]>
Commit | Line | Data |
---|---|---|
a37f6605 AB |
1 | #------------------------------------------------------------------------------\r |
2 | #\r | |
3 | # CopyMem() worker for ARM\r | |
4 | #\r | |
5 | # This file started out as C code that did 64 bit moves if the buffer was\r | |
6 | # 32-bit aligned, else it does a byte copy. It also does a byte copy for\r | |
7 | # any trailing bytes. It was updated to do 32-byte copies using stm/ldm.\r | |
8 | #\r | |
9 | # Copyright (c) 2008 - 2010, Apple Inc. All rights reserved.<BR>\r | |
10 | # Copyright (c) 2016, Linaro Ltd. All rights reserved.<BR>\r | |
11 | # This program and the accompanying materials\r | |
12 | # are licensed and made available under the terms and conditions of the BSD License\r | |
13 | # which accompanies this distribution. The full text of the license may be found at\r | |
14 | # http://opensource.org/licenses/bsd-license.php\r | |
15 | #\r | |
16 | # THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r | |
17 | # WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r | |
18 | #\r | |
19 | #------------------------------------------------------------------------------\r | |
20 | \r | |
21 | .text\r | |
22 | .thumb\r | |
23 | .syntax unified\r | |
24 | \r | |
25 | /**\r | |
26 | Copy Length bytes from Source to Destination. Overlap is OK.\r | |
27 | \r | |
28 | This implementation\r | |
29 | \r | |
30 | @param Destination Target of copy\r | |
31 | @param Source Place to copy from\r | |
32 | @param Length Number of bytes to copy\r | |
33 | \r | |
34 | @return Destination\r | |
35 | \r | |
36 | \r | |
37 | VOID *\r | |
38 | EFIAPI\r | |
39 | InternalMemCopyMem (\r | |
40 | OUT VOID *DestinationBuffer,\r | |
41 | IN CONST VOID *SourceBuffer,\r | |
42 | IN UINTN Length\r | |
43 | )\r | |
44 | **/\r | |
45 | ASM_GLOBAL ASM_PFX(InternalMemCopyMem)\r | |
46 | ASM_PFX(InternalMemCopyMem):\r | |
47 | push {r4-r11, lr}\r | |
48 | // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length)\r | |
49 | mov r11, r0\r | |
50 | mov r10, r0\r | |
51 | mov r12, r2\r | |
52 | mov r14, r1\r | |
53 | \r | |
54 | cmp r11, r1\r | |
55 | // If (dest < source)\r | |
56 | bcc memcopy_check_optim_default\r | |
57 | \r | |
58 | // If (source + length < dest)\r | |
59 | rsb r3, r1, r11\r | |
60 | cmp r12, r3\r | |
61 | bcc memcopy_check_optim_default\r | |
62 | b memcopy_check_optim_overlap\r | |
63 | \r | |
64 | memcopy_check_optim_default:\r | |
65 | // Check if we can use an optimized path ((length >= 32) && destination word-aligned && source word-aligned) for the memcopy (optimized path if r0 == 1)\r | |
66 | tst r0, #0xF\r | |
67 | it ne\r | |
68 | movne r0, #0\r | |
69 | bne memcopy_default\r | |
70 | tst r1, #0xF\r | |
71 | ite ne\r | |
72 | movne r3, #0\r | |
73 | moveq r3, #1\r | |
74 | cmp r2, #31\r | |
75 | ite ls\r | |
76 | movls r0, #0\r | |
77 | andhi r0, r3, #1\r | |
78 | b memcopy_default\r | |
79 | \r | |
80 | memcopy_check_optim_overlap:\r | |
81 | // r10 = dest_end, r14 = source_end\r | |
82 | add r10, r11, r12\r | |
83 | add r14, r12, r1\r | |
84 | \r | |
85 | // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned)\r | |
86 | cmp r2, #31\r | |
87 | ite ls\r | |
88 | movls r0, #0\r | |
89 | movhi r0, #1\r | |
90 | tst r10, #0xF\r | |
91 | it ne\r | |
92 | movne r0, #0\r | |
93 | tst r14, #0xF\r | |
94 | it ne\r | |
95 | movne r0, #0\r | |
96 | b memcopy_overlapped\r | |
97 | \r | |
98 | memcopy_overlapped_non_optim:\r | |
99 | // We read 1 byte from the end of the source buffer\r | |
100 | sub r3, r14, #1\r | |
101 | sub r12, r12, #1\r | |
102 | ldrb r3, [r3, #0]\r | |
103 | sub r2, r10, #1\r | |
104 | cmp r12, #0\r | |
105 | // We write 1 byte at the end of the dest buffer\r | |
106 | sub r10, r10, #1\r | |
107 | sub r14, r14, #1\r | |
108 | strb r3, [r2, #0]\r | |
109 | bne memcopy_overlapped_non_optim\r | |
110 | b memcopy_end\r | |
111 | \r | |
112 | // r10 = dest_end, r14 = source_end\r | |
113 | memcopy_overlapped:\r | |
114 | // Are we in the optimized case ?\r | |
115 | cmp r0, #0\r | |
116 | beq memcopy_overlapped_non_optim\r | |
117 | \r | |
118 | // Optimized Overlapped - Read 32 bytes\r | |
119 | sub r14, r14, #32\r | |
120 | sub r12, r12, #32\r | |
121 | cmp r12, #31\r | |
122 | ldmia r14, {r2-r9}\r | |
123 | \r | |
124 | // If length is less than 32 then disable optim\r | |
125 | it ls\r | |
126 | movls r0, #0\r | |
127 | \r | |
128 | cmp r12, #0\r | |
129 | \r | |
130 | // Optimized Overlapped - Write 32 bytes\r | |
131 | sub r10, r10, #32\r | |
132 | stmia r10, {r2-r9}\r | |
133 | \r | |
134 | // while (length != 0)\r | |
135 | bne memcopy_overlapped\r | |
136 | b memcopy_end\r | |
137 | \r | |
138 | memcopy_default_non_optim:\r | |
139 | // Byte copy\r | |
140 | ldrb r3, [r14], #1\r | |
141 | sub r12, r12, #1\r | |
142 | strb r3, [r10], #1\r | |
143 | \r | |
144 | memcopy_default:\r | |
145 | cmp r12, #0\r | |
146 | beq memcopy_end\r | |
147 | \r | |
148 | // r10 = dest, r14 = source\r | |
149 | memcopy_default_loop:\r | |
150 | cmp r0, #0\r | |
151 | beq memcopy_default_non_optim\r | |
152 | \r | |
153 | // Optimized memcopy - Read 32 Bytes\r | |
154 | sub r12, r12, #32\r | |
155 | cmp r12, #31\r | |
156 | ldmia r14!, {r2-r9}\r | |
157 | \r | |
158 | // If length is less than 32 then disable optim\r | |
159 | it ls\r | |
160 | movls r0, #0\r | |
161 | \r | |
162 | cmp r12, #0\r | |
163 | \r | |
164 | // Optimized memcopy - Write 32 Bytes\r | |
165 | stmia r10!, {r2-r9}\r | |
166 | \r | |
167 | // while (length != 0)\r | |
168 | bne memcopy_default_loop\r | |
169 | \r | |
170 | memcopy_end:\r | |
171 | mov r0, r11\r | |
172 | pop {r4-r11, pc}\r |