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_flrp32 = true,
  46                         .lower_flrp64 = true,
  47                         .lower_ffract = true,
  48                         .fuse_ffma = true,
  49                         .native_integers = true,
  50                         .vertex_id_zero_based = true,
  51                         .lower_extract_byte = true,
  52                         .lower_extract_word = true,
  53         };
  54         return tgsi_to_nir(tokens, &options);
  55 }
  56
  57 /* for given shader key, are any steps handled in nir? */
  58 bool
  59 ir3_key_lowers_nir(const struct ir3_shader_key *key)
  60 {
  61         return key->fsaturate_s | key->fsaturate_t | key->fsaturate_r |
  62                         key->vsaturate_s | key->vsaturate_t | key->vsaturate_r |
  63                         key->ucp_enables | key->color_two_side |
  64                         key->fclamp_color | key->vclamp_color;
  65 }
  66
  67 #define OPT(nir, pass, ...) ({                             \
  68    bool this_progress = false;                             \
  69    NIR_PASS(this_progress, nir, pass, ##__VA_ARGS__);      \
  70    this_progress;                                          \
  71 })
  72
  73 #define OPT_V(nir, pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
  74
  75 struct nir_shader *
  76 ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s,
  77                 const struct ir3_shader_key *key)
  78 {
  79         struct nir_lower_tex_options tex_options = {
  80                         .lower_rect = 0,
  81         };
  82         bool progress;
  83
  84         if (key) {
  85                 switch (shader->type) {
  86                 case SHADER_FRAGMENT:
  87                 case SHADER_COMPUTE:
  88                         tex_options.saturate_s = key->fsaturate_s;
  89                         tex_options.saturate_t = key->fsaturate_t;
  90                         tex_options.saturate_r = key->fsaturate_r;
  91                         break;
  92                 case SHADER_VERTEX:
  93                         tex_options.saturate_s = key->vsaturate_s;
  94                         tex_options.saturate_t = key->vsaturate_t;
  95                         tex_options.saturate_r = key->vsaturate_r;
  96                         break;
  97                 }
  98         }
  99
 100         if (shader->compiler->gpu_id >= 400) {
 101                 /* a4xx seems to have *no* sam.p */
 102                 tex_options.lower_txp = ~0;  /* lower all txp */
 103         } else {
 104                 /* a3xx just needs to avoid sam.p for 3d tex */
 105                 tex_options.lower_txp = (1 << GLSL_SAMPLER_DIM_3D);
 106         }
 107
 108         if (fd_mesa_debug & FD_DBG_DISASM) {
 109                 debug_printf("----------------------\n");
 110                 nir_print_shader(s, stdout);
 111                 debug_printf("----------------------\n");
 112         }
 113
 114         OPT_V(s, nir_opt_global_to_local);
 115         OPT_V(s, nir_convert_to_ssa);
 116
 117         if (key) {
 118                 if (s->stage == MESA_SHADER_VERTEX) {
 119                         OPT_V(s, nir_lower_clip_vs, key->ucp_enables);
 120                         if (key->vclamp_color)
 121                                 OPT_V(s, nir_lower_clamp_color_outputs);
 122                 } else if (s->stage == MESA_SHADER_FRAGMENT) {
 123                         OPT_V(s, nir_lower_clip_fs, key->ucp_enables);
 124                         if (key->fclamp_color)
 125                                 OPT_V(s, nir_lower_clamp_color_outputs);
 126                 }
 127                 if (key->color_two_side) {
 128                         OPT_V(s, nir_lower_two_sided_color);
 129                 }
 130         } else {
 131                 /* only want to do this the first time (when key is null)
 132                  * and not again on any potential 2nd variant lowering pass:
 133                  */
 134                 OPT_V(s, ir3_nir_apply_trig_workarounds);
 135         }
 136
 137         OPT_V(s, nir_lower_tex, &tex_options);
 138         OPT_V(s, nir_lower_idiv);
 139         OPT_V(s, nir_lower_load_const_to_scalar);
 140
 141         do {
 142                 progress = false;
 143
 144                 OPT_V(s, nir_lower_vars_to_ssa);
 145                 OPT_V(s, nir_lower_alu_to_scalar);
 146                 OPT_V(s, nir_lower_phis_to_scalar);
 147
 148                 progress |= OPT(s, nir_copy_prop);
 149                 progress |= OPT(s, nir_opt_dce);
 150                 progress |= OPT(s, nir_opt_cse);
 151                 progress |= OPT(s, ir3_nir_lower_if_else);
 152                 progress |= OPT(s, nir_opt_algebraic);
 153                 progress |= OPT(s, nir_opt_constant_folding);
 154
 155         } while (progress);
 156
 157         OPT_V(s, nir_remove_dead_variables, nir_var_local);
 158
 159         if (fd_mesa_debug & FD_DBG_DISASM) {
 160                 debug_printf("----------------------\n");
 161                 nir_print_shader(s, stdout);
 162                 debug_printf("----------------------\n");
 163         }
 164
 165         nir_sweep(s);
 166
 167         return s;
 168 }