Am Mi., 19. Dez. 2018 um 17:44 Uhr schrieb Jonathan Marek <jonat...@marek.ca>: > > Mainly a copy of nir_lower_bool_to_int32, but with float opcodes. >
Hmmm.. are you aware of https://patchwork.freedesktop.org/patch/257867/ and https://gitlab.freedesktop.org/jekstrand/mesa/commit/cf819c8a3fa99ccedf423ea77cf710dbd852066b ? I am going to send out a lager patch series with that version of bool to float during my christmas break. Keep in mind that I did not looked very closely at your lowering pass. > Signed-off-by: Jonathan Marek <jonat...@marek.ca> > --- > src/compiler/Makefile.sources | 1 + > src/compiler/nir/meson.build | 3 +- > src/compiler/nir/nir.h | 1 + > src/compiler/nir/nir_lower_bool_to_float.c | 165 +++++++++++++++++++++ > 4 files changed, 169 insertions(+), 1 deletion(-) > create mode 100644 src/compiler/nir/nir_lower_bool_to_float.c > > diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources > index ef47bdb33b..39eaedc658 100644 > --- a/src/compiler/Makefile.sources > +++ b/src/compiler/Makefile.sources > @@ -231,6 +231,7 @@ NIR_FILES = \ > nir/nir_lower_atomics_to_ssbo.c \ > nir/nir_lower_bitmap.c \ > nir/nir_lower_bit_size.c \ > + nir/nir_lower_bool_to_float.c \ > nir/nir_lower_bool_to_int32.c \ > nir/nir_lower_clamp_color_outputs.c \ > nir/nir_lower_clip.c \ > diff --git a/src/compiler/nir/meson.build b/src/compiler/nir/meson.build > index e252f64539..f1016104af 100644 > --- a/src/compiler/nir/meson.build > +++ b/src/compiler/nir/meson.build > @@ -114,6 +114,7 @@ files_libnir = files( > 'nir_lower_alpha_test.c', > 'nir_lower_atomics_to_ssbo.c', > 'nir_lower_bitmap.c', > + 'nir_lower_bool_to_float.c', > 'nir_lower_bool_to_int32.c', > 'nir_lower_clamp_color_outputs.c', > 'nir_lower_clip.c', > @@ -248,7 +249,7 @@ if with_tests > include_directories : [inc_common], > dependencies : [dep_thread, idep_gtest, idep_nir], > link_with : libmesa_util, > - ), > + ), > suite : ['compiler', 'nir'], > ) > > diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h > index 54f9c64a3a..f6d0bdf7ec 100644 > --- a/src/compiler/nir/nir.h > +++ b/src/compiler/nir/nir.h > @@ -2905,6 +2905,7 @@ void nir_lower_alpha_test(nir_shader *shader, enum > compare_func func, > bool alpha_to_one); > bool nir_lower_alu(nir_shader *shader); > bool nir_lower_alu_to_scalar(nir_shader *shader); > +bool nir_lower_bool_to_float(nir_shader *shader); > bool nir_lower_bool_to_int32(nir_shader *shader); > bool nir_lower_load_const_to_scalar(nir_shader *shader); > bool nir_lower_read_invocation_to_scalar(nir_shader *shader); > diff --git a/src/compiler/nir/nir_lower_bool_to_float.c > b/src/compiler/nir/nir_lower_bool_to_float.c > new file mode 100644 > index 0000000000..2756a1815f > --- /dev/null > +++ b/src/compiler/nir/nir_lower_bool_to_float.c > @@ -0,0 +1,165 @@ > +/* > + * Copyright © 2018 Intel Corporation > + * > + * Permission is hereby granted, free of charge, to any person obtaining a > + * copy of this software and associated documentation files (the "Software"), > + * to deal in the Software without restriction, including without limitation > + * the rights to use, copy, modify, merge, publish, distribute, sublicense, > + * and/or sell copies of the Software, and to permit persons to whom the > + * Software is furnished to do so, subject to the following conditions: > + * > + * The above copyright notice and this permission notice (including the next > + * paragraph) shall be included in all copies or substantial portions of the > + * Software. > + * > + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR > + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, > + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL > + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER > + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING > + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER > DEALINGS > + * IN THE SOFTWARE. > + */ > + > +#include "nir.h" > + > +static bool > +assert_ssa_def_is_not_1bit(nir_ssa_def *def, UNUSED void *unused) > +{ > + assert(def->bit_size > 1); > + return true; > +} > + > +static bool > +rewrite_1bit_ssa_def_to_32bit(nir_ssa_def *def, void *_progress) > +{ > + bool *progress = _progress; > + if (def->bit_size == 1) { > + def->bit_size = 32; > + *progress = true; > + } > + return true; > +} > + > +static bool > +lower_alu_instr(nir_alu_instr *alu) > +{ > + const nir_op_info *op_info = &nir_op_infos[alu->op]; > + > + switch (alu->op) { > + case nir_op_vec2: > + case nir_op_vec3: > + case nir_op_vec4: > + /* These we expect to have booleans but the opcode doesn't change */ > + break; > + > + case nir_op_b2f32: alu->op = nir_op_fmov; break; > + > + /* Note: we only expect these 5 opcodes with bools */ > + case nir_op_imov: alu->op = nir_op_fmov; break; > + case nir_op_inot: alu->op = nir_op_fnot; break; > + case nir_op_iand: alu->op = nir_op_fand; break; > + case nir_op_ior: alu->op = nir_op_for; break; > + case nir_op_ixor: alu->op = nir_op_fxor; break; > + > + /* We might want a new opcode (for the (x != 0.0) f2b op) */ > + case nir_op_f2b1: alu->op = nir_op_f2b32; break; > + case nir_op_i2b1: alu->op = nir_op_f2b32; break; > + > + case nir_op_flt: alu->op = nir_op_slt; break; > + case nir_op_fge: alu->op = nir_op_sge; break; > + case nir_op_feq: alu->op = nir_op_seq; break; > + case nir_op_fne: alu->op = nir_op_sne; break; > + case nir_op_ilt: alu->op = nir_op_slt; break; > + case nir_op_ige: alu->op = nir_op_sge; break; > + case nir_op_ieq: alu->op = nir_op_seq; break; > + case nir_op_ine: alu->op = nir_op_sne; break; > + case nir_op_ult: alu->op = nir_op_slt; break; > + case nir_op_uge: alu->op = nir_op_sge; break; > + > + case nir_op_ball_fequal2: alu->op = nir_op_fall_equal2; break; > + case nir_op_ball_fequal3: alu->op = nir_op_fall_equal3; break; > + case nir_op_ball_fequal4: alu->op = nir_op_fall_equal4; break; > + case nir_op_bany_fnequal2: alu->op = nir_op_fany_nequal2; break; > + case nir_op_bany_fnequal3: alu->op = nir_op_fany_nequal3; break; > + case nir_op_bany_fnequal4: alu->op = nir_op_fany_nequal4; break; > + case nir_op_ball_iequal2: alu->op = nir_op_fall_equal2; break; > + case nir_op_ball_iequal3: alu->op = nir_op_fall_equal3; break; > + case nir_op_ball_iequal4: alu->op = nir_op_fall_equal4; break; > + case nir_op_bany_inequal2: alu->op = nir_op_fany_nequal2; break; > + case nir_op_bany_inequal3: alu->op = nir_op_fany_nequal3; break; > + case nir_op_bany_inequal4: alu->op = nir_op_fany_nequal4; break; > + > + case nir_op_bcsel: alu->op = nir_op_fcsel; break; > + > + default: > + assert(alu->dest.dest.ssa.bit_size > 1); > + for (unsigned i = 0; i < op_info->num_inputs; i++) > + assert(alu->src[i].src.ssa->bit_size > 1); > + return false; > + } > + > + if (alu->dest.dest.ssa.bit_size == 1) > + alu->dest.dest.ssa.bit_size = 32; > + > + return true; > +} > + > +static bool > +nir_lower_bool_to_float_impl(nir_function_impl *impl) > +{ > + bool progress = false; > + > + nir_foreach_block(block, impl) { > + nir_foreach_instr_safe(instr, block) { > + switch (instr->type) { > + case nir_instr_type_alu: > + progress |= lower_alu_instr(nir_instr_as_alu(instr)); > + break; > + > + case nir_instr_type_load_const: { > + nir_load_const_instr *load = nir_instr_as_load_const(instr); > + if (load->def.bit_size == 1) { > + nir_const_value value = load->value; > + for (unsigned i = 0; i < load->def.num_components; i++) > + load->value.f32[i] = value.b[i] ? 1.0f : 0.0f; > + load->def.bit_size = 32; > + progress = true; > + } > + break; > + } > + > + case nir_instr_type_intrinsic: > + case nir_instr_type_ssa_undef: > + case nir_instr_type_phi: > + case nir_instr_type_tex: > + nir_foreach_ssa_def(instr, rewrite_1bit_ssa_def_to_32bit, > + &progress); > + break; > + > + default: > + nir_foreach_ssa_def(instr, assert_ssa_def_is_not_1bit, NULL); > + } > + } > + } > + > + if (progress) { > + nir_metadata_preserve(impl, nir_metadata_block_index | > + nir_metadata_dominance); > + } > + > + return progress; > +} > + > +bool > +nir_lower_bool_to_float(nir_shader *shader) > +{ > + bool progress = false; > + > + nir_foreach_function(function, shader) { > + if (function->impl && nir_lower_bool_to_float_impl(function->impl)) > + progress = true; > + } > + > + return progress; > +} > -- > 2.17.1 > > _______________________________________________ > mesa-dev mailing list > mesa-dev@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/mesa-dev -- greets -- Christian Gmeiner, MSc https://christian-gmeiner.info _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
