[Mesa-dev] [PATCH v2 00/29] nir: Use a 1-bit data type for booleans
Jason Ekstrand
jason at jlekstrand.net
Thu Dec 6 19:44:51 UTC 2018
This is a v2 of my series to switch NIR over to 1-bit Booleans. The first
version of the series can be found here:
https://patchwork.freedesktop.org/series/51351/
Since then, a bit of work has been done on NIR to make the transition a bit
smoother. Connor rewrote the entire bit-size inference infrastructure in
nir_algebraic.py to be more dynamic and accurate. On top of that, I
reworked x2b conversions to work like other conversions and have a sized
destination instead of having a fixed size on the source and unsized
destination. This means that switching to 1-bit booleans doesn't involve
those weird b322x opcodes that were in the first version. A few of the
patches from the first version have also landed in the intervening time.
Shader-db results on Kaby Lake for the entire series:
total instructions in shared programs: 15072525 -> 14977978 (-0.63%)
instructions in affected programs: 1178699 -> 1084152 (-8.02%)
helped: 2126
HURT: 14
total cycles in shared programs: 369737290 -> 359790175 (-2.69%)
cycles in affected programs: 127174789 -> 117227674 (-7.82%)
helped: 2020
HURT: 325
total loops in shared programs: 4401 -> 4401 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0
total spills in shared programs: 10158 -> 10051 (-1.05%)
spills in affected programs: 1429 -> 1322 (-7.49%)
helped: 8
HURT: 15
total fills in shared programs: 22105 -> 21732 (-1.69%)
fills in affected programs: 2532 -> 2159 (-14.73%)
helped: 9
HURT: 15
I've tried, by adding a lowering pass and enabling it, to ensure that this
series doesn't break anyone's drivers. However, as with any other core NIR
rework, please test. I don't have the capability to really test anything
other than Intel.
Cc: Connor Abbott <cwabbott0 at gmail.com>
Cc: Timothy Arceri <tarceri at itsqueeze.com>
Cc: Eric Anholt <eric at anholt.net>
Cc: Rob Clark <robdclark at gmail.com>
Cc: Karol Herbst <karolherbst at gmail.com>
Cc: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
Jason Ekstrand (28):
nir/opcodes: Pull in the type helpers from constant_expressions
nir/opcodes: Rename tbool to tbool32
nir/algebraic: Clean up some __str__ cruft
nir/algebraic: Refactor codegen a bit
nir/algebraic: Add support for unsized conversion opcodes
nir/opt_algebraic: Simplify an optimization using the new search ops
nir/opt_algebraic: Drop bit-size suffixes from conversions
nir/opt_algebraic: Add 32-bit specifiers to a bunch of booleans
nir: Make boolean conversions sized just like the others
nir/algebraic: Optimize x2b(xneg(a)) -> a
nir: Drop support for lower_b2f
nir/algebraic: Make an optimization more specific
nir: Rename Boolean-related opcodes to include 32 in the name
FIXUP: nir/builder: Generate 32-bit bool opcodes transparently
FIXUP: nir/algebraic: Remap Boolean opcodes to the 32-bit variant
FIXUP: Use 32-bit opcodes in the NIR producers and optimizations
FIXUP: Use 32-bit opcodes in the NIR back-ends
nir/constant_expressions: Rework Boolean handling
nir: Add support for 1-bit data types
nir/large_constants: Properly handle 1-bit bools
nir/algebraic: Generalize an optimization
nir: Add 1-bit Boolean opcodes
nir: Add a bool to int32 lowering pass
glsl,spirv: Generate 1-bit Booleans
FIXUP: Revert "Use 32-bit opcodes in the NIR producers and
optimizations"
FIXUP: Revert "nir/builder: Generate 32-bit bool opcodes
transparently"
FIXUP: nir/builder: Generate 1-bit Booleans in nir_build_imm_bool
nir/algebraic: Optimize 1-bit Booleans
nir/algebraic: Add some optimizations for D3D-style Booleans
src/amd/common/ac_nir_to_llvm.c | 34 +--
src/amd/vulkan/radv_shader.c | 2 +
src/broadcom/compiler/nir_to_vir.c | 52 ++---
src/broadcom/compiler/vir.c | 2 +
src/compiler/Makefile.sources | 1 +
src/compiler/glsl/glsl_to_nir.cpp | 17 +-
src/compiler/nir/meson.build | 1 +
src/compiler/nir/nir.c | 15 +-
src/compiler/nir/nir.h | 31 +--
src/compiler/nir/nir_algebraic.py | 84 ++++++--
src/compiler/nir/nir_builder.h | 28 ++-
src/compiler/nir/nir_constant_expressions.py | 74 +++----
src/compiler/nir/nir_instr_set.c | 23 +-
src/compiler/nir/nir_lower_alu_to_scalar.c | 4 +
src/compiler/nir/nir_lower_bool_to_int32.c | 160 ++++++++++++++
src/compiler/nir/nir_lower_idiv.c | 2 +-
src/compiler/nir/nir_lower_int64.c | 2 +-
.../nir/nir_lower_load_const_to_scalar.c | 3 +
src/compiler/nir/nir_opcodes.py | 98 ++++++---
src/compiler/nir/nir_opcodes_c.py | 41 ++--
src/compiler/nir/nir_opt_algebraic.py | 215 +++++++++---------
src/compiler/nir/nir_opt_constant_folding.c | 3 +
src/compiler/nir/nir_opt_if.c | 2 +-
src/compiler/nir/nir_opt_large_constants.c | 14 +-
src/compiler/nir/nir_print.c | 3 +
src/compiler/nir/nir_search.c | 114 +++++++++-
src/compiler/nir/nir_search.h | 17 +-
src/compiler/nir/nir_validate.c | 2 +-
src/compiler/nir_types.cpp | 2 +-
src/compiler/nir_types.h | 4 +-
src/compiler/spirv/spirv_to_nir.c | 11 +-
src/compiler/spirv/vtn_glsl450.c | 4 +-
src/freedreno/ir3/ir3_compiler_nir.c | 34 +--
src/gallium/auxiliary/nir/tgsi_to_nir.c | 20 +-
src/gallium/drivers/radeonsi/si_shader_nir.c | 2 +
src/gallium/drivers/vc4/vc4_program.c | 50 ++---
src/intel/compiler/brw_fs_nir.cpp | 85 ++++----
src/intel/compiler/brw_nir.c | 2 +
.../brw_nir_analyze_boolean_resolves.c | 24 +--
src/intel/compiler/brw_vec4_nir.cpp | 123 +++++------
src/mesa/program/prog_to_nir.c | 4 +-
41 files changed, 927 insertions(+), 465 deletions(-)
create mode 100644 src/compiler/nir/nir_lower_bool_to_int32.c
--
2.19.2
More information about the mesa-dev
mailing list