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_fmod = true,
  47    .lower_fsqrt = true,
  48    .lower_sub = true,
  49    .lower_flrp32 = true,
  50    .lower_flrp64 = true,
  51    .lower_ftrunc = true,
  52    /* could be implemented by clamp */
  53    .lower_fsat = true,
  54    .lower_bitops = true,
  55    .lower_rotate = true,
  56    .lower_sincos = true,
  57 };
  58
  59 static const nir_shader_compiler_options fs_nir_options = {
  60    .lower_ffma = true,
  61    .lower_fpow = true,
  62    .lower_fdiv = true,
  63    .lower_fmod = true,
  64    .lower_sub = true,
  65    .lower_flrp32 = true,
  66    .lower_flrp64 = true,
  67    .lower_fsign = true,
  68    .lower_rotate = true,
  69    .lower_fdot = true,
  70    .lower_bitops = true,
  71 };
  72
  73 static const struct nir_lower_tex_options tex_options = {
  74    .lower_txp = ~0u,
  75 };
  76
  77 const void *
  78 lima_program_get_compiler_options(enum pipe_shader_type shader)
  79 {
  80    switch (shader) {
  81    case PIPE_SHADER_VERTEX:
  82       return &vs_nir_options;
  83    case PIPE_SHADER_FRAGMENT:
  84       return &fs_nir_options;
  85    default:
  86       return NULL;
  87    }
  88 }
  89
  90 static int
  91 type_size(const struct glsl_type *type, bool bindless)
  92 {
  93    return glsl_count_attribute_slots(type, false);
  94 }
  95
  96 void
  97 lima_program_optimize_vs_nir(struct nir_shader *s)
  98 {
  99    bool progress;
 100
 101    NIR_PASS_V(s, nir_lower_viewport_transform);
 102    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 103    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
 104    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
 105    NIR_PASS_V(s, nir_lower_io_to_scalar,
 106               nir_var_shader_in|nir_var_shader_out);
 107
 108    do {
 109       progress = false;
 110
 111       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 112       NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL);
 113       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 114       NIR_PASS(progress, s, nir_copy_prop);
 115       NIR_PASS(progress, s, nir_opt_remove_phis);
 116       NIR_PASS(progress, s, nir_opt_dce);
 117       NIR_PASS(progress, s, nir_opt_dead_cf);
 118       NIR_PASS(progress, s, nir_opt_cse);
 119       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 120       NIR_PASS(progress, s, nir_opt_algebraic);
 121       NIR_PASS(progress, s, nir_opt_constant_folding);
 122       NIR_PASS(progress, s, nir_opt_undef);
 123       NIR_PASS(progress, s, nir_opt_loop_unroll,
 124                nir_var_shader_in |
 125                nir_var_shader_out |
 126                nir_var_function_temp);
 127    } while (progress);
 128
 129    NIR_PASS_V(s, nir_lower_int_to_float);
 130    NIR_PASS_V(s, nir_lower_bool_to_float);
 131    NIR_PASS_V(s, nir_copy_prop);
 132    NIR_PASS_V(s, nir_opt_dce);
 133    NIR_PASS_V(s, nir_lower_locals_to_regs);
 134    NIR_PASS_V(s, nir_convert_from_ssa, true);
 135    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 136    nir_sweep(s);
 137 }
 138
 139 void
 140 lima_program_optimize_fs_nir(struct nir_shader *s)
 141 {
 142    BITSET_DECLARE(alu_lower, nir_num_opcodes) = {0};
 143    bool progress;
 144
 145    BITSET_SET(alu_lower, nir_op_frcp);
 146    BITSET_SET(alu_lower, nir_op_frsq);
 147    BITSET_SET(alu_lower, nir_op_flog2);
 148    BITSET_SET(alu_lower, nir_op_fexp2);
 149    BITSET_SET(alu_lower, nir_op_fsqrt);
 150    BITSET_SET(alu_lower, nir_op_fsin);
 151    BITSET_SET(alu_lower, nir_op_fcos);
 152
 153    NIR_PASS_V(s, nir_lower_fragcoord_wtrans);
 154    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 155    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 156    NIR_PASS_V(s, nir_lower_tex, &tex_options);
 157
 158    do {
 159       progress = false;
 160
 161       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 162       NIR_PASS(progress, s, nir_lower_alu_to_scalar, alu_lower);
 163       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 164       NIR_PASS(progress, s, nir_copy_prop);
 165       NIR_PASS(progress, s, nir_opt_remove_phis);
 166       NIR_PASS(progress, s, nir_opt_dce);
 167       NIR_PASS(progress, s, nir_opt_dead_cf);
 168       NIR_PASS(progress, s, nir_opt_cse);
 169       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 170       NIR_PASS(progress, s, nir_opt_algebraic);
 171       NIR_PASS(progress, s, nir_opt_constant_folding);
 172       NIR_PASS(progress, s, nir_opt_undef);
 173       NIR_PASS(progress, s, nir_opt_loop_unroll,
 174                nir_var_shader_in |
 175                nir_var_shader_out |
 176                nir_var_function_temp);
 177    } while (progress);
 178
 179    NIR_PASS_V(s, nir_lower_int_to_float);
 180    NIR_PASS_V(s, nir_lower_bool_to_float);
 181
 182    /* Lower modifiers */
 183    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 184    NIR_PASS_V(s, nir_copy_prop);
 185    NIR_PASS_V(s, nir_opt_dce);
 186
 187    NIR_PASS_V(s, nir_lower_locals_to_regs);
 188    NIR_PASS_V(s, nir_convert_from_ssa, true);
 189    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 190
 191    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 192    NIR_PASS_V(s, nir_lower_vec_to_movs);
 193
 194    nir_sweep(s);
 195 }
 196
 197 static void *
 198 lima_create_fs_state(struct pipe_context *pctx,
 199                      const struct pipe_shader_state *cso)
 200 {
 201    struct lima_screen *screen = lima_screen(pctx->screen);
 202    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 203
 204    if (!so)
 205       return NULL;
 206
 207    nir_shader *nir;
 208    if (cso->type == PIPE_SHADER_IR_NIR)
 209       nir = cso->ir.nir;
 210    else {
 211       assert(cso->type == PIPE_SHADER_IR_TGSI);
 212
 213       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 214    }
 215
 216    lima_program_optimize_fs_nir(nir);
 217
 218    if (lima_debug & LIMA_DEBUG_PP)
 219       nir_print_shader(nir, stdout);
 220
 221    if (!ppir_compile_nir(so, nir, screen->pp_ra)) {
 222       ralloc_free(so);
 223       return NULL;
 224    }
 225
 226    return so;
 227 }
 228
 229 static void
 230 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 231 {
 232    struct lima_context *ctx = lima_context(pctx);
 233
 234    ctx->fs = hwcso;
 235    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 236 }
 237
 238 static void
 239 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 240 {
 241    struct lima_fs_shader_state *so = hwcso;
 242
 243    if (so->bo)
 244       lima_bo_free(so->bo);
 245
 246    ralloc_free(so);
 247 }
 248
 249 bool
 250 lima_update_vs_state(struct lima_context *ctx)
 251 {
 252    struct lima_vs_shader_state *vs = ctx->vs;
 253    if (!vs->bo) {
 254       struct lima_screen *screen = lima_screen(ctx->base.screen);
 255       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 256       if (!vs->bo) {
 257          fprintf(stderr, "lima: create vs shader bo fail\n");
 258          return false;
 259       }
 260
 261       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 262       ralloc_free(vs->shader);
 263       vs->shader = NULL;
 264    }
 265
 266    return true;
 267 }
 268
 269 bool
 270 lima_update_fs_state(struct lima_context *ctx)
 271 {
 272    struct lima_fs_shader_state *fs = ctx->fs;
 273    if (!fs->bo) {
 274       struct lima_screen *screen = lima_screen(ctx->base.screen);
 275       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 276       if (!fs->bo) {
 277          fprintf(stderr, "lima: create fs shader bo fail\n");
 278          return false;
 279       }
 280
 281       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 282       ralloc_free(fs->shader);
 283       fs->shader = NULL;
 284    }
 285
 286    return true;
 287 }
 288
 289 static void *
 290 lima_create_vs_state(struct pipe_context *pctx,
 291                      const struct pipe_shader_state *cso)
 292 {
 293    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 294
 295    if (!so)
 296       return NULL;
 297
 298    nir_shader *nir;
 299    if (cso->type == PIPE_SHADER_IR_NIR)
 300       nir = cso->ir.nir;
 301    else {
 302       assert(cso->type == PIPE_SHADER_IR_TGSI);
 303
 304       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 305    }
 306
 307    lima_program_optimize_vs_nir(nir);
 308
 309    if (lima_debug & LIMA_DEBUG_GP)
 310       nir_print_shader(nir, stdout);
 311
 312    if (!gpir_compile_nir(so, nir)) {
 313       ralloc_free(so);
 314       return NULL;
 315    }
 316
 317    return so;
 318 }
 319
 320 static void
 321 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 322 {
 323    struct lima_context *ctx = lima_context(pctx);
 324
 325    ctx->vs = hwcso;
 326    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 327 }
 328
 329 static void
 330 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 331 {
 332    struct lima_vs_shader_state *so = hwcso;
 333
 334    if (so->bo)
 335       lima_bo_free(so->bo);
 336
 337    ralloc_free(so);
 338 }
 339
 340 void
 341 lima_program_init(struct lima_context *ctx)
 342 {
 343    ctx->base.create_fs_state = lima_create_fs_state;
 344    ctx->base.bind_fs_state = lima_bind_fs_state;
 345    ctx->base.delete_fs_state = lima_delete_fs_state;
 346
 347    ctx->base.create_vs_state = lima_create_vs_state;
 348    ctx->base.bind_vs_state = lima_bind_vs_state;
 349    ctx->base.delete_vs_state = lima_delete_vs_state;
 350 }