From 11da511c784eca003deb90c23570f0873954e0de Mon Sep 17 00:00:00 2001 From: Duncan Wilkie Date: Sat, 18 Nov 2023 06:11:09 -0600 Subject: Initial commit. --- gmp-6.3.0/mpn/arm64/aors_n.asm | 125 ++++++++++ gmp-6.3.0/mpn/arm64/aorsmul_1.asm | 145 +++++++++++ gmp-6.3.0/mpn/arm64/aorsorrlsh1_n.asm | 43 ++++ gmp-6.3.0/mpn/arm64/aorsorrlsh2_n.asm | 43 ++++ gmp-6.3.0/mpn/arm64/aorsorrlshC_n.asm | 139 +++++++++++ gmp-6.3.0/mpn/arm64/applem1/addaddmul_1msb0.asm | 92 +++++++ gmp-6.3.0/mpn/arm64/applem1/aorsmul_1.asm | 161 ++++++++++++ gmp-6.3.0/mpn/arm64/applem1/gmp-mparam.h | 187 ++++++++++++++ gmp-6.3.0/mpn/arm64/applem1/sqr_basecase.asm | 318 ++++++++++++++++++++++++ gmp-6.3.0/mpn/arm64/arm64-defs.m4 | 53 ++++ gmp-6.3.0/mpn/arm64/bdiv_dbm1c.asm | 111 +++++++++ gmp-6.3.0/mpn/arm64/bdiv_q_1.asm | 122 +++++++++ gmp-6.3.0/mpn/arm64/cnd_aors_n.asm | 129 ++++++++++ gmp-6.3.0/mpn/arm64/com.asm | 92 +++++++ gmp-6.3.0/mpn/arm64/copyd.asm | 85 +++++++ gmp-6.3.0/mpn/arm64/copyi.asm | 82 ++++++ gmp-6.3.0/mpn/arm64/cora53/cnd_aors_n.asm | 99 ++++++++ gmp-6.3.0/mpn/arm64/cora53/gmp-mparam.h | 242 ++++++++++++++++++ gmp-6.3.0/mpn/arm64/cora57/gmp-mparam.h | 188 ++++++++++++++ gmp-6.3.0/mpn/arm64/cora72/gmp-mparam.h | 242 ++++++++++++++++++ gmp-6.3.0/mpn/arm64/cora73/gmp-mparam.h | 225 +++++++++++++++++ gmp-6.3.0/mpn/arm64/darwin.m4 | 50 ++++ gmp-6.3.0/mpn/arm64/divrem_1.asm | 231 +++++++++++++++++ gmp-6.3.0/mpn/arm64/gcd_11.asm | 70 ++++++ gmp-6.3.0/mpn/arm64/gcd_22.asm | 112 +++++++++ gmp-6.3.0/mpn/arm64/gmp-mparam.h | 192 ++++++++++++++ gmp-6.3.0/mpn/arm64/hamdist.asm | 181 ++++++++++++++ gmp-6.3.0/mpn/arm64/invert_limb.asm | 83 +++++++ gmp-6.3.0/mpn/arm64/logops_n.asm | 139 +++++++++++ gmp-6.3.0/mpn/arm64/lshift.asm | 138 ++++++++++ gmp-6.3.0/mpn/arm64/lshiftc.asm | 141 +++++++++++ gmp-6.3.0/mpn/arm64/mod_34lsub1.asm | 124 +++++++++ gmp-6.3.0/mpn/arm64/mul_1.asm | 128 ++++++++++ gmp-6.3.0/mpn/arm64/popcount.asm | 157 ++++++++++++ gmp-6.3.0/mpn/arm64/rsh1aors_n.asm | 168 +++++++++++++ gmp-6.3.0/mpn/arm64/rshift.asm | 136 ++++++++++ gmp-6.3.0/mpn/arm64/sec_tabselect.asm | 122 +++++++++ gmp-6.3.0/mpn/arm64/sqr_diag_addlsh1.asm | 102 ++++++++ gmp-6.3.0/mpn/arm64/xgene1/gmp-mparam.h | 182 ++++++++++++++ 39 files changed, 5379 insertions(+) create mode 100644 gmp-6.3.0/mpn/arm64/aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/aorsmul_1.asm create mode 100644 gmp-6.3.0/mpn/arm64/aorsorrlsh1_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/aorsorrlsh2_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/aorsorrlshC_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/applem1/addaddmul_1msb0.asm create mode 100644 gmp-6.3.0/mpn/arm64/applem1/aorsmul_1.asm create mode 100644 gmp-6.3.0/mpn/arm64/applem1/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/applem1/sqr_basecase.asm create mode 100644 gmp-6.3.0/mpn/arm64/arm64-defs.m4 create mode 100644 gmp-6.3.0/mpn/arm64/bdiv_dbm1c.asm create mode 100644 gmp-6.3.0/mpn/arm64/bdiv_q_1.asm create mode 100644 gmp-6.3.0/mpn/arm64/cnd_aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/com.asm create mode 100644 gmp-6.3.0/mpn/arm64/copyd.asm create mode 100644 gmp-6.3.0/mpn/arm64/copyi.asm create mode 100644 gmp-6.3.0/mpn/arm64/cora53/cnd_aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/cora53/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/cora57/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/cora72/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/cora73/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/darwin.m4 create mode 100644 gmp-6.3.0/mpn/arm64/divrem_1.asm create mode 100644 gmp-6.3.0/mpn/arm64/gcd_11.asm create mode 100644 gmp-6.3.0/mpn/arm64/gcd_22.asm create mode 100644 gmp-6.3.0/mpn/arm64/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm64/hamdist.asm create mode 100644 gmp-6.3.0/mpn/arm64/invert_limb.asm create mode 100644 gmp-6.3.0/mpn/arm64/logops_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/lshift.asm create mode 100644 gmp-6.3.0/mpn/arm64/lshiftc.asm create mode 100644 gmp-6.3.0/mpn/arm64/mod_34lsub1.asm create mode 100644 gmp-6.3.0/mpn/arm64/mul_1.asm create mode 100644 gmp-6.3.0/mpn/arm64/popcount.asm create mode 100644 gmp-6.3.0/mpn/arm64/rsh1aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm64/rshift.asm create mode 100644 gmp-6.3.0/mpn/arm64/sec_tabselect.asm create mode 100644 gmp-6.3.0/mpn/arm64/sqr_diag_addlsh1.asm create mode 100644 gmp-6.3.0/mpn/arm64/xgene1/gmp-mparam.h (limited to 'gmp-6.3.0/mpn/arm64') diff --git a/gmp-6.3.0/mpn/arm64/aors_n.asm b/gmp-6.3.0/mpn/arm64/aors_n.asm new file mode 100644 index 0000000..b4a6da6 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/aors_n.asm @@ -0,0 +1,125 @@ +dnl ARM64 mpn_add_n and mpn_sub_n + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2013, 2017 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 Cortex-A53 2.75-3.25 +C Cortex-A57 1.5 +C X-Gene 2.0 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`vp', `x2') +define(`n', `x3') + +ifdef(`OPERATION_add_n', ` + define(`ADDSUBC', adcs) + define(`CLRCY', `cmn xzr, xzr') + define(`SETCY', `cmp $1, #1') + define(`RETVAL', `cset x0, cs') + define(`func_n', mpn_add_n) + define(`func_nc', mpn_add_nc)') +ifdef(`OPERATION_sub_n', ` + define(`ADDSUBC', sbcs) + define(`CLRCY', `cmp xzr, xzr') + define(`SETCY', `cmp xzr, $1') + define(`RETVAL', `cset x0, cc') + define(`func_n', mpn_sub_n) + define(`func_nc', mpn_sub_nc)') + +MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc) + +ASM_START() +PROLOGUE(func_nc) + SETCY( x4) + b L(ent) +EPILOGUE() +PROLOGUE(func_n) + CLRCY +L(ent): lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x7, [up] + ldr x11, [vp] + ADDSUBC x13, x7, x11 + str x13, [rp],#8 + tbnz n, #1, L(b11) + +L(b01): cbz x17, L(ret) + ldp x4, x5, [up,#8] + ldp x8, x9, [vp,#8] + sub up, up, #8 + sub vp, vp, #8 + b L(mid) + +L(b11): ldp x6, x7, [up,#8] + ldp x10, x11, [vp,#8] + add up, up, #8 + add vp, vp, #8 + cbz x17, L(end) + b L(top) + +L(bx0): tbnz n, #1, L(b10) + +L(b00): ldp x4, x5, [up] + ldp x8, x9, [vp] + sub up, up, #16 + sub vp, vp, #16 + b L(mid) + +L(b10): ldp x6, x7, [up] + ldp x10, x11, [vp] + cbz x17, L(end) + + ALIGN(16) +L(top): ldp x4, x5, [up,#16] + ldp x8, x9, [vp,#16] + ADDSUBC x12, x6, x10 + ADDSUBC x13, x7, x11 + stp x12, x13, [rp],#16 +L(mid): ldp x6, x7, [up,#32]! + ldp x10, x11, [vp,#32]! + ADDSUBC x12, x4, x8 + ADDSUBC x13, x5, x9 + stp x12, x13, [rp],#16 + sub x17, x17, #1 + cbnz x17, L(top) + +L(end): ADDSUBC x12, x6, x10 + ADDSUBC x13, x7, x11 + stp x12, x13, [rp] +L(ret): RETVAL + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/aorsmul_1.asm b/gmp-6.3.0/mpn/arm64/aorsmul_1.asm new file mode 100644 index 0000000..81ec1da --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/aorsmul_1.asm @@ -0,0 +1,145 @@ +dnl ARM64 mpn_addmul_1 and mpn_submul_1 + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2013, 2015, 2017 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 addmul_1 submul_1 +C cycles/limb cycles/limb +C Cortex-A53 9.3-9.8 9.3-9.8 +C Cortex-A55 9.0-9.5 9.3-9.8 +C Cortex-A57 7 7 +C Cortex-A72 +C Cortex-A73 6 6 +C X-Gene 5 5 +C Apple M1 1.75 1.75 + +C NOTES +C * It is possible to keep the carry chain alive between the addition blocks +C and thus avoid csinc, but only for addmul_1. Since that saves no time +C on the tested pipelines, we keep addmul_1 and submul_1 similar. +C * We could separate feed-in into 4 blocks, one for each residue (mod 4). +C That is likely to save a few cycles. + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') +define(`v0', `x3') + +ifdef(`OPERATION_addmul_1', ` + define(`ADDSUB', adds) + define(`ADDSUBC', adcs) + define(`COND', `cc') + define(`func', mpn_addmul_1)') +ifdef(`OPERATION_submul_1', ` + define(`ADDSUB', subs) + define(`ADDSUBC', sbcs) + define(`COND', `cs') + define(`func', mpn_submul_1)') + +MULFUNC_PROLOGUE(mpn_addmul_1 mpn_submul_1) + +PROLOGUE(func) + adds x15, xzr, xzr + + tbz n, #0, L(1) + + ldr x4, [up],#8 + mul x8, x4, v0 + umulh x12, x4, v0 + ldr x4, [rp] + ADDSUB x8, x4, x8 + csinc x15, x12, x12, COND + str x8, [rp],#8 + +L(1): tbz n, #1, L(2) + + ldp x4, x5, [up],#16 + mul x8, x4, v0 + umulh x12, x4, v0 + mul x9, x5, v0 + umulh x13, x5, v0 + adds x8, x8, x15 + adcs x9, x9, x12 + ldp x4, x5, [rp] + adc x15, x13, xzr + ADDSUB x8, x4, x8 + ADDSUBC x9, x5, x9 + csinc x15, x15, x15, COND + stp x8, x9, [rp],#16 + +L(2): lsr n, n, #2 + cbz n, L(le3) + ldp x4, x5, [up],#32 + ldp x6, x7, [up,#-16] + b L(mid) +L(le3): mov x0, x15 + ret + + ALIGN(16) +L(top): ldp x4, x5, [up],#32 + ldp x6, x7, [up,#-16] + ADDSUB x8, x16, x8 + ADDSUBC x9, x17, x9 + stp x8, x9, [rp],#32 + ADDSUBC x10, x12, x10 + ADDSUBC x11, x13, x11 + stp x10, x11, [rp,#-16] + csinc x15, x15, x15, COND +L(mid): sub n, n, #1 + mul x8, x4, v0 + umulh x12, x4, v0 + mul x9, x5, v0 + umulh x13, x5, v0 + adds x8, x8, x15 + mul x10, x6, v0 + umulh x14, x6, v0 + adcs x9, x9, x12 + mul x11, x7, v0 + umulh x15, x7, v0 + adcs x10, x10, x13 + ldp x16, x17, [rp] + adcs x11, x11, x14 + ldp x12, x13, [rp,#16] + adc x15, x15, xzr + cbnz n, L(top) + + ADDSUB x8, x16, x8 + ADDSUBC x9, x17, x9 + ADDSUBC x10, x12, x10 + ADDSUBC x11, x13, x11 + stp x8, x9, [rp] + stp x10, x11, [rp,#16] + csinc x0, x15, x15, COND + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/aorsorrlsh1_n.asm b/gmp-6.3.0/mpn/arm64/aorsorrlsh1_n.asm new file mode 100644 index 0000000..c617a67 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/aorsorrlsh1_n.asm @@ -0,0 +1,43 @@ +dnl ARM64 mpn_addlsh1_n, mpn_sublsh1_n, mpn_rsblsh1_n. + +dnl Copyright 2017 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') + + +define(LSH, 1) +define(RSH, 63) + +ifdef(`OPERATION_addlsh1_n',`define(`DO_add')') +ifdef(`OPERATION_sublsh1_n',`define(`DO_sub')') +ifdef(`OPERATION_rsblsh1_n',`define(`DO_rsb')') + +MULFUNC_PROLOGUE(mpn_addlsh1_n mpn_sublsh1_n mpn_rsblsh1_n) + +include_mpn(`arm64/aorsorrlshC_n.asm') diff --git a/gmp-6.3.0/mpn/arm64/aorsorrlsh2_n.asm b/gmp-6.3.0/mpn/arm64/aorsorrlsh2_n.asm new file mode 100644 index 0000000..852d117 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/aorsorrlsh2_n.asm @@ -0,0 +1,43 @@ +dnl ARM64 mpn_addlsh2_n, mpn_sublsh2_n, mpn_rsblsh2_n. + +dnl Copyright 2017 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') + + +define(LSH, 2) +define(RSH, 62) + +ifdef(`OPERATION_addlsh2_n',`define(`DO_add')') +ifdef(`OPERATION_sublsh2_n',`define(`DO_sub')') +ifdef(`OPERATION_rsblsh2_n',`define(`DO_rsb')') + +MULFUNC_PROLOGUE(mpn_addlsh2_n mpn_sublsh2_n mpn_rsblsh2_n) + +include_mpn(`arm64/aorsorrlshC_n.asm') diff --git a/gmp-6.3.0/mpn/arm64/aorsorrlshC_n.asm b/gmp-6.3.0/mpn/arm64/aorsorrlshC_n.asm new file mode 100644 index 0000000..1718b77 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/aorsorrlshC_n.asm @@ -0,0 +1,139 @@ +dnl ARM64 mpn_addlshC_n, mpn_sublshC_n, mpn_rsblshC_n. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2017 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 Cortex-A53 3.25-3.75 +C Cortex-A57 2.18 +C X-Gene 2.5 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`vp', `x2') +define(`n', `x3') + +ifdef(`DO_add', ` + define(`ADDSUB', `adds $1, $2, $3') + define(`ADDSUBC', `adcs $1, $2, $3') + define(`CLRRCY', `adds $1, xzr, xzr') + define(`RETVAL', `adc x0, $1, xzr') + define(`func_n', mpn_addlsh`'LSH`'_n)') +ifdef(`DO_sub', ` + define(`ADDSUB', `subs $1, $3, $2') + define(`ADDSUBC', `sbcs $1, $3, $2') + define(`CLRRCY', `subs $1, xzr, xzr') + define(`RETVAL', `cinc x0, $1, cc') + define(`func_n', mpn_sublsh`'LSH`'_n)') +ifdef(`DO_rsb', ` + define(`ADDSUB', `subs $1, $2, $3') + define(`ADDSUBC', `sbcs $1, $2, $3') + define(`CLRRCY', `subs $1, xzr, xzr') + define(`RETVAL', `sbc x0, $1, xzr') + define(`func_n', mpn_rsblsh`'LSH`'_n)') + +ASM_START() +PROLOGUE(func_n) + lsr x6, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x5, [up] + tbnz n, #1, L(b11) + +L(b01): ldr x11, [vp] + cbz x6, L(1) + ldp x8, x9, [vp,#8] + lsl x13, x11, #LSH + ADDSUB( x15, x13, x5) + str x15, [rp],#8 + sub up, up, #24 + sub vp, vp, #8 + b L(mid) + +L(1): lsl x13, x11, #LSH + ADDSUB( x15, x13, x5) + str x15, [rp] + lsr x0, x11, RSH + RETVAL( x0, x1) + ret + +L(b11): ldr x9, [vp] + ldp x10, x11, [vp,#8]! + lsl x13, x9, #LSH + ADDSUB( x17, x13, x5) + str x17, [rp],#8 + sub up, up, #8 + cbz x6, L(end) + b L(top) + +L(bx0): tbnz n, #1, L(b10) + +L(b00): CLRRCY( x11) + ldp x8, x9, [vp],#-16 + sub up, up, #32 + b L(mid) + +L(b10): CLRRCY( x9) + ldp x10, x11, [vp] + sub up, up, #16 + cbz x6, L(end) + + ALIGN(16) +L(top): ldp x4, x5, [up,#16] + extr x12, x10, x9, #RSH + ldp x8, x9, [vp,#16] + extr x13, x11, x10, #RSH + ADDSUBC(x14, x12, x4) + ADDSUBC(x15, x13, x5) + stp x14, x15, [rp],#16 +L(mid): ldp x4, x5, [up,#32]! + extr x12, x8, x11, #RSH + ldp x10, x11, [vp,#32]! + extr x13, x9, x8, #RSH + ADDSUBC(x16, x12, x4) + ADDSUBC(x17, x13, x5) + stp x16, x17, [rp],#16 + sub x6, x6, #1 + cbnz x6, L(top) + +L(end): ldp x4, x5, [up,#16] + extr x12, x10, x9, #RSH + extr x13, x11, x10, #RSH + ADDSUBC(x14, x12, x4) + ADDSUBC(x15, x13, x5) + stp x14, x15, [rp] + lsr x0, x11, RSH + RETVAL( x0, x1) + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/applem1/addaddmul_1msb0.asm b/gmp-6.3.0/mpn/arm64/applem1/addaddmul_1msb0.asm new file mode 100644 index 0000000..03cbf97 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/applem1/addaddmul_1msb0.asm @@ -0,0 +1,92 @@ +dnl ARM64 mpn_addaddmul_1msb0, R = Au + Bv, u,v < 2^63. + +dnl Copyright 2021 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 Cortex-A53 +C Cortex-A55 +C Cortex-A57 +C Cortex-A72 +C Cortex-A73 +C X-Gene +C Apple M1 2.0 + +changecom(blah) + +define(`rp', x0) +define(`ap', x1) +define(`bp', x2) +define(`n', x3) +define(`u0', x4) +define(`v0', x5) + +C TODO +C * Use fewer distinct registers, should be trivial. + +PROLOGUE(mpn_addaddmul_1msb0) + lsr x7, n, #1 + adds x6, xzr, xzr + tbz n, #0, L(top) + + ldr x11, [ap], #8 C 0 + ldr x15, [bp], #8 C 0 + mul x10, x11, u0 C 0 + umulh x11, x11, u0 C 1 + mul x14, x15, v0 C 0 + umulh x15, x15, v0 C 1 + adds x10, x10, x14 C 0 + adcs x6, x11, x15 C 1 + str x10, [rp], #8 C 0 + cbz x7, L(end) + +L(top): ldp x11, x13, [ap], #16 C 0 1 + ldp x15, x17, [bp], #16 C 0 1 + mul x10, x11, u0 C 0 + umulh x11, x11, u0 C 1 + mul x14, x15, v0 C 0 + umulh x15, x15, v0 C 1 + adcs x10, x10, x14 C 0 + adc x11, x11, x15 C 1 + adds x10, x10, x6 C 0 + mul x12, x13, u0 C 1 + umulh x13, x13, u0 C 2 + mul x14, x17, v0 C 1 + umulh x17, x17, v0 C 2 + adcs x12, x12, x14 C 1 + adc x6, x13, x17 C 2 + adds x11, x12, x11 C 1 + stp x10, x11, [rp], #16 C 0 1 + sub x7, x7, #1 + cbnz x7, L(top) + +L(end): adc x0, x6, xzr + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/applem1/aorsmul_1.asm b/gmp-6.3.0/mpn/arm64/applem1/aorsmul_1.asm new file mode 100644 index 0000000..aa87c2a --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/applem1/aorsmul_1.asm @@ -0,0 +1,161 @@ +dnl ARM64 mpn_addmul_1 and mpn_submul_1. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2020 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 Cortex-A53 +C Cortex-A55 +C Cortex-A57 +C Cortex-A72 +C Cortex-A73 +C X-Gene +C Apple M1 1.25 + +changecom(blah) + +define(`rp', x0) +define(`up', x1) +define(`n', x2) +define(`v0', x3) +define(`cin',x4) + +define(`CY',x17) + +ifdef(`OPERATION_addmul_1', ` + define(`ADDSUB', adds) + define(`ADDSUBC', adcs) + define(`COND', `cc') + define(`func', mpn_addmul_1)') +ifdef(`OPERATION_submul_1', ` + define(`ADDSUB', subs) + define(`ADDSUBC', sbcs) + define(`COND', `cs') + define(`func', mpn_submul_1)') + +MULFUNC_PROLOGUE(mpn_addmul_1 mpn_submul_1 mpn_addmul_1c) + +ifdef(`OPERATION_addmul_1', ` +PROLOGUE(mpn_addmul_1c) + mov CY, cin + b L(ent) +EPILOGUE() +') + +PROLOGUE(func) + mov CY, #0 C W0 +L(ent): lsr x16, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x4, [up], #8 + mul x8, x4, v0 + umulh x4, x4, v0 + tbz n, #1, L(b01) + +L(b11): ldp x5,x6, [up], #16 + ldp x12,x13, [rp] + ldr x14, [rp,#16] + mul x9, x5, v0 + umulh x5, x5, v0 + mul x10, x6, v0 + umulh x6, x6, v0 + ADDSUB x8, x12, x8 + ADDSUBC x4, x13, x4 + ADDSUBC x5, x14, x5 + csinc x6, x6, x6, COND + ADDSUB x8, x8, CY + ADDSUBC x4, x4, x9 + ADDSUBC x5, x5, x10 + csinc CY, x6, x6, COND + stp x8, x4, [rp], #16 + str x5, [rp], #8 + cbnz x16, L(top) + mov x0, CY + ret + +L(b01): ldr x12, [rp] + ADDSUB x8, x12, x8 + csinc x4, x4, x4, COND + ADDSUB x8, x8, CY + csinc CY, x4, x4, COND + str x8, [rp], #8 + cbnz x16, L(top) + mov x0, CY + ret + +L(bx0): ldp x4,x5, [up], #16 + tbz n, #1, L(top)+4 + +L(b10): ldp x12,x13, [rp] + mul x8, x4, v0 + umulh x4, x4, v0 + mul x9, x5, v0 + umulh x5, x5, v0 + ADDSUB x8, x12, x8 + ADDSUBC x4, x13, x4 + csinc x5, x5, x5, COND + ADDSUB x8, x8, CY + ADDSUBC x4, x4, x9 + csinc CY, x5, x5, COND + stp x8, x4, [rp], #16 + cbz x16, L(done) + +L(top): ldp x4,x5, [up], #16 C W0 W1 + ldp x6,x7, [up], #16 C W2 W3 + ldp x12,x13, [rp] C W0 W1 + ldp x14,x15, [rp,#16] C W2 W3 + mul x8, x4, v0 C W0 + umulh x4, x4, v0 C W1 + mul x9, x5, v0 C W1 + umulh x5, x5, v0 C W2 + mul x10, x6, v0 C W2 + umulh x6, x6, v0 C W3 + mul x11, x7, v0 C W3 + umulh x7, x7, v0 C W4 + ADDSUB x8, x12, x8 C W0 + ADDSUBC x4, x13, x4 C W1 + ADDSUBC x5, x14, x5 C W2 + ADDSUBC x6, x15, x6 C W3 + csinc x7, x7, x7, COND C W4 + ADDSUB x8, x8, CY C W0 carry-in + ADDSUBC x4, x4, x9 C W1 + ADDSUBC x5, x5, x10 C W2 + ADDSUBC x6, x6, x11 C W2 + csinc CY, x7, x7, COND C W3 carry-out + stp x8, x4, [rp], #16 + stp x5, x6, [rp], #16 + sub x16, x16, #1 + cbnz x16, L(top) + +L(done):mov x0, CY + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/applem1/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/applem1/gmp-mparam.h new file mode 100644 index 0000000..d08262f --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/applem1/gmp-mparam.h @@ -0,0 +1,187 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file. + +Copyright 2020 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 3200 MHz Apple M1 */ +/* FFT tuning limit = 1 M */ +/* Generated by tuneup.c, 2020-12-25, gcc 4.2 */ + +#define MOD_1_1P_METHOD 2 /* 42.96% faster than 1 */ +#define MOD_1_NORM_THRESHOLD 0 /* always */ +#define MOD_1_UNNORM_THRESHOLD 0 /* always */ +#define MOD_1N_TO_MOD_1_1_THRESHOLD 5 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 3 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 11 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 18 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 11 +#define USE_PREINV_DIVREM_1 1 /* native */ +/* From m1.gmplib.org, 2023-07-21 */ +#define DIV_QR_1N_PI1_METHOD 3 /* 13.35% faster than 1 */ +#define DIV_QR_1_NORM_THRESHOLD 2 +#define DIV_QR_1_UNNORM_THRESHOLD 1 +#define DIV_QR_2_PI2_THRESHOLD 9 +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 28 + +#define DIV_1_VS_MUL_1_PERCENT 659 + +#define MUL_TOOM22_THRESHOLD 26 +#define MUL_TOOM33_THRESHOLD 77 +#define MUL_TOOM44_THRESHOLD 153 +#define MUL_TOOM6H_THRESHOLD 446 +#define MUL_TOOM8H_THRESHOLD 626 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 94 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 81 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 41 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 99 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 133 + +#define SQR_BASECASE_THRESHOLD 0 /* always (native) */ +#define SQR_TOOM2_THRESHOLD 47 +#define SQR_TOOM3_THRESHOLD 74 +#define SQR_TOOM4_THRESHOLD 372 +#define SQR_TOOM6_THRESHOLD 462 +#define SQR_TOOM8_THRESHOLD 592 + +#define MULMID_TOOM42_THRESHOLD 44 + +#define MULMOD_BNM1_THRESHOLD 9 +#define SQRMOD_BNM1_THRESHOLD 11 + +#define MUL_FFT_MODF_THRESHOLD 216 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 216, 5}, { 7, 4}, { 19, 5}, { 19, 6}, \ + { 10, 5}, { 21, 6}, { 21, 7}, { 11, 6}, \ + { 23, 7}, { 21, 8}, { 11, 7}, { 24, 8}, \ + { 21, 9}, { 11, 8}, { 27, 9}, { 15, 8}, \ + { 33, 9}, { 19, 8}, { 39, 9}, { 23, 8}, \ + { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \ + { 23, 9}, { 47,11}, { 15,10}, { 31, 9}, \ + { 63,10}, { 39, 9}, { 79,10}, { 55,11}, \ + { 31,10}, { 79,11}, { 47,12}, { 31,11}, \ + { 63,10}, { 127, 9}, { 255, 8}, { 511,11}, \ + { 79,10}, { 159, 9}, { 319, 8}, { 639,11}, \ + { 95,10}, { 191, 9}, { 383,12}, { 63,11}, \ + { 127,10}, { 255, 9}, { 511, 8}, { 1023,10}, \ + { 271, 9}, { 543, 8}, { 1087,11}, { 143,10}, \ + { 287, 9}, { 575, 8}, { 1151,11}, { 159,10}, \ + { 319, 9}, { 639,12}, { 95,11}, { 191,10}, \ + { 383,13}, { 63,12}, { 127,11}, { 255,10}, \ + { 511, 9}, { 1023,11}, { 271,10}, { 543, 9}, \ + { 1087, 8}, { 2175,11}, { 287,10}, { 575, 9}, \ + { 1151,12}, { 159,11}, { 319,10}, { 639, 9}, \ + { 1279,11}, { 351,10}, { 703, 9}, { 1407,12}, \ + { 191,11}, { 383,10}, { 767,11}, { 415,12}, \ + { 223,11}, { 447,10}, { 895,11}, { 479,10}, \ + { 959,13}, { 8192,14}, { 16384,15}, { 32768,16}, \ + { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \ + {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 104 +#define MUL_FFT_THRESHOLD 2368 + +#define SQR_FFT_MODF_THRESHOLD 304 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 304, 5}, { 10, 4}, { 21, 5}, { 11, 4}, \ + { 23, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \ + { 21, 7}, { 11, 6}, { 23, 7}, { 21, 8}, \ + { 11, 7}, { 24, 8}, { 15, 7}, { 31, 8}, \ + { 21, 9}, { 11, 8}, { 27, 9}, { 15, 8}, \ + { 33, 9}, { 19, 8}, { 39, 9}, { 23, 8}, \ + { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \ + { 23, 9}, { 47,11}, { 15,10}, { 31, 9}, \ + { 63,10}, { 39, 9}, { 79,10}, { 47,11}, \ + { 31,10}, { 79,11}, { 47,12}, { 31,11}, \ + { 63,10}, { 127, 9}, { 255, 8}, { 511,11}, \ + { 79,10}, { 159, 9}, { 319, 8}, { 639,11}, \ + { 95,10}, { 191, 9}, { 383,12}, { 63,10}, \ + { 255, 9}, { 511, 8}, { 1023,10}, { 271, 9}, \ + { 543, 8}, { 1087,10}, { 287, 9}, { 575, 8}, \ + { 1151,11}, { 159,10}, { 319, 9}, { 639,11}, \ + { 175,12}, { 95,11}, { 191,10}, { 383, 9}, \ + { 767,13}, { 63,12}, { 127,11}, { 255,10}, \ + { 511, 9}, { 1023,11}, { 271,10}, { 543, 9}, \ + { 1087, 8}, { 2175,10}, { 575, 9}, { 1151,11}, \ + { 303,12}, { 159,11}, { 319,10}, { 639, 9}, \ + { 1279,11}, { 351,10}, { 703, 9}, { 1407,12}, \ + { 191,11}, { 383,10}, { 767,11}, { 415,10}, \ + { 831, 9}, { 1663,12}, { 223,11}, { 447,10}, \ + { 895,11}, { 479,10}, { 959, 9}, { 1919,13}, \ + { 8192,14}, { 16384,15}, { 32768,16}, { 65536,17}, \ + { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \ + {2097152,22}, {4194304,23}, {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 111 +#define SQR_FFT_THRESHOLD 1856 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 76 +#define MULLO_MUL_N_THRESHOLD 4292 +#define SQRLO_BASECASE_THRESHOLD 6 +#define SQRLO_DC_THRESHOLD 186 +#define SQRLO_SQR_THRESHOLD 3688 + +#define DC_DIV_QR_THRESHOLD 67 +#define DC_DIVAPPR_Q_THRESHOLD 242 +#define DC_BDIV_QR_THRESHOLD 68 +#define DC_BDIV_Q_THRESHOLD 129 + +#define INV_MULMOD_BNM1_THRESHOLD 82 +#define INV_NEWTON_THRESHOLD 157 +#define INV_APPR_THRESHOLD 157 + +#define BINV_NEWTON_THRESHOLD 99 +#define REDC_1_TO_REDC_N_THRESHOLD 68 + +#define MU_DIV_QR_THRESHOLD 979 +#define MU_DIVAPPR_Q_THRESHOLD 1210 +#define MUPI_DIV_QR_THRESHOLD 76 +#define MU_BDIV_QR_THRESHOLD 942 +#define MU_BDIV_Q_THRESHOLD 1341 + +#define POWM_SEC_TABLE 11,75,137,712,2177 + +#define GET_STR_DC_THRESHOLD 12 +#define GET_STR_PRECOMPUTE_THRESHOLD 18 +#define SET_STR_DC_THRESHOLD 632 +#define SET_STR_PRECOMPUTE_THRESHOLD 1215 + +#define FAC_DSC_THRESHOLD 252 +#define FAC_ODD_THRESHOLD 0 /* always */ + +#define MATRIX22_STRASSEN_THRESHOLD 9 +#define HGCD2_DIV1_METHOD 1 /* 8.52% faster than 3 */ +#define HGCD_THRESHOLD 131 +#define HGCD_APPR_THRESHOLD 144 +#define HGCD_REDUCE_THRESHOLD 1962 +#define GCD_DC_THRESHOLD 435 +#define GCDEXT_DC_THRESHOLD 199 +#define JACOBI_BASE_METHOD 4 /* 0.80% faster than 1 */ diff --git a/gmp-6.3.0/mpn/arm64/applem1/sqr_basecase.asm b/gmp-6.3.0/mpn/arm64/applem1/sqr_basecase.asm new file mode 100644 index 0000000..22246cf --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/applem1/sqr_basecase.asm @@ -0,0 +1,318 @@ +dnl ARM64 mpn_sqr_basecase + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2020 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') + +dnl TODO +dnl * Replace the mul_1 code with less scheduled and thus simpler code. If +dnl we base it on the addmul_1 loop, the corner code could benefit from +dnl similar incoming register state, which could eliminate some loads. +dnl * Handle n = 4 early. +dnl * Duplicate addmul loop into 4 loops which fall into each other. Perhaps +dnl stick to one mul_1 loop, but do the (mod 4) stuff at its end instead of +dnl its beginning. + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') + +define(`v0', `x3') +define(`CY', `x17') + +PROLOGUE(mpn_sqr_basecase) + cmp n, #3 + b.ls L(le3) + + ldr v0, [up],#8 + sub n, n, #1 + mul x6, v0, v0 + umulh x4, v0, v0 + str x6, [rp],#8 + lsl v0, v0, 1 + lsl n, n, #3 + lsr x16, n, #5 + tbnz n, #3, L(mbx1) + +L(mbx0):adds x11, x4, xzr C move and clear cy + tbz n, #4, L(mb00) + +L(mb10):ldp x4, x5, [up],#16 + mul x8, x4, v0 + umulh x10, x4, v0 + cbz x16, L(m2e) + ldp x6, x7, [up],#16 + mul x9, x5, v0 + b L(mmid)-8 + +L(mbx1):ldr x7, [up],#8 + mul x9, x7, v0 + umulh x11, x7, v0 + adds x9, x9, x4 + str x9, [rp],#8 + tbnz n, #4, L(mb10) +L(mb00):ldp x6, x7, [up],#16 + mul x8, x6, v0 + umulh x10, x6, v0 + ldp x4, x5, [up],#16 + mul x9, x7, v0 + adcs x12, x8, x11 + umulh x11, x7, v0 + sub x16, x16, #1 + cbz x16, L(mend) + + ALIGN(16) +L(mtop):mul x8, x4, v0 + ldp x6, x7, [up],#16 + adcs x13, x9, x10 + umulh x10, x4, v0 + mul x9, x5, v0 + stp x12, x13, [rp],#16 + adcs x12, x8, x11 + umulh x11, x5, v0 +L(mmid):mul x8, x6, v0 + ldp x4, x5, [up],#16 + adcs x13, x9, x10 + umulh x10, x6, v0 + mul x9, x7, v0 + stp x12, x13, [rp],#16 + adcs x12, x8, x11 + umulh x11, x7, v0 + sub x16, x16, #1 + cbnz x16, L(mtop) + +L(mend):mul x8, x4, v0 + adcs x13, x9, x10 + umulh x10, x4, v0 + stp x12, x13, [rp],#16 +L(m2e): mul x9, x5, v0 + adcs x12, x8, x11 + umulh x11, x5, v0 + adcs x13, x9, x10 + stp x12, x13, [rp],#16 + adc x11, x11, xzr + str x11, [rp],#8 + +L(outer): + sub n, n, #8 + sub rp, rp, n + sub up, up, n + ldp x6, x7, [up,#-16] + ldr v0, [rp,#-8] + and x8, x7, x6, asr 63 + mul x9, x7, x7 + adds v0, v0, x8 + umulh x4, x7, x7 + adc x4, x4, xzr + adds v0, v0, x9 + str v0, [rp,#-8] + adc CY, x4, xzr + adds xzr, x6, x6 + adc v0, x7, x7 + cmp n, #16 + beq L(cor2) + + lsr x16, n, #5 + tbz n, #3, L(bx0) + +L(bx1): ldr x4, [up],#8 + mul x8, x4, v0 + umulh x4, x4, v0 + tbz n, #4, L(b01) + +L(b11): ldp x5, x6, [up],#16 + ldp x12, x13, [rp] + ldr x14, [rp,#16] + mul x9, x5, v0 + umulh x5, x5, v0 + mul x10, x6, v0 + umulh x6, x6, v0 + adds x8, x12, x8 + adcs x4, x13, x4 + adcs x5, x14, x5 + adc x6, x6, xzr + adds x8, x8, CY + adcs x4, x4, x9 + adcs x5, x5, x10 + adc CY, x6, xzr + stp x8, x4, [rp],#16 + str x5, [rp],#8 + cbnz x16, L(top) + b L(end) + +L(b01): ldr x12, [rp] + adds x8, x12, x8 + adc x4, x4, xzr + adds x8, x8, CY + adc CY, x4, xzr + str x8, [rp],#8 + b L(top) + +L(bx0): ldp x4, x5, [up],#16 + tbz n, #4, L(top)+4 + +L(b10): ldp x12, x13, [rp] + mul x8, x4, v0 + umulh x4, x4, v0 + mul x9, x5, v0 + umulh x5, x5, v0 + adds x8, x12, x8 + adcs x4, x13, x4 + adc x5, x5, xzr + adds x8, x8, CY + adcs x4, x4, x9 + adc CY, x5, xzr + stp x8, x4, [rp],#16 + + ALIGN(16) +L(top): ldp x4, x5, [up],#16 + ldp x6, x7, [up],#16 + ldp x12, x13, [rp] + ldp x14, x15, [rp,#16] + mul x8, x4, v0 + umulh x4, x4, v0 + mul x9, x5, v0 + umulh x5, x5, v0 + mul x10, x6, v0 + umulh x6, x6, v0 + mul x11, x7, v0 + umulh x7, x7, v0 + adds x8, x12, x8 + adcs x4, x13, x4 + adcs x5, x14, x5 + adcs x6, x15, x6 + adc x7, x7, xzr + adds x8, x8, CY + adcs x4, x4, x9 + adcs x5, x5, x10 + adcs x6, x6, x11 + adc CY, x7, xzr + stp x8, x4, [rp],#16 + stp x5, x6, [rp],#16 + sub x16, x16, #1 + cbnz x16, L(top) + +L(end): str CY, [rp],#8 + b L(outer) + +L(cor2):ldp x10, x11, [up] + ldp x12, x13, [rp] + mul x8, x10, v0 + umulh x4, x10, v0 + mul x9, x11, v0 + umulh x5, x11, v0 + adds x8, x12, x8 + adcs x4, x13, x4 + adc x5, x5, xzr + adds x8, x8, CY + adcs x13, x4, x9 + adc x12, x5, xzr + str x8, [rp] + and x8, x10, x7, asr 63 + mul x9, x10, x10 + adds x13, x13, x8 + umulh x4, x10, x10 + adc x4, x4, xzr + adds x13, x13, x9 + adc CY, x4, xzr + adds xzr, x7, x7 + adc v0, x10, x10 + mul x8, x11, v0 + umulh x4, x11, v0 + adds x8, x12, x8 + adc x4, x4, xzr + adds x8, x8, CY + adc v0, x4, xzr + stp x13, x8, [rp,#8] + and x2, x11, x10, asr 63 + mul x5, x11, x11 + adds v0, v0, x2 + umulh x4, x11, x11 + adc x4, x4, xzr + adds v0, v0, x5 + adc x4, x4, xzr + stp v0, x4, [rp,#24] + ret + +L(le3): ldr v0, [up] + mul x4, v0, v0 C W0 + umulh x5, v0, v0 C W1 + cmp n, #2 + b.hs L(2o3) + stp x4, x5, [rp] + ret + +L(2o3): ldr x6, [up,#8] + mul x7, x6, x6 C W2 + umulh x8, x6, x6 C W3 + mul x9, v0, x6 C W1+1/64 + umulh x10, v0, x6 C W2+1/64 + b.hi L(3) + adds x5, x5, x9 C W1 + adcs x7, x7, x10 C W2 + adc x8, x8, xzr C W3 + adds x5, x5, x9 C W1 + adcs x7, x7, x10 C W2 + adc x8, x8, xzr C W3 + stp x4, x5, [rp] + stp x7, x8, [rp,#16] + ret + +L(3): ldr x11, [up,#16] + mul x12, x11, x11 C W4 + umulh x13, x11, x11 C W5 + mul x14, v0, x11 C W2+1/64 + umulh x15, v0, x11 C W3+1/64 + mul x16, x6, x11 C W3+1/64 + umulh x17, x6, x11 C W4+1/64 + adds x5, x5, x9 + adcs x7, x7, x10 + adcs x8, x8, x15 + adcs x12, x12, x17 + adc x13, x13, xzr + adds x5, x5, x9 + adcs x7, x7, x10 + adcs x8, x8, x15 + adcs x12, x12, x17 + adc x13, x13, xzr + adds x7, x7, x14 + adcs x8, x8, x16 + adcs x12, x12, xzr + adc x13, x13, xzr + adds x7, x7, x14 + adcs x8, x8, x16 + adcs x12, x12, xzr + adc x13, x13, xzr + stp x4, x5, [rp] + stp x7, x8, [rp,#16] + stp x12, x13, [rp,#32] + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/arm64-defs.m4 b/gmp-6.3.0/mpn/arm64/arm64-defs.m4 new file mode 100644 index 0000000..46149f7 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/arm64-defs.m4 @@ -0,0 +1,53 @@ +divert(-1) + +dnl m4 macros for ARM64 ELF assembler. + +dnl Copyright 2020 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/. + + +dnl Standard commenting is with @, the default m4 # is for constants and we +dnl don't want to disable macro expansions in or after them. + +changecom + + +dnl LEA_HI(reg,gmp_symbol), LEA_LO(reg,gmp_symbol) +dnl +dnl Load the address of gmp_symbol into a register. We split this into two +dnl parts to allow separation for manual insn scheduling. + +ifdef(`PIC',`dnl +define(`LEA_HI', `adrp $1, :got:$2')dnl +define(`LEA_LO', `ldr $1, [$1, #:got_lo12:$2]')dnl +',`dnl +define(`LEA_HI', `adrp $1, $2')dnl +define(`LEA_LO', `add $1, $1, :lo12:$2')dnl +')dnl + +divert`'dnl diff --git a/gmp-6.3.0/mpn/arm64/bdiv_dbm1c.asm b/gmp-6.3.0/mpn/arm64/bdiv_dbm1c.asm new file mode 100644 index 0000000..78984b4 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/bdiv_dbm1c.asm @@ -0,0 +1,111 @@ +dnl ARM64 mpn_bdiv_dbm1c. + +dnl Copyright 2008, 2011, 2012, 2014 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 Cortex-A53 8 +C Cortex-A57 7 +C X-Gene 4.25 + +define(`qp', `x0') +define(`up', `x1') +define(`n', `x2') +define(`bd', `x3') +define(`cy', `x4') + +ASM_START() + TEXT + ALIGN(16) +PROLOGUE(mpn_bdiv_dbm1c) + ldr x5, [up], #8 + ands x6, n, #3 + b.eq L(fi0) + cmp x6, #2 + b.cc L(fi1) + b.eq L(fi2) + +L(fi3): mul x12, x5, bd + umulh x13, x5, bd + ldr x5, [up], #8 + b L(lo3) + +L(fi0): mul x10, x5, bd + umulh x11, x5, bd + ldr x5, [up], #8 + b L(lo0) + +L(fi1): subs n, n, #1 + mul x12, x5, bd + umulh x13, x5, bd + b.ls L(wd1) + ldr x5, [up], #8 + b L(lo1) + +L(fi2): mul x10, x5, bd + umulh x11, x5, bd + ldr x5, [up], #8 + b L(lo2) + +L(top): ldr x5, [up], #8 + subs x4, x4, x10 + str x4, [qp], #8 + sbc x4, x4, x11 +L(lo1): mul x10, x5, bd + umulh x11, x5, bd + ldr x5, [up], #8 + subs x4, x4, x12 + str x4, [qp], #8 + sbc x4, x4, x13 +L(lo0): mul x12, x5, bd + umulh x13, x5, bd + ldr x5, [up], #8 + subs x4, x4, x10 + str x4, [qp], #8 + sbc x4, x4, x11 +L(lo3): mul x10, x5, bd + umulh x11, x5, bd + ldr x5, [up], #8 + subs x4, x4, x12 + str x4, [qp], #8 + sbc x4, x4, x13 +L(lo2): subs n, n, #4 + mul x12, x5, bd + umulh x13, x5, bd + b.hi L(top) + +L(wd2): subs x4, x4, x10 + str x4, [qp], #8 + sbc x4, x4, x11 +L(wd1): subs x4, x4, x12 + str x4, [qp] + sbc x0, x4, x13 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/bdiv_q_1.asm b/gmp-6.3.0/mpn/arm64/bdiv_q_1.asm new file mode 100644 index 0000000..7fffc93 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/bdiv_q_1.asm @@ -0,0 +1,122 @@ +dnl ARM64 mpn_bdiv_q_1, mpn_pi1_bdiv_q_1 -- Hensel division by 1-limb divisor. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2012, 2017 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 norm unorm +C Cortex-A53 12 15 +C Cortex-A57 12 12 +C Cortex-A72 +C Cortex-A73 +C X-Gene 11 11 + +C TODO +C * Scheduling of umulh later in the unorm loop brings A53 time to 12 c/l. +C Unfortunately, that requires software pipelining. + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') +define(`d', `x3') +define(`di', `x4') C just mpn_pi1_bdiv_q_1 +define(`cnt', `x5') C just mpn_pi1_bdiv_q_1 + +define(`cy', `r7') +define(`tnc', `x8') + +ASM_START() +PROLOGUE(mpn_bdiv_q_1) + + rbit x6, d + clz cnt, x6 + lsr d, d, cnt + + LEA_HI( x7, binvert_limb_table) + ubfx x6, d, 1, 7 + LEA_LO( x7, binvert_limb_table) + ldrb w6, [x7, x6] + ubfiz x7, x6, 1, 8 + umull x6, w6, w6 + msub x6, x6, d, x7 + lsl x7, x6, 1 + mul x6, x6, x6 + msub x6, x6, d, x7 + lsl x7, x6, 1 + mul x6, x6, x6 + msub di, x6, d, x7 + + b GSYM_PREFIX`'mpn_pi1_bdiv_q_1 +EPILOGUE() + +PROLOGUE(mpn_pi1_bdiv_q_1) + sub n, n, #1 + subs x6, x6, x6 C clear r6 and C flag + ldr x9, [up],#8 + cbz cnt, L(norm) + +L(unorm): + lsr x12, x9, cnt + cbz n, L(eu1) + sub tnc, xzr, cnt + +L(tpu): ldr x9, [up],#8 + lsl x7, x9, tnc + orr x7, x7, x12 + sbcs x6, x7, x6 + mul x7, x6, di + str x7, [rp],#8 + lsr x12, x9, cnt + umulh x6, x7, d + sub n, n, #1 + cbnz n, L(tpu) + +L(eu1): sbcs x6, x12, x6 + mul x6, x6, di + str x6, [rp] + ret + +L(norm): + mul x5, x9, di + str x5, [rp],#8 + cbz n, L(en1) + +L(tpn): ldr x9, [up],#8 + umulh x5, x5, d + sbcs x5, x9, x5 + mul x5, x5, di + str x5, [rp],#8 + sub n, n, #1 + cbnz n, L(tpn) + +L(en1): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/cnd_aors_n.asm b/gmp-6.3.0/mpn/arm64/cnd_aors_n.asm new file mode 100644 index 0000000..397aa51 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cnd_aors_n.asm @@ -0,0 +1,129 @@ +dnl ARM64 mpn_cnd_add_n, mpn_cnd_sub_n + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2012, 2013, 2017 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 Cortex-A53 3.87-4.37 +C Cortex-A57 1.75 +C X-Gene 2.0 + +changecom(blah) + +define(`cnd', `x0') +define(`rp', `x1') +define(`up', `x2') +define(`vp', `x3') +define(`n', `x4') + +ifdef(`OPERATION_cnd_add_n', ` + define(`ADDSUBC', adcs) + define(`CLRCY', `cmn xzr, xzr') + define(`RETVAL', `cset x0, cs') + define(`func', mpn_cnd_add_n)') +ifdef(`OPERATION_cnd_sub_n', ` + define(`ADDSUBC', sbcs) + define(`CLRCY', `cmp xzr, xzr') + define(`RETVAL', `cset x0, cc') + define(`func', mpn_cnd_sub_n)') + +MULFUNC_PROLOGUE(mpn_cnd_add_n mpn_cnd_sub_n) + +ASM_START() +PROLOGUE(func) + cmp cnd, #1 + sbc cnd, cnd, cnd + + CLRCY + + lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x13, [vp] + ldr x11, [up] + bic x7, x13, cnd + ADDSUBC x9, x11, x7 + str x9, [rp] + tbnz n, #1, L(b11) + +L(b01): cbz x17, L(rt) + ldp x12, x13, [vp,#8] + ldp x10, x11, [up,#8] + sub up, up, #8 + sub vp, vp, #8 + sub rp, rp, #24 + b L(mid) + +L(b11): ldp x12, x13, [vp,#8]! + ldp x10, x11, [up,#8]! + sub rp, rp, #8 + cbz x17, L(end) + b L(top) + +L(bx0): ldp x12, x13, [vp] + ldp x10, x11, [up] + tbnz n, #1, L(b10) + +L(b00): sub up, up, #16 + sub vp, vp, #16 + sub rp, rp, #32 + b L(mid) + +L(b10): sub rp, rp, #16 + cbz x17, L(end) + + ALIGN(16) +L(top): bic x6, x12, cnd + bic x7, x13, cnd + ldp x12, x13, [vp,#16] + ADDSUBC x8, x10, x6 + ADDSUBC x9, x11, x7 + ldp x10, x11, [up,#16] + stp x8, x9, [rp,#16] +L(mid): bic x6, x12, cnd + bic x7, x13, cnd + ldp x12, x13, [vp,#32]! + ADDSUBC x8, x10, x6 + ADDSUBC x9, x11, x7 + ldp x10, x11, [up,#32]! + stp x8, x9, [rp,#32]! + sub x17, x17, #1 + cbnz x17, L(top) + +L(end): bic x6, x12, cnd + bic x7, x13, cnd + ADDSUBC x8, x10, x6 + ADDSUBC x9, x11, x7 + stp x8, x9, [rp,#16] +L(rt): RETVAL + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/com.asm b/gmp-6.3.0/mpn/arm64/com.asm new file mode 100644 index 0000000..d594943 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/com.asm @@ -0,0 +1,92 @@ +dnl ARM64 mpn_com. + +dnl Copyright 2013, 2020 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 Cortex-A53 +C Cortex-A55 +C Cortex-A57 +C Cortex-A72 +C Cortex-A73 +C X-Gene +C Apple M1 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') + +ASM_START() +PROLOGUE(mpn_com) + cmp n, #3 + b.le L(bc) + +C Copy until rp is 128-bit aligned + tbz rp, #3, L(al2) + ldr x4, [up],#8 + sub n, n, #1 + mvn x4, x4 + str x4, [rp],#8 + +L(al2): ldp x4,x5, [up],#16 + sub n, n, #6 + tbnz n, #63, L(end) + + ALIGN(16) +L(top): ldp x6,x7, [up],#32 + mvn x4, x4 + mvn x5, x5 + stp x4,x5, [rp],#32 + ldp x4,x5, [up,#-16] + mvn x6, x6 + mvn x7, x7 + stp x6,x7, [rp,#-16] + sub n, n, #4 + tbz n, #63, L(top) + +L(end): mvn x4, x4 + mvn x5, x5 + stp x4,x5, [rp],#16 + +C Copy last 0-3 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tbz n, #1, L(tl1) + ldp x4,x5, [up],#16 + mvn x4, x4 + mvn x5, x5 + stp x4,x5, [rp],#16 +L(tl1): tbz n, #0, L(tl2) + ldr x4, [up] + mvn x4, x4 + str x4, [rp] +L(tl2): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/copyd.asm b/gmp-6.3.0/mpn/arm64/copyd.asm new file mode 100644 index 0000000..d542970 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/copyd.asm @@ -0,0 +1,85 @@ +dnl ARM64 mpn_copyd. + +dnl Copyright 2013, 2020 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 Cortex-A53 1.8 +C Cortex-A55 1.28 +C Cortex-A57 +C Cortex-A72 1 +C Cortex-A73 1.1-1.35 (alignment dependent) +C X-Gene 1 +C Apple M1 0.31 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') + +ASM_START() +PROLOGUE(mpn_copyd) + add rp, rp, n, lsl #3 + add up, up, n, lsl #3 + + cmp n, #3 + b.le L(bc) + +C Copy until rp is 128-bit aligned + tbz rp, #3, L(al2) + ldr x4, [up,#-8]! + sub n, n, #1 + str x4, [rp,#-8]! + +L(al2): ldp x4,x5, [up,#-16]! + sub n, n, #6 + tbnz n, #63, L(end) + + ALIGN(16) +L(top): ldp x6,x7, [up,#-16] + stp x4,x5, [rp,#-16] + ldp x4,x5, [up,#-32]! + stp x6,x7, [rp,#-32]! + sub n, n, #4 + tbz n, #63, L(top) + +L(end): stp x4,x5, [rp,#-16]! + +C Copy last 0-3 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tbz n, #1, L(tl1) + ldp x4,x5, [up,#-16]! + stp x4,x5, [rp,#-16]! +L(tl1): tbz n, #0, L(tl2) + ldr x4, [up,#-8] + str x4, [rp,#-8] +L(tl2): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/copyi.asm b/gmp-6.3.0/mpn/arm64/copyi.asm new file mode 100644 index 0000000..0de40c5 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/copyi.asm @@ -0,0 +1,82 @@ +dnl ARM64 mpn_copyi. + +dnl Copyright 2013, 2020 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 Cortex-A53 1.8 +C Cortex-A55 1.28 +C Cortex-A57 +C Cortex-A72 1 +C Cortex-A73 1.1-1.35 (alignment dependent) +C X-Gene 1 +C Apple M1 0.31 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') + +ASM_START() +PROLOGUE(mpn_copyi) + cmp n, #3 + b.le L(bc) + +C Copy until rp is 128-bit aligned + tbz rp, #3, L(al2) + ldr x4, [up],#8 + sub n, n, #1 + str x4, [rp],#8 + +L(al2): ldp x4,x5, [up],#16 + sub n, n, #6 + tbnz n, #63, L(end) + + ALIGN(16) +L(top): ldp x6,x7, [up],#32 + stp x4,x5, [rp],#32 + ldp x4,x5, [up,#-16] + stp x6,x7, [rp,#-16] + sub n, n, #4 + tbz n, #63, L(top) + +L(end): stp x4,x5, [rp],#16 + +C Copy last 0-3 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tbz n, #1, L(tl1) + ldp x4,x5, [up],#16 + stp x4,x5, [rp],#16 +L(tl1): tbz n, #0, L(tl2) + ldr x4, [up] + str x4, [rp] +L(tl2): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/cora53/cnd_aors_n.asm b/gmp-6.3.0/mpn/arm64/cora53/cnd_aors_n.asm new file mode 100644 index 0000000..1b227da --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cora53/cnd_aors_n.asm @@ -0,0 +1,99 @@ +dnl ARM64 mpn_cnd_add_n, mpn_cnd_sub_n + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2012, 2013, 2017 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 Cortex-A53 3.5-4 +C Cortex-A57 2.25 +C X-Gene 3.5 + +changecom(blah) + +define(`cnd', `x0') +define(`rp', `x1') +define(`up', `x2') +define(`vp', `x3') +define(`n', `x4') + +ifdef(`OPERATION_cnd_add_n', ` + define(`ADDSUBC', adcs) + define(`CLRCY', `cmn xzr, xzr') + define(`RETVAL', `cset x0, cs') + define(`func', mpn_cnd_add_n)') +ifdef(`OPERATION_cnd_sub_n', ` + define(`ADDSUBC', sbcs) + define(`CLRCY', `cmp xzr, xzr') + define(`RETVAL', `cset x0, cc') + define(`func', mpn_cnd_sub_n)') + +MULFUNC_PROLOGUE(mpn_cnd_add_n mpn_cnd_sub_n) + +ASM_START() +PROLOGUE(func) + cmp cnd, #1 + sbc cnd, cnd, cnd + + CLRCY C really only needed for n = 0 (mod 4) + + tbz n, #0, L(1) + ldr x10, [up], #8 + ldr x12, [vp], #8 + bic x6, x12, cnd + ADDSUBC x8, x10, x6 + sub n, n, #1 + str x8, [rp], #8 + cbz n, L(rt) + +L(1): ldp x10, x11, [up], #16 + ldp x12, x13, [vp], #16 + sub n, n, #2 + cbz n, L(end) + +L(top): bic x6, x12, cnd + bic x7, x13, cnd + ldp x12, x13, [vp], #16 + ADDSUBC x8, x10, x6 + ADDSUBC x9, x11, x7 + ldp x10, x11, [up], #16 + sub n, n, #2 + stp x8, x9, [rp], #16 + cbnz n, L(top) + +L(end): bic x6, x12, cnd + bic x7, x13, cnd + ADDSUBC x8, x10, x6 + ADDSUBC x9, x11, x7 + stp x8, x9, [rp] +L(rt): RETVAL + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/cora53/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/cora53/gmp-mparam.h new file mode 100644 index 0000000..f4e258d --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cora53/gmp-mparam.h @@ -0,0 +1,242 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file for a53. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 1536 MHz Cortex-A53 */ +/* FFT tuning limit = 21,583,800 */ +/* Generated by tuneup.c, 2019-10-22, gcc 5.4 */ + +#define DIVREM_1_NORM_THRESHOLD 3 +#define DIVREM_1_UNNORM_THRESHOLD 4 +#define MOD_1_1P_METHOD 2 /* 4.84% faster than 1 */ +#define MOD_1_NORM_THRESHOLD 3 +#define MOD_1_UNNORM_THRESHOLD 4 +#define MOD_1N_TO_MOD_1_1_THRESHOLD 7 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 7 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 12 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 18 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 22 +#define USE_PREINV_DIVREM_1 1 +#define DIV_QR_1N_PI1_METHOD 1 /* 39.05% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD 21 +#define DIV_QR_1_UNNORM_THRESHOLD 21 +#define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 38 + +#define DIV_1_VS_MUL_1_PERCENT 161 + +#define MUL_TOOM22_THRESHOLD 14 +#define MUL_TOOM33_THRESHOLD 49 +#define MUL_TOOM44_THRESHOLD 73 +#define MUL_TOOM6H_THRESHOLD 173 +#define MUL_TOOM8H_THRESHOLD 236 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 81 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 77 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 81 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 88 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 65 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 18 +#define SQR_TOOM3_THRESHOLD 68 +#define SQR_TOOM4_THRESHOLD 183 +#define SQR_TOOM6_THRESHOLD 230 +#define SQR_TOOM8_THRESHOLD 357 + +#define MULMID_TOOM42_THRESHOLD 23 + +#define MULMOD_BNM1_THRESHOLD 9 +#define SQRMOD_BNM1_THRESHOLD 11 + +#define MUL_FFT_MODF_THRESHOLD 316 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 316, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 13, 7}, { 7, 6}, { 15, 7}, { 8, 6}, \ + { 17, 7}, { 9, 6}, { 19, 7}, { 17, 8}, \ + { 9, 7}, { 20, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 19, 9}, { 11, 8}, \ + { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \ + { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \ + { 15, 9}, { 39,10}, { 23, 9}, { 51,11}, \ + { 15,10}, { 31, 9}, { 71,10}, { 39, 9}, \ + { 83,10}, { 47, 9}, { 99,10}, { 55,11}, \ + { 31,10}, { 63, 8}, { 255,10}, { 71, 8}, \ + { 287,10}, { 79, 9}, { 159, 8}, { 319,10}, \ + { 87,11}, { 47,10}, { 95, 9}, { 191, 8}, \ + { 383,10}, { 103, 9}, { 207, 8}, { 415,10}, \ + { 111, 9}, { 223,12}, { 31,11}, { 63, 9}, \ + { 255, 8}, { 511,10}, { 135, 9}, { 287, 8}, \ + { 575,11}, { 79,10}, { 159, 9}, { 319, 8}, \ + { 639,10}, { 175, 9}, { 351, 8}, { 703,11}, \ + { 95,10}, { 191, 9}, { 383, 8}, { 767,10}, \ + { 207, 9}, { 415, 8}, { 831,10}, { 223, 9}, \ + { 447,12}, { 63,10}, { 255, 9}, { 511, 8}, \ + { 1023, 9}, { 543,10}, { 287, 9}, { 575, 8}, \ + { 1151,11}, { 159,10}, { 319, 9}, { 639,11}, \ + { 175,10}, { 351, 9}, { 703, 8}, { 1407,12}, \ + { 95,11}, { 191,10}, { 383, 9}, { 767,11}, \ + { 207,10}, { 415, 9}, { 831,11}, { 223,10}, \ + { 447,13}, { 63,11}, { 255,10}, { 543,11}, \ + { 287,10}, { 575, 9}, { 1151,12}, { 159,11}, \ + { 319,10}, { 639,11}, { 351,10}, { 703, 9}, \ + { 1407, 8}, { 2815,12}, { 191,11}, { 383,10}, \ + { 767,11}, { 415,10}, { 831,12}, { 223,11}, \ + { 447,10}, { 895,11}, { 479,10}, { 959, 9}, \ + { 1919,12}, { 255,11}, { 511,10}, { 1023,11}, \ + { 543,10}, { 1087,12}, { 287,11}, { 575,10}, \ + { 1151,12}, { 319,11}, { 639,12}, { 351,11}, \ + { 703,10}, { 1407, 9}, { 2815,13}, { 191,12}, \ + { 383,11}, { 767,12}, { 415,11}, { 831,10}, \ + { 1663,12}, { 447,11}, { 895,10}, { 1791,12}, \ + { 479,11}, { 959,13}, { 255,12}, { 511,11}, \ + { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \ + { 1151,13}, { 319,12}, { 703,11}, { 1407,10}, \ + { 2815,13}, { 383,12}, { 831,11}, { 1663,13}, \ + { 447,12}, { 895,11}, { 1791,12}, { 959,11}, \ + { 1919,14}, { 255,13}, { 511,12}, { 1087,13}, \ + { 575,12}, { 1151,13}, { 703,12}, { 1407,11}, \ + { 2815,14}, { 383,13}, { 831,12}, { 1663,13}, \ + { 895,12}, { 1791,13}, { 959,12}, { 1919,15}, \ + { 255,14}, { 511,13}, { 1087,12}, { 2175,13}, \ + { 1215,14}, { 639,13}, { 1407,12}, { 2815,14}, \ + { 767,13}, { 1663,14}, { 895,13}, { 1919,12}, \ + { 3839,15}, { 511,14}, { 1023,13}, { 2175,14}, \ + { 1151,13}, { 2431,12}, { 4863,14}, { 16384,15}, \ + { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \ + { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \ + {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 217 +#define MUL_FFT_THRESHOLD 3200 + +#define SQR_FFT_MODF_THRESHOLD 276 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 276, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 17, 7}, { 17, 8}, { 9, 7}, { 20, 8}, \ + { 11, 7}, { 23, 8}, { 13, 9}, { 7, 8}, \ + { 15, 7}, { 31, 8}, { 19, 9}, { 11, 8}, \ + { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \ + { 39, 9}, { 23, 8}, { 47, 9}, { 27,10}, \ + { 15, 9}, { 39,10}, { 23, 9}, { 47,11}, \ + { 15,10}, { 31, 9}, { 67,10}, { 39, 9}, \ + { 79,10}, { 47, 9}, { 95, 8}, { 191,10}, \ + { 55,11}, { 31,10}, { 63, 8}, { 255,10}, \ + { 71, 9}, { 143, 8}, { 287,10}, { 79, 9}, \ + { 159,11}, { 47,10}, { 95, 9}, { 191, 8}, \ + { 383, 7}, { 767,10}, { 103,12}, { 31,11}, \ + { 63, 9}, { 255, 8}, { 511, 7}, { 1023,10}, \ + { 143, 9}, { 287,11}, { 79,10}, { 159, 9}, \ + { 319, 8}, { 639,10}, { 175, 9}, { 351, 8}, \ + { 703,11}, { 95,10}, { 191, 9}, { 383, 8}, \ + { 767,10}, { 207, 9}, { 415, 8}, { 831,10}, \ + { 223, 9}, { 447,12}, { 63,10}, { 255, 9}, \ + { 511, 8}, { 1023,11}, { 143,10}, { 287, 9}, \ + { 575, 8}, { 1151,11}, { 159,10}, { 319, 9}, \ + { 639,11}, { 175,10}, { 351, 9}, { 703,12}, \ + { 95,11}, { 191,10}, { 383, 9}, { 767,11}, \ + { 207,10}, { 415, 9}, { 831,11}, { 223,10}, \ + { 447,13}, { 63,11}, { 255,10}, { 511, 9}, \ + { 1023,11}, { 287,10}, { 575, 9}, { 1151,12}, \ + { 159,11}, { 319,10}, { 639,11}, { 351,10}, \ + { 703, 9}, { 1407,12}, { 191,11}, { 383,10}, \ + { 767,11}, { 415,10}, { 831,12}, { 223,11}, \ + { 447,10}, { 895,11}, { 479,10}, { 959,12}, \ + { 255,11}, { 511,10}, { 1023,12}, { 287,11}, \ + { 575,10}, { 1151,12}, { 319,11}, { 639,12}, \ + { 351,11}, { 703,10}, { 1407,13}, { 191,12}, \ + { 383,11}, { 767,12}, { 415,11}, { 831,10}, \ + { 1663,12}, { 447,11}, { 895,12}, { 479,11}, \ + { 959,10}, { 1919,13}, { 255,12}, { 511,11}, \ + { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \ + { 1151,13}, { 319,12}, { 703,11}, { 1407,10}, \ + { 2815,13}, { 383,12}, { 831,11}, { 1663,13}, \ + { 447,12}, { 895,11}, { 1791,12}, { 959,14}, \ + { 255,13}, { 511,12}, { 1087,13}, { 575,12}, \ + { 1151,13}, { 703,12}, { 1407,11}, { 2815,14}, \ + { 383,13}, { 831,12}, { 1663,13}, { 895,12}, \ + { 1791,13}, { 959,12}, { 1919,15}, { 255,14}, \ + { 511,13}, { 1087,12}, { 2175,13}, { 1151,14}, \ + { 639,13}, { 1407,12}, { 2815,14}, { 767,13}, \ + { 1663,14}, { 895,13}, { 1919,12}, { 3839,15}, \ + { 511,14}, { 1023,13}, { 2175,14}, { 1151,13}, \ + { 2431,12}, { 4863,14}, { 16384,15}, { 32768,16}, \ + { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \ + {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 204 +#define SQR_FFT_THRESHOLD 2688 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 38 +#define MULLO_MUL_N_THRESHOLD 6253 +#define SQRLO_BASECASE_THRESHOLD 4 +#define SQRLO_DC_THRESHOLD 67 +#define SQRLO_SQR_THRESHOLD 5240 + +#define DC_DIV_QR_THRESHOLD 43 +#define DC_DIVAPPR_Q_THRESHOLD 155 +#define DC_BDIV_QR_THRESHOLD 39 +#define DC_BDIV_Q_THRESHOLD 89 + +#define INV_MULMOD_BNM1_THRESHOLD 34 +#define INV_NEWTON_THRESHOLD 163 +#define INV_APPR_THRESHOLD 161 + +#define BINV_NEWTON_THRESHOLD 196 +#define REDC_1_TO_REDC_N_THRESHOLD 43 + +#define MU_DIV_QR_THRESHOLD 998 +#define MU_DIVAPPR_Q_THRESHOLD 998 +#define MUPI_DIV_QR_THRESHOLD 91 +#define MU_BDIV_QR_THRESHOLD 807 +#define MU_BDIV_Q_THRESHOLD 924 + +#define POWM_SEC_TABLE 6,30,125,579,1730 + +#define GET_STR_DC_THRESHOLD 15 +#define GET_STR_PRECOMPUTE_THRESHOLD 30 +#define SET_STR_DC_THRESHOLD 802 +#define SET_STR_PRECOMPUTE_THRESHOLD 1815 + +#define FAC_DSC_THRESHOLD 258 +#define FAC_ODD_THRESHOLD 24 + +#define MATRIX22_STRASSEN_THRESHOLD 10 +#define HGCD2_DIV1_METHOD 1 /* 7.05% faster than 3 */ +#define HGCD_THRESHOLD 107 +#define HGCD_APPR_THRESHOLD 112 +#define HGCD_REDUCE_THRESHOLD 1679 +#define GCD_DC_THRESHOLD 324 +#define GCDEXT_DC_THRESHOLD 242 +#define JACOBI_BASE_METHOD 4 /* 22.41% faster than 1 */ + +/* Tuneup completed successfully, took 66624 seconds */ diff --git a/gmp-6.3.0/mpn/arm64/cora57/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/cora57/gmp-mparam.h new file mode 100644 index 0000000..e034f02 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cora57/gmp-mparam.h @@ -0,0 +1,188 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file for a57, a72-a75. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 1800 MHz Cortex-A72 */ +/* FFT tuning limit = 0.5 M */ +/* Generated by tuneup.c, 2019-10-02, gcc 7.4 */ + +#define DIVREM_1_NORM_THRESHOLD 3 +#define DIVREM_1_UNNORM_THRESHOLD 4 +#define MOD_1_1P_METHOD 1 /* 2.21% faster than 2 */ +#define MOD_1_NORM_THRESHOLD 3 +#define MOD_1_UNNORM_THRESHOLD 4 +#define MOD_1N_TO_MOD_1_1_THRESHOLD 6 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 4 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 8 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 42 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 15 +#define USE_PREINV_DIVREM_1 1 +/* From gcc117.osuosl.org, 2023-07-27 */ +#define DIV_QR_1N_PI1_METHOD 4 /* 8.57% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD 5 +#define DIV_QR_1_UNNORM_THRESHOLD 5 +#define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 33 + +#define DIV_1_VS_MUL_1_PERCENT 168 + +#define MUL_TOOM22_THRESHOLD 10 +#define MUL_TOOM33_THRESHOLD 41 +#define MUL_TOOM44_THRESHOLD 99 +#define MUL_TOOM6H_THRESHOLD 142 +#define MUL_TOOM8H_THRESHOLD 199 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 65 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 69 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 63 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 66 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 55 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 18 +#define SQR_TOOM3_THRESHOLD 65 +#define SQR_TOOM4_THRESHOLD 166 +#define SQR_TOOM6_THRESHOLD 222 +#define SQR_TOOM8_THRESHOLD 309 + +#define MULMID_TOOM42_THRESHOLD 22 + +#define MULMOD_BNM1_THRESHOLD 7 +#define SQRMOD_BNM1_THRESHOLD 12 + +#define MUL_FFT_MODF_THRESHOLD 276 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 276, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 13, 7}, { 7, 6}, { 15, 7}, { 8, 6}, \ + { 17, 7}, { 9, 6}, { 19, 7}, { 13, 8}, \ + { 7, 7}, { 17, 8}, { 9, 7}, { 20, 8}, \ + { 11, 7}, { 23, 8}, { 13, 9}, { 7, 8}, \ + { 21, 9}, { 11, 8}, { 25,10}, { 7, 9}, \ + { 15, 8}, { 33, 9}, { 19, 8}, { 39, 9}, \ + { 23, 8}, { 49, 9}, { 27,10}, { 15, 9}, \ + { 39,10}, { 23, 9}, { 51,11}, { 15,10}, \ + { 31, 9}, { 67,10}, { 39, 9}, { 79,10}, \ + { 47, 9}, { 99,10}, { 55,11}, { 31,10}, \ + { 63, 8}, { 255,10}, { 71, 9}, { 143, 8}, \ + { 287,10}, { 79, 9}, { 159, 8}, { 319,11}, \ + { 47,10}, { 95, 9}, { 191,10}, { 103,12}, \ + { 31,11}, { 63, 9}, { 255, 8}, { 511,10}, \ + { 143, 8}, { 575,11}, { 79,10}, { 159, 9}, \ + { 319,10}, { 175, 9}, { 351, 8}, { 703,11}, \ + { 95,10}, { 191, 9}, { 383,10}, { 207, 9}, \ + { 415,10}, { 223, 9}, { 447, 8}, { 895,12}, \ + { 63,10}, { 255, 9}, { 511, 8}, { 1023, 9}, \ + { 543,11}, { 143,10}, { 287, 9}, { 575, 8}, \ + { 1151,10}, { 319, 9}, { 639,11}, { 175,10}, \ + { 351, 9}, { 703,12}, { 95,10}, { 383, 9}, \ + { 767,11}, { 207, 9}, { 831,11}, { 223,10}, \ + { 447, 9}, { 895,13}, { 8192,14}, { 16384,15}, \ + { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \ + { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \ + {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 109 +#define MUL_FFT_THRESHOLD 3200 + +#define SQR_FFT_MODF_THRESHOLD 244 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 244, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 8, 5}, { 17, 6}, { 17, 7}, { 9, 6}, \ + { 19, 7}, { 17, 8}, { 9, 7}, { 20, 8}, \ + { 11, 7}, { 23, 8}, { 13, 9}, { 7, 8}, \ + { 19, 9}, { 11, 8}, { 25,10}, { 7, 9}, \ + { 15, 8}, { 33, 9}, { 19, 8}, { 39, 9}, \ + { 27,10}, { 15, 9}, { 39,10}, { 23, 9}, \ + { 47,11}, { 15,10}, { 31, 9}, { 67,10}, \ + { 39, 9}, { 79,10}, { 47, 9}, { 99,10}, \ + { 55,11}, { 31,10}, { 63, 9}, { 127, 8}, \ + { 255,10}, { 71, 8}, { 287, 7}, { 575, 9}, \ + { 159, 8}, { 319,11}, { 47,10}, { 95, 9}, \ + { 191, 8}, { 383,12}, { 31,11}, { 63,10}, \ + { 127, 9}, { 255,10}, { 143, 9}, { 287, 8}, \ + { 575,11}, { 79,10}, { 159, 9}, { 319, 8}, \ + { 639, 9}, { 351,10}, { 191, 9}, { 383,10}, \ + { 207, 9}, { 415,10}, { 239,12}, { 63,10}, \ + { 255, 9}, { 511,10}, { 271,11}, { 143,10}, \ + { 287, 9}, { 575,11}, { 159,10}, { 319, 9}, \ + { 639,10}, { 351, 9}, { 703,11}, { 191,10}, \ + { 383, 9}, { 767,11}, { 207,10}, { 415, 9}, \ + { 831,11}, { 223,13}, { 8192,14}, { 16384,15}, \ + { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \ + { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \ + {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 97 +#define SQR_FFT_THRESHOLD 2496 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 39 +#define MULLO_MUL_N_THRESHOLD 6253 +#define SQRLO_BASECASE_THRESHOLD 4 +#define SQRLO_DC_THRESHOLD 56 +#define SQRLO_SQR_THRESHOLD 4940 + +#define DC_DIV_QR_THRESHOLD 41 +#define DC_DIVAPPR_Q_THRESHOLD 136 +#define DC_BDIV_QR_THRESHOLD 39 +#define DC_BDIV_Q_THRESHOLD 89 + +#define INV_MULMOD_BNM1_THRESHOLD 22 +#define INV_NEWTON_THRESHOLD 154 +#define INV_APPR_THRESHOLD 141 + +#define BINV_NEWTON_THRESHOLD 182 +#define REDC_1_TO_REDC_N_THRESHOLD 39 + +#define MU_DIV_QR_THRESHOLD 979 +#define MU_DIVAPPR_Q_THRESHOLD 1078 +#define MUPI_DIV_QR_THRESHOLD 75 +#define MU_BDIV_QR_THRESHOLD 872 +#define MU_BDIV_Q_THRESHOLD 942 + +#define POWM_SEC_TABLE 1,19,117,539,1730 + +#define GET_STR_DC_THRESHOLD 10 +#define GET_STR_PRECOMPUTE_THRESHOLD 21 +#define SET_STR_DC_THRESHOLD 572 +#define SET_STR_PRECOMPUTE_THRESHOLD 1036 + +#define FAC_DSC_THRESHOLD 142 +#define FAC_ODD_THRESHOLD 23 + +#define MATRIX22_STRASSEN_THRESHOLD 11 +#define HGCD2_DIV1_METHOD 1 /* 8.83% faster than 3 */ +#define HGCD_THRESHOLD 80 +#define HGCD_APPR_THRESHOLD 70 +#define HGCD_REDUCE_THRESHOLD 1962 +#define GCD_DC_THRESHOLD 273 +#define GCDEXT_DC_THRESHOLD 198 +#define JACOBI_BASE_METHOD 1 /* 7.49% faster than 4 */ diff --git a/gmp-6.3.0/mpn/arm64/cora72/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/cora72/gmp-mparam.h new file mode 100644 index 0000000..fc66fd3 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cora72/gmp-mparam.h @@ -0,0 +1,242 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file for a72. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 1800 MHz Cortex-A72 */ +/* FFT tuning limit = 50,811,960 */ +/* Generated by tuneup.c, 2019-10-22, gcc 7.3 */ + +#define DIVREM_1_NORM_THRESHOLD 3 +#define DIVREM_1_UNNORM_THRESHOLD 3 +#define MOD_1_1P_METHOD 2 /* 12.09% faster than 1 */ +#define MOD_1_NORM_THRESHOLD 3 +#define MOD_1_UNNORM_THRESHOLD 3 +#define MOD_1N_TO_MOD_1_1_THRESHOLD 8 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 5 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 0 /* never mpn_mod_1_1p */ +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 26 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 15 +#define USE_PREINV_DIVREM_1 1 +#define DIV_QR_1N_PI1_METHOD 1 /* 13.42% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD 4 +#define DIV_QR_1_UNNORM_THRESHOLD 4 +#define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 38 + +#define DIV_1_VS_MUL_1_PERCENT 168 + +#define MUL_TOOM22_THRESHOLD 8 +#define MUL_TOOM33_THRESHOLD 57 +#define MUL_TOOM44_THRESHOLD 153 +#define MUL_TOOM6H_THRESHOLD 222 +#define MUL_TOOM8H_THRESHOLD 333 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 57 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 108 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 104 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 56 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 82 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 16 +#define SQR_TOOM3_THRESHOLD 73 +#define SQR_TOOM4_THRESHOLD 154 +#define SQR_TOOM6_THRESHOLD 206 +#define SQR_TOOM8_THRESHOLD 333 + +#define MULMID_TOOM42_THRESHOLD 18 + +#define MULMOD_BNM1_THRESHOLD 8 +#define SQRMOD_BNM1_THRESHOLD 10 + +#define MUL_FFT_MODF_THRESHOLD 268 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 268, 5}, { 11, 6}, { 6, 5}, { 13, 6}, \ + { 15, 7}, { 13, 8}, { 7, 7}, { 16, 8}, \ + { 9, 7}, { 19, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 15, 7}, { 31, 8}, \ + { 19, 9}, { 11, 8}, { 27,10}, { 7, 9}, \ + { 15, 8}, { 33, 9}, { 19, 8}, { 39, 9}, \ + { 27,10}, { 15, 9}, { 39,10}, { 23, 9}, \ + { 51,11}, { 15,10}, { 31, 9}, { 71,10}, \ + { 39, 9}, { 79, 8}, { 159, 7}, { 319, 9}, \ + { 83,10}, { 47, 9}, { 95, 7}, { 383, 9}, \ + { 99,10}, { 55,11}, { 31,10}, { 63, 8}, \ + { 255, 7}, { 511, 9}, { 131,10}, { 71, 9}, \ + { 143, 8}, { 287, 7}, { 575, 6}, { 1151,10}, \ + { 79, 8}, { 319, 7}, { 639,10}, { 87, 8}, \ + { 351,11}, { 47,10}, { 95, 8}, { 383, 7}, \ + { 767,10}, { 103, 8}, { 415, 7}, { 831, 6}, \ + { 1663, 9}, { 223, 8}, { 447,12}, { 31,11}, \ + { 63, 9}, { 255, 8}, { 511, 7}, { 1023, 9}, \ + { 287, 8}, { 575, 7}, { 1151, 6}, { 2303, 7}, \ + { 1215,11}, { 79, 9}, { 319, 8}, { 639, 7}, \ + { 1279, 9}, { 351, 8}, { 703, 7}, { 1407, 6}, \ + { 2815, 9}, { 383, 8}, { 831, 7}, { 1663, 9}, \ + { 447, 8}, { 895, 7}, { 1791, 6}, { 3583, 8}, \ + { 959, 6}, { 3839, 5}, { 7679, 9}, { 511, 8}, \ + { 1023, 7}, { 2175, 9}, { 575, 8}, { 1151, 7}, \ + { 2303, 8}, { 1215,10}, { 351, 9}, { 703, 7}, \ + { 3071, 8}, { 1663, 9}, { 895, 8}, { 1791, 7}, \ + { 3583, 8}, { 1919, 6}, { 7679, 7}, { 3967, 9}, \ + { 1023,10}, { 575, 9}, { 1151, 8}, { 2559,10}, \ + { 703, 8}, { 2815, 9}, { 1471, 7}, { 5887,10}, \ + { 767,11}, { 415, 9}, { 1791, 8}, { 3583,11}, \ + { 479,10}, { 959, 8}, { 3967,11}, { 511, 9}, \ + { 2175,10}, { 1151, 8}, { 4607, 9}, { 2815,10}, \ + { 1471, 9}, { 2943,11}, { 767,10}, { 1535,11}, \ + { 831,10}, { 1791,11}, { 959,10}, { 1919, 9}, \ + { 3839, 8}, { 7679,10}, { 1983,12}, { 511,10}, \ + { 2047,11}, { 1215,12}, { 639,11}, { 1407,10}, \ + { 2815,11}, { 1471,12}, { 767,11}, { 1663,12}, \ + { 895,11}, { 1791,12}, { 959,11}, { 1919,10}, \ + { 3839,14}, { 255,13}, { 511,12}, { 1023,11}, \ + { 2047,12}, { 1215,13}, { 639,12}, { 1279,13}, \ + { 703,12}, { 1407,11}, { 2815,13}, { 767,12}, \ + { 1535,13}, { 831,12}, { 1663,13}, { 895,12}, \ + { 1791,11}, { 3583,13}, { 959,12}, { 1919,11}, \ + { 3839,14}, { 511,13}, { 1023,12}, { 2047,13}, \ + { 1215,12}, { 2431,14}, { 639,13}, { 1407,12}, \ + { 2815,13}, { 1471,12}, { 2943,14}, { 767,13}, \ + { 1535,12}, { 3071,13}, { 1791,12}, { 3583,13}, \ + { 1919,14}, { 1023,13}, { 2175,14}, { 1151,13}, \ + { 2431,14}, { 1279,13}, { 2559,15}, { 767,14}, \ + { 1791,13}, { 3839,15}, { 1023,14}, { 2431,13}, \ + { 4863,15}, { 1279,14}, { 2943,15}, { 1535,14}, \ + { 16384,15}, { 32768,16}, { 65536,17}, { 131072,18}, \ + { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \ + {4194304,23}, {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 218 +#define MUL_FFT_THRESHOLD 2688 + +#define SQR_FFT_MODF_THRESHOLD 236 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 236, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 15, 7}, { 8, 6}, { 17, 7}, { 13, 8}, \ + { 7, 7}, { 17, 8}, { 9, 7}, { 20, 8}, \ + { 11, 7}, { 23, 8}, { 13, 9}, { 7, 8}, \ + { 19, 9}, { 11, 8}, { 25,10}, { 7, 9}, \ + { 15, 8}, { 33, 9}, { 19, 8}, { 39, 9}, \ + { 27,10}, { 15, 9}, { 39,10}, { 23, 9}, \ + { 47,11}, { 15,10}, { 31, 9}, { 67,10}, \ + { 39, 9}, { 79, 8}, { 159,10}, { 47, 9}, \ + { 95, 8}, { 191, 7}, { 383,10}, { 55,11}, \ + { 31,10}, { 63, 9}, { 127, 8}, { 255, 7}, \ + { 511,10}, { 71, 9}, { 143, 8}, { 287, 7}, \ + { 575,10}, { 79, 8}, { 319, 7}, { 639,11}, \ + { 47,10}, { 95, 8}, { 383, 7}, { 767, 8}, \ + { 415,12}, { 31,11}, { 63,10}, { 127, 9}, \ + { 255, 8}, { 543, 9}, { 287, 8}, { 575, 7}, \ + { 1151, 9}, { 319, 8}, { 639, 9}, { 351, 8}, \ + { 703, 7}, { 1407, 6}, { 2815,10}, { 191, 9}, \ + { 383, 8}, { 767, 9}, { 415, 8}, { 831, 7}, \ + { 1663,10}, { 223, 9}, { 447, 8}, { 895, 7}, \ + { 1791, 9}, { 479, 8}, { 959,12}, { 63,11}, \ + { 127, 9}, { 543, 8}, { 1087,10}, { 287, 9}, \ + { 575, 8}, { 1151,10}, { 319, 9}, { 639,10}, \ + { 351, 9}, { 703, 8}, { 1407, 7}, { 2815, 8}, \ + { 1471, 5}, { 11775, 9}, { 767, 8}, { 1535,10}, \ + { 415, 9}, { 895, 8}, { 1919, 6}, { 7679, 7}, \ + { 3967,11}, { 255,10}, { 543, 9}, { 1087, 8}, \ + { 2175,10}, { 575, 9}, { 1151, 8}, { 2431,10}, \ + { 639, 9}, { 1279,10}, { 703, 9}, { 1407, 8}, \ + { 2943,11}, { 383,10}, { 767,11}, { 447,10}, \ + { 895,11}, { 479,10}, { 959, 9}, { 1919, 8}, \ + { 3839,10}, { 1023, 9}, { 2175,10}, { 1215, 9}, \ + { 2431,11}, { 703, 9}, { 2815,10}, { 1471,11}, \ + { 767,10}, { 1663,11}, { 895,10}, { 1791,11}, \ + { 959, 9}, { 3839,12}, { 511,11}, { 1087,10}, \ + { 2175,11}, { 1215,10}, { 2431,12}, { 639,11}, \ + { 1279,12}, { 703,11}, { 1471,12}, { 767,11}, \ + { 1663,12}, { 895,11}, { 1919,10}, { 3839,13}, \ + { 511,12}, { 1087,11}, { 2175,12}, { 1215,11}, \ + { 2431,13}, { 639,12}, { 1279,13}, { 703,12}, \ + { 1407,13}, { 767,12}, { 1535,13}, { 831,12}, \ + { 1791,13}, { 1151,12}, { 2303,13}, { 1215,14}, \ + { 639,12}, { 2559,13}, { 1407,14}, { 767,12}, \ + { 3071,14}, { 895,13}, { 1919,12}, { 3839,14}, \ + { 1023,13}, { 2175,14}, { 1151,12}, { 4607,14}, \ + { 1279,13}, { 2559,14}, { 1407,13}, { 2943,15}, \ + { 767,14}, { 1663,13}, { 3583,14}, { 1919,15}, \ + { 1023,14}, { 2047,13}, { 4095,14}, { 2943,15}, \ + { 1535,14}, { 16384,15}, { 32768,16}, { 65536,17}, \ + { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \ + {2097152,22}, {4194304,23}, {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 203 +#define SQR_FFT_THRESHOLD 2176 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 33 +#define MULLO_MUL_N_THRESHOLD 5240 +#define SQRLO_BASECASE_THRESHOLD 6 +#define SQRLO_DC_THRESHOLD 45 +#define SQRLO_SQR_THRESHOLD 4265 + +#define DC_DIV_QR_THRESHOLD 38 +#define DC_DIVAPPR_Q_THRESHOLD 108 +#define DC_BDIV_QR_THRESHOLD 36 +#define DC_BDIV_Q_THRESHOLD 71 + +#define INV_MULMOD_BNM1_THRESHOLD 14 +#define INV_NEWTON_THRESHOLD 132 +#define INV_APPR_THRESHOLD 124 + +#define BINV_NEWTON_THRESHOLD 199 +#define REDC_1_TO_REDC_N_THRESHOLD 34 + +#define MU_DIV_QR_THRESHOLD 979 +#define MU_DIVAPPR_Q_THRESHOLD 979 +#define MUPI_DIV_QR_THRESHOLD 61 +#define MU_BDIV_QR_THRESHOLD 734 +#define MU_BDIV_Q_THRESHOLD 942 + +#define POWM_SEC_TABLE 6,30,110,579,1730 + +#define GET_STR_DC_THRESHOLD 12 +#define GET_STR_PRECOMPUTE_THRESHOLD 19 +#define SET_STR_DC_THRESHOLD 458 +#define SET_STR_PRECOMPUTE_THRESHOLD 875 + +#define FAC_DSC_THRESHOLD 153 +#define FAC_ODD_THRESHOLD 24 + +#define MATRIX22_STRASSEN_THRESHOLD 15 +#define HGCD2_DIV1_METHOD 1 /* 8.41% faster than 3 */ +#define HGCD_THRESHOLD 81 +#define HGCD_APPR_THRESHOLD 80 +#define HGCD_REDUCE_THRESHOLD 1494 +#define GCD_DC_THRESHOLD 268 +#define GCDEXT_DC_THRESHOLD 189 +#define JACOBI_BASE_METHOD 1 /* 10.80% faster than 4 */ + +/* Tuneup completed successfully, took 96906 seconds */ diff --git a/gmp-6.3.0/mpn/arm64/cora73/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/cora73/gmp-mparam.h new file mode 100644 index 0000000..7fc7f4e --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/cora73/gmp-mparam.h @@ -0,0 +1,225 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file for a73. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 1800 MHz Cortex-A72 */ +/* FFT tuning limit = 48,820,337 */ +/* Generated by tuneup.c, 2019-10-22, gcc 7.4 */ + +#define DIVREM_1_NORM_THRESHOLD 3 +#define DIVREM_1_UNNORM_THRESHOLD 3 +#define MOD_1_1P_METHOD 1 /* 2.28% faster than 2 */ +#define MOD_1_NORM_THRESHOLD 3 +#define MOD_1_UNNORM_THRESHOLD 4 +#define MOD_1N_TO_MOD_1_1_THRESHOLD 8 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 5 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 8 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 44 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 16 +#define USE_PREINV_DIVREM_1 1 +#define DIV_QR_1N_PI1_METHOD 1 /* 35.13% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD 5 +#define DIV_QR_1_UNNORM_THRESHOLD 5 +#define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 33 + +#define DIV_1_VS_MUL_1_PERCENT 168 + +#define MUL_TOOM22_THRESHOLD 10 +#define MUL_TOOM33_THRESHOLD 57 +#define MUL_TOOM44_THRESHOLD 89 +#define MUL_TOOM6H_THRESHOLD 141 +#define MUL_TOOM8H_THRESHOLD 199 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 61 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 69 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 65 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 66 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 58 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 18 +#define SQR_TOOM3_THRESHOLD 62 +#define SQR_TOOM4_THRESHOLD 166 +#define SQR_TOOM6_THRESHOLD 222 +#define SQR_TOOM8_THRESHOLD 309 + +#define MULMID_TOOM42_THRESHOLD 22 + +#define MULMOD_BNM1_THRESHOLD 8 +#define SQRMOD_BNM1_THRESHOLD 11 + +#define MUL_FFT_MODF_THRESHOLD 276 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 276, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 15, 7}, { 8, 6}, { 17, 7}, { 9, 6}, \ + { 19, 7}, { 13, 8}, { 7, 7}, { 17, 8}, \ + { 9, 7}, { 19, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 19, 9}, { 11, 8}, \ + { 27,10}, { 7, 9}, { 15, 8}, { 33, 9}, \ + { 19, 8}, { 39, 9}, { 23, 8}, { 47, 9}, \ + { 27,10}, { 15, 9}, { 43,10}, { 23, 9}, \ + { 51,11}, { 15,10}, { 31, 9}, { 67,10}, \ + { 39, 9}, { 83,10}, { 47, 9}, { 99,10}, \ + { 55,11}, { 31,10}, { 63, 9}, { 127, 8}, \ + { 255, 9}, { 131,10}, { 71, 9}, { 143, 8}, \ + { 287,10}, { 79, 9}, { 159, 8}, { 319,11}, \ + { 47, 9}, { 191, 8}, { 383, 7}, { 767, 8}, \ + { 415,12}, { 31,11}, { 63, 9}, { 255, 8}, \ + { 511,10}, { 143, 9}, { 287, 8}, { 575,11}, \ + { 79,10}, { 159, 9}, { 319,10}, { 175, 9}, \ + { 351, 8}, { 703,11}, { 95,10}, { 191, 9}, \ + { 383, 8}, { 767,10}, { 207, 9}, { 415,10}, \ + { 223, 9}, { 447,12}, { 63,10}, { 255, 9}, \ + { 511, 8}, { 1023, 9}, { 543,11}, { 143, 9}, \ + { 575,10}, { 319, 9}, { 639,10}, { 351, 9}, \ + { 703,12}, { 95,11}, { 191,10}, { 383,11}, \ + { 207,10}, { 415,11}, { 223,10}, { 447, 9}, \ + { 895,13}, { 63,11}, { 255,10}, { 511,11}, \ + { 287,10}, { 575,12}, { 159,11}, { 319,10}, \ + { 639,11}, { 351,10}, { 703, 9}, { 1407,12}, \ + { 191,11}, { 383,10}, { 767,11}, { 415,12}, \ + { 223,11}, { 447,10}, { 895,11}, { 479,10}, \ + { 959,12}, { 255,11}, { 543,10}, { 1087,11}, \ + { 575,12}, { 319,11}, { 639,12}, { 351,11}, \ + { 703,13}, { 191,12}, { 383,11}, { 767,12}, \ + { 415,11}, { 831,12}, { 447,11}, { 895,12}, \ + { 479,13}, { 255,12}, { 511,11}, { 1023,12}, \ + { 575,13}, { 319,12}, { 703,13}, { 383,12}, \ + { 831,13}, { 447,12}, { 959,14}, { 255,13}, \ + { 511,12}, { 1023,13}, { 575,12}, { 1151,13}, \ + { 703,12}, { 1407,14}, { 383,13}, { 831,12}, \ + { 1663,13}, { 959,15}, { 255,14}, { 511,13}, \ + { 1151,14}, { 639,13}, { 1407,14}, { 767,13}, \ + { 1663,14}, { 895,13}, { 1791,15}, { 511,14}, \ + { 1023,13}, { 2047,14}, { 1151,13}, { 2431,14}, \ + { 1407,15}, { 767,14}, { 1791,16}, { 511,15}, \ + { 1023,14}, { 2431,15}, { 1279,14}, { 2815,15}, \ + { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \ + { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \ + {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 185 +#define MUL_FFT_THRESHOLD 3200 + +#define SQR_FFT_MODF_THRESHOLD 244 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 244, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 17, 7}, { 9, 6}, { 19, 7}, { 17, 8}, \ + { 9, 7}, { 20, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 19, 9}, { 11, 8}, \ + { 25,10}, { 7, 9}, { 15, 8}, { 31, 9}, \ + { 19, 8}, { 39, 9}, { 23, 8}, { 47,10}, \ + { 15, 9}, { 39,10}, { 23, 9}, { 47,11}, \ + { 15,10}, { 31, 9}, { 63,10}, { 39, 9}, \ + { 79,10}, { 47, 9}, { 95,10}, { 55,11}, \ + { 31,10}, { 63, 8}, { 255,10}, { 71, 9}, \ + { 143, 8}, { 287,10}, { 79, 9}, { 159, 8}, \ + { 319,11}, { 47,10}, { 95, 9}, { 191, 8}, \ + { 383,12}, { 31,11}, { 63,10}, { 127, 9}, \ + { 287, 8}, { 575,11}, { 79,10}, { 159, 9}, \ + { 319, 8}, { 639,10}, { 175, 9}, { 351, 8}, \ + { 703,11}, { 95, 9}, { 383, 8}, { 767,10}, \ + { 207, 9}, { 415,10}, { 223, 8}, { 895,10}, \ + { 239,12}, { 63,11}, { 127,10}, { 255, 9}, \ + { 511,10}, { 287, 9}, { 575,11}, { 159,10}, \ + { 319, 9}, { 639,11}, { 175,10}, { 351, 9}, \ + { 703,11}, { 191,10}, { 383,11}, { 207,10}, \ + { 415,11}, { 223,10}, { 479,11}, { 255,10}, \ + { 511,11}, { 287,10}, { 575,12}, { 159,11}, \ + { 351,12}, { 191,11}, { 383,10}, { 767,12}, \ + { 223,11}, { 447,10}, { 895,11}, { 479,13}, \ + { 127,12}, { 255,11}, { 511,12}, { 287,10}, \ + { 1151,12}, { 319,11}, { 639,12}, { 351,11}, \ + { 703,13}, { 191,12}, { 383,11}, { 767,12}, \ + { 415,11}, { 831,12}, { 447,11}, { 895,12}, \ + { 479,11}, { 959,12}, { 511,11}, { 1023,12}, \ + { 575,11}, { 1151,13}, { 319,12}, { 639,11}, \ + { 1279,13}, { 383,12}, { 831,13}, { 447,12}, \ + { 895,14}, { 255,13}, { 511,12}, { 1023,13}, \ + { 703,14}, { 383,13}, { 831,12}, { 1663,13}, \ + { 895,15}, { 255,14}, { 511,13}, { 1151,14}, \ + { 639,13}, { 1407,14}, { 767,13}, { 1535,14}, \ + { 895,15}, { 511,14}, { 1151,13}, { 2431,14}, \ + { 1407,15}, { 767,14}, { 1791,16}, { 511,15}, \ + { 1023,14}, { 2431,15}, { 1279,14}, { 2815,15}, \ + { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \ + { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \ + {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 165 +#define SQR_FFT_THRESHOLD 2496 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 39 +#define MULLO_MUL_N_THRESHOLD 6253 +#define SQRLO_BASECASE_THRESHOLD 4 +#define SQRLO_DC_THRESHOLD 56 +#define SQRLO_SQR_THRESHOLD 4940 + +#define DC_DIV_QR_THRESHOLD 36 +#define DC_DIVAPPR_Q_THRESHOLD 136 +#define DC_BDIV_QR_THRESHOLD 35 +#define DC_BDIV_Q_THRESHOLD 88 + +#define INV_MULMOD_BNM1_THRESHOLD 30 +#define INV_NEWTON_THRESHOLD 149 +#define INV_APPR_THRESHOLD 139 + +#define BINV_NEWTON_THRESHOLD 166 +#define REDC_1_TO_REDC_N_THRESHOLD 38 + +#define MU_DIV_QR_THRESHOLD 1120 +#define MU_DIVAPPR_Q_THRESHOLD 1078 +#define MUPI_DIV_QR_THRESHOLD 68 +#define MU_BDIV_QR_THRESHOLD 889 +#define MU_BDIV_Q_THRESHOLD 942 + +#define POWM_SEC_TABLE 4,22,102,473,1730 + +#define GET_STR_DC_THRESHOLD 11 +#define GET_STR_PRECOMPUTE_THRESHOLD 22 +#define SET_STR_DC_THRESHOLD 381 +#define SET_STR_PRECOMPUTE_THRESHOLD 1042 + +#define FAC_DSC_THRESHOLD 140 +#define FAC_ODD_THRESHOLD 23 + +#define MATRIX22_STRASSEN_THRESHOLD 11 +#define HGCD2_DIV1_METHOD 1 /* 7.84% faster than 3 */ +#define HGCD_THRESHOLD 80 +#define HGCD_APPR_THRESHOLD 80 +#define HGCD_REDUCE_THRESHOLD 1679 +#define GCD_DC_THRESHOLD 273 +#define GCDEXT_DC_THRESHOLD 201 +#define JACOBI_BASE_METHOD 1 /* 1.03% faster than 4 */ + +/* Tuneup completed successfully, took 64972 seconds */ diff --git a/gmp-6.3.0/mpn/arm64/darwin.m4 b/gmp-6.3.0/mpn/arm64/darwin.m4 new file mode 100644 index 0000000..36e72fe --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/darwin.m4 @@ -0,0 +1,50 @@ +divert(-1) + +dnl m4 macros for ARM64 Darwin assembler. + +dnl Copyright 2020 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/. + + +dnl Standard commenting is with @, the default m4 # is for constants and we +dnl don't want to disable macro expansions in or after them. + +changecom + + +dnl LEA_HI(reg,gmp_symbol), LEA_LO(reg,gmp_symbol) +dnl +dnl Load the address of gmp_symbol into a register. We split this into two +dnl parts to allow separation for manual insn scheduling. TODO: Darwin allows +dnl for relaxing these two insns into an adr and a nop, but that requires the +dnl .loh pseudo for connecting them. + +define(`LEA_HI',`adrp $1, $2@GOTPAGE')dnl +define(`LEA_LO',`ldr $1, [$1, $2@GOTPAGEOFF]')dnl + +divert`'dnl diff --git a/gmp-6.3.0/mpn/arm64/divrem_1.asm b/gmp-6.3.0/mpn/arm64/divrem_1.asm new file mode 100644 index 0000000..9d5bb59 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/divrem_1.asm @@ -0,0 +1,231 @@ +dnl ARM64 mpn_divrem_1 and mpn_preinv_divrem_1. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2020 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') + +dnl TODO +dnl * Handle the most significant quotient limb for the unnormalised case +dnl specially, just like in the C code. (It is very often 0.) + +define(`qp_arg', x0) +define(`fn_arg', x1) +define(`np_arg', x2) +define(`n_arg', x3) +define(`d_arg', x4) +define(`dinv_arg', x5) +define(`cnt_arg', x6) + +define(`qp', x19) +define(`np', x20) +define(`n', x21) +define(`d', x22) +define(`fn', x24) +define(`dinv', x0) +define(`cnt', x23) +define(`tnc', x8) + +dnl mp_limb_t +dnl mpn_divrem_1 (mp_ptr qp, mp_size_t fn, +dnl mp_srcptr np, mp_size_t n, +dnl mp_limb_t d_unnorm) + +dnl mp_limb_t +dnl mpn_preinv_divrem_1 (mp_ptr qp, mp_size_t fn, +dnl mp_srcptr np, mp_size_t n, +dnl mp_limb_t d_unnorm, mp_limb_t dinv, int cnt) + +ASM_START() + +PROLOGUE(mpn_preinv_divrem_1) + cbz n_arg, L(fz) + stp x29, x30, [sp, #-80]! + mov x29, sp + stp x19, x20, [sp, #16] + stp x21, x22, [sp, #32] + stp x23, x24, [sp, #48] + + sub n, n_arg, #1 + add x7, n, fn_arg + add np, np_arg, n, lsl #3 + add qp, qp_arg, x7, lsl #3 + mov fn, fn_arg + mov d, d_arg + mov dinv, dinv_arg + tbnz d_arg, #63, L(nentry) + mov cnt, cnt_arg + b L(uentry) +EPILOGUE() + +PROLOGUE(mpn_divrem_1) + cbz n_arg, L(fz) + stp x29, x30, [sp, #-80]! + mov x29, sp + stp x19, x20, [sp, #16] + stp x21, x22, [sp, #32] + stp x23, x24, [sp, #48] + + sub n, n_arg, #1 + add x7, n, fn_arg + add np, np_arg, n, lsl #3 + add qp, qp_arg, x7, lsl #3 + mov fn, fn_arg + mov d, d_arg + tbnz d_arg, #63, L(normalised) + +L(unnorm): + clz cnt, d + lsl x0, d, cnt + bl GSYM_PREFIX`'MPN(invert_limb) +L(uentry): + lsl d, d, cnt + ldr x7, [np], #-8 + sub tnc, xzr, cnt + lsr x11, x7, tnc C r + lsl x1, x7, cnt + cbz n, L(uend) + +L(utop):ldr x7, [np], #-8 + add x2, x11, #1 + mul x10, x11, dinv + umulh x17, x11, dinv + lsr x9, x7, tnc + orr x1, x1, x9 + adds x10, x1, x10 + adc x2, x2, x17 + msub x11, d, x2, x1 + lsl x1, x7, cnt + cmp x10, x11 + add x14, x11, d + csel x11, x14, x11, cc + sbc x2, x2, xzr + cmp x11, d + bcs L(ufx) +L(uok): str x2, [qp], #-8 + sub n, n, #1 + cbnz n, L(utop) + +L(uend):add x2, x11, #1 + mul x10, x11, dinv + umulh x17, x11, dinv + adds x10, x1, x10 + adc x2, x2, x17 + msub x11, d, x2, x1 + cmp x10, x11 + add x14, x11, d + csel x11, x14, x11, cc + sbc x2, x2, xzr + subs x14, x11, d + adc x2, x2, xzr + csel x11, x14, x11, cs + str x2, [qp], #-8 + + cbnz fn, L(ftop) + lsr x0, x11, cnt + ldp x19, x20, [sp, #16] + ldp x21, x22, [sp, #32] + ldp x23, x24, [sp, #48] + ldp x29, x30, [sp], #80 + ret + +L(ufx): add x2, x2, #1 + sub x11, x11, d + b L(uok) + + +L(normalised): + mov x0, d + bl GSYM_PREFIX`'MPN(invert_limb) +L(nentry): + ldr x7, [np], #-8 + subs x14, x7, d + adc x2, xzr, xzr C hi q limb + csel x11, x14, x7, cs + b L(nok) + +L(ntop):ldr x1, [np], #-8 + add x2, x11, #1 + mul x10, x11, dinv + umulh x17, x11, dinv + adds x10, x1, x10 + adc x2, x2, x17 + msub x11, d, x2, x1 + cmp x10, x11 + add x14, x11, d + csel x11, x14, x11, cc C remainder + sbc x2, x2, xzr + cmp x11, d + bcs L(nfx) +L(nok): str x2, [qp], #-8 + sub n, n, #1 + tbz n, #63, L(ntop) + +L(nend):cbnz fn, L(frac) + mov x0, x11 + ldp x19, x20, [sp, #16] + ldp x21, x22, [sp, #32] + ldp x23, x24, [sp, #48] + ldp x29, x30, [sp], #80 + ret + +L(nfx): add x2, x2, #1 + sub x11, x11, d + b L(nok) + +L(frac):mov cnt, #0 +L(ftop):add x2, x11, #1 + mul x10, x11, dinv + umulh x17, x11, dinv + add x2, x2, x17 + msub x11, d, x2, xzr + cmp x10, x11 + add x14, x11, d + csel x11, x14, x11, cc C remainder + sbc x2, x2, xzr + str x2, [qp], #-8 + sub fn, fn, #1 + cbnz fn, L(ftop) + + lsr x0, x11, cnt + ldp x19, x20, [sp, #16] + ldp x21, x22, [sp, #32] + ldp x23, x24, [sp, #48] + ldp x29, x30, [sp], #80 + ret + +C Block zero. We need this for the degenerated case of n = 0, fn != 0. +L(fz): cbz fn_arg, L(zend) +L(ztop):str xzr, [qp_arg], #8 + sub fn_arg, fn_arg, #1 + cbnz fn_arg, L(ztop) +L(zend):mov x0, #0 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/gcd_11.asm b/gmp-6.3.0/mpn/arm64/gcd_11.asm new file mode 100644 index 0000000..d8cc3e2 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/gcd_11.asm @@ -0,0 +1,70 @@ +dnl ARM v8a mpn_gcd_11. + +dnl Based on the K7 gcd_1.asm, by Kevin Ryde. Rehacked for ARM by Torbjorn +dnl Granlund. + +dnl Copyright 2000-2002, 2005, 2009, 2011-2013 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') + +changecom(blah) + +C cycles/bit (approx) +C Cortex-A35 ? +C Cortex-A53 ? +C Cortex-A55 ? +C Cortex-A57 ? +C Cortex-A72 ? +C Cortex-A73 ? +C Cortex-A75 ? +C Cortex-A76 ? +C Cortex-A77 ? +C Numbers measured with: speed -CD -s16-64 -t48 mpn_gcd_1 + +define(`u0', `x0') +define(`v0', `x1') + +ASM_START() + TEXT + ALIGN(16) +PROLOGUE(mpn_gcd_11) + subs x3, u0, v0 C 0 + b.eq L(end) C + + ALIGN(16) +L(top): rbit x12, x3 C 1,5 + clz x12, x12 C 2 + csneg x3, x3, x3, cs C v = abs(u-v), even 1 + csel u0, v0, u0, cs C u = min(u,v) 1 + lsr v0, x3, x12 C 3 + subs x3, u0, v0 C 4 + b.ne L(top) C + +L(end): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/gcd_22.asm b/gmp-6.3.0/mpn/arm64/gcd_22.asm new file mode 100644 index 0000000..5367fea --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/gcd_22.asm @@ -0,0 +1,112 @@ +dnl ARM v8a mpn_gcd_22. + +dnl Copyright 2019 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') + +changecom(blah) + +C cycles/bit (approx) +C Cortex-A35 ? +C Cortex-A53 7.26 +C Cortex-A55 ? +C Cortex-A57 ? +C Cortex-A72 5.72 +C Cortex-A73 6.43 +C Cortex-A75 ? +C Cortex-A76 ? +C Cortex-A77 ? + + +define(`u1', `x0') +define(`u0', `x1') +define(`v1', `x2') +define(`v0', `x3') + +define(`t0', `x5') +define(`t1', `x6') +define(`cnt', `x7') +define(`tnc', `x8') + +ASM_START() +PROLOGUE(mpn_gcd_22) + + ALIGN(16) +L(top): subs t0, u0, v0 C 0 6 + cbz t0, L(lowz) + sbcs t1, u1, v1 C 1 7 + + rbit cnt, t0 C 1 + + cneg t0, t0, cc C 2 + cinv t1, t1, cc C 2 u = |u - v| +L(bck): csel v0, v0, u0, cs C 2 + csel v1, v1, u1, cs C 2 v = min(u,v) + + clz cnt, cnt C 2 + sub tnc, xzr, cnt C 3 + + lsr u0, t0, cnt C 3 + lsl x14, t1, tnc C 4 + lsr u1, t1, cnt C 3 + orr u0, u0, x14 C 5 + + orr x11, u1, v1 + cbnz x11, L(top) + + + subs x4, u0, v0 C 0 + b.eq L(end1) C + + ALIGN(16) +L(top1):rbit x12, x4 C 1,5 + clz x12, x12 C 2 + csneg x4, x4, x4, cs C v = abs(u-v), even 1 + csel u0, v0, u0, cs C u = min(u,v) 1 + lsr v0, x4, x12 C 3 + subs x4, u0, v0 C 4 + b.ne L(top1) C +L(end1):mov x0, u0 + mov x1, #0 + ret + +L(lowz):C We come here when v0 - u0 = 0 + C 1. If v1 - u1 = 0, then gcd is u = v. + C 2. Else compute gcd_21({v1,v0}, |u1-v1|) + subs t0, u1, v1 + b.eq L(end) + mov t1, #0 + rbit cnt, t0 C 1 + cneg t0, t0, cc C 2 + b L(bck) C FIXME: make conditional + +L(end): mov x0, v0 + mov x1, v1 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/gmp-mparam.h new file mode 100644 index 0000000..7c0c193 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/gmp-mparam.h @@ -0,0 +1,192 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 1536 MHz Cortex-A53 */ +/* FFT tuning limit = 0.5 M */ +/* Generated by tuneup.c, 2019-09-29, gcc 5.4 */ + +#define DIVREM_1_NORM_THRESHOLD 3 +#define DIVREM_1_UNNORM_THRESHOLD 4 +#define MOD_1_1P_METHOD 2 /* 2.08% faster than 1 */ +#define MOD_1_NORM_THRESHOLD 3 +#define MOD_1_UNNORM_THRESHOLD 4 +#define MOD_1N_TO_MOD_1_1_THRESHOLD 8 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 6 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 10 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 20 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 21 +#define USE_PREINV_DIVREM_1 1 +#define DIV_QR_1N_PI1_METHOD 1 /* 38.26% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD 13 +#define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */ +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 40 + +#define DIV_1_VS_MUL_1_PERCENT 159 + +#define MUL_TOOM22_THRESHOLD 14 +#define MUL_TOOM33_THRESHOLD 49 +#define MUL_TOOM44_THRESHOLD 82 +#define MUL_TOOM6H_THRESHOLD 173 +#define MUL_TOOM8H_THRESHOLD 236 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 81 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 76 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 81 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 80 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 74 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 18 +#define SQR_TOOM3_THRESHOLD 67 +#define SQR_TOOM4_THRESHOLD 166 +#define SQR_TOOM6_THRESHOLD 222 +#define SQR_TOOM8_THRESHOLD 333 + +#define MULMID_TOOM42_THRESHOLD 20 + +#define MULMOD_BNM1_THRESHOLD 10 +#define SQRMOD_BNM1_THRESHOLD 11 + +#define MUL_FFT_MODF_THRESHOLD 316 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 316, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 13, 7}, { 7, 6}, { 15, 7}, { 8, 6}, \ + { 17, 7}, { 9, 6}, { 19, 7}, { 17, 8}, \ + { 9, 7}, { 20, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 19, 9}, { 11, 8}, \ + { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \ + { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \ + { 15, 9}, { 39,10}, { 23, 9}, { 51,11}, \ + { 15,10}, { 31, 9}, { 71,10}, { 39, 9}, \ + { 83,10}, { 47, 9}, { 99,10}, { 55,11}, \ + { 31,10}, { 63, 9}, { 127, 8}, { 255, 9}, \ + { 131,10}, { 71, 8}, { 287,10}, { 79, 9}, \ + { 159, 8}, { 319,10}, { 87,11}, { 47,10}, \ + { 95, 9}, { 191, 8}, { 383,10}, { 103, 9}, \ + { 207, 8}, { 415,10}, { 111, 9}, { 223,12}, \ + { 31,11}, { 63, 9}, { 255, 8}, { 511,10}, \ + { 135, 9}, { 287, 8}, { 575,11}, { 79,10}, \ + { 159, 9}, { 319, 8}, { 639,10}, { 175, 9}, \ + { 351, 8}, { 703,11}, { 95,10}, { 191, 9}, \ + { 383, 8}, { 767,10}, { 207, 9}, { 415,11}, \ + { 111,10}, { 223, 9}, { 447,12}, { 63,10}, \ + { 255, 9}, { 511, 8}, { 1023, 9}, { 543,10}, \ + { 287, 9}, { 575, 8}, { 1151,11}, { 159,10}, \ + { 319, 9}, { 639,11}, { 175,10}, { 351, 9}, \ + { 703, 8}, { 1407,12}, { 95,11}, { 191,10}, \ + { 383, 9}, { 767,11}, { 207,10}, { 415, 9}, \ + { 831,11}, { 223,10}, { 447,13}, { 8192,14}, \ + { 16384,15}, { 32768,16}, { 65536,17}, { 131072,18}, \ + { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \ + {4194304,23}, {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 118 +#define MUL_FFT_THRESHOLD 3200 + +#define SQR_FFT_MODF_THRESHOLD 272 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 272, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \ + { 8, 5}, { 17, 6}, { 17, 7}, { 17, 8}, \ + { 9, 7}, { 19, 8}, { 11, 7}, { 23, 8}, \ + { 13, 9}, { 7, 8}, { 15, 7}, { 31, 8}, \ + { 19, 9}, { 11, 8}, { 27, 9}, { 15, 8}, \ + { 33, 9}, { 19, 8}, { 39, 9}, { 23, 8}, \ + { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \ + { 23, 9}, { 47,11}, { 15,10}, { 31, 9}, \ + { 67,10}, { 39, 9}, { 79,10}, { 47, 9}, \ + { 95, 8}, { 191,10}, { 55,11}, { 31,10}, \ + { 63, 8}, { 255,10}, { 71, 9}, { 143, 8}, \ + { 287,10}, { 79, 9}, { 159,11}, { 47,10}, \ + { 95, 9}, { 191, 8}, { 383, 7}, { 767,10}, \ + { 103, 9}, { 207,12}, { 31,11}, { 63, 9}, \ + { 255, 8}, { 511, 7}, { 1023, 9}, { 271,10}, \ + { 143, 9}, { 287,11}, { 79,10}, { 159, 9}, \ + { 319, 8}, { 639,10}, { 175, 9}, { 351, 8}, \ + { 703,11}, { 95,10}, { 191, 9}, { 383, 8}, \ + { 767,10}, { 207, 9}, { 415, 8}, { 831,10}, \ + { 223,12}, { 63,10}, { 255, 9}, { 511, 8}, \ + { 1023,10}, { 271,11}, { 143,10}, { 287, 9}, \ + { 575, 8}, { 1151,11}, { 159,10}, { 319, 9}, \ + { 639,11}, { 175,10}, { 351, 9}, { 703,12}, \ + { 95,11}, { 191,10}, { 383, 9}, { 767,11}, \ + { 207,10}, { 415, 9}, { 831,11}, { 223,10}, \ + { 447,13}, { 8192,14}, { 16384,15}, { 32768,16}, \ + { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \ + {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 112 +#define SQR_FFT_THRESHOLD 2688 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 38 +#define MULLO_MUL_N_THRESHOLD 6253 +#define SQRLO_BASECASE_THRESHOLD 4 +#define SQRLO_DC_THRESHOLD 67 +#define SQRLO_SQR_THRESHOLD 5240 + +#define DC_DIV_QR_THRESHOLD 42 +#define DC_DIVAPPR_Q_THRESHOLD 152 +#define DC_BDIV_QR_THRESHOLD 39 +#define DC_BDIV_Q_THRESHOLD 93 + +#define INV_MULMOD_BNM1_THRESHOLD 37 +#define INV_NEWTON_THRESHOLD 163 +#define INV_APPR_THRESHOLD 162 + +#define BINV_NEWTON_THRESHOLD 194 +#define REDC_1_TO_REDC_N_THRESHOLD 43 + +#define MU_DIV_QR_THRESHOLD 998 +#define MU_DIVAPPR_Q_THRESHOLD 998 +#define MUPI_DIV_QR_THRESHOLD 98 +#define MU_BDIV_QR_THRESHOLD 807 +#define MU_BDIV_Q_THRESHOLD 924 + +#define POWM_SEC_TABLE 6,30,194,579,1730 + +#define GET_STR_DC_THRESHOLD 15 +#define GET_STR_PRECOMPUTE_THRESHOLD 29 +#define SET_STR_DC_THRESHOLD 788 +#define SET_STR_PRECOMPUTE_THRESHOLD 1816 + +#define FAC_DSC_THRESHOLD 236 +#define FAC_ODD_THRESHOLD 24 + +#define MATRIX22_STRASSEN_THRESHOLD 10 +#define HGCD2_DIV1_METHOD 1 /* 7.05% faster than 3 */ +#define HGCD_THRESHOLD 101 +#define HGCD_APPR_THRESHOLD 104 +#define HGCD_REDUCE_THRESHOLD 1679 +#define GCD_DC_THRESHOLD 330 +#define GCDEXT_DC_THRESHOLD 242 +#define JACOBI_BASE_METHOD 4 /* 20.00% faster than 1 */ diff --git a/gmp-6.3.0/mpn/arm64/hamdist.asm b/gmp-6.3.0/mpn/arm64/hamdist.asm new file mode 100644 index 0000000..c72ca55 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/hamdist.asm @@ -0,0 +1,181 @@ +dnl ARM64 Neon mpn_hamdist -- mpn bit hamming distance. + +dnl Copyright 2013, 2014 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 Cortex-A53 4.5 +C Cortex-A57 1.9 +C X-Gene 4.36 + +C TODO +C * Consider greater unrolling. +C * Arrange to align the pointer, if that helps performance. Use the same +C read-and-mask trick we use on PCs, for simplicity and performance. (Sorry +C valgrind!) +C * Explore if explicit align directives, e.g., "[ptr:128]" help. +C * See rth's gmp-devel 2013-02/03 messages about final summation tricks. + +changecom(blah) + +C INPUT PARAMETERS +define(`ap', x0) +define(`bp', x1) +define(`n', x2) + +C We sum into 16 16-bit counters in v4,v5, but at the end we sum them and end +C up with 8 16-bit counters. Therefore, we can sum to 8(2^16-1) bits, or +C (8*2^16-1)/64 = 0x1fff limbs. We use a chunksize close to that, but which +C allows the huge count code to jump deep into the code (at L(chu)). + +define(`maxsize', 0x1fff) +define(`chunksize',0x1ff0) + +ASM_START() +PROLOGUE(mpn_hamdist) + + mov x11, #maxsize + cmp n, x11 + b.hi L(gt8k) + +L(lt8k): + movi v4.16b, #0 C clear summation register + movi v5.16b, #0 C clear summation register + + tbz n, #0, L(xx0) + sub n, n, #1 + ld1 {v0.1d}, [ap], #8 C load 1 limb + ld1 {v16.1d}, [bp], #8 C load 1 limb + eor v0.16b, v0.16b, v16.16b + cnt v6.16b, v0.16b + uadalp v4.8h, v6.16b C could also splat + +L(xx0): tbz n, #1, L(x00) + sub n, n, #2 + ld1 {v0.2d}, [ap], #16 C load 2 limbs + ld1 {v16.2d}, [bp], #16 C load 2 limbs + eor v0.16b, v0.16b, v16.16b + cnt v6.16b, v0.16b + uadalp v4.8h, v6.16b + +L(x00): tbz n, #2, L(000) + subs n, n, #4 + ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + ld1 {v16.2d,v17.2d}, [bp], #32 C load 4 limbs + b.ls L(sum) + +L(gt4): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + ld1 {v18.2d,v19.2d}, [bp], #32 C load 4 limbs + eor v0.16b, v0.16b, v16.16b + eor v1.16b, v1.16b, v17.16b + sub n, n, #4 + cnt v6.16b, v0.16b + cnt v7.16b, v1.16b + b L(mid) + +L(000): subs n, n, #8 + b.lo L(e0) + +L(chu): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + ld1 {v18.2d,v19.2d}, [bp], #32 C load 4 limbs + ld1 {v16.2d,v17.2d}, [bp], #32 C load 4 limbs + eor v2.16b, v2.16b, v18.16b + eor v3.16b, v3.16b, v19.16b + cnt v6.16b, v2.16b + cnt v7.16b, v3.16b + subs n, n, #8 + b.lo L(end) + +L(top): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + ld1 {v18.2d,v19.2d}, [bp], #32 C load 4 limbs + eor v0.16b, v0.16b, v16.16b + eor v1.16b, v1.16b, v17.16b + uadalp v4.8h, v6.16b + cnt v6.16b, v0.16b + uadalp v5.8h, v7.16b + cnt v7.16b, v1.16b +L(mid): ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + ld1 {v16.2d,v17.2d}, [bp], #32 C load 4 limbs + eor v2.16b, v2.16b, v18.16b + eor v3.16b, v3.16b, v19.16b + subs n, n, #8 + uadalp v4.8h, v6.16b + cnt v6.16b, v2.16b + uadalp v5.8h, v7.16b + cnt v7.16b, v3.16b + b.hs L(top) + +L(end): uadalp v4.8h, v6.16b + uadalp v5.8h, v7.16b +L(sum): eor v0.16b, v0.16b, v16.16b + eor v1.16b, v1.16b, v17.16b + cnt v6.16b, v0.16b + cnt v7.16b, v1.16b + uadalp v4.8h, v6.16b + uadalp v5.8h, v7.16b + add v4.8h, v4.8h, v5.8h + C we have 8 16-bit counts +L(e0): uaddlp v4.4s, v4.8h C we have 4 32-bit counts + uaddlp v4.2d, v4.4s C we have 2 64-bit counts + mov x0, v4.d[0] + mov x1, v4.d[1] + add x0, x0, x1 + ret + +C Code for count > maxsize. Splits operand and calls above code. +define(`ap2', x5) C caller-saves reg not used above +define(`bp2', x6) C caller-saves reg not used above +L(gt8k): + mov x8, x30 + mov x7, n C full count (caller-saves reg not used above) + mov x4, #0 C total sum (caller-saves reg not used above) + mov x9, #chunksize*8 C caller-saves reg not used above + mov x10, #chunksize C caller-saves reg not used above + +1: add ap2, ap, x9 C point at subsequent block + add bp2, bp, x9 C point at subsequent block + mov n, #chunksize-8 C count for this invocation, adjusted for entry pt + movi v4.16b, #0 C clear chunk summation register + movi v5.16b, #0 C clear chunk summation register + bl L(chu) C jump deep inside code + add x4, x4, x0 + mov ap, ap2 C put chunk pointer in place for calls + mov bp, bp2 C put chunk pointer in place for calls + sub x7, x7, x10 + cmp x7, x11 + b.hi 1b + + mov n, x7 C count for final invocation + bl L(lt8k) + add x0, x4, x0 + mov x30, x8 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/invert_limb.asm b/gmp-6.3.0/mpn/arm64/invert_limb.asm new file mode 100644 index 0000000..6a99bf0 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/invert_limb.asm @@ -0,0 +1,83 @@ +dnl ARM64 mpn_invert_limb -- Invert a normalized limb. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2013 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 Cortex-A53 ? +C Cortex-A57 ? + +C Compiler generated, mildly edited. Could surely be further optimised. + +ASM_START() +PROLOGUE(mpn_invert_limb) + lsr x2, x0, #54 + LEA_HI( x1, approx_tab) + and x2, x2, #0x1fe + LEA_LO( x1, approx_tab) + ldrh w3, [x1,x2] + lsr x4, x0, #24 + add x4, x4, #1 + ubfiz x2, x3, #11, #16 + umull x3, w3, w3 + mul x3, x3, x4 + sub x2, x2, #1 + sub x2, x2, x3, lsr #40 + lsl x3, x2, #60 + mul x1, x2, x2 + msub x1, x1, x4, x3 + lsl x2, x2, #13 + add x1, x2, x1, lsr #47 + and x2, x0, #1 + neg x3, x2 + and x3, x3, x1, lsr #1 + add x2, x2, x0, lsr #1 + msub x2, x1, x2, x3 + umulh x2, x2, x1 + lsl x1, x1, #31 + add x1, x1, x2, lsr #1 + mul x3, x1, x0 + umulh x2, x1, x0 + adds x4, x3, x0 + adc x0, x2, x0 + sub x0, x1, x0 + ret +EPILOGUE() + + RODATA + ALIGN(2) + TYPE( approx_tab, object) + SIZE( approx_tab, 512) +approx_tab: +forloop(i,256,512-1,dnl +` .hword eval(0x7fd00/i) +')dnl diff --git a/gmp-6.3.0/mpn/arm64/logops_n.asm b/gmp-6.3.0/mpn/arm64/logops_n.asm new file mode 100644 index 0000000..e959abc --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/logops_n.asm @@ -0,0 +1,139 @@ +dnl ARM64 mpn_and_n, mpn_andn_n. mpn_nand_n, etc. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2013 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 cycles/limb +C nand,nior all other +C Cortex-A53 3.25-3.5 2.75-3 +C Cortex-A57 2.0 1.5 +C X-Gene 2.14 2.0 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`vp', `x2') +define(`n', `x3') + +define(`POSTOP', `dnl') + +ifdef(`OPERATION_and_n',` + define(`func', `mpn_and_n') + define(`LOGOP', `and $1, $2, $3')') +ifdef(`OPERATION_andn_n',` + define(`func', `mpn_andn_n') + define(`LOGOP', `bic $1, $2, $3')') +ifdef(`OPERATION_nand_n',` + define(`func', `mpn_nand_n') + define(`POSTOP', `mvn $1, $1') + define(`LOGOP', `and $1, $2, $3')') +ifdef(`OPERATION_ior_n',` + define(`func', `mpn_ior_n') + define(`LOGOP', `orr $1, $2, $3')') +ifdef(`OPERATION_iorn_n',` + define(`func', `mpn_iorn_n') + define(`LOGOP', `orn $1, $2, $3')') +ifdef(`OPERATION_nior_n',` + define(`func', `mpn_nior_n') + define(`POSTOP', `mvn $1, $1') + define(`LOGOP', `orr $1, $2, $3')') +ifdef(`OPERATION_xor_n',` + define(`func', `mpn_xor_n') + define(`LOGOP', `eor $1, $2, $3')') +ifdef(`OPERATION_xnor_n',` + define(`func', `mpn_xnor_n') + define(`LOGOP', `eon $1, $2, $3')') + +MULFUNC_PROLOGUE(mpn_and_n mpn_andn_n mpn_nand_n mpn_ior_n mpn_iorn_n mpn_nior_n mpn_xor_n mpn_xnor_n) + +ASM_START() +PROLOGUE(func) + lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x7, [up] + ldr x11, [vp] + LOGOP( x15, x7, x11) + POSTOP( x15) + str x15, [rp],#8 + tbnz n, #1, L(b11) + +L(b01): cbz x17, L(ret) + ldp x4, x5, [up,#8] + ldp x8, x9, [vp,#8] + sub up, up, #8 + sub vp, vp, #8 + b L(mid) + +L(b11): ldp x6, x7, [up,#8] + ldp x10, x11, [vp,#8] + add up, up, #8 + add vp, vp, #8 + cbz x17, L(end) + b L(top) + +L(bx0): tbnz n, #1, L(b10) + +L(b00): ldp x4, x5, [up],#-16 + ldp x8, x9, [vp],#-16 + b L(mid) + +L(b10): ldp x6, x7, [up] + ldp x10, x11, [vp] + cbz x17, L(end) + + ALIGN(16) +L(top): ldp x4, x5, [up,#16] + ldp x8, x9, [vp,#16] + LOGOP( x12, x6, x10) + LOGOP( x13, x7, x11) + POSTOP( x12) + POSTOP( x13) + stp x12, x13, [rp],#16 +L(mid): ldp x6, x7, [up,#32]! + ldp x10, x11, [vp,#32]! + LOGOP( x12, x4, x8) + LOGOP( x13, x5, x9) + POSTOP( x12) + POSTOP( x13) + stp x12, x13, [rp],#16 + sub x17, x17, #1 + cbnz x17, L(top) + +L(end): LOGOP( x12, x6, x10) + LOGOP( x13, x7, x11) + POSTOP( x12) + POSTOP( x13) + stp x12, x13, [rp] +L(ret): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/lshift.asm b/gmp-6.3.0/mpn/arm64/lshift.asm new file mode 100644 index 0000000..fe8a1aa --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/lshift.asm @@ -0,0 +1,138 @@ +dnl ARM64 mpn_lshift. + +dnl Copyright 2013, 2014, 2017 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 assumed optimal c/l +C Cortex-A53 3.5-4.0 3.25 +C Cortex-A57 2.0 2.0 +C X-Gene 2.67 2.5 + +C TODO +C * The feed-in code used 1 ldr for odd sized and 2 ldr for even sizes. These +C numbers should be 1 and 0, respectively. The str in wind-down should also +C go. +C * Using extr and with 63 separate loops we might reach 1.25 c/l on A57. +C * A53's speed depends on alignment, tune/speed -w1 gives 3.5, -w0 gives 4.0. + +changecom(blah) + +define(`rp_arg', `x0') +define(`up', `x1') +define(`n', `x2') +define(`cnt', `x3') + +define(`rp', `x16') + +define(`tnc',`x8') + +define(`PSHIFT', lsl) +define(`NSHIFT', lsr) + +ASM_START() +PROLOGUE(mpn_lshift) + add rp, rp_arg, n, lsl #3 + add up, up, n, lsl #3 + sub tnc, xzr, cnt + lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x4, [up,#-8] + tbnz n, #1, L(b11) + +L(b01): NSHIFT x0, x4, tnc + PSHIFT x2, x4, cnt + cbnz x17, L(gt1) + str x2, [rp,#-8] + ret +L(gt1): ldp x4, x5, [up,#-24] + sub up, up, #8 + add rp, rp, #16 + b L(lo2) + +L(b11): NSHIFT x0, x4, tnc + PSHIFT x2, x4, cnt + ldp x6, x7, [up,#-24]! + b L(lo3) + +L(bx0): ldp x4, x5, [up,#-16] + tbz n, #1, L(b00) + +L(b10): NSHIFT x0, x5, tnc + PSHIFT x13, x5, cnt + NSHIFT x10, x4, tnc + PSHIFT x2, x4, cnt + cbnz x17, L(gt2) + orr x10, x10, x13 + stp x2, x10, [rp,#-16] + ret +L(gt2): ldp x4, x5, [up,#-32] + orr x10, x10, x13 + str x10, [rp,#-8] + sub up, up, #16 + add rp, rp, #8 + b L(lo2) + +L(b00): NSHIFT x0, x5, tnc + PSHIFT x13, x5, cnt + NSHIFT x10, x4, tnc + PSHIFT x2, x4, cnt + ldp x6, x7, [up,#-32]! + orr x10, x10, x13 + str x10, [rp,#-8]! + b L(lo0) + + ALIGN(16) +L(top): ldp x4, x5, [up,#-16] + orr x10, x10, x13 + orr x11, x12, x2 + stp x10, x11, [rp,#-16] + PSHIFT x2, x6, cnt +L(lo2): NSHIFT x10, x4, tnc + PSHIFT x13, x5, cnt + NSHIFT x12, x5, tnc + ldp x6, x7, [up,#-32]! + orr x10, x10, x13 + orr x11, x12, x2 + stp x10, x11, [rp,#-32]! + PSHIFT x2, x4, cnt +L(lo0): sub x17, x17, #1 +L(lo3): NSHIFT x10, x6, tnc + PSHIFT x13, x7, cnt + NSHIFT x12, x7, tnc + cbnz x17, L(top) + +L(end): orr x10, x10, x13 + orr x11, x12, x2 + PSHIFT x2, x6, cnt + stp x10, x11, [rp,#-16] + str x2, [rp,#-24] + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/lshiftc.asm b/gmp-6.3.0/mpn/arm64/lshiftc.asm new file mode 100644 index 0000000..6bf5844 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/lshiftc.asm @@ -0,0 +1,141 @@ +dnl ARM64 mpn_lshiftc. + +dnl Copyright 2013, 2014, 2017 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 assumed optimal c/l +C Cortex-A53 3.5-4.0 3.25 +C Cortex-A57 2.0 2.0 +C X-Gene 2.67 2.5 + +C TODO +C * The feed-in code used 1 ldr for odd sized and 2 ldr for even sizes. These +C numbers should be 1 and 0, respectively. The str in wind-down should also +C go. +C * Using extr and with 63 separate loops we might reach 1.5 c/l on A57. +C * A53's speed depends on alignment, tune/speed -w1 gives 3.5, -w0 gives 4.0. + +changecom(blah) + +define(`rp_arg', `x0') +define(`up', `x1') +define(`n', `x2') +define(`cnt', `x3') + +define(`rp', `x16') + +define(`tnc',`x8') + +define(`PSHIFT', lsl) +define(`NSHIFT', lsr) + +ASM_START() +PROLOGUE(mpn_lshiftc) + add rp, rp_arg, n, lsl #3 + add up, up, n, lsl #3 + sub tnc, xzr, cnt + lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x4, [up,#-8] + tbnz n, #1, L(b11) + +L(b01): NSHIFT x0, x4, tnc + PSHIFT x2, x4, cnt + cbnz x17, L(gt1) + mvn x2, x2 + str x2, [rp,#-8] + ret +L(gt1): ldp x4, x5, [up,#-24] + sub up, up, #8 + add rp, rp, #16 + b L(lo2) + +L(b11): NSHIFT x0, x4, tnc + PSHIFT x2, x4, cnt + ldp x6, x7, [up,#-24]! + b L(lo3) + +L(bx0): ldp x4, x5, [up,#-16] + tbz n, #1, L(b00) + +L(b10): NSHIFT x0, x5, tnc + PSHIFT x13, x5, cnt + NSHIFT x10, x4, tnc + PSHIFT x2, x4, cnt + cbnz x17, L(gt2) + eon x10, x10, x13 + mvn x2, x2 + stp x2, x10, [rp,#-16] + ret +L(gt2): ldp x4, x5, [up,#-32] + eon x10, x10, x13 + str x10, [rp,#-8] + sub up, up, #16 + add rp, rp, #8 + b L(lo2) + +L(b00): NSHIFT x0, x5, tnc + PSHIFT x13, x5, cnt + NSHIFT x10, x4, tnc + PSHIFT x2, x4, cnt + ldp x6, x7, [up,#-32]! + eon x10, x10, x13 + str x10, [rp,#-8]! + b L(lo0) + + ALIGN(16) +L(top): ldp x4, x5, [up,#-16] + eon x10, x10, x13 + eon x11, x12, x2 + stp x10, x11, [rp,#-16] + PSHIFT x2, x6, cnt +L(lo2): NSHIFT x10, x4, tnc + PSHIFT x13, x5, cnt + NSHIFT x12, x5, tnc + ldp x6, x7, [up,#-32]! + eon x10, x10, x13 + eon x11, x12, x2 + stp x10, x11, [rp,#-32]! + PSHIFT x2, x4, cnt +L(lo0): sub x17, x17, #1 +L(lo3): NSHIFT x10, x6, tnc + PSHIFT x13, x7, cnt + NSHIFT x12, x7, tnc + cbnz x17, L(top) + +L(end): eon x10, x10, x13 + eon x11, x12, x2 + PSHIFT x2, x6, cnt + stp x10, x11, [rp,#-16] + mvn x2, x2 + str x2, [rp,#-24] + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/mod_34lsub1.asm b/gmp-6.3.0/mpn/arm64/mod_34lsub1.asm new file mode 100644 index 0000000..7945fe7 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/mod_34lsub1.asm @@ -0,0 +1,124 @@ +dnl ARM64 mpn_mod_34lsub1 -- remainder modulo 2^48-1. + +dnl Copyright 2012-2014 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 Cortex-A53 2 +C Cortex-A57 1 +C X-Gene 1.45 + +define(`ap', x0) +define(`n', x1) + +changecom(blah) + +C mp_limb_t mpn_mod_34lsub1 (mp_srcptr up, mp_size_t n) + +C TODO +C * An alternative inner loop which could run at 0.722 c/l on A57: +C adds x8, x8, x2 +C adcs x9, x9, x3 +C ldp x2, x3, [ap, #-32] +C adcs x10, x10, x4 +C adc x12, x12, xzr +C adds x8, x8, x5 +C ldp x4, x5, [ap, #-16] +C sub n, n, #6 +C adcs x9, x9, x6 +C adcs x10, x10, x7 +C ldp x6, x7, [ap], #48 +C adc x12, x12, xzr +C tbz n, #63, L(top) + +ASM_START() + TEXT + ALIGN(32) +PROLOGUE(mpn_mod_34lsub1) + subs n, n, #3 + mov x8, #0 + b.lt L(le2) C n <= 2 + + ldp x2, x3, [ap, #0] + ldr x4, [ap, #16] + add ap, ap, #24 + subs n, n, #3 + b.lt L(sum) C n <= 5 + cmn x0, #0 C clear carry + +L(top): ldp x5, x6, [ap, #0] + ldr x7, [ap, #16] + add ap, ap, #24 + sub n, n, #3 + adcs x2, x2, x5 + adcs x3, x3, x6 + adcs x4, x4, x7 + tbz n, #63, L(top) + + adc x8, xzr, xzr C x8 <= 1 + +L(sum): cmn n, #2 + mov x5, #0 + b.lo 1f + ldr x5, [ap], #8 +1: mov x6, #0 + b.ls 1f + ldr x6, [ap], #8 +1: adds x2, x2, x5 + adcs x3, x3, x6 + adcs x4, x4, xzr + adc x8, x8, xzr C x8 <= 2 + +L(sum2): + and x0, x2, #0xffffffffffff + add x0, x0, x2, lsr #48 + add x0, x0, x8 + + lsl x8, x3, #16 + and x1, x8, #0xffffffffffff + add x0, x0, x1 + add x0, x0, x3, lsr #32 + + lsl x8, x4, #32 + and x1, x8, #0xffffffffffff + add x0, x0, x1 + add x0, x0, x4, lsr #16 + ret + +L(le2): cmn n, #1 + b.ne L(1) + ldp x2, x3, [ap] + mov x4, #0 + b L(sum2) +L(1): ldr x2, [ap] + and x0, x2, #0xffffffffffff + add x0, x0, x2, lsr #48 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/mul_1.asm b/gmp-6.3.0/mpn/arm64/mul_1.asm new file mode 100644 index 0000000..fb965ef --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/mul_1.asm @@ -0,0 +1,128 @@ +dnl ARM64 mpn_mul_1 + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2013, 2015, 2017 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 Cortex-A53 7.5-8 +C Cortex-A57 7 +C Cortex-A72 +C X-Gene 4 +C Apple M1 1 + +C TODO +C * Start first multiply earlier. + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`n', `x2') +define(`v0', `x3') + + +PROLOGUE(mpn_mul_1c) + adds xzr, xzr, xzr C clear cy flag + b L(com) +EPILOGUE() + +PROLOGUE(mpn_mul_1) + adds x4, xzr, xzr C clear register and cy flag +L(com): lsr x17, n, #2 + tbnz n, #0, L(bx1) + +L(bx0): mov x11, x4 + tbz n, #1, L(b00) + +L(b10): ldp x4, x5, [up] + mul x8, x4, v0 + umulh x10, x4, v0 + cbz x17, L(2) + ldp x6, x7, [up,#16]! + mul x9, x5, v0 + b L(mid)-8 + +L(2): mul x9, x5, v0 + b L(2e) + +L(bx1): ldr x7, [up],#8 + mul x9, x7, v0 + umulh x11, x7, v0 + adds x9, x9, x4 + str x9, [rp],#8 + tbnz n, #1, L(b10) + +L(b01): cbz x17, L(1) + +L(b00): ldp x6, x7, [up] + mul x8, x6, v0 + umulh x10, x6, v0 + ldp x4, x5, [up,#16] + mul x9, x7, v0 + adcs x12, x8, x11 + umulh x11, x7, v0 + add rp, rp, #16 + sub x17, x17, #1 + cbz x17, L(end) + + ALIGN(16) +L(top): mul x8, x4, v0 + ldp x6, x7, [up,#32]! + adcs x13, x9, x10 + umulh x10, x4, v0 + mul x9, x5, v0 + stp x12, x13, [rp,#-16] + adcs x12, x8, x11 + umulh x11, x5, v0 +L(mid): mul x8, x6, v0 + ldp x4, x5, [up,#16] + adcs x13, x9, x10 + umulh x10, x6, v0 + mul x9, x7, v0 + stp x12, x13, [rp],#32 + adcs x12, x8, x11 + umulh x11, x7, v0 + sub x17, x17, #1 + cbnz x17, L(top) + +L(end): mul x8, x4, v0 + adcs x13, x9, x10 + umulh x10, x4, v0 + mul x9, x5, v0 + stp x12, x13, [rp,#-16] +L(2e): adcs x12, x8, x11 + umulh x11, x5, v0 + adcs x13, x9, x10 + stp x12, x13, [rp] +L(1): adc x0, x11, xzr + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/popcount.asm b/gmp-6.3.0/mpn/arm64/popcount.asm new file mode 100644 index 0000000..74de3fc --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/popcount.asm @@ -0,0 +1,157 @@ +dnl ARM64 Neon mpn_popcount -- mpn bit population count. + +dnl Copyright 2013, 2014 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 Cortex-A53 2.5 +C Cortex-A57 1.14 +C X-Gene 3 + +C TODO +C * Consider greater unrolling. +C * Arrange to align the pointer, if that helps performance. Use the same +C read-and-mask trick we use on PCs, for simplicity and performance. (Sorry +C valgrind!) +C * Explore if explicit align directives, e.g., "[ptr:128]" help. +C * See rth's gmp-devel 2013-02/03 messages about final summation tricks. + +changecom(blah) + +C INPUT PARAMETERS +define(`ap', x0) +define(`n', x1) + +C We sum into 16 16-bit counters in v4,v5, but at the end we sum them and end +C up with 8 16-bit counters. Therefore, we can sum to 8(2^16-1) bits, or +C (8*2^16-1)/64 = 0x1fff limbs. We use a chunksize close to that, but which +C allows the huge count code to jump deep into the code (at L(chu)). + +define(`maxsize', 0x1fff) +define(`chunksize',0x1ff0) + +ASM_START() +PROLOGUE(mpn_popcount) + + mov x11, #maxsize + cmp n, x11 + b.hi L(gt8k) + +L(lt8k): + movi v4.16b, #0 C clear summation register + movi v5.16b, #0 C clear summation register + + tbz n, #0, L(xx0) + sub n, n, #1 + ld1 {v0.1d}, [ap], #8 C load 1 limb + cnt v6.16b, v0.16b + uadalp v4.8h, v6.16b C could also splat + +L(xx0): tbz n, #1, L(x00) + sub n, n, #2 + ld1 {v0.2d}, [ap], #16 C load 2 limbs + cnt v6.16b, v0.16b + uadalp v4.8h, v6.16b + +L(x00): tbz n, #2, L(000) + subs n, n, #4 + ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + b.ls L(sum) + +L(gt4): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + sub n, n, #4 + cnt v6.16b, v0.16b + cnt v7.16b, v1.16b + b L(mid) + +L(000): subs n, n, #8 + b.lo L(e0) + +L(chu): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + cnt v6.16b, v2.16b + cnt v7.16b, v3.16b + subs n, n, #8 + b.lo L(end) + +L(top): ld1 {v2.2d,v3.2d}, [ap], #32 C load 4 limbs + uadalp v4.8h, v6.16b + cnt v6.16b, v0.16b + uadalp v5.8h, v7.16b + cnt v7.16b, v1.16b +L(mid): ld1 {v0.2d,v1.2d}, [ap], #32 C load 4 limbs + subs n, n, #8 + uadalp v4.8h, v6.16b + cnt v6.16b, v2.16b + uadalp v5.8h, v7.16b + cnt v7.16b, v3.16b + b.hs L(top) + +L(end): uadalp v4.8h, v6.16b + uadalp v5.8h, v7.16b +L(sum): cnt v6.16b, v0.16b + cnt v7.16b, v1.16b + uadalp v4.8h, v6.16b + uadalp v5.8h, v7.16b + add v4.8h, v4.8h, v5.8h + C we have 8 16-bit counts +L(e0): uaddlp v4.4s, v4.8h C we have 4 32-bit counts + uaddlp v4.2d, v4.4s C we have 2 64-bit counts + mov x0, v4.d[0] + mov x1, v4.d[1] + add x0, x0, x1 + ret + +C Code for count > maxsize. Splits operand and calls above code. +define(`ap2', x5) C caller-saves reg not used above +L(gt8k): + mov x8, x30 + mov x7, n C full count (caller-saves reg not used above) + mov x4, #0 C total sum (caller-saves reg not used above) + mov x9, #chunksize*8 C caller-saves reg not used above + mov x10, #chunksize C caller-saves reg not used above + +1: add ap2, ap, x9 C point at subsequent block + mov n, #chunksize-8 C count for this invocation, adjusted for entry pt + movi v4.16b, #0 C clear chunk summation register + movi v5.16b, #0 C clear chunk summation register + bl L(chu) C jump deep inside code + add x4, x4, x0 + mov ap, ap2 C put chunk pointer in place for calls + sub x7, x7, x10 + cmp x7, x11 + b.hi 1b + + mov n, x7 C count for final invocation + bl L(lt8k) + add x0, x4, x0 + mov x30, x8 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/rsh1aors_n.asm b/gmp-6.3.0/mpn/arm64/rsh1aors_n.asm new file mode 100644 index 0000000..afd3d5b --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/rsh1aors_n.asm @@ -0,0 +1,168 @@ +dnl ARM64 mpn_rsh1add_n and mpn_rsh1sub_n. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2017 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 assumed optimal c/l +C Cortex-A53 3.25-3.75 3.0 steady +C Cortex-A57 2.15 1.75 +C X-Gene 2.75 2.5 + +changecom(blah) + +define(`rp', `x0') +define(`up', `x1') +define(`vp', `x2') +define(`n', `x3') + +ifdef(`OPERATION_rsh1add_n', ` + define(`ADDSUB', adds) + define(`ADDSUBC', adcs) + define(`COND', `cs') + define(`func_n', mpn_rsh1add_n)') +ifdef(`OPERATION_rsh1sub_n', ` + define(`ADDSUB', subs) + define(`ADDSUBC', sbcs) + define(`COND', `cc') + define(`func_n', mpn_rsh1sub_n)') + +MULFUNC_PROLOGUE(mpn_rsh1add_n mpn_rsh1sub_n) + +ASM_START() +PROLOGUE(func_n) + lsr x6, n, #2 + + tbz n, #0, L(bx0) + +L(bx1): ldr x5, [up],#8 + ldr x9, [vp],#8 + tbnz n, #1, L(b11) + +L(b01): ADDSUB x13, x5, x9 + and x10, x13, #1 + cbz x6, L(1) + ldp x4, x5, [up],#48 + ldp x8, x9, [vp],#48 + ADDSUBC x14, x4, x8 + ADDSUBC x15, x5, x9 + ldp x4, x5, [up,#-32] + ldp x8, x9, [vp,#-32] + extr x17, x14, x13, #1 + ADDSUBC x12, x4, x8 + ADDSUBC x13, x5, x9 + str x17, [rp], #24 + sub x6, x6, #1 + cbz x6, L(end) + b L(top) + +L(1): cset x14, COND + extr x17, x14, x13, #1 + str x17, [rp] + mov x0, x10 + ret + +L(b11): ADDSUB x15, x5, x9 + and x10, x15, #1 + + ldp x4, x5, [up],#32 + ldp x8, x9, [vp],#32 + ADDSUBC x12, x4, x8 + ADDSUBC x13, x5, x9 + cbz x6, L(3) + ldp x4, x5, [up,#-16] + ldp x8, x9, [vp,#-16] + extr x17, x12, x15, #1 + ADDSUBC x14, x4, x8 + ADDSUBC x15, x5, x9 + str x17, [rp], #8 + b L(mid) + +L(3): extr x17, x12, x15, #1 + str x17, [rp], #8 + b L(2) + +L(bx0): tbz n, #1, L(b00) + +L(b10): ldp x4, x5, [up],#32 + ldp x8, x9, [vp],#32 + ADDSUB x12, x4, x8 + ADDSUBC x13, x5, x9 + and x10, x12, #1 + cbz x6, L(2) + ldp x4, x5, [up,#-16] + ldp x8, x9, [vp,#-16] + ADDSUBC x14, x4, x8 + ADDSUBC x15, x5, x9 + b L(mid) + +L(b00): ldp x4, x5, [up],#48 + ldp x8, x9, [vp],#48 + ADDSUB x14, x4, x8 + ADDSUBC x15, x5, x9 + and x10, x14, #1 + ldp x4, x5, [up,#-32] + ldp x8, x9, [vp,#-32] + ADDSUBC x12, x4, x8 + ADDSUBC x13, x5, x9 + add rp, rp, #16 + sub x6, x6, #1 + cbz x6, L(end) + + ALIGN(16) +L(top): ldp x4, x5, [up,#-16] + ldp x8, x9, [vp,#-16] + extr x16, x15, x14, #1 + extr x17, x12, x15, #1 + ADDSUBC x14, x4, x8 + ADDSUBC x15, x5, x9 + stp x16, x17, [rp,#-16] +L(mid): ldp x4, x5, [up],#32 + ldp x8, x9, [vp],#32 + extr x16, x13, x12, #1 + extr x17, x14, x13, #1 + ADDSUBC x12, x4, x8 + ADDSUBC x13, x5, x9 + stp x16, x17, [rp],#32 + sub x6, x6, #1 + cbnz x6, L(top) + +L(end): extr x16, x15, x14, #1 + extr x17, x12, x15, #1 + stp x16, x17, [rp,#-16] +L(2): cset x14, COND + extr x16, x13, x12, #1 + extr x17, x14, x13, #1 + stp x16, x17, [rp] + +L(ret): mov x0, x10 + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/rshift.asm b/gmp-6.3.0/mpn/arm64/rshift.asm new file mode 100644 index 0000000..90187ad --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/rshift.asm @@ -0,0 +1,136 @@ +dnl ARM64 mpn_rshift. + +dnl Copyright 2013, 2014, 2017 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 assumed optimal c/l +C Cortex-A53 3.5-4.0 3.25 +C Cortex-A57 2.0 2.0 +C X-Gene 2.67 2.5 + +C TODO +C * The feed-in code used 1 ldr for odd sized and 2 ldr for even sizes. These +C numbers should be 1 and 0, respectively. The str in wind-down should also +C go. +C * Using extr and with 63 separate loops we might reach 1.25 c/l on A57. +C * A53's speed depends on alignment, but not as simply as for lshift/lshiftc. + +changecom(blah) + +define(`rp_arg', `x0') +define(`up', `x1') +define(`n', `x2') +define(`cnt', `x3') + +define(`rp', `x16') + +define(`tnc',`x8') + +define(`PSHIFT', lsr) +define(`NSHIFT', lsl) + +ASM_START() +PROLOGUE(mpn_rshift) + mov rp, rp_arg + sub tnc, xzr, cnt + lsr x17, n, #2 + tbz n, #0, L(bx0) + +L(bx1): ldr x5, [up] + tbnz n, #1, L(b11) + +L(b01): NSHIFT x0, x5, tnc + PSHIFT x2, x5, cnt + cbnz x17, L(gt1) + str x2, [rp] + ret +L(gt1): ldp x4, x5, [up,#8] + sub up, up, #8 + sub rp, rp, #32 + b L(lo2) + +L(b11): NSHIFT x0, x5, tnc + PSHIFT x2, x5, cnt + ldp x6, x7, [up,#8]! + sub rp, rp, #16 + b L(lo3) + +L(bx0): ldp x4, x5, [up] + tbz n, #1, L(b00) + +L(b10): NSHIFT x0, x4, tnc + PSHIFT x13, x4, cnt + NSHIFT x10, x5, tnc + PSHIFT x2, x5, cnt + cbnz x17, L(gt2) + orr x10, x10, x13 + stp x10, x2, [rp] + ret +L(gt2): ldp x4, x5, [up,#16] + orr x10, x10, x13 + str x10, [rp],#-24 + b L(lo2) + +L(b00): NSHIFT x0, x4, tnc + PSHIFT x13, x4, cnt + NSHIFT x10, x5, tnc + PSHIFT x2, x5, cnt + ldp x6, x7, [up,#16]! + orr x10, x10, x13 + str x10, [rp],#-8 + b L(lo0) + + ALIGN(16) +L(top): ldp x4, x5, [up,#16] + orr x10, x10, x13 + orr x11, x12, x2 + stp x11, x10, [rp,#16] + PSHIFT x2, x7, cnt +L(lo2): NSHIFT x10, x5, tnc + NSHIFT x12, x4, tnc + PSHIFT x13, x4, cnt + ldp x6, x7, [up,#32]! + orr x10, x10, x13 + orr x11, x12, x2 + stp x11, x10, [rp,#32]! + PSHIFT x2, x5, cnt +L(lo0): sub x17, x17, #1 +L(lo3): NSHIFT x10, x7, tnc + NSHIFT x12, x6, tnc + PSHIFT x13, x6, cnt + cbnz x17, L(top) + +L(end): orr x10, x10, x13 + orr x11, x12, x2 + PSHIFT x2, x7, cnt + stp x11, x10, [rp,#16] + str x2, [rp,#32] + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/sec_tabselect.asm b/gmp-6.3.0/mpn/arm64/sec_tabselect.asm new file mode 100644 index 0000000..18a268a --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/sec_tabselect.asm @@ -0,0 +1,122 @@ +dnl ARM64 Neon mpn_sec_tabselect. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2011-2014 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 Cortex-A53 2.25 +C Cortex-A57 1.33 +C X-Gene 2 + +C void +C mpn_sec_tabselect (mp_ptr rp, mp_srcptr *tab, +C mp_size_t n, mp_size_t nents, mp_size_t which) + +changecom(blah) + +define(`rp', `x0') +define(`tp', `x1') +define(`n', `x2') +define(`nents', `x3') +define(`which', `x4') + +define(`i', `x5') +define(`j', `x6') + +define(`maskq', `v4') + +ASM_START() +PROLOGUE(mpn_sec_tabselect) + dup v7.2d, x4 C 2 `which' copies + + mov x10, #1 + dup v6.2d, x10 C 2 copies of 1 + + subs j, n, #4 + b.mi L(outer_end) + +L(outer_top): + mov i, nents + mov x12, tp C preserve tp + movi v5.16b, #0 C zero 2 counter copies + movi v2.16b, #0 + movi v3.16b, #0 + ALIGN(16) +L(tp4): cmeq maskq.2d, v5.2d, v7.2d C compare idx copies to `which' copies + ld1 {v0.2d,v1.2d}, [tp] + add v5.2d, v5.2d, v6.2d + bit v2.16b, v0.16b, maskq.16b + bit v3.16b, v1.16b, maskq.16b + add tp, tp, n, lsl #3 + sub i, i, #1 + cbnz i, L(tp4) + st1 {v2.2d,v3.2d}, [rp], #32 + add tp, x12, #32 C restore tp, point to next slice + subs j, j, #4 + b.pl L(outer_top) +L(outer_end): + + tbz n, #1, L(b0x) + mov i, nents + mov x12, tp + movi v5.16b, #0 C zero 2 counter copies + movi v2.16b, #0 + ALIGN(16) +L(tp2): cmeq maskq.2d, v5.2d, v7.2d + ld1 {v0.2d}, [tp] + add v5.2d, v5.2d, v6.2d + bit v2.16b, v0.16b, maskq.16b + add tp, tp, n, lsl #3 + sub i, i, #1 + cbnz i, L(tp2) + st1 {v2.2d}, [rp], #16 + add tp, x12, #16 + +L(b0x): tbz n, #0, L(b00) + mov i, nents + mov x12, tp + movi v5.16b, #0 C zero 2 counter copies + movi v2.16b, #0 + ALIGN(16) +L(tp1): cmeq maskq.2d, v5.2d, v7.2d + ld1 {v0.1d}, [tp] + add v5.2d, v5.2d, v6.2d C FIXME size should be `1d' + bit v2.8b, v0.8b, maskq.8b + add tp, tp, n, lsl #3 + sub i, i, #1 + cbnz i, L(tp1) + st1 {v2.1d}, [rp], #8 + add tp, x12, #8 + +L(b00): ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/sqr_diag_addlsh1.asm b/gmp-6.3.0/mpn/arm64/sqr_diag_addlsh1.asm new file mode 100644 index 0000000..39f1cb1 --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/sqr_diag_addlsh1.asm @@ -0,0 +1,102 @@ +dnl ARM64 mpn_sqr_diag_addlsh1. + +dnl Contributed to the GNU project by Torbjörn Granlund. + +dnl Copyright 2016, 2017 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 Cortex-A53 5.65 +C Cortex-A57 3.5 +C X-Gene 3.38 + +changecom(blah) + +define(`rp', `x0') +define(`tp', `x1') +define(`up', `x2') +define(`n', `x3') + +ASM_START() +PROLOGUE(mpn_sqr_diag_addlsh1) + ldr x15, [up],#8 + lsr x14, n, #1 + tbz n, #0, L(bx0) + +L(bx1): adds x7, xzr, xzr + mul x12, x15, x15 + ldr x16, [up],#8 + ldp x4, x5, [tp],#16 + umulh x11, x15, x15 + b L(mid) + +L(bx0): adds x5, xzr, xzr + mul x12, x15, x15 + ldr x17, [up],#16 + ldp x6, x7, [tp],#32 + umulh x11, x15, x15 + sub x14, x14, #1 + cbz x14, L(end) + + ALIGN(16) +L(top): extr x9, x6, x5, #63 + mul x10, x17, x17 + ldr x16, [up,#-8] + adcs x13, x9, x11 + ldp x4, x5, [tp,#-16] + umulh x11, x17, x17 + extr x8, x7, x6, #63 + stp x12, x13, [rp],#16 + adcs x12, x8, x10 +L(mid): extr x9, x4, x7, #63 + mul x10, x16, x16 + ldr x17, [up],#16 + adcs x13, x9, x11 + ldp x6, x7, [tp],#32 + umulh x11, x16, x16 + extr x8, x5, x4, #63 + stp x12, x13, [rp],#16 + adcs x12, x8, x10 + sub x14, x14, #1 + cbnz x14, L(top) + +L(end): extr x9, x6, x5, #63 + mul x10, x17, x17 + adcs x13, x9, x11 + umulh x11, x17, x17 + extr x8, x7, x6, #63 + stp x12, x13, [rp] + adcs x12, x8, x10 + extr x9, xzr, x7, #63 + adcs x13, x9, x11 + stp x12, x13, [rp,#16] + + ret +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm64/xgene1/gmp-mparam.h b/gmp-6.3.0/mpn/arm64/xgene1/gmp-mparam.h new file mode 100644 index 0000000..c8020bb --- /dev/null +++ b/gmp-6.3.0/mpn/arm64/xgene1/gmp-mparam.h @@ -0,0 +1,182 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file. + +Copyright 2019 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/. */ + +#define GMP_LIMB_BITS 64 +#define GMP_LIMB_BYTES 8 + +/* 2400 MHz AppliedMicro X-Gene */ +/* FFT tuning limit = 0.5 M */ +/* Generated by tuneup.c, 2019-09-28, gcc 4.8 */ + +#define DIVREM_1_NORM_THRESHOLD 0 /* always */ +#define DIVREM_1_UNNORM_THRESHOLD 0 /* always */ +#define MOD_1_1P_METHOD 1 /* 2.00% faster than 2 */ +#define MOD_1_NORM_THRESHOLD 0 /* always */ +#define MOD_1_UNNORM_THRESHOLD 0 /* always */ +#define MOD_1N_TO_MOD_1_1_THRESHOLD 7 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 5 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 0 /* never mpn_mod_1_1p */ +#define MOD_1_2_TO_MOD_1_4_THRESHOLD 22 +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 13 +#define USE_PREINV_DIVREM_1 1 +/* From gcc185.osuosl.org, 2023-07-26 */ +#define DIV_QR_1N_PI1_METHOD 3 /* 5.60% faster than 4 */ +#define DIV_QR_1_NORM_THRESHOLD 1 +#define DIV_QR_1_UNNORM_THRESHOLD 1 +#define DIV_QR_2_PI2_THRESHOLD 14 +#define DIVEXACT_1_THRESHOLD 0 /* always */ +#define BMOD_1_TO_MOD_1_THRESHOLD 27 + +#define DIV_1_VS_MUL_1_PERCENT 249 + +#define MUL_TOOM22_THRESHOLD 18 +#define MUL_TOOM33_THRESHOLD 61 +#define MUL_TOOM44_THRESHOLD 112 +#define MUL_TOOM6H_THRESHOLD 242 +#define MUL_TOOM8H_THRESHOLD 321 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 73 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 99 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 109 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 72 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 106 + +#define SQR_BASECASE_THRESHOLD 0 /* always */ +#define SQR_TOOM2_THRESHOLD 28 +#define SQR_TOOM3_THRESHOLD 81 +#define SQR_TOOM4_THRESHOLD 154 +#define SQR_TOOM6_THRESHOLD 214 +#define SQR_TOOM8_THRESHOLD 284 + +#define MULMID_TOOM42_THRESHOLD 46 + +#define MULMOD_BNM1_THRESHOLD 11 +#define SQRMOD_BNM1_THRESHOLD 13 + +#define MUL_FFT_MODF_THRESHOLD 412 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 412, 5}, { 15, 6}, { 8, 5}, { 17, 6}, \ + { 19, 7}, { 12, 6}, { 25, 7}, { 17, 8}, \ + { 9, 7}, { 20, 8}, { 11, 7}, { 25, 8}, \ + { 13, 7}, { 27, 8}, { 15, 7}, { 31, 8}, \ + { 17, 7}, { 35, 8}, { 21, 9}, { 11, 8}, \ + { 27, 9}, { 15, 8}, { 35, 9}, { 19, 8}, \ + { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \ + { 15, 9}, { 31, 8}, { 63, 9}, { 39,10}, \ + { 23, 9}, { 55,11}, { 15,10}, { 31, 9}, \ + { 71,10}, { 39, 9}, { 83,10}, { 47, 9}, \ + { 99,10}, { 55,11}, { 31,10}, { 63, 9}, \ + { 127,10}, { 71, 9}, { 143,10}, { 79,11}, \ + { 47,10}, { 103,12}, { 31,11}, { 63,10}, \ + { 127, 9}, { 255,10}, { 135, 9}, { 271,10}, \ + { 143,11}, { 79, 9}, { 319,10}, { 167, 9}, \ + { 351,11}, { 95, 9}, { 383, 8}, { 767,10}, \ + { 207, 9}, { 415,11}, { 111,12}, { 63,11}, \ + { 127,10}, { 255, 9}, { 511,10}, { 271, 9}, \ + { 543,11}, { 143,10}, { 287, 9}, { 575,10}, \ + { 319, 9}, { 639,10}, { 351,12}, { 95,10}, \ + { 383, 9}, { 767,11}, { 207,10}, { 415, 9}, \ + { 831,11}, { 223,10}, { 447,13}, { 8192,14}, \ + { 16384,15}, { 32768,16}, { 65536,17}, { 131072,18}, \ + { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \ + {4194304,23}, {8388608,24} } +#define MUL_FFT_TABLE3_SIZE 98 +#define MUL_FFT_THRESHOLD 4736 + +#define SQR_FFT_MODF_THRESHOLD 340 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 340, 5}, { 17, 6}, { 9, 5}, { 19, 6}, \ + { 19, 7}, { 10, 6}, { 21, 7}, { 11, 6}, \ + { 23, 7}, { 21, 8}, { 11, 7}, { 24, 8}, \ + { 15, 7}, { 31, 8}, { 21, 9}, { 11, 8}, \ + { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \ + { 39, 9}, { 23, 8}, { 47, 9}, { 27,10}, \ + { 15, 9}, { 31, 8}, { 63, 9}, { 39,10}, \ + { 23, 9}, { 51,11}, { 15,10}, { 31, 9}, \ + { 67,10}, { 39, 9}, { 79,10}, { 47, 9}, \ + { 95,10}, { 55,11}, { 31,10}, { 79,11}, \ + { 47,10}, { 95, 9}, { 191,12}, { 31,11}, \ + { 63,10}, { 127, 9}, { 255, 8}, { 511,10}, \ + { 135, 9}, { 271,11}, { 79, 9}, { 319, 8}, \ + { 639,10}, { 175,11}, { 95,10}, { 191, 9}, \ + { 383,10}, { 207,12}, { 63,11}, { 127,10}, \ + { 255, 9}, { 511,10}, { 271, 9}, { 543,10}, \ + { 287, 9}, { 575,10}, { 319, 9}, { 639,11}, \ + { 175,10}, { 351,12}, { 95,11}, { 191,10}, \ + { 383, 9}, { 767,11}, { 207,10}, { 415,13}, \ + { 8192,14}, { 16384,15}, { 32768,16}, { 65536,17}, \ + { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \ + {2097152,22}, {4194304,23}, {8388608,24} } +#define SQR_FFT_TABLE3_SIZE 87 +#define SQR_FFT_THRESHOLD 3264 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 45 +#define MULLO_MUL_N_THRESHOLD 8648 +#define SQRLO_BASECASE_THRESHOLD 0 /* always */ +#define SQRLO_DC_THRESHOLD 108 +#define SQRLO_SQR_THRESHOLD 6461 + +#define DC_DIV_QR_THRESHOLD 64 +#define DC_DIVAPPR_Q_THRESHOLD 222 +#define DC_BDIV_QR_THRESHOLD 63 +#define DC_BDIV_Q_THRESHOLD 132 + +#define INV_MULMOD_BNM1_THRESHOLD 38 +#define INV_NEWTON_THRESHOLD 242 +#define INV_APPR_THRESHOLD 222 + +#define BINV_NEWTON_THRESHOLD 254 +#define REDC_1_TO_REDC_N_THRESHOLD 66 + +#define MU_DIV_QR_THRESHOLD 1234 +#define MU_DIVAPPR_Q_THRESHOLD 1234 +#define MUPI_DIV_QR_THRESHOLD 122 +#define MU_BDIV_QR_THRESHOLD 1210 +#define MU_BDIV_Q_THRESHOLD 1234 + +#define POWM_SEC_TABLE 3,23,194,712,2499 + +#define GET_STR_DC_THRESHOLD 11 +#define GET_STR_PRECOMPUTE_THRESHOLD 22 +#define SET_STR_DC_THRESHOLD 381 +#define SET_STR_PRECOMPUTE_THRESHOLD 2503 + +#define FAC_DSC_THRESHOLD 216 +#define FAC_ODD_THRESHOLD 26 + +#define MATRIX22_STRASSEN_THRESHOLD 14 +#define HGCD2_DIV1_METHOD 5 /* 2.01% faster than 3 */ +#define HGCD_THRESHOLD 122 +#define HGCD_APPR_THRESHOLD 171 +#define HGCD_REDUCE_THRESHOLD 2479 +#define GCD_DC_THRESHOLD 541 +#define GCDEXT_DC_THRESHOLD 386 +#define JACOBI_BASE_METHOD 4 /* 7.46% faster than 1 */ -- cgit v1.2.3