AuroraMiddleware/qt5base-lts
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

rhi: Properly squeeze in trimOpLists

Browse Source 
...and improve the comments since it is not directly obvious why
things are done the way they are.

Pick-to: 6.5
Change-Id: I6aa5f09b1b2f7f3fc18f74a4af2977d24278aae6
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
This commit is contained in:
Laszlo Agocs 2023-04-24 13:21:54 +02:00
parent 3d2eb6c191
commit 7a79302e4e
1 changed files with 28 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
src/gui/rhi/qrhi.cpp
View File

@ -6287,6 +6287,16 @@ void QRhiResourceUpdateBatch::generateMips(QRhiTexture *tex)
*/
QRhiResourceUpdateBatch *QRhi::nextResourceUpdateBatch()
{
// By default we prefer spreading out the utilization of the 64 batches as
// much as possible, meaning we won't pick the first one even if it's free,
// but prefer picking one after the last picked one. Relevant due to how
// QVLA and QRhiBufferData allocations behind the bufferOps are reused; in
// typical Qt Quick scenes this leads to a form of (eventually) seeding all
// the 64 resource batches with buffer operation data allocations which are
// then reused in subsequent frames. This comes at the expense of using
// more memory, but has proven good results when (CPU) profiling typical
// Quick/Quick3D apps.
auto nextFreeBatch = [this]() -> QRhiResourceUpdateBatch * {
auto isFree = [this](int i) -> QRhiResourceUpdateBatch * {
const quint64 mask = 1ULL << quint64(i);
@ -6337,7 +6347,15 @@ void QRhiResourceUpdateBatchPrivate::free()
rhi->resUpdPoolMap &= ~mask;
poolIndex = -1;
// textureOps is cleared, to not keep the potentially large image pixel
// data alive, but it is expected that the container keeps the list alloc
// at least. Only trimOpList() goes for the more aggressive route with squeeze.
textureOps.clear();
// bufferOps is not touched, to allow reusing allocations (incl. in the
// elements' QRhiBufferData) as much as possible when this batch is used
// again in the future, which is important for performance, in particular
// with Qt Quick.
}
void QRhiResourceUpdateBatchPrivate::merge(QRhiResourceUpdateBatchPrivate *other)
@ -6367,11 +6385,21 @@ void QRhiResourceUpdateBatchPrivate::trimOpLists()
{
Q_ASSERT(poolIndex == -1); // must not be in use
// Unlike free(), this is expected to aggressively deallocate all memory
// used by both the buffer and texture operation lists. (i.e. using
// squeeze() to only keep the stack prealloc of the QVLAs)
//
// This (e.g. just the destruction of bufferOps elements) may have a
// non-negligible performance impact e.g. with Qt Quick with scenes where
// there are lots of buffer operations per frame.
activeBufferOpCount = 0;
bufferOps.clear();
bufferOps.squeeze();
activeTextureOpCount = 0;
textureOps.clear();
textureOps.squeeze();
}
/*!