[Mesa-dev] [PATCH v2 00/73] Experimental NIR backend for radeonsi
Connor Abbott
cwabbott0 at gmail.com
Wed Jul 5 21:48:23 UTC 2017
Hi,
On Wed, Jul 5, 2017 at 3:47 AM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>
> Hi all,
>
> Here's an updated version of the NIR backend series. Most of the comments
> have been integrated, and where they haven't been fully integrated, I think
> they could be addressed as a follow-up.
>
> The series has become shorter, since Dave pulled several of the ac/nir
> changes into master already, and I've pushed a bunch of other non-controversial
> changes as well.
>
> The main changes of the remaining patches relative to v1:
>
> 1. nir_lower_io is moved from st/mesa into the drivers (patch 10). This
> is a major change that affects freedreno / vc4, which I can't test.
> st/mesa is still responsible for assigning all the locations, e.g. for
> uniforms, var->data.driver_location is setup by st/mesa, and then the
> driver calls nir_lower_io, which uses that information.
>
> 2. nir_lower_uniforms_to_ubo now converts from load_uniform intrinsics to
> load_ubo, rather than directly from uniform variable load_vars.
>
> 3. GLSLFragCoordIsSysVal is handled cleanly with a NIR intrinsic.
>
> 4. radeonsi calls nir_lower_tex itself.
>
> 5. ac_shader_abi uses LLVMValueRefs instead of function parameter indices.
>
> We still don't apply nir_lower_io to input/output variables, keeping aligned
> with how radv does things. But this is now an issue entirely internal to
> radeonsi/radv, and can be cleaned up at our leisure if we think it makes
> sense.
Ok, that makes sense.
>
> FWIW, dynamic array indexing still makes things awkward, because while
> nir_lower_io_to_temporaries helps with VS-PS inputs and outputs, it'd
> actually be preferable *not* to use it on API VS outputs when compiling
> VS-as-LS or VS-as-ES.
Hmm... what's the exact issue? Is it that we might have to compile the
same NIR shader with and without lower_io_to_temporaries? In that
case, it might make sense to clone the shader before doing the
lowering and keep around the original shader in case we need to
compile it again in a different mode.
>
> Please review!
> Cheers,
> Nicolai
> --
> src/amd/common/ac_llvm_build.c | 14 +-
> src/amd/common/ac_llvm_build.h | 4 +-
> src/amd/common/ac_nir_to_llvm.c | 1427 ++++++++++++---------
> src/amd/common/ac_nir_to_llvm.h | 9 +-
> src/amd/common/ac_shader_abi.h | 95 ++
> src/amd/common/ac_shader_info.c | 9 +
> src/amd/common/ac_shader_info.h | 2 +
> src/amd/vulkan/radv_cmd_buffer.c | 2 +-
> src/amd/vulkan/radv_pipeline.c | 2 +-
> src/compiler/Makefile.sources | 2 +
> src/compiler/glsl_types.cpp | 4 +-
> src/compiler/nir/nir.c | 17 +
> src/compiler/nir/nir.h | 5 +
> src/compiler/nir/nir_gather_info.c | 1 +
> src/compiler/nir/nir_intrinsics.h | 1 +
> src/compiler/nir/nir_lower_samplers_as_deref.c | 243 ++++
> src/compiler/nir/nir_lower_uniforms_to_ubo.c | 97 ++
> src/compiler/nir/nir_lower_wpos_ytransform.c | 6 +-
> src/compiler/nir_types.cpp | 7 +
> src/compiler/nir_types.h | 2 +
> src/gallium/docs/source/screen.rst | 2 +
> src/gallium/drivers/etnaviv/etnaviv_screen.c | 1 +
> src/gallium/drivers/freedreno/freedreno_screen.c | 1 +
> src/gallium/drivers/freedreno/ir3/ir3_cmdline.c | 15 +-
> src/gallium/drivers/freedreno/ir3/ir3_shader.c | 12 +
> src/gallium/drivers/freedreno/ir3/ir3_shader.h | 5 +
> src/gallium/drivers/i915/i915_screen.c | 1 +
> src/gallium/drivers/llvmpipe/lp_screen.c | 1 +
> src/gallium/drivers/nouveau/nv30/nv30_screen.c | 1 +
> src/gallium/drivers/nouveau/nv50/nv50_screen.c | 1 +
> src/gallium/drivers/nouveau/nvc0/nvc0_screen.c | 1 +
> src/gallium/drivers/r300/r300_screen.c | 1 +
> src/gallium/drivers/r600/r600_pipe.c | 1 +
> src/gallium/drivers/radeon/r600_pipe_common.c | 1 +
> src/gallium/drivers/radeon/r600_pipe_common.h | 2 +-
> src/gallium/drivers/radeonsi/Automake.inc | 3 +
> src/gallium/drivers/radeonsi/Makefile.am | 1 +
> src/gallium/drivers/radeonsi/Makefile.sources | 1 +
> src/gallium/drivers/radeonsi/si_pipe.c | 40 +
> src/gallium/drivers/radeonsi/si_shader.c | 864 +++++++------
> src/gallium/drivers/radeonsi/si_shader.h | 8 +
> src/gallium/drivers/radeonsi/si_shader_internal.h | 36 +-
> src/gallium/drivers/radeonsi/si_shader_nir.c | 508 ++++++++
> src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c | 63 +-
> src/gallium/drivers/radeonsi/si_state_shaders.c | 46 +-
> src/gallium/drivers/softpipe/sp_screen.c | 1 +
> src/gallium/drivers/svga/svga_screen.c | 1 +
> src/gallium/drivers/swr/swr_screen.cpp | 1 +
> src/gallium/drivers/vc4/vc4_program.c | 13 +-
> src/gallium/drivers/vc4/vc4_screen.c | 1 +
> src/gallium/drivers/virgl/virgl_screen.c | 1 +
> src/gallium/include/pipe/p_defines.h | 1 +
> src/gallium/targets/d3dadapter9/Makefile.am | 1 +
> src/gallium/targets/omx/Makefile.am | 1 +
> src/gallium/targets/va/Makefile.am | 1 +
> src/gallium/targets/vdpau/Makefile.am | 1 +
> src/gallium/targets/xvmc/Makefile.am | 1 +
> src/mesa/Makefile.sources | 2 -
> src/mesa/state_tracker/st_glsl_to_nir.cpp | 28 +-
> src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 5 +-
> src/mesa/state_tracker/st_glsl_types.cpp | 107 --
> src/mesa/state_tracker/st_glsl_types.h | 44 -
> 62 files changed, 2542 insertions(+), 1233 deletions(-)
> create mode 100644 src/amd/common/ac_shader_abi.h
> create mode 100644 src/compiler/nir/nir_lower_samplers_as_deref.c
> create mode 100644 src/compiler/nir/nir_lower_uniforms_to_ubo.c
> create mode 100644 src/gallium/drivers/radeonsi/si_shader_nir.c
> delete mode 100644 src/mesa/state_tracker/st_glsl_types.cpp
> delete mode 100644 src/mesa/state_tracker/st_glsl_types.h
>
> --
> 2.9.3
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list