result, i.e. we
+   want the maxima of the polynom to increase linearily.
+   Given an interval [X0,X2) over which to approximate,
+   with X1 := (X0+X2)/2,  D := X1-X0, F := 1/D, and S := D/X1 we have,
+   like for Tchebycheff polynoms:
+   P(0) := 1
+   but then we have:
+   P(1) := X + S*D
+   P(2) := 2 * X^2 + S*D * X - D^2
+   Then again:
+   P(n+1) := 2 * X * P(n) - D^2 * P (n-1)
+ */
+
+static long double merr = 42.;
+
+double
+err (long double a0, long double a1, long double x)
+{
+  long double y0 = a0 + (x-1)*a1;
+
+  long double approx = 2. * y0 - y0 * x * y0;
+  long double true = 1./x;
+  long double err = approx - true;
+
+  if (err <= -1./65536./16384.)
+    printf ("ERROR EXCEEDS 1 ULP %.15f %.15f %.15f\n",
+	    (double)x, (double)approx, (double)true);
+  if (merr > err)
+    merr = err;
+  return err;
+}
+
+int
+main (void)
+{
+  long double T[5]; /* Taylor polynom */
+  long double P[5][5];
+  int i, j;
+  long double X0, X1, X2, S;
+  long double inc = 1./64;
+  long double D = inc*0.5;
+  long i0, i1, i2, io;
+
+  memset (P, 0, sizeof (P));
+  P[0][0] = 1.;
+  for (i = 1; i < 5; i++)
+    P[i][i] = 1 << i-1;
+  P[2][0] = -D*D;
+  for (X0 = 1.; X0 < 2.; X0 += inc)
+    {
+      X1 = X0 + inc * 0.5;
+      X2 = X0 + inc;
+      S = D / X1;
+      T[0] = 1./X1;
+      for (i = 1; i < 5; i++)
+	T[i] = T[i-1] * -T[0];
+#if 0
+      printf ("T %1.8f %f %f %f %f\n", (double)T[0], (double)T[1], (double)T[2],
+(double)T[3], (double)T[4]);
+#endif
+      P[1][0] = S*D;
+      P[2][1] = S*D;
+      for (i = 3; i < 5; i++)
+	{
+	  P[i][0] = -D*D*P[i-2][0];
+	  for (j = 1; j < i; j++)
+	    P[i][j] = 2*P[i-1][j-1]-D*D*P[i-2][j];
+	}
+#if 0
+      printf ("P3 %1.8f %f %f %f %f\n", (double)P[3][0], (double)P[3][1], (double)P[3][2],
+(double)P[3][3], (double)P[3][4]);
+      printf ("P4 %1.8f %f %f %f %f\n", (double)P[4][0], (double)P[4][1], (double)P[4][2],
+(double)P[4][3], (double)P[4][4]);
+#endif
+      for (i = 4; i > 1; i--)
+	{
+	  long double a = T[i]/P[i][i];
+
+	  for (j = 0; j < i; j++)
+	    T[j] -= a * P[i][j];
+	}
+#if 0
+      printf ("A %1.8f %f %f\n", (double)T[0], (double)T[1], (double)T[2]);
+#endif
+#if 0
+      i2 = T[2]*1024;
+      long double a = (T[2]-i/1024.)/P[2][2];
+      for (j = 0; j < 2; j++)
+	T[j] -= a * P[2][j];
+#else
+      i2 = 0;
+#endif
+	  long double T0, Ti1;
+      for (i = 0, i0 = 0; i < 4; i++)
+	{
+
+	  i1 = T[1]*4096. + i0 / (long double)(1 << 20) - 0.5;
+	  i1 = - (-i1 & 0x0fff);
+	  Ti1 = ((unsigned)(-i1 << 20) | i0) /-(long double)(1LL<<32LL);
+	  T0 = T[0] - (T[1]-Ti1)/P[1][1] * P[1][0] - (X1 - 1) * Ti1;
+	  i0 = T0 * 1024 * 1024 + 0.5;
+	  i0 &= 0xfffff;
+	}
+#if 0
+      printf ("A %1.8f %f %f\n", (double)T[0], (double)T[1], (double)T[2]);
+#endif
+      io = (unsigned)(-i1 << 20) | i0;
+      long double A1 = (unsigned)io/-65536./65536.;
+      long double A0 =  (unsigned)(io << 12)/65536./65536.;
+      long double Xm0 = 1./sqrt (-A1);
+      long double Xm1 = 0.5+0.5*-A0/A1;
+#if 0
+      printf ("%f %f %f %f\n", (double)A0, (double)A1, (double) Ti1, (double)X0);
+      printf ("%.12f %.12f %.12f\n",
+	      err (A0, A1, X0), err (A0, A1, X1), err (A0, A1, X2));
+      printf ("%.12f %.12f\n", (double)Xm0, (double)Xm1);
+      printf ("%.12f %.12f\n", err (A0, A1, Xm0), err (A0, A1, Xm1));
+#endif
+      printf ("\t.long 0x%x\n", io);
+   }
+#if 0
+  printf ("maximum error: %.15f %x %f\n", (double)merr, (unsigned)(long long)(-merr * 65536 * 65536), (double)log(-merr)/log(2));
+#endif
+  return 0;
+}

libgcc/config/arc/ieee-754/divtab-arc-sf.c

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+/* We use a polynom similar to a Tchebycheff polynom to get an initial
+   seed, and then use a newton-raphson iteration step to get an
+   approximate result
+   If this result can't be rounded to the exact result with confidence, we
+   round to the value between the two closest representable values, and
+   test if the correctly rounded value is above or below this value.
+ 
+   Because of the Newton-raphson iteration step, an error in the seed at X
+   is amplified by X.  Therefore, we don't want a Tchebycheff polynom
+   or a polynom that is close to optimal according to the maximum norm
+   on the errro of the seed value; we want one that is close to optimal
+   according to the maximum norm on the error of the result, i.e. we
+   want the maxima of the polynom to increase linearily.
+   Given an interval [X0,X2) over which to approximate,
+   with X1 := (X0+X2)/2,  D := X1-X0, F := 1/D, and S := D/X1 we have,
+   like for Tchebycheff polynoms:
+   P(0) := 1
+   but then we have:
+   P(1) := X + S*D
+   P(2) := 2 * X^2 + S*D * X - D^2
+   Then again:
+   P(n+1) := 2 * X * P(n) - D^2 * P (n-1)
+ */
+
+int
+main (void)
+{
+  long double T[5]; /* Taylor polynom */
+  long double P[5][5];
+  int i, j;
+  long double X0, X1, X2, S;
+  long double inc = 1./64;
+  long double D = inc*0.5;
+  long i0, i1, i2;
+
+  memset (P, 0, sizeof (P));
+  P[0][0] = 1.;
+  for (i = 1; i < 5; i++)
+    P[i][i] = 1 << i-1;
+  P[2][0] = -D*D;
+  for (X0 = 1.; X0 < 2.; X0 += inc)
+    {
+      X1 = X0 + inc * 0.5;
+      X2 = X1 + inc;
+      S = D / X1;
+      T[0] = 1./X1;
+      for (i = 1; i < 5; i++)
+	T[i] = T[i-1] * -T[0];
+#if 0
+      printf ("T %1.8f %f %f %f %f\n", (double)T[0], (double)T[1], (double)T[2],
+(double)T[3], (double)T[4]);
+#endif
+      P[1][0] = S*D;
+      P[2][1] = S*D;
+      for (i = 3; i < 5; i++)
+	{
+	  P[i][0] = -D*D*P[i-2][0];
+	  for (j = 1; j < i; j++)
+	    P[i][j] = 2*P[i-1][j-1]-D*D*P[i-2][j];
+	}
+#if 0
+      printf ("P3 %1.8f %f %f %f %f\n", (double)P[3][0], (double)P[3][1], (double)P[3][2],
+(double)P[3][3], (double)P[3][4]);
+      printf ("P4 %1.8f %f %f %f %f\n", (double)P[4][0], (double)P[4][1], (double)P[4][2],
+(double)P[4][3], (double)P[4][4]);
+#endif
+      for (i = 4; i > 1; i--)
+	{
+	  long double a = T[i]/P[i][i];
+
+	  for (j = 0; j < i; j++)
+	    T[j] -= a * P[i][j];
+	}
+#if 0
+      printf ("A %1.8f %f %f\n", (double)T[0], (double)T[1], (double)T[2]);
+#endif
+#if 0
+      i2 = T[2]*512;
+      long double a = (T[2]-i/512.)/P[2][2];
+      for (j = 0; j < 2; j++)
+	T[j] -= a * P[2][j];
+#else
+      i2 = 0;
+#endif
+      for (i = 0, i0 = 0; i < 4; i++)
+	{
+	  long double T0, Ti1;
+
+	  i1 = T[1]*8192. + i0 / (long double)(1 << 19) - 0.5;
+	  i1 = - (-i1 & 0x1fff);
+	  Ti1 = ((unsigned)(-i1 << 19) | i0) /-(long double)(1LL<<32LL);
+	  T0 = T[0] - (T[1]-Ti1)/P[1][1] * P[1][0] - (X1 - 1) * Ti1;
+	  i0 = T0 * 512 * 1024 + 0.5;
+	  i0 &= 0x7ffff;
+	}
+#if 0
+      printf ("A %1.8f %f %f\n", (double)T[0], (double)T[1], (double)T[2]);
+#endif
+      printf ("\t.long 0x%x\n", (-i1 << 19) | i0);
+   }
+  return 0;
+}

libgcc/config/arc/ieee-754/eqdf2.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+/* inputs: DBL0, DBL1
+   output: z flag
+   clobber: r12, flags
+   For NaNs, bit 19.. bit 30 of the high word must be set.  */
+#if 0 /* DEBUG */
+	.global __eqdf2
+	.balign 4
+	FUNC(__eqdf2)
+__eqdf2:
+	st.a r11,[sp,-4]` push_s blink` st.a r10,[sp,-4]` st.a r9,[sp,-4]
+	st.a r8,[sp,-4]` st.a r7,[sp,-4]` st.a r6,[sp,-4]` st.a r5,[sp,-4]
+	st.a r4,[sp,-4]` push_s r3` push_s r2` push_s r1`
+	bl.d __eqdf2_c` push_s r0
+	mov r11,r0` pop_s r0` pop_s r1` pop_s r2` pop_s r3
+	ld.ab r4,[sp,4]` ld.ab r5,[sp,4]` ld.ab r6,[sp,4]`
+	ld.ab r7,[sp,4]` ld.ab r8,[sp,4]` ld.ab r9,[sp,4]
+	bl.d __eqdf2_asm` ld.ab r10,[sp,4]
+	pop_s blink
+	breq.d r11,0,0f
+	ld.ab r11,[sp,4]
+	jne_s [blink]
+	bl abort
+0:	jeq_s [blink]
+	bl abort
+	ENDFUNC(__eqdf2)
+#define __eqdf2 __eqdf2_asm
+#endif /* DEBUG */
+	.global __eqdf2
+	.balign 4
+	HIDDEN_FUNC(__eqdf2)
+	/* Good performance as long as the difference in high word is
+	   well predictable (as seen from the branch predictor).  */
+__eqdf2:
+	brne.d DBL0H,DBL1H,.Lhighdiff
+	bmsk r12,DBL0H,20
+#ifdef DPFP_COMPAT
+	or.f 0,DBL0L,DBL1L
+	bset.ne r12,r12,21
+#endif /* DPFP_COMPAT */
+	add1.f	r12,r12,DBL0H /* set c iff NaN; also, clear z if NaN.  */
+	j_s.d	[blink]
+	cmp.cc	DBL0L,DBL1L
+	.balign 4
+.Lhighdiff:
+	or	r12,DBL0H,DBL1H
+	or.f	0,DBL0L,DBL1L
+	j_s.d	[blink]
+	bmsk.eq.f r12,r12,30
+	ENDFUNC(__eqdf2)
+/* ??? could we do better by speeding up some 'common' case of inequality?  */

libgcc/config/arc/ieee-754/eqsf2.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+/* inputs: r0, r1
+   output: z flag
+   clobber: r12, flags
+   For NaNs, bit 22 .. bit 30 must be set.  */
+#if 0 /* DEBUG */
+	.global __eqsf2
+	.balign 4
+	FUNC(__eqsf2)
+__eqsf2:
+	st.a r11,[sp,-4]` push_s blink` st.a r10,[sp,-4]` st.a r9,[sp,-4]
+	st.a r8,[sp,-4]` st.a r7,[sp,-4]` st.a r6,[sp,-4]` st.a r5,[sp,-4]
+	st.a r4,[sp,-4]` push_s r3` push_s r2` push_s r1`
+	bl.d __eqsf2_c` push_s r0
+	mov r11,r0` pop_s r0` pop_s r1` pop_s r2` pop_s r3
+	ld.ab r4,[sp,4]` ld.ab r5,[sp,4]` ld.ab r6,[sp,4]`
+	ld.ab r7,[sp,4]` ld.ab r8,[sp,4]` ld.ab r9,[sp,4]
+	bl.d __eqsf2_asm` ld.ab r10,[sp,4]
+	pop_s blink
+	breq.d r11,0,0f
+	ld.ab r11,[sp,4]
+	jne_s [blink]
+	bl abort
+0:	jeq_s [blink]
+	bl abort
+	ENDFUNC(__eqsf2)
+#define __eqsf2 __eqsf2_asm
+#endif /* DEBUG */
+	/* Good performance as long as the binary difference is
+	   well predictable (as seen from the branch predictor).  */
+	.global __eqsf2
+	.balign 4
+	HIDDEN_FUNC(__eqsf2)
+__eqsf2:
+	breq	r0, r1,.Lno_bdiff
+	or	r12,r0,r1
+	j_s.d	[blink]
+	bmsk.f	0,r12,30
+.Lno_bdiff:
+	bmsk r12,r0,23
+	add1.f	r12,r12,r0 /* set c iff NaN; also, clear z if NaN.  */
+	j_s.d	[blink]
+	cmp.cc	r0,r1
+	ENDFUNC(__eqsf2)

libgcc/config/arc/ieee-754/extendsfdf2.S

+/* Copyright (C) 2006, 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __extendsfdf2
+	.balign 4
+	FUNC(__extendsfdf2)
+__extendsfdf2:
+	push_s blink
+	bl.d __extendsfdf2_c
+	push_s r0
+	ld_s r2,[sp]
+	st_s r1,[sp]
+	push_s r0
+	bl.d __extendsfdf2_asm
+	mov_s r0,r2
+	pop_s r2
+	pop_s r3
+	pop_s blink
+	cmp r0,r2
+	cmp.eq r1,r3
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__extendsfdf2)
+#define __extendsfdf2 __extendsfdf2_asm
+#endif /* DEBUG */
+#if 0 /* ARC600 */
+__extendsfdf2:
+	lsr r2,r0,23
+	tst r2,0xff
+	bic.ne.f r2,0xff
+	beq_s .Linf_nan_denorm_0
+..
+.Linf_nan_denorm:
+	bbit1 r0,30,.Linf_nan
+#endif
+	.global __extendsfdf2
+	.balign 4
+	FUNC(__extendsfdf2)
+__extendsfdf2:
+	add.f r1,r0,r0
+	norm r3,r1
+#ifdef __LITTLE_ENDIAN__
+	lsr_s DBL0H,r1,4
+	brhs r3,7,.Linf_nan_denorm_0
+	asl_s DBL0L,r0,29
+	add_s DBL0H,DBL0H, \
+		0x38000000
+#else
+	lsr r2,r1,4
+	brhs r3,7,.Linf_nan_denorm_0
+	asl_s DBL0L,r1,28
+	add DBL0H,r2, \
+		0x38000000
+#endif
+	j_s.d [blink]
+	bxor.cs DBL0H,DBL0H,31
+	.balign 4
+.Linf_nan_denorm_0:
+#ifdef __LITTLE_ENDIAN__
+	mov_s DBL0H,r0
+	jeq.d [blink]
+	mov.eq DBL0L,0
+#else
+	jeq_s [blink]
+#endif
+	bmi .Linf_nan
+	asl_s r0,r0,r3
+	rsub r3,r3,0x380+6
+#ifdef __LITTLE_ENDIAN__
+	asl_s r3,r3,20
+	lsr DBL0H,r0,9
+	asl_s DBL0L,r0,23
+	add_s DBL0H,DBL0H,r3
+	j_s.d [blink]
+	bxor.cs DBL0H,DBL0H,31
+#else
+	asl DBL0L,r0,23
+	lsr_s DBL0H,r0,9
+	asl_s r3,r3,20
+	bxor.cs DBL0H,DBL0H,31
+	j_s.d [blink]
+	add_l DBL0H,DBL0H,r3
+#endif
+.Linf_nan:
+#ifdef __LITTLE_ENDIAN__
+	lsr DBL0H,r0,3
+
+	or_s DBL0H,DBL0H,r0
+	j_s.d [blink]
+	mov_l DBL0L,0
+#else
+	lsr r3,r0,3
+	mov_s DBL0L,0
+	j_s.d [blink]
+	or_l DBL0H,r0,r3
+#endif
+	ENDFUNC(__extendsfdf2)

libgcc/config/arc/ieee-754/fixdfsi.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	FUNC(__fixdfsi)
+	.global __fixdfsi
+	.balign 4
+__fixdfsi:
+	push_s blink
+	push_s r0
+	bl.d __fixdfsi_c
+	push_s r1
+	mov_s r2,r0
+	pop_s r1
+	ld r0,[sp]
+	bl.d __fixdfsi_asm
+	st r2,[sp]
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__fixdfsi)
+#define __fixdfsi __fixdfsi_asm
+#endif /* DEBUG */
+
+/*      If the fraction has to be shifted left by a positive non-zero amount,
+        we have to combine bits from DBL0L and DBL0H.  If we shift right,
+        or shift by zero, we only want to have the bits from DBL0H in r0.  */
+
+	.global __fixdfsi
+	FUNC(__fixdfsi)
+	.balign 4
+__fixdfsi:
+	bbit0 DBL0H,30,.Lret0or1
+	asr r2,DBL0H,20
+	bmsk_s DBL0H,DBL0H,19
+	sub_s r2,r2,19; 0x3ff+20-0x400
+	neg_s r3,r2
+	asr.f 0,r3,11
+	bset_s DBL0H,DBL0H,20
+#ifdef __LITTLE_ENDIAN__
+	mov.cs DBL0L,DBL0H
+	asl DBL0H,DBL0H,r2
+#else
+	asl.cc DBL0H,DBL0H,r2
+	lsr.cs DBL0H,DBL0H,r3
+#endif
+	lsr_s DBL0L,DBL0L,r3
+
+	add.cc r0,r0,r1
+	j_s.d [blink]
+	neg.pl r0,r0
+.Lret0or1:
+	add.f r0,DBL0H,0x100000
+	lsr_s r0,r0,30
+
+	bmsk_s r0,r0,0
+	j_s.d [blink]
+	neg.mi r0,r0
+	ENDFUNC(__fixdfsi)

libgcc/config/arc/ieee-754/fixsfsi.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __fixsfsi
+	FUNC(__fixsfsi)
+	.balign 4
+__fixsfsi:
+	push_s blink
+	bl.d __fixsfsi_c
+	push_s r0
+	ld_s r1,[sp]
+	st_s r0,[sp]
+	bl.d __fixsfsi_asm
+	mov_s r0,r1
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__fixsfsi)
+#define __fixsfsi __fixsfsi_asm
+#endif /* DEBUG */
+
+	.global __fixsfsi
+	FUNC(__fixsfsi)
+	.balign 4
+__fixsfsi:
+	bbit0 r0,30,.Lret0or1
+	lsr r2,r0,23
+	bmsk_s r0,r0,22
+	bset_s r0,r0,23
+	sub_s r2,r2,22;0x7f+23-0x80
+	asl.f 0,r2,24
+	neg r3,r2
+	asl.mi r0,r0,r2
+	lsr.pl r0,r0,r3
+	j_s.d [blink]
+	neg.cs r0,r0
+.Lret0or1:
+	add.f r0,r0,0x800000
+	lsr_s r0,r0,30
+
+	bmsk_s r0,r0,0
+	j_s.d [blink]
+	neg.mi r0,r0
+	ENDFUNC(__fixsfsi)

libgcc/config/arc/ieee-754/fixunsdfsi.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	FUNC(__fixunsdfsi)
+	.global __fixunsdfsi
+	.balign 4
+__fixunsdfsi:
+	push_s blink
+	push_s r0
+	bl.d __fixunsdfsi_c
+	push_s r1
+	mov_s r2,r0
+	pop_s r1
+	ld r0,[sp]
+	bl.d __fixunsdfsi_asm
+	st r2,[sp]
+	pop_s r1
+	pop_s blink
+	cmp r0,r1
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__fixunsdfsi)
+#define __fixunsdfsi __fixunsdfsi_asm
+#endif /* DEBUG */
+
+	.global __fixunsdfsi
+	FUNC(__fixunsdfsi)
+	.balign 4
+__fixunsdfsi:
+	bbit0 DBL0H,30,.Lret0or1
+	lsr r2,DBL0H,20
+	bmsk_s DBL0H,DBL0H,19
+	sub_s r2,r2,19; 0x3ff+20-0x400
+	neg_s r3,r2
+	btst_s r3,10
+	bset_s DBL0H,DBL0H,20
+#ifdef __LITTLE_ENDIAN__
+	mov.ne DBL0L,DBL0H
+	asl DBL0H,DBL0H,r2
+#else
+	asl.eq DBL0H,DBL0H,r2
+	lsr.ne DBL0H,DBL0H,r3
+#endif
+	lsr DBL0L,DBL0L,r3
+	j_s.d [blink]
+	add.eq r0,r0,r1
+.Lret0:
+	j_s.d [blink]
+	mov_l r0,0
+.Lret0or1:
+	add_s DBL0H,DBL0H,0x100000
+	lsr_s DBL0H,DBL0H,30
+	j_s.d [blink]
+	bmsk_l r0,DBL0H,0
+	ENDFUNC(__fixunsdfsi)

libgcc/config/arc/ieee-754/floatsidf.S

+/* Copyright (C) 2008-2012 Free Software Foundation, Inc.  
+   Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
+		on behalf of Synopsys 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/>.  */
+
+#include "arc-ieee-754.h"
+
+#if 0 /* DEBUG */
+	.global __floatsidf
+	.balign 4
+	FUNC(__floatsidf)
+__floatsidf:
+	push_s blink
+	bl.d __floatsidf_c
+	push_s r0
+	ld_s r2,[sp]
+	st_s r1,[sp]
+	push_s r0
+	bl.d __floatsidf_asm
+	mov_s r0,r2
+	pop_s r2
+	pop_s r3
+	pop_s blink
+	cmp r0,r2
+	cmp.eq r1,r3
+	jeq_s [blink]
+	bl abort
+	ENDFUNC(__floatsidf)
+#define __floatsidf __floatsidf_asm
+#endif /* DEBUG */
+
+	.global __floatsidf
+	.balign 4
+	FUNC(__floatsidf)
+__floatsidf:
+	abs.f r1,r0
+	jeq_s [blink]
+	lsr r2,r1
+	mov r12,-0x41d ; -(0x3ff+31-1)
+	norm r2,r2
+	bclr.cs r12,r12,11
+	rsub.f r3,r2,11
+	add_s r12,r2,r12
+	add_s r2,r2,21
+#ifdef __LITTLE_ENDIAN__
+	asl DBL0L,r1,r2
+	lsr_s DBL0H,r1,r3
+#else
+	lsr DBL0H,r1,r3
+	asl_s DBL0L,r1,r2
+#endif
+	asl_s r12,r12,20
+	mov.lo DBL0H,DBL0L
+	sub_s DBL0H,DBL0H,r12
+	j_s.d [blink]
+	mov.ls DBL0L,0
+	ENDFUNC(__floatsidf)