[Mesa-dev] [PATCH 00/11] nir: Add a pass management framework

[Jason Ekstrand]

FYI: I just force-pushed my wip/nir-pass branch with the following changes:

1) nir_pass no longer contains a data pointer
2) All of the nir_shader_run_pass-type functions now take a
nir_shader** so we can replace it for nir_clone testing
3) The nir_shader_optimize function now takes a null-terminated
nir_pass** instead of a nir_pass* and a number of passes.

On Wed, Oct 28, 2015 at 2:32 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> This series adds a nir_pass datastructure and some helpers for managing
> optimization and lowering passes.  I've been meaning to get around to this
> for some time.  There are a couple of primary benifits to this:
>
> First, this gives us a central place to put things such as validating the
> shader, printing it if it changes, etc.  Right now, the i965 backend calls
> nir_validate_shader after each pass.  We would also like to add something
> like we have in the i965 backend where it can be set to dump the IR to a
> file after every pass that changess it.
>
> Mor importantly, though, it moves metadata out of the passes them selves
> and into the runner.  In the process of putting this series together, I
> found at least 3 or 4 optimization passes that don't properly invalidate
> metadata.  By putting a metadata_preserved field in nir_pass and handling
> metadata in the pass runner, we make it much less likely that a pass will
> get this wrong.  LLVM has a similar optimization pass architecture for
> precicely this reason.
>
> As a nice little side-benifit, we no longer have to iterate over all of the
> overloads with non-NULL impl pointers in each pass.
>
> Jason Ekstrand (11):
>   nir: Move nir_metadata.c to nir_pass.c
>   nir: Add a pass-running infastructure
>   nir: Unexpose _impl versions of copy_prop and dce
>   nir/alu_to_scalar: Use the nir_pass framework internally
>   nir/phis_to_scalar: Use the nir_pass framework internally
>   nir/lower_vars_to_ssa: Use the nir_pass framework internally
>   nir: Use the nir_pass framework internally for copy_prop, dce, and cse
>   nir/algebraic: Use the nir_pass framework internally
>   nir: Use the nir_pass framework internally for more passes
>   ir3: Use nir_pass for lower_if_else
>   nir: Expose nir_pass structures rather than functions for several
>     passes
>
>  .../drivers/freedreno/ir3/ir3_compiler_nir.c       |  28 +++--
>  src/gallium/drivers/freedreno/ir3/ir3_nir.h        |   2 +-
>  .../drivers/freedreno/ir3/ir3_nir_lower_if_else.c  |  22 ++--
>  src/gallium/drivers/vc4/vc4_program.c              |  28 +++--
>  src/glsl/Makefile.sources                          |   2 +-
>  src/glsl/nir/nir.h                                 |  57 ++++++----
>  src/glsl/nir/nir_algebraic.py                      |  39 +++----
>  src/glsl/nir/nir_lower_alu_to_scalar.c             |  58 +++++++----
>  src/glsl/nir/nir_lower_phis_to_scalar.c            |  27 ++---
>  src/glsl/nir/nir_lower_vars_to_ssa.c               |  18 ++--
>  src/glsl/nir/nir_metadata.c                        |  54 ----------
>  src/glsl/nir/nir_opt_constant_folding.c            |  23 ++--
>  src/glsl/nir/nir_opt_copy_propagate.c              |  21 ++--
>  src/glsl/nir/nir_opt_cse.c                         |  24 ++---
>  src/glsl/nir/nir_opt_dce.c                         |  24 ++---
>  src/glsl/nir/nir_opt_dead_cf.c                     |  25 ++---
>  src/glsl/nir/nir_opt_peephole_select.c             |  22 ++--
>  src/glsl/nir/nir_opt_remove_phis.c                 |  19 ++--
>  src/glsl/nir/nir_opt_undef.c                       |  20 ++--
>  src/glsl/nir/nir_pass.c                            | 116 +++++++++++++++++++++
>  src/mesa/drivers/dri/i965/brw_nir.c                |  67 ++++--------
>  21 files changed, 328 insertions(+), 368 deletions(-)
>  delete mode 100644 src/glsl/nir/nir_metadata.c
>  create mode 100644 src/glsl/nir/nir_pass.c
>
> --
> 2.5.0.400.gff86faf
>