radv/ci: run two jobs in parallel to reduce the execution time

This should bring the overall execution time at around 15 minutes \o/,
but it exposes more issues in the driver:

 - [...].linear_interpolation.no_offset_8_samples (#7220)
 - Timeouts on some dedicated_alloc.image.full.1048577.* tests
 - Timeout on memory.pipeline_barrier.host_read_host_write

More performance improvements should come later, as the load average
remains in the [12, 16] range when running on my machine on a navi22
gets me a 31 load average and a lower execution time by a couple of
minutes.

References: #7220
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19111>

src/amd/ci/gitlab-ci.yml

@@ -323,6 +323,7 @@ vkcts-navi10-valve:
     GPU_VERSION: radv-navi10-aco
 
 vkcts-navi21-valve:
+  parallel: 2
   extends:
     - .vkcts-test-valve
     - .navi21-test-valve

src/amd/ci/radv-navi21-aco-flakes.txt

@@ -1,3 +1,16 @@
 dEQP-VK.dynamic_rendering.basic.*
 dEQP-VK.pipeline.monolithic.image.suballocation.sampling_type.separate.view_type.2d.format.a2b10g10r10_uint_pack32.count_4.size.13x23
 dEQP-VK.pipeline.monolithic.image.suballocation.sampling_type.combined.view_type.cube_array.format.r8g8_sint.count_4.size.13x13_array_of_36
+
+# Came with the split of the test lists in CI
+dEQP-VK.memory.mapping.dedicated_alloc.image.full.1048577.*
+dEQP-VK.memory.pipeline_barrier.host_read_host_write.65536
+dEQP-VK.rasterization.primitives_multisample_2_bit.static_stipple.bresenham_line_strip_wide
+
+# This fail is 100% reproducible when running vkcts in parallel mode,
+# but isn't reproducible when running with the normal full test list.
+# Given that we want both CI and developers to be able to get clean
+# runs, let's mark this test as a flake until it gets fixed.
+#
+# See: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7220
+dEQP-VK.draw.renderpass.linear_interpolation.no_offset_8_samples