URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [libgcc/] [config/] [sh/] [lib1funcs-4-300.S] - Rev 734
Compare with Previous | Blame | View Log
/* Copyright (C) 2004, 2006, 2009 Free Software Foundation, Inc.This file is free software; you can redistribute it and/or modify itunder the terms of the GNU General Public License as published by theFree Software Foundation; either version 3, or (at your option) anylater version.This file is distributed in the hope that it will be useful, butWITHOUT ANY WARRANTY; without even the implied warranty ofMERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNUGeneral Public License for more details.Under Section 7 of GPL version 3, you are granted additionalpermissions described in the GCC Runtime Library Exception, version3.1, as published by the Free Software Foundation.You should have received a copy of the GNU General Public License anda copy of the GCC Runtime Library Exception along with this program;see the files COPYING3 and COPYING.RUNTIME respectively. If not, see<http://www.gnu.org/licenses/>. *//* libgcc routines for the STMicroelectronics ST40-300 CPU.Contributed by J"orn Rennecke joern.rennecke@st.com. */#include "lib1funcs.h"#if !__SHMEDIA__#ifdef L_div_table#if defined (__SH3__) || defined (__SH3E__) || defined (__SH4__) || defined (__SH4_SINGLE__) || defined (__SH4_SINGLE_ONLY__) || defined (__SH4_NOFPU__)/* This code used shld, thus is not suitable for SH1 / SH2. *//* Signed / unsigned division without use of FPU, optimized for SH4-300.Uses a lookup table for divisors in the range -128 .. +127, anddiv1 with case distinction for larger divisors in three more ranges.The code is lumped together with the table to allow the use of mova. */#ifdef __LITTLE_ENDIAN__#define L_LSB 0#define L_LSWMSB 1#define L_MSWLSB 2#else#define L_LSB 3#define L_LSWMSB 2#define L_MSWLSB 1#endif.global GLOBAL(udivsi3_i4i).global GLOBAL(sdivsi3_i4i)FUNC(GLOBAL(udivsi3_i4i))FUNC(GLOBAL(sdivsi3_i4i)).balign 4LOCAL(div_ge8m): ! 10 cycles up to hererotcr r1 ! signed shift must use original sign from r4div0s r5,r4mov #24,r7shld r7,r6shad r0,r1rotcl r6div1 r5,r1swap.w r5,r0 ! detect -0x80000000 : 0x800000rotcl r6swap.w r4,r7div1 r5,r1swap.b r7,r7rotcl r6or r7,r0div1 r5,r1swap.w r0,r7rotcl r6or r7,r0div1 r5,r1add #-0x80,r0rotcl r6extu.w r0,r0div1 r5,r1neg r0,r0rotcl r6swap.w r0,r0div1 r5,r1mov.l @r15+,r7and r6,r0rotcl r6div1 r5,r1shll2 r0rotcl r6exts.b r0,r0div1 r5,r1swap.w r0,r0exts.w r0,r1exts.b r6,r0mov.l @r15+,r6rotcl r0rtssub r1,r0! 31 cycles up to here.balign 4LOCAL(udiv_ge64k): ! 3 cycles up to heremov r4,r0shlr8 r0div0ucmp/hi r0,r5bt LOCAL(udiv_r8)mov.l r5,@-r15shll8 r5! 7 cycles up to here.rept 8div1 r5,r0.endrextu.b r4,r1 ! 15 cycles up to hereextu.b r0,r6xor r1,r0xor r6,r0swap.b r6,r6.rept 8div1 r5,r0.endr ! 25 cycles up to hereextu.b r0,r0mov.l @r15+,r5or r6,r0mov.l @r15+,r6rtsrotcl r0 ! 28 cycles up to here.balign 4LOCAL(udiv_r8): ! 6 cycles up to heremov.l r4,@-r15shll16 r4shll8 r4!shll r4mov r0,r1div1 r5,r1mov r4,r0rotcl r0mov.l @r15+,r4div1 r5,r1! 12 cycles up to here.rept 6rotcl r0; div1 r5,r1.endrmov.l @r15+,r6 ! 24 cycles up to herertsrotcl r0.balign 4LOCAL(div_ge32k): ! 6 cycles up to heremov.l r7,@-r15swap.w r5,r6exts.b r6,r7exts.w r6,r6cmp/eq r6,r7extu.b r1,r6bf/s LOCAL(div_ge8m)cmp/hi r1,r4 ! copy sign bit of r4 into Trotcr r1 ! signed shift must use original sign from r4div0s r5,r4shad r0,r1shll8 r5div1 r5,r1mov r5,r7 ! detect r4 == 0x80000000 && r5 == 0x8000(00)div1 r5,r1shlr8 r7div1 r5,r1swap.w r4,r0div1 r5,r1swap.b r0,r0div1 r5,r1or r0,r7div1 r5,r1add #-80,r7div1 r5,r1swap.w r7,r0div1 r5,r1or r0,r7extu.b r1,r0xor r6,r1xor r0,r1exts.b r0,r0div1 r5,r1extu.w r7,r7div1 r5,r1neg r7,r7 ! upper 16 bit of r7 == 0 if r4 == 0x80000000 && r5 == 0x8000div1 r5,r1and r0,r7div1 r5,r1swap.w r7,r7 ! 26 cycles up to here.div1 r5,r1shll8 r0div1 r5,r1exts.w r7,r7div1 r5,r1add r0,r0div1 r5,r1sub r7,r0extu.b r1,r1mov.l @r15+,r7rotcl r1mov.l @r15+,r6add r1,r0mov #-8,r1rtsshad r1,r5 ! 34 cycles up to here.balign 4GLOBAL(udivsi3_i4i):mov.l r6,@-r15extu.w r5,r6cmp/eq r5,r6mov #0x7f,r0bf LOCAL(udiv_ge64k)cmp/hi r0,r5bf LOCAL(udiv_le128)mov r4,r1shlr8 r1div0ushlr r1shll16 r6div1 r6,r1extu.b r4,r0 ! 7 cycles up to here.rept 8div1 r6,r1.endr ! 15 cycles up to herexor r1,r0 ! xor dividend with result lsb.rept 6div1 r6,r1.endrmov.l r7,@-r15 ! 21 cycles up to herediv1 r6,r1extu.b r0,r7div1 r6,r1shll8 r7extu.w r1,r0xor r7,r1 ! replace lsb of result with lsb of dividenddiv1 r6,r1mov #0,r7div1 r6,r1!div1 r6,r1bra LOCAL(div_end)div1 r6,r1 ! 28 cycles up to here/* This is link-compatible with a GLOBAL(sdivsi3) call,but we effectively clobber only r1, macl and mach *//* Because negative quotients are calculated as one's complements,-0x80000000 divided by the smallest positive number of a numberrange (0x80, 0x8000, 0x800000) causes saturation in the one'scomplement representation, and we have to suppress theone's -> two's complement adjustment. Since positive numbersdon't get such an adjustment, it's OK to also compute one's -> two'scomplement adjustment suppression for a dividend of 0. */.balign 4GLOBAL(sdivsi3_i4i):mov.l r6,@-r15exts.b r5,r6cmp/eq r5,r6mov #-1,r1bt/s LOCAL(div_le128)cmp/pz r4addc r4,r1exts.w r5,r6cmp/eq r5,r6mov #-7,r0bf/s LOCAL(div_ge32k)cmp/hi r1,r4 ! copy sign bit of r4 into Trotcr r1shll16 r6 ! 7 cycles up to hereshad r0,r1div0s r5,r4div1 r6,r1mov.l r7,@-r15div1 r6,r1mov r4,r0 ! re-compute adjusted dividenddiv1 r6,r1mov #-31,r7div1 r6,r1shad r7,r0div1 r6,r1add r4,r0 ! adjusted dividenddiv1 r6,r1mov.l r8,@-r15div1 r6,r1swap.w r4,r8 ! detect special case r4 = 0x80000000, r5 = 0x80div1 r6,r1swap.b r8,r8xor r1,r0 ! xor dividend with result lsbdiv1 r6,r1div1 r6,r1or r5,r8div1 r6,r1add #-0x80,r8 ! r8 is 0 iff there is a matchdiv1 r6,r1swap.w r8,r7 ! or upper 16 bits...div1 r6,r1or r7,r8 !...into lower 16 bitsdiv1 r6,r1extu.w r8,r8div1 r6,r1extu.b r0,r7div1 r6,r1shll8 r7exts.w r1,r0xor r7,r1 ! replace lsb of result with lsb of dividenddiv1 r6,r1neg r8,r8 ! upper 16 bits of r8 are now 0xffff iff we want end adjm.div1 r6,r1and r0,r8div1 r6,r1swap.w r8,r7div1 r6,r1mov.l @r15+,r8 ! 58 insns, 29 cycles up to hereLOCAL(div_end):div1 r6,r1shll8 r0div1 r6,r1exts.w r7,r7div1 r6,r1add r0,r0div1 r6,r1sub r7,r0extu.b r1,r1mov.l @r15+,r7rotcl r1mov.l @r15+,r6rtsadd r1,r0.balign 4LOCAL(udiv_le128): ! 4 cycles up to here (or 7 for mispredict)mova LOCAL(div_table_inv),r0shll2 r6mov.l @(r0,r6),r1mova LOCAL(div_table_clz),r0lds r4,mach!!!tst r1,r1!bt 0fdmulu.l r1,r40: mov.b @(r0,r5),r1clrt!!sts mach,r0addc r4,r0rotcr r0mov.l @r15+,r6rtsshld r1,r0.balign 4LOCAL(div_le128): ! 3 cycles up to here (or 6 for mispredict)mova LOCAL(div_table_inv),r0shll2 r6mov.l @(r0,r6),r1mova LOCAL(div_table_clz),r0neg r4,r6bf 0fmov r4,r60: lds r6,machtst r1,r1bt 0fdmulu.l r1,r60: div0s r4,r5mov.b @(r0,r5),r1bt/s LOCAL(le128_neg)clrt!sts mach,r0addc r6,r0rotcr r0mov.l @r15+,r6rtsshld r1,r0/* Could trap divide by zero for the cost of one cycle more mispredict penalty:...dmulu.l r1,r60: div0s r4,r5bt/s LOCAL(le128_neg)tst r5,r5bt LOCAL(div_by_zero)mov.b @(r0,r5),r1sts mach,r0addc r6,r0...LOCAL(div_by_zero):trapa #.balign 4LOCAL(le128_neg):bt LOCAL(div_by_zero)mov.b @(r0,r5),r1sts mach,r0addc r6,r0... */.balign 4LOCAL(le128_neg):sts mach,r0addc r6,r0rotcr r0mov.l @r15+,r6shad r1,r0rtsneg r0,r0ENDFUNC(GLOBAL(udivsi3_i4i))ENDFUNC(GLOBAL(sdivsi3_i4i))/* This table has been generated by divtab-sh4.c. */.balign 4.byte -7.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -2.byte -2.byte -2.byte -2.byte -1.byte -1.byte 0LOCAL(div_table_clz):.byte 0.byte 0.byte -1.byte -1.byte -2.byte -2.byte -2.byte -2.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -3.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -4.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -5.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6.byte -6/* 1/-128 .. 1/127, normalized. There is an implicit leading 1 in bit 32,or in bit 33 for powers of two. */.balign 4.long 0x0.long 0x2040811.long 0x4104105.long 0x624DD30.long 0x8421085.long 0xA6810A7.long 0xC9714FC.long 0xECF56BF.long 0x11111112.long 0x135C8114.long 0x15B1E5F8.long 0x18118119.long 0x1A7B9612.long 0x1CF06ADB.long 0x1F7047DD.long 0x21FB7813.long 0x24924925.long 0x27350B89.long 0x29E4129F.long 0x2C9FB4D9.long 0x2F684BDB.long 0x323E34A3.long 0x3521CFB3.long 0x38138139.long 0x3B13B13C.long 0x3E22CBCF.long 0x41414142.long 0x446F8657.long 0x47AE147B.long 0x4AFD6A06.long 0x4E5E0A73.long 0x51D07EAF.long 0x55555556.long 0x58ED2309.long 0x5C9882BA.long 0x60581606.long 0x642C8591.long 0x68168169.long 0x6C16C16D.long 0x702E05C1.long 0x745D1746.long 0x78A4C818.long 0x7D05F418.long 0x81818182.long 0x86186187.long 0x8ACB90F7.long 0x8F9C18FA.long 0x948B0FCE.long 0x9999999A.long 0x9EC8E952.long 0xA41A41A5.long 0xA98EF607.long 0xAF286BCB.long 0xB4E81B4F.long 0xBACF914D.long 0xC0E07039.long 0xC71C71C8.long 0xCD856891.long 0xD41D41D5.long 0xDAE6076C.long 0xE1E1E1E2.long 0xE9131AC0.long 0xF07C1F08.long 0xF81F81F9.long 0x0.long 0x4104105.long 0x8421085.long 0xC9714FC.long 0x11111112.long 0x15B1E5F8.long 0x1A7B9612.long 0x1F7047DD.long 0x24924925.long 0x29E4129F.long 0x2F684BDB.long 0x3521CFB3.long 0x3B13B13C.long 0x41414142.long 0x47AE147B.long 0x4E5E0A73.long 0x55555556.long 0x5C9882BA.long 0x642C8591.long 0x6C16C16D.long 0x745D1746.long 0x7D05F418.long 0x86186187.long 0x8F9C18FA.long 0x9999999A.long 0xA41A41A5.long 0xAF286BCB.long 0xBACF914D.long 0xC71C71C8.long 0xD41D41D5.long 0xE1E1E1E2.long 0xF07C1F08.long 0x0.long 0x8421085.long 0x11111112.long 0x1A7B9612.long 0x24924925.long 0x2F684BDB.long 0x3B13B13C.long 0x47AE147B.long 0x55555556.long 0x642C8591.long 0x745D1746.long 0x86186187.long 0x9999999A.long 0xAF286BCB.long 0xC71C71C8.long 0xE1E1E1E2.long 0x0.long 0x11111112.long 0x24924925.long 0x3B13B13C.long 0x55555556.long 0x745D1746.long 0x9999999A.long 0xC71C71C8.long 0x0.long 0x24924925.long 0x55555556.long 0x9999999A.long 0x0.long 0x55555556.long 0x0.long 0x0LOCAL(div_table_inv):.long 0x0.long 0x0.long 0x0.long 0x55555556.long 0x0.long 0x9999999A.long 0x55555556.long 0x24924925.long 0x0.long 0xC71C71C8.long 0x9999999A.long 0x745D1746.long 0x55555556.long 0x3B13B13C.long 0x24924925.long 0x11111112.long 0x0.long 0xE1E1E1E2.long 0xC71C71C8.long 0xAF286BCB.long 0x9999999A.long 0x86186187.long 0x745D1746.long 0x642C8591.long 0x55555556.long 0x47AE147B.long 0x3B13B13C.long 0x2F684BDB.long 0x24924925.long 0x1A7B9612.long 0x11111112.long 0x8421085.long 0x0.long 0xF07C1F08.long 0xE1E1E1E2.long 0xD41D41D5.long 0xC71C71C8.long 0xBACF914D.long 0xAF286BCB.long 0xA41A41A5.long 0x9999999A.long 0x8F9C18FA.long 0x86186187.long 0x7D05F418.long 0x745D1746.long 0x6C16C16D.long 0x642C8591.long 0x5C9882BA.long 0x55555556.long 0x4E5E0A73.long 0x47AE147B.long 0x41414142.long 0x3B13B13C.long 0x3521CFB3.long 0x2F684BDB.long 0x29E4129F.long 0x24924925.long 0x1F7047DD.long 0x1A7B9612.long 0x15B1E5F8.long 0x11111112.long 0xC9714FC.long 0x8421085.long 0x4104105.long 0x0.long 0xF81F81F9.long 0xF07C1F08.long 0xE9131AC0.long 0xE1E1E1E2.long 0xDAE6076C.long 0xD41D41D5.long 0xCD856891.long 0xC71C71C8.long 0xC0E07039.long 0xBACF914D.long 0xB4E81B4F.long 0xAF286BCB.long 0xA98EF607.long 0xA41A41A5.long 0x9EC8E952.long 0x9999999A.long 0x948B0FCE.long 0x8F9C18FA.long 0x8ACB90F7.long 0x86186187.long 0x81818182.long 0x7D05F418.long 0x78A4C818.long 0x745D1746.long 0x702E05C1.long 0x6C16C16D.long 0x68168169.long 0x642C8591.long 0x60581606.long 0x5C9882BA.long 0x58ED2309.long 0x55555556.long 0x51D07EAF.long 0x4E5E0A73.long 0x4AFD6A06.long 0x47AE147B.long 0x446F8657.long 0x41414142.long 0x3E22CBCF.long 0x3B13B13C.long 0x38138139.long 0x3521CFB3.long 0x323E34A3.long 0x2F684BDB.long 0x2C9FB4D9.long 0x29E4129F.long 0x27350B89.long 0x24924925.long 0x21FB7813.long 0x1F7047DD.long 0x1CF06ADB.long 0x1A7B9612.long 0x18118119.long 0x15B1E5F8.long 0x135C8114.long 0x11111112.long 0xECF56BF.long 0xC9714FC.long 0xA6810A7.long 0x8421085.long 0x624DD30.long 0x4104105.long 0x2040811/* maximum error: 0.987342 scaled: 0.921875*/#endif /* SH3 / SH4 */#endif /* L_div_table */#endif /* !__SHMEDIA__ */
