Created attachment 67592 [details] apitrace Attached is an apitrace, and screenshots. Bad behaviour with --enable-r600-llvm-compiler: OpenGL vendor string: X.Org OpenGL renderer string: Gallium 0.4 on AMD RV730 OpenGL version string: 3.0 Mesa 9.1-devel (git-fb40f88) OpenGL shading language version string: 1.30 If you look at the 'bad.png' screenshot you'll see that the 4th column is light blue. It should be grey. With R600_LLVM=0 the good behaviour is seen: good.png. See the apitrace for the full shader, here is the relevant bit: else if (gl_FragCoord.x < 640) color = vec4(1.0f, 1.0f, 1.0f, 1.0f);//white [...] vec4 color0, color1, color2, color3; if (gl_FragCoord.y < 200) { color0 = color1 = color2 = color;//75% [...] } else if (xmod < 55 || xmod >= 105 || ymod < 75 || ymod >= 125) { // real color // TODO: buggy on r600g with LLVM? gl_FragColor = (color0 + color1 + color2 + color3) / 4.0f; P.S.: the shader is not optimal, most of the branches could be moved out of it, I haven't tried to minimize the testcase though.
Created attachment 67593 [details] another.trace Another apitrace that doesn't use 3.0 features and can thus be replayed with LIBGL_ALWAYS_SOFTWARE too. The software rendered shows the same (good) behaviour as the R600_LLVM=0. Note: the color values have a different gamma in this trace.
Created attachment 67594 [details] bad.png
Created attachment 67595 [details] good.png
Graphics card type: 01:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI RV730 PRO [Radeon HD 4650] xorg.conf: Section "Device" Identifier "Radeon" Driver "radeon" Option "ColorTiling2D" "True" BusID "PCI:1:0:0" EndSection kernel: Linux debian 3.6.0-rc6 x86_64 Mesa build flags: ./configure --prefix=/opt/xorg --with-driver=dri --with-state-trackers="egl dri" --with-dri-drivers=i965 --with-gallium-drivers="r600 swrast" LLVM_CONFIG=/usr/bin/llvm-config-3.1 --enable-r600-llvm-compiler --enable-openvg --enable-vdpau --enable-glx-tls --enable-shared-glapi --enable-texture-float --enable-egl --enable-gallium-egl --enable-gbm --enable-gallium-gbm --with-egl-platforms=x11,drm,fbdev --enable-gles2 /opt/xorg has xserver, drm, libkms, etc. from git.
Sorry, chose wrong component: bug is about r600g.
Created attachment 67624 [details] bug.shader_test A small piglit .shader_test that reproduces the bug: $ bin/shader_runner -auto bug.shader_test Probe at (0,0) Expected: 0.500000 0.500000 0.500000 1.000000 Observed: 0.501961 1.000000 1.000000 1.000000 PIGLIT: {'result': 'fail' } $ R600_LLVM=0 bin/shader_runner -auto bug.shader_test PIGLIT: {'result': 'pass' } $ LIBGL_ALWAYS_SOFTWARE=1 bin/shader_runner -auto bug.shader_test PIGLIT: {'result': 'pass' }
There are some asic specific shader instruction behaviors for r6xx/r7xx chips that need to be ported to the LLVM backend. A couple that come to mind: - Trig functions (see tgsi_setup_trig() in r600_shader.c) - AR register handling (see load_ar() in r600_asm.c)
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/419.
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.