diff options
Diffstat (limited to 'roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64')
12 files changed, 697 insertions, 0 deletions
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CompareMem.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CompareMem.nasm new file mode 100644 index 000000000..5514ab663 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CompareMem.nasm @@ -0,0 +1,52 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006 - 2008, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; CompareMem.Asm
+;
+; Abstract:
+;
+; CompareMem function
+;
+; Notes:
+;
+; The following BaseMemoryLib instances contain the same copy of this file:
+;
+; BaseMemoryLibRepStr
+; BaseMemoryLibMmx
+; BaseMemoryLibSse2
+; BaseMemoryLibOptDxe
+; BaseMemoryLibOptPei
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+; IN CONST VOID *DestinationBuffer,
+; IN CONST VOID *SourceBuffer,
+; IN UINTN Length
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemCompareMem)
+ASM_PFX(InternalMemCompareMem):
+ push rsi
+ push rdi
+ mov rsi, rcx
+ mov rdi, rdx
+ mov rcx, r8
+ repe cmpsb
+ movzx rax, byte [rsi - 1]
+ movzx rdx, byte [rdi - 1]
+ sub rax, rdx
+ pop rdi
+ pop rsi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CopyMem.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CopyMem.nasm new file mode 100644 index 000000000..33545712b --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/CopyMem.nasm @@ -0,0 +1,77 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; CopyMem.nasm
+;
+; Abstract:
+;
+; CopyMem function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; EFIAPI
+; InternalMemCopyMem (
+; IN VOID *Destination,
+; IN VOID *Source,
+; IN UINTN Count
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemCopyMem)
+ASM_PFX(InternalMemCopyMem):
+ push rsi
+ push rdi
+ mov rsi, rdx ; rsi <- Source
+ mov rdi, rcx ; rdi <- Destination
+ lea r9, [rsi + r8 - 1] ; r9 <- Last byte of Source
+ cmp rsi, rdi
+ mov rax, rdi ; rax <- Destination as return value
+ jae .0 ; Copy forward if Source > Destination
+ cmp r9, rdi ; Overlapped?
+ jae @CopyBackward ; Copy backward if overlapped
+.0:
+ xor rcx, rcx
+ sub rcx, rdi ; rcx <- -rdi
+ and rcx, 15 ; rcx + rsi should be 16 bytes aligned
+ jz .1 ; skip if rcx == 0
+ cmp rcx, r8
+ cmova rcx, r8
+ sub r8, rcx
+ rep movsb
+.1:
+ mov rcx, r8
+ and r8, 15
+ shr rcx, 4 ; rcx <- # of DQwords to copy
+ jz @CopyBytes
+ movdqa [rsp + 0x18], xmm0 ; save xmm0 on stack
+.2:
+ movdqu xmm0, [rsi] ; rsi may not be 16-byte aligned
+ movntdq [rdi], xmm0 ; rdi should be 16-byte aligned
+ add rsi, 16
+ add rdi, 16
+ loop .2
+ mfence
+ movdqa xmm0, [rsp + 0x18] ; restore xmm0
+ jmp @CopyBytes ; copy remaining bytes
+@CopyBackward:
+ mov rsi, r9 ; rsi <- Last byte of Source
+ lea rdi, [rdi + r8 - 1] ; rdi <- Last byte of Destination
+ std
+@CopyBytes:
+ mov rcx, r8
+ rep movsb
+ cld
+ pop rdi
+ pop rsi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/IsZeroBuffer.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/IsZeroBuffer.nasm new file mode 100644 index 000000000..42e5f3f84 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/IsZeroBuffer.nasm @@ -0,0 +1,70 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2016, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; IsZeroBuffer.nasm
+;
+; Abstract:
+;
+; IsZeroBuffer function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; BOOLEAN
+; EFIAPI
+; InternalMemIsZeroBuffer (
+; IN CONST VOID *Buffer,
+; IN UINTN Length
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemIsZeroBuffer)
+ASM_PFX(InternalMemIsZeroBuffer):
+ push rdi
+ mov rdi, rcx ; rdi <- Buffer
+ xor rcx, rcx ; rcx <- 0
+ sub rcx, rdi
+ and rcx, 15 ; rcx + rdi aligns on 16-byte boundary
+ jz @Is16BytesZero
+ cmp rcx, rdx ; Length already in rdx
+ cmova rcx, rdx ; bytes before the 16-byte boundary
+ sub rdx, rcx
+ xor rax, rax ; rax <- 0, also set ZF
+ repe scasb
+ jnz @ReturnFalse ; ZF=0 means non-zero element found
+@Is16BytesZero:
+ mov rcx, rdx
+ and rdx, 15
+ shr rcx, 4
+ jz @IsBytesZero
+.0:
+ pxor xmm0, xmm0 ; xmm0 <- 0
+ pcmpeqb xmm0, [rdi] ; check zero for 16 bytes
+ pmovmskb eax, xmm0 ; eax <- compare results
+ ; nasm doesn't support 64-bit destination
+ ; for pmovmskb
+ cmp eax, 0xffff
+ jnz @ReturnFalse
+ add rdi, 16
+ loop .0
+@IsBytesZero:
+ mov rcx, rdx
+ xor rax, rax ; rax <- 0, also set ZF
+ repe scasb
+ jnz @ReturnFalse ; ZF=0 means non-zero element found
+ pop rdi
+ mov rax, 1 ; return TRUE
+ ret
+@ReturnFalse:
+ pop rdi
+ xor rax, rax
+ ret ; return FALSE
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem16.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem16.nasm new file mode 100644 index 000000000..eb385a4a2 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem16.nasm @@ -0,0 +1,49 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006 - 2008, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; ScanMem16.Asm
+;
+; Abstract:
+;
+; ScanMem16 function
+;
+; Notes:
+;
+; The following BaseMemoryLib instances contain the same copy of this file:
+;
+; BaseMemoryLibRepStr
+; BaseMemoryLibMmx
+; BaseMemoryLibSse2
+; BaseMemoryLibOptDxe
+; BaseMemoryLibOptPei
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+; IN CONST VOID *Buffer,
+; IN UINTN Length,
+; IN UINT16 Value
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemScanMem16)
+ASM_PFX(InternalMemScanMem16):
+ push rdi
+ mov rdi, rcx
+ mov rax, r8
+ mov rcx, rdx
+ repne scasw
+ lea rax, [rdi - 2]
+ cmovnz rax, rcx
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem32.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem32.nasm new file mode 100644 index 000000000..6fccdf2d1 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem32.nasm @@ -0,0 +1,49 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006 - 2008, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; ScanMem32.Asm
+;
+; Abstract:
+;
+; ScanMem32 function
+;
+; Notes:
+;
+; The following BaseMemoryLib instances contain the same copy of this file:
+;
+; BaseMemoryLibRepStr
+; BaseMemoryLibMmx
+; BaseMemoryLibSse2
+; BaseMemoryLibOptDxe
+; BaseMemoryLibOptPei
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+; IN CONST VOID *Buffer,
+; IN UINTN Length,
+; IN UINT32 Value
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemScanMem32)
+ASM_PFX(InternalMemScanMem32):
+ push rdi
+ mov rdi, rcx
+ mov rax, r8
+ mov rcx, rdx
+ repne scasd
+ lea rax, [rdi - 4]
+ cmovnz rax, rcx
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem64.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem64.nasm new file mode 100644 index 000000000..6659aef44 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem64.nasm @@ -0,0 +1,49 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006 - 2008, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; ScanMem64.Asm
+;
+; Abstract:
+;
+; ScanMem64 function
+;
+; Notes:
+;
+; The following BaseMemoryLib instances contain the same copy of this file:
+;
+; BaseMemoryLibRepStr
+; BaseMemoryLibMmx
+; BaseMemoryLibSse2
+; BaseMemoryLibOptDxe
+; BaseMemoryLibOptPei
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+; IN CONST VOID *Buffer,
+; IN UINTN Length,
+; IN UINT64 Value
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemScanMem64)
+ASM_PFX(InternalMemScanMem64):
+ push rdi
+ mov rdi, rcx
+ mov rax, r8
+ mov rcx, rdx
+ repne scasq
+ lea rax, [rdi - 8]
+ cmovnz rax, rcx
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem8.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem8.nasm new file mode 100644 index 000000000..16fd931c0 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ScanMem8.nasm @@ -0,0 +1,49 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006 - 2008, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; ScanMem8.Asm
+;
+; Abstract:
+;
+; ScanMem8 function
+;
+; Notes:
+;
+; The following BaseMemoryLib instances contain the same copy of this file:
+;
+; BaseMemoryLibRepStr
+; BaseMemoryLibMmx
+; BaseMemoryLibSse2
+; BaseMemoryLibOptDxe
+; BaseMemoryLibOptPei
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+; IN CONST VOID *Buffer,
+; IN UINTN Length,
+; IN UINT8 Value
+; );
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemScanMem8)
+ASM_PFX(InternalMemScanMem8):
+ push rdi
+ mov rdi, rcx
+ mov rcx, rdx
+ mov rax, r8
+ repne scasb
+ lea rax, [rdi - 1]
+ cmovnz rax, rcx ; set rax to 0 if not found
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem.nasm new file mode 100644 index 000000000..5bd1c2262 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem.nasm @@ -0,0 +1,65 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; SetMem.nasm
+;
+; Abstract:
+;
+; SetMem function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; InternalMemSetMem (
+; IN VOID *Buffer,
+; IN UINTN Count,
+; IN UINT8 Value
+; )
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemSetMem)
+ASM_PFX(InternalMemSetMem):
+ push rdi
+ mov rdi, rcx ; rdi <- Buffer
+ mov al, r8b ; al <- Value
+ mov r9, rdi ; r9 <- Buffer as return value
+ xor rcx, rcx
+ sub rcx, rdi
+ and rcx, 15 ; rcx + rdi aligns on 16-byte boundary
+ jz .0
+ cmp rcx, rdx
+ cmova rcx, rdx
+ sub rdx, rcx
+ rep stosb
+.0:
+ mov rcx, rdx
+ and rdx, 15
+ shr rcx, 4
+ jz @SetBytes
+ mov ah, al ; ax <- Value repeats twice
+ movdqa [rsp + 0x10], xmm0 ; save xmm0
+ movd xmm0, eax ; xmm0[0..16] <- Value repeats twice
+ pshuflw xmm0, xmm0, 0 ; xmm0[0..63] <- Value repeats 8 times
+ movlhps xmm0, xmm0 ; xmm0 <- Value repeats 16 times
+.1:
+ movntdq [rdi], xmm0 ; rdi should be 16-byte aligned
+ add rdi, 16
+ loop .1
+ mfence
+ movdqa xmm0, [rsp + 0x10] ; restore xmm0
+@SetBytes:
+ mov ecx, edx ; high 32 bits of rcx are always zero
+ rep stosb
+ mov rax, r9 ; rax <- Return value
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem16.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem16.nasm new file mode 100644 index 000000000..90d159820 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem16.nasm @@ -0,0 +1,63 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; SetMem16.nasm
+;
+; Abstract:
+;
+; SetMem16 function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; InternalMemSetMem16 (
+; IN VOID *Buffer,
+; IN UINTN Count,
+; IN UINT16 Value
+; )
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemSetMem16)
+ASM_PFX(InternalMemSetMem16):
+ push rdi
+ mov rdi, rcx
+ mov r9, rdi
+ xor rcx, rcx
+ sub rcx, rdi
+ and rcx, 15
+ mov rax, r8
+ jz .0
+ shr rcx, 1
+ cmp rcx, rdx
+ cmova rcx, rdx
+ sub rdx, rcx
+ rep stosw
+.0:
+ mov rcx, rdx
+ and edx, 7
+ shr rcx, 3
+ jz @SetWords
+ movd xmm0, eax
+ pshuflw xmm0, xmm0, 0
+ movlhps xmm0, xmm0
+.1:
+ movntdq [rdi], xmm0
+ add rdi, 16
+ loop .1
+ mfence
+@SetWords:
+ mov ecx, edx
+ rep stosw
+ mov rax, r9
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem32.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem32.nasm new file mode 100644 index 000000000..928e08688 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem32.nasm @@ -0,0 +1,62 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; SetMem32.nasm
+;
+; Abstract:
+;
+; SetMem32 function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; InternalMemSetMem32 (
+; IN VOID *Buffer,
+; IN UINTN Count,
+; IN UINT8 Value
+; )
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemSetMem32)
+ASM_PFX(InternalMemSetMem32):
+ push rdi
+ mov rdi, rcx
+ mov r9, rdi
+ xor rcx, rcx
+ sub rcx, rdi
+ and rcx, 15
+ mov rax, r8
+ jz .0
+ shr rcx, 2
+ cmp rcx, rdx
+ cmova rcx, rdx
+ sub rdx, rcx
+ rep stosd
+.0:
+ mov rcx, rdx
+ and edx, 3
+ shr rcx, 2
+ jz @SetDwords
+ movd xmm0, eax
+ pshufd xmm0, xmm0, 0
+.1:
+ movntdq [rdi], xmm0
+ add rdi, 16
+ loop .1
+ mfence
+@SetDwords:
+ mov ecx, edx
+ rep stosd
+ mov rax, r9
+ pop rdi
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem64.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem64.nasm new file mode 100644 index 000000000..d77181054 --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/SetMem64.nasm @@ -0,0 +1,53 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; SetMem64.nasm
+;
+; Abstract:
+;
+; SetMem64 function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; InternalMemSetMem64 (
+; IN VOID *Buffer,
+; IN UINTN Count,
+; IN UINT64 Value
+; )
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemSetMem64)
+ASM_PFX(InternalMemSetMem64):
+ mov rax, rcx ; rax <- Buffer
+ xchg rcx, rdx ; rcx <- Count & rdx <- Buffer
+ test dl, 8
+ movq xmm0, r8
+ jz .0
+ mov [rdx], r8
+ add rdx, 8
+ dec rcx
+.0:
+ shr rcx, 1
+ jz @SetQwords
+ movlhps xmm0, xmm0
+.1:
+ movntdq [rdx], xmm0
+ lea rdx, [rdx + 16]
+ loop .1
+ mfence
+@SetQwords:
+ jnc .2
+ mov [rdx], r8
+.2:
+ ret
+
diff --git a/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ZeroMem.nasm b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ZeroMem.nasm new file mode 100644 index 000000000..5ddcae9ca --- /dev/null +++ b/roms/edk2/MdePkg/Library/BaseMemoryLibSse2/X64/ZeroMem.nasm @@ -0,0 +1,59 @@ +;------------------------------------------------------------------------------
+;
+; Copyright (c) 2006, Intel Corporation. All rights reserved.<BR>
+; SPDX-License-Identifier: BSD-2-Clause-Patent
+;
+; Module Name:
+;
+; ZeroMem.nasm
+;
+; Abstract:
+;
+; ZeroMem function
+;
+; Notes:
+;
+;------------------------------------------------------------------------------
+
+ DEFAULT REL
+ SECTION .text
+
+;------------------------------------------------------------------------------
+; VOID *
+; InternalMemZeroMem (
+; IN VOID *Buffer,
+; IN UINTN Count
+; )
+;------------------------------------------------------------------------------
+global ASM_PFX(InternalMemZeroMem)
+ASM_PFX(InternalMemZeroMem):
+ push rdi
+ mov rdi, rcx
+ xor rcx, rcx
+ xor eax, eax
+ sub rcx, rdi
+ and rcx, 15
+ mov r8, rdi
+ jz .0
+ cmp rcx, rdx
+ cmova rcx, rdx
+ sub rdx, rcx
+ rep stosb
+.0:
+ mov rcx, rdx
+ and edx, 15
+ shr rcx, 4
+ jz @ZeroBytes
+ pxor xmm0, xmm0
+.1:
+ movntdq [rdi], xmm0 ; rdi should be 16-byte aligned
+ add rdi, 16
+ loop .1
+ mfence
+@ZeroBytes:
+ mov ecx, edx
+ rep stosb
+ mov rax, r8
+ pop rdi
+ ret
+
|