1 ;------------------------------------------------------------------------------ 2 ; 3 ; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR> 4 ; This program and the accompanying materials 5 ; are licensed and made available under the terms and conditions of the BSD License 6 ; which accompanies this distribution. The full text of the license may be found at 7 ; http://opensource.org/licenses/bsd-license.php. 8 ; 9 ; THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS, 10 ; WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED. 11 ; 12 ; Module Name: 13 ; 14 ; CopyMem.nasm 15 ; 16 ; Abstract: 17 ; 18 ; CopyMem function 19 ; 20 ; Notes: 21 ; 22 ;------------------------------------------------------------------------------ 23 24 DEFAULT REL 25 SECTION .text 26 27 ;------------------------------------------------------------------------------ 28 ; VOID * 29 ; EFIAPI 30 ; InternalMemCopyMem ( 31 ; OUT VOID *DestinationBuffer, 32 ; IN CONST VOID *SourceBuffer, 33 ; IN UINTN Length 34 ; ); 35 ;------------------------------------------------------------------------------ 36 global ASM_PFX(InternalMemCopyMem) 37 ASM_PFX(InternalMemCopyMem): 38 push rsi 39 push rdi 40 mov rsi, rdx ; rsi <- Source 41 mov rdi, rcx ; rdi <- Destination 42 lea r9, [rsi + r8 - 1] ; r9 <- End of Source 43 cmp rsi, rdi 44 mov rax, rdi ; rax <- Destination as return value 45 jae .0 46 cmp r9, rdi 47 jae @CopyBackward ; Copy backward if overlapped 48 .0: 49 mov rcx, r8 50 and r8, 7 51 shr rcx, 3 ; rcx <- # of Qwords to copy 52 jz @CopyBytes 53 DB 0x49, 0xf, 0x7e, 0xc2 ; movd r10, mm0 (Save mm0 in r10) 54 .1: 55 DB 0xf, 0x6f, 0x6 ; movd mm0, [rsi] 56 DB 0xf, 0xe7, 0x7 ; movntq [rdi], mm0 57 add rsi, 8 58 add rdi, 8 59 loop .1 60 mfence 61 DB 0x49, 0xf, 0x6e, 0xc2 ; movd mm0, r10 (Restore mm0) 62 jmp @CopyBytes 63 @CopyBackward: 64 mov rsi, r9 ; rsi <- End of Source 65 lea rdi, [rdi + r8 - 1] ; rdi <- End of Destination 66 std ; set direction flag 67 @CopyBytes: 68 mov rcx, r8 69 rep movsb ; Copy bytes backward 70 cld 71 pop rdi 72 pop rsi 73 ret 74 75
