diff options
Diffstat (limited to 'sysdeps/sparc/add_n.S')
| -rw-r--r-- | sysdeps/sparc/add_n.S | 42 |
1 files changed, 22 insertions, 20 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] |