[Mesa-dev] [PATCH v2 00/31] Nir support for Nouveau

Thu Jan 4 17:52:56 UTC 2018

On January 4, 2018 11:50:12 Karol Herbst <kherbst at redhat.com> wrote:

> On Thu, Jan 4, 2018 at 6:39 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>> I'm fairly pleased that you were able to get this far with only 3k lines.
>> How much is that compared to your TGSI consumer?
>>
>>
>
> around 4,3k, but keep in mind, that we really only want to convert
> things here and all the RA, optimizations and emitting stuff is shared
> (on top of the 4,3k loc)

Yeah, I assumed as much.  We also have a decent sized back-end IR (two of 
them, actually).

>>
>> On January 4, 2018 09:02:24 Karol Herbst <kherbst at redhat.com> wrote:
>>
>>> significant changes to last series:
>>> * support for geometry, tessellation and compute shaders
>>> * support for more intrinsics and operations
>>> * lot of fixes for 64 bit types
>>> * NV50_PROG_USE_NIR=1 can be used to enable NIR for non debug builds
>>>
>>> There are some outstanding issues triggered by having 64 bit loads and
>>> nir_compact_varyings also does some harm. But I think the current state is
>>> quite good and a lot of things will work out of the box already.
>>>
>>> I mainly tested with piglit, unigine heaven and valley. RealisticRendering
>>> Demo
>>> doesn't run yet, but this might be related to some 64bit or memory related
>>> issues.
>>>
>>> It would be nice if others could just run some games/tests with it and
>>> report
>>> back where things are still broken, so that the focus can be on fixing
>>> real
>>> issues instead of trying to pass all tests first.
>>>
>>> ./piglit run -x glx -x egl -x streaming-texture-leak -x max-texture-size
>>> tests/gpu.py:
>>> [26073/26073] skip: 1640, pass: 20773, warn: 9, fail: 1548, crash: 2103
>>>
>>> the biggest todos looking at the piglit results are:
>>> * arb_gpu_shader5 interpolateat* (those nir ops don't map well to nvir)
>>>   no good plan on how to properly implement those
>>> * arb_gpu_shader5.texturegatheroffsets (nir internal assert)
>>>   glsl_to_nir.cpp:2082: virtual void
>>> {anonymous}::nir_visitor::visit(ir_texture*): Assertion
>>> `ir->offset->type->is_vector() || ir->offset->type->is_scalar()' failed.
>>> * some int64 stuff related to compound types
>>> * various extensions
>>> * variable-indexing (related to above mentioned packing issue)
>>> * glsl-4.20.execution.vs_in
>>> * some variable-indexing issues related to unaligned memory accesses
>>> * some geometry shader fails
>>>
>>> Karol Herbst (31):
>>>   st/nir: treat tess shader input/outputs the same as geom shaders
>>>   nvir: print the shader type when dumping headers
>>>   nvir: move common converter code in base class
>>>   nvir: add lowering helper
>>>   nvc0: add support for NIR
>>>   nvc0/debug: add env var to make nir default
>>>   nvir/nir: run some passes to make the conversion easier
>>>   nvir/nir: track defs and provide easy access functions
>>>   nvir/nir: add nir type helper functions
>>>   nvir/nir: use lowering helper
>>>   nvir/nir: run assignSlots
>>>   nvir/nir: parse NIR shader info
>>>   nvir/nir: implement CFG handling
>>>   nvir/nir: implement nir_load_const_instr
>>>   nvir/nir: add skeleton for nir_intrinsic_instr
>>>   nvir/nir: implement nir_alu_instr handling
>>>   nvir/nir: implement nir_intrinsic_load_uniform
>>>   nvir/nir: implement nir_intrinsic_store_(per_vertex_)output
>>>   nvir/nir: implement nir_intrinsic_load_input
>>>   nvir/nir: implement intrinsic_discard(_if)
>>>   nvir/nir: implement loading system values
>>>   nvir/nir: implement nir_ssa_undef_instr
>>>   nvir/nir: implement nir_instr_type_tex
>>>   nvir/nir: add getOperation for intrinsics
>>>   nvir/nir: implement vote and ballot
>>>   nvir/nir: implement variable indexing
>>>   nvir/nir: implement geometry shader nir_intrinsics
>>>   nvir/nir: implement nir_intrinsic_load_ubo
>>>   nvir/nir: implement ssbo intrinsics
>>>   nvir/nir: implement images
>>>   nvir/nir: add memory barriers
>>>
>>>  src/gallium/drivers/nouveau/Makefile.sources       |    5 +
>>>  src/gallium/drivers/nouveau/codegen/nv50_ir.cpp    |    3 +
>>>  src/gallium/drivers/nouveau/codegen/nv50_ir.h      |    1 +
>>>  .../nouveau/codegen/nv50_ir_from_common.cpp        |  107 +
>>>  .../drivers/nouveau/codegen/nv50_ir_from_common.h  |   58 +
>>>  .../drivers/nouveau/codegen/nv50_ir_from_nir.cpp   | 2630
>>> ++++++++++++++++++++
>>>  .../drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp  |  106 +-
>>>  .../nouveau/codegen/nv50_ir_lowering_helper.cpp    |  204 ++
>>>  .../nouveau/codegen/nv50_ir_lowering_helper.h      |   50 +
>>>  src/gallium/drivers/nouveau/meson.build            |   14 +-
>>>  src/gallium/drivers/nouveau/nvc0/nvc0_program.c    |   17 +-
>>>  src/gallium/drivers/nouveau/nvc0/nvc0_screen.c     |   50 +-
>>>  src/gallium/drivers/nouveau/nvc0/nvc0_state.c      |   27 +-
>>>  src/mesa/state_tracker/st_glsl_to_nir.cpp          |    4 +-
>>>  14 files changed, 3162 insertions(+), 114 deletions(-)
>>>  create mode 100644
>>> src/gallium/drivers/nouveau/codegen/nv50_ir_from_common.cpp
>>>  create mode 100644
>>> src/gallium/drivers/nouveau/codegen/nv50_ir_from_common.h
>>>  create mode 100644
>>> src/gallium/drivers/nouveau/codegen/nv50_ir_from_nir.cpp
>>>  create mode 100644
>>> src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_helper.cpp
>>>  create mode 100644
>>> src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_helper.h
>>>
>>> --
>>> 2.14.3
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>
>>
>>