<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Fri, Jan 26, 2018 at 5:59 PM, Jason Ekstrand <<a href="mailto:jason@jlekstrand.net" target="_blank">jason@jlekstrand.net</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">This series adds additional format support to BLORP. In particular, BLORP's blit and copy functions now handle: * Source and destination swizzles on all gens * Blitting to RGB formats (previously we only supported copying to RGB) * Blitting to RGBX even if it's not supported for rendering * Blitting to more exotic formats such as D24_UNORM and RGB9_E5 * Copying to and from 10_10_10_2 and 11_11_10 formats with CCS_E </blockquote><div> </div><div>I finally got around to running this series on CNL today and the 11_11_10 stuff doesn't quite work. There are some YcBcR tests in the Vulkan CTS which copy to/from 11_11_10 and they fail due to nan-flushing issues. In particular, the tests upload some pixels where one of the values is the 11-bit representation of infinity. Somewhere in the process of converting to/from float, the mantissa bits get lost (it's inf after all). I could spend a bunch of time trying to debug it or we could just leave CCS_E disabled on ISL_FORMAT_R11G11B10_FLOAT for now. In light of this, I've dropped patch 28 and left the R11G11B10_FLOAT restriction in place in patch 29. With that, this series passes the Vulkan CTS on CNL. </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> As a side-effect of this, BLORP should now be capable of doing basically all of OpenGL format conversion for texture upload. There are still a few edge-cases around packing and maybe luminance/alpha but most of it should be possible. I have no intentions of actually doing this but some enterprising person could pick it up. This series also does some refactoring and improving of ISL since that seemed better than just extending everything inside blorp. We also get a little format conversion library for NIR out of the deal. If we ever want to do image_load_store lowering in NIR, most of the format conversion code is now written. Happy Friday! Jason Ekstrand (29): blorp: Handle the RGB workaround more like other workarounds i965/surface_state: Use an identity swizzle pre-Haswell intel/isl: Add an isl_swizzle_supports_rendering helper intel/isl: Add a helper for composing swizzles intel/isl: Add a helper for inverting swizzles intel/blorp: Add swizzle support for all hardware nir: Add the start of a format conversion helper header nir/format_convert: Add linear <-> sRGB helpers nir/format_convert: Add pack/unpack for R11F_G11F_B10F nir/format_convert: Add a function to pack RGB9_E5 formats intel/blorp: Include nir_format_convert.h in blorp_blit.c intel/blorp: Handle more exotic destination formats intel/isl: Add some helpers for working with RGBX formats intel/blorp: Silently convert RGBX destination formats to RGBA intel/blorp: Support the RGB workaround on more formats anv: Allow blitting to/from any supported format i965/blorp: Remove a pile of blorp_blit restrictions util/srgb: Add a float sRGB -> linear helper intel/isl/format: Add a column for channel order to the table intel/isl/format: Add field locations informations to channel_layout intel/isl/format: Get rid of the ALPHA colorspace intel/isl: Add format conversion code intel/blorp: Use ISL instead of bitcast_color_value_to_uint nir/format_convert: Add code for bitcasting vectors intel/blorp: Use nir_format_bitcast_uint_vec_unmasked intel/blorp: Add support for more format bitcasting intel/blorp: Allow CCS copies of 1010102 formats intel/blorp: Add blorp_copy support for ISL_FORMAT_R11G11B10_FLOAT intel/isl: Allow CCS_E on more formats src/compiler/Makefile.sources | 1 + src/compiler/nir/meson.build | 1 + src/compiler/nir/nir_format_convert.h | 287 ++++++++++++ src/intel/<a href="http://Makefile.compiler.am" rel="noreferrer" target="_blank">Makefile.compiler.am</a> | 2 +- src/intel/<a href="http://Makefile.isl.am" rel="noreferrer" target="_blank">Makefile.isl.am</a> | 1 + src/intel/blorp/blorp_blit.c | 502 ++++++++++++-------- src/intel/blorp/blorp_priv.h | 23 +- src/intel/isl/gen_format_layout.py | 21 +- src/intel/isl/isl.c | 118 +++++ src/intel/isl/isl.h | 37 ++ src/intel/isl/isl_format.c | 270 ++++++++++- src/intel/isl/isl_format_layout.csv | 558 +++++++++++------------ src/intel/isl/isl_surface_state.c | 37 +- src/intel/isl/meson.build | 2 +- src/intel/vulkan/anv_blorp.c | 3 +- src/intel/vulkan/anv_formats.c | 10 +- src/mesa/drivers/dri/i965/brw_blorp.c | 67 ++- src/mesa/drivers/dri/i965/brw_wm_surface_state.c | 6 + src/util/format_srgb.h | 14 + 19 files changed, 1390 insertions(+), 570 deletions(-) create mode 100644 src/compiler/nir/nir_format_convert.h -- 2.5.0.400.gff86faf </blockquote></div> </div></div>