dnl x86-64 mpn_rsh1add_n/mpn_rsh1sub_n optimized for Pentium 4. dnl Contributed to the GNU project by Torbjorn Granlund. dnl Copyright 2007, 2008, 2010-2012 Free Software Foundation, Inc. dnl This file is part of the GNU MP Library. dnl dnl The GNU MP Library is free software; you can redistribute it and/or modify dnl it under the terms of either: dnl dnl * the GNU Lesser General Public License as published by the Free dnl Software Foundation; either version 3 of the License, or (at your dnl option) any later version. dnl dnl or dnl dnl * the GNU General Public License as published by the Free Software dnl Foundation; either version 2 of the License, or (at your option) any dnl later version. dnl dnl or both in parallel, as here. dnl dnl The GNU MP Library is distributed in the hope that it will be useful, but dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License dnl for more details. dnl dnl You should have received copies of the GNU General Public License and the dnl GNU Lesser General Public License along with the GNU MP Library. If not, dnl see https://www.gnu.org/licenses/. include(`../config.m4') C cycles/limb C AMD K8,K9 4.13 C AMD K10 4.13 C Intel P4 5.70 C Intel core2 4.75 C Intel corei 5 C Intel atom 8.75 C VIA nano 5.25 C TODO C * Try to make this smaller, 746 bytes seem excessive for this 2nd class C function. Less sw pipelining would help, and since we now probably C pipeline somewhat too deeply, it might not affect performance too much. C * A separate small-n loop might speed things as well as make things smaller. C That loop should be selected before pushing registers. C INPUT PARAMETERS define(`rp', `%rdi') define(`up', `%rsi') define(`vp', `%rdx') define(`n', `%rcx') define(`cy', `%r8') ifdef(`OPERATION_rsh1add_n', ` define(ADDSUB, add) define(func, mpn_rsh1add_n) define(func_nc, mpn_rsh1add_nc)') ifdef(`OPERATION_rsh1sub_n', ` define(ADDSUB, sub) define(func, mpn_rsh1sub_n) define(func_nc, mpn_rsh1sub_nc)') ABI_SUPPORT(DOS64) ABI_SUPPORT(STD64) MULFUNC_PROLOGUE(mpn_rsh1add_n mpn_rsh1add_nc mpn_rsh1sub_n mpn_rsh1sub_nc) ASM_START() TEXT PROLOGUE(func) FUNC_ENTRY(4) xor %r8, %r8 IFDOS(` jmp L(ent) ') EPILOGUE() PROLOGUE(func_nc) FUNC_ENTRY(4) IFDOS(` mov 56(%rsp), %r8 ') L(ent): push %rbx push %r12 push %r13 push %r14 push %r15 mov (vp), %r9 mov (up), %r15 mov R32(n), R32(%rax) and $3, R32(%rax) jne L(n00) mov R32(%r8), R32(%rbx) C n = 0, 4, 8, ... mov 8(up), %r10 ADDSUB %r9, %r15 mov 8(vp), %r9 setc R8(%rax) ADDSUB %rbx, %r15 C return bit jnc 1f mov $1, R8(%rax) 1: mov 16(up), %r12 ADDSUB %r9, %r10 mov 16(vp), %r9 setc R8(%rbx) mov %r15, %r13 ADDSUB %rax, %r10 jnc 1f mov $1, R8(%rbx) 1: mov 24(up), %r11 ADDSUB %r9, %r12 lea 32(up), up mov 24(vp), %r9 lea 32(vp), vp setc R8(%rax) mov %r10, %r14 shl $63, %r10 shr %r13 jmp L(L00) L(n00): cmp $2, R32(%rax) jnc L(n01) xor R32(%rbx), R32(%rbx) C n = 1, 5, 9, ... lea -24(rp), rp mov R32(%r8), R32(%rax) dec n jnz L(gt1) ADDSUB %r9, %r15 setc R8(%rbx) ADDSUB %rax, %r15 jnc 1f mov $1, R8(%rbx) 1: mov %r15, %r14 shl $63, %rbx shr %r14 jmp L(cj1) L(gt1): mov 8(up), %r8 ADDSUB %r9, %r15 mov 8(vp), %r9 setc R8(%rbx) ADDSUB %rax, %r15 jnc 1f mov $1, R8(%rbx) 1: mov 16(up), %r10 ADDSUB %r9, %r8 mov 16(vp), %r9 setc R8(%rax) mov %r15, %r14 ADDSUB %rbx, %r8 jnc 1f mov $1, R8(%rax) 1: mov 24(up), %r12 ADDSUB %r9, %r10 mov 24(vp), %r9 setc R8(%rbx) mov %r8, %r13 shl $63, %r8 shr %r14 lea 8(up), up lea 8(vp), vp jmp L(L01) L(n01): jne L(n10) lea -16(rp), rp C n = 2, 6, 10, ... mov R32(%r8), R32(%rbx) mov 8(up), %r11 ADDSUB %r9, %r15 mov 8(vp), %r9 setc R8(%rax) ADDSUB %rbx, %r15 jnc 1f mov $1, R8(%rax) 1: sub $2, n jnz L(gt2) ADDSUB %r9, %r11 setc R8(%rbx) mov %r15, %r13 ADDSUB %rax, %r11 jnc 1f mov $1, R8(%rbx) 1: mov %r11, %r14 shl $63, %r11 shr %r13 jmp L(cj2) L(gt2): mov 16(up), %r8 ADDSUB %r9, %r11 mov 16(vp), %r9 setc R8(%rbx) mov %r15, %r13 ADDSUB %rax, %r11 jnc 1f mov $1, R8(%rbx) 1: mov 24(up), %r10 ADDSUB %r9, %r8 mov 24(vp), %r9 setc R8(%rax) mov %r11, %r14 shl $63, %r11 shr %r13 lea 16(up), up lea 16(vp), vp jmp L(L10) L(n10): xor R32(%rbx), R32(%rbx) C n = 3, 7, 11, ... lea -8(rp), rp mov R32(%r8), R32(%rax) mov 8(up), %r12 ADDSUB %r9, %r15 mov 8(vp), %r9 setc R8(%rbx) ADDSUB %rax, %r15 jnc 1f mov $1, R8(%rbx) 1: mov 16(up), %r11 ADDSUB %r9, %r12 mov 16(vp), %r9 setc R8(%rax) mov %r15, %r14 ADDSUB %rbx, %r12 jnc 1f mov $1, R8(%rax) 1: sub $3, n jnz L(gt3) ADDSUB %r9, %r11 setc R8(%rbx) mov %r12, %r13 shl $63, %r12 shr %r14 jmp L(cj3) L(gt3): mov 24(up), %r8 ADDSUB %r9, %r11 mov 24(vp), %r9 setc R8(%rbx) mov %r12, %r13 shl $63, %r12 shr %r14 lea 24(up), up lea 24(vp), vp jmp L(L11) L(c0): mov $1, R8(%rbx) jmp L(rc0) L(c1): mov $1, R8(%rax) jmp L(rc1) L(c2): mov $1, R8(%rbx) jmp L(rc2) ALIGN(16) L(top): mov (up), %r8 C not on critical path or %r13, %r10 ADDSUB %r9, %r11 C not on critical path mov (vp), %r9 C not on critical path setc R8(%rbx) C save carry out mov %r12, %r13 C new for later shl $63, %r12 C shift new right shr %r14 C shift old left mov %r10, (rp) L(L11): ADDSUB %rax, %r11 C apply previous carry out jc L(c0) C jump if ripple L(rc0): mov 8(up), %r10 or %r14, %r12 ADDSUB %r9, %r8 mov 8(vp), %r9 setc R8(%rax) mov %r11, %r14 shl $63, %r11 shr %r13 mov %r12, 8(rp) L(L10): ADDSUB %rbx, %r8 jc L(c1) L(rc1): mov 16(up), %r12 or %r13, %r11 ADDSUB %r9, %r10 mov 16(vp), %r9 setc R8(%rbx) mov %r8, %r13 shl $63, %r8 shr %r14 mov %r11, 16(rp) L(L01): ADDSUB %rax, %r10 jc L(c2) L(rc2): mov 24(up), %r11 or %r14, %r8 ADDSUB %r9, %r12 lea 32(up), up mov 24(vp), %r9 lea 32(vp), vp setc R8(%rax) mov %r10, %r14 shl $63, %r10 shr %r13 mov %r8, 24(rp) lea 32(rp), rp L(L00): ADDSUB %rbx, %r12 jc L(c3) L(rc3): sub $4, n ja L(top) L(end): or %r13, %r10 ADDSUB %r9, %r11 setc R8(%rbx) mov %r12, %r13 shl $63, %r12 shr %r14 mov %r10, (rp) L(cj3): ADDSUB %rax, %r11 jnc 1f mov $1, R8(%rbx) 1: or %r14, %r12 mov %r11, %r14 shl $63, %r11 shr %r13 mov %r12, 8(rp) L(cj2): or %r13, %r11 shl $63, %rbx shr %r14 mov %r11, 16(rp) L(cj1): or %r14, %rbx mov %rbx, 24(rp) mov R32(%r15), R32(%rax) and $1, R32(%rax) pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx FUNC_EXIT() ret L(c3): mov $1, R8(%rax) jmp L(rc3) EPILOGUE()