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/arm/v7a/cora15/addmul_1.asm | 145 ++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/aors_n.asm | 162 +++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/bdiv_q_1.asm | 36 +++ gmp-6.3.0/mpn/arm/v7a/cora15/cnd_aors_n.asm | 158 +++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/com.asm | 180 +++++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/gmp-mparam.h | 212 +++++++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/logops_n.asm | 253 +++++++++++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/mul_1.asm | 104 +++++++++ .../mpn/arm/v7a/cora15/neon/aorsorrlsh1_n.asm | 43 ++++ .../mpn/arm/v7a/cora15/neon/aorsorrlsh2_n.asm | 43 ++++ .../mpn/arm/v7a/cora15/neon/aorsorrlshC_n.asm | 144 ++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/neon/com.asm | 97 ++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyd.asm | 110 +++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyi.asm | 90 ++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/neon/rsh1aors_n.asm | 177 ++++++++++++++ gmp-6.3.0/mpn/arm/v7a/cora15/submul_1.asm | 159 +++++++++++++ 16 files changed, 2113 insertions(+) create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/addmul_1.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/bdiv_q_1.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/cnd_aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/com.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/gmp-mparam.h create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/logops_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/mul_1.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh1_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh2_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlshC_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/com.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyd.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyi.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/neon/rsh1aors_n.asm create mode 100644 gmp-6.3.0/mpn/arm/v7a/cora15/submul_1.asm (limited to 'gmp-6.3.0/mpn/arm/v7a/cora15') diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/addmul_1.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/addmul_1.asm new file mode 100644 index 0000000..c2277b3 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/addmul_1.asm @@ -0,0 +1,145 @@ +dnl ARM mpn_addmul_1 optimised for A15. + +dnl Copyright 2012, 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 best +C StrongARM: - +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 6 3.25 +C Cortex-A15 2 this + +C This code uses umlal for adding in the rp[] data, keeping the recurrency path +C separate from any multiply instructions. It performs well on A15, at umlal's +C bandwidth. +C +C An A9 variant should perhaps stick to 3-way unrolling, and use ldm and stm +C for all loads and stores. Alternatively, it could do 2-way or 4-way, but +C then alignment aware code will be necessary (adding O(1) bookkeeping +C overhead). +C +C We don't use r12 due to ldrd and strd limitations. + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 - +C v6t2 - +C v7a - + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') +define(`v0', `r3') + +define(`w0', `r10') define(`w1', `r11') +define(`u0', `r8') define(`u1', `r9') + +ASM_START() +PROLOGUE(mpn_addmul_1) + push { r4-r11 } + + ands r6, n, #3 + sub n, n, #3 + beq L(b00) + cmp r6, #2 + bcc L(b01) + beq L(b10) + +L(b11): mov r6, #0 + cmn r13, #0 C carry clear + ldr u1, [up], #-4 + ldr w1, [rp], #-4 + mov r7, #0 + b L(mid) + +L(b00): ldrd u0, u1, [up] + ldrd w0, w1, [rp] + mov r6, #0 + umlal w0, r6, u0, v0 + cmn r13, #0 C carry clear + mov r7, #0 + str w0, [rp] + b L(mid) + +L(b10): ldrd u0, u1, [up], #8 + ldrd w0, w1, [rp] + mov r4, #0 + umlal w0, r4, u0, v0 + cmn r13, #0 C carry clear + mov r5, #0 + str w0, [rp], #8 + umlal w1, r5, u1, v0 + tst n, n + bmi L(end) + b L(top) + +L(b01): mov r4, #0 + ldr u1, [up], #4 + ldr w1, [rp], #4 + mov r5, #0 + umlal w1, r5, u1, v0 + tst n, n + bmi L(end) + + ALIGN(16) +L(top): ldrd u0, u1, [up, #0] + adcs r4, r4, w1 + ldrd w0, w1, [rp, #0] + mov r6, #0 + umlal w0, r6, u0, v0 C 1 2 + adcs r5, r5, w0 + mov r7, #0 + strd r4, r5, [rp, #-4] +L(mid): umlal w1, r7, u1, v0 C 2 3 + ldrd u0, u1, [up, #8] + adcs r6, r6, w1 + ldrd w0, w1, [rp, #8] + mov r4, #0 + umlal w0, r4, u0, v0 C 3 4 + adcs r7, r7, w0 + mov r5, #0 + strd r6, r7, [rp, #4] + umlal w1, r5, u1, v0 C 0 1 + sub n, n, #4 + add up, up, #16 + add rp, rp, #16 + tst n, n + bpl L(top) + +L(end): adcs r4, r4, w1 + str r4, [rp, #-4] + adc r0, r5, #0 + pop { r4-r11 } + bx r14 +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/aors_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/aors_n.asm new file mode 100644 index 0000000..dc3f839 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/aors_n.asm @@ -0,0 +1,162 @@ +dnl ARM mpn_add_n/mpn_sub_n optimised for A15. + +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 best +C StrongARM: - +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 3.55 2.5 +C Cortex-A15 1.27 this + +C This was a major improvement compared to the code we had before, but it might +C not be the best 8-way code possible. We've tried some permutations of auto- +C increments and separate pointer updates, but they all ran at the same speed +C on A15. + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 - +C v6t2 - +C v7a - + +define(`rp', `r0') +define(`up', `r1') +define(`vp', `r2') +define(`n', `r3') + +ifdef(`OPERATION_add_n', ` + define(`ADDSUBC', adcs) + define(`IFADD', `$1') + define(`SETCY', `cmp $1, #1') + define(`RETVAL', `adc r0, n, #0') + define(`RETVAL2', `adc r0, n, #1') + define(`func', mpn_add_n) + define(`func_nc', mpn_add_nc)') +ifdef(`OPERATION_sub_n', ` + define(`ADDSUBC', sbcs) + define(`IFADD', `') + define(`SETCY', `rsbs $1, $1, #0') + define(`RETVAL', `sbc r0, r0, r0 + and r0, r0, #1') + define(`RETVAL2', `RETVAL') + define(`func', 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) + ldr r12, [sp] + b L(ent) +EPILOGUE() +PROLOGUE(func) + mov r12, #0 +L(ent): push { r4-r9 } + + ands r6, n, #3 + mov n, n, lsr #2 + beq L(b00) + cmp r6, #2 + bcc L(b01) + beq L(b10) + +L(b11): ldr r5, [up], #4 + ldr r7, [vp], #4 + SETCY( r12) + ADDSUBC r9, r5, r7 + ldrd r4, r5, [up, #0] + ldrd r6, r7, [vp, #0] + str r9, [rp], #-4 + b L(lo) + +L(b00): ldrd r4, r5, [up], #-8 + ldrd r6, r7, [vp], #-8 + SETCY( r12) + sub rp, rp, #16 + b L(mid) + +L(b01): ldr r5, [up], #-4 + ldr r7, [vp], #-4 + SETCY( r12) + ADDSUBC r9, r5, r7 + str r9, [rp], #-12 + tst n, n + beq L(wd1) +L(gt1): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + b L(mid) + +L(b10): ldrd r4, r5, [up] + ldrd r6, r7, [vp] + SETCY( r12) + sub rp, rp, #8 + b L(lo) + + ALIGN(16) +L(top): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + strd r8, r9, [rp, #8] +L(mid): ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + ldrd r4, r5, [up, #16] + ldrd r6, r7, [vp, #16] + strd r8, r9, [rp, #16] + ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + sub n, n, #2 + tst n, n + bmi L(dne) + ldrd r4, r5, [up, #24] + ldrd r6, r7, [vp, #24] + strd r8, r9, [rp, #24] + ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + ldrd r4, r5, [up, #32]! + ldrd r6, r7, [vp, #32]! + strd r8, r9, [rp, #32]! +L(lo): ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + tst n, n + bne L(top) + +L(end): strd r8, r9, [rp, #8] +L(wd1): RETVAL + pop { r4-r9 } + bx r14 +L(dne): strd r8, r9, [rp, #24] + RETVAL2 + pop { r4-r9 } + bx r14 +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/bdiv_q_1.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/bdiv_q_1.asm new file mode 100644 index 0000000..245b371 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/bdiv_q_1.asm @@ -0,0 +1,36 @@ +dnl ARM 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, 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') + +MULFUNC_PROLOGUE(mpn_bdiv_q_1 mpn_pi1_bdiv_q_1) +include_mpn(`arm/v7a/cora8/bdiv_q_1.asm') diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/cnd_aors_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/cnd_aors_n.asm new file mode 100644 index 0000000..b9e5cd3 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/cnd_aors_n.asm @@ -0,0 +1,158 @@ +dnl ARM mpn_cnd_add_n/mpn_cnd_sub_n optimised for A15. + +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 best +C StrongARM: - +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 3.75 3 +C Cortex-A15 1.78 this + +C This code does not run as well as one could have hoped, since 1.5 c/l seems +C realistic for this insn mix. + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 - +C v6t2 - +C v7a - + +define(`cnd',`r0') +define(`rp', `r1') +define(`up', `r2') +define(`vp', `r3') +define(`n', `r12') + +ifdef(`OPERATION_cnd_add_n', ` + define(`ADDSUB', adds) + define(`ADDSUBC', adcs) + define(`IFADD', `$1') + define(`INITCY', `cmn r0, #0') + define(`RETVAL', `adc r0, n, #0') + define(`RETVAL2', `adc r0, n, #1') + define(`func', mpn_cnd_add_n) + define(`func_nc', mpn_add_nc)') +ifdef(`OPERATION_cnd_sub_n', ` + define(`ADDSUB', subs) + define(`ADDSUBC', sbcs) + define(`IFADD', `') + define(`INITCY', `cmp r0, #0') + define(`RETVAL', `sbc r0, r0, r0 + and r0, r0, #1') + define(`RETVAL2', `RETVAL') + define(`func', mpn_cnd_sub_n) + define(`func_nc', mpn_sub_nc)') + +MULFUNC_PROLOGUE(mpn_cnd_add_n mpn_cnd_sub_n) + +ASM_START() +PROLOGUE(func) + ldr n, [sp] + push { r4-r9 } + + cmp cnd, #1 + sbc cnd, cnd, cnd C conditionally set to 0xffffffff + + ands r6, n, #3 + mov n, n, lsr #2 + beq L(b00) + cmp r6, #2 + bcc L(b01) + beq L(b10) + +L(b11): ldr r5, [up], #4 + ldr r7, [vp], #4 + bic r7, r7, cnd + ADDSUB r9, r5, r7 + ldrd r4, r5, [up, #0] + ldrd r6, r7, [vp, #0] + bic r6, r6, cnd + bic r7, r7, cnd + str r9, [rp], #-4 + b L(lo) + +L(b00): ldrd r4, r5, [up], #-8 + ldrd r6, r7, [vp], #-8 + bic r6, r6, cnd + bic r7, r7, cnd + INITCY + sub rp, rp, #16 + b L(mid) + +L(b01): ldr r5, [up], #-4 + ldr r7, [vp], #-4 + bic r7, r7, cnd + ADDSUB r9, r5, r7 + str r9, [rp], #-12 + tst n, n + beq L(wd1) +L(gt1): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + bic r6, r6, cnd + bic r7, r7, cnd + b L(mid) + +L(b10): ldrd r4, r5, [up] + ldrd r6, r7, [vp] + bic r6, r6, cnd + bic r7, r7, cnd + INITCY + sub rp, rp, #8 + b L(lo) + + ALIGN(16) +L(top): ldrd r6, r7, [vp, #8] + ldrd r4, r5, [up, #8] + bic r6, r6, cnd + bic r7, r7, cnd + strd r8, r9, [rp, #8] +L(mid): ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + ldrd r6, r7, [vp, #16]! + ldrd r4, r5, [up, #16]! + bic r6, r6, cnd + bic r7, r7, cnd + sub n, n, #1 + strd r8, r9, [rp, #16]! +L(lo): ADDSUBC r8, r4, r6 + ADDSUBC r9, r5, r7 + tst n, n + bne L(top) + +L(end): strd r8, r9, [rp, #8] +L(wd1): RETVAL + pop { r4-r9 } + bx r14 +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/com.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/com.asm new file mode 100644 index 0000000..a258afe --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/com.asm @@ -0,0 +1,180 @@ +dnl ARM mpn_com optimised for A15. + +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 StrongARM ? +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 2.5 +C Cortex-A15 1.0 + +C This is great A15 core register code, but it is a bit large. +C We use FEEDIN_VARIANT 1 to save some space, but use 8-way unrolling. + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 - +C v6t2 - +C v7a - + +define(`FEEDIN_VARIANT', 1) C alternatives: 0 1 2 +define(`UNROLL', 4x2) C alternatives: 4 4x2 + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') + +ASM_START() +PROLOGUE(mpn_com) + push { r4-r5,r8-r9 } + +ifelse(FEEDIN_VARIANT,0,` + ands r12, n, #3 + mov n, n, lsr #2 + beq L(b00a) + tst r12, #1 + beq L(bx0) + ldr r5, [up], #4 + mvn r9, r5 + str r9, [rp], #4 + tst r12, #2 + beq L(b00) +L(bx0): ldrd r4, r5, [up, #0] + sub rp, rp, #8 + b L(lo) +L(b00): tst n, n + beq L(wd1) +L(b00a):ldrd r4, r5, [up], #-8 + sub rp, rp, #16 + b L(mid) +') +ifelse(FEEDIN_VARIANT,1,` + and r12, n, #3 + mov n, n, lsr #2 + tst r12, #1 + beq L(bx0) + ldr r5, [up], #4 + mvn r9, r5 + str r9, [rp], #4 +L(bx0): tst r12, #2 + beq L(b00) + ldrd r4, r5, [up, #0] + sub rp, rp, #8 + b L(lo) +L(b00): tst n, n + beq L(wd1) + ldrd r4, r5, [up], #-8 + sub rp, rp, #16 + b L(mid) +') +ifelse(FEEDIN_VARIANT,2,` + ands r12, n, #3 + mov n, n, lsr #2 + beq L(b00) + cmp r12, #2 + bcc L(b01) + beq L(b10) + +L(b11): ldr r5, [up], #4 + mvn r9, r5 + ldrd r4, r5, [up, #0] + str r9, [rp], #-4 + b L(lo) + +L(b00): ldrd r4, r5, [up], #-8 + sub rp, rp, #16 + b L(mid) + +L(b01): ldr r5, [up], #-4 + mvn r9, r5 + str r9, [rp], #-12 + tst n, n + beq L(wd1) +L(gt1): ldrd r4, r5, [up, #8] + b L(mid) + +L(b10): ldrd r4, r5, [up] + sub rp, rp, #8 + b L(lo) +') + ALIGN(16) +ifelse(UNROLL,4,` +L(top): ldrd r4, r5, [up, #8] + strd r8, r9, [rp, #8] +L(mid): mvn r8, r4 + mvn r9, r5 + ldrd r4, r5, [up, #16]! + strd r8, r9, [rp, #16]! + sub n, n, #1 +L(lo): mvn r8, r4 + mvn r9, r5 + tst n, n + bne L(top) +') +ifelse(UNROLL,4x2,` +L(top): ldrd r4, r5, [up, #8] + strd r8, r9, [rp, #8] +L(mid): mvn r8, r4 + mvn r9, r5 + ldrd r4, r5, [up, #16] + strd r8, r9, [rp, #16] + mvn r8, r4 + mvn r9, r5 + sub n, n, #2 + tst n, n + bmi L(dne) + ldrd r4, r5, [up, #24] + strd r8, r9, [rp, #24] + mvn r8, r4 + mvn r9, r5 + ldrd r4, r5, [up, #32]! + strd r8, r9, [rp, #32]! +L(lo): mvn r8, r4 + mvn r9, r5 + tst n, n + bne L(top) +') + +L(end): strd r8, r9, [rp, #8] +L(wd1): pop { r4-r5,r8-r9 } + bx r14 +ifelse(UNROLL,4x2,` +L(dne): strd r8, r9, [rp, #24] + pop { r4-r5,r8-r9 } + bx r14 +') +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/gmp-mparam.h b/gmp-6.3.0/mpn/arm/v7a/cora15/gmp-mparam.h new file mode 100644 index 0000000..409cbbb --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/gmp-mparam.h @@ -0,0 +1,212 @@ +/* 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 32 +#define GMP_LIMB_BYTES 4 + +/* 2000 MHz Cortex-A15 with Neon (in spite of file position) */ +/* FFT tuning limit = 50,736,668 */ +/* Generated by tuneup.c, 2019-10-22, gcc 5.4 */ + +#define MOD_1_NORM_THRESHOLD 0 /* always */ +#define MOD_1_UNNORM_THRESHOLD 0 /* always */ +#define MOD_1N_TO_MOD_1_1_THRESHOLD 3 +#define MOD_1U_TO_MOD_1_1_THRESHOLD 2 +#define MOD_1_1_TO_MOD_1_2_THRESHOLD 10 +#define MOD_1_2_TO_MOD_1_4_THRESHOLD MP_SIZE_T_MAX +#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 8 +#define USE_PREINV_DIVREM_1 1 /* native */ +#define DIV_QR_1N_PI1_METHOD 1 /* 49.14% faster than 2 */ +#define DIV_QR_1_NORM_THRESHOLD MP_SIZE_T_MAX /* never */ +#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 (native) */ +#define BMOD_1_TO_MOD_1_THRESHOLD 17 + +#define DIV_1_VS_MUL_1_PERCENT 267 + +#define MUL_TOOM22_THRESHOLD 28 +#define MUL_TOOM33_THRESHOLD 114 +#define MUL_TOOM44_THRESHOLD 178 +#define MUL_TOOM6H_THRESHOLD 238 +#define MUL_TOOM8H_THRESHOLD 597 + +#define MUL_TOOM32_TO_TOOM43_THRESHOLD 113 +#define MUL_TOOM32_TO_TOOM53_THRESHOLD 115 +#define MUL_TOOM42_TO_TOOM53_THRESHOLD 115 +#define MUL_TOOM42_TO_TOOM63_THRESHOLD 115 +#define MUL_TOOM43_TO_TOOM54_THRESHOLD 154 + +#define SQR_BASECASE_THRESHOLD 0 /* always (native) */ +#define SQR_TOOM2_THRESHOLD 38 +#define SQR_TOOM3_THRESHOLD 126 +#define SQR_TOOM4_THRESHOLD 336 +#define SQR_TOOM6_THRESHOLD 446 +#define SQR_TOOM8_THRESHOLD 650 + +#define MULMID_TOOM42_THRESHOLD 52 + +#define MULMOD_BNM1_THRESHOLD 23 +#define SQRMOD_BNM1_THRESHOLD 17 + +#define MUL_FFT_MODF_THRESHOLD 575 /* k = 5 */ +#define MUL_FFT_TABLE3 \ + { { 575, 5}, { 25, 6}, { 13, 5}, { 27, 6}, \ + { 27, 7}, { 15, 6}, { 31, 7}, { 19, 6}, \ + { 39, 7}, { 25, 6}, { 51, 7}, { 27, 8}, \ + { 15, 7}, { 33, 8}, { 19, 7}, { 41, 8}, \ + { 23, 7}, { 51, 8}, { 27, 9}, { 15, 8}, \ + { 31, 7}, { 63, 8}, { 39, 9}, { 23, 8}, \ + { 55,10}, { 15, 9}, { 31, 8}, { 67, 9}, \ + { 39, 8}, { 79, 9}, { 47, 8}, { 95, 9}, \ + { 55,10}, { 31, 9}, { 71, 8}, { 143, 9}, \ + { 87,10}, { 47, 9}, { 111,11}, { 31,10}, \ + { 63, 9}, { 143,10}, { 79, 9}, { 159,10}, \ + { 95,11}, { 63,10}, { 143, 9}, { 287,10}, \ + { 159,11}, { 95,10}, { 191,12}, { 63,11}, \ + { 127,10}, { 255, 9}, { 511,10}, { 271, 9}, \ + { 543,10}, { 287,11}, { 159,10}, { 335, 9}, \ + { 671,10}, { 367, 9}, { 735,11}, { 191,10}, \ + { 383, 9}, { 799,10}, { 415,11}, { 223,12}, \ + { 127,10}, { 543,11}, { 287,10}, { 575,11}, \ + { 319,10}, { 639,11}, { 351,10}, { 703,12}, \ + { 191,11}, { 383,10}, { 767,11}, { 415,10}, \ + { 831,11}, { 447,13}, { 127,12}, { 255,11}, \ + { 543,10}, { 1087,11}, { 575,12}, { 319,11}, \ + { 671,10}, { 1343,11}, { 735,12}, { 383,11}, \ + { 831,12}, { 447,11}, { 959,13}, { 255,12}, \ + { 511,11}, { 1087,12}, { 575,11}, { 1151,12}, \ + { 639,11}, { 1343,12}, { 703,13}, { 383,12}, \ + { 767,11}, { 1599,12}, { 895,14}, { 255,13}, \ + { 511,12}, { 1087,13}, { 639,12}, { 1407,13}, \ + { 767,12}, { 1599,13}, { 895,14}, { 511,13}, \ + { 1023,12}, { 2111,13}, { 1151,12}, { 2431,13}, \ + { 1279,12}, { 2559,13}, { 1407,14}, { 767,13}, \ + { 1535,12}, { 3135,13}, { 1663,15}, { 511,14}, \ + { 1023,13}, { 2303,14}, { 1279,13}, { 2559,12}, \ + { 5119,13}, { 2687,14}, { 1535,13}, { 3071,12}, \ + { 6143,13}, { 3199,12}, { 6399,14}, { 1791,15}, \ + { 1023,14}, { 2047,13}, { 4095,14}, { 2303,13}, \ + { 4607,12}, { 9215,13}, { 4863,12}, { 9727,14}, \ + { 2559,13}, { 5119,15}, { 1535,14}, { 3071,13}, \ + { 8192,14}, { 16384,15}, { 32768,16} } +#define MUL_FFT_TABLE3_SIZE 155 +#define MUL_FFT_THRESHOLD 5760 + +#define SQR_FFT_MODF_THRESHOLD 525 /* k = 5 */ +#define SQR_FFT_TABLE3 \ + { { 525, 5}, { 25, 6}, { 27, 7}, { 15, 6}, \ + { 31, 7}, { 17, 6}, { 35, 7}, { 19, 6}, \ + { 39, 7}, { 25, 6}, { 51, 7}, { 27, 8}, \ + { 15, 7}, { 33, 8}, { 19, 7}, { 41, 8}, \ + { 23, 7}, { 51, 8}, { 27, 7}, { 55, 9}, \ + { 15, 8}, { 31, 7}, { 63, 8}, { 39, 9}, \ + { 23, 8}, { 51,10}, { 15, 9}, { 31, 8}, \ + { 67, 9}, { 39, 8}, { 79, 9}, { 47, 8}, \ + { 99, 9}, { 55,10}, { 31, 9}, { 79,10}, \ + { 47, 9}, { 95,11}, { 31,10}, { 63, 9}, \ + { 135,10}, { 95,11}, { 63,10}, { 143, 9}, \ + { 287, 8}, { 575, 9}, { 303,10}, { 159,11}, \ + { 95,10}, { 191, 9}, { 383,12}, { 63,11}, \ + { 127,10}, { 255, 9}, { 511,10}, { 271, 9}, \ + { 543,10}, { 287,11}, { 159,10}, { 319, 9}, \ + { 639,10}, { 335, 9}, { 671,10}, { 351,11}, \ + { 191,10}, { 399, 9}, { 799,10}, { 415, 9}, \ + { 831,11}, { 223,10}, { 447,12}, { 127,10}, \ + { 543,11}, { 287,10}, { 575,11}, { 319,10}, \ + { 639,11}, { 351,10}, { 703,12}, { 191,11}, \ + { 383,10}, { 799,11}, { 415,10}, { 831,11}, \ + { 447,13}, { 127,11}, { 543,10}, { 1087,11}, \ + { 607,12}, { 319,11}, { 735,12}, { 383,11}, \ + { 831,12}, { 447,11}, { 959,12}, { 511,11}, \ + { 1023,12}, { 575,11}, { 1151,12}, { 639,11}, \ + { 1279,12}, { 703,13}, { 383,12}, { 767,11}, \ + { 1535,12}, { 831,11}, { 1663,12}, { 895,14}, \ + { 255,13}, { 511,12}, { 1087,13}, { 639,12}, \ + { 1343,13}, { 767,12}, { 1599,13}, { 895,14}, \ + { 511,13}, { 1023,12}, { 2111,13}, { 1151,12}, \ + { 2303,13}, { 1279,14}, { 767,13}, { 1535,12}, \ + { 3135,13}, { 1663,15}, { 511,14}, { 1023,13}, \ + { 2047,12}, { 4095,13}, { 2303,14}, { 1279,13}, \ + { 2559,12}, { 5119,14}, { 1535,13}, { 3071,12}, \ + { 6143,13}, { 3199,12}, { 6399,14}, { 1791,15}, \ + { 1023,14}, { 2047,13}, { 4095,14}, { 2303,13}, \ + { 4607,12}, { 9215,13}, { 4863,12}, { 9727,14}, \ + { 2559,15}, { 1535,14}, { 3071,13}, { 8192,14}, \ + { 16384,15}, { 32768,16} } +#define SQR_FFT_TABLE3_SIZE 154 +#define SQR_FFT_THRESHOLD 5312 + +#define MULLO_BASECASE_THRESHOLD 0 /* always */ +#define MULLO_DC_THRESHOLD 38 +#define MULLO_MUL_N_THRESHOLD 10950 +#define SQRLO_BASECASE_THRESHOLD 10 +#define SQRLO_DC_THRESHOLD 35 +#define SQRLO_SQR_THRESHOLD 10323 + +#define DC_DIV_QR_THRESHOLD 57 +#define DC_DIVAPPR_Q_THRESHOLD 254 +#define DC_BDIV_QR_THRESHOLD 48 +#define DC_BDIV_Q_THRESHOLD 286 + +#define INV_MULMOD_BNM1_THRESHOLD 55 +#define INV_NEWTON_THRESHOLD 252 +#define INV_APPR_THRESHOLD 252 + +#define BINV_NEWTON_THRESHOLD 372 +#define REDC_1_TO_REDC_2_THRESHOLD 61 +#define REDC_2_TO_REDC_N_THRESHOLD 0 /* always */ + +#define MU_DIV_QR_THRESHOLD 1858 +#define MU_DIVAPPR_Q_THRESHOLD 1787 +#define MUPI_DIV_QR_THRESHOLD 122 +#define MU_BDIV_QR_THRESHOLD 1528 +#define MU_BDIV_Q_THRESHOLD 1836 + +#define POWM_SEC_TABLE 1,14,200,480,1532 + +#define GET_STR_DC_THRESHOLD 16 +#define GET_STR_PRECOMPUTE_THRESHOLD 33 +#define SET_STR_DC_THRESHOLD 104 +#define SET_STR_PRECOMPUTE_THRESHOLD 1120 + +#define FAC_DSC_THRESHOLD 164 +#define FAC_ODD_THRESHOLD 27 + +#define MATRIX22_STRASSEN_THRESHOLD 19 +#define HGCD2_DIV1_METHOD 1 /* 3.70% faster than 3 */ +#define HGCD_THRESHOLD 137 +#define HGCD_APPR_THRESHOLD 157 +#define HGCD_REDUCE_THRESHOLD 3389 +#define GCD_DC_THRESHOLD 610 +#define GCDEXT_DC_THRESHOLD 443 +#define JACOBI_BASE_METHOD 4 /* 12.66% faster than 1 */ + +/* Tuneup completed successfully, took 69757 seconds */ diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/logops_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/logops_n.asm new file mode 100644 index 0000000..0602614 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/logops_n.asm @@ -0,0 +1,253 @@ +dnl ARM mpn_and_n, mpn_andn_n. mpn_nand_n, etc, optimised for A15. + +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 and andn ior xor nand iorn nior xnor +C StrongARM ? ? +C XScale ? ? +C Cortex-A7 ? ? +C Cortex-A8 ? ? +C Cortex-A9 3.5 3.56 +C Cortex-A15 1.27 1.64 + +C This is great A15 core register code, but it is a bit large. +C We use FEEDIN_VARIANT 1 to save some space, but use 8-way unrolling. + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 - +C v6t2 - +C v7a - + +define(`FEEDIN_VARIANT', 1) C alternatives: 0 1 2 +define(`UNROLL', 4x2) C alternatives: 4 4x2 + +define(`rp', `r0') +define(`up', `r1') +define(`vp', `r2') +define(`n', `r3') + +define(`POSTOP') + +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(`POSTOP', `mvn $1, $1') + define(`LOGOP', `bic $1, $3, $2')') +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(`POSTOP', `mvn $1, $1') + define(`LOGOP', `eor $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) + push { r4-r9 } + +ifelse(FEEDIN_VARIANT,0,` + ands r6, n, #3 + mov n, n, lsr #2 + beq L(b00a) + tst r6, #1 + beq L(bx0) + ldr r5, [up], #4 + ldr r7, [vp], #4 + LOGOP( r9, r5, r7) + POSTOP( r9) + str r9, [rp], #4 + tst r6, #2 + beq L(b00) +L(bx0): ldrd r4, r5, [up, #0] + ldrd r6, r7, [vp, #0] + sub rp, rp, #8 + b L(lo) +L(b00): tst n, n + beq L(wd1) +L(b00a):ldrd r4, r5, [up], #-8 + ldrd r6, r7, [vp], #-8 + sub rp, rp, #16 + b L(mid) +') +ifelse(FEEDIN_VARIANT,1,` + and r6, n, #3 + mov n, n, lsr #2 + tst r6, #1 + beq L(bx0) + ldr r5, [up], #4 + ldr r7, [vp], #4 + LOGOP( r9, r5, r7) + POSTOP( r9) + str r9, [rp], #4 +L(bx0): tst r6, #2 + beq L(b00) + ldrd r4, r5, [up, #0] + ldrd r6, r7, [vp, #0] + sub rp, rp, #8 + b L(lo) +L(b00): tst n, n + beq L(wd1) + ldrd r4, r5, [up], #-8 + ldrd r6, r7, [vp], #-8 + sub rp, rp, #16 + b L(mid) +') +ifelse(FEEDIN_VARIANT,2,` + ands r6, n, #3 + mov n, n, lsr #2 + beq L(b00) + cmp r6, #2 + bcc L(b01) + beq L(b10) + +L(b11): ldr r5, [up], #4 + ldr r7, [vp], #4 + LOGOP( r9, r5, r7) + ldrd r4, r5, [up, #0] + ldrd r6, r7, [vp, #0] + POSTOP( r9) + str r9, [rp], #-4 + b L(lo) + +L(b00): ldrd r4, r5, [up], #-8 + ldrd r6, r7, [vp], #-8 + sub rp, rp, #16 + b L(mid) + +L(b01): ldr r5, [up], #-4 + ldr r7, [vp], #-4 + LOGOP( r9, r5, r7) + POSTOP( r9) + str r9, [rp], #-12 + tst n, n + beq L(wd1) +L(gt1): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + b L(mid) + +L(b10): ldrd r4, r5, [up] + ldrd r6, r7, [vp] + sub rp, rp, #8 + b L(lo) +') + ALIGN(16) +ifelse(UNROLL,4,` +L(top): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #8] +L(mid): LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + ldrd r4, r5, [up, #16]! + ldrd r6, r7, [vp, #16]! + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #16]! + sub n, n, #1 +L(lo): LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + tst n, n + bne L(top) +') +ifelse(UNROLL,4x2,` +L(top): ldrd r4, r5, [up, #8] + ldrd r6, r7, [vp, #8] + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #8] +L(mid): LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + ldrd r4, r5, [up, #16] + ldrd r6, r7, [vp, #16] + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #16] + LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + sub n, n, #2 + tst n, n + bmi L(dne) + ldrd r4, r5, [up, #24] + ldrd r6, r7, [vp, #24] + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #24] + LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + ldrd r4, r5, [up, #32]! + ldrd r6, r7, [vp, #32]! + POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #32]! +L(lo): LOGOP( r8, r4, r6) + LOGOP( r9, r5, r7) + tst n, n + bne L(top) +') + +L(end): POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #8] +L(wd1): pop { r4-r9 } + bx r14 +ifelse(UNROLL,4x2,` +L(dne): POSTOP( r8) + POSTOP( r9) + strd r8, r9, [rp, #24] + pop { r4-r9 } + bx r14 +') +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/mul_1.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/mul_1.asm new file mode 100644 index 0000000..766ba5c --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/mul_1.asm @@ -0,0 +1,104 @@ +dnl ARM mpn_mul_1 optimised for A15. + +dnl Copyright 2012, 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 best +C StrongARM: - +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 5.25 3.25 +C Cortex-A15 2.25 this + + +C This runs well on A15 but very poorly on A9. By scheduling loads and adds +C it is possible to get good A9 performance as well, but at the cost of using +C many more (callee-saves) registers. + +C This is armv5 code, optimized for the armv7a cpu A15. Its location in the +C GMP file structure might be misleading. + + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') +define(`v0', `r3') + +ASM_START() +PROLOGUE(mpn_mul_1c) + ldr r12, [sp] + b L(ent) +EPILOGUE() +PROLOGUE(mpn_mul_1) + mov r12, #0 +L(ent): push {r4-r7} + + ldr r6, [up], #4 + tst n, #1 + beq L(bx0) + +L(bx1): umull r4, r7, r6, v0 + adds r4, r4, r12 + tst n, #2 + beq L(lo1) + b L(lo3) + +L(bx0): umull r4, r5, r6, v0 + adds r4, r4, r12 + tst n, #2 + beq L(lo0) + b L(lo2) + +L(top): ldr r6, [up], #4 + str r4, [rp], #4 + umull r4, r5, r6, v0 + adds r4, r4, r7 +L(lo0): ldr r6, [up], #4 + str r4, [rp], #4 + umull r4, r7, r6, v0 + adcs r4, r4, r5 +L(lo3): ldr r6, [up], #4 + str r4, [rp], #4 + umull r4, r5, r6, v0 + adcs r4, r4, r7 +L(lo2): ldr r6, [up], #4 + str r4, [rp], #4 + umull r4, r7, r6, v0 + adcs r4, r4, r5 +L(lo1): adc r7, r7, #0 + subs n, n, #4 + bgt L(top) + + str r4, [rp] + mov r0, r7 + pop {r4-r7} + bx lr +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh1_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh1_n.asm new file mode 100644 index 0000000..d8cfe3f --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh1_n.asm @@ -0,0 +1,43 @@ +dnl ARM mpn_addlshC_n, mpn_sublshC_n, mpn_rsblshC_n + +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') + +define(LSH, 1) + +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(`arm/v7a/cora15/neon/aorsorrlshC_n.asm') diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh2_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh2_n.asm new file mode 100644 index 0000000..b48204d --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlsh2_n.asm @@ -0,0 +1,43 @@ +dnl ARM mpn_addlshC_n, mpn_sublshC_n, mpn_rsblshC_n + +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') + +define(LSH, 2) + +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(`arm/v7a/cora15/neon/aorsorrlshC_n.asm') diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlshC_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlshC_n.asm new file mode 100644 index 0000000..51f93c1 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/aorsorrlshC_n.asm @@ -0,0 +1,144 @@ +dnl ARM mpn_addlshC_n, mpn_sublshC_n, mpn_rsblshC_n + +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/. + + +C cycles/limb +C StrongARM - +C XScale - +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 5.25 +C Cortex-A15 2.25 + +C TODO +C * Consider using 4-way feed-in code. +C * This is ad-hoc scheduled, perhaps unnecessarily so for A15, and perhaps +C insufficiently for A7 and A8. + +define(`rp', `r0') +define(`up', `r1') +define(`vp', `r2') +define(`n', `r3') + +ifdef(`DO_add', ` + define(`ADCSBCS', `adcs $1, $2, $3') + define(`CLRCY', `cmn r13, #1') + define(`RETVAL', `adc r0, $1, #0') + define(`func', mpn_addlsh`'LSH`'_n)') +ifdef(`DO_sub', ` + define(`ADCSBCS', `sbcs $1, $2, $3') + define(`CLRCY', `cmp r13, #0') + define(`RETVAL', `sbc $2, $2, $2 + cmn $2, #1 + adc r0, $1, #0') + define(`func', mpn_sublsh`'LSH`'_n)') +ifdef(`DO_rsb', ` + define(`ADCSBCS', `sbcs $1, $3, $2') + define(`CLRCY', `cmp r13, #0') + define(`RETVAL', `sbc r0, $1, #0') + define(`func', mpn_rsblsh`'LSH`'_n)') + + +ASM_START() +PROLOGUE(func) + push {r4-r10} + vmov.i8 d0, #0 C could feed carry through here + CLRCY + tst n, #1 + beq L(bb0) + +L(bb1): vld1.32 {d3[0]}, [vp]! + vsli.u32 d0, d3, #LSH + ldr r12, [up], #4 + vmov.32 r5, d0[0] + vshr.u32 d0, d3, #32-LSH + ADCSBCS( r12, r12, r5) + str r12, [rp], #4 + bics n, n, #1 + beq L(rtn) + +L(bb0): tst n, #2 + beq L(b00) + +L(b10): vld1.32 {d3}, [vp]! + vsli.u64 d0, d3, #LSH + ldmia up!, {r10,r12} + vmov r4, r5, d0 + vshr.u64 d0, d3, #64-LSH + ADCSBCS( r10, r10, r4) + ADCSBCS( r12, r12, r5) + stmia rp!, {r10,r12} + bics n, n, #2 + beq L(rtn) + +L(b00): vld1.32 {d2}, [vp]! + vsli.u64 d0, d2, #LSH + vshr.u64 d1, d2, #64-LSH + vld1.32 {d3}, [vp]! + vsli.u64 d1, d3, #LSH + vmov r6, r7, d0 + vshr.u64 d0, d3, #64-LSH + sub n, n, #4 + tst n, n + beq L(end) + + ALIGN(16) +L(top): ldmia up!, {r8,r9,r10,r12} + vld1.32 {d2}, [vp]! + vsli.u64 d0, d2, #LSH + vmov r4, r5, d1 + vshr.u64 d1, d2, #64-LSH + ADCSBCS( r8, r8, r6) + ADCSBCS( r9, r9, r7) + vld1.32 {d3}, [vp]! + vsli.u64 d1, d3, #LSH + vmov r6, r7, d0 + vshr.u64 d0, d3, #64-LSH + ADCSBCS( r10, r10, r4) + ADCSBCS( r12, r12, r5) + stmia rp!, {r8,r9,r10,r12} + sub n, n, #4 + tst n, n + bne L(top) + +L(end): ldmia up!, {r8,r9,r10,r12} + vmov r4, r5, d1 + ADCSBCS( r8, r8, r6) + ADCSBCS( r9, r9, r7) + ADCSBCS( r10, r10, r4) + ADCSBCS( r12, r12, r5) + stmia rp!, {r8,r9,r10,r12} +L(rtn): vmov.32 r0, d0[0] + RETVAL( r0, r1) + pop {r4-r10} + bx r14 +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/com.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/com.asm new file mode 100644 index 0000000..9e7a629 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/com.asm @@ -0,0 +1,97 @@ +dnl ARM Neon mpn_com optimised for A15. + +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 StrongARM ? +C XScale ? +C Cortex-A8 ? +C Cortex-A9 2.1 +C Cortex-A15 0.65 + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') + +ASM_START() +PROLOGUE(mpn_com) + cmp n, #7 + ble L(bc) + +C Perform a few initial operation until rp is 128-bit aligned + tst rp, #4 + beq L(al1) + vld1.32 {d0[0]}, [up]! + sub n, n, #1 + vmvn d0, d0 + vst1.32 {d0[0]}, [rp]! +L(al1): tst rp, #8 + beq L(al2) + vld1.32 {d0}, [up]! + sub n, n, #2 + vmvn d0, d0 + vst1.32 {d0}, [rp:64]! +L(al2): vld1.32 {q2}, [up]! + subs n, n, #12 + blt L(end) + + ALIGN(16) +L(top): vld1.32 {q0}, [up]! + vmvn q2, q2 + subs n, n, #8 + vst1.32 {q2}, [rp:128]! + vld1.32 {q2}, [up]! + vmvn q0, q0 + vst1.32 {q0}, [rp:128]! + bge L(top) + +L(end): vmvn q2, q2 + vst1.32 {q2}, [rp:128]! + +C Handle last 0-7 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tst n, #4 + beq L(tl1) + vld1.32 {q0}, [up]! + vmvn q0, q0 + vst1.32 {q0}, [rp]! +L(tl1): tst n, #2 + beq L(tl2) + vld1.32 {d0}, [up]! + vmvn d0, d0 + vst1.32 {d0}, [rp]! +L(tl2): tst n, #1 + beq L(tl3) + vld1.32 {d0[0]}, [up] + vmvn d0, d0 + vst1.32 {d0[0]}, [rp] +L(tl3): bx lr +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyd.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyd.asm new file mode 100644 index 0000000..98fe535 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyd.asm @@ -0,0 +1,110 @@ +dnl ARM Neon mpn_copyd optimised for A15. + +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 StrongARM - +C XScale - +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 1.75 slower than core register code +C Cortex-A15 0.52 + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') + +ASM_START() +PROLOGUE(mpn_copyd) + add rp, rp, n, lsl #2 + add up, up, n, lsl #2 + + cmp n, #7 + ble L(bc) + +C Copy until rp is 128-bit aligned + tst rp, #4 + beq L(al1) + sub up, up, #4 + vld1.32 {d22[0]}, [up] + sub n, n, #1 + sub rp, rp, #4 + vst1.32 {d22[0]}, [rp] +L(al1): tst rp, #8 + beq L(al2) + sub up, up, #8 + vld1.32 {d22}, [up] + sub n, n, #2 + sub rp, rp, #8 + vst1.32 {d22}, [rp:64] +L(al2): sub up, up, #16 + vld1.32 {d26-d27}, [up] + subs n, n, #12 + sub rp, rp, #16 C offset rp for loop + blt L(end) + + sub up, up, #16 C offset up for loop + mov r12, #-16 + + ALIGN(16) +L(top): vld1.32 {d22-d23}, [up], r12 + vst1.32 {d26-d27}, [rp:128], r12 + vld1.32 {d26-d27}, [up], r12 + vst1.32 {d22-d23}, [rp:128], r12 + subs n, n, #8 + bge L(top) + + add up, up, #16 C undo up offset + C rp offset undoing folded +L(end): vst1.32 {d26-d27}, [rp:128] + +C Copy last 0-7 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tst n, #4 + beq L(tl1) + sub up, up, #16 + vld1.32 {d22-d23}, [up] + sub rp, rp, #16 + vst1.32 {d22-d23}, [rp] +L(tl1): tst n, #2 + beq L(tl2) + sub up, up, #8 + vld1.32 {d22}, [up] + sub rp, rp, #8 + vst1.32 {d22}, [rp] +L(tl2): tst n, #1 + beq L(tl3) + sub up, up, #4 + vld1.32 {d22[0]}, [up] + sub rp, rp, #4 + vst1.32 {d22[0]}, [rp] +L(tl3): bx lr +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyi.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyi.asm new file mode 100644 index 0000000..2e05afe --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/copyi.asm @@ -0,0 +1,90 @@ +dnl ARM Neon mpn_copyi optimised for A15. + +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 StrongARM - +C XScale - +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 1.75 slower than core register code +C Cortex-A15 0.52 + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') + +ASM_START() +PROLOGUE(mpn_copyi) + cmp n, #7 + ble L(bc) + +C Copy until rp is 128-bit aligned + tst rp, #4 + beq L(al1) + vld1.32 {d22[0]}, [up]! + sub n, n, #1 + vst1.32 {d22[0]}, [rp]! +L(al1): tst rp, #8 + beq L(al2) + vld1.32 {d22}, [up]! + sub n, n, #2 + vst1.32 {d22}, [rp:64]! +L(al2): vld1.32 {d26-d27}, [up]! + subs n, n, #12 + blt L(end) + + ALIGN(16) +L(top): vld1.32 {d22-d23}, [up]! + vst1.32 {d26-d27}, [rp:128]! + vld1.32 {d26-d27}, [up]! + vst1.32 {d22-d23}, [rp:128]! + subs n, n, #8 + bge L(top) + +L(end): vst1.32 {d26-d27}, [rp:128]! + +C Copy last 0-7 limbs. Note that rp is aligned after loop, but not when we +C arrive here via L(bc) +L(bc): tst n, #4 + beq L(tl1) + vld1.32 {d22-d23}, [up]! + vst1.32 {d22-d23}, [rp]! +L(tl1): tst n, #2 + beq L(tl2) + vld1.32 {d22}, [up]! + vst1.32 {d22}, [rp]! +L(tl2): tst n, #1 + beq L(tl3) + vld1.32 {d22[0]}, [up] + vst1.32 {d22[0]}, [rp] +L(tl3): bx lr +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/neon/rsh1aors_n.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/rsh1aors_n.asm new file mode 100644 index 0000000..2c11d6d --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/neon/rsh1aors_n.asm @@ -0,0 +1,177 @@ +dnl ARM Neon mpn_rsh1add_n, mpn_rsh1sub_n. + +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 StrongARM - +C XScale - +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 4-5 +C Cortex-A15 2.5 + +C TODO +C * Try to make this smaller, its size (384 bytes) is excessive. +C * Try to reach 2.25 c/l on A15, to match the addlsh_1 family. +C * This is ad-hoc scheduled, perhaps unnecessarily so for A15, and perhaps +C insufficiently for A7 and A8. + +define(`rp', `r0') +define(`up', `r1') +define(`vp', `r2') +define(`n', `r3') + +ifdef(`OPERATION_rsh1add_n', ` + define(`ADDSUBS', `adds $1, $2, $3') + define(`ADCSBCS', `adcs $1, $2, $3') + define(`IFADD', `$1') + define(`IFSUB', `') + define(`func', mpn_rsh1add_n)') +ifdef(`OPERATION_rsh1sub_n', ` + define(`ADDSUBS', `subs $1, $2, $3') + define(`ADCSBCS', `sbcs $1, $2, $3') + define(`IFADD', `') + define(`IFSUB', `$1') + define(`func', mpn_rsh1sub_n)') + +MULFUNC_PROLOGUE(mpn_rsh1add_n mpn_rsh1sub_n) + +ASM_START() +PROLOGUE(func) + push {r4-r10} + + ands r4, n, #3 + beq L(b00) + cmp r4, #2 + blo L(b01) + beq L(b10) + +L(b11): ldmia up!, {r9,r10,r12} + ldmia vp!, {r5,r6,r7} + ADDSUBS( r9, r9, r5) + vmov d4, r9, r9 + ADCSBCS( r10, r10, r6) + ADCSBCS( r12, r12, r7) + vshr.u64 d3, d4, #1 + vmov d1, r10, r12 + vsli.u64 d3, d1, #31 + vshr.u64 d2, d1, #1 + vst1.32 d3[0], [rp]! + bics n, n, #3 + beq L(wd2) +L(gt3): ldmia up!, {r8,r9,r10,r12} + ldmia vp!, {r4,r5,r6,r7} + b L(mi0) + +L(b10): ldmia up!, {r10,r12} + ldmia vp!, {r6,r7} + ADDSUBS( r10, r10, r6) + ADCSBCS( r12, r12, r7) + vmov d4, r10, r12 + bics n, n, #2 + vshr.u64 d2, d4, #1 + beq L(wd2) +L(gt2): ldmia up!, {r8,r9,r10,r12} + ldmia vp!, {r4,r5,r6,r7} + b L(mi0) + +L(b01): ldr r12, [up], #4 + ldr r7, [vp], #4 + ADDSUBS( r12, r12, r7) + vmov d4, r12, r12 + bics n, n, #1 + bne L(gt1) + mov r5, r12, lsr #1 +IFADD(` adc r1, n, #0') +IFSUB(` adc r1, n, #1') + bfi r5, r1, #31, #1 + str r5, [rp] + and r0, r12, #1 + pop {r4-r10} + bx r14 +L(gt1): ldmia up!, {r8,r9,r10,r12} + ldmia vp!, {r4,r5,r6,r7} + vshr.u64 d2, d4, #1 + ADCSBCS( r8, r8, r4) + ADCSBCS( r9, r9, r5) + vmov d0, r8, r9 + ADCSBCS( r10, r10, r6) + ADCSBCS( r12, r12, r7) + vsli.u64 d2, d0, #31 + vshr.u64 d3, d0, #1 + vst1.32 d2[0], [rp]! + b L(mi1) + +L(b00): ldmia up!, {r8,r9,r10,r12} + ldmia vp!, {r4,r5,r6,r7} + ADDSUBS( r8, r8, r4) + ADCSBCS( r9, r9, r5) + vmov d4, r8, r9 + ADCSBCS( r10, r10, r6) + ADCSBCS( r12, r12, r7) + vshr.u64 d3, d4, #1 + b L(mi1) + + ALIGN(16) +L(top): ldmia up!, {r8,r9,r10,r12} + ldmia vp!, {r4,r5,r6,r7} + vsli.u64 d3, d1, #63 + vshr.u64 d2, d1, #1 + vst1.32 d3, [rp]! +L(mi0): ADCSBCS( r8, r8, r4) + ADCSBCS( r9, r9, r5) + vmov d0, r8, r9 + ADCSBCS( r10, r10, r6) + ADCSBCS( r12, r12, r7) + vsli.u64 d2, d0, #63 + vshr.u64 d3, d0, #1 + vst1.32 d2, [rp]! +L(mi1): vmov d1, r10, r12 + sub n, n, #4 + tst n, n + bne L(top) + +L(end): vsli.u64 d3, d1, #63 + vshr.u64 d2, d1, #1 + vst1.32 d3, [rp]! +L(wd2): vmov r4, r5, d2 +IFADD(` adc r1, n, #0') +IFSUB(` adc r1, n, #1') + bfi r5, r1, #31, #1 + stm rp, {r4,r5} + +L(rtn): vmov.32 r0, d4[0] + and r0, r0, #1 + pop {r4-r10} + bx r14 +EPILOGUE() diff --git a/gmp-6.3.0/mpn/arm/v7a/cora15/submul_1.asm b/gmp-6.3.0/mpn/arm/v7a/cora15/submul_1.asm new file mode 100644 index 0000000..ed7bfe8 --- /dev/null +++ b/gmp-6.3.0/mpn/arm/v7a/cora15/submul_1.asm @@ -0,0 +1,159 @@ +dnl ARM mpn_submul_1 optimised for A15. + +dnl Copyright 2012, 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 best +C StrongARM: - +C XScale ? +C Cortex-A7 ? +C Cortex-A8 ? +C Cortex-A9 5.75 3.75 +C Cortex-A15 2.32 this + +C This code uses umlal and umaal for adding in the rp[] data, keeping the +C recurrency path separate from any multiply instructions. It performs well on +C A15, but not quite at the multiply bandwidth like the corresponding addmul_1 +C code. +C +C We don't use r12 due to ldrd and strd limitations. +C +C This loop complements U on the fly, +C U' = B^n - 1 - U +C and then uses that +C R - U*v = R + U'*v + v - B^n v + +C Architecture requirements: +C v5 - +C v5t - +C v5te ldrd strd +C v6 umaal +C v6t2 - +C v7a - + +define(`rp', `r0') +define(`up', `r1') +define(`n', `r2') +define(`v0', `r3') + +define(`w0', `r10') define(`w1', `r11') +define(`u0', `r8') define(`u1', `r9') + +ASM_START() +PROLOGUE(mpn_submul_1) + sub sp, sp, #32 + strd r10, r11, [sp, #24] + strd r8, r9, [sp, #16] + strd r6, r7, [sp, #8] + strd r4, r5, [sp, #0] +C push { r4-r11 } + + ands r6, n, #3 + sub n, n, #3 + beq L(b00) + cmp r6, #2 + bcc L(b01) + beq L(b10) + +L(b11): mov r6, #0 + ldr u1, [up], #-4 + ldr w1, [rp], #-16 + mvn u1, u1 + adds r7, v0, #0 + b L(mid) + +L(b00): ldrd u0, u1, [up] + ldrd w0, w1, [rp], #-12 + mvn u0, u0 + mvn u1, u1 + mov r6, v0 + umaal w0, r6, u0, v0 + cmn r13, #0 C carry clear + mov r7, #0 + str w0, [rp, #12] + b L(mid) + +L(b10): ldrd u0, u1, [up], #8 + ldrd w0, w1, [rp] + mvn u0, u0 + mvn u1, u1 + mov r4, v0 + umaal w0, r4, u0, v0 + mov r5, #0 + str w0, [rp], #-4 + umlal w1, r5, u1, v0 + adds n, n, #0 + bmi L(end) + b L(top) + +L(b01): ldr u1, [up], #4 + ldr w1, [rp], #-8 + mvn u1, u1 + mov r5, v0 + mov r4, #0 + umaal w1, r5, u1, v0 + tst n, n + bmi L(end) + +C ALIGN(16) +L(top): ldrd u0, u1, [up, #0] + adcs r4, r4, w1 + mvn u0, u0 + ldrd w0, w1, [rp, #12] + mvn u1, u1 + mov r6, #0 + umlal w0, r6, u0, v0 C 1 2 + adcs r5, r5, w0 + mov r7, #0 + strd r4, r5, [rp, #8] +L(mid): umaal w1, r7, u1, v0 C 2 3 + ldrd u0, u1, [up, #8] + add up, up, #16 + adcs r6, r6, w1 + mvn u0, u0 + ldrd w0, w1, [rp, #20] + mvn u1, u1 + mov r4, #0 + umlal w0, r4, u0, v0 C 3 4 + adcs r7, r7, w0 + mov r5, #0 + strd r6, r7, [rp, #16]! + sub n, n, #4 + umlal w1, r5, u1, v0 C 0 1 + tst n, n + bpl L(top) + +L(end): adcs r4, r4, w1 + str r4, [rp, #8] + adc r0, r5, #0 + sub r0, v0, r0 + pop { r4-r11 } + bx r14 +EPILOGUE() -- cgit v1.2.3