Warning: That file was not part of the compilation database. It may have many parsing errors.

1 | /* Copyright (C) 1996-2019 Free Software Foundation, Inc. |
---|---|

2 | Contributed by Richard Henderson (rth@tamu.edu) |

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, see |

17 | <http://www.gnu.org/licenses/>. */ |

18 | |

19 | /* Bytewise compare two null-terminated strings. */ |

20 | |

21 | #include <sysdep.h> |

22 | |

23 | .set noat |

24 | .set noreorder |

25 | |

26 | .text |

27 | |

28 | ENTRY(strcmp) |

29 | #ifdef PROF |

30 | ldgp gp, 0(pv) |

31 | lda AT, _mcount |

32 | jmp AT, (AT), _mcount |

33 | .prologue 1 |

34 | #else |

35 | .prologue 0 |

36 | #endif |

37 | |

38 | ldq_u t0, 0(a0) # e0 : give cache time to catch up |

39 | xor a0, a1, t2 # .. e1 : are s1 and s2 co-aligned? |

40 | ldq_u t1, 0(a1) # e0 : |

41 | and t2, 7, t2 # .. e1 : |

42 | lda t3, -1 # e0 : |

43 | bne t2, $unaligned # .. e1 : |

44 | |

45 | /* On entry to this basic block: |

46 | t0 == the first destination word for masking back in |

47 | t1 == the first source word. |

48 | t3 == -1. */ |

49 | |

50 | $aligned: |

51 | mskqh t3, a0, t3 # e0 : |

52 | nop # .. e1 : |

53 | ornot t1, t3, t1 # e0 : |

54 | ornot t0, t3, t0 # .. e1 : |

55 | cmpbge zero, t1, t7 # e0 : bits set iff null found |

56 | bne t7, $eos # e1 (zdb) |

57 | |

58 | /* Aligned compare main loop. |

59 | On entry to this basic block: |

60 | t0 == an s1 word. |

61 | t1 == an s2 word not containing a null. */ |

62 | |

63 | $a_loop: |

64 | xor t0, t1, t2 # e0 : |

65 | bne t2, $wordcmp # .. e1 (zdb) |

66 | ldq_u t1, 8(a1) # e0 : |

67 | ldq_u t0, 8(a0) # .. e1 : |

68 | addq a1, 8, a1 # e0 : |

69 | addq a0, 8, a0 # .. e1 : |

70 | cmpbge zero, t1, t7 # e0 : |

71 | beq t7, $a_loop # .. e1 (zdb) |

72 | br $eos # e1 : |

73 | |

74 | /* The two strings are not co-aligned. Align s1 and cope. */ |

75 | |

76 | $unaligned: |

77 | and a0, 7, t4 # e0 : find s1 misalignment |

78 | and a1, 7, t5 # .. e1 : find s2 misalignment |

79 | subq a1, t4, a1 # e0 : |

80 | |

81 | /* If s2 misalignment is larger than s2 misalignment, we need |

82 | extra startup checks to avoid SEGV. */ |

83 | |

84 | cmplt t4, t5, t8 # .. e1 : |

85 | beq t8, $u_head # e1 : |

86 | |

87 | mskqh t3, t5, t3 # e0 : |

88 | ornot t1, t3, t3 # e0 : |

89 | cmpbge zero, t3, t7 # e1 : is there a zero? |

90 | beq t7, $u_head # e1 : |

91 | |

92 | /* We've found a zero in the first partial word of s2. Align |

93 | our current s1 and s2 words and compare what we've got. */ |

94 | |

95 | extql t1, t5, t1 # e0 : |

96 | extql t0, a0, t0 # e0 : |

97 | cmpbge zero, t1, t7 # .. e1 : find that zero again |

98 | br $eos # e1 : and finish up |

99 | |

100 | .align 3 |

101 | $u_head: |

102 | /* We know just enough now to be able to assemble the first |

103 | full word of s2. We can still find a zero at the end of it. |

104 | |

105 | On entry to this basic block: |

106 | t0 == first word of s1 |

107 | t1 == first partial word of s2. */ |

108 | |

109 | ldq_u t2, 8(a1) # e0 : load second partial s2 word |

110 | lda t3, -1 # .. e1 : create leading garbage mask |

111 | extql t1, a1, t1 # e0 : create first s2 word |

112 | mskqh t3, a0, t3 # e0 : |

113 | extqh t2, a1, t4 # e0 : |

114 | ornot t0, t3, t0 # .. e1 : kill s1 garbage |

115 | or t1, t4, t1 # e0 : s2 word now complete |

116 | cmpbge zero, t0, t7 # .. e1 : find zero in first s1 word |

117 | ornot t1, t3, t1 # e0 : kill s2 garbage |

118 | lda t3, -1 # .. e1 : |

119 | mskql t3, a1, t3 # e0 : mask for s2[1] bits we have seen |

120 | bne t7, $eos # .. e1 : |

121 | xor t0, t1, t4 # e0 : compare aligned words |

122 | bne t4, $wordcmp # .. e1 (zdb) |

123 | or t2, t3, t3 # e0 : |

124 | cmpbge zero, t3, t7 # e1 : |

125 | bne t7, $u_final # e1 : |

126 | |

127 | /* Unaligned copy main loop. In order to avoid reading too much, |

128 | the loop is structured to detect zeros in aligned words from s2. |

129 | This has, unfortunately, effectively pulled half of a loop |

130 | iteration out into the head and half into the tail, but it does |

131 | prevent nastiness from accumulating in the very thing we want |

132 | to run as fast as possible. |

133 | |

134 | On entry to this basic block: |

135 | t2 == the unshifted low-bits from the next s2 word. */ |

136 | |

137 | .align 3 |

138 | $u_loop: |

139 | extql t2, a1, t3 # e0 : |

140 | ldq_u t2, 16(a1) # .. e1 : load next s2 high bits |

141 | ldq_u t0, 8(a0) # e0 : load next s1 word |

142 | addq a1, 8, a1 # .. e1 : |

143 | addq a0, 8, a0 # e0 : |

144 | nop # .. e1 : |

145 | extqh t2, a1, t1 # e0 : |

146 | cmpbge zero, t0, t7 # .. e1 : find zero in current s1 word |

147 | or t1, t3, t1 # e0 : |

148 | bne t7, $eos # .. e1 : |

149 | xor t0, t1, t4 # e0 : compare the words |

150 | bne t4, $wordcmp # .. e1 (zdb) |

151 | cmpbge zero, t2, t4 # e0 : find zero in next low bits |

152 | beq t4, $u_loop # .. e1 (zdb) |

153 | |

154 | /* We've found a zero in the low bits of the last s2 word. Get |

155 | the next s1 word and align them. */ |

156 | $u_final: |

157 | ldq_u t0, 8(a0) # e1 : |

158 | extql t2, a1, t1 # .. e0 : |

159 | cmpbge zero, t1, t7 # e0 : |

160 | |

161 | /* We've found a zero somewhere in a word we just read. |

162 | On entry to this basic block: |

163 | t0 == s1 word |

164 | t1 == s2 word |

165 | t7 == cmpbge mask containing the zero. */ |

166 | |

167 | .align 3 |

168 | $eos: |

169 | negq t7, t6 # e0 : create bytemask of valid data |

170 | and t6, t7, t8 # e1 : |

171 | subq t8, 1, t6 # e0 : |

172 | or t6, t8, t7 # e1 : |

173 | zapnot t0, t7, t0 # e0 : kill the garbage |

174 | zapnot t1, t7, t1 # .. e1 : |

175 | xor t0, t1, v0 # e0 : and compare |

176 | beq v0, $done # .. e1 : |

177 | |

178 | /* Here we have two differing co-aligned words in t0 & t1. |

179 | Bytewise compare them and return (t0 > t1 ? 1 : -1). */ |

180 | $wordcmp: |

181 | cmpbge t0, t1, t2 # e0 : comparison yields bit mask of ge |

182 | cmpbge t1, t0, t3 # .. e1 : |

183 | xor t2, t3, t0 # e0 : bits set iff t0/t1 bytes differ |

184 | negq t0, t1 # e1 : clear all but least bit |

185 | and t0, t1, t0 # e0 : |

186 | lda v0, -1 # .. e1 : |

187 | and t0, t2, t1 # e0 : was bit set in t0 > t1? |

188 | cmovne t1, 1, v0 # .. e1 (zdb) |

189 | |

190 | $done: |

191 | ret # e1 : |

192 | |

193 | END(strcmp) |

194 | libc_hidden_builtin_def (strcmp) |

195 |

Warning: That file was not part of the compilation database. It may have many parsing errors.