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

Working descriptorset creation cycle

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-22 11:21:34 +01:00
parent 9d97ca07a7
commit 03688bc5b2
2 changed files with 9 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

11
src/OpMult.cpp
View file

@ -36,16 +36,22 @@ OpMult::init(std::vector<std::shared_ptr<Tensor>> tensors)
throw std::runtime_error(
"Kompute OpMult called with less than 1 tensor");
} else if (tensors.size() > 3) {
spdlog::warn("Kompute OpMult called with more than 2 tensor");
spdlog::warn("Kompute OpMult called with more than 3 tensors");
}
this->mTensorLHS = tensors[0];
this->mTensorRHS = tensors[1];
this->mTensorOutput = tensors[2];
this->mTensorOutputStaging= std::make_shared<Tensor>(
if (!(this->mTensorLHS->isInit() && this->mTensorRHS->isInit() && this->mTensorOutput->isInit())) {
throw std::runtime_error("Kompute OpMult all tensor parameters must be initialised. LHS: " + std::to_string(this->mTensorLHS->isInit()) + " RHS: " + std::to_string(this->mTensorRHS->isInit()) + " Output: " + std::to_string(this->mTensorOutput->isInit()));
}
this->mTensorOutputStaging = std::make_shared<Tensor>(
this->mTensorOutput->data(), Tensor::TensorTypes::eStaging);
this->mTensorOutputStaging->init(this->mPhysicalDevice, this->mDevice, this->mCommandBuffer, this->mTensorOutput->data());
this->mAlgorithm->init(
"shaders/glsl/computeheadless.comp.spv", tensors);
}
@ -65,6 +71,7 @@ void OpMult::postSubmit()
SPDLOG_DEBUG("Kompute OpCreateTensor postSubmit called");
this->mTensorOutputStaging->copyDataFromHostBuffer();
this->mTensorOutput->setData(this->mTensorOutputStaging->data());
}