)\r
**/\r
EXPORT InternalMemCopyMem\r
-
- AREA AsmMemStuff, CODE, READONLY
+\r
+ AREA AsmMemStuff, CODE, READONLY\r
\r
InternalMemCopyMem\r
- stmfd sp!, {r4-r11, lr}
- // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length)
- mov r11, r0
- mov r10, r0
- mov r12, r2
- mov r14, r1
+ stmfd sp!, {r4-r11, lr}\r
+ // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length)\r
+ mov r11, r0\r
+ mov r10, r0\r
+ mov r12, r2\r
+ mov r14, r1\r
\r
memcopy_check_overlapped\r
cmp r11, r1\r
memcopy_check_optim_default\r
// 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
tst r0, #0xF\r
- movne r0, #0
- bne memcopy_default
- tst r1, #0xF
- movne r3, #0
- moveq r3, #1
- cmp r2, #31
- movls r0, #0
- andhi r0, r3, #1
- b memcopy_default
-
-memcopy_check_optim_overlap
- // r10 = dest_end, r14 = source_end
- add r10, r11, r12
- add r14, r12, r1
-
- // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned)
- cmp r2, #31
- movls r0, #0
- movhi r0, #1
- tst r10, #0xF
- movne r0, #0
- tst r14, #0xF
- movne r0, #0
+ movne r0, #0\r
+ bne memcopy_default\r
+ tst r1, #0xF\r
+ movne r3, #0\r
+ moveq r3, #1\r
+ cmp r2, #31\r
+ movls r0, #0\r
+ andhi r0, r3, #1\r
+ b memcopy_default\r
+ \r
+memcopy_check_optim_overlap\r
+ // r10 = dest_end, r14 = source_end\r
+ add r10, r11, r12\r
+ add r14, r12, r1\r
+\r
+ // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned)\r
+ cmp r2, #31\r
+ movls r0, #0\r
+ movhi r0, #1\r
+ tst r10, #0xF\r
+ movne r0, #0\r
+ tst r14, #0xF\r
+ movne r0, #0\r
b memcopy_overlapped\r
-
+ \r
memcopy_overlapped_non_optim\r
// We read 1 byte from the end of the source buffer\r
- sub r3, r14, #1
- sub r12, r12, #1
- ldrb r3, [r3, #0]
- sub r2, r10, #1
- cmp r12, #0
+ sub r3, r14, #1\r
+ sub r12, r12, #1\r
+ ldrb r3, [r3, #0] \r
+ sub r2, r10, #1\r
+ cmp r12, #0\r
// We write 1 byte at the end of the dest buffer\r
- sub r10, r10, #1
- sub r14, r14, #1
- strb r3, [r2, #0]
+ sub r10, r10, #1\r
+ sub r14, r14, #1\r
+ strb r3, [r2, #0]\r
bne memcopy_overlapped_non_optim\r
b memcopy_end\r
\r
// Are we in the optimized case ?\r
cmp r0, #0\r
beq memcopy_overlapped_non_optim\r
-
+ \r
// Optimized Overlapped - Read 32 bytes\r
sub r14, r14, #32\r
sub r12, r12, #32\r
cmp r12, #31\r
ldmia r14, {r2-r9}\r
-
+ \r
// If length is less than 32 then disable optim\r
movls r0, #0\r
-
+ \r
cmp r12, #0\r
\r
// Optimized Overlapped - Write 32 bytes \r
\r
memcopy_default_non_optim\r
// Byte copy\r
- ldrb r3, [r14], #1
- sub r12, r12, #1
- strb r3, [r10], #1
-
+ ldrb r3, [r14], #1 \r
+ sub r12, r12, #1\r
+ strb r3, [r10], #1\r
+ \r
memcopy_default\r
- cmp r12, #0
- beq memcopy_end
-
+ cmp r12, #0\r
+ beq memcopy_end\r
+ \r
// r10 = dest, r14 = source\r
memcopy_default_loop\r
- cmp r0, #0
+ cmp r0, #0\r
beq memcopy_default_non_optim\r
-
+ \r
// Optimized memcopy - Read 32 Bytes\r
- sub r12, r12, #32
- cmp r12, #31
- ldmia r14!, {r2-r9}
+ sub r12, r12, #32\r
+ cmp r12, #31\r
+ ldmia r14!, {r2-r9}\r
\r
// If length is less than 32 then disable optim\r
- movls r0, #0
+ movls r0, #0\r
\r
- cmp r12, #0
+ cmp r12, #0\r
\r
// Optimized memcopy - Write 32 Bytes\r
- stmia r10!, {r2-r9}
-
+ stmia r10!, {r2-r9}\r
+\r
// while (length != 0)\r
bne memcopy_default_loop\r
-
+ \r
memcopy_end\r
- mov r0, r11
+ mov r0, r11\r
ldmfd sp!, {r4-r11, pc}\r
\r
END\r