Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

try to port ASCIIUtility.WidenAsciiToUtf16 to x-plat intrinsics #73055

Merged
merged 16 commits into from
Sep 9, 2022
+79 −187
Merged

try to port ASCIIUtility.WidenAsciiToUtf16 to x-plat intrinsics #73055

Changes from 1 commit
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
80ebdc4
port WidenFourAsciiBytesToUtf16AndWriteToBuffer
adamsitnik Jul 29, 2022
c37ef61
port WidenAsciiToUtf16*
adamsitnik Jul 29, 2022
53cdf1d
enforce the inlining of Vector128.WidenVector128<byte>) to remove per…
adamsitnik Jul 29, 2022
a3b10cb
don't use Vector128.Widen as it less optimal on ARM64
adamsitnik Jul 29, 2022
865721c
remove dead code: the condition was impossible to meet (jump to NonAs…
adamsitnik Jul 29, 2022
349aca7
simplify the code and get perf boost on both x64 and ARM64
adamsitnik Jul 29, 2022
764d038
add Vector256 code path
adamsitnik Jul 29, 2022
f0b227f
Merge remote-tracking branch 'upstream/main' into WidenAsciiToUtf16
adamsitnik Aug 19, 2022
5d7ef96
don't use Vector128.ExtractMostSignificantBits as it's expensive on ARM
adamsitnik Aug 19, 2022
a84c712
these methods are now getting inlined
adamsitnik Aug 19, 2022
e44f2a7
fix the size check
adamsitnik Aug 19, 2022
20a2f23
prefer arm-specific instructions to avoid regressions for smaller inp…
adamsitnik Aug 19, 2022
bc0b8be
comment out the problematic assert
adamsitnik Aug 19, 2022
7ee2c7b
Merge branch 'dotnet:main' into WidenAsciiToUtf16
adamsitnik Aug 22, 2022
5325199
it's beneficial to do it for fewer elements as well
adamsitnik Aug 22, 2022
5501301
inline the helper methods to simply get better perf and avoid R2R issue
adamsitnik Aug 23, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
prefer arm-specific instructions to avoid regressions for smaller inp… 
…uts on ARM
@adamsitnik
adamsitnik committed Aug 19, 2022
commit 20a2f2313c34d2c48052336167895f66916e9cb5
10 changes: 8 additions & 2 deletions src/libraries/System.Private.CoreLib/src/System/Text/ASCIIUtility.cs
View file
Original file line number Diff line number Diff line change
@@ -1380,7 +1380,7 @@ private static bool VectorContainsNonAsciiChar(Vector128<byte> asciiVector)
{
return !Sse41.TestZ(asciiVector, Vector128.Create((byte)0x80));
}
if (AdvSimd.Arm64.IsSupported)
else if (AdvSimd.Arm64.IsSupported)
{
Vector128<byte> maxBytes = AdvSimd.Arm64.MaxPairwise(asciiVector, asciiVector);
return (maxBytes.AsUInt64().ToScalar() & 0x8080808080808080) != 0;
@@ -1793,7 +1793,13 @@ internal static void WidenFourAsciiBytesToUtf16AndWriteToBuffer(ref char outputB
{
Debug.Assert(AllBytesInUInt32AreAscii(value));

if (Vector128.IsHardwareAccelerated)
if (AdvSimd.Arm64.IsSupported)
{
Vector128<byte> vecNarrow = AdvSimd.DuplicateToVector128(value).AsByte();
Vector128<ulong> vecWide = AdvSimd.Arm64.ZipLow(vecNarrow, Vector128<byte>.Zero).AsUInt64();
Unsafe.WriteUnaligned<ulong>(ref Unsafe.As<char, byte>(ref outputBuffer), vecWide.ToScalar());
}
else if (Vector128.IsHardwareAccelerated)
{
Vector128<byte> vecNarrow = Vector128.CreateScalar(value).AsByte();
Vector128<ulong> vecWide = Vector128.WidenLower(vecNarrow).AsUInt64();