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