-;------------------------------------------------------------------------------ \r
+;------------------------------------------------------------------------------\r
;\r
; CopyMem() worker for ARM\r
;\r
/**\r
Copy Length bytes from Source to Destination. Overlap is OK.\r
\r
- This implementation \r
+ This implementation\r
\r
@param Destination Target of copy\r
@param Source Place to copy from\r
IN UINTN Length\r
)\r
**/\r
-\s\sEXPORT InternalMemCopyMem\r
-
-\s\sAREA AsmMemStuff, CODE, READONLY
-\r
-InternalMemCopyMem\r
-\s\sstmfd\s\ssp!, {r4-r11, lr}
-\s\stst\s\sr0, #3
-\s\smov\s\sr11, r0
-\s\smov\s\sr10, r0
-\s\smov\s\sip, r2
-\s\smov\s\slr, r1
-\s\smovne\s\sr0, #0
-\s\sbne\s\sL4
-\s\stst\s\sr1, #3
-\s\smovne\s\sr3, #0
-\s\smoveq\s\sr3, #1
-\s\scmp\s\sr2, #31
-\s\smovls\s\sr0, #0
-\s\sandhi\s\sr0, r3, #1
-L4
-\s\scmp\s\sr11, r1
-\s\sbcc\s\sL26
-\s\sbls\s\sL7
-\s\srsb\s\sr3, r1, r11
-\s\scmp\s\sip, r3
-\s\sbcc\s\sL26
-\s\scmp\s\sip, #0
-\s\sbeq\s\sL7
-\s\sadd\s\sr10, r11, ip
-\s\sadd\s\slr, ip, r1
-\s\sb\s\sL16
-L29
-\s\ssub\s\sip, ip, #8
-\s\scmp\s\sip, #7
-\s\sldrd\s\sr2, [lr, #-8]!
-\s\smovls\s\sr0, #0
-\s\scmp\s\sip, #0
-\s\sstrd\s\sr2, [r10, #-8]!
-\s\sbeq\s\sL7
-L16
-\s\scmp\s\sr0, #0
-\s\sbne\s\sL29
-\s\ssub\s\sr3, lr, #1
-\s\ssub\s\sip, ip, #1
-\s\sldrb\s\sr3, [r3, #0]\s\s
-\s\ssub\s\sr2, r10, #1
-\s\scmp\s\sip, #0
-\s\ssub\s\sr10, r10, #1
-\s\ssub\s\slr, lr, #1
-\s\sstrb\s\sr3, [r2, #0]
-\s\sbne\s\sL16
-\s\sb L7
-L11
-\s\sldrb\s\sr3, [lr], #1\s\s
-\s\ssub\s\sip, ip, #1
-\s\sstrb\s\sr3, [r10], #1
-L26
-\s\scmp\s\sip, #0
-\s\sbeq\s\sL7
-L30
-\s\scmp\s\sr0, #0
-\s\sbeq\s\sL11
-\s\ssub\s\sip, ip, #32
-\s\scmp\s\sip, #31
-\s\sldmia\s\slr!, {r2-r9}
-\s\smovls\s\sr0, #0
-\s\scmp\s\sip, #0
-\s\sstmia\s\sr10!, {r2-r9}
-\s\sbne\s\sL30
-L7
- mov\s\sr0, r11
-\s\sldmfd\s\ssp!, {r4-r11, pc}\r
-\s\s\r
+\r
+ INCLUDE AsmMacroExport.inc\r
+\r
+ RVCT_ASM_EXPORT InternalMemCopyMem\r
+ 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
+ // If (dest < source)\r
+ bcc memcopy_check_optim_default\r
+ // If (dest <= source). But with the previous condition -> If (dest == source)\r
+ bls memcopy_end\r
+\r
+ // If (source + length < dest)\r
+ rsb r3, r1, r11\r
+ cmp r12, r3\r
+ bcc memcopy_check_optim_default\r
+\r
+ // If (length == 0)\r
+ cmp r12, #0\r
+ beq memcopy_end\r
+\r
+ b memcopy_check_optim_overlap\r
+\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\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\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\r
+ sub r14, r14, #1\r
+ strb r3, [r2, #0]\r
+ bne memcopy_overlapped_non_optim\r
+ b memcopy_end\r
+\r
+// r10 = dest_end, r14 = source_end\r
+memcopy_overlapped\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
+ sub r10, r10, #32\r
+ stmia r10, {r2-r9}\r
+\r
+ // while (length != 0)\r
+ bne memcopy_overlapped\r
+ b memcopy_end\r
+\r
+memcopy_default_non_optim\r
+ // Byte copy\r
+ ldrb r3, [r14], #1\r
+ sub r12, r12, #1\r
+ strb r3, [r10], #1\r
+\r
+memcopy_default\r
+ cmp r12, #0\r
+ beq memcopy_end\r
+\r
+// r10 = dest, r14 = source\r
+memcopy_default_loop\r
+ cmp r0, #0\r
+ beq memcopy_default_non_optim\r
+\r
+ // Optimized memcopy - Read 32 Bytes\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 memcopy - Write 32 Bytes\r
+ stmia r10!, {r2-r9}\r
+\r
+ // while (length != 0)\r
+ bne memcopy_default_loop\r
+\r
+memcopy_end\r
+ mov r0, r11\r
+ ldmfd sp!, {r4-r11, pc}\r
+\r
END\r
- \r
+\r
projects / mirror_edk2.git / blobdiff