src/gallium/drivers/freedreno/ir3/ir3_nir.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2015 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29
  30 #include "freedreno_util.h"
  31
  32 #include "ir3_nir.h"
  33 #include "ir3_compiler.h"
  34 #include "ir3_shader.h"
  35
  36 #include "nir/tgsi_to_nir.h"
  37
  38 struct nir_shader *
  39 ir3_tgsi_to_nir(const struct tgsi_token *tokens)
  40 {
  41         static const nir_shader_compiler_options options = {
  42                         .lower_fpow = true,
  43                         .lower_fsat = true,
  44                         .lower_scmp = true,
  45                         .lower_flrp = true,
  46                         .lower_ffract = true,
  47                         .native_integers = true,
  48                         .vertex_id_zero_based = true,
  49                         .lower_extract_byte = true,
  50                         .lower_extract_word = true,
  51         };
  52         return tgsi_to_nir(tokens, &options);
  53 }
  54
  55 /* for given shader key, are any steps handled in nir? */
  56 bool
  57 ir3_key_lowers_nir(const struct ir3_shader_key *key)
  58 {
  59         return key->fsaturate_s | key->fsaturate_t | key->fsaturate_r |
  60                         key->vsaturate_s | key->vsaturate_t | key->vsaturate_r |
  61                         key->ucp_enables | key->color_two_side;
  62 }
  63
  64 #define OPT(nir, pass, ...) ({                             \
  65    bool this_progress = false;                             \
  66    NIR_PASS(this_progress, nir, pass, ##__VA_ARGS__);      \
  67    this_progress;                                          \
  68 })
  69
  70 #define OPT_V(nir, pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
  71
  72 struct nir_shader *
  73 ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s,
  74                 const struct ir3_shader_key *key)
  75 {
  76         struct nir_lower_tex_options tex_options = {
  77                         .lower_rect = 0,
  78         };
  79         bool progress;
  80
  81         if (key) {
  82                 switch (shader->type) {
  83                 case SHADER_FRAGMENT:
  84                 case SHADER_COMPUTE:
  85                         tex_options.saturate_s = key->fsaturate_s;
  86                         tex_options.saturate_t = key->fsaturate_t;
  87                         tex_options.saturate_r = key->fsaturate_r;
  88                         break;
  89                 case SHADER_VERTEX:
  90                         tex_options.saturate_s = key->vsaturate_s;
  91                         tex_options.saturate_t = key->vsaturate_t;
  92                         tex_options.saturate_r = key->vsaturate_r;
  93                         break;
  94                 }
  95         }
  96
  97         if (shader->compiler->gpu_id >= 400) {
  98                 /* a4xx seems to have *no* sam.p */
  99                 tex_options.lower_txp = ~0;  /* lower all txp */
 100         } else {
 101                 /* a3xx just needs to avoid sam.p for 3d tex */
 102                 tex_options.lower_txp = (1 << GLSL_SAMPLER_DIM_3D);
 103         }
 104
 105         if (fd_mesa_debug & FD_DBG_DISASM) {
 106                 debug_printf("----------------------\n");
 107                 nir_print_shader(s, stdout);
 108                 debug_printf("----------------------\n");
 109         }
 110
 111         OPT_V(s, nir_opt_global_to_local);
 112         OPT_V(s, nir_convert_to_ssa);
 113
 114         if (key) {
 115                 if (s->stage == MESA_SHADER_VERTEX) {
 116                         OPT_V(s, nir_lower_clip_vs, key->ucp_enables);
 117                 } else if (s->stage == MESA_SHADER_FRAGMENT) {
 118                         OPT_V(s, nir_lower_clip_fs, key->ucp_enables);
 119                 }
 120                 if (key->color_two_side) {
 121                         OPT_V(s, nir_lower_two_sided_color);
 122                 }
 123         } else {
 124                 /* only want to do this the first time (when key is null)
 125                  * and not again on any potential 2nd variant lowering pass:
 126                  */
 127                 OPT_V(s, ir3_nir_apply_trig_workarounds);
 128         }
 129
 130         OPT_V(s, nir_lower_tex, &tex_options);
 131         OPT_V(s, nir_lower_idiv);
 132         OPT_V(s, nir_lower_load_const_to_scalar);
 133
 134         do {
 135                 progress = false;
 136
 137                 OPT_V(s, nir_lower_vars_to_ssa);
 138                 OPT_V(s, nir_lower_alu_to_scalar);
 139                 OPT_V(s, nir_lower_phis_to_scalar);
 140
 141                 progress |= OPT(s, nir_copy_prop);
 142                 progress |= OPT(s, nir_opt_dce);
 143                 progress |= OPT(s, nir_opt_cse);
 144                 progress |= OPT(s, ir3_nir_lower_if_else);
 145                 progress |= OPT(s, nir_opt_algebraic);
 146                 progress |= OPT(s, nir_opt_constant_folding);
 147
 148         } while (progress);
 149
 150         OPT_V(s, nir_remove_dead_variables, nir_var_local);
 151
 152         if (fd_mesa_debug & FD_DBG_DISASM) {
 153                 debug_printf("----------------------\n");
 154                 nir_print_shader(s, stdout);
 155                 debug_printf("----------------------\n");
 156         }
 157
 158         nir_sweep(s);
 159
 160         return s;
 161 }