public static void GetIndexOfFirstNonAsciiByte_Boundaries()
{
// The purpose of this test is to make sure we're hitting all of the vectorized
// and draining logic correctly both in the SSE2 and in the non-SSE2 enlightened
// code paths. We shouldn't be reading beyond the boundaries we were given.
// The 5 * Vector test should make sure that we're exercising all possible
// code paths across both implementations.
using (BoundedMemory <byte> mem = BoundedMemory.Allocate <byte>(5 * Vector <byte> .Count))
{
Span <byte> bytes = mem.Span;
// First, try it with all-ASCII buffers.
for (int i = 0; i < bytes.Length; i++)
{
bytes[i] &= 0x7F; // make sure each byte (of the pre-populated random data) is ASCII
}
for (int i = bytes.Length; i >= 0; i--)
{
Assert.Equal(i, CallGetIndexOfFirstNonAsciiByte(bytes.Slice(0, i)));
}
// Then, try it with non-ASCII bytes.
for (int i = bytes.Length; i >= 1; i--)
{
bytes[i - 1] = 0x80; // set non-ASCII
Assert.Equal(i - 1, CallGetIndexOfFirstNonAsciiByte(bytes.Slice(0, i)));
}
}
}
public static void GetIndexOfFirstNonAsciiByte_Vector128InnerLoop()
{
// The purpose of this test is to make sure we're identifying the correct
// vector (of the two that we're reading simultaneously) when performing
// the final ASCII drain at the end of the method once we've broken out
// of the inner loop.
using (BoundedMemory <byte> mem = BoundedMemory.Allocate <byte>(1024))
{
Span <byte> bytes = mem.Span;
for (int i = 0; i < bytes.Length; i++)
{
bytes[i] &= 0x7F; // make sure each byte (of the pre-populated random data) is ASCII
}
// Two vectors have offsets 0 .. 31. We'll go backward to avoid having to
// re-clear the vector every time.
for (int i = 2 * SizeOfVector128 - 1; i >= 0; i--)
{
bytes[100 + i * 13] = 0x80; // 13 is relatively prime to 32, so it ensures all possible positions are hit
Assert.Equal(100 + i * 13, CallGetIndexOfFirstNonAsciiByte(bytes));
}
}
}
public static void GetIndexOfFirstNonAsciiChar_Vector128InnerLoop()
{
// The purpose of this test is to make sure we're identifying the correct
// vector (of the two that we're reading simultaneously) when performing
// the final ASCII drain at the end of the method once we've broken out
// of the inner loop.
//
// Use U+0123 instead of U+0080 for this test because if our implementation
// uses pminuw / pmovmskb incorrectly, U+0123 will incorrectly show up as ASCII,
// causing our test to produce a false negative.
using (BoundedMemory <char> mem = BoundedMemory.Allocate <char>(1024))
{
Span <char> chars = mem.Span;
for (int i = 0; i < chars.Length; i++)
{
chars[i] &= '\u007F'; // make sure each char (of the pre-populated random data) is ASCII
}
// Two vectors have offsets 0 .. 31. We'll go backward to avoid having to
// re-clear the vector every time.
for (int i = 2 * SizeOfVector128 - 1; i >= 0; i--)
{
chars[100 + i * 13] = '\u0123'; // 13 is relatively prime to 32, so it ensures all possible positions are hit
Assert.Equal(100 + i * 13, CallGetIndexOfFirstNonAsciiChar(chars));
}
}
}
private static void ToBytes_Test_Core(ReadOnlySpan <char> utf16Input, int destinationSize, bool replaceInvalidSequences, bool isFinalChunk, OperationStatus expectedOperationStatus, int expectedNumCharsRead, ReadOnlySpan <byte> expectedUtf8Transcoding)
{
// Arrange
using (BoundedMemory <char> boundedSource = BoundedMemory.AllocateFromExistingData(utf16Input))
using (BoundedMemory <byte> boundedDestination = BoundedMemory.Allocate <byte>(destinationSize))
{
boundedSource.MakeReadonly();
// Act
OperationStatus actualOperationStatus = Utf8.FromUtf16(boundedSource.Span, boundedDestination.Span, out int actualNumCharsRead, out int actualNumBytesWritten, replaceInvalidSequences, isFinalChunk);
// Assert
Assert.Equal(expectedOperationStatus, actualOperationStatus);
Assert.Equal(expectedNumCharsRead, actualNumCharsRead);
Assert.Equal(expectedUtf8Transcoding.Length, actualNumBytesWritten);
Assert.Equal(expectedUtf8Transcoding.ToArray(), boundedDestination.Span.Slice(0, actualNumBytesWritten).ToArray());
}
}
public static void GetIndexOfFirstNonAsciiChar_Boundaries()
{
// The purpose of this test is to make sure we're hitting all of the vectorized
// and draining logic correctly both in the SSE2 and in the non-SSE2 enlightened
// code paths. We shouldn't be reading beyond the boundaries we were given.
//
// The 5 * Vector test should make sure that we're exercising all possible
// code paths across both implementations. The sizeof(char) is because we're
// specifying element count, but underlying implementation reintepret casts to bytes.
//
// Use U+0123 instead of U+0080 for this test because if our implementation
// uses pminuw / pmovmskb incorrectly, U+0123 will incorrectly show up as ASCII,
// causing our test to produce a false negative.
using (BoundedMemory <char> mem = BoundedMemory.Allocate <char>(5 * Vector <byte> .Count / sizeof(char)))
{
Span <char> chars = mem.Span;
for (int i = 0; i < chars.Length; i++)
{
chars[i] &= '\u007F'; // make sure each char (of the pre-populated random data) is ASCII
}
for (int i = chars.Length; i >= 0; i--)
{
Assert.Equal(i, CallGetIndexOfFirstNonAsciiChar(chars.Slice(0, i)));
}
// Then, try it with non-ASCII bytes.
for (int i = chars.Length; i >= 1; i--)
{
chars[i - 1] = '\u0123'; // set non-ASCII
Assert.Equal(i - 1, CallGetIndexOfFirstNonAsciiChar(chars.Slice(0, i)));
}
}
}