Followers 0

DX9 with MRT, access violation

3 posts in this topic

So I'm attempting to work with deferred rendering for the first time, and I'm having issues with access violations. I am currently trying to distill the problem down to a short example, so I don't have much code to show at the moment.

1) I create 4 textures using device->createTexture() all using the same width, height, mips,  format, and usage(d3dusage_rendertarget);

everything seems to work ok on that front.

2)  I bind the 4 targets to slots 1-4, and render a simple cube.

3) after rendering the gbuffer, i set RT0 to the backbuffer of my swapchain, and render a quad for the light.

4) when i call present it fails.

so I scrap that, and try to figure out if something is wrong pass by pass, so after step 2, I try and write out each texture surface to a bmp file and check the results. on the second pass it crashes when trying to save the surfaces. looking at the rts from the first pass, my object is not being drawn to any of the targets. If i don't save the surfaces, I can render a few frames before I end up with an access violation while rendering the cube(right after locking the dynamic VB), which has me even more confused, as I've never had an issue with my object cache before.

EDIT: if i don't render anything(just clear buffers) everything seems to be ok, I can save the buffers which have been cleared to pink, and I can see the backbuffer, which i clear to a time dependent color, working as intended.

Does anyone have any ideas I can run with at this point? Helpful would be a short example, or insight as to why i would get an access violation, that seems to shift around in the code.

As I said above, i'm trying to distill the code down to a simple example without any of my framework code in that may be causing the issue(though it works fine in all other demo projects I've used it in). d3d9 tutorials with multiple render targets are few and far between(I've yet to find one that is c++, to be honest) and MSDN doesn't yield much else.

Edited by Burnt_Fyr
0

Share on other sites

Have you tried enabling the debug runtime and checking the debug output?

1

Share on other sites

Have you tried enabling the debug runtime and checking the debug output?

ugh I feel dumb. I thought that I had no control over that since the update that broke pix, but I see i can still use the debug data in VS2010. Thanks for that, it will help immensely.

I'm getting a failed assert:

file s:\gfx_aug09\windows\directx\dxg\inactive\d3d9\d3d\inc\vsinout.h line 397 pRstRngEnd[ begin] == SentinelMax();

after binding the pixelshader. Any ideas?

0

Share on other sites

A simple missed break in a switch statement, which allowed my vertex shader to bind to both the vertex and pixel stage was the issue. That cleared up the assert, and i'll be well on my way to multiple render targets shortly with any luck.

EDIT: the other issue was trying to render a cube with 8 verts, and 36 indices. Neither of these would have been a problem with my last my last stable version, as it would have identified the draw primitive error, and my shaders were not based on an interface. In order to get a cleaner interface that will be mostly compatible with directx11, I'm rewriting much of how my code handles DX9 objects, and there are still many bugs that need to be patched. regardless, I have the MRTS rendering, so now it's on to the second pass.

Edited by Burnt_Fyr
1

Create an account

Register a new account