thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

Added buffer memory barriers

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-22 13:01:04 +01:00
parent 5bb9046b49
commit 25d5df21ce
4 changed files with 40 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

12
src/OpMult.cpp
View file

@ -85,7 +85,19 @@ OpMult::record()
this->mAlgorithm->recordDispatch(1, 1, 1);
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensorOutputStaging->recordCopyFrom(this->mTensorOutput);
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
}
void