diff options
Diffstat (limited to 'sysdeps/sparc/sparc32')
36 files changed, 4740 insertions, 0 deletions
diff --git a/sysdeps/sparc/sparc32/Dist b/sysdeps/sparc/sparc32/Dist new file mode 100644 index 0000000000..6c3b23baa5 --- /dev/null +++ b/sysdeps/sparc/sparc32/Dist @@ -0,0 +1,4 @@ +dotmul.S umul.S +divrem.m4 sdiv.S udiv.S rem.S urem.S +alloca.S +sys/trap.h diff --git a/sysdeps/sparc/sparc32/Implies b/sysdeps/sparc/sparc32/Implies new file mode 100644 index 0000000000..39a34c5f57 --- /dev/null +++ b/sysdeps/sparc/sparc32/Implies @@ -0,0 +1 @@ +wordsize-32 diff --git a/sysdeps/sparc/sparc32/Makefile b/sysdeps/sparc/sparc32/Makefile new file mode 100644 index 0000000000..deec2f8145 --- /dev/null +++ b/sysdeps/sparc/sparc32/Makefile @@ -0,0 +1,51 @@ +# Copyright (C) 1991, 92, 93, 94, 95, 96, 97 Free Software Foundation, Inc. +# This file is part of the GNU C Library. + +# The GNU C Library is free software; you can redistribute it and/or +# modify it under the terms of the GNU Library General Public License +# as published by the Free Software Foundation; either version 2 of +# the License, or (at your option) any later version. + +# The GNU C Library 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 +# Library General Public License for more details. + +# You should have received a copy of the GNU Library General Public +# License along with the GNU C Library; see the file COPYING.LIB. If not, +# write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, +# Boston, MA 02111-1307, USA. + +ifeq ($(subdir),gnulib) +routines = dotmul umul $(divrem) alloca +endif # gnulib + +# We distribute these files, even though they are generated, +# so as to avoid the need for a functioning m4 to build the library. +divrem := sdiv udiv rem urem + ++divrem-NAME-sdiv := div ++divrem-NAME-udiv := udiv ++divrem-NAME-rem := rem ++divrem-NAME-urem := urem ++divrem-NAME = $(+divrem-NAME-$(basename $(notdir $@))) ++divrem-OP-div := div ++divrem-OP-udiv := div ++divrem-OP-rem := rem ++divrem-OP-urem := rem ++divrem-S-div := true ++divrem-S-rem := true ++divrem-S-udiv := false ++divrem-S-urem := false +$(divrem:%=$(sysdep_dir)/sparc/%.S): $(sysdep_dir)/sparc/divrem.m4 + (echo "define(NAME,\`.$(+divrem-NAME)')\ + define(OP,\`$(+divrem-OP-$(+divrem-NAME))')\ + define(S,\`$(+divrem-S-$(+divrem-NAME))')\ + /* This file is generated from divrem.m4; DO NOT EDIT! */"; \ + cat $<) | $(M4) > $@-tmp +# Make it unwritable so noone will edit it by mistake. + -chmod a-w $@-tmp + mv -f $@-tmp $@ + test ! -d CVS || cvs commit -m'Regenerated from $<' $@ + +sysdep-realclean := $(sysdep-realclean) $(divrem:%=sysdeps/sparc/%.S) diff --git a/sysdeps/sparc/sparc32/__longjmp.S b/sysdeps/sparc/sparc32/__longjmp.S new file mode 100644 index 0000000000..36e1c170dd --- /dev/null +++ b/sysdeps/sparc/sparc32/__longjmp.S @@ -0,0 +1,59 @@ +/* Copyright (C) 1991, 1993, 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <sysdep.h> + +#define _ASM 1 +#include <bits/setjmp.h> +#define ENV(reg) [%g1 + (reg * 4)] + +ENTRY (__longjmp) + /* Store our arguments in global registers so we can still + use them while unwinding frames and their register windows. */ + mov %o0, %g1 /* ENV in %g1 */ + orcc %o1, %g0, %g6 /* VAL in %g6 */ + be,a 0f /* Branch if zero; else skip delay slot. */ + mov 1, %g6 /* Delay slot only hit if zero: VAL = 1. */ +0: + + /* Cache target FP in register %g7. */ + ld ENV (JB_FP), %g7 + + /* Now we will loop, unwinding the register windows up the stack + until the restored %fp value matches the target value in %g7. */ + +loop: cmp %fp, %g7 /* Have we reached the target frame? */ + bl,a loop /* Loop while current fp is below target. */ + restore /* Unwind register window in delay slot. */ + be,a found /* Better have hit it exactly. */ + ld ENV (JB_SP), %o0 /* Delay slot: extract target SP. */ + +bogus: /* Get here only if the jmp_buf or stack is clobbered. */ + call C_SYMBOL_NAME (abort) + nop + unimp 0 + +found: /* We have unwound register windows so %fp matches the target. */ + cmp %o0, %sp /* Check jmp_buf SP vs register window. */ + bge,a sp_ok /* Saved must not be deeper than register. */ + mov %o0, %sp /* OK, install new SP. */ + b,a bogus /* Bogus, we lose. */ + +sp_ok: ld ENV (JB_PC), %o0 /* Extract target return PC. */ + jmp %o0 + 8 /* Return there. */ + mov %g6, %o0 /* Delay slot: set return value. */ diff --git a/sysdeps/sparc/sparc32/add_n.S b/sysdeps/sparc/sparc32/add_n.S new file mode 100644 index 0000000000..9852c256aa --- /dev/null +++ b/sysdeps/sparc/sparc32/add_n.S @@ -0,0 +1,226 @@ +! SPARC __mpn_add_n -- Add two limb vectors of the same length > 0 and store +! sum in a third limb vector. + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +#define res_ptr %o0 +#define s1_ptr %o1 +#define s2_ptr %o2 +#define size %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_add_n) +C_SYMBOL_NAME(__mpn_add_n): + xor s2_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L1 ! branch if alignment differs + nop +! ** V1a ** +L0: andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1 ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + addcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s1_ptr+0],%g4 + addcc size,-10,size + ld [s1_ptr+4],%g1 + ldd [s2_ptr+0],%g2 + blt Lfin1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1: addxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addxcc %g4,%g2,%o4 + ld [s1_ptr+16],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+20],%g1 + ldd [s2_ptr+16],%g2 + std %o4,[res_ptr+8] + addxcc %g4,%g2,%o4 + ld [s1_ptr+24],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+28],%g1 + ldd [s2_ptr+24],%g2 + std %o4,[res_ptr+16] + addxcc %g4,%g2,%o4 + ld [s1_ptr+32],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+36],%g1 + ldd [s2_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1 + subcc %g0,%o4,%g0 ! restore cy + +Lfin1: addcc size,8-2,size + blt Lend1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: addxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1 + subcc %g0,%o4,%g0 ! restore cy +Lend1: addxcc %g4,%g2,%o4 + addxcc %g1,%g3,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s1_ptr+8],%g4 + ld [s2_ptr+8],%g2 + addxcc %g4,%g2,%o4 + st %o4,[res_ptr+8] + +Lret1: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +L1: xor s1_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L2 + nop +! ** V1b ** + mov s2_ptr,%g1 + mov s1_ptr,s2_ptr + b L0 + mov %g1,s1_ptr + +! ** V2 ** +/* If we come here, the alignment of s1_ptr and res_ptr as well as the + alignment of s2_ptr and res_ptr differ. Since there are only two ways + things can be aligned (that we care about) we now know that the alignment + of s1_ptr and s2_ptr are the same. */ + +L2: cmp size,1 + be Ljone + nop + andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 + be L_v2 ! if no, branch + nop +/* Add least significant limb separately to align s1_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + addcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr + +L_v2: addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + blt Lfin2 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + ldd [s1_ptr+8],%g2 + ldd [s2_ptr+8],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+8] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+12] + ldd [s1_ptr+16],%g2 + ldd [s2_ptr+16],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+16] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+20] + ldd [s1_ptr+24],%g2 + ldd [s2_ptr+24],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+24] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+28] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop2 + subcc %g0,%o4,%g0 ! restore cy + +Lfin2: addcc size,8-2,size + blt Lend2 + subcc %g0,%o4,%g0 ! restore cy +Loope2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope2 + subcc %g0,%o4,%g0 ! restore cy +Lend2: andcc size,1,%g0 + be Lret2 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ +Ljone: ld [s1_ptr],%g4 + ld [s2_ptr],%g2 + addxcc %g4,%g2,%o4 + st %o4,[res_ptr] + +Lret2: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/sparc32/addmul_1.S b/sysdeps/sparc/sparc32/addmul_1.S new file mode 100644 index 0000000000..375d25db6b --- /dev/null +++ b/sysdeps/sparc/sparc32/addmul_1.S @@ -0,0 +1,147 @@ +! SPARC __mpn_addmul_1 -- Multiply a limb vector with a limb and add +! the result to a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_addmul_1) +C_SYMBOL_NAME(__mpn_addmul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + addcc %o5,%g1,%g1 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne Loop0 + ld [%o4+%o2],%o5 + + addcc %o5,%g1,%g1 + addx %o0,%g0,%o0 + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + addcc %o5,%g3,%g3 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 + addcc %o2,4,%o2 + bne Loop + ld [%o4+%o2],%o5 + + addcc %o5,%g3,%g3 + addx %o0,%g0,%o0 + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/alloca.S b/sysdeps/sparc/sparc32/alloca.S new file mode 100644 index 0000000000..dcbd171163 --- /dev/null +++ b/sysdeps/sparc/sparc32/alloca.S @@ -0,0 +1,32 @@ +/* Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "sysdep.h" + +/* Code produced by Sun's C compiler calls this function with two extra + arguments which it makes relocatable symbols but seem always to be + the constant 96; I have no idea what they are for. */ + +#ifndef NO_UNDERSCORES +#define __builtin_alloca ___builtin_alloca +#endif + +ENTRY (__builtin_alloca) + sub %sp, %o0, %sp /* Push some stack space. */ + retl /* Return; the returned buffer leaves 96 */ + add %sp, 96, %o0 /* bytes of register save area at the top. */ diff --git a/sysdeps/sparc/sparc32/bits/endian.h b/sysdeps/sparc/sparc32/bits/endian.h new file mode 100644 index 0000000000..f1a75c0652 --- /dev/null +++ b/sysdeps/sparc/sparc32/bits/endian.h @@ -0,0 +1,3 @@ +/* SPARC is big-endian. */ + +#define __BYTE_ORDER __BIG_ENDIAN diff --git a/sysdeps/sparc/sparc32/bits/setjmp.h b/sysdeps/sparc/sparc32/bits/setjmp.h new file mode 100644 index 0000000000..43bae1a496 --- /dev/null +++ b/sysdeps/sparc/sparc32/bits/setjmp.h @@ -0,0 +1,16 @@ +/* Define the machine-dependent type `jmp_buf'. SPARC version. */ + +#if defined (__USE_MISC) || defined (_ASM) +#define JB_SP 0 +#define JB_FP 1 +#define JB_PC 2 +#endif + +#ifndef _ASM +typedef int __jmp_buf[3]; +#endif + +/* Test if longjmp to JMPBUF would unwind the frame + containing a local variable at ADDRESS. */ +#define _JMPBUF_UNWINDS(jmpbuf, address) \ + ((int) (address) < (jmpbuf)[JB_SP]) diff --git a/sysdeps/sparc/sparc32/bsd-_setjmp.S b/sysdeps/sparc/sparc32/bsd-_setjmp.S new file mode 100644 index 0000000000..bf49765f4d --- /dev/null +++ b/sysdeps/sparc/sparc32/bsd-_setjmp.S @@ -0,0 +1,40 @@ +/* BSD `_setjmp' entry point to `sigsetjmp (..., 0)'. Sparc version. + Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <sysdep.h> + +ENTRY (_setjmp) + +#ifdef PIC + save %sp, -64, %sp +1: call 2f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 +2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 + add %g1, %o7, %g1 + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 + restore + or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 + ld [%g1+%g2], %g1 +#else + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 + or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 +#endif + + jmp %g1 + mov %g0, %o1 /* Pass second argument of zero. */ diff --git a/sysdeps/sparc/sparc32/bsd-setjmp.S b/sysdeps/sparc/sparc32/bsd-setjmp.S new file mode 100644 index 0000000000..16dc260765 --- /dev/null +++ b/sysdeps/sparc/sparc32/bsd-setjmp.S @@ -0,0 +1,40 @@ +/* BSD `setjmp' entry point to `sigsetjmp (..., 1)'. Sparc version. + Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <sysdep.h> + +ENTRY (setjmp) + +#ifdef PIC + save %sp, -64, %sp +1: call 2f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 +2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 + add %g1, %o7, %g1 + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 + restore + or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 + ld [%g1+%g2], %g1 +#else + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 + or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 +#endif + + jmp %g1 + mov 1, %o1 /* Pass second argument of one. */ diff --git a/sysdeps/sparc/sparc32/divrem.m4 b/sysdeps/sparc/sparc32/divrem.m4 new file mode 100644 index 0000000000..665abf11ae --- /dev/null +++ b/sysdeps/sparc/sparc32/divrem.m4 @@ -0,0 +1,238 @@ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * NAME name of function to generate + * OP OP=div => %o0 / %o1; OP=rem => %o0 % %o1 + * S S=true => signed; S=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top `decade' of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + +define(N, `4')dnl +define(WORDSIZE, `32')dnl +define(TOPBITS, eval(WORDSIZE - N*((WORDSIZE-1)/N)))dnl +dnl +define(dividend, `%o0')dnl +define(divisor, `%o1')dnl +define(Q, `%o2')dnl +define(R, `%o3')dnl +define(ITER, `%o4')dnl +define(V, `%o5')dnl +dnl +dnl m4 reminder: ifelse(a,b,c,d) => if a is b, then c, else d +define(T, `%g1')dnl +define(SC, `%g7')dnl +ifelse(S, `true', `define(SIGN, `%g6')')dnl + +dnl +dnl This is the recursive definition for developing quotient digits. +dnl +dnl Parameters: +dnl $1 the current depth, 1 <= $1 <= N +dnl $2 the current accumulation of quotient bits +dnl N max depth +dnl +dnl We add a new bit to $2 and either recurse or insert the bits in +dnl the quotient. R, Q, and V are inputs and outputs as defined above; +dnl the condition codes are expected to reflect the input R, and are +dnl modified to reflect the output R. +dnl +define(DEVELOP_QUOTIENT_BITS, +` ! depth $1, accumulated bits $2 + bl L.$1.eval(2**N+$2) + srl V,1,V + ! remainder is positive + subcc R,V,R + ifelse($1, N, + ` b 9f + add Q, ($2*2+1), Q + ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2+1)')') +L.$1.eval(2**N+$2): + ! remainder is negative + addcc R,V,R + ifelse($1, N, + ` b 9f + add Q, ($2*2-1), Q + ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2-1)')') + ifelse($1, 1, `9:')')dnl + +#include "sysdep.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include <machine/trap.h> +#endif +#endif + +FUNC(NAME) +ifelse(S, `true', +` ! compute sign of result; if neither is negative, no problem + orcc divisor, dividend, %g0 ! either negative? + bge 2f ! no, go do the divide +ifelse(OP, `div', +` xor divisor, dividend, SIGN ! compute sign in any case', +` mov dividend, SIGN ! sign of remainder matches dividend') + tst divisor + bge 1f + tst dividend + ! divisor is definitely negative; dividend might also be negative + bge 2f ! if dividend not negative... + sub %g0, divisor, divisor ! in any case, make divisor nonneg +1: ! dividend is negative, divisor is nonnegative + sub %g0, dividend, dividend ! make dividend nonnegative +2: +') + ! Ready to divide. Compute size of quotient; scale comparand. + orcc divisor, %g0, V + bne 1f + mov dividend, R + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp R, V ! if divisor exceeds dividend, done + blu Lgot_result ! (and algorithm fails otherwise) + clr Q + sethi %hi(1 << (WORDSIZE - TOPBITS - 1)), T + cmp R, T + blu Lnot_really_big + clr ITER + + ! `Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R.' + 1: + cmp V, T + bgeu 3f + mov 1, SC + sll V, N, V + b 1b + add ITER, 1, ITER + + ! Now compute SC. + 2: addcc V, V, V + bcc Lnot_too_big + add SC, 1, SC + + ! We get here if the divisor overflowed while shifting. + ! This means that R has the high-order bit set. + ! Restore V and subtract from R. + sll T, TOPBITS, T ! high order bit + srl V, 1, V ! rest of V + add V, T, V + b Ldo_single_div + sub SC, 1, SC + + Lnot_too_big: + 3: cmp V, R + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! V > R: went too far: back up 1 step + ! srl V, 1, V + ! dec SC + ! do single-bit divide steps + ! + ! We have to be careful here. We know that R >= V, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if R >= 0. Because both R and V may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc SC, 1, SC + bl Lend_regular_divide + nop + sub R, V, R + mov 1, Q + b Lend_single_divloop + nop + Lsingle_divloop: + sll Q, 1, Q + bl 1f + srl V, 1, V + ! R >= 0 + sub R, V, R + b 2f + add Q, 1, Q + 1: ! R < 0 + add R, V, R + sub Q, 1, Q + 2: + Lend_single_divloop: + subcc SC, 1, SC + bge Lsingle_divloop + tst R + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll V, N, V + cmp V, R + bleu 1b + addcc ITER, 1, ITER + be Lgot_result + sub ITER, 1, ITER + + tst R ! set up for initial iteration +Ldivloop: + sll Q, N, Q + DEVELOP_QUOTIENT_BITS(1, 0) +Lend_regular_divide: + subcc ITER, 1, ITER + bge Ldivloop + tst R + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) +ifelse(OP, `div', +` sub Q, 1, Q +', ` add R, divisor, R +') + +Lgot_result: +ifelse(S, `true', +` ! check to see if answer should be < 0 + tst SIGN + bl,a 1f + ifelse(OP, `div', `sub %g0, Q, Q', `sub %g0, R, R') +1:') + retl + ifelse(OP, `div', `mov Q, %o0', `mov R, %o0') diff --git a/sysdeps/sparc/sparc32/dl-machine.h b/sysdeps/sparc/sparc32/dl-machine.h new file mode 100644 index 0000000000..f7bcba0c2f --- /dev/null +++ b/sysdeps/sparc/sparc32/dl-machine.h @@ -0,0 +1,363 @@ +/* Machine-dependent ELF dynamic relocation inline functions. SPARC version. + Copyright (C) 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, write to the Free Software Foundation, Inc., + 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + +#define ELF_MACHINE_NAME "sparc" + +#include <assert.h> +#include <string.h> +#include <link.h> +#include <sys/param.h> + + +/* Some SPARC opcodes we need to use for self-modifying code. */ +#define OPCODE_NOP 0x01000000 /* nop */ +#define OPCODE_CALL 0x40000000 /* call ?; add PC-rel word address */ +#define OPCODE_SETHI_G1 0x03000000 /* sethi ?, %g1; add value>>10 */ +#define OPCODE_JMP_G1 0x81c06000 /* jmp %g1+?; add lo 10 bits of value */ +#define OPCODE_SAVE_SP64 0x9de3bfc0 /* save %sp, -64, %sp */ + + +/* Return nonzero iff E_MACHINE is compatible with the running host. */ +static inline int +elf_machine_matches_host (Elf32_Half e_machine) +{ + return e_machine == EM_SPARC; +} + + +/* Return the link-time address of _DYNAMIC. Conveniently, this is the + first element of the GOT. This must be inlined in a function which + uses global data. */ +static inline Elf32_Addr +elf_machine_dynamic (void) +{ + register Elf32_Addr *got asm ("%l7"); + return *got; +} + +/* Return the run-time load address of the shared object. */ +static inline Elf32_Addr +elf_machine_load_address (void) +{ + register Elf32_Addr pc __asm("%o7"), got; + + /* Utilize the fact that a local .got entry will be partially + initialized at startup awaiting its RELATIVE fixup. */ + + __asm("sethi %%hi(.Load_address),%1\n" + ".Load_address:\n\t" + "call 1f\n\t" + "or %1,%%lo(.Load_address),%1\n" + "1:\tld [%%l7+%1],%1" + : "=r"(pc), "=r"(got)); + + return pc - got; +} + + Elf32_Addr addr; + + asm ( + "add %%fp,0x44,%%o2\n\t" /* o2 = point to argc */ + "ld [%%o2 - 4],%%o0\n\t" /* o0 = load argc */ + "sll %%o0, 2, %%o0\n\t" /* o0 = argc * sizeof (int) */ + "add %%o2,%%o0,%%o2\n\t" /* o2 = skip over argv */ + "add %%o2,4,%%o2\n\t" /* skip over null after argv */ + + /* Now %o2 is pointing to env, skip over that as well. */ + "1:\n\t" + "ld [%%o2],%%o0\n\t" + "cmp %%o0,0\n\t" + "bnz 1b\n\t" + "add %%o2,4,%%o2\n\t" + + /* Note that above, we want to advance the NULL after envp so + we always add 4. */ + + /* Now, search for the AT_BASE property. */ + "2:\n\t" + "ld [%%o2],%%o0\n\t" + "cmp %%o0,0\n\t" + "be,a 3f\n\t" + "or %%g0,%%g0,%0\n\t" + "cmp %%o0,7\n\t" /* AT_BASE = 7 */ + "be,a 3f\n\t" + "ld [%%o2+4],%0\n\t" + "b 2b\n\t" + "add %%o2,8,%%o2\n\t" + /* At this point %0 has the load address for the interpreter */ + "3:\n\t" + : "=r" (addr) + : /* no inputs */ + : "o0", "o2"); + return addr; +} + +#ifdef RESOLVE +/* Perform the relocation specified by RELOC and SYM (which is fully resolved). + MAP is the object containing the reloc. */ + +static inline void +elf_machine_rela (struct link_map *map, const Elf32_Rela *reloc, + const Elf32_Sym *sym, const struct r_found_version *version, + Elf32_Addr *const reloc_addr) +{ + Elf32_Addr loadbase; + + if (ELF32_R_TYPE (reloc->r_info) == R_SPARC_RELATIVE) + { +#ifndef RTLD_BOOTSTRAP + if (map != &_dl_rtld_map) /* Already done in rtld itself. */ +#endif + *reloc_addr += map->l_addr + reloc->r_addend; + } + else + { + const Elf32_Sym *const refsym = sym; + Elf32_Addr value; + if (sym->st_shndx != SHN_UNDEF && + ELF32_ST_BIND (sym->st_info) == STB_LOCAL) + value = map->l_addr; + else + { + value = RESOLVE (&sym, version, ELF32_R_TYPE (reloc->r_info)); + if (sym) + value += sym->st_value; + } + value += reloc->r_addend; /* Assume copy relocs have zero addend. */ + + switch (ELF32_R_TYPE (reloc->r_info)) + { + case R_SPARC_COPY: + if (sym->st_size > refsym->st_size + || (_dl_verbose && sym->st_size < refsym->st_size)) + { + extern char **_dl_argv; + const char *strtab; + + strtab = ((void *) map->l_addr + + map->l_info[DT_STRTAB]->d_un.d_ptr); + _dl_sysdep_error (_dl_argv[0] ?: "<program name unknown>", + ": Symbol `", strtab + refsym->st_name, + "' has different size in shared object, " + "consider re-linking\n", NULL); + } + memcpy (reloc_addr, (void *) value, MIN (sym->st_size, + refsym->st_size)); + break; + case R_SPARC_GLOB_DAT: + case R_SPARC_32: + *reloc_addr = value; + break; + case R_SPARC_JMP_SLOT: + reloc_addr[1] = OPCODE_SETHI_G1 | (value >> 10); + reloc_addr[2] = OPCODE_JMP_G1 | (value & 0x3ff); + break; + case R_SPARC_8: + *(char *) reloc_addr = value; + break; + case R_SPARC_16: + *(short *) reloc_addr = value; + break; + case R_SPARC_DISP8: + *(char *) reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_DISP16: + *(short *) reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_DISP32: + *reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_LO10: + *reloc_addr = (*reloc_addr & ~0x3ff) | (value & 0x3ff); + break; + case R_SPARC_WDISP30: + *reloc_addr = ((*reloc_addr & 0xc0000000) + | ((value - (unsigned int) reloc_addr) >> 2)); + break; + case R_SPARC_HI22: + *reloc_addr = (*reloc_addr & 0xffc00000) | (value >> 10); + break; + case R_SPARC_NONE: /* Alright, Wilbur. */ + break; + default: + assert (! "unexpected dynamic reloc type"); + break; + } + } +} + +static inline void +elf_machine_lazy_rel (struct link_map *map, const Elf32_Rela *reloc) +{ + switch (ELF32_R_TYPE (reloc->r_info)) + { + case R_SPARC_NONE: + break; + case R_SPARC_JMP_SLOT: + break; + default: + assert (! "unexpected PLT reloc type"); + break; + } +} + +#endif /* RESOLVE */ + +/* Nonzero iff TYPE should not be allowed to resolve to one of + the main executable's symbols, as for a COPY reloc. */ +#define elf_machine_lookup_noexec_p(type) ((type) == R_SPARC_COPY) + +/* Nonzero iff TYPE describes relocation of a PLT entry, so + PLT entries should not be allowed to define the value. */ +#define elf_machine_lookup_noplt_p(type) ((type) == R_SPARC_JMP_SLOT) + +/* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */ +#define ELF_MACHINE_RELOC_NOPLT R_SPARC_JMP_SLOT + +/* The SPARC never uses Elf32_Rel relocations. */ +#define ELF_MACHINE_NO_REL 1 + +/* The SPARC overlaps DT_RELA and DT_PLTREL. */ +#define ELF_MACHINE_PLTREL_OVERLAP 1 + +/* Set up the loaded object described by L so its unrelocated PLT + entries will jump to the on-demand fixup code in dl-runtime.c. */ + +static inline int +elf_machine_runtime_setup (struct link_map *l, int lazy, int profile) +{ + Elf32_Addr *plt; + extern void _dl_runtime_resolve (Elf32_Word); + + if (l->l_info[DT_JMPREL] && lazy) + { + /* The entries for functions in the PLT have not yet been filled in. + Their initial contents will arrange when called to set the high 22 + bits of %g1 with an offset into the .rela.plt section and jump to + the beginning of the PLT. */ + plt = (Elf32_Addr *) (l->l_addr + l->l_info[DT_PLTGOT]->d_un.d_ptr); + + /* The beginning of the PLT does: + + save %sp, -64, %sp + pltpc: call _dl_runtime_resolve + nop + .word MAP + + This saves the register window containing the arguments, and the + PC value (pltpc) implicitly saved in %o7 by the call points near the + location where we store the link_map pointer for this object. */ + + plt[0] = OPCODE_SAVE_SP64; /* save %sp, -64, %sp */ + /* Construct PC-relative word address. */ + plt[1] = OPCODE_CALL | (((Elf32_Addr) &_dl_runtime_resolve - + (Elf32_Addr) &plt[1]) >> 2); + plt[2] = OPCODE_NOP; /* Fill call delay slot. */ + plt[3] = (Elf32_Addr *) l; + } + + return lazy; +} + +/* This code is used in dl-runtime.c to call the `fixup' function + and then redirect to the address it returns. */ +#define ELF_MACHINE_RUNTIME_TRAMPOLINE asm ("\ +# Trampoline for _dl_runtime_resolver + .globl _dl_runtime_resolve + .type _dl_runtime_resolve, @function +_dl_runtime_resolve: + t 1 + #call %g0 + # Pass two args to fixup: the PLT address computed from the PC saved + # in the PLT's call insn, and the reloc offset passed in %g1. + #ld [%o7 + 8], %o1 | Second arg, loaded from PLTPC[2]. + #call fixup + #shrl %g1, 22, %o0 | First arg, set in delay slot of call. + # Jump to the real function. + #jmpl %o0, %g0 + # In the delay slot of that jump, restore the register window + # saved by the first insn of the PLT. + #restore + .size _dl_runtime_resolve, . - _dl_runtime_resolve +"); + +/* The PLT uses Elf32_Rela relocs. */ +#define elf_machine_relplt elf_machine_rela + + +/* Mask identifying addresses reserved for the user program, + where the dynamic linker should not map anything. */ +#define ELF_MACHINE_USER_ADDRESS_MASK ??? + +/* Initial entry point code for the dynamic linker. + The C function `_dl_start' is the real entry point; + its return value is the user program's entry point. */ + +#define RTLD_START __asm__ ( \ +".text\n\ + .globl _start\n\ + .type _start,@function\n\ +_start:\n\ + /* Pass pointer to argument block to _dl_start. */\n\ + add %sp,64,%o0\n\ + call _dl_start\n\ + nop\n\ + \n\ + mov %o0,%l0\n\ + \n\ +2:\n\ + call 1f\n\ + nop\n\ +1:\n\ + sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ + sethi %hi(_dl_default_scope),%l3\n\ + or %l2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ + or %l3,%lo(_dl_default_scope),%l3\n\ + add %o7,%l2,%l1\n\ + # %l1 has the GOT. %l3 has _dl_default_scope GOT offset\n\ + ld [%l1+%l3],%l4\n\ + # %l4 has pointer to _dl_default_scope. Now, load _dl_default_scope [2]\n\ + ld [%l4+8],%l4\n\ + # %l4 has _dl_default_scope [2]\n\ + # call _dl_init_next until it returns 0, pass _dl_default_scope [2]\n\ +3:\n\ + call _dl_init_next\n\ + mov %l4,%o0\n\ + cmp %o0,%g0\n\ + bz,a 4f\n\ + nop\n\ + call %o0\n\ + /* Pass pointer to argument block to this init function */\n\ + add %sp,64,%o0\n\ + b,a 3b\n\ +4:\n\ + # Clear the _dl_starting_up variable and pass _dl_fini in %g1 as per ELF ABI.\n\ + sethi %hi(_dl_starting_up),%l4\n\ + sethi %hi(_dl_fini),%l3\n\ + or %l4,%lo(_dl_starting_up),%l4\n\ + or %l3,%lo(_dl_fini),%l3\n\ + # clear _dl_starting_up\n\ + ld [%l1+%l4],%l5\n\ + st %g0,[%l5]\n\ + # load out fini function for atexit in %g1\n\ + ld [%l3+%l1],%g1\n\ + # jump to the user program entry point.\n\ + jmpl %l0,%g0\n\ + nop\n\ +"); diff --git a/sysdeps/sparc/sparc32/dotmul.S b/sysdeps/sparc/sparc32/dotmul.S new file mode 100644 index 0000000000..7ce695cb7a --- /dev/null +++ b/sysdeps/sparc/sparc32/dotmul.S @@ -0,0 +1,123 @@ +/* + * Signed multiply, from Appendix E of the Sparc Version 8 + * Architecture Manual. + */ + +/* + * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of + * the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. + */ + +#include "sysdep.h" +ENTRY(.mul) + mov %o0, %y ! multiplier -> Y + andncc %o0, 0xfff, %g0 ! test bits 12..31 + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + ! If %o0 was negative, the result is + ! (%o0 * %o1) + (%o1 << 32)) + ! We fix that here. + +#if 0 + tst %o0 + bge 1f + rd %y, %o0 + + ! %o0 was indeed negative; fix upper 32 bits of result by subtracting + ! %o1 (i.e., return %o4 - %o1 in %o1). + retl + sub %o4, %o1, %o1 + +1: + retl + mov %o4, %o1 +#else + /* Faster code adapted from tege@sics.se's code for umul.S. */ + sra %o0, 31, %o2 ! make mask from sign bit + and %o1, %o2, %o2 ! %o2 = 0 or %o1, depending on sign of %o0 + rd %y, %o0 ! get lower half of product + retl + sub %o4, %o2, %o1 ! subtract compensation + ! and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the low part of the + * result; %y has the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * --hi-- ----low-part---- + * + * The upper 12 bits of %o4 should be sign-extended to form the + * high part of the product (i.e., highpart = %o4 >> 20). + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20, zero fill at left + or %o5, %o0, %o0 ! construct low part of result + retl + sra %o4, 20, %o1 ! ... and extract high part of result diff --git a/sysdeps/sparc/sparc32/e_sqrt.c b/sysdeps/sparc/sparc32/e_sqrt.c new file mode 100644 index 0000000000..a98ae21ace --- /dev/null +++ b/sysdeps/sparc/sparc32/e_sqrt.c @@ -0,0 +1,34 @@ +/* Copyright (C) 1991, 1992, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <errno.h> +#include <math.h> + +#ifndef __GNUC__ + #error This file uses GNU C extensions; you must compile with GCC. +#endif + +/* Return the square root of X. */ +double +__ieee754_sqrt (x) + double x; +{ + register double result; + asm ("fsqrtd %1, %0" : "=f" (result) : "f" (x)); + return result; +} diff --git a/sysdeps/sparc/sparc32/elf/Makefile b/sysdeps/sparc/sparc32/elf/Makefile new file mode 100644 index 0000000000..319fbdef10 --- /dev/null +++ b/sysdeps/sparc/sparc32/elf/Makefile @@ -0,0 +1,4 @@ +# Sparc/ELF specific definitions. + +# The assembler on SPARC needs the -fPIC flag even when it's assembler code. +ASFLAGS-.so = -fPIC diff --git a/sysdeps/sparc/sparc32/elf/start.c b/sysdeps/sparc/sparc32/elf/start.c new file mode 100644 index 0000000000..f1e80195ca --- /dev/null +++ b/sysdeps/sparc/sparc32/elf/start.c @@ -0,0 +1,68 @@ +/* Copyright (C) 1991, 1992, 1993, 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <errno.h> +#include <stdlib.h> +#include <unistd.h> +#include <fcntl.h> + +extern char **__environ; + +extern void __libc_init_first __P ((int argc, char **argv, char **envp)); +extern int main __P ((int argc, char **argv, char **envp)); + +register long int sp asm("%sp"), fp asm("%fp"); + +void +_start (void) +{ + /* It is important that these be declared `register'. + Otherwise, when compiled without optimization, they are put on the + stack, which loses completely after we zero the FP. */ + register int argc; + register char **argv, **envp; + register long int g1 asm ("%g1"); + unsigned long int copy_g1 = g1; + + /* Unwind the frame built when we entered the function. */ + asm("restore"); + if (copy_g1) + atexit (copy_g1); + + /* And clear the frame pointer. */ + fp = 0; + + /* The argument info starts after one register + window (64 bytes) past the SP. */ + argc = ((int *) sp)[16]; + argv = (char **) &((int *) sp)[17]; + envp = &argv[argc + 1]; + __environ = envp; + + /* Allocate 24 bytes of stack space for the register save area. */ + sp -= 24; + __libc_init_first (argc, argv, envp); +#ifdef ELF_INIT_FINI + { + extern void _fini (void); + _init (); + atexit (_fini); + } +#endif + exit (main (argc, argv, envp)); +} diff --git a/sysdeps/sparc/sparc32/fpu/bits/fenv.h b/sysdeps/sparc/sparc32/fpu/bits/fenv.h new file mode 100644 index 0000000000..0560dd99b8 --- /dev/null +++ b/sysdeps/sparc/sparc32/fpu/bits/fenv.h @@ -0,0 +1,76 @@ +/* Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FENV_H +#error "Never use <bits/fenv.h> directly; include <fenv.h> instead." +#endif + + +/* Define bits representing the exception. We use the bit positions + of the appropriate accrued exception bits from the FSR. */ +enum + { + FE_INVALID = (1 << 9), +#define FE_INVALID FE_INVALID + FE_OVERFLOW = (1 << 8), +#define FE_OVERFLOW FE_OVERFLOW + FE_UNDERFLOW = (1 << 7), +#define FE_UNDERFLOW FE_UNDERFLOW + FE_DIVBYZERO = (1 << 6), +#define FE_DIVBYZERO FE_DIVBYZERO + FE_INEXACT = (1 << 5) +#define FE_INEXACT FE_INEXACT + }; + +#define FE_ALL_EXCEPT \ + (FE_INEXACT | FE_DIVBYZERO | FE_UNDERFLOW | FE_OVERFLOW | FE_INVALID) + +/* The Sparc FPU supports all of the four defined rounding modes. We + use again the bit positions in the FPU control word as the values + for the appropriate macros. */ +enum + { + FE_TONEAREST = (0U << 30), +#define FE_TONEAREST FE_TONEAREST + FE_TOWARDSZERO = (1U << 30), +#define FE_TOWARDSZERO FE_TOWARDSZERO + FE_UPWARD = (2U << 30), +#define FE_UPWARD FE_UPWARD + FE_DOWNWARD = (3U << 30) +#define FE_DOWNWARD FE_DOWNWARD + }; + +#define __FE_ROUND_MASK (3U << 30) + +/* Type representing exception flags. */ +typedef unsigned int fexcept_t; + +/* Type representing floating-point environment. */ +typedef unsigned int fenv_t; + +/* If the default argument is used we use this value. */ +#define FE_DFL_ENV ((fenv_t *) -1) + +#ifdef __USE_GNU +/* Floating-point environment where none of the exception is masked. */ +# define FE_NOMASK_ENV ((fenv_t *) -2) +#endif + +/* For internal use only: access the fp state register. */ +#define __fenv_stfsr(X) __asm__("stfsr %0" : "=m"(X)) +#define __fenv_ldfsr(X) __asm__ __volatile__("ldfsr %0" : : "m"(X)) diff --git a/sysdeps/sparc/sparc32/fpu/fpu_control.h b/sysdeps/sparc/sparc32/fpu/fpu_control.h new file mode 100644 index 0000000000..5ef3824702 --- /dev/null +++ b/sysdeps/sparc/sparc32/fpu/fpu_control.h @@ -0,0 +1,69 @@ +/* FPU control word bits. SPARC version. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Miguel de Icaza + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FPU_CONTROL_H +#define _FPU_CONTROL_H 1 + + +#include <features.h> + +/* precision control */ +#define _FPU_EXTENDED 0x00000000 /* RECOMMENDED */ +#define _FPU_DOUBLE 0x20000000 +#define _FPU_80BIT 0x30000000 +#define _FPU_SINGLE 0x10000000 /* DO NOT USE */ + +/* rounding control / Sparc */ +#define _FPU_RC_DOWN 0xc0000000 +#define _FPU_RC_UP 0x80000000 +#define _FPU_RC_ZERO 0x40000000 +#define _FPU_RC_NEAREST 0x0 /* RECOMMENDED */ + +#define _FPU_RESERVED 0x30300000 /* Reserved bits in cw */ + + +/* Now two recommended cw */ + +/* Linux default: + - extended precision + - rounding to nearest + - exceptions on overflow, zero divide and NaN */ +#define _FPU_DEFAULT 0x1e + +/* IEEE: same as above, but exceptions */ +#define _FPU_IEEE 0x0 + +/* Type of the control word. */ +typedef unsigned int fpu_control_t; + +#define _FPU_GETCW(cw) __asm__ ("st %%fsr,%0" : "=m" (*&cw)) +#define _FPU_SETCW(cw) __asm__ ("ld %0,%%fsr" : : "m" (*&cw)) + +/* Default control word set at startup. */ +extern fpu_control_t __fpu_control; + +__BEGIN_DECLS + +/* Called at startup. It can be used to manipulate fpu control register. */ +extern void __setfpucw __P ((fpu_control_t)); + +__END_DECLS + +#endif /* fpu_control.h */ diff --git a/sysdeps/sparc/sparc32/lshift.S b/sysdeps/sparc/sparc32/lshift.S new file mode 100644 index 0000000000..4f0595f2fb --- /dev/null +++ b/sysdeps/sparc/sparc32/lshift.S @@ -0,0 +1,95 @@ +! sparc __mpn_lshift -- + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_lshift) +C_SYMBOL_NAME(__mpn_lshift): + sll %o2,2,%g1 + add %o1,%g1,%o1 ! make %o1 point at end of src + ld [%o1-4],%g2 ! load first limb + sub %g0,%o3,%o5 ! negate shift count + add %o0,%g1,%o0 ! make %o0 point at end of res + add %o2,-1,%o2 + andcc %o2,4-1,%g4 ! number of limbs in first loop + srl %g2,%o5,%g1 ! compute function result + be L0 ! if multiple of 4 limbs, skip first loop + st %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +Loop0: ld [%o1-8],%g3 + add %o0,-4,%o0 + add %o1,-4,%o1 + addcc %g4,-1,%g4 + sll %g2,%o3,%o4 + srl %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + bne Loop0 + st %o4,[%o0+0] + +L0: tst %o2 + be Lend + nop + +Loop: ld [%o1-8],%g3 + add %o0,-16,%o0 + addcc %o2,-4,%o2 + sll %g2,%o3,%o4 + srl %g3,%o5,%g1 + + ld [%o1-12],%g2 + sll %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0+12] + srl %g2,%o5,%g1 + + ld [%o1-16],%g3 + sll %g2,%o3,%o4 + or %g4,%g1,%g4 + st %g4,[%o0+8] + srl %g3,%o5,%g1 + + ld [%o1-20],%g2 + sll %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0+4] + srl %g2,%o5,%g1 + + add %o1,-16,%o1 + or %g4,%g1,%g4 + bne Loop + st %g4,[%o0+0] + +Lend: sll %g2,%o3,%g2 + st %g2,[%o0-4] + retl + ld [%sp+80],%o0 diff --git a/sysdeps/sparc/sparc32/memcopy.h b/sysdeps/sparc/sparc32/memcopy.h new file mode 100644 index 0000000000..ff73f453da --- /dev/null +++ b/sysdeps/sparc/sparc32/memcopy.h @@ -0,0 +1,21 @@ +/* Copyright (C) 1991, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <sysdeps/generic/memcopy.h> +#undef reg_char +#define reg_char int diff --git a/sysdeps/sparc/sparc32/mul_1.S b/sysdeps/sparc/sparc32/mul_1.S new file mode 100644 index 0000000000..142fd8ba2a --- /dev/null +++ b/sysdeps/sparc/sparc32/mul_1.S @@ -0,0 +1,199 @@ +! SPARC __mpn_mul_1 -- Multiply a limb vector with a limb and store +! the result in a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +! ADD CODE FOR SMALL MULTIPLIERS! +!1: ld +! st +! +!2: ld ,a +! addxcc a,a,x +! st x, +! +!3_unrolled: +! ld ,a +! addxcc a,a,x1 ! 2a + cy +! addx %g0,%g0,x2 +! addcc a,x1,x ! 3a + c +! st x, +! +! ld ,a +! addxcc a,a,y1 +! addx %g0,%g0,y2 +! addcc a,y1,x +! st x, +! +!4_unrolled: +! ld ,a +! srl a,2,x1 ! 4a +! addxcc y2,x1,x +! sll a,30,x2 +! st x, +! +! ld ,a +! srl a,2,y1 +! addxcc x2,y1,y +! sll a,30,y2 +! st x, +! +!5_unrolled: +! ld ,a +! srl a,2,x1 ! 4a +! addxcc a,x1,x ! 5a + c +! sll a,30,x2 +! addx %g0,x2,x2 +! st x, +! +! ld ,a +! srl a,2,y1 +! addxcc a,y1,x +! sll a,30,y2 +! addx %g0,y2,y2 +! st x, +! +!8_unrolled: +! ld ,a +! srl a,3,x1 ! 8a +! addxcc y2,x1,x +! sll a,29,x2 +! st x, +! +! ld ,a +! srl a,3,y1 +! addxcc x2,y1,y +! sll a,29,y2 +! st x, + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_mul_1) +C_SYMBOL_NAME(__mpn_mul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne,a Loop0 + ld [%o1+%o2],%o5 + + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 ! g2 = S1_LIMB iff S2_LIMB < 0, else 0 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne,a Loop + ld [%o1+%o2],%o5 + + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/rem.S b/sysdeps/sparc/sparc32/rem.S new file mode 100644 index 0000000000..d50f1af630 --- /dev/null +++ b/sysdeps/sparc/sparc32/rem.S @@ -0,0 +1,369 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .rem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include <machine/trap.h> +#endif +#endif + +ENTRY(.rem) + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + mov %o0, %g6 ! sign of remainder matches %o0 + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +Lgot_result: + ! check to see if answer should be < 0 + tst %g6 + bl,a 1f + sub %g0, %o3, %o3 +1: + retl + mov %o3, %o0 diff --git a/sysdeps/sparc/sparc32/rshift.S b/sysdeps/sparc/sparc32/rshift.S new file mode 100644 index 0000000000..fea4f3b926 --- /dev/null +++ b/sysdeps/sparc/sparc32/rshift.S @@ -0,0 +1,92 @@ +! sparc __mpn_rshift -- + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_rshift) +C_SYMBOL_NAME(__mpn_rshift): + ld [%o1],%g2 ! load first limb + sub %g0,%o3,%o5 ! negate shift count + add %o2,-1,%o2 + andcc %o2,4-1,%g4 ! number of limbs in first loop + sll %g2,%o5,%g1 ! compute function result + be L0 ! if multiple of 4 limbs, skip first loop + st %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +Loop0: ld [%o1+4],%g3 + add %o0,4,%o0 + add %o1,4,%o1 + addcc %g4,-1,%g4 + srl %g2,%o3,%o4 + sll %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + bne Loop0 + st %o4,[%o0-4] + +L0: tst %o2 + be Lend + nop + +Loop: ld [%o1+4],%g3 + add %o0,16,%o0 + addcc %o2,-4,%o2 + srl %g2,%o3,%o4 + sll %g3,%o5,%g1 + + ld [%o1+8],%g2 + srl %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0-16] + sll %g2,%o5,%g1 + + ld [%o1+12],%g3 + srl %g2,%o3,%o4 + or %g4,%g1,%g4 + st %g4,[%o0-12] + sll %g3,%o5,%g1 + + ld [%o1+16],%g2 + srl %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0-8] + sll %g2,%o5,%g1 + + add %o1,16,%o1 + or %g4,%g1,%g4 + bne Loop + st %g4,[%o0-4] + +Lend: srl %g2,%o3,%g2 + st %g2,[%o0-0] + retl + ld [%sp+80],%o0 diff --git a/sysdeps/sparc/sparc32/sdiv.S b/sysdeps/sparc/sparc32/sdiv.S new file mode 100644 index 0000000000..02ed2e973c --- /dev/null +++ b/sysdeps/sparc/sparc32/sdiv.S @@ -0,0 +1,369 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .div name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include <machine/trap.h> +#endif +#endif + +ENTRY(.div) + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + xor %o1, %o0, %g6 ! compute sign in any case + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + ! check to see if answer should be < 0 + tst %g6 + bl,a 1f + sub %g0, %o2, %o2 +1: + retl + mov %o2, %o0 diff --git a/sysdeps/sparc/sparc32/setjmp.S b/sysdeps/sparc/sparc32/setjmp.S new file mode 100644 index 0000000000..772ec723a3 --- /dev/null +++ b/sysdeps/sparc/sparc32/setjmp.S @@ -0,0 +1,54 @@ +/* Copyright (C) 1991, 1993, 1994, 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <sysdep.h> +#include <sys/trap.h> + +#define _ASM 1 +#include <bits/setjmp.h> + +ENTRY (__sigsetjmp) + /* Save our SP and FP; in the delay slot of the jump, save our + return PC. Save the signal mask if requested with a tail-call + for simplicity; it always returns zero. */ + ta ST_FLUSH_WINDOWS +#ifdef PIC + mov %o7,%g1 +2: + call 1f + nop +1: + sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 + or %g2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 + add %g2,%o7,%g2 + sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g3 + or %g3,%lo(C_SYMBOL_NAME (__sigjmp_save)), %g3 + st %sp, [%o0 + (JB_SP * 4)] + st %fp, [%o0 + (JB_FP * 4)] + mov %g1,%o7 + ld [%g2+%g3],%g1 + jmp %g1 + st %o7, [%o0+(JB_PC*4)] +#else + sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g1 + st %sp, [%o0 + (JB_SP*4)] + or %lo(C_SYMBOL_NAME (__sigjmp_save)), %g1, %g1 + st %fp, [%o0 + (JB_FP*4)] + jmp %g1 + st %o7, [%o0 + (JB_PC*4)] +#endif /* PIC */ diff --git a/sysdeps/sparc/sparc32/sparcv8/addmul_1.S b/sysdeps/sparc/sparc32/sparcv8/addmul_1.S new file mode 100644 index 0000000000..fb9ea7cf0e --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/addmul_1.S @@ -0,0 +1,124 @@ +! SPARC v8 __mpn_addmul_1 -- Multiply a limb vector with a limb and +! add the result to a second limb vector. + +! Copyright (C) 1992, 1993, 1994, 1995 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_addmul_1) +C_SYMBOL_NAME(__mpn_addmul_1): + orcc %g0,%g0,%g2 + ld [%o1+0],%o4 ! 1 + + sll %o2,4,%g1 + and %g1,(4-1)<<4,%g1 +#if PIC + mov %o7,%g4 ! Save return address register + call 1f + add %o7,LL-1f,%g3 +1: mov %g4,%o7 ! Restore return address register +#else + sethi %hi(LL),%g3 + or %g3,%lo(LL),%g3 +#endif + jmp %g3+%g1 + nop +LL: +LL00: add %o0,-4,%o0 + b Loop00 /* 4, 8, 12, ... */ + add %o1,-4,%o1 + nop +LL01: b Loop01 /* 1, 5, 9, ... */ + nop + nop + nop +LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ + b Loop10 + add %o1,4,%o1 + nop +LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ + b Loop11 + add %o1,-8,%o1 + nop + +1: addcc %g3,%g2,%g3 ! 1 + ld [%o1+4],%o4 ! 2 + rd %y,%g2 ! 1 + addx %g0,%g2,%g2 + ld [%o0+0],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+0] ! 1 +Loop00: umul %o4,%o3,%g3 ! 2 + ld [%o0+4],%g1 ! 2 + addxcc %g3,%g2,%g3 ! 2 + ld [%o1+8],%o4 ! 3 + rd %y,%g2 ! 2 + addx %g0,%g2,%g2 + nop + addcc %g1,%g3,%g3 + st %g3,[%o0+4] ! 2 +Loop11: umul %o4,%o3,%g3 ! 3 + addxcc %g3,%g2,%g3 ! 3 + ld [%o1+12],%o4 ! 4 + rd %y,%g2 ! 3 + add %o1,16,%o1 + addx %g0,%g2,%g2 + ld [%o0+8],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+8] ! 3 +Loop10: umul %o4,%o3,%g3 ! 4 + addxcc %g3,%g2,%g3 ! 4 + ld [%o1+0],%o4 ! 1 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 + ld [%o0+12],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+12] ! 4 + add %o0,16,%o0 + addx %g0,%g2,%g2 +Loop01: addcc %o2,-4,%o2 + bg 1b + umul %o4,%o3,%g3 ! 1 + + addcc %g3,%g2,%g3 ! 4 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 + ld [%o0+0],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+0] ! 4 + addx %g0,%g2,%o0 + + retl + nop + + +! umul, ld, addxcc, rd, st + +! umul, ld, addxcc, rd, ld, addcc, st, addx + diff --git a/sysdeps/sparc/sparc32/sparcv8/mul_1.S b/sysdeps/sparc/sparc32/sparcv8/mul_1.S new file mode 100644 index 0000000000..b641feb453 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/mul_1.S @@ -0,0 +1,99 @@ +! SPARC v8 __mpn_mul_1 -- Multiply a limb vector with a single limb and +! store the product in a second limb vector. + +! Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 8 + .global C_SYMBOL_NAME(__mpn_mul_1) +C_SYMBOL_NAME(__mpn_mul_1): + sll %o2,4,%g1 + and %g1,(4-1)<<4,%g1 +#if PIC + mov %o7,%g4 ! Save return address register + call 1f + add %o7,LL-1f,%g3 +1: mov %g4,%o7 ! Restore return address register +#else + sethi %hi(LL),%g3 + or %g3,%lo(LL),%g3 +#endif + jmp %g3+%g1 + ld [%o1+0],%o4 ! 1 +LL: +LL00: add %o0,-4,%o0 + add %o1,-4,%o1 + b Loop00 /* 4, 8, 12, ... */ + orcc %g0,%g0,%g2 +LL01: b Loop01 /* 1, 5, 9, ... */ + orcc %g0,%g0,%g2 + nop + nop +LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ + add %o1,4,%o1 + b Loop10 + orcc %g0,%g0,%g2 + nop +LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ + add %o1,-8,%o1 + b Loop11 + orcc %g0,%g0,%g2 + +Loop: addcc %g3,%g2,%g3 ! 1 + ld [%o1+4],%o4 ! 2 + st %g3,[%o0+0] ! 1 + rd %y,%g2 ! 1 +Loop00: umul %o4,%o3,%g3 ! 2 + addxcc %g3,%g2,%g3 ! 2 + ld [%o1+8],%o4 ! 3 + st %g3,[%o0+4] ! 2 + rd %y,%g2 ! 2 +Loop11: umul %o4,%o3,%g3 ! 3 + addxcc %g3,%g2,%g3 ! 3 + ld [%o1+12],%o4 ! 4 + add %o1,16,%o1 + st %g3,[%o0+8] ! 3 + rd %y,%g2 ! 3 +Loop10: umul %o4,%o3,%g3 ! 4 + addxcc %g3,%g2,%g3 ! 4 + ld [%o1+0],%o4 ! 1 + st %g3,[%o0+12] ! 4 + add %o0,16,%o0 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 +Loop01: addcc %o2,-4,%o2 + bg Loop + umul %o4,%o3,%g3 ! 1 + + addcc %g3,%g2,%g3 ! 4 + st %g3,[%o0+0] ! 4 + rd %y,%g2 ! 4 + + retl + addx %g0,%g2,%o0 diff --git a/sysdeps/sparc/sparc32/sparcv8/submul_1.S b/sysdeps/sparc/sparc32/sparcv8/submul_1.S new file mode 100644 index 0000000000..e40119d011 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/submul_1.S @@ -0,0 +1,58 @@ +! SPARC v8 __mpn_submul_1 -- Multiply a limb vector with a limb and +! subtract the result from a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_submul_1) +C_SYMBOL_NAME(__mpn_submul_1): + sub %g0,%o2,%o2 ! negate ... + sll %o2,2,%o2 ! ... and scale size + sub %o1,%o2,%o1 ! o1 is offset s1_ptr + sub %o0,%o2,%g1 ! g1 is offset res_ptr + + mov 0,%o0 ! clear cy_limb + +Loop: ld [%o1+%o2],%o4 + ld [%g1+%o2],%g2 + umul %o4,%o3,%o5 + rd %y,%g3 + addcc %o5,%o0,%o5 + addx %g3,0,%o0 + subcc %g2,%o5,%g2 + addx %o0,0,%o0 + st %g2,[%g1+%o2] + + addcc %o2,4,%o2 + bne Loop + nop + + retl + nop diff --git a/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S b/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S new file mode 100644 index 0000000000..49c2398806 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S @@ -0,0 +1,186 @@ +! SPARC __udiv_qrnnd division support, used from longlong.h. + +! Copyright (C) 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! rem_ptr o0 +! n1 o1 +! n0 o2 +! d o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__udiv_qrnnd) +C_SYMBOL_NAME(__udiv_qrnnd): + tst %o3 + bneg Largedivisor + mov 8,%g1 + + b Lp1 + addxcc %o2,%o2,%o2 + +Lplop: bcc Ln1 + addxcc %o2,%o2,%o2 +Lp1: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln2 + addxcc %o2,%o2,%o2 +Lp2: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln3 + addxcc %o2,%o2,%o2 +Lp3: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln4 + addxcc %o2,%o2,%o2 +Lp4: addx %o1,%o1,%o1 + addcc %g1,-1,%g1 + bne Lplop + subcc %o1,%o3,%o4 + bcc Ln5 + addxcc %o2,%o2,%o2 +Lp5: st %o1,[%o0] + retl + xnor %g0,%o2,%o0 + +Lnlop: bcc Lp1 + addxcc %o2,%o2,%o2 +Ln1: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp2 + addxcc %o2,%o2,%o2 +Ln2: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp3 + addxcc %o2,%o2,%o2 +Ln3: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp4 + addxcc %o2,%o2,%o2 +Ln4: addx %o4,%o4,%o4 + addcc %g1,-1,%g1 + bne Lnlop + subcc %o4,%o3,%o1 + bcc Lp5 + addxcc %o2,%o2,%o2 +Ln5: st %o4,[%o0] + retl + xnor %g0,%o2,%o0 + +Largedivisor: + and %o2,1,%o5 ! %o5 = n0 & 1 + + srl %o2,1,%o2 + sll %o1,31,%g2 + or %g2,%o2,%o2 ! %o2 = lo(n1n0 >> 1) + srl %o1,1,%o1 ! %o1 = hi(n1n0 >> 1) + + and %o3,1,%g2 + srl %o3,1,%g3 ! %g3 = floor(d / 2) + add %g3,%g2,%g3 ! %g3 = ceil(d / 2) + + b LLp1 + addxcc %o2,%o2,%o2 + +LLplop: bcc LLn1 + addxcc %o2,%o2,%o2 +LLp1: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn2 + addxcc %o2,%o2,%o2 +LLp2: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn3 + addxcc %o2,%o2,%o2 +LLp3: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn4 + addxcc %o2,%o2,%o2 +LLp4: addx %o1,%o1,%o1 + addcc %g1,-1,%g1 + bne LLplop + subcc %o1,%g3,%o4 + bcc LLn5 + addxcc %o2,%o2,%o2 +LLp5: add %o1,%o1,%o1 ! << 1 + tst %g2 + bne Oddp + add %o5,%o1,%o1 + st %o1,[%o0] + retl + xnor %g0,%o2,%o0 + +LLnlop: bcc LLp1 + addxcc %o2,%o2,%o2 +LLn1: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp2 + addxcc %o2,%o2,%o2 +LLn2: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp3 + addxcc %o2,%o2,%o2 +LLn3: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp4 + addxcc %o2,%o2,%o2 +LLn4: addx %o4,%o4,%o4 + addcc %g1,-1,%g1 + bne LLnlop + subcc %o4,%g3,%o1 + bcc LLp5 + addxcc %o2,%o2,%o2 +LLn5: add %o4,%o4,%o4 ! << 1 + tst %g2 + bne Oddn + add %o5,%o4,%o4 + st %o4,[%o0] + retl + xnor %g0,%o2,%o0 + +Oddp: xnor %g0,%o2,%o2 + ! q' in %o2. r' in %o1 + addcc %o1,%o2,%o1 + bcc LLp6 + addx %o2,0,%o2 + sub %o1,%o3,%o1 +LLp6: subcc %o1,%o3,%g0 + bcs LLp7 + subx %o2,-1,%o2 + sub %o1,%o3,%o1 +LLp7: st %o1,[%o0] + retl + mov %o2,%o0 + +Oddn: xnor %g0,%o2,%o2 + ! q' in %o2. r' in %o4 + addcc %o4,%o2,%o4 + bcc LLn6 + addx %o2,0,%o2 + sub %o4,%o3,%o4 +LLn6: subcc %o4,%o3,%g0 + bcs LLn7 + subx %o2,-1,%o2 + sub %o4,%o3,%o4 +LLn7: st %o4,[%o0] + retl + mov %o2,%o0 diff --git a/sysdeps/sparc/sparc32/sub_n.S b/sysdeps/sparc/sparc32/sub_n.S new file mode 100644 index 0000000000..b7a11958e2 --- /dev/null +++ b/sysdeps/sparc/sparc32/sub_n.S @@ -0,0 +1,311 @@ +! SPARC __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and +! store difference in a third limb vector. + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +#define res_ptr %o0 +#define s1_ptr %o1 +#define s2_ptr %o2 +#define size %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_sub_n) +C_SYMBOL_NAME(__mpn_sub_n): + xor s2_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L1 ! branch if alignment differs + nop +! ** V1a ** + andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1 ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + subcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s1_ptr+0],%g4 + addcc size,-10,size + ld [s1_ptr+4],%g1 + ldd [s2_ptr+0],%g2 + blt Lfin1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1: subxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + subxcc %g4,%g2,%o4 + ld [s1_ptr+16],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+20],%g1 + ldd [s2_ptr+16],%g2 + std %o4,[res_ptr+8] + subxcc %g4,%g2,%o4 + ld [s1_ptr+24],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+28],%g1 + ldd [s2_ptr+24],%g2 + std %o4,[res_ptr+16] + subxcc %g4,%g2,%o4 + ld [s1_ptr+32],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+36],%g1 + ldd [s2_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1 + subcc %g0,%o4,%g0 ! restore cy + +Lfin1: addcc size,8-2,size + blt Lend1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: subxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1 + subcc %g0,%o4,%g0 ! restore cy +Lend1: subxcc %g4,%g2,%o4 + subxcc %g1,%g3,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s1_ptr+8],%g4 + ld [s2_ptr+8],%g2 + subxcc %g4,%g2,%o4 + st %o4,[res_ptr+8] + +Lret1: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +L1: xor s1_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L2 + nop +! ** V1b ** + andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1b ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s1_ptr */ + ld [s2_ptr],%g4 + add s2_ptr,4,s2_ptr + ld [s1_ptr],%g2 + add s1_ptr,4,s1_ptr + add size,-1,size + subcc %g2,%g4,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1b: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s2_ptr+0],%g4 + addcc size,-10,size + ld [s2_ptr+4],%g1 + ldd [s1_ptr+0],%g2 + blt Lfin1b + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1b: subxcc %g2,%g4,%o4 + ld [s2_ptr+8],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+12],%g1 + ldd [s1_ptr+8],%g2 + std %o4,[res_ptr+0] + subxcc %g2,%g4,%o4 + ld [s2_ptr+16],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+20],%g1 + ldd [s1_ptr+16],%g2 + std %o4,[res_ptr+8] + subxcc %g2,%g4,%o4 + ld [s2_ptr+24],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+28],%g1 + ldd [s1_ptr+24],%g2 + std %o4,[res_ptr+16] + subxcc %g2,%g4,%o4 + ld [s2_ptr+32],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+36],%g1 + ldd [s1_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1b + subcc %g0,%o4,%g0 ! restore cy + +Lfin1b: addcc size,8-2,size + blt Lend1b + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1b:subxcc %g2,%g4,%o4 + ld [s2_ptr+8],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+12],%g1 + ldd [s1_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1b + subcc %g0,%o4,%g0 ! restore cy +Lend1b: subxcc %g2,%g4,%o4 + subxcc %g3,%g1,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1b + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s2_ptr+8],%g4 + ld [s1_ptr+8],%g2 + subxcc %g2,%g4,%o4 + st %o4,[res_ptr+8] + +Lret1b: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +! ** V2 ** +/* If we come here, the alignment of s1_ptr and res_ptr as well as the + alignment of s2_ptr and res_ptr differ. Since there are only two ways + things can be aligned (that we care about) we now know that the alignment + of s1_ptr and s2_ptr are the same. */ + +L2: cmp size,1 + be Ljone + nop + andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 + be L_v2 ! if no, branch + nop +/* Add least significant limb separately to align s1_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + subcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr + +L_v2: addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + blt Lfin2 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + ldd [s1_ptr+8],%g2 + ldd [s2_ptr+8],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+8] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+12] + ldd [s1_ptr+16],%g2 + ldd [s2_ptr+16],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+16] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+20] + ldd [s1_ptr+24],%g2 + ldd [s2_ptr+24],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+24] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+28] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop2 + subcc %g0,%o4,%g0 ! restore cy + +Lfin2: addcc size,8-2,size + blt Lend2 + subcc %g0,%o4,%g0 ! restore cy +Loope2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope2 + subcc %g0,%o4,%g0 ! restore cy +Lend2: andcc size,1,%g0 + be Lret2 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ +Ljone: ld [s1_ptr],%g4 + ld [s2_ptr],%g2 + subxcc %g4,%g2,%o4 + st %o4,[res_ptr] + +Lret2: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/sparc32/submul_1.S b/sysdeps/sparc/sparc32/submul_1.S new file mode 100644 index 0000000000..a8ebd501a7 --- /dev/null +++ b/sysdeps/sparc/sparc32/submul_1.S @@ -0,0 +1,147 @@ +! SPARC __mpn_submul_1 -- Multiply a limb vector with a limb and subtract +! the result from a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_submul_1) +C_SYMBOL_NAME(__mpn_submul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + subcc %o5,%g1,%g1 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne Loop0 + ld [%o4+%o2],%o5 + + subcc %o5,%g1,%g1 + addx %o0,%g0,%o0 + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + subcc %o5,%g3,%g3 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 + addcc %o2,4,%o2 + bne Loop + ld [%o4+%o2],%o5 + + subcc %o5,%g3,%g3 + addx %o0,%g0,%o0 + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/udiv.S b/sysdeps/sparc/sparc32/udiv.S new file mode 100644 index 0000000000..4a7d1526b0 --- /dev/null +++ b/sysdeps/sparc/sparc32/udiv.S @@ -0,0 +1,352 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .udiv name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include <machine/trap.h> +#endif +#endif + +ENTRY(.udiv) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + + retl + mov %o2, %o0 diff --git a/sysdeps/sparc/sparc32/udiv_qrnnd.S b/sysdeps/sparc/sparc32/udiv_qrnnd.S new file mode 100644 index 0000000000..5b67f874c3 --- /dev/null +++ b/sysdeps/sparc/sparc32/udiv_qrnnd.S @@ -0,0 +1,165 @@ +! SPARC __udiv_qrnnd division support, used from longlong.h. + +! Copyright (C) 1993, 1994, 1997 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library 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 Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + +! Added PIC support - May/96, Miguel de Icaza + +! INPUT PARAMETERS +! rem_ptr i0 +! n1 i1 +! n0 i2 +! d i3 + +#include "sysdep.h" +#undef ret /* Kludge for glibc */ + + .text + .align 8 + + .type two_to_32,@object +two_to_32: + .double 0r4294967296 + .size two_to_32,8 + + .type two_to_31,@object +two_to_31: + .double 0r2147483648 + .size two_to_31,8 + + .align 4 + .global __udiv_qrnnd + .type __udiv_qrnnd,@function +ENTRY(__udiv_qrnnd) + !#PROLOGUE# 0 + save %sp,-104,%sp + !#PROLOGUE# 1 + st %i1,[%fp-8] + ld [%fp-8],%f10 +#ifdef PIC +.Lbase: call 1f + fitod %f10,%f4 +1: ldd [%o7-(.Lbase-two_to_32)],%f8 +#else + sethi %hi(two_to_32),%o7 + fitod %f10,%f4 + ldd [%o7+%lo(two_to_32)],%f8 +#endif + cmp %i1,0 + bge L248 + mov %i0,%i5 + faddd %f4,%f8,%f4 +.L248: + st %i2,[%fp-8] + ld [%fp-8],%f10 + fmuld %f4,%f8,%f6 + cmp %i2,0 + bge L249 + fitod %f10,%f2 + faddd %f2,%f8,%f2 +.L249: + st %i3,[%fp-8] + faddd %f6,%f2,%f2 + ld [%fp-8],%f10 + cmp %i3,0 + bge L250 + fitod %f10,%f4 + faddd %f4,%f8,%f4 +.L250: + fdivd %f2,%f4,%f2 +#ifdef PIC + ldd [%o7-(.Lbase-two_to_31)],%f4 +#else + sethi %hi(two_to_31),%o7 + ldd [%o7+%lo(two_to_31)],%f4 +#endif + fcmped %f2,%f4 + nop + fbge,a L251 + fsubd %f2,%f4,%f2 + fdtoi %f2,%f2 + st %f2,[%fp-8] + b L252 + ld [%fp-8],%i4 +.L251: + fdtoi %f2,%f2 + st %f2,[%fp-8] + ld [%fp-8],%i4 + sethi %hi(-2147483648),%g2 + xor %i4,%g2,%i4 +.L252: + wr %g0,%i4,%y + sra %i3,31,%g2 + and %i4,%g2,%g2 + andcc %g0,0,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,0,%g1 + add %g1,%g2,%i0 + rd %y,%g3 + subcc %i2,%g3,%o7 + subxcc %i1,%i0,%g0 + be L253 + cmp %o7,%i3 + + add %i4,-1,%i0 + add %o7,%i3,%o7 + st %o7,[%i5] + ret + restore +.L253: + blu L246 + mov %i4,%i0 + add %i4,1,%i0 + sub %o7,%i3,%o7 +.L246: + st %o7,[%i5] + ret + restore + + .size __udiv_qrnnd, .-__udiv_qrnnd diff --git a/sysdeps/sparc/sparc32/umul.S b/sysdeps/sparc/sparc32/umul.S new file mode 100644 index 0000000000..7a26c295cb --- /dev/null +++ b/sysdeps/sparc/sparc32/umul.S @@ -0,0 +1,153 @@ +/* + * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the + * upper 32 bits of the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. Short + * multiplies require 25 instruction cycles, and long ones require + * 45 instruction cycles. + * + * On return, overflow has occurred (%o1 is not zero) if and only if + * the Z condition code is clear, allowing, e.g., the following: + * + * call .umul + * nop + * bnz overflow (or tnz) + */ + +#include "DEFS.h" +FUNC(.umul) + or %o0, %o1, %o4 + mov %o0, %y ! multiplier -> Y + andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + + /* + * Normally, with the shift-and-add approach, if both numbers are + * positive you get the correct result. With 32-bit two's-complement + * numbers, -x is represented as + * + * x 32 + * ( 2 - ------ ) mod 2 * 2 + * 32 + * 2 + * + * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, + * we can treat this as if the radix point were just to the left + * of the sign bit (multiply by 2^32), and get + * + * -x = (2 - x) mod 2 + * + * Then, ignoring the `mod 2's for convenience: + * + * x * y = xy + * -x * y = 2y - xy + * x * -y = 2x - xy + * -x * -y = 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */ +#if 0 + tst %o1 + bl,a 1f ! if %o1 < 0 (high order bit = 1), + add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) +1: rd %y, %o0 ! get lower half of product + retl + addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 +#else + /* Faster code from tege@sics.se. */ + sra %o1, 31, %o2 ! make mask from sign bit + and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %o2, %o1 ! add compensation and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above), + * and overflow is impossible (the answer is at most 24 bits long). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the result; %y has + * the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * -----result----- + * + * The 12 bits of %o4 left of the `result' area are all zero; + * in fact, all top 20 bits of %o4 are zero. + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20 + or %o5, %o0, %o0 + retl + addcc %g0, %g0, %o1 ! %o1 = zero, and set Z diff --git a/sysdeps/sparc/sparc32/urem.S b/sysdeps/sparc/sparc32/urem.S new file mode 100644 index 0000000000..e72c33dd63 --- /dev/null +++ b/sysdeps/sparc/sparc32/urem.S @@ -0,0 +1,352 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .urem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include <machine/trap.h> +#endif +#endif + +ENTRY(.urem) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +Lgot_result: + + retl + mov %o3, %o0 |