[Piglit] [PARTCH 00/13] More tests for atomics

Ian Romanick idr at freedesktop.org
Fri Jun 1 22:43:05 UTC 2018


This series has two parts.

The second part adds tests for NV_shader_atomic_float.  I worked on an
implementation of this extension for i965 until I realized the hardware
couldn't actually do it.  I had the tests, so it seems a shame to let
them rot in a branch somewhere.  Patches 10, 11, and 13 were fully
tested on my implementation for Mesa.  Patch 12 is not tested, but read
on.

The first part adds a bunch of tests for SSBO and compute shader shared
variable integer atomics.  As I was implementing the tests for
NV_shader_atomic_float, I realized that I needed to test my tests before
I had a working implementatation.  Since integer atomics are fully
deployed, I modified the in-progress float tests to be integer tests.
The float tests are mostly s/int/float/g.  Along the way I also
discovered a bug in NIR with SSBOs (see
https://bugs.freedesktop.org/show_bug.cgi?id=106774).

The third part, which should go out next week or the week after, will
add tests for a new floating-point atomic extension called
INTEL_shader_atomic_float_minmax.  I'm still nailing down bits of the
spec and the associated test cases.  Most of the tests for the new
extension are nearly identical to tests in this series (especially the
test for atomicExchange(float, float)).  If reviewers would like to see
changes in these tests, I will apply those changes to the
INTEL_shader_atomic_float_minmax tests before sending them out.



More information about the Piglit mailing list