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    .lower_vector_cmp = true,
  72 };
  73
  74 static const struct nir_lower_tex_options tex_options = {
  75    .lower_txp = ~0u,
  76 };
  77
  78 const void *
  79 lima_program_get_compiler_options(enum pipe_shader_type shader)
  80 {
  81    switch (shader) {
  82    case PIPE_SHADER_VERTEX:
  83       return &vs_nir_options;
  84    case PIPE_SHADER_FRAGMENT:
  85       return &fs_nir_options;
  86    default:
  87       return NULL;
  88    }
  89 }
  90
  91 static int
  92 type_size(const struct glsl_type *type, bool bindless)
  93 {
  94    return glsl_count_attribute_slots(type, false);
  95 }
  96
  97 void
  98 lima_program_optimize_vs_nir(struct nir_shader *s)
  99 {
 100    bool progress;
 101
 102    NIR_PASS_V(s, nir_lower_viewport_transform);
 103    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 104    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
 105    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
 106    NIR_PASS_V(s, nir_lower_io_to_scalar,
 107               nir_var_shader_in|nir_var_shader_out);
 108
 109    do {
 110       progress = false;
 111
 112       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 113       NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL, NULL);
 114       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 115       NIR_PASS(progress, s, nir_copy_prop);
 116       NIR_PASS(progress, s, nir_opt_remove_phis);
 117       NIR_PASS(progress, s, nir_opt_dce);
 118       NIR_PASS(progress, s, nir_opt_dead_cf);
 119       NIR_PASS(progress, s, nir_opt_cse);
 120       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 121       NIR_PASS(progress, s, nir_opt_algebraic);
 122       NIR_PASS(progress, s, nir_opt_constant_folding);
 123       NIR_PASS(progress, s, nir_opt_undef);
 124       NIR_PASS(progress, s, nir_opt_loop_unroll,
 125                nir_var_shader_in |
 126                nir_var_shader_out |
 127                nir_var_function_temp);
 128    } while (progress);
 129
 130    NIR_PASS_V(s, nir_lower_int_to_float);
 131    NIR_PASS_V(s, nir_lower_bool_to_float);
 132
 133    /* Some ops must be lowered after being converted from int ops,
 134     * so re-run nir_opt_algebraic after int lowering. */
 135    do {
 136       progress = false;
 137       NIR_PASS(progress, s, nir_opt_algebraic);
 138    } while (progress);
 139
 140    NIR_PASS_V(s, nir_copy_prop);
 141    NIR_PASS_V(s, nir_opt_dce);
 142    NIR_PASS_V(s, nir_lower_locals_to_regs);
 143    NIR_PASS_V(s, nir_convert_from_ssa, true);
 144    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 145    nir_sweep(s);
 146 }
 147
 148 static bool
 149 lima_alu_to_scalar_filter_cb(const nir_instr *instr, const void *data)
 150 {
 151    if (instr->type != nir_instr_type_alu)
 152       return false;
 153
 154    nir_alu_instr *alu = nir_instr_as_alu(instr);
 155    switch (alu->op) {
 156    case nir_op_frcp:
 157    case nir_op_frsq:
 158    case nir_op_flog2:
 159    case nir_op_fexp2:
 160    case nir_op_fsqrt:
 161    case nir_op_fsin:
 162    case nir_op_fcos:
 163    /* nir vec4 fcsel assumes that each component of the condition will be
 164     * used to select the same component from the two options, but lima
 165     * can't implement that since we only have 1 component condition */
 166    case nir_op_fcsel:
 167    case nir_op_bcsel:
 168       return true;
 169    default:
 170       break;
 171    }
 172
 173    return false;
 174 }
 175
 176 void
 177 lima_program_optimize_fs_nir(struct nir_shader *s)
 178 {
 179    bool progress;
 180
 181    NIR_PASS_V(s, nir_lower_fragcoord_wtrans);
 182    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 183    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 184    NIR_PASS_V(s, nir_lower_tex, &tex_options);
 185
 186    do {
 187       progress = false;
 188       NIR_PASS(progress, s, nir_opt_vectorize);
 189    } while (progress);
 190
 191    do {
 192       progress = false;
 193
 194       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 195       NIR_PASS(progress, s, nir_lower_alu_to_scalar, lima_alu_to_scalar_filter_cb, NULL);
 196       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 197       NIR_PASS(progress, s, nir_copy_prop);
 198       NIR_PASS(progress, s, nir_opt_remove_phis);
 199       NIR_PASS(progress, s, nir_opt_dce);
 200       NIR_PASS(progress, s, nir_opt_dead_cf);
 201       NIR_PASS(progress, s, nir_opt_cse);
 202       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 203       NIR_PASS(progress, s, nir_opt_algebraic);
 204       NIR_PASS(progress, s, nir_opt_constant_folding);
 205       NIR_PASS(progress, s, nir_opt_undef);
 206       NIR_PASS(progress, s, nir_opt_loop_unroll,
 207                nir_var_shader_in |
 208                nir_var_shader_out |
 209                nir_var_function_temp);
 210    } while (progress);
 211
 212    NIR_PASS_V(s, nir_lower_int_to_float);
 213    NIR_PASS_V(s, nir_lower_bool_to_float);
 214
 215    /* Some ops must be lowered after being converted from int ops,
 216     * so re-run nir_opt_algebraic after int lowering. */
 217    do {
 218       progress = false;
 219       NIR_PASS(progress, s, nir_opt_algebraic);
 220    } while (progress);
 221
 222    /* Must be run after optimization loop */
 223    NIR_PASS_V(s, lima_nir_scale_trig);
 224
 225    /* Lower modifiers */
 226    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 227    NIR_PASS_V(s, nir_copy_prop);
 228    NIR_PASS_V(s, nir_opt_dce);
 229
 230    NIR_PASS_V(s, nir_lower_locals_to_regs);
 231    NIR_PASS_V(s, nir_convert_from_ssa, true);
 232    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 233
 234    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 235    NIR_PASS_V(s, nir_lower_vec_to_movs);
 236
 237    nir_sweep(s);
 238 }
 239
 240 static void *
 241 lima_create_fs_state(struct pipe_context *pctx,
 242                      const struct pipe_shader_state *cso)
 243 {
 244    struct lima_context *ctx = lima_context(pctx);
 245    struct lima_screen *screen = lima_screen(pctx->screen);
 246    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 247
 248    if (!so)
 249       return NULL;
 250
 251    nir_shader *nir;
 252    if (cso->type == PIPE_SHADER_IR_NIR)
 253       nir = cso->ir.nir;
 254    else {
 255       assert(cso->type == PIPE_SHADER_IR_TGSI);
 256
 257       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 258    }
 259
 260    lima_program_optimize_fs_nir(nir);
 261
 262    if (lima_debug & LIMA_DEBUG_PP)
 263       nir_print_shader(nir, stdout);
 264
 265    if (!ppir_compile_nir(so, nir, screen->pp_ra, &ctx->debug)) {
 266       ralloc_free(so);
 267       return NULL;
 268    }
 269
 270    return so;
 271 }
 272
 273 static void
 274 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 275 {
 276    struct lima_context *ctx = lima_context(pctx);
 277
 278    ctx->fs = hwcso;
 279    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 280 }
 281
 282 static void
 283 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 284 {
 285    struct lima_fs_shader_state *so = hwcso;
 286
 287    if (so->bo)
 288       lima_bo_free(so->bo);
 289
 290    ralloc_free(so);
 291 }
 292
 293 bool
 294 lima_update_vs_state(struct lima_context *ctx)
 295 {
 296    struct lima_vs_shader_state *vs = ctx->vs;
 297    if (!vs->bo) {
 298       struct lima_screen *screen = lima_screen(ctx->base.screen);
 299       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 300       if (!vs->bo) {
 301          fprintf(stderr, "lima: create vs shader bo fail\n");
 302          return false;
 303       }
 304
 305       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 306       ralloc_free(vs->shader);
 307       vs->shader = NULL;
 308    }
 309
 310    return true;
 311 }
 312
 313 bool
 314 lima_update_fs_state(struct lima_context *ctx)
 315 {
 316    struct lima_fs_shader_state *fs = ctx->fs;
 317    if (!fs->bo) {
 318       struct lima_screen *screen = lima_screen(ctx->base.screen);
 319       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 320       if (!fs->bo) {
 321          fprintf(stderr, "lima: create fs shader bo fail\n");
 322          return false;
 323       }
 324
 325       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 326       ralloc_free(fs->shader);
 327       fs->shader = NULL;
 328    }
 329
 330    ctx->pp_max_stack_size = MAX2(ctx->pp_max_stack_size, ctx->fs->stack_size);
 331
 332    return true;
 333 }
 334
 335 static void *
 336 lima_create_vs_state(struct pipe_context *pctx,
 337                      const struct pipe_shader_state *cso)
 338 {
 339    struct lima_context *ctx = lima_context(pctx);
 340    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 341
 342    if (!so)
 343       return NULL;
 344
 345    nir_shader *nir;
 346    if (cso->type == PIPE_SHADER_IR_NIR)
 347       nir = cso->ir.nir;
 348    else {
 349       assert(cso->type == PIPE_SHADER_IR_TGSI);
 350
 351       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 352    }
 353
 354    lima_program_optimize_vs_nir(nir);
 355
 356    if (lima_debug & LIMA_DEBUG_GP)
 357       nir_print_shader(nir, stdout);
 358
 359    if (!gpir_compile_nir(so, nir, &ctx->debug)) {
 360       ralloc_free(so);
 361       return NULL;
 362    }
 363
 364    return so;
 365 }
 366
 367 static void
 368 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 369 {
 370    struct lima_context *ctx = lima_context(pctx);
 371
 372    ctx->vs = hwcso;
 373    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 374 }
 375
 376 static void
 377 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 378 {
 379    struct lima_vs_shader_state *so = hwcso;
 380
 381    if (so->bo)
 382       lima_bo_free(so->bo);
 383
 384    ralloc_free(so);
 385 }
 386
 387 void
 388 lima_program_init(struct lima_context *ctx)
 389 {
 390    ctx->base.create_fs_state = lima_create_fs_state;
 391    ctx->base.bind_fs_state = lima_bind_fs_state;
 392    ctx->base.delete_fs_state = lima_delete_fs_state;
 393
 394    ctx->base.create_vs_state = lima_create_vs_state;
 395    ctx->base.bind_vs_state = lima_bind_vs_state;
 396    ctx->base.delete_vs_state = lima_delete_vs_state;
 397 }