src/gallium/drivers/lima/lima_program.c

   1 /*
   2  * Copyright (c) 2017-2019 Lima Project
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sub license,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the
  12  * next paragraph) shall be included in all copies or substantial portions
  13  * of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  *
  23  */
  24
  25 #include "util/u_memory.h"
  26 #include "util/ralloc.h"
  27 #include "util/u_debug.h"
  28
  29 #include "tgsi/tgsi_dump.h"
  30 #include "compiler/nir/nir.h"
  31 #include "nir/tgsi_to_nir.h"
  32
  33 #include "pipe/p_state.h"
  34
  35 #include "lima_screen.h"
  36 #include "lima_context.h"
  37 #include "lima_program.h"
  38 #include "lima_bo.h"
  39 #include "ir/lima_ir.h"
  40
  41 static const nir_shader_compiler_options vs_nir_options = {
  42    .lower_ffma = true,
  43    .lower_fpow = true,
  44    .lower_ffract = true,
  45    .lower_fdiv = true,
  46    .lower_fsqrt = true,
  47    .lower_sub = true,
  48    .lower_flrp32 = true,
  49    .lower_flrp64 = true,
  50    /* could be implemented by clamp */
  51    .lower_fsat = true,
  52 };
  53
  54 static const nir_shader_compiler_options fs_nir_options = {
  55    .lower_fpow = true,
  56    .lower_fdiv = true,
  57    .lower_sub = true,
  58    .lower_flrp32 = true,
  59    .lower_flrp64 = true,
  60 };
  61
  62 const void *
  63 lima_program_get_compiler_options(enum pipe_shader_type shader)
  64 {
  65    switch (shader) {
  66    case PIPE_SHADER_VERTEX:
  67       return &vs_nir_options;
  68    case PIPE_SHADER_FRAGMENT:
  69       return &fs_nir_options;
  70    default:
  71       return NULL;
  72    }
  73 }
  74
  75 static int
  76 type_size(const struct glsl_type *type)
  77 {
  78    return glsl_count_attribute_slots(type, false);
  79 }
  80
  81 static void
  82 lima_program_optimize_vs_nir(struct nir_shader *s)
  83 {
  84    bool progress;
  85
  86    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
  87    NIR_PASS_V(s, nir_lower_regs_to_ssa);
  88    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
  89    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
  90    NIR_PASS_V(s, nir_lower_io_to_scalar,
  91               nir_var_shader_in|nir_var_shader_out);
  92    NIR_PASS_V(s, nir_lower_bool_to_float);
  93
  94    do {
  95       progress = false;
  96
  97       NIR_PASS_V(s, nir_lower_vars_to_ssa);
  98       NIR_PASS(progress, s, nir_lower_alu_to_scalar);
  99       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 100       NIR_PASS(progress, s, nir_copy_prop);
 101       NIR_PASS(progress, s, nir_opt_remove_phis);
 102       NIR_PASS(progress, s, nir_opt_dce);
 103       NIR_PASS(progress, s, nir_opt_dead_cf);
 104       NIR_PASS(progress, s, nir_opt_cse);
 105       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 106       NIR_PASS(progress, s, nir_opt_algebraic);
 107       NIR_PASS(progress, s, nir_opt_constant_folding);
 108       NIR_PASS(progress, s, nir_opt_undef);
 109       NIR_PASS(progress, s, nir_opt_loop_unroll,
 110                nir_var_shader_in |
 111                nir_var_shader_out |
 112                nir_var_function_temp);
 113    } while (progress);
 114
 115    NIR_PASS_V(s, nir_lower_locals_to_regs);
 116    NIR_PASS_V(s, nir_convert_from_ssa, true);
 117    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 118    nir_sweep(s);
 119 }
 120
 121 static void
 122 lima_program_optimize_fs_nir(struct nir_shader *s)
 123 {
 124    bool progress;
 125
 126    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 127    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 128    NIR_PASS_V(s, nir_lower_bool_to_float);
 129
 130    do {
 131       progress = false;
 132
 133       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 134       //NIR_PASS(progress, s, nir_lower_alu_to_scalar);
 135       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 136       NIR_PASS(progress, s, nir_copy_prop);
 137       NIR_PASS(progress, s, nir_opt_remove_phis);
 138       NIR_PASS(progress, s, nir_opt_dce);
 139       NIR_PASS(progress, s, nir_opt_dead_cf);
 140       NIR_PASS(progress, s, nir_opt_cse);
 141       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 142       NIR_PASS(progress, s, nir_opt_algebraic);
 143       NIR_PASS(progress, s, nir_opt_constant_folding);
 144       NIR_PASS(progress, s, nir_opt_undef);
 145       NIR_PASS(progress, s, nir_opt_loop_unroll,
 146                nir_var_shader_in |
 147                nir_var_shader_out |
 148                nir_var_function_temp);
 149    } while (progress);
 150
 151    /* Lower modifiers */
 152    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 153    NIR_PASS_V(s, nir_copy_prop);
 154    NIR_PASS_V(s, nir_opt_dce);
 155
 156    NIR_PASS_V(s, nir_lower_locals_to_regs);
 157    NIR_PASS_V(s, nir_convert_from_ssa, true);
 158    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 159
 160    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 161    NIR_PASS_V(s, nir_lower_vec_to_movs);
 162
 163    nir_sweep(s);
 164 }
 165
 166 static void *
 167 lima_create_fs_state(struct pipe_context *pctx,
 168                      const struct pipe_shader_state *cso)
 169 {
 170    struct lima_screen *screen = lima_screen(pctx->screen);
 171    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 172
 173    if (!so)
 174       return NULL;
 175
 176    nir_shader *nir;
 177    if (cso->type == PIPE_SHADER_IR_NIR)
 178       nir = cso->ir.nir;
 179    else {
 180       assert(cso->type == PIPE_SHADER_IR_TGSI);
 181
 182       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 183    }
 184
 185    lima_program_optimize_fs_nir(nir);
 186
 187    if (lima_debug & LIMA_DEBUG_PP)
 188       nir_print_shader(nir, stdout);
 189
 190    if (!ppir_compile_nir(so, nir, screen->pp_ra)) {
 191       ralloc_free(so);
 192       return NULL;
 193    }
 194
 195    return so;
 196 }
 197
 198 static void
 199 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 200 {
 201    struct lima_context *ctx = lima_context(pctx);
 202
 203    ctx->fs = hwcso;
 204    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 205 }
 206
 207 static void
 208 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 209 {
 210    struct lima_fs_shader_state *so = hwcso;
 211
 212    if (so->bo)
 213       lima_bo_free(so->bo);
 214
 215    ralloc_free(so);
 216 }
 217
 218 bool
 219 lima_update_vs_state(struct lima_context *ctx)
 220 {
 221    struct lima_vs_shader_state *vs = ctx->vs;
 222    if (!vs->bo) {
 223       struct lima_screen *screen = lima_screen(ctx->base.screen);
 224       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 225       if (!vs->bo) {
 226          fprintf(stderr, "lima: create vs shader bo fail\n");
 227          return false;
 228       }
 229
 230       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 231       ralloc_free(vs->shader);
 232       vs->shader = NULL;
 233    }
 234
 235    return true;
 236 }
 237
 238 bool
 239 lima_update_fs_state(struct lima_context *ctx)
 240 {
 241    struct lima_fs_shader_state *fs = ctx->fs;
 242    if (!fs->bo) {
 243       struct lima_screen *screen = lima_screen(ctx->base.screen);
 244       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 245       if (!fs->bo) {
 246          fprintf(stderr, "lima: create fs shader bo fail\n");
 247          return false;
 248       }
 249
 250       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 251       ralloc_free(fs->shader);
 252       fs->shader = NULL;
 253    }
 254
 255    return true;
 256 }
 257
 258 static void *
 259 lima_create_vs_state(struct pipe_context *pctx,
 260                      const struct pipe_shader_state *cso)
 261 {
 262    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 263
 264    if (!so)
 265       return NULL;
 266
 267    nir_shader *nir;
 268    if (cso->type == PIPE_SHADER_IR_NIR)
 269       nir = cso->ir.nir;
 270    else {
 271       assert(cso->type == PIPE_SHADER_IR_TGSI);
 272
 273       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 274    }
 275
 276    lima_program_optimize_vs_nir(nir);
 277
 278    if (lima_debug & LIMA_DEBUG_GP)
 279       nir_print_shader(nir, stdout);
 280
 281    if (!gpir_compile_nir(so, nir)) {
 282       ralloc_free(so);
 283       return NULL;
 284    }
 285
 286    return so;
 287 }
 288
 289 static void
 290 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 291 {
 292    struct lima_context *ctx = lima_context(pctx);
 293
 294    ctx->vs = hwcso;
 295    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 296 }
 297
 298 static void
 299 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 300 {
 301    struct lima_vs_shader_state *so = hwcso;
 302
 303    if (so->bo)
 304       lima_bo_free(so->bo);
 305
 306    ralloc_free(so);
 307 }
 308
 309 void
 310 lima_program_init(struct lima_context *ctx)
 311 {
 312    ctx->base.create_fs_state = lima_create_fs_state;
 313    ctx->base.bind_fs_state = lima_bind_fs_state;
 314    ctx->base.delete_fs_state = lima_delete_fs_state;
 315
 316    ctx->base.create_vs_state = lima_create_vs_state;
 317    ctx->base.bind_vs_state = lima_bind_vs_state;
 318    ctx->base.delete_vs_state = lima_delete_vs_state;
 319 }