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

Added option for creating barrier on copyfrom tensor

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-29 15:49:43 +01:00
parent a2efc441db
commit 2298159586
4 changed files with 20 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

11
src/Tensor.cpp
View file

@ -95,7 +95,7 @@ Tensor::setData(const std::vector<uint32_t>& data)
}
void
Tensor::recordCopyFrom(std::shared_ptr<Tensor> copyFromTensor)
Tensor::recordCopyFrom(std::shared_ptr<Tensor> copyFromTensor, bool createBarrier)
{
SPDLOG_DEBUG("Kompute Tensor recordCopyFrom called");
@ -114,6 +114,15 @@ Tensor::recordCopyFrom(std::shared_ptr<Tensor> copyFromTensor)
// TODO: Ensure command buffer is in same device from buffer
this->mCommandBuffer->copyBuffer(
*copyFromTensor->mBuffer, *this->mBuffer, copyRegion);
if (createBarrier) {
// Buffer to ensure wait until data is copied to staging buffer
this->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
}
}
void