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

Added optensorcopy operation

Browse source
This commit is contained in:
Alejandro Saucedo 2020-09-06 11:07:32 +01:00
parent 93c1ba126e
commit 236c349aa0
13 changed files with 238 additions and 45 deletions
Download patch file Download diff file Expand all files Collapse all files

10
src/OpCreateTensor.cpp
View file

@ -48,13 +48,13 @@ OpCreateTensor::init()
}
if (tensor->tensorType() == Tensor::TensorTypes::eDevice) {
tensor->init(
this->mPhysicalDevice, this->mDevice, this->mCommandBuffer);
this->mPhysicalDevice, this->mDevice);
std::shared_ptr<Tensor> stagingTensor = std::make_shared<Tensor>(
tensor->data(), Tensor::TensorTypes::eStaging);
stagingTensor->init(
this->mPhysicalDevice, this->mDevice, this->mCommandBuffer);
this->mPhysicalDevice, this->mDevice);
stagingTensor->mapDataIntoHostMemory();
@ -63,7 +63,7 @@ OpCreateTensor::init()
} else {
tensor->init(
this->mPhysicalDevice, this->mDevice, this->mCommandBuffer);
this->mPhysicalDevice, this->mDevice);
// We push a nullptr when no staging tensor is needed to match
// index number in array to have one to one mapping with tensors
@ -79,7 +79,9 @@ OpCreateTensor::record()
for (size_t i = 0; i < this->mTensors.size(); i++) {
if (this->mTensors[i]->tensorType() == Tensor::TensorTypes::eDevice) {
this->mTensors[i]->recordCopyFrom(this->mStagingTensors[i], false);
this->mTensors[i]->recordCopyFrom(this->mCommandBuffer, this->mStagingTensors[i], false);
} else if (this->mTensors[i]->tensorType() == Tensor::TensorTypes::eStaging) {
this->mTensors[i]->mapDataIntoHostMemory();
}
}
}