projects / thirdparty / glibc.git / blame
| Commit | Line | Data |
|---|---|---|
| 7df11052 UD |
1 | /* Optimized strcmp implementation for PowerPC64. |
| 2 | Copyright (C) 2003 Free Software Foundation, Inc. | |
| 3 | This file is part of the GNU C Library. | |
| 4 | ||
| 5 | The GNU C Library is free software; you can redistribute it and/or | |
| 6 | modify it under the terms of the GNU Lesser General Public | |
| 7 | License as published by the Free Software Foundation; either | |
| 8 | version 2.1 of the License, or (at your option) any later version. | |
| 9 | ||
| 10 | The GNU C Library is distributed in the hope that it will be useful, | |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 13 | Lesser General Public License for more details. | |
| 14 | ||
| 15 | You should have received a copy of the GNU Lesser General Public | |
| 16 | License along with the GNU C Library; if not, write to the Free | |
| 17 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA | |
| 18 | 02111-1307 USA. */ | |
| 19 | ||
| 20 | #include <sysdep.h> | |
| 21 | #include <bp-sym.h> | |
| 22 | #include <bp-asm.h> | |
| 23 | ||
| 24 | /* See strlen.s for comments on how the end-of-string testing works. */ | |
| 25 | ||
| 26 | /* int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5]) */ | |
| 27 | ||
| 28 | EALIGN (BP_SYM(strncmp), 4, 0) | |
| d7d06f79 | 29 | CALL_MCOUNT 3 |
| 7df11052 UD |
30 | |
| 31 | #define rTMP r0 | |
| 32 | #define rRTN r3 | |
| 33 | #define rSTR1 r3 /* first string arg */ | |
| 34 | #define rSTR2 r4 /* second string arg */ | |
| 35 | #define rN r5 /* max string length */ | |
| 36 | /* Note: The Bounded pointer support in this code is broken. This code | |
| 37 | was inherited from PPC32 and and that support was never completed. | |
| 38 | Current PPC gcc does not support -fbounds-check or -fbounded-pointers. */ | |
| 39 | #define rWORD1 r6 /* current word in s1 */ | |
| 40 | #define rWORD2 r7 /* current word in s2 */ | |
| 41 | #define rFEFE r8 /* constant 0xfefefefefefefeff (-0x0101010101010101) */ | |
| 42 | #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ | |
| 43 | #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ | |
| 44 | #define rBITDIF r11 /* bits that differ in s1 & s2 words */ | |
| 45 | ||
| 46 | dcbt 0,rSTR1 | |
| 47 | or rTMP, rSTR2, rSTR1 | |
| 48 | lis r7F7F, 0x7f7f | |
| 49 | dcbt 0,rSTR2 | |
| 50 | clrldi. rTMP, rTMP, 61 | |
| 51 | lis rFEFE, -0x101 | |
| 52 | bne L(unaligned) | |
| 53 | /* We are doubleword alligned so set up for two loops. first a double word | |
| 54 | loop, then fall into the byte loop if any residual. */ | |
| 55 | srdi. rTMP, rN, 3 | |
| 56 | clrldi rN, rN, 61 | |
| 57 | addi rFEFE, rFEFE, -0x101 | |
| a334319f | 58 | addi r7F7F, r7F7F, 0x7f7f |
| 7df11052 UD |
59 | beq L(unaligned) |
| 60 | ||
| 61 | mtctr rTMP /* Power4 wants mtctr 1st in dispatch group. */ | |
| 62 | ld rWORD1, 0(rSTR1) | |
| 63 | ld rWORD2, 0(rSTR2) | |
| 64 | sldi rTMP, rFEFE, 32 | |
| 65 | insrdi r7F7F, r7F7F, 32, 0 | |
| 66 | add rFEFE, rFEFE, rTMP | |
| 67 | b L(g1) | |
| 68 | ||
| 69 | L(g0): | |
| 70 | ldu rWORD1, 8(rSTR1) | |
| 71 | bne- cr1, L(different) | |
| 72 | ldu rWORD2, 8(rSTR2) | |
| 73 | L(g1): add rTMP, rFEFE, rWORD1 | |
| 74 | nor rNEG, r7F7F, rWORD1 | |
| 75 | bdz L(tail) | |
| 76 | and. rTMP, rTMP, rNEG | |
| 77 | cmpd cr1, rWORD1, rWORD2 | |
| 78 | beq+ L(g0) | |
| 79 | ||
| 80 | /* OK. We've hit the end of the string. We need to be careful that | |
| 81 | we don't compare two strings as different because of gunk beyond | |
| 82 | the end of the strings... */ | |
| 83 | ||
| 84 | L(endstring): | |
| 85 | and rTMP, r7F7F, rWORD1 | |
| 86 | beq cr1, L(equal) | |
| 87 | add rTMP, rTMP, r7F7F | |
| 88 | xor. rBITDIF, rWORD1, rWORD2 | |
| 89 | ||
| 90 | andc rNEG, rNEG, rTMP | |
| 91 | blt- L(highbit) | |
| 92 | cntlzd rBITDIF, rBITDIF | |
| 93 | cntlzd rNEG, rNEG | |
| 94 | addi rNEG, rNEG, 7 | |
| 95 | cmpd cr1, rNEG, rBITDIF | |
| 96 | sub rRTN, rWORD1, rWORD2 | |
| 97 | blt- cr1, L(equal) | |
| 98 | sradi rRTN, rRTN, 63 | |
| 99 | ori rRTN, rRTN, 1 | |
| 100 | blr | |
| 101 | L(equal): | |
| 102 | li rRTN, 0 | |
| 103 | blr | |
| 104 | ||
| 105 | L(different): | |
| 106 | ldu rWORD1, -8(rSTR1) | |
| 107 | xor. rBITDIF, rWORD1, rWORD2 | |
| 108 | sub rRTN, rWORD1, rWORD2 | |
| 109 | blt- L(highbit) | |
| 110 | sradi rRTN, rRTN, 63 | |
| 111 | ori rRTN, rRTN, 1 | |
| 112 | blr | |
| 113 | L(highbit): | |
| 114 | srdi rWORD2, rWORD2, 56 | |
| 115 | srdi rWORD1, rWORD1, 56 | |
| 116 | sub rRTN, rWORD1, rWORD2 | |
| 117 | blr | |
| 118 | ||
| 119 | ||
| 120 | /* Oh well. In this case, we just do a byte-by-byte comparison. */ | |
| 121 | .align 4 | |
| 122 | L(tail): | |
| 123 | and. rTMP, rTMP, rNEG | |
| 124 | cmpd cr1, rWORD1, rWORD2 | |
| 125 | bne- L(endstring) | |
| 126 | addi rSTR1, rSTR1, 8 | |
| 127 | bne- cr1, L(different) | |
| 128 | addi rSTR2, rSTR2, 8 | |
| 129 | L(unaligned): | |
| 130 | mtctr rN /* Power4 wants mtctr 1st in dispatch group */ | |
| a334319f UD |
131 | cmpdi rN,0 |
| 132 | lbz rWORD1, 0(rSTR1) | |
| 133 | lbz rWORD2, 0(rSTR2) | |
| 134 | bgt L(u1) | |
| 7df11052 UD |
135 | L(ux): |
| 136 | li rRTN, 0 | |
| 137 | blr | |
| a334319f | 138 | |
| 7df11052 UD |
139 | L(u0): |
| 140 | lbzu rWORD2, 1(rSTR2) | |
| 141 | L(u1): | |
| 142 | bdz L(u3) | |
| 143 | cmpdi cr1, rWORD1, 0 | |
| 144 | cmpd rWORD1, rWORD2 | |
| 145 | beq- cr1, L(u3) | |
| 146 | lbzu rWORD1, 1(rSTR1) | |
| 147 | bne- L(u2) | |
| 148 | lbzu rWORD2, 1(rSTR2) | |
| 149 | bdz L(u3) | |
| 150 | cmpdi cr1, rWORD1, 0 | |
| 151 | cmpd rWORD1, rWORD2 | |
| 152 | bne- L(u3) | |
| 153 | lbzu rWORD1, 1(rSTR1) | |
| 154 | bne+ cr1, L(u0) | |
| 155 | ||
| 156 | L(u2): lbzu rWORD1, -1(rSTR1) | |
| 157 | L(u3): sub rRTN, rWORD1, rWORD2 | |
| 158 | blr | |
| 159 | END (BP_SYM (strncmp)) | |
| 160 | libc_hidden_builtin_def (strncmp) |