head 1.1; branch 1.1.1; access; symbols netbsd-11-0-RC5:1.1.1.2 netbsd-11-0-RC4:1.1.1.2 netbsd-11-0-RC3:1.1.1.2 netbsd-11-0-RC2:1.1.1.2 netbsd-11-0-RC1:1.1.1.2 perseant-exfatfs-base-20250801:1.1.1.2 netbsd-11:1.1.1.2.0.18 netbsd-11-base:1.1.1.2 netbsd-10-1-RELEASE:1.1.1.2 perseant-exfatfs-base-20240630:1.1.1.2 perseant-exfatfs:1.1.1.2.0.16 perseant-exfatfs-base:1.1.1.2 netbsd-8-3-RELEASE:1.1.1.1 netbsd-9-4-RELEASE:1.1.1.2 netbsd-10-0-RELEASE:1.1.1.2 netbsd-10-0-RC6:1.1.1.2 netbsd-10-0-RC5:1.1.1.2 netbsd-10-0-RC4:1.1.1.2 netbsd-10-0-RC3:1.1.1.2 netbsd-10-0-RC2:1.1.1.2 netbsd-10-0-RC1:1.1.1.2 netbsd-10:1.1.1.2.0.14 netbsd-10-base:1.1.1.2 netbsd-9-3-RELEASE:1.1.1.2 gmp-6-2-1:1.1.1.2 cjep_sun2x-base1:1.1.1.2 cjep_sun2x:1.1.1.2.0.12 cjep_sun2x-base:1.1.1.2 cjep_staticlib_x-base1:1.1.1.2 netbsd-9-2-RELEASE:1.1.1.2 cjep_staticlib_x:1.1.1.2.0.10 cjep_staticlib_x-base:1.1.1.2 netbsd-9-1-RELEASE:1.1.1.2 gmp-6-2-0:1.1.1.2 phil-wifi-20200421:1.1.1.2 phil-wifi-20200411:1.1.1.2 is-mlppp:1.1.1.2.0.8 is-mlppp-base:1.1.1.2 phil-wifi-20200406:1.1.1.2 netbsd-8-2-RELEASE:1.1.1.1 netbsd-9-0-RELEASE:1.1.1.2 netbsd-9-0-RC2:1.1.1.2 netbsd-9-0-RC1:1.1.1.2 phil-wifi-20191119:1.1.1.2 netbsd-9:1.1.1.2.0.6 netbsd-9-base:1.1.1.2 phil-wifi-20190609:1.1.1.2 netbsd-8-1-RELEASE:1.1.1.1 netbsd-8-1-RC1:1.1.1.1 pgoyette-compat-merge-20190127:1.1.1.2 pgoyette-compat-20190127:1.1.1.2 pgoyette-compat-20190118:1.1.1.2 pgoyette-compat-1226:1.1.1.2 pgoyette-compat-1126:1.1.1.2 pgoyette-compat-1020:1.1.1.2 pgoyette-compat-0930:1.1.1.2 pgoyette-compat-0906:1.1.1.2 netbsd-7-2-RELEASE:1.1.1.1 pgoyette-compat-0728:1.1.1.2 netbsd-8-0-RELEASE:1.1.1.1 phil-wifi:1.1.1.2.0.4 phil-wifi-base:1.1.1.2 pgoyette-compat-0625:1.1.1.2 netbsd-8-0-RC2:1.1.1.1 pgoyette-compat-0521:1.1.1.2 pgoyette-compat-0502:1.1.1.2 pgoyette-compat-0422:1.1.1.2 netbsd-8-0-RC1:1.1.1.1 pgoyette-compat-0415:1.1.1.2 pgoyette-compat-0407:1.1.1.2 pgoyette-compat-0330:1.1.1.2 pgoyette-compat-0322:1.1.1.2 pgoyette-compat-0315:1.1.1.2 netbsd-7-1-2-RELEASE:1.1.1.1 pgoyette-compat:1.1.1.2.0.2 pgoyette-compat-base:1.1.1.2 netbsd-7-1-1-RELEASE:1.1.1.1 matt-nb8-mediatek:1.1.1.1.0.38 matt-nb8-mediatek-base:1.1.1.1 gmp-6-1-2:1.1.1.2 perseant-stdc-iso10646:1.1.1.1.0.36 perseant-stdc-iso10646-base:1.1.1.1 netbsd-8:1.1.1.1.0.34 netbsd-8-base:1.1.1.1 prg-localcount2-base3:1.1.1.1 prg-localcount2-base2:1.1.1.1 prg-localcount2-base1:1.1.1.1 prg-localcount2:1.1.1.1.0.32 prg-localcount2-base:1.1.1.1 pgoyette-localcount-20170426:1.1.1.1 bouyer-socketcan-base1:1.1.1.1 pgoyette-localcount-20170320:1.1.1.1 netbsd-7-1:1.1.1.1.0.30 netbsd-7-1-RELEASE:1.1.1.1 netbsd-7-1-RC2:1.1.1.1 netbsd-7-nhusb-base-20170116:1.1.1.1 bouyer-socketcan:1.1.1.1.0.28 bouyer-socketcan-base:1.1.1.1 pgoyette-localcount-20170107:1.1.1.1 netbsd-7-1-RC1:1.1.1.1 pgoyette-localcount-20161104:1.1.1.1 netbsd-7-0-2-RELEASE:1.1.1.1 localcount-20160914:1.1.1.1 netbsd-7-nhusb:1.1.1.1.0.26 netbsd-7-nhusb-base:1.1.1.1 pgoyette-localcount-20160806:1.1.1.1 pgoyette-localcount-20160726:1.1.1.1 pgoyette-localcount:1.1.1.1.0.24 pgoyette-localcount-base:1.1.1.1 netbsd-7-0-1-RELEASE:1.1.1.1 netbsd-7-0:1.1.1.1.0.22 netbsd-7-0-RELEASE:1.1.1.1 netbsd-7-0-RC3:1.1.1.1 netbsd-7-0-RC2:1.1.1.1 netbsd-7-0-RC1:1.1.1.1 netbsd-6-0-6-RELEASE:1.1.1.1 netbsd-6-1-5-RELEASE:1.1.1.1 netbsd-7:1.1.1.1.0.20 netbsd-7-base:1.1.1.1 yamt-pagecache-base9:1.1.1.1 yamt-pagecache-tag8:1.1.1.1 netbsd-6-1-4-RELEASE:1.1.1.1 netbsd-6-0-5-RELEASE:1.1.1.1 tls-earlyentropy:1.1.1.1.0.18 tls-earlyentropy-base:1.1.1.1 riastradh-xf86-video-intel-2-7-1-pre-2-21-15:1.1.1.1 riastradh-drm2-base3:1.1.1.1 netbsd-6-1-3-RELEASE:1.1.1.1 netbsd-6-0-4-RELEASE:1.1.1.1 gmp-5-1-3:1.1.1.1 netbsd-6-1-2-RELEASE:1.1.1.1 netbsd-6-0-3-RELEASE:1.1.1.1 netbsd-6-1-1-RELEASE:1.1.1.1 riastradh-drm2-base2:1.1.1.1 riastradh-drm2-base1:1.1.1.1 riastradh-drm2:1.1.1.1.0.12 riastradh-drm2-base:1.1.1.1 netbsd-6-1:1.1.1.1.0.16 netbsd-6-0-2-RELEASE:1.1.1.1 netbsd-6-1-RELEASE:1.1.1.1 netbsd-6-1-RC4:1.1.1.1 netbsd-6-1-RC3:1.1.1.1 agc-symver:1.1.1.1.0.14 agc-symver-base:1.1.1.1 netbsd-6-1-RC2:1.1.1.1 netbsd-6-1-RC1:1.1.1.1 yamt-pagecache-base8:1.1.1.1 netbsd-6-0-1-RELEASE:1.1.1.1 yamt-pagecache-base7:1.1.1.1 matt-nb6-plus-nbase:1.1.1.1 yamt-pagecache-base6:1.1.1.1 netbsd-6-0:1.1.1.1.0.10 netbsd-6-0-RELEASE:1.1.1.1 netbsd-6-0-RC2:1.1.1.1 tls-maxphys:1.1.1.1.0.8 tls-maxphys-base:1.1.1.1 matt-nb6-plus:1.1.1.1.0.6 matt-nb6-plus-base:1.1.1.1 netbsd-6-0-RC1:1.1.1.1 yamt-pagecache-base5:1.1.1.1 yamt-pagecache-base4:1.1.1.1 netbsd-6:1.1.1.1.0.4 netbsd-6-base:1.1.1.1 yamt-pagecache-base3:1.1.1.1 yamt-pagecache-base2:1.1.1.1 yamt-pagecache:1.1.1.1.0.2 yamt-pagecache-base:1.1.1.1 gmp-5-0-2:1.1.1.1 gmp:1.1.1; locks; strict; comment @;; @; 1.1 date 2011.06.20.05.54.43; author mrg; state Exp; branches 1.1.1.1; next ; 1.1.1.1 date 2011.06.20.05.54.43; author mrg; state Exp; branches; next 1.1.1.2; 1.1.1.2 date 2017.08.22.09.40.49; author mrg; state Exp; branches; next ; commitid W5kmAIk8hwVpSb4A; desc @@ 1.1 log @Initial revision @ text @dnl AMD K6 mpn_divexact_1 -- mpn by limb exact division. dnl Copyright 2000, 2001, 2002, 2007 Free Software Foundation, Inc. dnl 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 dnl modify it under the terms of the GNU Lesser General Public License as dnl published by the Free Software Foundation; either version 3 of the dnl License, or (at your option) any later version. dnl dnl The GNU MP Library is distributed in the hope that it will be useful, dnl but WITHOUT ANY WARRANTY; without even the implied warranty of dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU dnl Lesser General Public License for more details. dnl dnl You should have received a copy of the GNU Lesser General Public License dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/. include(`../config.m4') C divisor C odd even C K6: 10.0 12.0 cycles/limb C K6-2: 10.0 11.5 C void mpn_divexact_1 (mp_ptr dst, mp_srcptr src, mp_size_t size, C mp_limb_t divisor); C C A simple divl is used for size==1. This is about 10 cycles faster for an C odd divisor or 20 cycles for an even divisor. C C The loops are quite sensitive to code alignment, speeds should be C rechecked (odd and even divisor, pic and non-pic) if contemplating C changing anything. defframe(PARAM_DIVISOR,16) defframe(PARAM_SIZE, 12) defframe(PARAM_SRC, 8) defframe(PARAM_DST, 4) dnl re-use parameter space define(VAR_INVERSE,`PARAM_DST') TEXT ALIGN(32) PROLOGUE(mpn_divexact_1) deflit(`FRAME',0) movl PARAM_SIZE, %ecx movl PARAM_SRC, %eax xorl %edx, %edx cmpl $1, %ecx jnz L(two_or_more) movl (%eax), %eax divl PARAM_DIVISOR movl PARAM_DST, %ecx movl %eax, (%ecx) ret L(two_or_more): movl PARAM_DIVISOR, %eax pushl %ebx FRAME_pushl() movl PARAM_SRC, %ebx pushl %ebp FRAME_pushl() L(strip_twos): shrl %eax incl %edx C will get shift+1 jnc L(strip_twos) pushl %esi FRAME_pushl() leal 1(%eax,%eax), %esi C d without twos andl $127, %eax C d/2, 7 bits ifdef(`PIC',` LEA( binvert_limb_table, %ebp) Zdisp( movzbl, 0,(%eax,%ebp), %eax) ',` movzbl binvert_limb_table(%eax), %eax C inv 8 bits ') pushl %edi FRAME_pushl() leal (%eax,%eax), %ebp C 2*inv imull %eax, %eax C inv*inv movl PARAM_DST, %edi imull %esi, %eax C inv*inv*d subl %eax, %ebp C inv = 2*inv - inv*inv*d leal (%ebp,%ebp), %eax C 2*inv imull %ebp, %ebp C inv*inv movl %esi, PARAM_DIVISOR C d without twos leal (%ebx,%ecx,4), %ebx C src end imull %esi, %ebp C inv*inv*d leal (%edi,%ecx,4), %edi C dst end negl %ecx C -size subl %ebp, %eax C inv = 2*inv - inv*inv*d subl $1, %edx C shift amount, and clear carry ASSERT(e,` C expect d*inv == 1 mod 2^GMP_LIMB_BITS pushl %eax FRAME_pushl() imull PARAM_DIVISOR, %eax cmpl $1, %eax popl %eax FRAME_popl()') movl %eax, VAR_INVERSE jnz L(even) movl (%ebx,%ecx,4), %esi C src low limb jmp L(odd_entry) ALIGN(16) nop C code alignment L(odd_top): C eax scratch C ebx src end C ecx counter, limbs, negative C edx inverse C esi next limb, adjusted for carry C edi dst end C ebp carry bit, 0 or -1 imull %edx, %esi movl PARAM_DIVISOR, %eax movl %esi, -4(%edi,%ecx,4) mull %esi C carry limb in edx subl %ebp, %edx C apply carry bit movl (%ebx,%ecx,4), %esi L(odd_entry): subl %edx, %esi C apply carry limb movl VAR_INVERSE, %edx sbbl %ebp, %ebp C 0 or -1 incl %ecx jnz L(odd_top) imull %edx, %esi movl %esi, -4(%edi,%ecx,4) popl %edi popl %esi popl %ebp popl %ebx ret L(even): C eax C ebx src end C ecx -size C edx twos C esi C edi dst end C ebp xorl %ebp, %ebp Zdisp( movq, 0,(%ebx,%ecx,4), %mm0) C src[0,1] movd %edx, %mm7 movl VAR_INVERSE, %edx addl $2, %ecx psrlq %mm7, %mm0 movd %mm0, %esi jz L(even_two) C if only two limbs C Out-of-order execution is good enough to hide the load/rshift/movd C latency. Having imul at the top of the loop gives 11.5 c/l instead of 12, C on K6-2. In fact there's only 11 of decode, but nothing running at 11 has C been found. Maybe the fact every second movq is unaligned costs the extra C 0.5. L(even_top): C eax scratch C ebx src end C ecx counter, limbs, negative C edx inverse C esi next limb, adjusted for carry C edi dst end C ebp carry bit, 0 or -1 C C mm0 scratch, source limbs C mm7 twos imull %edx, %esi movl %esi, -8(%edi,%ecx,4) movl PARAM_DIVISOR, %eax mull %esi C carry limb in edx movq -4(%ebx,%ecx,4), %mm0 psrlq %mm7, %mm0 movd %mm0, %esi subl %ebp, %edx C apply carry bit subl %edx, %esi C apply carry limb movl VAR_INVERSE, %edx sbbl %ebp, %ebp C 0 or -1 incl %ecx jnz L(even_top) L(even_two): movd -4(%ebx), %mm0 C src high limb psrlq %mm7, %mm0 imull %edx, %esi movl %esi, -8(%edi) movl PARAM_DIVISOR, %eax mull %esi C carry limb in edx movd %mm0, %esi subl %ebp, %edx C apply carry bit movl VAR_INVERSE, %eax subl %edx, %esi C apply carry limb imull %eax, %esi movl %esi, -4(%edi) popl %edi popl %esi popl %ebp popl %ebx emms_or_femms ret EPILOGUE() @ 1.1.1.1 log @initial import of GMP 5.0.2. GNU MP is a library for arbitrary precision arithmetic, operating on signed integers, rational numbers, and floating point numbers. It has a rich set of functions, and the functions have a regular interface. GMP is necessary for GCC >= 4.2. @ text @@ 1.1.1.2 log @initial import of GMP 6.1.2. main changes from 5.1.3 below. notes: - support for thumb-less ARM chips was in our port of 5.1.3, but a similar method has been provided upstream now - someone should look at the AVX failure reports, and fix them Changes between GMP version 6.1.0 and 6.1.1 FEATURES * Work around faulty cpuid on some recent Intel chips (this allows GMP to run on Skylake Pentiums). * Support thumb-less ARM chips. Changes between GMP version 6.0.* and 6.1.0 BUGS FIXED * The public function mpn_com is now correctly declared in gmp.h. * Healed possible failures of mpn_sec_sqr for non-cryptographic sizes for some obsolete CPUs. * Various problems related to precision for mpf have been fixed. * Fixed ABI incompatible stack alignment in calls from assembly code. * Fixed PIC bug in popcount affecting Intel processors using the 32-bit ABI. SPEEDUPS * Speedup for Intel Broadwell and Skylake through assembly code making use of new ADX instructions. * Square root is now faster when the remainder is not needed. Also the speed to compute the k-th root improved, for small sizes. FEATURES * New C++ functions gcd and lcm for mpz_class. * New public mpn functions mpn_divexact_1, mpn_zero_p, and mpn_cnd_swap. * New public mpq_cmp_z function, to efficiently compare rationals with integers. * Support for more 32-bit arm processors. * Support for AVX-less modern x86 CPUs. (Such support might be missing either because the CPU vendor chose to disable AVX, or because the running kernel lacks AVX context switch support.) * Support for NetBSD under Xen; we switch off AVX unconditionally under NetBSD since a bug in NetBSD makes AVX fail under Xen. MISC * Tuned values for FFT multiplications are provided for larger number on many platforms. Changes between GMP version 5.1.* and 6.0.0 BUGS FIXED * The function mpz_invert now considers any number invertible in Z/1Z. * The mpn multiply code now handles operands of more than 2^31 limbs correctly. (Note however that the mpz code is limited to 2^32 bits on 32-bit hosts and 2^37 bits on 64-bit hosts.) SPEEDUPS * Plain division of large operands is faster and more monotonous in operand size. * Major speedup for ARM, in particular ARM Cortex-A15, thanks to improved assembly. * Speedup for Intel Sandy Bridge, Ivy Bridge, Haswell, thanks to rewritten and vastly expanded assembly support. Speedup also for the older Core 2 and Nehalem. * Faster mixed arithmetic between mpq_class and double. FEATURES * Support for new Intel and AMD CPUs. * New public functions mpn_sec_mul and mpn_sec_sqr, implementing side-channel silent multiplication and squaring. * New public functions mpn_sec_div_qr and mpn_sec_div_r, implementing side-channel silent division. * New public functions mpn_cnd_add_n and mpn_cnd_sub_n. Side-channel silent conditional addition and subtraction. * New public function mpn_sec_powm, implementing side-channel silent modexp. * New public function mpn_sec_invert, implementing side-channel silent modular inversion. * Better support for applications which use the mpz_t type, but nevertheless need to call some of the lower-level mpn functions. See the documentation for mpz_limbs_read and related functions. @ text @d3 2 a4 2 dnl Copyright 2000-2002, 2007 Free Software Foundation, Inc. d7 4 a10 14 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. d12 4 a15 4 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. d17 2 a18 3 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/. a270 1 ASM_END() @