We've had this optimization in the GL driver since July but never got
around to hooking it up in Vulkan. This lets us turn UBOs into push
constants which are significantly faster since they read the UBO data
once and the shader dispatch shoves it into the shader prior to program
execution.
I have zero benchmark numbers at the moment but I expect this to help
anything which uses UBOs and has constant offsets into them which is
basically everything. Unfortunately, I'm at home and not near any
hardware with stable enough performance characteristics to get decent
numbers today.
Cc: Kenneth Graunke <kenn...@whitecape.org>
Cc: Eero Tamminen <eero.t.tammi...@intel.com>
Jason Ekstrand (6):
anv/pipeline: Translate vulkan_resource_index to a constant when
possible
anv/cmd_buffer: Add some helpers for working with descriptor sets
anv/cmd_buffer: Add some stage asserts
anv/cmd_buffer: Add support for pushing UBO ranges
anv/device: Increase the UBO alignment requirement to 32
anv: Enable UBO pushing
src/intel/vulkan/anv_device.c | 13 +-
src/intel/vulkan/anv_nir_apply_pipeline_layout.c | 15 +-
src/intel/vulkan/anv_pipeline.c | 6 +
src/intel/vulkan/genX_cmd_buffer.c | 191 ++++++++++++++++++-----
src/intel/vulkan/genX_pipeline.c | 3 +-
5 files changed, 179 insertions(+), 49 deletions(-)
--
2.5.0.400.gff86faf
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
