diff options
Diffstat (limited to 'sysdeps/sparc')
| -rw-r--r-- | sysdeps/sparc/add_n.S | 42 | ||||
| -rw-r--r-- | sysdeps/sparc/lshift.S | 6 | ||||
| -rw-r--r-- | sysdeps/sparc/rshift.S | 6 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/add_n.s | 57 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/addmul_1.s | 88 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/gmp-mparam.h | 26 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/lshift.s | 95 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/mul_1.s | 85 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/rshift.s | 92 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/sub_n.s | 57 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc64/submul_1.s | 88 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc8/addmul_1.S | 2 | ||||
| -rw-r--r-- | sysdeps/sparc/sparc8/mul_1.S | 2 | ||||
| -rw-r--r-- | sysdeps/sparc/sub_n.S | 12 |
14 files changed, 624 insertions, 34 deletions
diff --git a/sysdeps/sparc/add_n.S b/sysdeps/sparc/add_n.S index 80c3b99640..49b31fc660 100644 --- a/sysdeps/sparc/add_n.S +++ b/sysdeps/sparc/add_n.S @@ -1,7 +1,7 @@ -! sparc __mpn_add_n -- Add two limb vectors of the same length > 0 and store +! SPARC __mpn_add_n -- Add two limb vectors of the same length > 0 and store ! sum in a third limb vector. -! Copyright (C) 1995 Free Software Foundation, Inc. +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. @@ -32,18 +32,14 @@ .align 4 .global C_SYMBOL_NAME(__mpn_add_n) C_SYMBOL_NAME(__mpn_add_n): - cmp size,8 - mov 0,%o4 ! clear cy-save register - blt,a Ltriv - addcc size,-2,size 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 - beq L_v1 ! if no, branch + be L_v1 ! if no, branch nop -! ** V1a ** /* Add least significant limb separately to align res_ptr and s2_ptr */ ld [s1_ptr],%g4 add s1_ptr,4,s1_ptr @@ -53,12 +49,15 @@ L0: andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 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 -L_v1: ld [s1_ptr+0],%g4 + ld [s1_ptr+0],%g4 + addcc size,-10,size ld [s1_ptr+4],%g1 ldd [s2_ptr+0],%g2 - addx %g0,%g0,%o4 ! save cy in register - addcc size,-10,size blt Lfin1 subcc %g0,%o4,%g0 ! restore cy /* Add blocks of 8 limbs until less than 8 limbs remain */ @@ -98,7 +97,7 @@ 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 */ -Loop1b: addxcc %g4,%g2,%o4 +Loope1: addxcc %g4,%g2,%o4 ld [s1_ptr+8],%g4 addxcc %g1,%g3,%o5 ld [s1_ptr+12],%g1 @@ -109,7 +108,7 @@ Loop1b: addxcc %g4,%g2,%o4 add s1_ptr,8,s1_ptr add s2_ptr,8,s2_ptr add res_ptr,8,res_ptr - bge Loop1b + bge Loope1 subcc %g0,%o4,%g0 ! restore cy Lend1: addxcc %g4,%g2,%o4 addxcc %g1,%g3,%o5 @@ -144,10 +143,13 @@ L1: xor s1_ptr,res_ptr,%g1 things can be aligned (that we care about) we now know that the alignment of s1_ptr and s2_ptr are the same. */ -L2: andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 - beq L_v2 ! if no, branch +L2: cmp size,1 + be Ljone nop -/* Add least significant limb separately to align res_ptr and s2_ptr */ + 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 @@ -195,9 +197,9 @@ Loop2: ldd [s1_ptr+0],%g2 subcc %g0,%o4,%g0 ! restore cy Lfin2: addcc size,8-2,size -Ltriv: blt Lend2 + blt Lend2 subcc %g0,%o4,%g0 ! restore cy -Loop2b: ldd [s1_ptr+0],%g2 +Loope2: ldd [s1_ptr+0],%g2 ldd [s2_ptr+0],%o4 addxcc %g2,%o4,%g2 st %g2,[res_ptr+0] @@ -208,13 +210,13 @@ Loop2b: ldd [s1_ptr+0],%g2 add s1_ptr,8,s1_ptr add s2_ptr,8,s2_ptr add res_ptr,8,res_ptr - bge Loop2b + bge Loope2 subcc %g0,%o4,%g0 ! restore cy Lend2: andcc size,1,%g0 be Lret2 subcc %g0,%o4,%g0 ! restore cy /* Add last limb */ - ld [s1_ptr],%g4 +Ljone: ld [s1_ptr],%g4 ld [s2_ptr],%g2 addxcc %g4,%g2,%o4 st %o4,[res_ptr] diff --git a/sysdeps/sparc/lshift.S b/sysdeps/sparc/lshift.S index 497272af0f..6844fa2475 100644 --- a/sysdeps/sparc/lshift.S +++ b/sysdeps/sparc/lshift.S @@ -1,6 +1,6 @@ ! sparc __mpn_lshift -- -! Copyright (C) 1995 Free Software Foundation, Inc. +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. @@ -39,7 +39,7 @@ C_SYMBOL_NAME(__mpn_lshift): add %o2,-1,%o2 andcc %o2,4-1,%g4 ! number of limbs in first loop srl %g2,%o5,%g1 ! compute function result - beq L0 ! if multiple of 4 limbs, skip first loop + be L0 ! if multiple of 4 limbs, skip first loop st %g1,[%sp+80] sub %o2,%g4,%o2 ! adjust count for main loop @@ -56,7 +56,7 @@ Loop0: ld [%o1-8],%g3 st %o4,[%o0+0] L0: tst %o2 - beq Lend + be Lend nop Loop: ld [%o1-8],%g3 diff --git a/sysdeps/sparc/rshift.S b/sysdeps/sparc/rshift.S index 3428cfe9a4..5a47926302 100644 --- a/sysdeps/sparc/rshift.S +++ b/sysdeps/sparc/rshift.S @@ -1,6 +1,6 @@ ! sparc __mpn_rshift -- -! Copyright (C) 1995 Free Software Foundation, Inc. +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. @@ -36,7 +36,7 @@ C_SYMBOL_NAME(__mpn_rshift): add %o2,-1,%o2 andcc %o2,4-1,%g4 ! number of limbs in first loop sll %g2,%o5,%g1 ! compute function result - beq L0 ! if multiple of 4 limbs, skip first loop + be L0 ! if multiple of 4 limbs, skip first loop st %g1,[%sp+80] sub %o2,%g4,%o2 ! adjust count for main loop @@ -53,7 +53,7 @@ Loop0: ld [%o1+4],%g3 st %o4,[%o0-4] L0: tst %o2 - beq Lend + be Lend nop Loop: ld [%o1+4],%g3 diff --git a/sysdeps/sparc/sparc64/add_n.s b/sysdeps/sparc/sparc64/add_n.s new file mode 100644 index 0000000000..104a89e78c --- /dev/null +++ b/sysdeps/sparc/sparc64/add_n.s @@ -0,0 +1,57 @@ +! SPARC v9 __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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! s1_ptr %o1 +! s2_ptr %o2 +! size %o3 + +.section ".text" + .align 4 + .global __mpn_add_n + .type __mpn_add_n,#function + .proc 04 +__mpn_add_n: + sub %g0,%o3,%g3 + sllx %o3,3,%g1 + add %o1,%g1,%o1 ! make s1_ptr point at end + add %o2,%g1,%o2 ! make s2_ptr point at end + add %o0,%g1,%o0 ! make res_ptr point at end + mov 0,%o4 ! clear carry variable + sllx %g3,3,%o5 ! compute initial address index + +.Loop: ldx [%o2+%o5],%g1 ! load s2 limb + add %g3,1,%g3 ! increment loop count + ldx [%o1+%o5],%g2 ! load s1 limb + addcc %g1,%o4,%g1 ! add s2 limb and carry variable + movcc %xcc,0,%o4 ! if carry-out, o4 was 1; clear it + addcc %g1,%g2,%g1 ! add s1 limb to sum + stx %g1,[%o0+%o5] ! store result + add %o5,8,%o5 ! increment address index + brnz,pt %g3,.Loop + movcs %xcc,1,%o4 ! if s1 add gave carry, record it + + retl + mov %o4,%o0 +.LLfe1: + .size __mpn_add_n,.LLfe1-__mpn_add_n diff --git a/sysdeps/sparc/sparc64/addmul_1.s b/sysdeps/sparc/sparc64/addmul_1.s new file mode 100644 index 0000000000..ef013eeb91 --- /dev/null +++ b/sysdeps/sparc/sparc64/addmul_1.s @@ -0,0 +1,88 @@ +! SPARC v9 __mpn_addmul_1 -- Multiply a limb vector with a single limb and +! add the product to a second limb vector. + +! Copyright (C) 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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +.section ".text" + .align 4 + .global __mpn_addmul_1 + .type __mpn_addmul_1,#function + .proc 016 +__mpn_addmul_1: + !#PROLOGUE# 0 + save %sp,-160,%sp + !#PROLOGUE# 1 + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +.Loop: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + ldx [%o4+%g1],%l1 + addcc %l1,%i0,%i0 + movcs %xcc,1,%g5 + stx %i0,[%o4+%g1] + brnz %o7,.Loop + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + ret + restore +.LLfe1: + .size __mpn_addmul_1,.LLfe1-__mpn_addmul_1 diff --git a/sysdeps/sparc/sparc64/gmp-mparam.h b/sysdeps/sparc/sparc64/gmp-mparam.h new file mode 100644 index 0000000000..05c893f790 --- /dev/null +++ b/sysdeps/sparc/sparc64/gmp-mparam.h @@ -0,0 +1,26 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file. + +Copyright (C) 1991, 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. */ + +#define BITS_PER_MP_LIMB 64 +#define BYTES_PER_MP_LIMB 8 +#define BITS_PER_LONGINT 64 +#define BITS_PER_INT 32 +#define BITS_PER_SHORTINT 16 +#define BITS_PER_CHAR 8 diff --git a/sysdeps/sparc/sparc64/lshift.s b/sysdeps/sparc/sparc64/lshift.s new file mode 100644 index 0000000000..bd7fa0111f --- /dev/null +++ b/sysdeps/sparc/sparc64/lshift.s @@ -0,0 +1,95 @@ +! SPARC v9 __mpn_lshift -- + +! Copyright (C) 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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +.section ".text" + .align 4 + .global __mpn_lshift + .type __mpn_lshift,#function + .proc 04 +__mpn_lshift: + sllx %o2,3,%g1 + add %o1,%g1,%o1 ! make %o1 point at end of src + ldx [%o1-8],%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 + and %o2,4-1,%g4 ! number of limbs in first loop + srlx %g2,%o5,%g1 ! compute function result + brz,pn %g4,.L0 ! if multiple of 4 limbs, skip first loop + stx %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +.Loop0: ldx [%o1-16],%g3 + add %o0,-8,%o0 + add %o1,-8,%o1 + add %g4,-1,%g4 + sllx %g2,%o3,%o4 + srlx %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + brnz,pt %g4,.Loop0 + stx %o4,[%o0+0] + +.L0: brz,pn %o2,.Lend + nop + +.Loop: ldx [%o1-16],%g3 + add %o0,-32,%o0 + add %o2,-4,%o2 + sllx %g2,%o3,%o4 + srlx %g3,%o5,%g1 + + ldx [%o1-24],%g2 + sllx %g3,%o3,%g4 + or %o4,%g1,%o4 + stx %o4,[%o0+24] + srlx %g2,%o5,%g1 + + ldx [%o1-32],%g3 + sllx %g2,%o3,%o4 + or %g4,%g1,%g4 + stx %g4,[%o0+16] + srlx %g3,%o5,%g1 + + ldx [%o1-40],%g2 + sllx %g3,%o3,%g4 + or %o4,%g1,%o4 + stx %o4,[%o0+8] + srlx %g2,%o5,%g1 + + add %o1,-32,%o1 + or %g4,%g1,%g4 + brnz,pt %o2,.Loop + stx %g4,[%o0+0] + +.Lend: sllx %g2,%o3,%g2 + stx %g2,[%o0-8] + retl + ldx [%sp+80],%o0 +.LLfe1: + .size __mpn_lshift,.LLfe1-__mpn_lshift diff --git a/sysdeps/sparc/sparc64/mul_1.s b/sysdeps/sparc/sparc64/mul_1.s new file mode 100644 index 0000000000..41be3705a8 --- /dev/null +++ b/sysdeps/sparc/sparc64/mul_1.s @@ -0,0 +1,85 @@ +! SPARC v9 __mpn_mul_1 -- Multiply a limb vector with a single limb and +! store the product in a second 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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +.section ".text" + .align 4 + .global __mpn_mul_1 + .type __mpn_mul_1,#function + .proc 016 +__mpn_mul_1: + !#PROLOGUE# 0 + save %sp,-160,%sp + !#PROLOGUE# 1 + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +.Loop: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + stx %i0,[%o4+%g1] + brnz %o7,.Loop + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + ret + restore +.LLfe1: + .size __mpn_mul_1,.LLfe1-__mpn_mul_1 diff --git a/sysdeps/sparc/sparc64/rshift.s b/sysdeps/sparc/sparc64/rshift.s new file mode 100644 index 0000000000..971deec2a4 --- /dev/null +++ b/sysdeps/sparc/sparc64/rshift.s @@ -0,0 +1,92 @@ +! SPARC v9 __mpn_rshift -- + +! Copyright (C) 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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +.section ".text" + .align 4 + .global __mpn_rshift + .type __mpn_rshift,#function + .proc 04 +__mpn_rshift: + ldx [%o1],%g2 ! load first limb + sub %g0,%o3,%o5 ! negate shift count + add %o2,-1,%o2 + and %o2,4-1,%g4 ! number of limbs in first loop + sllx %g2,%o5,%g1 ! compute function result + brz,pn %g4,.L0 ! if multiple of 4 limbs, skip first loop + stx %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +.Loop0: ldx [%o1+8],%g3 + add %o0,8,%o0 + add %o1,8,%o1 + add %g4,-1,%g4 + srlx %g2,%o3,%o4 + sllx %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + brnz,pt %g4,.Loop0 + stx %o4,[%o0-8] + +.L0: brz,pn %o2,.Lend + nop + +.Loop: ldx [%o1+8],%g3 + add %o0,32,%o0 + add %o2,-4,%o2 + srlx %g2,%o3,%o4 + sllx %g3,%o5,%g1 + + ldx [%o1+16],%g2 + srlx %g3,%o3,%g4 + or %o4,%g1,%o4 + stx %o4,[%o0-32] + sllx %g2,%o5,%g1 + + ldx [%o1+24],%g3 + srlx %g2,%o3,%o4 + or %g4,%g1,%g4 + stx %g4,[%o0-24] + sllx %g3,%o5,%g1 + + ldx [%o1+32],%g2 + srlx %g3,%o3,%g4 + or %o4,%g1,%o4 + stx %o4,[%o0-16] + sllx %g2,%o5,%g1 + + add %o1,32,%o1 + or %g4,%g1,%g4 + brnz %o2,.Loop + stx %g4,[%o0-8] + +.Lend: srlx %g2,%o3,%g2 + stx %g2,[%o0-0] + retl + ldx [%sp+80],%o0 +.LLfe1: + .size __mpn_rshift,.LLfe1-__mpn_rshift diff --git a/sysdeps/sparc/sparc64/sub_n.s b/sysdeps/sparc/sparc64/sub_n.s new file mode 100644 index 0000000000..7099bf43e9 --- /dev/null +++ b/sysdeps/sparc/sparc64/sub_n.s @@ -0,0 +1,57 @@ +! SPARC v9 __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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! s1_ptr %o1 +! s2_ptr %o2 +! size %o3 + +.section ".text" + .align 4 + .global __mpn_sub_n + .type __mpn_sub_n,#function + .proc 04 +__mpn_sub_n: + sub %g0,%o3,%g3 + sllx %o3,3,%g1 + add %o1,%g1,%o1 ! make s1_ptr point at end + add %o2,%g1,%o2 ! make s2_ptr point at end + add %o0,%g1,%o0 ! make res_ptr point at end + mov 0,%o4 ! clear carry variable + sllx %g3,3,%o5 ! compute initial address index + +.Loop: ldx [%o2+%o5],%g1 ! load s2 limb + add %g3,1,%g3 ! increment loop count + ldx [%o1+%o5],%g2 ! load s1 limb + addcc %g1,%o4,%g1 ! add s2 limb and carry variable + movcc %xcc,0,%o4 ! if carry-out, o4 was 1; clear it + subcc %g1,%g2,%g1 ! subtract s1 limb from sum + stx %g1,[%o0+%o5] ! store result + add %o5,8,%o5 ! increment address index + brnz,pt %g3,.Loop + movcs %xcc,1,%o4 ! if s1 subtract gave carry, record it + + retl + mov %o4,%o0 +.LLfe1: + .size __mpn_sub_n,.LLfe1-__mpn_sub_n diff --git a/sysdeps/sparc/sparc64/submul_1.s b/sysdeps/sparc/sparc64/submul_1.s new file mode 100644 index 0000000000..f0df38c3bc --- /dev/null +++ b/sysdeps/sparc/sparc64/submul_1.s @@ -0,0 +1,88 @@ +! SPARC v9 __mpn_submul_1 -- Multiply a limb vector with a single limb and +! subtract the product from a second limb vector. + +! Copyright (C) 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., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +.section ".text" + .align 4 + .global __mpn_submul_1 + .type __mpn_submul_1,#function + .proc 016 +__mpn_submul_1: + !#PROLOGUE# 0 + save %sp,-160,%sp + !#PROLOGUE# 1 + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +.Loop: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + ldx [%o4+%g1],%l1 + subcc %l1,%i0,%i0 + movcs %xcc,1,%g5 + stx %i0,[%o4+%g1] + brnz %o7,.Loop + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + ret + restore +.LLfe1: + .size __mpn_submul_1,.LLfe1-__mpn_submul_1 diff --git a/sysdeps/sparc/sparc8/addmul_1.S b/sysdeps/sparc/sparc8/addmul_1.S index d1de0c3649..1cf5128ee5 100644 --- a/sysdeps/sparc/sparc8/addmul_1.S +++ b/sysdeps/sparc/sparc8/addmul_1.S @@ -1,7 +1,7 @@ ! 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 Free Software Foundation, Inc. +! Copyright (C) 1992, 1993, 1994, 1995 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. diff --git a/sysdeps/sparc/sparc8/mul_1.S b/sysdeps/sparc/sparc8/mul_1.S index 42717be33b..d56394e135 100644 --- a/sysdeps/sparc/sparc8/mul_1.S +++ b/sysdeps/sparc/sparc8/mul_1.S @@ -1,7 +1,7 @@ ! 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 Free Software Foundation, Inc. +! Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. diff --git a/sysdeps/sparc/sub_n.S b/sysdeps/sparc/sub_n.S index 2e217ed679..9ff5b7b863 100644 --- a/sysdeps/sparc/sub_n.S +++ b/sysdeps/sparc/sub_n.S @@ -1,7 +1,7 @@ -! sparc __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and +! SPARC __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and ! store difference in a third limb vector. -! Copyright (C) 1995 Free Software Foundation, Inc. +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. ! This file is part of the GNU MP Library. @@ -38,7 +38,7 @@ C_SYMBOL_NAME(__mpn_sub_n): nop ! ** V1a ** andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 - beq L_v1 ! if no, branch + be L_v1 ! if no, branch nop /* Add least significant limb separately to align res_ptr and s2_ptr */ ld [s1_ptr],%g4 @@ -133,9 +133,9 @@ L1: xor s1_ptr,res_ptr,%g1 nop ! ** V1b ** andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 - beq L_v1b ! if no, branch + be L_v1b ! if no, branch nop -/* Add least significant limb separately to align res_ptr and s2_ptr */ +/* 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 @@ -232,7 +232,7 @@ L2: cmp size,1 be Ljone nop andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 - beq L_v2 ! if no, branch + be L_v2 ! if no, branch nop /* Add least significant limb separately to align s1_ptr and s2_ptr */ ld [s1_ptr],%g4 |