测试D3D压缩纹理进显存的速度_d3d 解码降低显存

硬件环境：NVIDIA GeForce 1050 ti 

视频：手上有一个hap-q的视频，大小为：5760*4320。通过CPU解码出来以后，数据大小为：24883200（字节），相当于RGBA裸数据的1/4.

1.通过D3DXLoadSurfaceFromMemory函数加载这个数据需要38ms.

2.通过cuda将同样大小数据送入显存只需要：7ms.

测试结果：

源代码如下：

#include "pch.h"
#include <iostream>
#include <windows.h>
#include <d3d9.h>
#include <d3dx9.h>
#include <cuda_runtime.h>
 
#pragma comment(lib,"d3d9.lib")
#pragma comment(lib,"d3dx9.lib")
#pragma comment(lib,"Winmm.lib")
 
 
int main()
{
	IDirect3D9Ex *pd3d;
	IDirect3DDevice9Ex *pd3dDevice;
	Direct3DCreate9Ex(D3D_SDK_VERSION, &pd3d);
	D3DPRESENT_PARAMETERS		m_d3dPrtPar;
	ZeroMemory(&m_d3dPrtPar, sizeof(m_d3dPrtPar));
	m_d3dPrtPar.Windowed = TRUE;
	m_d3dPrtPar.SwapEffect = D3DSWAPEFFECT_DISCARD;
	m_d3dPrtPar.BackBufferFormat = D3DFMT_UNKNOWN;
	m_d3dPrtPar.BackBufferWidth = 0;
	m_d3dPrtPar.BackBufferHeight = 0;
	m_d3dPrtPar.BackBufferCount = 1;
	m_d3dPrtPar.hDeviceWindow = GetDesktopWindow();
	m_d3dPrtPar.MultiSampleType = D3DMULTISAMPLE_NONE;
	m_d3dPrtPar.PresentationInterval = D3DPRESENT_INTERVAL_DEFAULT;
	m_d3dPrtPar.EnableAutoDepthStencil = false;//如果要启用深度缓冲区，则必须创建一个大的深度缓冲区，否则目标渲染区域大于深度缓冲区则渲染不出来。
	m_d3dPrtPar.AutoDepthStencilFormat = D3DFMT_D16;
	DWORD vp = D3DCREATE_HARDWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED;
	pd3d->CreateDeviceEx(
		0,
		D3DDEVTYPE_HAL,
		GetDesktopWindow(),
		vp,
		&m_d3dPrtPar,
		NULL,
		&pd3dDevice);
 
	int width =5760, height = 4320;
 
	LPDIRECT3DTEXTURE9 pTexture = NULL;
	D3DFORMAT format = D3DFMT_DXT5;
	pd3dDevice->CreateTexture(width, height,
		1, 0, format, D3DPOOL_DEFAULT, &pTexture, NULL);
 
	int dataLen = width * height * 4;
	dataLen = 24883200;
	byte *pData = new byte[dataLen];
	ZeroMemory(pData, dataLen);
	int srcPitch = width*4;
 
	RECT rc = { 0 };
	rc.right = width;
	rc.bottom = height;
	
	
	LPDIRECT3DSURFACE9 pSurface;
	pTexture->GetSurfaceLevel(0, &pSurface);
	int timeBegin = timeGetTime();
	D3DXLoadSurfaceFromMemory(pSurface, NULL, NULL, pData, format, srcPitch, NULL, &rc, D3DX_DEFAULT, 0);
	int timeEnd = timeGetTime();
	pSurface->Release();
	printf("压缩纹理内存加载耗时：%d\n", timeEnd - timeBegin);
 
	LPDIRECT3DSURFACE9 pOffSurface = NULL;
	pd3dDevice->CreateOffscreenPlainSurface(width, height, D3DFMT_A8R8G8B8, D3DPOOL_DEFAULT, &pOffSurface, NULL);
	D3DLOCKED_RECT rect;
	timeBegin = timeGetTime();
	pOffSurface->LockRect(&rect, NULL, 0);
	//ZeroMemory(rect.pBits, width*height);
	pOffSurface->UnlockRect();
	timeEnd = timeGetTime();
	printf("Surface显存到内存耗时：%d\n", timeEnd - timeBegin);
 
	LPDIRECT3DTEXTURE9 pTextureRGB = NULL;
	LPDIRECT3DTEXTURE9 pTextureRGB2 = NULL;
	format = D3DFMT_A8R8G8B8;
	pd3dDevice->CreateTexture(width, height,
		1, D3DUSAGE_RENDERTARGET, format, D3DPOOL_DEFAULT, &pTextureRGB, NULL);
	pd3dDevice->CreateTexture(width, height,
		1, 0, format, D3DPOOL_SYSTEMMEM, &pTextureRGB2, NULL);
	LPDIRECT3DSURFACE9 pTexSuf1, pTexSuf2;
	pTextureRGB->GetSurfaceLevel(0, &pTexSuf1);
	pTextureRGB2->GetSurfaceLevel(0, &pTexSuf2);
	HRESULT hr = pd3dDevice->GetRenderTargetData(pTexSuf1, pTexSuf2);
	
	timeBegin = timeGetTime();
	//pTextureRGB->LockRect(0,&rect, NULL, D3DLOCK_DISCARD);
	//((char*)rect.pBits)[0] = 1;
	//pTextureRGB->UnlockRect(0);
	
	pTexSuf2->LockRect(&rect, NULL, 0);
	pTexSuf2->UnlockRect();
	timeEnd = timeGetTime();
	printf("纹理显存到内存加载耗时：%d\n", timeEnd - timeBegin);
 
	byte *dev_c = NULL;
	cudaMalloc((void**)&dev_c, dataLen);
	timeBegin = timeGetTime();
	cudaMemcpy(dev_c, pData, dataLen, cudaMemcpyHostToDevice);
	timeEnd = timeGetTime();
	printf("cuda内存到显存加载耗时：%d\n", timeEnd - timeBegin);
 
	timeBegin = timeGetTime();
	cudaMemcpy( pData,dev_c, dataLen, cudaMemcpyDeviceToHost);
	timeEnd = timeGetTime();
	printf("cuda显存到内存加载耗时：%d\n", timeEnd - timeBegin);
 
	system("pause");
}

测试demo：demo