src/gallium/drivers/etnaviv/etnaviv_shader.c

   1 /*
   2  * Copyright (c) 2012-2015 Etnaviv 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  * Authors:
  24  *    Wladimir J. van der Laan <laanwj@gmail.com>
  25  */
  26
  27 #include "etnaviv_shader.h"
  28
  29 #include "etnaviv_compiler.h"
  30 #include "etnaviv_context.h"
  31 #include "etnaviv_debug.h"
  32 #include "etnaviv_screen.h"
  33 #include "etnaviv_util.h"
  34
  35 #include "tgsi/tgsi_parse.h"
  36 #include "nir/tgsi_to_nir.h"
  37 #include "util/u_math.h"
  38 #include "util/u_memory.h"
  39
  40 /* Upload shader code to bo, if not already done */
  41 static bool etna_icache_upload_shader(struct etna_context *ctx, struct etna_shader_variant *v)
  42 {
  43    if (v->bo)
  44       return true;
  45    v->bo = etna_bo_new(ctx->screen->dev, v->code_size*4, DRM_ETNA_GEM_CACHE_WC);
  46    if (!v->bo)
  47       return false;
  48
  49    void *buf = etna_bo_map(v->bo);
  50    etna_bo_cpu_prep(v->bo, DRM_ETNA_PREP_WRITE);
  51    memcpy(buf, v->code, v->code_size*4);
  52    etna_bo_cpu_fini(v->bo);
  53    DBG("Uploaded %s of %u words to bo %p", v->stage == MESA_SHADER_FRAGMENT ? "fs":"vs", v->code_size, v->bo);
  54    return true;
  55 }
  56
  57 /* Link vs and fs together: fill in shader_state from vs and fs
  58  * as this function is called every time a new fs or vs is bound, the goal is to
  59  * do little processing as possible here, and to precompute as much as possible in
  60  * the vs/fs shader_object.
  61  *
  62  * XXX we could cache the link result for a certain set of VS/PS; usually a pair
  63  * of VS and PS will be used together anyway.
  64  */
  65 static bool
  66 etna_link_shaders(struct etna_context *ctx, struct compiled_shader_state *cs,
  67                   struct etna_shader_variant *vs, struct etna_shader_variant *fs)
  68 {
  69    struct etna_shader_link_info link = { };
  70    bool failed;
  71
  72    assert(vs->stage == MESA_SHADER_VERTEX);
  73    assert(fs->stage == MESA_SHADER_FRAGMENT);
  74
  75 #ifdef DEBUG
  76    if (DBG_ENABLED(ETNA_DBG_DUMP_SHADERS)) {
  77       if (DBG_ENABLED(ETNA_DBG_NIR)) {
  78          etna_dump_shader_nir(vs);
  79          etna_dump_shader_nir(fs);
  80       } else {
  81          etna_dump_shader(vs);
  82          etna_dump_shader(fs);
  83       }
  84    }
  85 #endif
  86
  87    if (DBG_ENABLED(ETNA_DBG_NIR))
  88       failed = etna_link_shader_nir(&link, vs, fs);
  89    else
  90       failed = etna_link_shader(&link, vs, fs);
  91
  92    if (failed) {
  93       /* linking failed: some fs inputs do not have corresponding
  94        * vs outputs */
  95       assert(0);
  96
  97       return false;
  98    }
  99
 100    if (DBG_ENABLED(ETNA_DBG_LINKER_MSGS)) {
 101       debug_printf("link result:\n");
 102       debug_printf("  vs  -> fs  comps use     pa_attr\n");
 103
 104       for (int idx = 0; idx < link.num_varyings; ++idx)
 105          debug_printf("  t%-2u -> t%-2u %-5.*s %u,%u,%u,%u 0x%08x\n",
 106                       link.varyings[idx].reg, idx + 1,
 107                       link.varyings[idx].num_components, "xyzw",
 108                       link.varyings[idx].use[0], link.varyings[idx].use[1],
 109                       link.varyings[idx].use[2], link.varyings[idx].use[3],
 110                       link.varyings[idx].pa_attributes);
 111    }
 112
 113    /* set last_varying_2x flag if the last varying has 1 or 2 components */
 114    bool last_varying_2x = false;
 115    if (link.num_varyings > 0 && link.varyings[link.num_varyings - 1].num_components <= 2)
 116       last_varying_2x = true;
 117
 118    cs->RA_CONTROL = VIVS_RA_CONTROL_UNK0 |
 119                     COND(last_varying_2x, VIVS_RA_CONTROL_LAST_VARYING_2X);
 120
 121    cs->PA_ATTRIBUTE_ELEMENT_COUNT = VIVS_PA_ATTRIBUTE_ELEMENT_COUNT_COUNT(link.num_varyings);
 122    for (int idx = 0; idx < link.num_varyings; ++idx)
 123       cs->PA_SHADER_ATTRIBUTES[idx] = link.varyings[idx].pa_attributes;
 124
 125    cs->VS_END_PC = vs->code_size / 4;
 126    cs->VS_OUTPUT_COUNT = 1 + link.num_varyings; /* position + varyings */
 127
 128    /* vs outputs (varyings) */
 129    DEFINE_ETNA_BITARRAY(vs_output, 16, 8) = {0};
 130    int varid = 0;
 131    etna_bitarray_set(vs_output, 8, varid++, vs->vs_pos_out_reg);
 132    for (int idx = 0; idx < link.num_varyings; ++idx)
 133       etna_bitarray_set(vs_output, 8, varid++, link.varyings[idx].reg);
 134    if (vs->vs_pointsize_out_reg >= 0)
 135       etna_bitarray_set(vs_output, 8, varid++, vs->vs_pointsize_out_reg); /* pointsize is last */
 136
 137    for (int idx = 0; idx < ARRAY_SIZE(cs->VS_OUTPUT); ++idx)
 138       cs->VS_OUTPUT[idx] = vs_output[idx];
 139
 140    if (vs->vs_pointsize_out_reg != -1) {
 141       /* vertex shader outputs point coordinate, provide extra output and make
 142        * sure PA config is
 143        * not masked */
 144       cs->PA_CONFIG = ~0;
 145       cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT + 1;
 146    } else {
 147       /* vertex shader does not output point coordinate, make sure thate
 148        * POINT_SIZE_ENABLE is masked
 149        * and no extra output is given */
 150       cs->PA_CONFIG = ~VIVS_PA_CONFIG_POINT_SIZE_ENABLE;
 151       cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT;
 152    }
 153
 154    cs->VS_LOAD_BALANCING = vs->vs_load_balancing;
 155    cs->VS_START_PC = 0;
 156
 157    cs->PS_END_PC = fs->code_size / 4;
 158    cs->PS_OUTPUT_REG = fs->ps_color_out_reg;
 159    cs->PS_INPUT_COUNT =
 160       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */
 161       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 162    cs->PS_TEMP_REGISTER_CONTROL =
 163       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1));
 164    cs->PS_CONTROL = VIVS_PS_CONTROL_SATURATE_RT0; /* XXX when can we set BYPASS? */
 165    cs->PS_START_PC = 0;
 166
 167    /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA
 168     * mode, avoids some fumbling in sync_context. */
 169    cs->PS_INPUT_COUNT_MSAA =
 170       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */
 171       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 172    cs->PS_TEMP_REGISTER_CONTROL_MSAA =
 173       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2));
 174
 175    uint32_t total_components = 0;
 176    DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0};
 177    DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0};
 178    for (int idx = 0; idx < link.num_varyings; ++idx) {
 179       const struct etna_varying *varying = &link.varyings[idx];
 180
 181       etna_bitarray_set(num_components, 4, idx, varying->num_components);
 182       for (int comp = 0; comp < varying->num_components; ++comp) {
 183          etna_bitarray_set(component_use, 2, total_components, varying->use[comp]);
 184          total_components += 1;
 185       }
 186    }
 187
 188    cs->GL_VARYING_TOTAL_COMPONENTS =
 189       VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2));
 190    cs->GL_VARYING_NUM_COMPONENTS = num_components[0];
 191    cs->GL_VARYING_COMPONENT_USE[0] = component_use[0];
 192    cs->GL_VARYING_COMPONENT_USE[1] = component_use[1];
 193
 194    cs->GL_HALTI5_SH_SPECIALS =
 195       0x7f7f0000 | /* unknown bits, probably other PS inputs */
 196       /* pointsize is last (see above) */
 197       VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ?
 198                                               cs->VS_OUTPUT_COUNT * 4 : 0x00) |
 199       VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ?
 200                                               link.pcoord_varying_comp_ofs : 0x7f);
 201
 202    /* mask out early Z bit when frag depth is written */
 203    cs->PE_DEPTH_CONFIG = ~COND(fs->ps_depth_out_reg >= 0, VIVS_PE_DEPTH_CONFIG_EARLY_Z);
 204
 205    /* reference instruction memory */
 206    cs->vs_inst_mem_size = vs->code_size;
 207    cs->VS_INST_MEM = vs->code;
 208
 209    cs->ps_inst_mem_size = fs->code_size;
 210    cs->PS_INST_MEM = fs->code;
 211
 212    if (vs->needs_icache | fs->needs_icache) {
 213       /* If either of the shaders needs ICACHE, we use it for both. It is
 214        * either switched on or off for the entire shader processor.
 215        */
 216       if (!etna_icache_upload_shader(ctx, vs) ||
 217           !etna_icache_upload_shader(ctx, fs)) {
 218          assert(0);
 219          return false;
 220       }
 221
 222       cs->VS_INST_ADDR.bo = vs->bo;
 223       cs->VS_INST_ADDR.offset = 0;
 224       cs->VS_INST_ADDR.flags = ETNA_RELOC_READ;
 225       cs->PS_INST_ADDR.bo = fs->bo;
 226       cs->PS_INST_ADDR.offset = 0;
 227       cs->PS_INST_ADDR.flags = ETNA_RELOC_READ;
 228    } else {
 229       /* clear relocs */
 230       memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR));
 231       memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR));
 232    }
 233
 234    return true;
 235 }
 236
 237 bool
 238 etna_shader_link(struct etna_context *ctx)
 239 {
 240    if (!ctx->shader.vs || !ctx->shader.fs)
 241       return false;
 242
 243    /* re-link vs and fs if needed */
 244    return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs);
 245 }
 246
 247 static bool
 248 etna_shader_update_vs_inputs(struct compiled_shader_state *cs,
 249                              const struct etna_shader_variant *vs,
 250                              const struct compiled_vertex_elements_state *ves)
 251 {
 252    unsigned num_temps, cur_temp, num_vs_inputs;
 253
 254    if (!vs)
 255       return false;
 256
 257    /* Number of vertex elements determines number of VS inputs. Otherwise,
 258     * the GPU crashes. Allocate any unused vertex elements to VS temporary
 259     * registers. */
 260    num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg);
 261    if (num_vs_inputs != ves->num_elements) {
 262       BUG("Number of elements %u does not match the number of VS inputs %zu",
 263           ves->num_elements, vs->infile.num_reg);
 264       return false;
 265    }
 266
 267    cur_temp = vs->num_temps;
 268    num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp;
 269
 270    cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) |
 271                         VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8);
 272    cs->VS_TEMP_REGISTER_CONTROL =
 273       VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps);
 274
 275    /* vs inputs (attributes) */
 276    DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0};
 277    for (int idx = 0; idx < num_vs_inputs; ++idx) {
 278       if (idx < vs->infile.num_reg)
 279          etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg);
 280       else
 281          etna_bitarray_set(vs_input, 8, idx, cur_temp++);
 282    }
 283
 284    for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx)
 285       cs->VS_INPUT[idx] = vs_input[idx];
 286
 287    return true;
 288 }
 289
 290 static inline const char *
 291 etna_shader_stage(struct etna_shader_variant *shader)
 292 {
 293    switch (shader->stage) {
 294    case MESA_SHADER_VERTEX:     return "VERT";
 295    case MESA_SHADER_FRAGMENT:   return "FRAG";
 296    case MESA_SHADER_COMPUTE:    return "CL";
 297    default:
 298       unreachable("invalid type");
 299       return NULL;
 300    }
 301 }
 302
 303 static void
 304 dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debug)
 305 {
 306    if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB))
 307       return;
 308
 309    pipe_debug_message(debug, SHADER_INFO, "\n"
 310          "SHADER-DB: %s prog %d/%d: %u instructions %u temps\n"
 311          "SHADER-DB: %s prog %d/%d: %u immediates %u loops\n",
 312          etna_shader_stage(v),
 313          v->shader->id, v->id,
 314          v->code_size,
 315          v->num_temps,
 316          etna_shader_stage(v),
 317          v->shader->id, v->id,
 318          v->uniforms.imm_count,
 319          v->num_loops);
 320 }
 321
 322 bool
 323 etna_shader_update_vertex(struct etna_context *ctx)
 324 {
 325    return etna_shader_update_vs_inputs(&ctx->shader_state, ctx->shader.vs,
 326                                        ctx->vertex_elements);
 327 }
 328
 329 static struct etna_shader_variant *
 330 create_variant(struct etna_shader *shader, struct etna_shader_key key)
 331 {
 332    struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant);
 333    int ret;
 334
 335    if (!v)
 336       return NULL;
 337
 338    v->shader = shader;
 339    v->key = key;
 340
 341    ret = etna_compile_shader(v);
 342    if (!ret) {
 343       debug_error("compile failed!");
 344       goto fail;
 345    }
 346
 347    v->id = ++shader->variant_count;
 348
 349    return v;
 350
 351 fail:
 352    FREE(v);
 353    return NULL;
 354 }
 355
 356 struct etna_shader_variant *
 357 etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key,
 358                    struct pipe_debug_callback *debug)
 359 {
 360    struct etna_shader_variant *v;
 361
 362    for (v = shader->variants; v; v = v->next)
 363       if (etna_shader_key_equal(&key, &v->key))
 364          return v;
 365
 366    /* compile new variant if it doesn't exist already */
 367    v = create_variant(shader, key);
 368    if (v) {
 369       v->next = shader->variants;
 370       shader->variants = v;
 371       dump_shader_info(v, debug);
 372    }
 373
 374    return v;
 375 }
 376
 377 static void *
 378 etna_create_shader_state(struct pipe_context *pctx,
 379                          const struct pipe_shader_state *pss)
 380 {
 381    struct etna_context *ctx = etna_context(pctx);
 382    struct etna_shader *shader = CALLOC_STRUCT(etna_shader);
 383
 384    if (!shader)
 385       return NULL;
 386
 387    static uint32_t id;
 388    shader->id = id++;
 389    shader->specs = &ctx->specs;
 390
 391    if (DBG_ENABLED(ETNA_DBG_NIR))
 392       shader->nir = (pss->type == PIPE_SHADER_IR_NIR) ? pss->ir.nir :
 393                      tgsi_to_nir(pss->tokens, pctx->screen);
 394    else
 395       shader->tokens = tgsi_dup_tokens(pss->tokens);
 396
 397
 398
 399    if (etna_mesa_debug & ETNA_DBG_SHADERDB) {
 400       /* if shader-db run, create a standard variant immediately
 401        * (as otherwise nothing will trigger the shader to be
 402        * actually compiled).
 403        */
 404       struct etna_shader_key key = {};
 405       etna_shader_variant(shader, key, &ctx->debug);
 406    }
 407
 408    return shader;
 409 }
 410
 411 static void
 412 etna_delete_shader_state(struct pipe_context *pctx, void *ss)
 413 {
 414    struct etna_shader *shader = ss;
 415    struct etna_shader_variant *v, *t;
 416
 417    v = shader->variants;
 418    while (v) {
 419       t = v;
 420       v = v->next;
 421       if (t->bo)
 422          etna_bo_del(t->bo);
 423
 424       if (DBG_ENABLED(ETNA_DBG_NIR))
 425          etna_destroy_shader_nir(t);
 426       else
 427          etna_destroy_shader(t);
 428    }
 429
 430    ralloc_free(shader->nir);
 431    FREE(shader);
 432 }
 433
 434 static void
 435 etna_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 436 {
 437    struct etna_context *ctx = etna_context(pctx);
 438
 439    ctx->shader.bind_fs = hwcso;
 440    ctx->dirty |= ETNA_DIRTY_SHADER;
 441 }
 442
 443 static void
 444 etna_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 445 {
 446    struct etna_context *ctx = etna_context(pctx);
 447
 448    ctx->shader.bind_vs = hwcso;
 449    ctx->dirty |= ETNA_DIRTY_SHADER;
 450 }
 451
 452 void
 453 etna_shader_init(struct pipe_context *pctx)
 454 {
 455    pctx->create_fs_state = etna_create_shader_state;
 456    pctx->bind_fs_state = etna_bind_fs_state;
 457    pctx->delete_fs_state = etna_delete_shader_state;
 458    pctx->create_vs_state = etna_create_shader_state;
 459    pctx->bind_vs_state = etna_bind_vs_state;
 460    pctx->delete_vs_state = etna_delete_shader_state;
 461 }