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

Removed bug calling start/end during sequence submit phase

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-22 16:27:09 +01:00
parent ce00048d8c
commit 906533ae9d
5 changed files with 11 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

3
src/OpMult.cpp
View file

@ -109,6 +109,7 @@ OpMult<tX, tY, tZ>::record()
{
SPDLOG_DEBUG("Kompute OpMult record called");
// Barrier to ensure the data is finished writing to buffer memory
this->mTensorLHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eHostWrite,
vk::AccessFlagBits::eShaderRead,
@ -122,6 +123,7 @@ OpMult<tX, tY, tZ>::record()
this->mAlgorithm->recordDispatch(this->mX, this->mY, this->mZ);
// Barrier to ensure the shader code is executed before buffer read
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
@ -140,6 +142,7 @@ OpMult<tX, tY, tZ>::record()
this->mTensorOutputStaging->recordCopyFrom(this->mTensorLHS);
// Buffer to ensure wait until data is copied to staging buffer
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,