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

Fully working end to end flow

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-22 16:34:39 +01:00
parent 906533ae9d
commit 6f0203b863
1 changed files with 31 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

62
src/OpMult.cpp
View file

@ -110,37 +110,37 @@ 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,
vk::PipelineStageFlagBits::eHost,
vk::PipelineStageFlagBits::eComputeShader);
this->mTensorRHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eHostWrite,
vk::AccessFlagBits::eShaderRead,
vk::PipelineStageFlagBits::eHost,
vk::PipelineStageFlagBits::eComputeShader);
//this->mTensorLHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eHostWrite,
// vk::AccessFlagBits::eShaderRead,
// vk::PipelineStageFlagBits::eHost,
// vk::PipelineStageFlagBits::eComputeShader);
//this->mTensorRHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eHostWrite,
// vk::AccessFlagBits::eShaderRead,
// vk::PipelineStageFlagBits::eHost,
// vk::PipelineStageFlagBits::eComputeShader);
this->mAlgorithm->recordDispatch(this->mX, this->mY, this->mZ);
// Barrier to ensure the shader code is executed before buffer read
//this->mTensorLHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eShaderWrite,
// vk::AccessFlagBits::eTransferRead,
// vk::PipelineStageFlagBits::eComputeShader,
// vk::PipelineStageFlagBits::eTransfer);
//this->mTensorRHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eShaderWrite,
// vk::AccessFlagBits::eTransferRead,
// vk::PipelineStageFlagBits::eComputeShader,
// vk::PipelineStageFlagBits::eTransfer);
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensorLHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensorRHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensorOutputStaging->recordCopyFrom(this->mTensorLHS);
this->mTensorOutputStaging->recordCopyFrom(this->mTensorOutput);
// Buffer to ensure wait until data is copied to staging buffer
this->mTensorOutput->recordBufferMemoryBarrier(
@ -148,16 +148,16 @@ OpMult<tX, tY, tZ>::record()
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
this->mTensorLHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
this->mTensorRHS->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
//this->mTensorLHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eTransferWrite,
// vk::AccessFlagBits::eHostRead,
// vk::PipelineStageFlagBits::eTransfer,
// vk::PipelineStageFlagBits::eHost);
//this->mTensorRHS->recordBufferMemoryBarrier(
// vk::AccessFlagBits::eTransferWrite,
// vk::AccessFlagBits::eHostRead,
// vk::PipelineStageFlagBits::eTransfer,
// vk::PipelineStageFlagBits::eHost);
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>