| Summary: | [ILK Bisected]Piglit spec/ARB_depth_buffer_float_fbo-clear-formats_stencil Aborted | ||
|---|---|---|---|
| Product: | Mesa | Reporter: | lu hua <huax.lu> |
| Component: | Drivers/DRI/i965 | Assignee: | Anuj Phogat <anuj.phogat> |
| Status: | VERIFIED FIXED | QA Contact: | |
| Severity: | major | ||
| Priority: | high | CC: | chadversary, idr, vlee, xunx.fang |
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux (All) | ||
| Whiteboard: | |||
| i915 platform: | i915 features: | ||
|
Description
lu hua
2013-01-17 05:26:35 UTC
It fails with a different error on SandyBridge Mobile with mesa master (git-c1dd0e3) if hiz is enabled: ./bin/fbo-clear-formats GL_ARB_depth_buffer_float stencil -auto Using test set: GL_ARB_depth_buffer_float Testing GL_DEPTH_COMPONENT32F - no stencil. fbo-clear-formats: brw_misc_state.c:652: emit_depthbuffer: Assertion `!separate_stencil || hiz_mt' failed. Testing GL_DEPTH32F_STENCIL8Aborted Including chad in CC. I think he was looking in to a similar issue on haswell. Reproduced the error reported with hiz=false. I'll soon post more details. *** Bug 61827 has been marked as a duplicate of this bug. *** Test fails due to GL_INVALID_OPERATION in glBlitFrameBuffer() while testing format=GL_DEPTH32F_STENCIL8. glBlitFrameBuffer() called from test_format()->draw_stencil_mipmap() in fbo-clear-formats.c tries to blit between framebuffers of format GL_DEPTH32F_STENCIL8 and GL_DEPTH24_STENCIL8. Blitting between these formats is not allowed in OpenGL. Test produces GL_INVALID_OPERATION on nvidia drivers as well. So, the test need to be fixed so that it blits to a FB with matching format=GL_DEPTH32F_STENCIL8. (In reply to comment #3) > Blitting between these formats is not allowed in OpenGL. > So, the test need to be > fixed so that it blits to a FB with matching format=GL_DEPTH32F_STENCIL8. Yep. The test needs fixing. From the OpenGL 3.2 spec: If the color formats of the read and draw buffers do not match, and mask in- cludes COLOR_BUFFER_BIT, pixel groups are converted to match the destination format. However, colors are clamped only if all draw color buffers have fixed- point components. Format conversion is not supported for all data types, and an INVALID_OPERATION error is generated under any of the following conditions: • The read buffer contains fixed-point or floating-point values and any draw buffer contains neither fixed-point nor floating-point values. • The read buffer contains unsigned integer values and any draw buffer does not contain unsigned integer values. • The read buffer contains signed i Lu, Test don't pass on mobile SNB even before bisected commit(252573a). It just started throwing INVALID_OPERATION after this commit. Can you bisect the commit when it really started failing? I bisect it again. Bisect shows: f04dd3d0035f57854f0715c7e0f6a0fbe6b41f15 is the first bad commit commit f04dd3d0035f57854f0715c7e0f6a0fbe6b41f15 Author: Marek Olšák <maraeo@gmail.com> Date: Mon Jan 14 06:58:52 2013 +0100 gallium: remove PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE Reviewed-by: Brian Paul <brianp@vmware.com> Bisect is incorrect. All the changes made in this commit are are for gallium drivers. Test fails before and after this commit. Select good commit:1c9833ba70466906d3d2ad3aee6a10642a39abdd Select bad commit: 56c01d81094795286828ecb83713e043695e905a [1cedf7819b605cd78ae4762c2abc20988577d852] bad [d9d857e24fef2ae5576523b861f46c426b94ba0d] good [f727fc630462c91f19a59e8b48e3a4224dd0ec52] good [d0ce8d6ceb5f597e9cb132e6395bce8fbd51b58e] bad [b77243b9c245fd5957aa6d615a3d83d061f51cc1] bad [0c535ae7fc703e10c1df7a108e8704263d7b75d3] bad [252573ae0fe6c928885f005cf6e77200b45601fb] bad Bisect shows: 252573ae0fe6c928885f005cf6e77200b45601fb is the first bad commit. I kept seeing Hua talked about Ironlake while Anuj talked about SNB. Can you sync? Hua, I assume SNB mobile passes on your side, right? Anuj, I saw you mentioned the test needs fixing. Is there a fixed test now? It passes on Sandybridge, only happens on Ironlake. (In reply to comment #9) > I kept seeing Hua talked about Ironlake while Anuj talked about SNB. Can you > sync? > > Hua, I assume SNB mobile passes on your side, right? > The issue I'm seeing on SNB can be reproduced by setting hiz=false. SNB issue looks different from what Hua is seeing on Ironlake. I think a separate bug should be opened for SNB. More investigation is required to find the cause of failure on Ironlake. I'm not planning to investigate Ironlake in near future. > Anuj, I saw you mentioned the test needs fixing. Is there a fixed test now? Test case fix will only help failure on SNB. I didn't get chance to fix the test. Fixed on latest master branch. Verified.Fixed. |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.