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

7
src/Algorithm.cpp
View file

@ -49,7 +49,7 @@ Algorithm::createParameters(std::vector<std::shared_ptr<Tensor>>& tensorParams)
// TODO: Explore design for having multiple descriptor pool sizes
std::vector<vk::DescriptorPoolSize> descriptorPoolSizes = {
vk::DescriptorPoolSize(vk::DescriptorType::eStorageBuffer,
1 // Descriptor count
static_cast<uint32_t>(tensorParams.size()) // Descriptor count
)
};
@ -99,6 +99,11 @@ Algorithm::createParameters(std::vector<std::shared_ptr<Tensor>>& tensorParams)
this->mDevice->allocateDescriptorSets(&descriptorSetAllocateInfo,
this->mDescriptorSet.get());
std::vector<vk::DescriptorBufferInfo> descriptorBufferInfos;
for (size_t i = 0; i < tensorParams.size(); i++) {
descriptorBufferInfos.push_back(tensorParams[i]->constructDescriptorBufferInfo());
}
// TODO: Explore design exposing the destination array element
std::vector<vk::WriteDescriptorSet> computeWriteDescriptorSets;
for (size_t i = 0; i < tensorParams.size(); i++) {