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    /* if fragment shader doesn't read pointcoord, disable it */
 155    if (link.pcoord_varying_comp_ofs == -1)
 156       cs->PA_CONFIG &= ~VIVS_PA_CONFIG_POINT_SPRITE_ENABLE;
 157
 158    cs->VS_LOAD_BALANCING = vs->vs_load_balancing;
 159    cs->VS_START_PC = 0;
 160
 161    cs->PS_END_PC = fs->code_size / 4;
 162    cs->PS_OUTPUT_REG = fs->ps_color_out_reg;
 163    cs->PS_INPUT_COUNT =
 164       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */
 165       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 166    cs->PS_TEMP_REGISTER_CONTROL =
 167       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1));
 168    cs->PS_CONTROL = VIVS_PS_CONTROL_SATURATE_RT0; /* XXX when can we set BYPASS? */
 169    cs->PS_START_PC = 0;
 170
 171    /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA
 172     * mode, avoids some fumbling in sync_context. */
 173    cs->PS_INPUT_COUNT_MSAA =
 174       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */
 175       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 176    cs->PS_TEMP_REGISTER_CONTROL_MSAA =
 177       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2));
 178
 179    uint32_t total_components = 0;
 180    DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0};
 181    DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0};
 182    for (int idx = 0; idx < link.num_varyings; ++idx) {
 183       const struct etna_varying *varying = &link.varyings[idx];
 184
 185       etna_bitarray_set(num_components, 4, idx, varying->num_components);
 186       for (int comp = 0; comp < varying->num_components; ++comp) {
 187          etna_bitarray_set(component_use, 2, total_components, varying->use[comp]);
 188          total_components += 1;
 189       }
 190    }
 191
 192    cs->GL_VARYING_TOTAL_COMPONENTS =
 193       VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2));
 194    cs->GL_VARYING_NUM_COMPONENTS = num_components[0];
 195    cs->GL_VARYING_COMPONENT_USE[0] = component_use[0];
 196    cs->GL_VARYING_COMPONENT_USE[1] = component_use[1];
 197
 198    cs->GL_HALTI5_SH_SPECIALS =
 199       0x7f7f0000 | /* unknown bits, probably other PS inputs */
 200       /* pointsize is last (see above) */
 201       VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ?
 202                                               cs->VS_OUTPUT_COUNT * 4 : 0x00) |
 203       VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ?
 204                                               link.pcoord_varying_comp_ofs : 0x7f);
 205
 206    /* mask out early Z bit when frag depth is written */
 207    cs->PE_DEPTH_CONFIG = ~COND(fs->ps_depth_out_reg >= 0, VIVS_PE_DEPTH_CONFIG_EARLY_Z);
 208
 209    /* reference instruction memory */
 210    cs->vs_inst_mem_size = vs->code_size;
 211    cs->VS_INST_MEM = vs->code;
 212
 213    cs->ps_inst_mem_size = fs->code_size;
 214    cs->PS_INST_MEM = fs->code;
 215
 216    if (vs->needs_icache | fs->needs_icache) {
 217       /* If either of the shaders needs ICACHE, we use it for both. It is
 218        * either switched on or off for the entire shader processor.
 219        */
 220       if (!etna_icache_upload_shader(ctx, vs) ||
 221           !etna_icache_upload_shader(ctx, fs)) {
 222          assert(0);
 223          return false;
 224       }
 225
 226       cs->VS_INST_ADDR.bo = vs->bo;
 227       cs->VS_INST_ADDR.offset = 0;
 228       cs->VS_INST_ADDR.flags = ETNA_RELOC_READ;
 229       cs->PS_INST_ADDR.bo = fs->bo;
 230       cs->PS_INST_ADDR.offset = 0;
 231       cs->PS_INST_ADDR.flags = ETNA_RELOC_READ;
 232    } else {
 233       /* clear relocs */
 234       memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR));
 235       memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR));
 236    }
 237
 238    return true;
 239 }
 240
 241 bool
 242 etna_shader_link(struct etna_context *ctx)
 243 {
 244    if (!ctx->shader.vs || !ctx->shader.fs)
 245       return false;
 246
 247    /* re-link vs and fs if needed */
 248    return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs);
 249 }
 250
 251 static bool
 252 etna_shader_update_vs_inputs(struct compiled_shader_state *cs,
 253                              const struct etna_shader_variant *vs,
 254                              const struct compiled_vertex_elements_state *ves)
 255 {
 256    unsigned num_temps, cur_temp, num_vs_inputs;
 257
 258    if (!vs)
 259       return false;
 260
 261    /* Number of vertex elements determines number of VS inputs. Otherwise,
 262     * the GPU crashes. Allocate any unused vertex elements to VS temporary
 263     * registers. */
 264    num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg);
 265    if (num_vs_inputs != ves->num_elements) {
 266       BUG("Number of elements %u does not match the number of VS inputs %zu",
 267           ves->num_elements, vs->infile.num_reg);
 268       return false;
 269    }
 270
 271    cur_temp = vs->num_temps;
 272    num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp;
 273
 274    cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) |
 275                         VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8);
 276    cs->VS_TEMP_REGISTER_CONTROL =
 277       VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps);
 278
 279    /* vs inputs (attributes) */
 280    DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0};
 281    for (int idx = 0; idx < num_vs_inputs; ++idx) {
 282       if (idx < vs->infile.num_reg)
 283          etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg);
 284       else
 285          etna_bitarray_set(vs_input, 8, idx, cur_temp++);
 286    }
 287
 288    for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx)
 289       cs->VS_INPUT[idx] = vs_input[idx];
 290
 291    return true;
 292 }
 293
 294 static inline const char *
 295 etna_shader_stage(struct etna_shader_variant *shader)
 296 {
 297    switch (shader->stage) {
 298    case MESA_SHADER_VERTEX:     return "VERT";
 299    case MESA_SHADER_FRAGMENT:   return "FRAG";
 300    case MESA_SHADER_COMPUTE:    return "CL";
 301    default:
 302       unreachable("invalid type");
 303       return NULL;
 304    }
 305 }
 306
 307 static void
 308 dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debug)
 309 {
 310    if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB))
 311       return;
 312
 313    pipe_debug_message(debug, SHADER_INFO,
 314          "%s shader: %u instructions, %u temps, "
 315          "%u immediates, %u loops",
 316          etna_shader_stage(v),
 317          v->code_size,
 318          v->num_temps,
 319          v->uniforms.imm_count,
 320          v->num_loops);
 321 }
 322
 323 bool
 324 etna_shader_update_vertex(struct etna_context *ctx)
 325 {
 326    return etna_shader_update_vs_inputs(&ctx->shader_state, ctx->shader.vs,
 327                                        ctx->vertex_elements);
 328 }
 329
 330 static struct etna_shader_variant *
 331 create_variant(struct etna_shader *shader, struct etna_shader_key key)
 332 {
 333    struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant);
 334    int ret;
 335
 336    if (!v)
 337       return NULL;
 338
 339    v->shader = shader;
 340    v->key = key;
 341
 342    ret = etna_compile_shader(v);
 343    if (!ret) {
 344       debug_error("compile failed!");
 345       goto fail;
 346    }
 347
 348    v->id = ++shader->variant_count;
 349
 350    return v;
 351
 352 fail:
 353    FREE(v);
 354    return NULL;
 355 }
 356
 357 struct etna_shader_variant *
 358 etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key,
 359                    struct pipe_debug_callback *debug)
 360 {
 361    struct etna_shader_variant *v;
 362
 363    for (v = shader->variants; v; v = v->next)
 364       if (etna_shader_key_equal(&key, &v->key))
 365          return v;
 366
 367    /* compile new variant if it doesn't exist already */
 368    v = create_variant(shader, key);
 369    if (v) {
 370       v->next = shader->variants;
 371       shader->variants = v;
 372       dump_shader_info(v, debug);
 373    }
 374
 375    return v;
 376 }
 377
 378 static void *
 379 etna_create_shader_state(struct pipe_context *pctx,
 380                          const struct pipe_shader_state *pss)
 381 {
 382    struct etna_context *ctx = etna_context(pctx);
 383    struct etna_shader *shader = CALLOC_STRUCT(etna_shader);
 384
 385    if (!shader)
 386       return NULL;
 387
 388    static uint32_t id;
 389    shader->id = id++;
 390    shader->specs = &ctx->specs;
 391
 392    if (DBG_ENABLED(ETNA_DBG_NIR))
 393       shader->nir = (pss->type == PIPE_SHADER_IR_NIR) ? pss->ir.nir :
 394                      tgsi_to_nir(pss->tokens, pctx->screen);
 395    else
 396       shader->tokens = tgsi_dup_tokens(pss->tokens);
 397
 398
 399
 400    if (etna_mesa_debug & ETNA_DBG_SHADERDB) {
 401       /* if shader-db run, create a standard variant immediately
 402        * (as otherwise nothing will trigger the shader to be
 403        * actually compiled).
 404        */
 405       struct etna_shader_key key = {};
 406       etna_shader_variant(shader, key, &ctx->debug);
 407    }
 408
 409    return shader;
 410 }
 411
 412 static void
 413 etna_delete_shader_state(struct pipe_context *pctx, void *ss)
 414 {
 415    struct etna_shader *shader = ss;
 416    struct etna_shader_variant *v, *t;
 417
 418    v = shader->variants;
 419    while (v) {
 420       t = v;
 421       v = v->next;
 422       if (t->bo)
 423          etna_bo_del(t->bo);
 424
 425       if (DBG_ENABLED(ETNA_DBG_NIR))
 426          etna_destroy_shader_nir(t);
 427       else
 428          etna_destroy_shader(t);
 429    }
 430
 431    ralloc_free(shader->nir);
 432    FREE(shader);
 433 }
 434
 435 static void
 436 etna_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 437 {
 438    struct etna_context *ctx = etna_context(pctx);
 439
 440    ctx->shader.bind_fs = hwcso;
 441    ctx->dirty |= ETNA_DIRTY_SHADER;
 442 }
 443
 444 static void
 445 etna_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 446 {
 447    struct etna_context *ctx = etna_context(pctx);
 448
 449    ctx->shader.bind_vs = hwcso;
 450    ctx->dirty |= ETNA_DIRTY_SHADER;
 451 }
 452
 453 void
 454 etna_shader_init(struct pipe_context *pctx)
 455 {
 456    pctx->create_fs_state = etna_create_shader_state;
 457    pctx->bind_fs_state = etna_bind_fs_state;
 458    pctx->delete_fs_state = etna_delete_shader_state;
 459    pctx->create_vs_state = etna_create_shader_state;
 460    pctx->bind_vs_state = etna_bind_vs_state;
 461    pctx->delete_vs_state = etna_delete_shader_state;
 462 }