# delete lines from the test list. Be careful.
# Skip the perf/stress tests to keep runtime manageable
-dEQP-GLES[0-9]*.performance
-dEQP-GLES[0-9]*.stress
+dEQP-GLES[0-9]*.performance.*
+dEQP-GLES[0-9]*.stress.*
# These are really slow on tiling architectures (including llvmpipe).
-dEQP-GLES[0-9]*.functional.flush_finish
+dEQP-GLES[0-9]*.functional.flush_finish.*
-# Unstable test results
-dEQP-GLES3.functional.fragment_out.random.*
-dEQP-GLES3.functional.transform_feedback.*
-dEQP-GLES31.functional.primitive_bounding_box.*
+# Flakes reported more than once during Jan-Feb 2020
dEQP-GLES31.functional.layout_binding.ssbo.fragment_binding_array
-# Seen a couple flakes on this one. Note that valgrind complains about
-# some things in deqp reference renderer on this one. Not sure if that
-# is a real problem or perhaps valgrind gets confused about unitialized
-# z24 channel in z24s8?? Let's just skip this one for now:
-dEQP-GLES3.functional.fbo.msaa.2_samples.stencil_index8
+# This started failing, despite passing locally (and generating identical
+# cmdstream as before. Not sure what is going on, but adding it to skips
+# for now
+dEQP-GLES31.functional.compute.shared_var.atomic.compswap.lowp_int
-# Two reports of spurious failures on unrelated MRs (2019-09-27, 2019-10-05)
-dEQP-GLES3.functional.texture.specification.texsubimage2d_pbo.r16ui_2d
+# Non-sysmem flakes
+dEQP-VK.pipeline.spec_constant.compute.composite.matrix.mat3x2
-# Layered rendering is sysmem only and needs working clears
-dEQP-GLES31.functional.geometry_shading.layered.*
-dEQP-GLES31.functional.geometry_shading.instanced.*layer.*
+# Fails NIR_VALIDATE so probably flaky
+dEQP-VK.memory_model.write_after_read.core11.u32.coherent.fence_fence.atomicwrite.workgroup.payload_nonlocal.workgroup.guard_local.buffer.comp
+
+# Sysmem flake: this one is fairly frequent, but if you enable it then
+# it moves to dEQP-VK.renderpass.dedicated_allocation.attachment.3.393
+#
+#dEQP-VK.renderpass.suballocation.attachment_allocation.grow_shrink.89
+
+# At least some of the separate_channels tests fail on sysmem due to an
+# interaction of use of a UBWC buffer as both a render target and a
+# texture. Stores are done through both paths in separate channels,
+# and the UBWC updates don't get synced. The current a650 blob also
+# fails these tests and qcom apparently noted the failure at one point
+# https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/2017
+dEQP-VK.renderpass.*separate_channels.*
projects / mesa.git / blobdiff