Title: [133100] trunk/Source/WTF
- Revision
- 133100
- Author
- [email protected]
- Date
- 2012-10-31 17:31:14 -0700 (Wed, 31 Oct 2012)
Log Message
Add an optimized version of copyLCharsFromUCharSource for ARM
https://bugs.webkit.org/show_bug.cgi?id=94886
Patch by Benjamin Poulain <[email protected]> on 2012-10-31
Reviewed by Gavin Barraclough.
Michael Saboff added a SIMD version of copyLCharsFromUCharSource() in r125846.
This patch a similar optimization for ARMv7 by using the interleaved load/store available
in the NEON extension.
The performance gains:
-10000 characters: ~3.5 times faster.
-20 characters (2 vectors): ~55% faster.
-15 characters (1 vector): ~21% faster.
-3 characters (no vector, pure overhead): ~10% slower.
* wtf/text/ASCIIFastPath.h:
(WTF::copyLCharsFromUCharSource):
Modified Paths
Diff
Modified: trunk/Source/WTF/ChangeLog (133099 => 133100)
--- trunk/Source/WTF/ChangeLog 2012-11-01 00:09:54 UTC (rev 133099)
+++ trunk/Source/WTF/ChangeLog 2012-11-01 00:31:14 UTC (rev 133100)
@@ -1,3 +1,24 @@
+2012-10-31 Benjamin Poulain <[email protected]>
+
+ Add an optimized version of copyLCharsFromUCharSource for ARM
+ https://bugs.webkit.org/show_bug.cgi?id=94886
+
+ Reviewed by Gavin Barraclough.
+
+ Michael Saboff added a SIMD version of copyLCharsFromUCharSource() in r125846.
+
+ This patch a similar optimization for ARMv7 by using the interleaved load/store available
+ in the NEON extension.
+
+ The performance gains:
+ -10000 characters: ~3.5 times faster.
+ -20 characters (2 vectors): ~55% faster.
+ -15 characters (1 vector): ~21% faster.
+ -3 characters (no vector, pure overhead): ~10% slower.
+
+ * wtf/text/ASCIIFastPath.h:
+ (WTF::copyLCharsFromUCharSource):
+
2012-10-31 Christophe Dumez <[email protected]>
[EFL][WK2][AC] Use smart pointers for Evas_GL types
Modified: trunk/Source/WTF/wtf/text/ASCIIFastPath.h (133099 => 133100)
--- trunk/Source/WTF/wtf/text/ASCIIFastPath.h 2012-11-01 00:09:54 UTC (rev 133099)
+++ trunk/Source/WTF/wtf/text/ASCIIFastPath.h 2012-11-01 00:31:14 UTC (rev 133100)
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2011 Apple Inc. All rights reserved.
+ * Copyright (C) 2011, 2012 Apple Inc. All rights reserved.
* Copyright (C) 2011 Nokia Corporation and/or its subsidiary(-ies).
*
* This library is free software; you can redistribute it and/or
@@ -131,6 +131,31 @@
ASSERT(!(source[i] & 0xff00));
destination[i] = static_cast<LChar>(source[i]);
}
+#elif COMPILER(GCC) && CPU(ARM_NEON) && !(PLATFORM(BIG_ENDIAN) || PLATFORM(MIDDLE_ENDIAN)) && defined(NDEBUG)
+ const LChar* const end = destination + length;
+ const uintptr_t memoryAccessSize = 8;
+
+ if (length >= (2 * memoryAccessSize) - 1) {
+ // Prefix: align dst on 64 bits.
+ const uintptr_t memoryAccessMask = memoryAccessSize - 1;
+ do {
+ *destination++ = static_cast<LChar>(*source++);
+ } while (!isAlignedTo<memoryAccessMask>(destination));
+
+ // Vector interleaved unpack, we only store the lower 8 bits.
+ const uintptr_t lengthLeft = end - destination;
+ const LChar* const simdEnd = end - (lengthLeft % memoryAccessSize);
+ do {
+ asm("vld2.8 { d0-d1 }, [%[SOURCE]] !\n\t"
+ "vst1.8 { d0 }, [%[DESTINATION],:64] !\n\t"
+ : [SOURCE]"+r" (source), [DESTINATION]"+r" (destination)
+ :
+ : "memory", "d0", "d1");
+ } while (destination != simdEnd);
+ }
+
+ while (destination != end)
+ *destination++ = static_cast<LChar>(*source++);
#else
for (size_t i = 0; i < length; ++i) {
ASSERT(!(source[i] & 0xff00));
_______________________________________________
webkit-changes mailing list
[email protected]
http://lists.webkit.org/mailman/listinfo/webkit-changes
