Sign in to follow this  
Jimkm

OpenGL Need sugestion to optimize realtime bitblt->StretchBlt (scaling takes to much CPU)

Recommended Posts

Here's my problem. I need to copy and scale the screen, ~30 times a second (live screen drawing).
I do this by first getting a handle to the screen DC:

GetDC(NULL)
Then, I create a compatible DC using CreateCompatibleDC, create a DIB using CreateDIBSection, and select the DIB into the new DC. Now I StretchBlt from the screen DC to my own DC  (blit and scale the image) .

So we have

while(1)
{
StretchBlt ... (get the screen and resize from 1440*900 to 1200*700)

}

As you probable know the scaling takes most of the CPU and time.
If i do only

while(1){bitblt()} i get ~400 fps - i.e 400 call of bitblt per second.

if i use the while(1){StretchBlt} i get only ~25 fps - i.e 25 call of StretchBlt per second.

(I need 30). also since i am doing other things as well, CPU will not be idle so this code need to be ran at least at 34 FPS.

As said, the bottleneck is not the Blitting from the screen. the bottleneck  is the scaling (i use SetStretchBltMode(hCompatible, HALFTONE)).
with COLORONCOLOR i get 60FPS but quality is not good (if there only something in the middle of HALFTONE to COLORONCOLOR  ...).

For now i do not want to use directx or open GL.

Any idea how to make it faster (i only need ~10 more FPS :))
I have an idea how to make it faster  but i do know how to do that and if it is possible.
Idea is to:
Bitblt to get the screen into buffer A.
StretchBlt to scale the image into buffer B.
Now on the next bitblt's to use SRCINVERT (XOR) to blt only the differences between the new capture and the old and to use the differences to update the scaled buffer B so no need to do more scaling.

Again just an idea but i do not think it is possible.

I would love to hear your suggestions.

I tries to use bitblt with some custom scaling functions (instead of StretchBlt) i found on the Internet or with image libs like FreeImage but results were dramatically worst than StretchBlt  (even using BOX filtering)

Again, i know that using DirectX/opengl may be faster but currently i do not have the time to learn that or deal with that - so currently i would like to use GDI/GDI+ only.

Thanks!

 

Share this post


Link to post
Share on other sites

Hi. are you running this bitblt in a loop like here.

 

while(1)
{
StretchBlt ... (get the screen and resize from 1440*900 to 1200*700)

}

 

may be you need to just update it on a wm_paint or what ever the message is.

Share this post


Link to post
Share on other sites

I do not think that there is an option to catch when wm_paint is sent to the screen itself, i.e tocatch some event that is thrown when what you see on the computer screen is changed.

Share this post


Link to post
Share on other sites

It seems that halftone blitter is really slow, couldn belive it to be so slow and i did quick testing - it shows the same result youre talking here about

abput 40 ms - what system do you have, it may depend on the system 

 

i dont know some say that in win7 it may run even worse than xp ;/ ?

(found something on this

http://vjforums.info/threads/stretchblt-can-hang-windows-by-hogging-gdi-lock.38179/ )

 

anyway normal mode blitter working ok and you could rescale this by hand code or ev trying to do some "opengl/dx blitter" with fast hardware resizin onboard - which i was not testing yet

Share this post


Link to post
Share on other sites

Hi Fir.

I tried to use custom scalling - some even with ASM byut could not get even near strechblt - with the custom scaling fucntion i got ~10 FPS ....

And as say i do not want to use opengl/dx.

 

Am am using win7.

Share this post


Link to post
Share on other sites

Hi Fir.

I tried to use custom scalling - some even with ASM byut could not get even near strechblt - with the custom scaling fucntion i got ~10 FPS ....

And as say i do not want to use opengl/dx.

 

Am am using win7.

you may show the code, we can taka a lok here and maybe come advices or conclusions will appear -i could do the test of some down of upscalling in bitmap arrays im not sure but i think it could work at 50 (100?) hz or more - not 100% sure though

 

PS i did some simle test of down and up scaling frame bitmap into some buffer then copying this to framebuffer array again -and even with this double copying i got no trouble

 

for screens size like 500p it was 4 ms (scaling forth somewhere and copying back) for screens like 1000p it was 12 ms (for such two way: scaling+copying back), so not a problem here -

 

got no idea what this halftone is doing to be so slow but this linear scaling works ok

 

Edited by fir

Share this post


Link to post
Share on other sites

First i tried this:

http://www.geisswerks.com/ryan/FAQS/resize.html

Gave me around 10 FPS only.

I then tried FreeImage lib (FreeImage_Rescale) got around 7 FPS.

None came even close to StretchBlt preformace.

 

I think the answer should be

1) To find fastest rescal (with good quality) scaling funtion/

OR AND

2)to somehow use the scaling histroy.

I.E to try to scal only the parts that changed from the last scale - but not sure how to do that ...

Maybe with SRCINVERT (XOR) - somehow.

Share this post


Link to post
Share on other sites

 

 

got no idea what this halftone is doing to be so slow but this linear scaling works ok

 

 

 

Well without halftone the image quality is not good.

Text does not look good and somehting is unreadable.

The halftone smoth the image by avraging pixels before it scall or something like that.

 

Share this post


Link to post
Share on other sites

Scaling is expensive. This is why, for instance, Fraps only offers fullscreen or halfscreen resolution when it captures the screen, so that scaling is either unnecessary or very easy. It's just too costly to handle cases where you're not scaling down to a power of two of the original size, because you have to handle filtering of multiple overlapping pixels, which also blows your cache because the resulting memory access patterns are.. suboptimal to say the least. There is one thing you can try, which drastically helped me back when I was on my old laptop and trying to capture stuff, which is to set your screen resolution to 16-bit colors.

 

Otherwise, what I would try is instead get a copy of the screen in a GPU-bound texture, using your favorite graphics API, scale it using the pixel shader (which ought to be fast enough, even with bilinear filtering) and read it back on the CPU. But this might incur significant capture latency, and might not be feasible, and you said you don't want to do this anyway, so...

 

I'm not sure what your specific needs are, but have you considered simply getting one of those screen recording devices? They don't have any system overhead and you might be able to get one to feed back into the computer to deliver the screen contents in realtime.

Share this post


Link to post
Share on other sites

Hi Bacterius

Yes i know scaling is expensive. that is why using the scaling history might help.

I just do not know how to do that. I.E to scale only the parts from the image that changed from the last scale - that way i scale one time and on next scales i only scale

the parts that were changed - but not sure if that can be done and how.

 

P.S

Just tried CxImgLib and got only 20 FPS still lower than  StretchBlt ...

 

P.S2 lowering to 16 bit did not help and in anyway it should run on all systems without making changes - that is why recording device is not a solution.

Edited by Jimkm

Share this post


Link to post
Share on other sites

Could you try comparing each pixel to the previous pixel to determine a difference, and then divvy the screen up into 32x32 sized chunks and only StretchBlt the chunks that contained at least one differing pixel?

Share this post


Link to post
Share on other sites

Hi Samith.

I am not sure this will work since the scaling also does some smoting and i think that doing this in your way will make the borders of the 32*32 squares to be seen ...

I can try that but i am not sure how to do that.

Share this post


Link to post
Share on other sites
The simplest solution would be to perform the resizing of the images in parallel in multiple threads. You could put the captured images into a queue, and have the threads grab the images from said queue and resize them.

It doesn't change the fact, that 40ms is awefully slow for a resize, but if you have the CPU cores to spare, why not use them.

Share this post


Link to post
Share on other sites

Hi Samith.
I am not sure this will work since the scaling also does some smoting and i think that doing this in your way will make the borders of the 32*32 squares to be seen ...
I can try that but i am not sure how to do that.

 

You might be able to work around that by having the grid squares "overlap" with a border that is as wide as the kernel of the filter.

It would mean processing some pixels twice, but it would still be less then doing every pixel.

 

Apart from that, and using multiple cores to update squares in paralell, I don't think its much you can do unless using graphics hardware. (which is a massively parallell processor designed for image processing, seems pretty ideal for the task)

 

Why not using graphics hardware? It would be a lot less complex to get to work.

 

MAybe a bit off-topic, but why exactly do you need to scale anyway? (I'm not sure what you mean by "live screen drawing" and why this implies scaling?)

Edited by Olof Hedman

Share this post


Link to post
Share on other sites

Olof Hi

I grab screenshot, scall it to the destination device resolution and trasmit it to that device.

The device that the screens are sent to can not do the scalling.

I already use the extra CPU to encode the images so it will not help to scall using the extra cpu.

I am more looking for a way to decrease cpu time than to have more fps (if i will be able to decrease cpu time then the extra fps will come by themself).

Share this post


Link to post
Share on other sites
Hi. I don't know why you have that while(1) in your render
What breaks it.
Can you show your render code.

Share this post


Link to post
Share on other sites

Whats break the while(1) is the user that press the stop button in the gui to stop.

It is

while(user did not press the stop button)

{

capture screenshot and scale it (Streachblt)

}

Share this post


Link to post
Share on other sites

I say the slowness is comming from your code set up.

ankhd hi.

I do not think so since i think this is the most FPS my CPU can give me with StretchBlt which is done on the cpu (i use gpu monitor and i see that it does not do that on gpu).

But if you have more details why it is coming from my code then i would love to hear.

Share this post


Link to post
Share on other sites
What exactly are your requirements on the quality?

You wrote that "color on color", which is nearest downsampling, is not good enough. What about downsizing the image to exactly half it's original size with a box filter, compressing and sending that to the target device, and linearly upsampling there?
You can simulate that in GIMP/Photoshop to see, what it looks like.

Share this post


Link to post
Share on other sites

take ma advice and do it such way

 

1) wrote your own resizing code (inbeetween the raw scaling and halftone), this shouldnt be hard - you will not get as fast as simple resizing blitter  when doing something more complex than raw scalling but you also should get not as slow as halftone when writting something simpler than halftone

 

2) find some library - this should be the same, if you use something in between than raw sacling and halftone that should be both slower than raw scalling and faster than halftone 

 

thats all

Share this post


Link to post
Share on other sites

take ma advice and do it such way

 

1) wrote your own resizing code (inbeetween the raw scaling and halftone), this shouldnt be hard - you will not get as fast as simple resizing blitter  when doing something more complex than raw scalling but you also should get not as slow as halftone when writting something simpler than halftone

 

2) find some library - this should be the same, if you use something in between than raw sacling and halftone that should be both slower than raw scalling and faster than halftone 

 

thats all

I already tried few pro image library - all gave me worst FPS than strachblt

Share this post


Link to post
Share on other sites

I don't know how much time per day you've been spending on this, but in the time since you started this thread you could have already written a GPU solution. I know you say it's not what you want to do, but seriously, it wouldn't be that hard and you'll have excellent performance.

Share this post


Link to post
Share on other sites
what cpu are you using for benchmarking?

some possible solutions:
1. scale to half and then upscale to the resolution you wanted, e.g.
1440*900 -> 720*450 -> 1200*700

2. write a custom scaler for the solution you need.

3. maybe your source is in some YCbCr format or something? you could scale the source

4. interlace scaling (scale just every 2nd line per frame -> double fps)

5. scale to half and use a letterbox

6. don't interpolate, just fetch the closest pixel to show.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this  

  • Forum Statistics

    • Total Topics
      627740
    • Total Posts
      2978884
  • Similar Content

    • By DelicateTreeFrog
      Hello! As an exercise for delving into modern OpenGL, I'm creating a simple .obj renderer. I want to support things like varying degrees of specularity, geometry opacity, things like that, on a per-material basis. Different materials can also have different textures. Basic .obj necessities. I've done this in old school OpenGL, but modern OpenGL has its own thing going on, and I'd like to conform as closely to the standards as possible so as to keep the program running correctly, and I'm hoping to avoid picking up bad habits this early on.
      Reading around on the OpenGL Wiki, one tip in particular really stands out to me on this page:
      For something like a renderer for .obj files, this sort of thing seems almost ideal, but according to the wiki, it's a bad idea. Interesting to note!
      So, here's what the plan is so far as far as loading goes:
      Set up a type for materials so that materials can be created and destroyed. They will contain things like diffuse color, diffuse texture, geometry opacity, and so on, for each material in the .mtl file. Since .obj files are conveniently split up by material, I can load different groups of vertices/normals/UVs and triangles into different blocks of data for different models. When it comes to the rendering, I get a bit lost. I can either:
      Between drawing triangle groups, call glUseProgram to use a different shader for that particular geometry (so a unique shader just for the material that is shared by this triangle group). or
      Between drawing triangle groups, call glUniform a few times to adjust different parameters within the "master shader", such as specularity, diffuse color, and geometry opacity. In both cases, I still have to call glBindTexture between drawing triangle groups in order to bind the diffuse texture used by the material, so there doesn't seem to be a way around having the CPU do *something* during the rendering process instead of letting the GPU do everything all at once.
      The second option here seems less cluttered, however. There are less shaders to keep up with while one "master shader" handles it all. I don't have to duplicate any code or compile multiple shaders. Arguably, I could always have the shader program for each material be embedded in the material itself, and be auto-generated upon loading the material from the .mtl file. But this still leads to constantly calling glUseProgram, much more than is probably necessary in order to properly render the .obj. There seem to be a number of differing opinions on if it's okay to use hundreds of shaders or if it's best to just use tens of shaders.
      So, ultimately, what is the "right" way to do this? Does using a "master shader" (or a few variants of one) bog down the system compared to using hundreds of shader programs each dedicated to their own corresponding materials? Keeping in mind that the "master shaders" would have to track these additional uniforms and potentially have numerous branches of ifs, it may be possible that the ifs will lead to additional and unnecessary processing. But would that more expensive than constantly calling glUseProgram to switch shaders, or storing the shaders to begin with?
      With all these angles to consider, it's difficult to come to a conclusion. Both possible methods work, and both seem rather convenient for their own reasons, but which is the most performant? Please help this beginner/dummy understand. Thank you!
    • By JJCDeveloper
      I want to make professional java 3d game with server program and database,packet handling for multiplayer and client-server communicating,maps rendering,models,and stuffs Which aspect of java can I learn and where can I learn java Lwjgl OpenGL rendering Like minecraft and world of tanks
    • By AyeRonTarpas
      A friend of mine and I are making a 2D game engine as a learning experience and to hopefully build upon the experience in the long run.

      -What I'm using:
          C++;. Since im learning this language while in college and its one of the popular language to make games with why not.     Visual Studios; Im using a windows so yea.     SDL or GLFW; was thinking about SDL since i do some research on it where it is catching my interest but i hear SDL is a huge package compared to GLFW, so i may do GLFW to start with as learning since i may get overwhelmed with SDL.  
      -Questions
      Knowing what we want in the engine what should our main focus be in terms of learning. File managements, with headers, functions ect. How can i properly manage files with out confusing myself and my friend when sharing code. Alternative to Visual studios: My friend has a mac and cant properly use Vis studios, is there another alternative to it?  
    • By ferreiradaselva
      Both functions are available since 3.0, and I'm currently using `glMapBuffer()`, which works fine.
      But, I was wondering if anyone has experienced advantage in using `glMapBufferRange()`, which allows to specify the range of the mapped buffer. Could this be only a safety measure or does it improve performance?
      Note: I'm not asking about glBufferSubData()/glBufferData. Those two are irrelevant in this case.
    • By xhcao
      Before using void glBindImageTexture(    GLuint unit, GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum access, GLenum format), does need to make sure that texture is completeness. 
  • Popular Now