--- /dev/null
+;------------------------------------------------------------------------------\r
+;\r
+; Copyright (c) 2006, Intel Corporation\r
+; All rights reserved. This program and the accompanying materials\r
+; are licensed and made available under the terms and conditions of the BSD License\r
+; which accompanies this distribution. The full text of the license may be found at\r
+; http://opensource.org/licenses/bsd-license.php\r
+;\r
+; THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
+; WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
+;\r
+; Module Name:\r
+;\r
+; SetMem.asm\r
+;\r
+; Abstract:\r
+;\r
+; SetMem function\r
+;\r
+; Notes:\r
+;\r
+;------------------------------------------------------------------------------\r
+\r
+ .code\r
+\r
+;------------------------------------------------------------------------------\r
+; VOID *\r
+; InternalMemSetMem (\r
+; IN VOID *Buffer,\r
+; IN UINTN Count,\r
+; IN UINT8 Value\r
+; )\r
+;------------------------------------------------------------------------------\r
+InternalMemSetMem PROC USES rdi\r
+ mov rdi, rcx ; rdi <- Buffer\r
+ mov al, r8b ; al <- Value\r
+ mov r9, rdi ; r9 <- Buffer as return value\r
+ xor rcx, rcx\r
+ sub rcx, rdi\r
+ and rcx, 15 ; rcx + rdi aligns on 16-byte boundary\r
+ jz @F\r
+ cmp rcx, rdx\r
+ cmova rcx, rdx\r
+ sub rdx, rcx\r
+ rep stosb\r
+@@:\r
+ mov rcx, rdx\r
+ and rdx, 15\r
+ shr rcx, 4\r
+ jz @SetBytes\r
+ mov ah, al ; ax <- Value repeats twice\r
+ movdqa [rsp + 10h], xmm0 ; save xmm0\r
+ movd xmm0, eax ; xmm0[0..16] <- Value repeats twice\r
+ pshuflw xmm0, xmm0, 0 ; xmm0[0..63] <- Value repeats 8 times\r
+ movlhps xmm0, xmm0 ; xmm0 <- Value repeats 16 times\r
+@@:\r
+ movntdq [rdi], xmm0 ; rdi should be 16-byte aligned\r
+ add rdi, 16\r
+ loop @B\r
+ mfence\r
+ movdqa xmm0, [rsp + 10h] ; restore xmm0\r
+@SetBytes:\r
+ mov ecx, edx ; high 32 bits of rcx are always zero\r
+ rep stosb\r
+ mov rax, r9 ; rax <- Return value\r
+ ret\r
+InternalMemSetMem ENDP\r
+\r
+ END\r