I am trying to render to just a rectangular portion of the frame. To my knowledge there are two options for this - glScissors and glViewport.
When profiling with DS-5 Streamline on both Mali-400 MP and Mali-450 MP, the number of rasterized fragments is always equal to the number of pixels in the frame. It does not matter if only a portion of the frame is specified. My understanding is, that the driver does not understand, that I have set glScissors or glViewport only once and the fragment shader still processes all pixels of the frame.
Does anyone know of a way to really render to just a portion of the frame?
To be more precise, I am rendering to a texture using OpenGL ES 2.0. I was thinking to attach just a part of the texture to the frame buffer, but I am not aware of any such functionality.
Can you post a short example which reproduces the problem you are seeing? It's very hard to help debug why your API usage isn't working without knowing what your API usage actually is ...
Thanks, Pete
Hi sogartar,
As Pete has already mentioned, the best way for us to identify if there is a problem here is with a reproducer sample or code snippet to help us understand if the behavior you see is expected or not.
Some further information while we wait to potentially explain this:
Scissor and Viewports only block the rendered drawcall, not the readback. It is possible that what you are measuring with DS-5 is the readback.
You can avoid the readback as well, if you first do a fullscreen (non-scissored) glClear at the beginning of the frame.
Kind Regards,
Michael McGeagh
What does readback mean? Is this the copy of a part from the buffer in main memory to a tile before rendering?
Exactly that.
Below is a trace of the calls to OpenGL. Basically it draws into a texture one triangle that covers half of the viewport, which is 100x100. The whole texture is 1920x1200. The prior contents of the texture, where it is not drawn must be preserved, so glClear is not used. What I observe in streamline is that the total data written to and read from the bus is equal to the size in bytes of the whole buffer. This leads me to believe, that there is no optimization, where only the area under the viewport is rendered.
1 glCreateShader type=GL_VERTEX_SHADER
null glShaderSource shader=1 count=1 string="attribute vec4 vPosition;
void main()
{
gl_Position = vPosition;
}
" length=0x0
null glCompileShader shader=1
null glGetShaderiv shader=1 pname=GL_COMPILE_STATUS params=1
2 glCreateShader type=GL_FRAGMENT_SHADER
null glShaderSource shader=2 count=1 string="precision mediump float;
gl_FragColor = vec4(1.0, 0.0, 0.0, 1.0);
null glCompileShader shader=2
null glGetShaderiv shader=2 pname=GL_COMPILE_STATUS params=1
3 glCreateProgram
null glAttachShader program=3 shader=1
null glAttachShader program=3 shader=2
null glLinkProgram program=3
null glGetProgramiv program=3 pname=GL_LINK_STATUS params=1
0 glGetAttribLocation program=3 name="vPosition"
null glViewport x=0 y=0 width=100 height=100
null glGenFramebuffers n=1 framebuffers=[1]
null glBindFramebuffer target=GL_FRAMEBUFFER framebuffer=1
null glGenTextures n=1 textures=[1]
null glBindTexture target=GL_TEXTURE_2D texture=1
null glTexImage2D target=GL_TEXTURE_2D level=0 internalformat=GL_RGBA width=1920 height=1200 border=0 format=GL_RGBA type=GL_UNSIGNED_BYTE pixels=0x0
null glTexParameteri target=GL_TEXTURE_2D pname=GL_TEXTURE_MAG_FILTER param=GL_NEAREST
null glTexParameteri target=GL_TEXTURE_2D pname=GL_TEXTURE_MIN_FILTER param=GL_NEAREST
null glFramebufferTexture2D target=GL_FRAMEBUFFER attachment=GL_COLOR_ATTACHMENT0 textarget=GL_TEXTURE_2D texture=1 level=0
null glBindTexture target=GL_TEXTURE_2D texture=0
GL_FRAMEBUFFER_COMPLETE glCheckFramebufferStatus target=GL_FRAMEBUFFER
null glBindFramebuffer target=GL_FRAMEBUFFER framebuffer=0
null glUseProgram program=3
null glVertexAttribPointer index=0 size=2 type=GL_FLOAT normalized=GL_FALSE stride=0 pointer=0x77d66e2c
null glEnableVertexAttribArray index=0
null glDrawArrays mode=GL_TRIANGLES first=0 count=3