VARCem/munt
1
0
Fork 0
mirror of https://github.com/VARCem/munt.git synced 2026-09-25 08:15:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

Improved reliability of QRingBuffer wrt. memory visibility

Browse Source 
and added descriptions of used memory barriers
This commit is contained in:
sergm
2020-09-05 20:04:32 +03:00
parent a05476e5db
commit 9863c0347c
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
mt32emu_qt/src/QRingBuffer.cpp
View File

@@ -85,10 +85,14 @@ void QRingBuffer::advanceWritePointer(quint32 bytesWritten) {
if (bufferSize <= myWritePosition) {
myWritePosition -= bufferSize;
}
// Release barrier ensures that data is fully written prior to updating the index.
storeRelease(writePosition, myWritePosition);
}
void *QRingBuffer::readPointer(quint32 &bytesReady) const {
// Aquire barrier ensures that data is never read ahead of the indices.
// In practice, this should never be the case due to the implicit dependency,
// but shouldn't hurt anyway.
quint32 myReadPosition = loadRelaxed(readPosition);
quint32 myWritePosition = loadAcquire(writePosition);
bytesReady = (myWritePosition < myReadPosition ? bufferSize : myWritePosition) - myReadPosition;
@@ -96,10 +100,13 @@ void *QRingBuffer::readPointer(quint32 &bytesReady) const {
}
void QRingBuffer::advanceReadPointer(quint32 bytesRead) {
quint32 myReadPosition = loadRelaxed(readPosition);
// Full barrier here prevents late data reads from possible overlapping with early writes.
// Using a combination of loadAcquire and storeRelease here to simulate it,
// since Qt API lacks operations for simple atomic load or store with the ordered semantic.
quint32 myReadPosition = loadAcquire(readPosition);
myReadPosition += bytesRead;
if (bufferSize <= myReadPosition) {
myReadPosition -= bufferSize;
}
storeRelaxed(readPosition, myReadPosition);
readPosition.storeRelease(myReadPosition);
}