Vulkan tests regresson on i965: dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc1_rgb_srgb_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc1_rgb_unorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc1_rgba_srgb_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc1_rgba_unorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc4_snorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.basic.image_store.bc4_unorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc1_rgb_srgb_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc1_rgb_unorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc1_rgba_srgb_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc1_rgba_unorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc4_snorm_block.r16g16b16a16_sfloat dEQP-VK.image.texel_view_compatible.compute.extended.image_store.bc4_unorm_block.r16g16b16a16_sfloat In the output for these tests: error: src->ssa->num_components == num_components (../src/compiler/nir/nir_validate.c:206) Bisected to: 06fbcd2cd5cc5702c9039c26d20082a99bc157bf Author: Eric Anholt <eric@anholt.net> intel: Simplify the half-float packing in image load/store lowering. This was noted by Jason in review when I tried to make a helper for the old path. Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Also regressed by 06fbcd2cd5cc5702c9039c26d20082a99bc157bf: piglit.spec.arb_shader_image_load_store.qualifiers piglit.spec.arb_shader_image_load_store.invalid Separately, KHR-GL46.shader_image_load_store.multiple-uniforms was regressed by another patch in the series: Invalid result! Image format: rgb10_a2ui Original value: 0x40104208 Copied value: 0x401041ff Negated value: 0x5ff7fdff d3e046e76c06978d92bc7311bf02926e888159dc Author: Eric Anholt <eric@anholt.net> nir: Pull some of intel's image load/store format conversion to nir_format.h I needed the same functions for v3d. Note that the color value in the Intel lowering has already been cut down to image.chans num_components. v2: Drop the half float one, since it was a 1-liner after cleanup. Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
You'll need the CTS : commit cfb8d04724da69306602c91c077eebf0af03719c (HEAD) Merge: a3e64b16f 39bc6c57a Author: Alexander Galazin <alexander.galazin@arm.com> Date: Thu Aug 2 10:27:48 2018 +0200 Merge vk-gl-cts/vulkan-cts-1.1.0 into vk-gl-cts/vulkan-cts-1.1.1 Because those particular tests (bc* -> sfloat) kind of disappeared on more recent versions. Also seems to affect only BDW so stub your driver to the right platform!
(In reply to Lionel Landwerlin from comment #2) > Also seems to affect only BDW so stub your driver to the right platform! I should have specified that BDW and HSW regressed for the vulkan tests.
The actual commit introducing this issue is : commit 06fbcd2cd5cc5702c9039c26d20082a99bc157bf Author: Eric Anholt <eric@anholt.net> Date: Wed Dec 12 11:29:29 2018 -0800 intel: Simplify the half-float packing in image load/store lowering. But I can't really understand how this is correct because it reduces the number of channels. Do we need to stuff the other channels with 0s?
(In reply to Lionel Landwerlin from comment #4) > But I can't really understand how this is correct because it reduces the > number of channels. Do we need to stuff the other channels with 0s? Eric reverted the commit, fixing the vulkan tests.
With Eric's recent series, the piglit and vulkan regressions are fixed. The only remaining regression is: KHR-GL46.shader_image_load_store.multiple-uniforms Sample failure here: https://mesa-ci.01.org/mesa_master/builds/14758/results/521013591 The remaining regression was initially bisected to: d3e046e76c06978d92bc7311bf02926e888159dc nir: Pull some of intel's image load/store format conversion to nir_format.h
The last regression documented by this bug was fixed with: 708d8f4d0a557f509435e05696a4096a2900d748 Author: Eric Anholt <eric@anholt.net> Commit: Jason Ekstrand <jason@jlekstrand.net> nir: Fix clamping of uints for image store lowering. I botched some copy-and-paste and clamped to signed int max instead of uint max. Fixes KHR-GL46.shader_image_load_store.multiple-uniforms on skl. Fixes: d3e046e76c06 ("nir: Pull some of intel's image load/store format conversion to nir_format.h") Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
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.