Linux/arch/riscv/lib/memmove.S

Version: ~ [ linux-6.12-rc7 ] ~ [ linux-6.11.7 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.60 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.116 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.171 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.229 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.285 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.323 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.12 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

Diff markup

1 /* SPDX-License-Identifier: GPL-2.0-only */ << 2 /* 1 /* 3 * Copyright (C) 2022 Michael T. Kloos <michael !! 2 * arch/alpha/lib/memmove.S >> 3 * >> 4 * Barely optimized memmove routine for Alpha EV5. >> 5 * >> 6 * This is hand-massaged output from the original memcpy.c. We defer to >> 7 * memcpy whenever possible; the backwards copy loops are not unrolled. 4 */ 8 */ >> 9 >> 10 .set noat >> 11 .set noreorder >> 12 .text >> 13 >> 14 .align 4 >> 15 .globl memmove >> 16 .ent memmove >> 17 memmove: >> 18 ldgp $29, 0($27) >> 19 unop >> 20 nop >> 21 .prologue 1 >> 22 >> 23 addq $16,$18,$4 >> 24 addq $17,$18,$5 >> 25 cmpule $4,$17,$1 /* dest + n <= src */ >> 26 cmpule $5,$16,$2 /* dest >= src + n */ >> 27 >> 28 bis $1,$2,$1 >> 29 mov $16,$0 >> 30 xor $16,$17,$2 >> 31 bne $1,memcpy !samegp >> 32 >> 33 and $2,7,$2 /* Test for src/dest co-alignment. */ >> 34 and $16,7,$1 >> 35 cmpule $16,$17,$3 >> 36 bne $3,$memmove_up /* dest < src */ >> 37 >> 38 and $4,7,$1 >> 39 bne $2,$misaligned_dn >> 40 unop >> 41 beq $1,$skip_aligned_byte_loop_head_dn >> 42 >> 43 $aligned_byte_loop_head_dn: >> 44 lda $4,-1($4) >> 45 lda $5,-1($5) >> 46 unop >> 47 ble $18,$egress >> 48 >> 49 ldq_u $3,0($5) >> 50 ldq_u $2,0($4) >> 51 lda $18,-1($18) >> 52 extbl $3,$5,$1 >> 53 >> 54 insbl $1,$4,$1 >> 55 mskbl $2,$4,$2 >> 56 bis $1,$2,$1 >> 57 and $4,7,$6 >> 58 >> 59 stq_u $1,0($4) >> 60 bne $6,$aligned_byte_loop_head_dn >> 61 >> 62 $skip_aligned_byte_loop_head_dn: >> 63 lda $18,-8($18) >> 64 blt $18,$skip_aligned_word_loop_dn >> 65 >> 66 $aligned_word_loop_dn: >> 67 ldq $1,-8($5) >> 68 nop >> 69 lda $5,-8($5) >> 70 lda $18,-8($18) >> 71 >> 72 stq $1,-8($4) >> 73 nop >> 74 lda $4,-8($4) >> 75 bge $18,$aligned_word_loop_dn >> 76 >> 77 $skip_aligned_word_loop_dn: >> 78 lda $18,8($18) >> 79 bgt $18,$byte_loop_tail_dn >> 80 unop >> 81 ret $31,($26),1 >> 82 >> 83 .align 4 >> 84 $misaligned_dn: >> 85 nop >> 86 fnop >> 87 unop >> 88 beq $18,$egress >> 89 >> 90 $byte_loop_tail_dn: >> 91 ldq_u $3,-1($5) >> 92 ldq_u $2,-1($4) >> 93 lda $5,-1($5) >> 94 lda $4,-1($4) >> 95 >> 96 lda $18,-1($18) >> 97 extbl $3,$5,$1 >> 98 insbl $1,$4,$1 >> 99 mskbl $2,$4,$2 >> 100 >> 101 bis $1,$2,$1 >> 102 stq_u $1,0($4) >> 103 bgt $18,$byte_loop_tail_dn >> 104 br $egress >> 105 >> 106 $memmove_up: >> 107 mov $16,$4 >> 108 mov $17,$5 >> 109 bne $2,$misaligned_up >> 110 beq $1,$skip_aligned_byte_loop_head_up >> 111 >> 112 $aligned_byte_loop_head_up: >> 113 unop >> 114 ble $18,$egress >> 115 ldq_u $3,0($5) >> 116 ldq_u $2,0($4) >> 117 >> 118 lda $18,-1($18) >> 119 extbl $3,$5,$1 >> 120 insbl $1,$4,$1 >> 121 mskbl $2,$4,$2 >> 122 >> 123 bis $1,$2,$1 >> 124 lda $5,1($5) >> 125 stq_u $1,0($4) >> 126 lda $4,1($4) >> 127 >> 128 and $4,7,$6 >> 129 bne $6,$aligned_byte_loop_head_up >> 130 >> 131 $skip_aligned_byte_loop_head_up: >> 132 lda $18,-8($18) >> 133 blt $18,$skip_aligned_word_loop_up >> 134 >> 135 $aligned_word_loop_up: >> 136 ldq $1,0($5) >> 137 nop >> 138 lda $5,8($5) >> 139 lda $18,-8($18) >> 140 >> 141 stq $1,0($4) >> 142 nop >> 143 lda $4,8($4) >> 144 bge $18,$aligned_word_loop_up >> 145 >> 146 $skip_aligned_word_loop_up: >> 147 lda $18,8($18) >> 148 bgt $18,$byte_loop_tail_up >> 149 unop >> 150 ret $31,($26),1 >> 151 >> 152 .align 4 >> 153 $misaligned_up: >> 154 nop >> 155 fnop >> 156 unop >> 157 beq $18,$egress >> 158 >> 159 $byte_loop_tail_up: >> 160 ldq_u $3,0($5) >> 161 ldq_u $2,0($4) >> 162 lda $18,-1($18) >> 163 extbl $3,$5,$1 >> 164 >> 165 insbl $1,$4,$1 >> 166 mskbl $2,$4,$2 >> 167 bis $1,$2,$1 >> 168 stq_u $1,0($4) >> 169 >> 170 lda $5,1($5) >> 171 lda $4,1($4) >> 172 nop >> 173 bgt $18,$byte_loop_tail_up >> 174 >> 175 $egress: >> 176 ret $31,($26),1 >> 177 nop >> 178 nop >> 179 nop 5 180 6 #include <linux/linkage.h> !! 181 .end memmove 7 #include <asm/asm.h> << 8 << 9 SYM_FUNC_START(__memmove) << 10 /* << 11 * Returns << 12 * a0 - dest << 13 * << 14 * Parameters << 15 * a0 - Inclusive first byte of dest << 16 * a1 - Inclusive first byte of src << 17 * a2 - Length of copy n << 18 * << 19 * Because the return matches the para << 20 * we will not clobber or modify that << 21 * << 22 * Note: This currently only works on << 23 * To port to big-endian, reverse the << 24 * in the 2 misaligned fixup copy loop << 25 */ << 26 << 27 /* Return if nothing to do */ << 28 beq a0, a1, .Lreturn_from_memmove << 29 beqz a2, .Lreturn_from_memmove << 30 << 31 /* << 32 * Register Uses << 33 * Forward Copy: a1 - Index count << 34 * Reverse Copy: a4 - Index count << 35 * Forward Copy: t3 - Index count << 36 * Reverse Copy: t4 - Index count << 37 * Both Copy Modes: t5 - Inclusive f << 38 * Both Copy Modes: t6 - Non-Inclusi << 39 * Both Copy Modes: t0 - Link / Temp << 40 * Both Copy Modes: t1 - Temporary f << 41 * Both Copy Modes: t2 - Temporary f << 42 * Both Copy Modes: a5 - dest to src << 43 * Both Copy Modes: a6 - Shift ammou << 44 * Both Copy Modes: a7 - Inverse Shi << 45 * Both Copy Modes: a2 - Alternate b << 46 */ << 47 << 48 /* << 49 * Solve for some register values now. << 50 * Byte copy does not need t5 or t6. << 51 */ << 52 mv t3, a0 << 53 add t4, a0, a2 << 54 add a4, a1, a2 << 55 << 56 /* << 57 * Byte copy if copying less than (2 * << 58 * cause problems with the bulk copy i << 59 * small enough not to bother. << 60 */ << 61 andi t0, a2, -(2 * SZREG) << 62 beqz t0, .Lbyte_copy << 63 << 64 /* << 65 * Now solve for t5 and t6. << 66 */ << 67 andi t5, t3, -SZREG << 68 andi t6, t4, -SZREG << 69 /* << 70 * If dest(Register t3) rounded down t << 71 * aligned SZREG address, does not equ << 72 * to find the low-bound of SZREG alig << 73 * region. Note that this could overs << 74 * region if n is less than SZREG. Th << 75 * we always byte copy if n is less th << 76 * Otherwise, dest is already naturall << 77 */ << 78 beq t5, t3, 1f << 79 addi t5, t5, SZREG << 80 1: << 81 << 82 /* << 83 * If the dest and src are co-aligned << 84 * no need for the full rigmarole of a << 85 * Instead, do a simpler co-aligned co << 86 */ << 87 xor t0, a0, a1 << 88 andi t1, t0, (SZREG - 1) << 89 beqz t1, .Lcoaligned_copy << 90 /* Fall through to misaligned fixup co << 91 << 92 .Lmisaligned_fixup_copy: << 93 bltu a1, a0, .Lmisaligned_fixup_copy_r << 94 << 95 .Lmisaligned_fixup_copy_forward: << 96 jal t0, .Lbyte_copy_until_aligned_for << 97 << 98 andi a5, a1, (SZREG - 1) /* Find the a << 99 slli a6, a5, 3 /* Multiply by 8 to con << 100 sub a5, a1, t3 /* Find the difference << 101 andi a1, a1, -SZREG /* Align the src p << 102 addi a2, t6, SZREG /* The other breakp << 103 << 104 /* << 105 * Compute The Inverse Shift << 106 * a7 = XLEN - a6 = XLEN + -a6 << 107 * 2s complement negation to find the << 108 * Add that to XLEN. XLEN = SZREG * 8 << 109 */ << 110 not a7, a6 << 111 addi a7, a7, (SZREG * 8 + 1) << 112 << 113 /* << 114 * Fix Misalignment Copy Loop - Forwar << 115 * load_val0 = load_ptr[0]; << 116 * do { << 117 * load_val1 = load_ptr[1]; << 118 * store_ptr += 2; << 119 * store_ptr[0 - 2] = (load_val0 << 120 * << 121 * if (store_ptr == {a2}) << 122 * break; << 123 * << 124 * load_val0 = load_ptr[2]; << 125 * load_ptr += 2; << 126 * store_ptr[1 - 2] = (load_val1 << 127 * << 128 * } while (store_ptr != store_ptr_end << 129 * store_ptr = store_ptr_end; << 130 */ << 131 << 132 REG_L t0, (0 * SZREG)(a1) << 133 1: << 134 REG_L t1, (1 * SZREG)(a1) << 135 addi t3, t3, (2 * SZREG) << 136 srl t0, t0, a6 << 137 sll t2, t1, a7 << 138 or t2, t0, t2 << 139 REG_S t2, ((0 * SZREG) - (2 * SZREG))( << 140 << 141 beq t3, a2, 2f << 142 << 143 REG_L t0, (2 * SZREG)(a1) << 144 addi a1, a1, (2 * SZREG) << 145 srl t1, t1, a6 << 146 sll t2, t0, a7 << 147 or t2, t1, t2 << 148 REG_S t2, ((1 * SZREG) - (2 * SZREG))( << 149 << 150 bne t3, t6, 1b << 151 2: << 152 mv t3, t6 /* Fix the dest pointer i << 153 << 154 add a1, t3, a5 /* Restore the src poi << 155 j .Lbyte_copy_forward /* Copy any rema << 156 << 157 .Lmisaligned_fixup_copy_reverse: << 158 jal t0, .Lbyte_copy_until_aligned_rev << 159 << 160 andi a5, a4, (SZREG - 1) /* Find the a << 161 slli a6, a5, 3 /* Multiply by 8 to con << 162 sub a5, a4, t4 /* Find the difference << 163 andi a4, a4, -SZREG /* Align the src p << 164 addi a2, t5, -SZREG /* The other break << 165 << 166 /* << 167 * Compute The Inverse Shift << 168 * a7 = XLEN - a6 = XLEN + -a6 << 169 * 2s complement negation to find the << 170 * Add that to XLEN. XLEN = SZREG * 8 << 171 */ << 172 not a7, a6 << 173 addi a7, a7, (SZREG * 8 + 1) << 174 << 175 /* << 176 * Fix Misalignment Copy Loop - Revers << 177 * load_val1 = load_ptr[0]; << 178 * do { << 179 * load_val0 = load_ptr[-1]; << 180 * store_ptr -= 2; << 181 * store_ptr[1] = (load_val0 >> { << 182 * << 183 * if (store_ptr == {a2}) << 184 * break; << 185 * << 186 * load_val1 = load_ptr[-2]; << 187 * load_ptr -= 2; << 188 * store_ptr[0] = (load_val1 >> { << 189 * << 190 * } while (store_ptr != store_ptr_end << 191 * store_ptr = store_ptr_end; << 192 */ << 193 << 194 REG_L t1, ( 0 * SZREG)(a4) << 195 1: << 196 REG_L t0, (-1 * SZREG)(a4) << 197 addi t4, t4, (-2 * SZREG) << 198 sll t1, t1, a7 << 199 srl t2, t0, a6 << 200 or t2, t1, t2 << 201 REG_S t2, ( 1 * SZREG)(t4) << 202 << 203 beq t4, a2, 2f << 204 << 205 REG_L t1, (-2 * SZREG)(a4) << 206 addi a4, a4, (-2 * SZREG) << 207 sll t0, t0, a7 << 208 srl t2, t1, a6 << 209 or t2, t0, t2 << 210 REG_S t2, ( 0 * SZREG)(t4) << 211 << 212 bne t4, t5, 1b << 213 2: << 214 mv t4, t5 /* Fix the dest pointer i << 215 << 216 add a4, t4, a5 /* Restore the src poi << 217 j .Lbyte_copy_reverse /* Copy any rema << 218 << 219 /* << 220 * Simple copy loops for SZREG co-aligned memo << 221 * These also make calls to do byte copies for << 222 * data at their terminations. << 223 */ << 224 .Lcoaligned_copy: << 225 bltu a1, a0, .Lcoaligned_copy_reverse << 226 << 227 .Lcoaligned_copy_forward: << 228 jal t0, .Lbyte_copy_until_aligned_forw << 229 << 230 1: << 231 REG_L t1, ( 0 * SZREG)(a1) << 232 addi a1, a1, SZREG << 233 addi t3, t3, SZREG << 234 REG_S t1, (-1 * SZREG)(t3) << 235 bne t3, t6, 1b << 236 << 237 j .Lbyte_copy_forward /* Copy any rema << 238 << 239 .Lcoaligned_copy_reverse: << 240 jal t0, .Lbyte_copy_until_aligned_reve << 241 << 242 1: << 243 REG_L t1, (-1 * SZREG)(a4) << 244 addi a4, a4, -SZREG << 245 addi t4, t4, -SZREG << 246 REG_S t1, ( 0 * SZREG)(t4) << 247 bne t4, t5, 1b << 248 << 249 j .Lbyte_copy_reverse /* Copy any rema << 250 << 251 /* << 252 * These are basically sub-functions within th << 253 * are used to byte copy until the dest pointe << 254 * At which point, a bulk copy method can be u << 255 * calling code. These work on the same regis << 256 * copy loops. Therefore, the register values << 257 * up from where they were left and we avoid c << 258 * without any overhead except the call in and << 259 */ << 260 .Lbyte_copy_until_aligned_forward: << 261 beq t3, t5, 2f << 262 1: << 263 lb t1, 0(a1) << 264 addi a1, a1, 1 << 265 addi t3, t3, 1 << 266 sb t1, -1(t3) << 267 bne t3, t5, 1b << 268 2: << 269 jalr zero, 0x0(t0) /* Return to multib << 270 << 271 .Lbyte_copy_until_aligned_reverse: << 272 beq t4, t6, 2f << 273 1: << 274 lb t1, -1(a4) << 275 addi a4, a4, -1 << 276 addi t4, t4, -1 << 277 sb t1, 0(t4) << 278 bne t4, t6, 1b << 279 2: << 280 jalr zero, 0x0(t0) /* Return to multib << 281 << 282 /* << 283 * Simple byte copy loops. << 284 * These will byte copy until they reach the e << 285 * At that point, they will call to return fro << 286 */ << 287 .Lbyte_copy: << 288 bltu a1, a0, .Lbyte_copy_reverse << 289 << 290 .Lbyte_copy_forward: << 291 beq t3, t4, 2f << 292 1: << 293 lb t1, 0(a1) << 294 addi a1, a1, 1 << 295 addi t3, t3, 1 << 296 sb t1, -1(t3) << 297 bne t3, t4, 1b << 298 2: << 299 ret << 300 << 301 .Lbyte_copy_reverse: << 302 beq t4, t3, 2f << 303 1: << 304 lb t1, -1(a4) << 305 addi a4, a4, -1 << 306 addi t4, t4, -1 << 307 sb t1, 0(t4) << 308 bne t4, t3, 1b << 309 2: << 310 << 311 .Lreturn_from_memmove: << 312 ret << 313 << 314 SYM_FUNC_END(__memmove) << 315 SYM_FUNC_ALIAS_WEAK(memmove, __memmove) << 316 SYM_FUNC_ALIAS(__pi_memmove, __memmove) << 317 SYM_FUNC_ALIAS(__pi___memmove, __memmove) <<

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

TOMOYO Linux Cross Reference
Linux/arch/riscv/lib/memmove.S

Diff markup

Differences between /arch/riscv/lib/memmove.S (Version linux-6.12-rc7) and /arch/alpha/lib/memmove.S (Version linux-3.10.108)

TOMOYO Linux Cross Reference Linux/arch/riscv/lib/memmove.S

Diff markup

Differences between /arch/riscv/lib/memmove.S (Version linux-6.12-rc7) and /arch/alpha/lib/memmove.S (Version linux-3.10.108)

TOMOYO Linux Cross Reference
Linux/arch/riscv/lib/memmove.S