Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * arch/alpha/lib/stxcpy.S |
| 3 | * Contributed by Richard Henderson (rth@tamu.edu) |
| 4 | * |
| 5 | * Copy a null-terminated string from SRC to DST. |
| 6 | * |
| 7 | * This is an internal routine used by strcpy, stpcpy, and strcat. |
| 8 | * As such, it uses special linkage conventions to make implementation |
| 9 | * of these public functions more efficient. |
| 10 | * |
| 11 | * On input: |
| 12 | * t9 = return address |
| 13 | * a0 = DST |
| 14 | * a1 = SRC |
| 15 | * |
| 16 | * On output: |
| 17 | * t12 = bitmask (with one bit set) indicating the last byte written |
| 18 | * a0 = unaligned address of the last *word* written |
| 19 | * |
| 20 | * Furthermore, v0, a3-a5, t11, and t12 are untouched. |
| 21 | */ |
| 22 | |
| 23 | #include <asm/regdef.h> |
| 24 | |
| 25 | .set noat |
| 26 | .set noreorder |
| 27 | |
| 28 | .text |
| 29 | |
| 30 | /* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that |
| 31 | doesn't like putting the entry point for a procedure somewhere in the |
| 32 | middle of the procedure descriptor. Work around this by putting the |
| 33 | aligned copy in its own procedure descriptor */ |
| 34 | |
| 35 | .ent stxcpy_aligned |
| 36 | .align 3 |
| 37 | stxcpy_aligned: |
| 38 | .frame sp, 0, t9 |
| 39 | .prologue 0 |
| 40 | |
| 41 | /* On entry to this basic block: |
| 42 | t0 == the first destination word for masking back in |
| 43 | t1 == the first source word. */ |
| 44 | |
| 45 | /* Create the 1st output word and detect 0's in the 1st input word. */ |
| 46 | lda t2, -1 # e1 : build a mask against false zero |
| 47 | mskqh t2, a1, t2 # e0 : detection in the src word |
| 48 | mskqh t1, a1, t3 # e0 : |
| 49 | ornot t1, t2, t2 # .. e1 : |
| 50 | mskql t0, a1, t0 # e0 : assemble the first output word |
| 51 | cmpbge zero, t2, t8 # .. e1 : bits set iff null found |
| 52 | or t0, t3, t1 # e0 : |
| 53 | bne t8, $a_eos # .. e1 : |
| 54 | |
| 55 | /* On entry to this basic block: |
| 56 | t0 == the first destination word for masking back in |
| 57 | t1 == a source word not containing a null. */ |
| 58 | |
| 59 | $a_loop: |
| 60 | stq_u t1, 0(a0) # e0 : |
| 61 | addq a0, 8, a0 # .. e1 : |
| 62 | ldq_u t1, 0(a1) # e0 : |
| 63 | addq a1, 8, a1 # .. e1 : |
| 64 | cmpbge zero, t1, t8 # e0 (stall) |
| 65 | beq t8, $a_loop # .. e1 (zdb) |
| 66 | |
| 67 | /* Take care of the final (partial) word store. |
| 68 | On entry to this basic block we have: |
| 69 | t1 == the source word containing the null |
| 70 | t8 == the cmpbge mask that found it. */ |
| 71 | $a_eos: |
| 72 | negq t8, t6 # e0 : find low bit set |
| 73 | and t8, t6, t12 # e1 (stall) |
| 74 | |
| 75 | /* For the sake of the cache, don't read a destination word |
| 76 | if we're not going to need it. */ |
| 77 | and t12, 0x80, t6 # e0 : |
| 78 | bne t6, 1f # .. e1 (zdb) |
| 79 | |
| 80 | /* We're doing a partial word store and so need to combine |
| 81 | our source and original destination words. */ |
| 82 | ldq_u t0, 0(a0) # e0 : |
| 83 | subq t12, 1, t6 # .. e1 : |
| 84 | zapnot t1, t6, t1 # e0 : clear src bytes >= null |
| 85 | or t12, t6, t8 # .. e1 : |
| 86 | zap t0, t8, t0 # e0 : clear dst bytes <= null |
| 87 | or t0, t1, t1 # e1 : |
| 88 | |
| 89 | 1: stq_u t1, 0(a0) # e0 : |
| 90 | ret (t9) # .. e1 : |
| 91 | |
| 92 | .end stxcpy_aligned |
| 93 | |
| 94 | .align 3 |
| 95 | .ent __stxcpy |
| 96 | .globl __stxcpy |
| 97 | __stxcpy: |
| 98 | .frame sp, 0, t9 |
| 99 | .prologue 0 |
| 100 | |
| 101 | /* Are source and destination co-aligned? */ |
| 102 | xor a0, a1, t0 # e0 : |
| 103 | unop # : |
| 104 | and t0, 7, t0 # e0 : |
| 105 | bne t0, $unaligned # .. e1 : |
| 106 | |
| 107 | /* We are co-aligned; take care of a partial first word. */ |
| 108 | ldq_u t1, 0(a1) # e0 : load first src word |
| 109 | and a0, 7, t0 # .. e1 : take care not to load a word ... |
| 110 | addq a1, 8, a1 # e0 : |
| 111 | beq t0, stxcpy_aligned # .. e1 : ... if we wont need it |
| 112 | ldq_u t0, 0(a0) # e0 : |
| 113 | br stxcpy_aligned # .. e1 : |
| 114 | |
| 115 | |
| 116 | /* The source and destination are not co-aligned. Align the destination |
| 117 | and cope. We have to be very careful about not reading too much and |
| 118 | causing a SEGV. */ |
| 119 | |
| 120 | .align 3 |
| 121 | $u_head: |
| 122 | /* We know just enough now to be able to assemble the first |
| 123 | full source word. We can still find a zero at the end of it |
| 124 | that prevents us from outputting the whole thing. |
| 125 | |
| 126 | On entry to this basic block: |
| 127 | t0 == the first dest word, for masking back in, if needed else 0 |
| 128 | t1 == the low bits of the first source word |
| 129 | t6 == bytemask that is -1 in dest word bytes */ |
| 130 | |
| 131 | ldq_u t2, 8(a1) # e0 : |
| 132 | addq a1, 8, a1 # .. e1 : |
| 133 | |
| 134 | extql t1, a1, t1 # e0 : |
| 135 | extqh t2, a1, t4 # e0 : |
| 136 | mskql t0, a0, t0 # e0 : |
| 137 | or t1, t4, t1 # .. e1 : |
| 138 | mskqh t1, a0, t1 # e0 : |
| 139 | or t0, t1, t1 # e1 : |
| 140 | |
| 141 | or t1, t6, t6 # e0 : |
| 142 | cmpbge zero, t6, t8 # .. e1 : |
| 143 | lda t6, -1 # e0 : for masking just below |
| 144 | bne t8, $u_final # .. e1 : |
| 145 | |
| 146 | mskql t6, a1, t6 # e0 : mask out the bits we have |
| 147 | or t6, t2, t2 # e1 : already extracted before |
| 148 | cmpbge zero, t2, t8 # e0 : testing eos |
| 149 | bne t8, $u_late_head_exit # .. e1 (zdb) |
| 150 | |
| 151 | /* Finally, we've got all the stupid leading edge cases taken care |
| 152 | of and we can set up to enter the main loop. */ |
| 153 | |
| 154 | stq_u t1, 0(a0) # e0 : store first output word |
| 155 | addq a0, 8, a0 # .. e1 : |
| 156 | extql t2, a1, t0 # e0 : position ho-bits of lo word |
| 157 | ldq_u t2, 8(a1) # .. e1 : read next high-order source word |
| 158 | addq a1, 8, a1 # e0 : |
| 159 | cmpbge zero, t2, t8 # .. e1 : |
| 160 | nop # e0 : |
| 161 | bne t8, $u_eos # .. e1 : |
| 162 | |
| 163 | /* Unaligned copy main loop. In order to avoid reading too much, |
| 164 | the loop is structured to detect zeros in aligned source words. |
| 165 | This has, unfortunately, effectively pulled half of a loop |
| 166 | iteration out into the head and half into the tail, but it does |
| 167 | prevent nastiness from accumulating in the very thing we want |
| 168 | to run as fast as possible. |
| 169 | |
| 170 | On entry to this basic block: |
| 171 | t0 == the shifted high-order bits from the previous source word |
| 172 | t2 == the unshifted current source word |
| 173 | |
| 174 | We further know that t2 does not contain a null terminator. */ |
| 175 | |
| 176 | .align 3 |
| 177 | $u_loop: |
| 178 | extqh t2, a1, t1 # e0 : extract high bits for current word |
| 179 | addq a1, 8, a1 # .. e1 : |
| 180 | extql t2, a1, t3 # e0 : extract low bits for next time |
| 181 | addq a0, 8, a0 # .. e1 : |
| 182 | or t0, t1, t1 # e0 : current dst word now complete |
| 183 | ldq_u t2, 0(a1) # .. e1 : load high word for next time |
| 184 | stq_u t1, -8(a0) # e0 : save the current word |
| 185 | mov t3, t0 # .. e1 : |
| 186 | cmpbge zero, t2, t8 # e0 : test new word for eos |
| 187 | beq t8, $u_loop # .. e1 : |
| 188 | |
| 189 | /* We've found a zero somewhere in the source word we just read. |
| 190 | If it resides in the lower half, we have one (probably partial) |
| 191 | word to write out, and if it resides in the upper half, we |
| 192 | have one full and one partial word left to write out. |
| 193 | |
| 194 | On entry to this basic block: |
| 195 | t0 == the shifted high-order bits from the previous source word |
| 196 | t2 == the unshifted current source word. */ |
| 197 | $u_eos: |
| 198 | extqh t2, a1, t1 # e0 : |
| 199 | or t0, t1, t1 # e1 : first (partial) source word complete |
| 200 | |
| 201 | cmpbge zero, t1, t8 # e0 : is the null in this first bit? |
| 202 | bne t8, $u_final # .. e1 (zdb) |
| 203 | |
| 204 | $u_late_head_exit: |
| 205 | stq_u t1, 0(a0) # e0 : the null was in the high-order bits |
| 206 | addq a0, 8, a0 # .. e1 : |
| 207 | extql t2, a1, t1 # e0 : |
| 208 | cmpbge zero, t1, t8 # .. e1 : |
| 209 | |
| 210 | /* Take care of a final (probably partial) result word. |
| 211 | On entry to this basic block: |
| 212 | t1 == assembled source word |
| 213 | t8 == cmpbge mask that found the null. */ |
| 214 | $u_final: |
| 215 | negq t8, t6 # e0 : isolate low bit set |
| 216 | and t6, t8, t12 # e1 : |
| 217 | |
| 218 | and t12, 0x80, t6 # e0 : avoid dest word load if we can |
| 219 | bne t6, 1f # .. e1 (zdb) |
| 220 | |
| 221 | ldq_u t0, 0(a0) # e0 : |
| 222 | subq t12, 1, t6 # .. e1 : |
| 223 | or t6, t12, t8 # e0 : |
| 224 | zapnot t1, t6, t1 # .. e1 : kill source bytes >= null |
| 225 | zap t0, t8, t0 # e0 : kill dest bytes <= null |
| 226 | or t0, t1, t1 # e1 : |
| 227 | |
| 228 | 1: stq_u t1, 0(a0) # e0 : |
| 229 | ret (t9) # .. e1 : |
| 230 | |
| 231 | /* Unaligned copy entry point. */ |
| 232 | .align 3 |
| 233 | $unaligned: |
| 234 | |
| 235 | ldq_u t1, 0(a1) # e0 : load first source word |
| 236 | |
| 237 | and a0, 7, t4 # .. e1 : find dest misalignment |
| 238 | and a1, 7, t5 # e0 : find src misalignment |
| 239 | |
| 240 | /* Conditionally load the first destination word and a bytemask |
| 241 | with 0xff indicating that the destination byte is sacrosanct. */ |
| 242 | |
| 243 | mov zero, t0 # .. e1 : |
| 244 | mov zero, t6 # e0 : |
| 245 | beq t4, 1f # .. e1 : |
| 246 | ldq_u t0, 0(a0) # e0 : |
| 247 | lda t6, -1 # .. e1 : |
| 248 | mskql t6, a0, t6 # e0 : |
| 249 | 1: |
| 250 | subq a1, t4, a1 # .. e1 : sub dest misalignment from src addr |
| 251 | |
| 252 | /* If source misalignment is larger than dest misalignment, we need |
| 253 | extra startup checks to avoid SEGV. */ |
| 254 | |
| 255 | cmplt t4, t5, t12 # e0 : |
| 256 | beq t12, $u_head # .. e1 (zdb) |
| 257 | |
| 258 | lda t2, -1 # e1 : mask out leading garbage in source |
| 259 | mskqh t2, t5, t2 # e0 : |
| 260 | nop # e0 : |
| 261 | ornot t1, t2, t3 # .. e1 : |
| 262 | cmpbge zero, t3, t8 # e0 : is there a zero? |
| 263 | beq t8, $u_head # .. e1 (zdb) |
| 264 | |
| 265 | /* At this point we've found a zero in the first partial word of |
| 266 | the source. We need to isolate the valid source data and mask |
| 267 | it into the original destination data. (Incidentally, we know |
| 268 | that we'll need at least one byte of that original dest word.) */ |
| 269 | |
| 270 | ldq_u t0, 0(a0) # e0 : |
| 271 | |
| 272 | negq t8, t6 # .. e1 : build bitmask of bytes <= zero |
| 273 | and t6, t8, t12 # e0 : |
| 274 | and a1, 7, t5 # .. e1 : |
| 275 | subq t12, 1, t6 # e0 : |
| 276 | or t6, t12, t8 # e1 : |
| 277 | srl t12, t5, t12 # e0 : adjust final null return value |
| 278 | |
| 279 | zapnot t2, t8, t2 # .. e1 : prepare source word; mirror changes |
| 280 | and t1, t2, t1 # e1 : to source validity mask |
| 281 | extql t2, a1, t2 # .. e0 : |
| 282 | extql t1, a1, t1 # e0 : |
| 283 | |
| 284 | andnot t0, t2, t0 # .. e1 : zero place for source to reside |
| 285 | or t0, t1, t1 # e1 : and put it there |
| 286 | stq_u t1, 0(a0) # .. e0 : |
| 287 | ret (t9) # e1 : |
| 288 | |
| 289 | .end __stxcpy |