wehub-resource-sync
39 lines
1.4 KiB
C++
39 lines
1.4 KiB
C++
#ifndef VulkanGaussianRender_hpp
|
|
#define VulkanGaussianRender_hpp
|
|
#include "VulkanBasicExecution.hpp"
|
|
namespace MNN {
|
|
class VulkanRadixSort {
|
|
public:
|
|
struct Content;
|
|
VulkanRadixSort(Backend* bn, int needBit);
|
|
~VulkanRadixSort();
|
|
ErrorCode onExcute(std::pair<VulkanBuffer*, VkDeviceSize> srcIndex, std::pair<VulkanBuffer*, VkDeviceSize> dstIndex, const VulkanCommandPool::Buffer *cmdBuffer, int numberPoint, std::shared_ptr<VulkanBuffer> sortNumber);
|
|
void autoTune(std::pair<VulkanBuffer*, VkDeviceSize> srcIndex, std::pair<VulkanBuffer*, VkDeviceSize> dstIndex, int numberPoint, std::shared_ptr<VulkanBuffer> sortNumber);
|
|
private:
|
|
Content* mContent;
|
|
Backend* mBackend;
|
|
int mNeedBits;
|
|
int mPerSortBit = 4;
|
|
int mLocalSize = 256;
|
|
int mGroupSize = 16;
|
|
};
|
|
|
|
class VulkanRasterSort : public VulkanBasicExecution {
|
|
public:
|
|
struct Content;
|
|
VulkanRasterSort(Backend* bn);
|
|
virtual ~ VulkanRasterSort();
|
|
virtual ErrorCode onEncode(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs,
|
|
const VulkanCommandPool::Buffer *cmdBuffer) override;
|
|
void autoTune(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs);
|
|
private:
|
|
Content* mContent;
|
|
std::shared_ptr<VulkanRadixSort> mRadixSort;
|
|
int mLocalSize = 1024;
|
|
int mGroupSize = 1024;
|
|
};
|
|
|
|
};
|
|
|
|
#endif
|