| Fenghua Yu | 9599ec0 | 2011-01-17 17:39:15 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Normally compiler builtins are used, but sometimes the compiler calls out |
| 3 | * of line code. Based on asm-i386/string.h. |
| 4 | * |
| 5 | * This assembly file is re-written from memmove_64.c file. |
| 6 | * - Copyright 2011 Fenghua Yu <fenghua.yu@intel.com> |
| 7 | */ |
| 8 | #define _STRING_C |
| 9 | #include <linux/linkage.h> |
| 10 | #include <asm/dwarf2.h> |
| 11 | |
| 12 | #undef memmove |
| 13 | |
| 14 | /* |
| 15 | * Implement memmove(). This can handle overlap between src and dst. |
| 16 | * |
| 17 | * Input: |
| 18 | * rdi: dest |
| 19 | * rsi: src |
| 20 | * rdx: count |
| 21 | * |
| 22 | * Output: |
| 23 | * rax: dest |
| 24 | */ |
| 25 | ENTRY(memmove) |
| 26 | CFI_STARTPROC |
| 27 | /* Handle more 32bytes in loop */ |
| 28 | mov %rdi, %rax |
| 29 | cmp $0x20, %rdx |
| 30 | jb 1f |
| 31 | |
| 32 | /* Decide forward/backward copy mode */ |
| 33 | cmp %rdi, %rsi |
| 34 | jb 2f |
| 35 | |
| 36 | /* |
| 37 | * movsq instruction have many startup latency |
| 38 | * so we handle small size by general register. |
| 39 | */ |
| 40 | cmp $680, %rdx |
| 41 | jb 3f |
| 42 | /* |
| 43 | * movsq instruction is only good for aligned case. |
| 44 | */ |
| 45 | |
| 46 | cmpb %dil, %sil |
| 47 | je 4f |
| 48 | 3: |
| 49 | sub $0x20, %rdx |
| 50 | /* |
| 51 | * We gobble 32byts forward in each loop. |
| 52 | */ |
| 53 | 5: |
| 54 | sub $0x20, %rdx |
| 55 | movq 0*8(%rsi), %r11 |
| 56 | movq 1*8(%rsi), %r10 |
| 57 | movq 2*8(%rsi), %r9 |
| 58 | movq 3*8(%rsi), %r8 |
| 59 | leaq 4*8(%rsi), %rsi |
| 60 | |
| 61 | movq %r11, 0*8(%rdi) |
| 62 | movq %r10, 1*8(%rdi) |
| 63 | movq %r9, 2*8(%rdi) |
| 64 | movq %r8, 3*8(%rdi) |
| 65 | leaq 4*8(%rdi), %rdi |
| 66 | jae 5b |
| 67 | addq $0x20, %rdx |
| 68 | jmp 1f |
| 69 | /* |
| 70 | * Handle data forward by movsq. |
| 71 | */ |
| 72 | .p2align 4 |
| 73 | 4: |
| 74 | movq %rdx, %rcx |
| 75 | movq -8(%rsi, %rdx), %r11 |
| 76 | lea -8(%rdi, %rdx), %r10 |
| 77 | shrq $3, %rcx |
| 78 | rep movsq |
| 79 | movq %r11, (%r10) |
| 80 | jmp 13f |
| 81 | /* |
| 82 | * Handle data backward by movsq. |
| 83 | */ |
| 84 | .p2align 4 |
| 85 | 7: |
| 86 | movq %rdx, %rcx |
| 87 | movq (%rsi), %r11 |
| 88 | movq %rdi, %r10 |
| 89 | leaq -8(%rsi, %rdx), %rsi |
| 90 | leaq -8(%rdi, %rdx), %rdi |
| 91 | shrq $3, %rcx |
| 92 | std |
| 93 | rep movsq |
| 94 | cld |
| 95 | movq %r11, (%r10) |
| 96 | jmp 13f |
| 97 | |
| 98 | /* |
| 99 | * Start to prepare for backward copy. |
| 100 | */ |
| 101 | .p2align 4 |
| 102 | 2: |
| 103 | cmp $680, %rdx |
| 104 | jb 6f |
| 105 | cmp %dil, %sil |
| 106 | je 7b |
| 107 | 6: |
| 108 | /* |
| 109 | * Calculate copy position to tail. |
| 110 | */ |
| 111 | addq %rdx, %rsi |
| 112 | addq %rdx, %rdi |
| 113 | subq $0x20, %rdx |
| 114 | /* |
| 115 | * We gobble 32byts backward in each loop. |
| 116 | */ |
| 117 | 8: |
| 118 | subq $0x20, %rdx |
| 119 | movq -1*8(%rsi), %r11 |
| 120 | movq -2*8(%rsi), %r10 |
| 121 | movq -3*8(%rsi), %r9 |
| 122 | movq -4*8(%rsi), %r8 |
| 123 | leaq -4*8(%rsi), %rsi |
| 124 | |
| 125 | movq %r11, -1*8(%rdi) |
| 126 | movq %r10, -2*8(%rdi) |
| 127 | movq %r9, -3*8(%rdi) |
| 128 | movq %r8, -4*8(%rdi) |
| 129 | leaq -4*8(%rdi), %rdi |
| 130 | jae 8b |
| 131 | /* |
| 132 | * Calculate copy position to head. |
| 133 | */ |
| 134 | addq $0x20, %rdx |
| 135 | subq %rdx, %rsi |
| 136 | subq %rdx, %rdi |
| 137 | 1: |
| 138 | cmpq $16, %rdx |
| 139 | jb 9f |
| 140 | /* |
| 141 | * Move data from 16 bytes to 31 bytes. |
| 142 | */ |
| 143 | movq 0*8(%rsi), %r11 |
| 144 | movq 1*8(%rsi), %r10 |
| 145 | movq -2*8(%rsi, %rdx), %r9 |
| 146 | movq -1*8(%rsi, %rdx), %r8 |
| 147 | movq %r11, 0*8(%rdi) |
| 148 | movq %r10, 1*8(%rdi) |
| 149 | movq %r9, -2*8(%rdi, %rdx) |
| 150 | movq %r8, -1*8(%rdi, %rdx) |
| 151 | jmp 13f |
| 152 | .p2align 4 |
| 153 | 9: |
| 154 | cmpq $8, %rdx |
| 155 | jb 10f |
| 156 | /* |
| 157 | * Move data from 8 bytes to 15 bytes. |
| 158 | */ |
| 159 | movq 0*8(%rsi), %r11 |
| 160 | movq -1*8(%rsi, %rdx), %r10 |
| 161 | movq %r11, 0*8(%rdi) |
| 162 | movq %r10, -1*8(%rdi, %rdx) |
| 163 | jmp 13f |
| 164 | 10: |
| 165 | cmpq $4, %rdx |
| 166 | jb 11f |
| 167 | /* |
| 168 | * Move data from 4 bytes to 7 bytes. |
| 169 | */ |
| 170 | movl (%rsi), %r11d |
| 171 | movl -4(%rsi, %rdx), %r10d |
| 172 | movl %r11d, (%rdi) |
| 173 | movl %r10d, -4(%rdi, %rdx) |
| 174 | jmp 13f |
| 175 | 11: |
| 176 | cmp $2, %rdx |
| 177 | jb 12f |
| 178 | /* |
| 179 | * Move data from 2 bytes to 3 bytes. |
| 180 | */ |
| 181 | movw (%rsi), %r11w |
| 182 | movw -2(%rsi, %rdx), %r10w |
| 183 | movw %r11w, (%rdi) |
| 184 | movw %r10w, -2(%rdi, %rdx) |
| 185 | jmp 13f |
| 186 | 12: |
| 187 | cmp $1, %rdx |
| 188 | jb 13f |
| 189 | /* |
| 190 | * Move data for 1 byte. |
| 191 | */ |
| 192 | movb (%rsi), %r11b |
| 193 | movb %r11b, (%rdi) |
| 194 | 13: |
| 195 | retq |
| 196 | CFI_ENDPROC |
| 197 | ENDPROC(memmove) |