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