From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wm0-x22b.google.com (mail-wm0-x22b.google.com [IPv6:2a00:1450:400c:c09::22b]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by ml01.01.org (Postfix) with ESMTPS id 871EA1A1E02 for ; Tue, 6 Sep 2016 07:23:45 -0700 (PDT) Received: by mail-wm0-x22b.google.com with SMTP id w12so95375976wmf.0 for ; Tue, 06 Sep 2016 07:23:45 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=fCp4fN/ZgIldG17pkMhW2tRr7fn8r1AE4KOghxdC4cI=; b=HGbGf0ekhc4gzG26lK7HeofwpUmlKzKT3Lq4Mq+TzNWwhW0LcpOfTUOgxtrbMeUXgA fw2uxceT+tHrItxZAQUhZH3M1SnHVye/9B4A38kG+KfhbKQKyBRX/wOc/xqSy15FnnrD A2ui1+Q3JV1DRJazpKsN+phzI1wsOoKveq7eU= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20130820; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=fCp4fN/ZgIldG17pkMhW2tRr7fn8r1AE4KOghxdC4cI=; b=cP+CTl/bvcnqCyvp5TFQBB9AZpZ1zX6HPacKO0/ORF6Nsn8d8HfMF1Ve2i54ADHXaK 4Y2LxkGatLgj6EuyEJG+tIf1TpWojUXWo7Og0qsyvw7d9fDMXCgBvJtqk/JyKr0SEqyG CoLFAA1iQ5C8SM2Vjj4aNLVR+9jmVIP/P648Pqg5Z8ziybul6Gvv3dScRgFWHqhP9eA+ WApKQ/Z8qj201jI10EJwCjfBAwbSr0mdC2cYFIG4T5UjRRYK0xq3DyUinNB8p1ALq3RY SHi8/Jj6wAJU0FHjE8oNih52Ndimb7U5WNGMFXF1wAf4ZbjepRjxt6O9wI7Mj5XouXSU fvBQ== X-Gm-Message-State: AE9vXwMAGeHGPhMRVIDhhQmFuEQ3VpbApfNlQ/lYca7j8qE+GY/VDyO5r8V54ZLiNuwJlrb/ X-Received: by 10.28.125.80 with SMTP id y77mr21520191wmc.25.1473171824101; Tue, 06 Sep 2016 07:23:44 -0700 (PDT) Received: from localhost.localdomain ([197.130.133.164]) by smtp.gmail.com with ESMTPSA id f8sm19922140wjh.45.2016.09.06.07.23.42 (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Tue, 06 Sep 2016 07:23:43 -0700 (PDT) From: Ard Biesheuvel To: edk2-devel@lists.01.org, leif.lindholm@linaro.org, liming.gao@intel.com Cc: michael.d.kinney@intel.com, Ard Biesheuvel Date: Tue, 6 Sep 2016 15:23:31 +0100 Message-Id: <1473171813-24595-2-git-send-email-ard.biesheuvel@linaro.org> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1473171813-24595-1-git-send-email-ard.biesheuvel@linaro.org> References: <1473171813-24595-1-git-send-email-ard.biesheuvel@linaro.org> Subject: [PATCH v2 1/3] MdePkg/BaseMemoryLib: widen aligned accesses to 32 or 64 bits X-BeenThere: edk2-devel@lists.01.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: EDK II Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 06 Sep 2016 14:23:46 -0000 Since the default BaseMemoryLib should be callable from any context, including ones where unaligned accesses are not allowed, it implements InternalCopyMem() and InternalSetMem() using byte accesses only. However, especially in a context where the MMU is off, such narrow accesses may be disproportionately costly, and so if the size and alignment of the access allow it, use 32-bit or even 64-bit loads and stores (the latter may be beneficial even on a 32-bit architectures like ARM, which has load pair/store pair instructions) Contributed-under: TianoCore Contribution Agreement 1.0 Signed-off-by: Ard Biesheuvel --- MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf | 2 +- MdePkg/Library/BaseMemoryLib/CopyMem.c | 112 ++++++++++++++++++-- MdePkg/Library/BaseMemoryLib/SetMem.c | 40 ++++++- 3 files changed, 140 insertions(+), 14 deletions(-) diff --git a/MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf b/MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf index 6d906e93faf3..358eeed4f449 100644 --- a/MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf +++ b/MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf @@ -26,7 +26,7 @@ [Defines] # -# VALID_ARCHITECTURES = IA32 X64 IPF EBC +# VALID_ARCHITECTURES = IA32 X64 IPF EBC ARM AARCH64 # [Sources] diff --git a/MdePkg/Library/BaseMemoryLib/CopyMem.c b/MdePkg/Library/BaseMemoryLib/CopyMem.c index 37f03660df5f..6f4fd900df5d 100644 --- a/MdePkg/Library/BaseMemoryLib/CopyMem.c +++ b/MdePkg/Library/BaseMemoryLib/CopyMem.c @@ -4,6 +4,9 @@ particular platform easily if an optimized version is desired. Copyright (c) 2006 - 2010, Intel Corporation. All rights reserved.
+ Copyright (c) 2012 - 2013, ARM Ltd. All rights reserved.
+ Copyright (c) 2016, Linaro Ltd. All rights reserved.
+ This program and the accompanying materials are licensed and made available under the terms and conditions of the BSD License which accompanies this distribution. The full text of the license may be found at @@ -44,18 +47,107 @@ InternalMemCopyMem ( // volatile UINT8 *Destination8; CONST UINT8 *Source8; + volatile UINT32 *Destination32; + CONST UINT32 *Source32; + volatile UINT64 *Destination64; + CONST UINT64 *Source64; + UINTN Alignment; + + if ((((UINTN)DestinationBuffer & 0x7) == 0) && (((UINTN)SourceBuffer & 0x7) == 0) && (Length >= 8)) { + if (SourceBuffer > DestinationBuffer) { + Destination64 = (UINT64*)DestinationBuffer; + Source64 = (CONST UINT64*)SourceBuffer; + while (Length >= 8) { + *(Destination64++) = *(Source64++); + Length -= 8; + } + + // Finish if there are still some bytes to copy + Destination8 = (UINT8*)Destination64; + Source8 = (CONST UINT8*)Source64; + while (Length-- != 0) { + *(Destination8++) = *(Source8++); + } + } else if (SourceBuffer < DestinationBuffer) { + Destination64 = (UINT64*)((UINTN)DestinationBuffer + Length); + Source64 = (CONST UINT64*)((UINTN)SourceBuffer + Length); + + // Destination64 and Source64 were aligned on a 64-bit boundary + // but if length is not a multiple of 8 bytes then they won't be + // anymore. + + Alignment = Length & 0x7; + if (Alignment != 0) { + Destination8 = (UINT8*)Destination64; + Source8 = (CONST UINT8*)Source64; + + while (Alignment-- != 0) { + *(--Destination8) = *(--Source8); + --Length; + } + Destination64 = (UINT64*)Destination8; + Source64 = (CONST UINT64*)Source8; + } + + while (Length > 0) { + *(--Destination64) = *(--Source64); + Length -= 8; + } + } + } else if ((((UINTN)DestinationBuffer & 0x3) == 0) && (((UINTN)SourceBuffer & 0x3) == 0) && (Length >= 4)) { + if (SourceBuffer > DestinationBuffer) { + Destination32 = (UINT32*)DestinationBuffer; + Source32 = (CONST UINT32*)SourceBuffer; + while (Length >= 4) { + *(Destination32++) = *(Source32++); + Length -= 4; + } + + // Finish if there are still some bytes to copy + Destination8 = (UINT8*)Destination32; + Source8 = (CONST UINT8*)Source32; + while (Length-- != 0) { + *(Destination8++) = *(Source8++); + } + } else if (SourceBuffer < DestinationBuffer) { + Destination32 = (UINT32*)((UINTN)DestinationBuffer + Length); + Source32 = (CONST UINT32*)((UINTN)SourceBuffer + Length); + + // Destination32 and Source32 were aligned on a 32-bit boundary + // but if length is not a multiple of 4 bytes then they won't be + // anymore. + + Alignment = Length & 0x3; + if (Alignment != 0) { + Destination8 = (UINT8*)Destination32; + Source8 = (CONST UINT8*)Source32; + + while (Alignment-- != 0) { + *(--Destination8) = *(--Source8); + --Length; + } + Destination32 = (UINT32*)Destination8; + Source32 = (CONST UINT32*)Source8; + } - if (SourceBuffer > DestinationBuffer) { - Destination8 = (UINT8*)DestinationBuffer; - Source8 = (CONST UINT8*)SourceBuffer; - while (Length-- != 0) { - *(Destination8++) = *(Source8++); + while (Length > 0) { + *(--Destination32) = *(--Source32); + Length -= 4; + } } - } else if (SourceBuffer < DestinationBuffer) { - Destination8 = (UINT8*)DestinationBuffer + Length; - Source8 = (CONST UINT8*)SourceBuffer + Length; - while (Length-- != 0) { - *(--Destination8) = *(--Source8); + } else { + if (SourceBuffer > DestinationBuffer) { + Destination8 = (UINT8*)DestinationBuffer; + Source8 = (CONST UINT8*)SourceBuffer; + while (Length-- != 0) { + *(Destination8++) = *(Source8++); + } + } else if (SourceBuffer < DestinationBuffer) { + Destination8 = (UINT8*)DestinationBuffer + Length; + Source8 = (CONST UINT8*)SourceBuffer + Length; + while (Length-- != 0) { + *(--Destination8) = *(--Source8); + } } } return DestinationBuffer; diff --git a/MdePkg/Library/BaseMemoryLib/SetMem.c b/MdePkg/Library/BaseMemoryLib/SetMem.c index 5e74085c56f0..932c61e959f3 100644 --- a/MdePkg/Library/BaseMemoryLib/SetMem.c +++ b/MdePkg/Library/BaseMemoryLib/SetMem.c @@ -5,6 +5,9 @@ is desired. Copyright (c) 2006 - 2010, Intel Corporation. All rights reserved.
+ Copyright (c) 2012 - 2013, ARM Ltd. All rights reserved.
+ Copyright (c) 2016, Linaro Ltd. All rights reserved.
+ This program and the accompanying materials are licensed and made available under the terms and conditions of the BSD License which accompanies this distribution. The full text of the license may be found at @@ -43,11 +46,42 @@ InternalMemSetMem ( // volatile to prevent the optimizer from replacing this function with // the intrinsic memset() // - volatile UINT8 *Pointer; + volatile UINT8 *Pointer8; + volatile UINT32 *Pointer32; + volatile UINT64 *Pointer64; + UINT32 Value32; + UINT64 Value64; + + if ((((UINTN)Buffer & 0x7) == 0) && (Length >= 8)) { + // Generate the 64bit value + Value32 = (Value << 24) | (Value << 16) | (Value << 8) | Value; + Value64 = (((UINT64)Value32) << 32) | Value32; + + Pointer64 = (UINT64*)Buffer; + while (Length >= 8) { + *(Pointer64++) = Value64; + Length -= 8; + } - Pointer = (UINT8*)Buffer; + // Finish with bytes if needed + Pointer8 = (UINT8*)Pointer64; + } else if ((((UINTN)Buffer & 0x3) == 0) && (Length >= 4)) { + // Generate the 32bit value + Value32 = (Value << 24) | (Value << 16) | (Value << 8) | Value; + + Pointer32 = (UINT32*)Buffer; + while (Length >= 4) { + *(Pointer32++) = Value32; + Length -= 4; + } + + // Finish with bytes if needed + Pointer8 = (UINT8*)Pointer32; + } else { + Pointer8 = (UINT8*)Buffer; + } while (Length-- > 0) { - *(Pointer++) = Value; + *(Pointer8++) = Value; } return Buffer; } -- 2.7.4