OpenGL_and_CUDA.txt

PART 1, OPENGL, 

Pay special attention if you have your own OpenGL code.
The OpenGL-based implmentations of SiftGPU need valid OpenGL context to run properly.

1. If you use the same OpenGL context for SiftGPU and your own your visualization
	Make sure the OpenGL states are restored before calling SiftGPU. SiftGPU changes several 
	OpenGL internal states, including texture binding to GL_TEXTURE_RECTANGLE_ARB and current
	ViewPort. You might need to restore them for your own OpenGL part. To avoid this problem,
	you can create a seperate GL context, and activate different context for different part. 

	Note that GL_TEXTURE_RECTANGLE_ARB is always enabled in SiftGPU. When you have problem 
    displaying textures, you can try first glDisable(GL_TEXTURE_RECTANGLE_ARB) before painting, 
    but don't forget to call glEnable(GL_TEXTURE_RECTANGLE_ARB) after. (Thanks to Pilet)
	
2. How to create/setup an OpenGL context for SiftGPU

	If you choose to let SiftGPU to manage OpenGL context, you can simply do that by 
	SiftGPU::CreateContextGL and SiftMatchGPU::CreateContextGL.	When you mix your own OpenGL
	code with SiftGPU, you need to re-call CreateContextGL before calling SiftGPU functions,
	which will implicitly activate the internal OpenGL context. 
	
	
	If you choose to create openGL contexts yourself when mixing SiftGPU with other openGL 
	code, don't call SiftGPU::CreateContextGL or SiftMatchGPU::CreateContextGL; Instead you 
	should first activate your OpenGL context (WglMakeCurrent in win32), and set GL_FILL 
	for polygon mode, then call SiftGPU::VerifyContextGL or SiftMatchGPU::VerifyContextGL
    for initialization. You should also setup in the same way before calling SiftGPU functions. 
	

PART 2, CUDA
---------------------------------------------------------------------------------
1. How to enable CUDA

The CUDA implementation in the package is not compiled by default.

To enable it for visual stuio 2010, use msvc/SiftGPU_CUDA_Enabled.sln
To enable it for other OS, you need to change siftgpu_enable_cuda to 1 in the makefile


---------------------------------------------------------------------------------
2. Change CUDA build parameters.
For windows, you need to change the settings in the custom build command line of 
ProgramCU.cu. For example, add -use_fast_match for using fast match. 

For Other OS, you need to change the makefile. The top part of the makefile is 
the configuration section, which includes: 
	siftgpu_enable_cuda = 0             (Set 1 to enable CUDA-based SiftGPU)
	CUDA_INSTALL_PATH = /usr/local/cuda (Where to find CUDA)
	siftgpu_cuda_options = -arch sm_10  (Additional CUDA Compiling options)

	
------------------------------------------------------------------------------------
3. CUDA runtime parameters for SiftGPU::ParseParam
First, you need to specify "-cuda" to use CUDA-based SiftGPU. More parameters can 
be chagned at runtime in CUDA-based SiftGPU than in OpenGL-based version. Check out
the manual for details. 

NEW. You can choose GPU for CUDA computation by using "-cuda [device_index=0]"

One parameter for CUDA is "-di", which controls whether dynamic indexing is used
in descriptor generations. It is turned off by default. My experiments on 8800 
GTX show that unrolled loop of 8 if-assigns are faster than dynamic indexing, but
it might be different on other GPUs.


--------------------------------------------------------------------------------------
4. Speed of CUDA-based SiftGPU
If the size of the first octave (multiply the original size by 2 if upsample is used)
is less than or around 1024x768, CUDA version will be faster than OpenGL versions, 
otherwise the OpenGL versions are still faster. 

**************************************************************************************
This is observed on nVidia 8800 GTX, it might be different on other GPUs. Recent
experiments on GTX280 show that CUDA version is not as fast as OpenGL version. 

Note: the thread block settings are currently tuned on GPU nVidia GTX 8800, 
       which may not be optimized for other GPUs.
**************************************************************************************