Skip to content

Commit 744fa67

Browse files
Nick-NuonNick-Nuon
committed
different tail benches
1 parent 713d29b commit 744fa67

File tree

1 file changed

+87
-15
lines changed

1 file changed

+87
-15
lines changed

‎src/UTF8_validation.cs

Lines changed: 87 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -154,27 +154,64 @@ public static unsafe class Utf8Utility
154154

155155
// Helpers.CheckForGCCollections("After AVX2 procession");
156156

157-
if (processedLength < inputLength)
158-
{
159-
// Unfortunalely, this approach with stackalloc might be expensive.
160-
// TODO: replace it by a simple scalar routine. You need to handle
161-
// prev_incomplete but it should be doable.
157+
158+
// | Method | FileName | Mean | Error | StdDev | Allocated |
159+
// |---------------------------- |----------------------- |----------:|----------:|----------:|----------:|
160+
// | SIMDUtf8ValidationRealData | data/arabic.utf8.txt | 33.062 us | 0.5046 us | 0.4720 us | 56 B |
161+
// | SIMDUtf8ValidationErrorData | data/arabic.utf8.txt | 35.609 us | 0.3369 us | 0.3152 us | 56 B |
162+
// | SIMDUtf8ValidationRealData | data/chinese.utf8.txt | 11.603 us | 0.2232 us | 0.2293 us | 56 B |
163+
// | SIMDUtf8ValidationErrorData | data/chinese.utf8.txt | 12.317 us | 0.1826 us | 0.1708 us | 56 B |
164+
// | SIMDUtf8ValidationRealData | data/english.utf8.txt | 13.726 us | 0.2471 us | 0.2311 us | 56 B |
165+
// | SIMDUtf8ValidationErrorData | data/english.utf8.txt | 13.392 us | 0.0520 us | 0.0487 us | 56 B |
166+
// | SIMDUtf8ValidationRealData | data/french.utf8.txt | 24.345 us | 0.2012 us | 0.1882 us | 56 B |
167+
// | SIMDUtf8ValidationErrorData | data/french.utf8.txt | 23.778 us | 0.1892 us | 0.1769 us | 56 B |
168+
// | SIMDUtf8ValidationRealData | data/german.utf8.txt | 9.323 us | 0.0155 us | 0.0130 us | 56 B |
169+
// | SIMDUtf8ValidationErrorData | data/german.utf8.txt | 8.336 us | 0.0502 us | 0.0470 us | 56 B |
170+
// | SIMDUtf8ValidationRealData | data/japanese.utf8.txt | 10.728 us | 0.1370 us | 0.1282 us | 56 B |
171+
// | SIMDUtf8ValidationErrorData | data/japanese.utf8.txt | 10.837 us | 0.1389 us | 0.1300 us | 56 B |
172+
// | SIMDUtf8ValidationRealData | data/turkish.utf8.txt | 11.086 us | 0.1190 us | 0.1113 us | 56 B |
173+
// | SIMDUtf8ValidationErrorData | data/turkish.utf8.txt | 10.017 us | 0.0615 us | 0.0514 us | 56 B |
162174

163-
Span<byte> remainingBytes = stackalloc byte[32];
164-
for (int i = 0; i < inputLength - processedLength; i++)
165-
{
166-
remainingBytes[i] = pInputBuffer[processedLength + i];
167-
}
168175

169-
Vector256<byte> remainingBlock = Vector256.Create(remainingBytes.ToArray());
170-
Utf8Validation.utf8_checker.CheckNextInput(remainingBlock, ref prev_input_block, ref prev_incomplete, ref error);
171-
processedLength += inputLength - processedLength;
176+
// if (processedLength < inputLength)
177+
// {
178+
// // Unfortunalely, this approach with stackalloc might be expensive.
179+
// // TODO: replace it by a simple scalar routine. You need to handle
180+
// // prev_incomplete but it should be doable.
172181

173-
}
174182

175-
// CheckForGCCollections("After processed remaining bytes");
183+
// Span<byte> remainingBytes = stackalloc byte[32];
184+
// for (int i = 0; i < inputLength - processedLength; i++)
185+
// {
186+
// remainingBytes[i] = pInputBuffer[processedLength + i];
187+
// }
176188

189+
// Vector256<byte> remainingBlock = Vector256.Create(remainingBytes.ToArray());
190+
// Utf8Validation.utf8_checker.CheckNextInput(remainingBlock, ref prev_input_block, ref prev_incomplete, ref error);
191+
// processedLength += inputLength - processedLength;
192+
193+
// }
194+
195+
// CheckForGCCollections("After processed remaining bytes");
177196

197+
// | Method | FileName | Mean | Error | StdDev | Median | Allocated |
198+
// |---------------------------- |----------------------- |----------:|----------:|----------:|----------:|----------:|
199+
// | SIMDUtf8ValidationRealData | data/arabic.utf8.txt | 36.968 us | 0.4637 us | 0.4111 us | 37.018 us | 56 B |
200+
// | SIMDUtf8ValidationErrorData | data/arabic.utf8.txt | 31.851 us | 0.2252 us | 0.2107 us | 31.865 us | 56 B |
201+
// | SIMDUtf8ValidationRealData | data/chinese.utf8.txt | 10.541 us | 0.0870 us | 0.0814 us | 10.490 us | 56 B |
202+
// | SIMDUtf8ValidationErrorData | data/chinese.utf8.txt | 12.404 us | 0.2447 us | 0.2913 us | 12.355 us | 56 B |
203+
// | SIMDUtf8ValidationRealData | data/english.utf8.txt | 14.297 us | 0.2786 us | 0.4655 us | 14.225 us | 56 B |
204+
// | SIMDUtf8ValidationErrorData | data/english.utf8.txt | 14.207 us | 0.0272 us | 0.0241 us | 14.211 us | 56 B |
205+
// | SIMDUtf8ValidationRealData | data/french.utf8.txt | 23.993 us | 0.2287 us | 0.2140 us | 23.879 us | 56 B |
206+
// | SIMDUtf8ValidationErrorData | data/french.utf8.txt | 26.856 us | 0.5314 us | 1.4367 us | 27.005 us | 56 B |
207+
// | SIMDUtf8ValidationRealData | data/german.utf8.txt | 8.541 us | 0.1702 us | 0.2090 us | 8.456 us | 56 B |
208+
// | SIMDUtf8ValidationErrorData | data/german.utf8.txt | 8.728 us | 0.1618 us | 0.3938 us | 8.567 us | 56 B |
209+
// | SIMDUtf8ValidationRealData | data/japanese.utf8.txt | 11.108 us | 0.1767 us | 0.1653 us | 11.184 us | 56 B |
210+
// | SIMDUtf8ValidationErrorData | data/japanese.utf8.txt | 9.533 us | 0.1288 us | 0.1205 us | 9.490 us | 56 B |
211+
// | SIMDUtf8ValidationRealData | data/turkish.utf8.txt | 10.128 us | 0.0544 us | 0.0454 us | 10.126 us | 56 B |
212+
// | SIMDUtf8ValidationErrorData | data/turkish.utf8.txt | 10.078 us | 0.0499 us | 0.0467 us | 10.079 us | 56 B |
213+
214+
// scalar results:
178215
// if (processedLength < inputLength)
179216
// {
180217
// // Directly call the scalar function on the remaining part of the buffer
@@ -190,6 +227,41 @@ public static unsafe class Utf8Utility
190227
// }
191228

192229

230+
// | Method | FileName | Mean | Error | StdDev | Allocated |
231+
// |---------------------------- |----------------------- |----------:|----------:|----------:|----------:|
232+
// | SIMDUtf8ValidationRealData | data/arabic.utf8.txt | 20.136 us | 0.3869 us | 0.5031 us | - |
233+
// | SIMDUtf8ValidationErrorData | data/arabic.utf8.txt | 19.576 us | 0.2366 us | 0.2098 us | - |
234+
// | SIMDUtf8ValidationRealData | data/chinese.utf8.txt | 6.207 us | 0.0479 us | 0.0400 us | - |
235+
// | SIMDUtf8ValidationErrorData | data/chinese.utf8.txt | 6.169 us | 0.0541 us | 0.0506 us | - |
236+
// | SIMDUtf8ValidationRealData | data/english.utf8.txt | 9.212 us | 0.0121 us | 0.0107 us | - |
237+
// | SIMDUtf8ValidationErrorData | data/english.utf8.txt | 9.373 us | 0.0250 us | 0.0209 us | - |
238+
// | SIMDUtf8ValidationRealData | data/french.utf8.txt | 13.726 us | 0.2609 us | 0.2900 us | - |
239+
// | SIMDUtf8ValidationErrorData | data/french.utf8.txt | 13.948 us | 0.2122 us | 0.1985 us | - |
240+
// | SIMDUtf8ValidationRealData | data/german.utf8.txt | 4.916 us | 0.0176 us | 0.0147 us | - |
241+
// | SIMDUtf8ValidationErrorData | data/german.utf8.txt | 4.897 us | 0.0525 us | 0.0491 us | - |
242+
// | SIMDUtf8ValidationRealData | data/japanese.utf8.txt | 5.526 us | 0.0463 us | 0.0411 us | - |
243+
// | SIMDUtf8ValidationErrorData | data/japanese.utf8.txt | 5.538 us | 0.0405 us | 0.0379 us | - |
244+
// | SIMDUtf8ValidationRealData | data/turkish.utf8.txt | 5.838 us | 0.0363 us | 0.0340 us | - |
245+
// | SIMDUtf8ValidationErrorData | data/turkish.utf8.txt | 5.813 us | 0.0440 us | 0.0412 us | - |
246+
247+
248+
if (processedLength < inputLength)
249+
{
250+
251+
Span<byte> remainingBytes = stackalloc byte[32];
252+
for (int i = 0; i < inputLength - processedLength; i++)
253+
{
254+
remainingBytes[i] = pInputBuffer[processedLength + i];
255+
}
256+
257+
ReadOnlySpan<Byte> remainingBytesReadOnly = remainingBytes;
258+
Vector256<byte> remainingBlock = Vector256.Create(remainingBytesReadOnly);
259+
Utf8Validation.utf8_checker.CheckNextInput(remainingBlock, ref prev_input_block, ref prev_incomplete, ref error);
260+
processedLength += inputLength - processedLength;
261+
262+
}
263+
264+
193265
Utf8Validation.utf8_checker.CheckEof(ref error, prev_incomplete);
194266
if (Utf8Validation.utf8_checker.Errors(error))
195267
{

0 commit comments

Comments
 (0)