<div dir="ltr">I don't think they're blocked by mailman decided to give up over the week-end so it may have gotten lost. Here's a branch: <a href="https://cgit.freedesktop.org/~jekstrand/mesa/log/?h=wip/anv-proc-addr">https://cgit.freedesktop.org/~jekstrand/mesa/log/?h=wip/anv-proc-addr</a> </div><div class="gmail_extra"> <div class="gmail_quote">On Mon, Jan 22, 2018 at 1:03 AM, Samuel Iglesias Gonsálvez <<a href="mailto:siglesias@igalia.com" target="_blank">siglesias@igalia.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I don't see patches 4, 8, 9 and 18 in the mailing list. Are they blocked? Sam <div class="HOEnZb"><div class="h5"> On 20/01/18 20:11, Jason Ekstrand wrote: > Our previous scheme for Get*ProcAddr was to just return what we could and > not care about the details. This meant that GetInstanceProcAddr returned > all anv_ entrypoints and GetDeviceProcAddr would return the per-gen > entrypoint and fall back to anv_. We figured that this was a perfectly > reasonable and Vulkan thing to do and that the loader could sort out the > nasty details. We were wrong. > > The Vulkan spec has some very specific rules about what vkGet*ProcAddr is > supposed to do in various cases. In particular, you're supposed to return > NULL for any extension entrypoints which have not explicitly been enabled. > Also, vkGetInstanceProcAddr is supposed to return entrypoints for device > functionality even if they have to be trampoline entrypoints. In 99% of > case, the loader takes care of all these details for us. However, what I > hear from the loader people is that they can't do it all and that the > drivers should also follow the rules. > > On the upside, this means that our driver, short of exposing a few symbols, > is now a completely stand-alone Vulkan implementation and doesn't require a > loader. > > Cc: Samuel Iglesias Gonsálvez <<a href="mailto:siglesias@igalia.com">siglesias@igalia.com</a>> > > Jason Ekstrand (21): > anv/meson: Make anv_entrypoints_gen.py depend on anv_extensions.py > anv: Split anv_extensions.py into two files > anv/meson: Simplify some dependency and flag tracking > anv/extensions: Generate a header file with extension tables > anv: Use tables for instance extension wrangling > anv: Add a per-instance table of enabled extensions > anv: Use tables for device extension wrangling > anv: Add a per-device table of enabled extensions > anv/entrypoints: Add an Entrypoint class > anv/entrypoints: Add a LAYERS helper variable > anv/entrypoints: Split entrypoint index lookup into its own function > anv/entrypoints: Expose the different dispatch tables > anv/entrypoints: Parse entrypoints before extensions/features > anv/extensions: Fix VkVersion::c_vk_version for patch == None > anv: Properly NULL for GetInstanceProcAddr with a null instance > anv: Add a per-instance dispatch table > anv: Add a per-device dispatch table > anv: Only advertise enabled entrypoints > anv/entrypoints: Use an named tuple for params > anv: Return trampoline entrypoints from GetInstanceProcAddr > HACK: Return instance entrypoints from GetDeviceProcAddr > > src/intel/Makefile.sources | 3 +- > src/intel/<a href="http://Makefile.vulkan.am" rel="noreferrer" target="_blank">Makefile.vulkan.am</a> | 15 +- > src/intel/vulkan/anv_device.c | 174 ++++++++++++++++++- > src/intel/vulkan/anv_entrypoints_gen.py | 297 +++++++++++++++++++++++--------- > src/intel/vulkan/anv_extensions.py | 157 +---------------- > src/intel/vulkan/anv_extensions_gen.py | 202 ++++++++++++++++++++++ > src/intel/vulkan/anv_private.h | 16 +- > src/intel/vulkan/meson.build | 50 ++++-- > 8 files changed, 653 insertions(+), 261 deletions(-) > create mode 100644 src/intel/vulkan/anv_extensions_gen.py > </div></div></blockquote></div> </div>