From a89a14ef5da44684a16b204e7a70460cc8c4922a Mon Sep 17 00:00:00 2001 From: Thomas Voss Date: Fri, 21 Jun 2024 23:36:36 +0200 Subject: Basic constant folding implementation --- vendor/gmp-6.3.0/mpn/m88k/README | 61 ++++++ vendor/gmp-6.3.0/mpn/m88k/add_n.s | 113 +++++++++++ vendor/gmp-6.3.0/mpn/m88k/mc88110/add_n.S | 209 ++++++++++++++++++++ vendor/gmp-6.3.0/mpn/m88k/mc88110/addmul_1.s | 70 +++++++ vendor/gmp-6.3.0/mpn/m88k/mc88110/mul_1.s | 68 +++++++ vendor/gmp-6.3.0/mpn/m88k/mc88110/sub_n.S | 285 +++++++++++++++++++++++++++ vendor/gmp-6.3.0/mpn/m88k/mul_1.s | 136 +++++++++++++ vendor/gmp-6.3.0/mpn/m88k/sub_n.s | 115 +++++++++++ 8 files changed, 1057 insertions(+) create mode 100644 vendor/gmp-6.3.0/mpn/m88k/README create mode 100644 vendor/gmp-6.3.0/mpn/m88k/add_n.s create mode 100644 vendor/gmp-6.3.0/mpn/m88k/mc88110/add_n.S create mode 100644 vendor/gmp-6.3.0/mpn/m88k/mc88110/addmul_1.s create mode 100644 vendor/gmp-6.3.0/mpn/m88k/mc88110/mul_1.s create mode 100644 vendor/gmp-6.3.0/mpn/m88k/mc88110/sub_n.S create mode 100644 vendor/gmp-6.3.0/mpn/m88k/mul_1.s create mode 100644 vendor/gmp-6.3.0/mpn/m88k/sub_n.s (limited to 'vendor/gmp-6.3.0/mpn/m88k') diff --git a/vendor/gmp-6.3.0/mpn/m88k/README b/vendor/gmp-6.3.0/mpn/m88k/README new file mode 100644 index 0000000..1b51e83 --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/README @@ -0,0 +1,61 @@ +Copyright 2003 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 either: + + * the GNU Lesser General Public License as published by the Free + Software Foundation; either version 3 of the License, or (at your + option) any later version. + +or + + * the GNU General Public License as published by the Free Software + Foundation; either version 2 of the License, or (at your option) any + later version. + +or both in parallel, as here. + +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 General Public License +for more details. + +You should have received copies of the GNU General Public License and the +GNU Lesser General Public License along with the GNU MP Library. If not, +see https://www.gnu.org/licenses/. + + + + + + M88K MPN SUBROUTINES + +This directory contains mpn functions for various m88k family chips. + +CODE ORGANIZATION + + m88k m88000, m88100 + m88k/mc88110 m88110 + +STATUS + +The code herein is old and poorly maintained. + +* The .s files assume the system uses a "_" underscore prefix, which + should be controlled by configure. + +* The mc88110/*.S files are using the defunct "sysdep.h" configuration + scheme and won't compile. + +Conversion to the current m4 .asm style wouldn't be difficult. + + + + +---------------- +Local variables: +mode: text +fill-column: 76 +End: diff --git a/vendor/gmp-6.3.0/mpn/m88k/add_n.s b/vendor/gmp-6.3.0/mpn/m88k/add_n.s new file mode 100644 index 0000000..dbdb22f --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/add_n.s @@ -0,0 +1,113 @@ +; mc88100 mpn_add_n -- Add two limb vectors of the same length > 0 and store +; sum in a third limb vector. + +; Copyright 1992, 1994, 1995, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +; res_ptr r2 +; s1_ptr r3 +; s2_ptr r4 +; size r5 + +; This code has been optimized to run one instruction per clock, avoiding +; load stalls and writeback contention. As a result, the instruction +; order is not always natural. + +; The speed is about 4.6 clocks/limb + 18 clocks/limb-vector on an 88100, +; but on the 88110, it seems to run much slower, 6.6 clocks/limb. + + text + align 16 + global ___gmpn_add_n +___gmpn_add_n: + ld r6,r3,0 ; read first limb from s1_ptr + extu r10,r5,3 + ld r7,r4,0 ; read first limb from s2_ptr + + subu.co r5,r0,r5 ; (clear carry as side effect) + mak r5,r5,3<4> + bcnd eq0,r5,Lzero + + or r12,r0,lo16(Lbase) + or.u r12,r12,hi16(Lbase) + addu r12,r12,r5 ; r12 is address for entering in loop + + extu r5,r5,2 ; divide by 4 + subu r2,r2,r5 ; adjust res_ptr + subu r3,r3,r5 ; adjust s1_ptr + subu r4,r4,r5 ; adjust s2_ptr + + or r8,r6,r0 + + jmp.n r12 + or r9,r7,r0 + +Loop: addu r3,r3,32 + st r8,r2,28 + addu r4,r4,32 + ld r6,r3,0 + addu r2,r2,32 + ld r7,r4,0 +Lzero: subu r10,r10,1 ; add 0 + 8r limbs (adj loop cnt) +Lbase: ld r8,r3,4 + addu.cio r6,r6,r7 + ld r9,r4,4 + st r6,r2,0 + ld r6,r3,8 ; add 7 + 8r limbs + addu.cio r8,r8,r9 + ld r7,r4,8 + st r8,r2,4 + ld r8,r3,12 ; add 6 + 8r limbs + addu.cio r6,r6,r7 + ld r9,r4,12 + st r6,r2,8 + ld r6,r3,16 ; add 5 + 8r limbs + addu.cio r8,r8,r9 + ld r7,r4,16 + st r8,r2,12 + ld r8,r3,20 ; add 4 + 8r limbs + addu.cio r6,r6,r7 + ld r9,r4,20 + st r6,r2,16 + ld r6,r3,24 ; add 3 + 8r limbs + addu.cio r8,r8,r9 + ld r7,r4,24 + st r8,r2,20 + ld r8,r3,28 ; add 2 + 8r limbs + addu.cio r6,r6,r7 + ld r9,r4,28 + st r6,r2,24 + bcnd.n ne0,r10,Loop ; add 1 + 8r limbs + addu.cio r8,r8,r9 + + st r8,r2,28 ; store most significant limb + + jmp.n r1 + addu.ci r2,r0,r0 ; return carry-out from most sign. limb diff --git a/vendor/gmp-6.3.0/mpn/m88k/mc88110/add_n.S b/vendor/gmp-6.3.0/mpn/m88k/mc88110/add_n.S new file mode 100644 index 0000000..c3b12b3 --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/mc88110/add_n.S @@ -0,0 +1,209 @@ +; mc88110 __gmpn_add_n -- Add two limb vectors of the same length > 0 and store +; sum in a third limb vector. + +; Copyright 1995, 1996, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +#define res_ptr r2 +#define s1_ptr r3 +#define s2_ptr r4 +#define size r5 + +#include "sysdep.h" + + text + align 16 + global C_SYMBOL_NAME(__gmpn_add_n) +C_SYMBOL_NAME(__gmpn_add_n): + addu.co r0,r0,r0 ; clear cy flag + xor r12,s2_ptr,res_ptr + bb1 2,r12,L1 +; ** V1a ** +L0: bb0 2,res_ptr,L_v1 ; branch if res_ptr is aligned? +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld r10,s1_ptr,0 + addu s1_ptr,s1_ptr,4 + ld r8,s2_ptr,0 + addu s2_ptr,s2_ptr,4 + subu size,size,1 + addu.co r6,r10,r8 + st r6,res_ptr,0 + addu res_ptr,res_ptr,4 +L_v1: cmp r12,size,2 + bb1 lt,r12,Lend2 + + ld r10,s1_ptr,0 + ld r12,s1_ptr,4 + ld.d r8,s2_ptr,0 + subu size,size,10 + bcnd lt0,size,Lfin1 +/* Add blocks of 8 limbs until less than 8 limbs remain */ + align 8 +Loop1: subu size,size,8 + addu.cio r6,r10,r8 + ld r10,s1_ptr,8 + addu.cio r7,r12,r9 + ld r12,s1_ptr,12 + ld.d r8,s2_ptr,8 + st.d r6,res_ptr,0 + addu.cio r6,r10,r8 + ld r10,s1_ptr,16 + addu.cio r7,r12,r9 + ld r12,s1_ptr,20 + ld.d r8,s2_ptr,16 + st.d r6,res_ptr,8 + addu.cio r6,r10,r8 + ld r10,s1_ptr,24 + addu.cio r7,r12,r9 + ld r12,s1_ptr,28 + ld.d r8,s2_ptr,24 + st.d r6,res_ptr,16 + addu.cio r6,r10,r8 + ld r10,s1_ptr,32 + addu.cio r7,r12,r9 + ld r12,s1_ptr,36 + addu s1_ptr,s1_ptr,32 + ld.d r8,s2_ptr,32 + addu s2_ptr,s2_ptr,32 + st.d r6,res_ptr,24 + addu res_ptr,res_ptr,32 + bcnd ge0,size,Loop1 + +Lfin1: addu size,size,8-2 + bcnd lt0,size,Lend1 +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: addu.cio r6,r10,r8 + ld r10,s1_ptr,8 + addu.cio r7,r12,r9 + ld r12,s1_ptr,12 + ld.d r8,s2_ptr,8 + st.d r6,res_ptr,0 + subu size,size,2 + addu s1_ptr,s1_ptr,8 + addu s2_ptr,s2_ptr,8 + addu res_ptr,res_ptr,8 + bcnd ge0,size,Loope1 +Lend1: addu.cio r6,r10,r8 + addu.cio r7,r12,r9 + st.d r6,res_ptr,0 + + bb0 0,size,Lret1 +/* Add last limb */ + ld r10,s1_ptr,8 + ld r8,s2_ptr,8 + addu.cio r6,r10,r8 + st r6,res_ptr,8 + +Lret1: jmp.n r1 + addu.ci r2,r0,r0 ; return carry-out from most sign. limb + +L1: xor r12,s1_ptr,res_ptr + bb1 2,r12,L2 +; ** V1b ** + or r12,r0,s2_ptr + or s2_ptr,r0,s1_ptr + or s1_ptr,r0,r12 + br L0 + +; ** 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 r12,size,1 + bb1 eq,r12,Ljone + bb0 2,s1_ptr,L_v2 ; branch if s1_ptr is aligned +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld r10,s1_ptr,0 + addu s1_ptr,s1_ptr,4 + ld r8,s2_ptr,0 + addu s2_ptr,s2_ptr,4 + subu size,size,1 + addu.co r6,r10,r8 + st r6,res_ptr,0 + addu res_ptr,res_ptr,4 + +L_v2: subu size,size,8 + bcnd lt0,size,Lfin2 +/* Add blocks of 8 limbs until less than 8 limbs remain */ + align 8 +Loop2: subu size,size,8 + ld.d r8,s1_ptr,0 + ld.d r6,s2_ptr,0 + addu.cio r8,r8,r6 + st r8,res_ptr,0 + addu.cio r9,r9,r7 + st r9,res_ptr,4 + ld.d r8,s1_ptr,8 + ld.d r6,s2_ptr,8 + addu.cio r8,r8,r6 + st r8,res_ptr,8 + addu.cio r9,r9,r7 + st r9,res_ptr,12 + ld.d r8,s1_ptr,16 + ld.d r6,s2_ptr,16 + addu.cio r8,r8,r6 + st r8,res_ptr,16 + addu.cio r9,r9,r7 + st r9,res_ptr,20 + ld.d r8,s1_ptr,24 + ld.d r6,s2_ptr,24 + addu.cio r8,r8,r6 + st r8,res_ptr,24 + addu.cio r9,r9,r7 + st r9,res_ptr,28 + addu s1_ptr,s1_ptr,32 + addu s2_ptr,s2_ptr,32 + addu res_ptr,res_ptr,32 + bcnd ge0,size,Loop2 + +Lfin2: addu size,size,8-2 + bcnd lt0,size,Lend2 +Loope2: ld.d r8,s1_ptr,0 + ld.d r6,s2_ptr,0 + addu.cio r8,r8,r6 + st r8,res_ptr,0 + addu.cio r9,r9,r7 + st r9,res_ptr,4 + subu size,size,2 + addu s1_ptr,s1_ptr,8 + addu s2_ptr,s2_ptr,8 + addu res_ptr,res_ptr,8 + bcnd ge0,size,Loope2 +Lend2: bb0 0,size,Lret2 +/* Add last limb */ +Ljone: ld r10,s1_ptr,0 + ld r8,s2_ptr,0 + addu.cio r6,r10,r8 + st r6,res_ptr,0 + +Lret2: jmp.n r1 + addu.ci r2,r0,r0 ; return carry-out from most sign. limb diff --git a/vendor/gmp-6.3.0/mpn/m88k/mc88110/addmul_1.s b/vendor/gmp-6.3.0/mpn/m88k/mc88110/addmul_1.s new file mode 100644 index 0000000..321221f --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/mc88110/addmul_1.s @@ -0,0 +1,70 @@ +; mc88110 __gmpn_addmul_1 -- Multiply a limb vector with a single limb and +; store the product in a second limb vector. + +; Copyright 1996, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +; res_ptr r2 +; s1_ptr r3 +; size r4 +; s2_limb r5 + + text + align 16 + global ___gmpn_addmul_1 +___gmpn_addmul_1: + lda r3,r3[r4] + lda r8,r2[r4] ; RES_PTR in r8 since r2 is retval + subu r4,r0,r4 + addu.co r2,r0,r0 ; r2 = cy = 0 + + ld r6,r3[r4] + addu r4,r4,1 + subu r8,r8,4 + bcnd.n eq0,r4,Lend + mulu.d r10,r6,r5 + +Loop: ld r7,r8[r4] + ld r6,r3[r4] + addu.cio r9,r11,r2 + addu.ci r2,r10,r0 + addu.co r9,r9,r7 + st r9,r8[r4] + addu r4,r4,1 + mulu.d r10,r6,r5 + bcnd ne0,r4,Loop + +Lend: ld r7,r8,0 + addu.cio r9,r11,r2 + addu.ci r2,r10,r0 + addu.co r9,r9,r7 + st r9,r8,0 + jmp.n r1 + addu.ci r2,r2,r0 diff --git a/vendor/gmp-6.3.0/mpn/m88k/mc88110/mul_1.s b/vendor/gmp-6.3.0/mpn/m88k/mc88110/mul_1.s new file mode 100644 index 0000000..28fd14b --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/mc88110/mul_1.s @@ -0,0 +1,68 @@ +; mc88110 __gmpn_mul_1 -- Multiply a limb vector with a single limb and +; store the product in a second limb vector. + +; Copyright 1992, 1994, 1995, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +; res_ptr r2 +; s1_ptr r3 +; size r4 +; s2_limb r5 + + text + align 16 + global ___gmpn_mul_1 +___gmpn_mul_1: + ; Make S1_PTR and RES_PTR point at the end of their blocks + ; and negate SIZE. + lda r3,r3[r4] + lda r8,r2[r4] ; RES_PTR in r8 since r2 is retval + subu r4,r0,r4 + + addu.co r2,r0,r0 ; r2 = cy = 0 + + ld r6,r3[r4] + addu r4,r4,1 + mulu.d r10,r6,r5 + bcnd.n eq0,r4,Lend + subu r8,r8,8 + +Loop: ld r6,r3[r4] + addu.cio r9,r11,r2 + or r2,r10,r0 ; could be avoided if unrolled + addu r4,r4,1 + mulu.d r10,r6,r5 + bcnd.n ne0,r4,Loop + st r9,r8[r4] + +Lend: addu.cio r9,r11,r2 + st r9,r8,4 + jmp.n r1 + addu.ci r2,r10,r0 diff --git a/vendor/gmp-6.3.0/mpn/m88k/mc88110/sub_n.S b/vendor/gmp-6.3.0/mpn/m88k/mc88110/sub_n.S new file mode 100644 index 0000000..f0a8ecb --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/mc88110/sub_n.S @@ -0,0 +1,285 @@ +; mc88110 __gmpn_sub_n -- Subtract two limb vectors of the same length > 0 and +; store difference in a third limb vector. + +; Copyright 1995, 1996, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +#define res_ptr r2 +#define s1_ptr r3 +#define s2_ptr r4 +#define size r5 + +#include "sysdep.h" + + text + align 16 + global C_SYMBOL_NAME(__gmpn_sub_n) +C_SYMBOL_NAME(__gmpn_sub_n): + subu.co r0,r0,r0 ; set cy flag + xor r12,s2_ptr,res_ptr + bb1 2,r12,L1 +; ** V1a ** +L0: bb0 2,res_ptr,L_v1 ; branch if res_ptr is aligned +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld r10,s1_ptr,0 + addu s1_ptr,s1_ptr,4 + ld r8,s2_ptr,0 + addu s2_ptr,s2_ptr,4 + subu size,size,1 + subu.co r6,r10,r8 + st r6,res_ptr,0 + addu res_ptr,res_ptr,4 +L_v1: cmp r12,size,2 + bb1 lt,r12,Lend2 + + ld r10,s1_ptr,0 + ld r12,s1_ptr,4 + ld.d r8,s2_ptr,0 + subu size,size,10 + bcnd lt0,size,Lfin1 +/* Add blocks of 8 limbs until less than 8 limbs remain */ + align 8 +Loop1: subu size,size,8 + subu.cio r6,r10,r8 + ld r10,s1_ptr,8 + subu.cio r7,r12,r9 + ld r12,s1_ptr,12 + ld.d r8,s2_ptr,8 + st.d r6,res_ptr,0 + subu.cio r6,r10,r8 + ld r10,s1_ptr,16 + subu.cio r7,r12,r9 + ld r12,s1_ptr,20 + ld.d r8,s2_ptr,16 + st.d r6,res_ptr,8 + subu.cio r6,r10,r8 + ld r10,s1_ptr,24 + subu.cio r7,r12,r9 + ld r12,s1_ptr,28 + ld.d r8,s2_ptr,24 + st.d r6,res_ptr,16 + subu.cio r6,r10,r8 + ld r10,s1_ptr,32 + subu.cio r7,r12,r9 + ld r12,s1_ptr,36 + addu s1_ptr,s1_ptr,32 + ld.d r8,s2_ptr,32 + addu s2_ptr,s2_ptr,32 + st.d r6,res_ptr,24 + addu res_ptr,res_ptr,32 + bcnd ge0,size,Loop1 + +Lfin1: addu size,size,8-2 + bcnd lt0,size,Lend1 +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: subu.cio r6,r10,r8 + ld r10,s1_ptr,8 + subu.cio r7,r12,r9 + ld r12,s1_ptr,12 + ld.d r8,s2_ptr,8 + st.d r6,res_ptr,0 + subu size,size,2 + addu s1_ptr,s1_ptr,8 + addu s2_ptr,s2_ptr,8 + addu res_ptr,res_ptr,8 + bcnd ge0,size,Loope1 +Lend1: subu.cio r6,r10,r8 + subu.cio r7,r12,r9 + st.d r6,res_ptr,0 + + bb0 0,size,Lret1 +/* Add last limb */ + ld r10,s1_ptr,8 + ld r8,s2_ptr,8 + subu.cio r6,r10,r8 + st r6,res_ptr,8 + +Lret1: addu.ci r2,r0,r0 ; return carry-out from most sign. limb + jmp.n r1 + xor r2,r2,1 + +L1: xor r12,s1_ptr,res_ptr + bb1 2,r12,L2 +; ** V1b ** + bb0 2,res_ptr,L_v1b ; branch if res_ptr is aligned +/* Add least significant limb separately to align res_ptr and s1_ptr */ + ld r10,s2_ptr,0 + addu s2_ptr,s2_ptr,4 + ld r8,s1_ptr,0 + addu s1_ptr,s1_ptr,4 + subu size,size,1 + subu.co r6,r8,r10 + st r6,res_ptr,0 + addu res_ptr,res_ptr,4 +L_v1b: cmp r12,size,2 + bb1 lt,r12,Lend2 + + ld r10,s2_ptr,0 + ld r12,s2_ptr,4 + ld.d r8,s1_ptr,0 + subu size,size,10 + bcnd lt0,size,Lfin1b +/* Add blocks of 8 limbs until less than 8 limbs remain */ + align 8 +Loop1b: subu size,size,8 + subu.cio r6,r8,r10 + ld r10,s2_ptr,8 + subu.cio r7,r9,r12 + ld r12,s2_ptr,12 + ld.d r8,s1_ptr,8 + st.d r6,res_ptr,0 + subu.cio r6,r8,r10 + ld r10,s2_ptr,16 + subu.cio r7,r9,r12 + ld r12,s2_ptr,20 + ld.d r8,s1_ptr,16 + st.d r6,res_ptr,8 + subu.cio r6,r8,r10 + ld r10,s2_ptr,24 + subu.cio r7,r9,r12 + ld r12,s2_ptr,28 + ld.d r8,s1_ptr,24 + st.d r6,res_ptr,16 + subu.cio r6,r8,r10 + ld r10,s2_ptr,32 + subu.cio r7,r9,r12 + ld r12,s2_ptr,36 + addu s2_ptr,s2_ptr,32 + ld.d r8,s1_ptr,32 + addu s1_ptr,s1_ptr,32 + st.d r6,res_ptr,24 + addu res_ptr,res_ptr,32 + bcnd ge0,size,Loop1b + +Lfin1b: addu size,size,8-2 + bcnd lt0,size,Lend1b +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1b:subu.cio r6,r8,r10 + ld r10,s2_ptr,8 + subu.cio r7,r9,r12 + ld r12,s2_ptr,12 + ld.d r8,s1_ptr,8 + st.d r6,res_ptr,0 + subu size,size,2 + addu s1_ptr,s1_ptr,8 + addu s2_ptr,s2_ptr,8 + addu res_ptr,res_ptr,8 + bcnd ge0,size,Loope1b +Lend1b: subu.cio r6,r8,r10 + subu.cio r7,r9,r12 + st.d r6,res_ptr,0 + + bb0 0,size,Lret1b +/* Add last limb */ + ld r10,s2_ptr,8 + ld r8,s1_ptr,8 + subu.cio r6,r8,r10 + st r6,res_ptr,8 + +Lret1b: addu.ci r2,r0,r0 ; return carry-out from most sign. limb + jmp.n r1 + xor r2,r2,1 + +; ** 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 r12,size,1 + bb1 eq,r12,Ljone + bb0 2,s1_ptr,L_v2 ; branch if s1_ptr is aligned +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld r10,s1_ptr,0 + addu s1_ptr,s1_ptr,4 + ld r8,s2_ptr,0 + addu s2_ptr,s2_ptr,4 + subu size,size,1 + subu.co r6,r10,r8 + st r6,res_ptr,0 + addu res_ptr,res_ptr,4 + +L_v2: subu size,size,8 + bcnd lt0,size,Lfin2 +/* Add blocks of 8 limbs until less than 8 limbs remain */ + align 8 +Loop2: subu size,size,8 + ld.d r8,s1_ptr,0 + ld.d r6,s2_ptr,0 + subu.cio r8,r8,r6 + st r8,res_ptr,0 + subu.cio r9,r9,r7 + st r9,res_ptr,4 + ld.d r8,s1_ptr,8 + ld.d r6,s2_ptr,8 + subu.cio r8,r8,r6 + st r8,res_ptr,8 + subu.cio r9,r9,r7 + st r9,res_ptr,12 + ld.d r8,s1_ptr,16 + ld.d r6,s2_ptr,16 + subu.cio r8,r8,r6 + st r8,res_ptr,16 + subu.cio r9,r9,r7 + st r9,res_ptr,20 + ld.d r8,s1_ptr,24 + ld.d r6,s2_ptr,24 + subu.cio r8,r8,r6 + st r8,res_ptr,24 + subu.cio r9,r9,r7 + st r9,res_ptr,28 + addu s1_ptr,s1_ptr,32 + addu s2_ptr,s2_ptr,32 + addu res_ptr,res_ptr,32 + bcnd ge0,size,Loop2 + +Lfin2: addu size,size,8-2 + bcnd lt0,size,Lend2 +Loope2: ld.d r8,s1_ptr,0 + ld.d r6,s2_ptr,0 + subu.cio r8,r8,r6 + st r8,res_ptr,0 + subu.cio r9,r9,r7 + st r9,res_ptr,4 + subu size,size,2 + addu s1_ptr,s1_ptr,8 + addu s2_ptr,s2_ptr,8 + addu res_ptr,res_ptr,8 + bcnd ge0,size,Loope2 +Lend2: bb0 0,size,Lret2 +/* Add last limb */ +Ljone: ld r10,s1_ptr,0 + ld r8,s2_ptr,0 + subu.cio r6,r10,r8 + st r6,res_ptr,0 + +Lret2: addu.ci r2,r0,r0 ; return carry-out from most sign. limb + jmp.n r1 + xor r2,r2,1 diff --git a/vendor/gmp-6.3.0/mpn/m88k/mul_1.s b/vendor/gmp-6.3.0/mpn/m88k/mul_1.s new file mode 100644 index 0000000..c8abdc0 --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/mul_1.s @@ -0,0 +1,136 @@ +; mc88100 __gmpn_mul_1 -- Multiply a limb vector with a single limb and +; store the product in a second limb vector. + +; Copyright 1992, 1994, 1995, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +; res_ptr r2 +; s1_ptr r3 +; size r4 +; s2_limb r5 + +; Common overhead is about 11 cycles/invocation. + +; The speed for S2_LIMB >= 0x10000 is approximately 21 cycles/limb. (The +; pipeline stalls 2 cycles due to WB contention.) + +; The speed for S2_LIMB < 0x10000 is approximately 16 cycles/limb. (The +; pipeline stalls 2 cycles due to WB contention and 1 cycle due to latency.) + +; To enhance speed: +; 1. Unroll main loop 4-8 times. +; 2. Schedule code to avoid WB contention. It might be tempting to move the +; ld instruction in the loops down to save 2 cycles (less WB contention), +; but that looses because the ultimate value will be read from outside +; the allocated space. But if we handle the ultimate multiplication in +; the tail, we can do this. +; 3. Make the multiplication with less instructions. I think the code for +; (S2_LIMB >= 0x10000) is not minimal. +; With these techniques the (S2_LIMB >= 0x10000) case would run in 17 or +; less cycles/limb; the (S2_LIMB < 0x10000) case would run in 11 +; cycles/limb. (Assuming infinite unrolling.) + + text + align 16 + global ___gmpn_mul_1 +___gmpn_mul_1: + + ; Make S1_PTR and RES_PTR point at the end of their blocks + ; and negate SIZE. + lda r3,r3[r4] + lda r6,r2[r4] ; RES_PTR in r6 since r2 is retval + subu r4,r0,r4 + + addu.co r2,r0,r0 ; r2 = cy = 0 + ld r9,r3[r4] + mask r7,r5,0xffff ; r7 = lo(S2_LIMB) + extu r8,r5,16 ; r8 = hi(S2_LIMB) + bcnd.n eq0,r8,Lsmall ; jump if (hi(S2_LIMB) == 0) + subu r6,r6,4 + +; General code for any value of S2_LIMB. + + ; Make a stack frame and save r25 and r26 + subu r31,r31,16 + st.d r25,r31,8 + + ; Enter the loop in the middle + br.n L1 + addu r4,r4,1 + +Loop: ld r9,r3[r4] + st r26,r6[r4] +; bcnd ne0,r0,0 ; bubble + addu r4,r4,1 +L1: mul r26,r9,r5 ; low word of product mul_1 WB ld + mask r12,r9,0xffff ; r12 = lo(s1_limb) mask_1 + mul r11,r12,r7 ; r11 = prod_0 mul_2 WB mask_1 + mul r10,r12,r8 ; r10 = prod_1a mul_3 + extu r13,r9,16 ; r13 = hi(s1_limb) extu_1 WB mul_1 + mul r12,r13,r7 ; r12 = prod_1b mul_4 WB extu_1 + mul r25,r13,r8 ; r25 = prod_2 mul_5 WB mul_2 + extu r11,r11,16 ; r11 = hi(prod_0) extu_2 WB mul_3 + addu r10,r10,r11 ; addu_1 WB extu_2 +; bcnd ne0,r0,0 ; bubble WB addu_1 + addu.co r10,r10,r12 ; WB mul_4 + mask.u r10,r10,0xffff ; move the 16 most significant bits... + addu.ci r10,r10,r0 ; ...to the low half of the word... + rot r10,r10,16 ; ...and put carry in pos 16. + addu.co r26,r26,r2 ; add old carry limb + bcnd.n ne0,r4,Loop + addu.ci r2,r25,r10 ; compute new carry limb + + st r26,r6[r4] + ld.d r25,r31,8 + jmp.n r1 + addu r31,r31,16 + +; Fast code for S2_LIMB < 0x10000 +Lsmall: + ; Enter the loop in the middle + br.n SL1 + addu r4,r4,1 + +SLoop: ld r9,r3[r4] ; + st r8,r6[r4] ; + addu r4,r4,1 ; +SL1: mul r8,r9,r5 ; low word of product + mask r12,r9,0xffff ; r12 = lo(s1_limb) + extu r13,r9,16 ; r13 = hi(s1_limb) + mul r11,r12,r7 ; r11 = prod_0 + mul r12,r13,r7 ; r12 = prod_1b + addu.cio r8,r8,r2 ; add old carry limb + extu r10,r11,16 ; r11 = hi(prod_0) + addu r10,r10,r12 ; + bcnd.n ne0,r4,SLoop + extu r2,r10,16 ; r2 = new carry limb + + jmp.n r1 + st r8,r6[r4] diff --git a/vendor/gmp-6.3.0/mpn/m88k/sub_n.s b/vendor/gmp-6.3.0/mpn/m88k/sub_n.s new file mode 100644 index 0000000..2bd8f09 --- /dev/null +++ b/vendor/gmp-6.3.0/mpn/m88k/sub_n.s @@ -0,0 +1,115 @@ +; mc88100 mpn_sub_n -- Subtract two limb vectors of the same length > 0 and +; store difference in a third limb vector. + +; Copyright 1992, 1994, 1996, 2000 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 either: +; +; * the GNU Lesser General Public License as published by the Free +; Software Foundation; either version 3 of the License, or (at your +; option) any later version. +; +; or +; +; * the GNU General Public License as published by the Free Software +; Foundation; either version 2 of the License, or (at your option) any +; later version. +; +; or both in parallel, as here. +; +; 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 General Public License +; for more details. +; +; You should have received copies of the GNU General Public License and the +; GNU Lesser General Public License along with the GNU MP Library. If not, +; see https://www.gnu.org/licenses/. + + +; INPUT PARAMETERS +; res_ptr r2 +; s1_ptr r3 +; s2_ptr r4 +; size r5 + +; This code has been optimized to run one instruction per clock, avoiding +; load stalls and writeback contention. As a result, the instruction +; order is not always natural. + +; The speed is about 4.6 clocks/limb + 18 clocks/limb-vector on an 88100, +; but on the 88110, it seems to run much slower, 6.6 clocks/limb. + + text + align 16 + global ___gmpn_sub_n +___gmpn_sub_n: + ld r6,r3,0 ; read first limb from s1_ptr + extu r10,r5,3 + ld r7,r4,0 ; read first limb from s2_ptr + + subu r5,r0,r5 + mak r5,r5,3<4> + bcnd.n eq0,r5,Lzero + subu.co r0,r0,r0 ; initialize carry + + or r12,r0,lo16(Lbase) + or.u r12,r12,hi16(Lbase) + addu r12,r12,r5 ; r12 is address for entering in loop + + extu r5,r5,2 ; divide by 4 + subu r2,r2,r5 ; adjust res_ptr + subu r3,r3,r5 ; adjust s1_ptr + subu r4,r4,r5 ; adjust s2_ptr + + or r8,r6,r0 + + jmp.n r12 + or r9,r7,r0 + +Loop: addu r3,r3,32 + st r8,r2,28 + addu r4,r4,32 + ld r6,r3,0 + addu r2,r2,32 + ld r7,r4,0 +Lzero: subu r10,r10,1 ; subtract 0 + 8r limbs (adj loop cnt) +Lbase: ld r8,r3,4 + subu.cio r6,r6,r7 + ld r9,r4,4 + st r6,r2,0 + ld r6,r3,8 ; subtract 7 + 8r limbs + subu.cio r8,r8,r9 + ld r7,r4,8 + st r8,r2,4 + ld r8,r3,12 ; subtract 6 + 8r limbs + subu.cio r6,r6,r7 + ld r9,r4,12 + st r6,r2,8 + ld r6,r3,16 ; subtract 5 + 8r limbs + subu.cio r8,r8,r9 + ld r7,r4,16 + st r8,r2,12 + ld r8,r3,20 ; subtract 4 + 8r limbs + subu.cio r6,r6,r7 + ld r9,r4,20 + st r6,r2,16 + ld r6,r3,24 ; subtract 3 + 8r limbs + subu.cio r8,r8,r9 + ld r7,r4,24 + st r8,r2,20 + ld r8,r3,28 ; subtract 2 + 8r limbs + subu.cio r6,r6,r7 + ld r9,r4,28 + st r6,r2,24 + bcnd.n ne0,r10,Loop ; subtract 1 + 8r limbs + subu.cio r8,r8,r9 + + st r8,r2,28 ; store most significant limb + + addu.ci r2,r0,r0 ; return carry-out from most sign. limb + jmp.n r1 + xor r2,r2,1 -- cgit v1.2.3