src/gallium/drivers/lima/lima_draw.c

   1 /*
   2  * Copyright (c) 2011-2013 Luc Verhaegen <libv@skynet.be>
   3  * Copyright (c) 2017-2019 Lima Project
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sub license,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the
  13  * next paragraph) shall be included in all copies or substantial portions
  14  * of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  22  * DEALINGS IN THE SOFTWARE.
  23  *
  24  */
  25
  26 #include "util/format/u_format.h"
  27 #include "util/u_debug.h"
  28 #include "util/u_half.h"
  29 #include "util/u_helpers.h"
  30 #include "util/u_inlines.h"
  31 #include "util/u_pack_color.h"
  32 #include "util/u_split_draw.h"
  33 #include "util/u_upload_mgr.h"
  34 #include "util/u_prim.h"
  35 #include "util/u_vbuf.h"
  36 #include "util/hash_table.h"
  37
  38 #include "lima_context.h"
  39 #include "lima_screen.h"
  40 #include "lima_resource.h"
  41 #include "lima_program.h"
  42 #include "lima_bo.h"
  43 #include "lima_submit.h"
  44 #include "lima_texture.h"
  45 #include "lima_util.h"
  46 #include "lima_gpu.h"
  47
  48 #include <drm-uapi/lima_drm.h>
  49
  50 static bool
  51 lima_is_scissor_zero(struct lima_context *ctx)
  52 {
  53    if (!ctx->rasterizer || !ctx->rasterizer->base.scissor)
  54       return false;
  55
  56    struct pipe_scissor_state *scissor = &ctx->scissor;
  57    return
  58       scissor->minx == scissor->maxx
  59       && scissor->miny == scissor->maxy;
  60 }
  61
  62 static void
  63 lima_update_submit_wb(struct lima_context *ctx, unsigned buffers)
  64 {
  65    struct lima_submit *submit = lima_submit_get(ctx);
  66    struct lima_context_framebuffer *fb = &ctx->framebuffer;
  67
  68    /* add to submit when the buffer is dirty and resolve is clear (not added before) */
  69    if (fb->base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0) &&
  70        !(submit->resolve & PIPE_CLEAR_COLOR0)) {
  71       struct lima_resource *res = lima_resource(fb->base.cbufs[0]->texture);
  72       lima_flush_submit_accessing_bo(ctx, res->bo, true);
  73       _mesa_hash_table_insert(ctx->write_submits, &res->base, submit);
  74       lima_submit_add_bo(submit, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
  75    }
  76
  77    /* add to submit when the buffer is dirty and resolve is clear (not added before) */
  78    if (fb->base.zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
  79        !(submit->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
  80       struct lima_resource *res = lima_resource(fb->base.zsbuf->texture);
  81       lima_flush_submit_accessing_bo(ctx, res->bo, true);
  82       _mesa_hash_table_insert(ctx->write_submits, &res->base, submit);
  83       lima_submit_add_bo(submit, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
  84    }
  85
  86    submit->resolve |= buffers;
  87 }
  88
  89 static void
  90 lima_damage_rect_union(struct lima_context *ctx, unsigned minx, unsigned maxx, unsigned miny, unsigned maxy)
  91 {
  92    ctx->damage_rect.minx = MIN2(ctx->damage_rect.minx, minx);
  93    ctx->damage_rect.miny = MIN2(ctx->damage_rect.miny, miny);
  94    ctx->damage_rect.maxx = MAX2(ctx->damage_rect.maxx, maxx);
  95    ctx->damage_rect.maxy = MAX2(ctx->damage_rect.maxy, maxy);
  96 }
  97
  98 static void
  99 lima_clear(struct pipe_context *pctx, unsigned buffers,
 100            const union pipe_color_union *color, double depth, unsigned stencil)
 101 {
 102    struct lima_context *ctx = lima_context(pctx);
 103
 104    lima_flush(ctx);
 105
 106    lima_update_submit_wb(ctx, buffers);
 107
 108    /* no need to reload if cleared */
 109    if (ctx->framebuffer.base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0)) {
 110       struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
 111       surf->reload = false;
 112    }
 113
 114    struct lima_context_clear *clear = &ctx->clear;
 115    clear->buffers = buffers;
 116
 117    if (buffers & PIPE_CLEAR_COLOR0) {
 118       clear->color_8pc =
 119          ((uint32_t)float_to_ubyte(color->f[3]) << 24) |
 120          ((uint32_t)float_to_ubyte(color->f[2]) << 16) |
 121          ((uint32_t)float_to_ubyte(color->f[1]) << 8) |
 122          float_to_ubyte(color->f[0]);
 123
 124       clear->color_16pc =
 125          ((uint64_t)float_to_ushort(color->f[3]) << 48) |
 126          ((uint64_t)float_to_ushort(color->f[2]) << 32) |
 127          ((uint64_t)float_to_ushort(color->f[1]) << 16) |
 128          float_to_ushort(color->f[0]);
 129    }
 130
 131    if (buffers & PIPE_CLEAR_DEPTH)
 132       clear->depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
 133
 134    if (buffers & PIPE_CLEAR_STENCIL)
 135       clear->stencil = stencil;
 136
 137    ctx->dirty |= LIMA_CONTEXT_DIRTY_CLEAR;
 138
 139    lima_damage_rect_union(ctx, 0, ctx->framebuffer.base.width,
 140                                0, ctx->framebuffer.base.height);
 141 }
 142
 143 enum lima_attrib_type {
 144    LIMA_ATTRIB_FLOAT = 0x000,
 145    /* todo: find out what lives here. */
 146    LIMA_ATTRIB_I16   = 0x004,
 147    LIMA_ATTRIB_U16   = 0x005,
 148    LIMA_ATTRIB_I8    = 0x006,
 149    LIMA_ATTRIB_U8    = 0x007,
 150    LIMA_ATTRIB_I8N   = 0x008,
 151    LIMA_ATTRIB_U8N   = 0x009,
 152    LIMA_ATTRIB_I16N  = 0x00A,
 153    LIMA_ATTRIB_U16N  = 0x00B,
 154    /* todo: where is the 32 int */
 155    /* todo: find out what lives here. */
 156    LIMA_ATTRIB_FIXED = 0x101
 157 };
 158
 159 static enum lima_attrib_type
 160 lima_pipe_format_to_attrib_type(enum pipe_format format)
 161 {
 162    const struct util_format_description *desc = util_format_description(format);
 163    int i = util_format_get_first_non_void_channel(format);
 164    const struct util_format_channel_description *c = desc->channel + i;
 165
 166    switch (c->type) {
 167    case UTIL_FORMAT_TYPE_FLOAT:
 168       return LIMA_ATTRIB_FLOAT;
 169    case UTIL_FORMAT_TYPE_FIXED:
 170       return LIMA_ATTRIB_FIXED;
 171    case UTIL_FORMAT_TYPE_SIGNED:
 172       if (c->size == 8) {
 173          if (c->normalized)
 174             return LIMA_ATTRIB_I8N;
 175          else
 176             return LIMA_ATTRIB_I8;
 177       }
 178       else if (c->size == 16) {
 179          if (c->normalized)
 180             return LIMA_ATTRIB_I16N;
 181          else
 182             return LIMA_ATTRIB_I16;
 183       }
 184       break;
 185    case UTIL_FORMAT_TYPE_UNSIGNED:
 186       if (c->size == 8) {
 187          if (c->normalized)
 188             return LIMA_ATTRIB_U8N;
 189          else
 190             return LIMA_ATTRIB_U8;
 191       }
 192       else if (c->size == 16) {
 193          if (c->normalized)
 194             return LIMA_ATTRIB_U16N;
 195          else
 196             return LIMA_ATTRIB_U16;
 197       }
 198       break;
 199    }
 200
 201    return LIMA_ATTRIB_FLOAT;
 202 }
 203
 204 static void
 205 lima_pack_vs_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 206 {
 207    struct lima_submit *submit = lima_submit_get(ctx);
 208
 209    VS_CMD_BEGIN(&submit->vs_cmd_array, 24);
 210
 211    if (!info->index_size) {
 212       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1();
 213       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2();
 214    }
 215
 216    int uniform_size = ctx->vs->uniform_pending_offset + ctx->vs->constant_size + 32;
 217    VS_CMD_UNIFORMS_ADDRESS(
 218       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform),
 219       align(uniform_size, 16));
 220
 221    VS_CMD_SHADER_ADDRESS(ctx->vs->bo->va, ctx->vs->shader_size);
 222    VS_CMD_SHADER_INFO(ctx->vs->prefetch, ctx->vs->shader_size);
 223
 224    int num_outputs = ctx->vs->num_outputs;
 225    int num_attributes = ctx->vertex_elements->num_elements;
 226    VS_CMD_VARYING_ATTRIBUTE_COUNT(num_outputs, MAX2(1, num_attributes));
 227
 228    VS_CMD_UNKNOWN1();
 229
 230    VS_CMD_ATTRIBUTES_ADDRESS(
 231       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info),
 232       MAX2(1, num_attributes));
 233
 234    VS_CMD_VARYINGS_ADDRESS(
 235       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info),
 236       num_outputs);
 237
 238    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 239    VS_CMD_DRAW(num, info->index_size);
 240
 241    VS_CMD_UNKNOWN2();
 242
 243    VS_CMD_ARRAYS_SEMAPHORE_END(info->index_size);
 244
 245    VS_CMD_END();
 246 }
 247
 248 static void
 249 lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 250 {
 251    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 252    struct lima_vs_shader_state *vs = ctx->vs;
 253    unsigned minx, maxx, miny, maxy;
 254
 255    /* If it's zero scissor, we skip adding all other commands */
 256    if (lima_is_scissor_zero(ctx))
 257       return;
 258
 259    struct lima_submit *submit = lima_submit_get(ctx);
 260    PLBU_CMD_BEGIN(&submit->plbu_cmd_array, 32);
 261
 262    PLBU_CMD_VIEWPORT_LEFT(fui(ctx->viewport.left));
 263    PLBU_CMD_VIEWPORT_RIGHT(fui(ctx->viewport.right));
 264    PLBU_CMD_VIEWPORT_BOTTOM(fui(ctx->viewport.bottom));
 265    PLBU_CMD_VIEWPORT_TOP(fui(ctx->viewport.top));
 266
 267    if (!info->index_size)
 268       PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN();
 269
 270    int cf = ctx->rasterizer->base.cull_face;
 271    int ccw = ctx->rasterizer->base.front_ccw;
 272    uint32_t cull = 0;
 273    bool force_point_size = false;
 274
 275    if (cf != PIPE_FACE_NONE) {
 276       if (cf & PIPE_FACE_FRONT)
 277          cull |= ccw ? 0x00040000 : 0x00020000;
 278       if (cf & PIPE_FACE_BACK)
 279          cull |= ccw ? 0x00020000 : 0x00040000;
 280    }
 281
 282    /* Specify point size with PLBU command if shader doesn't write */
 283    if (info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1)
 284       force_point_size = true;
 285
 286    /* Specify line width with PLBU command for lines */
 287    if (info->mode > PIPE_PRIM_POINTS && info->mode < PIPE_PRIM_TRIANGLES)
 288       force_point_size = true;
 289
 290    PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, info->index_size);
 291
 292    PLBU_CMD_RSW_VERTEX_ARRAY(
 293       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw),
 294       ctx->gp_output->va);
 295
 296    /* TODO
 297     * - we should set it only for the first draw that enabled the scissor and for
 298     *   latter draw only if scissor is dirty
 299     */
 300    if (ctx->rasterizer->base.scissor) {
 301       struct pipe_scissor_state *scissor = &ctx->scissor;
 302       minx = scissor->minx;
 303       maxx = scissor->maxx;
 304       miny = scissor->miny;
 305       maxy = scissor->maxy;
 306    } else {
 307       minx = 0;
 308       maxx = fb->base.width;
 309       miny = 0;
 310       maxy = fb->base.height;
 311    }
 312
 313    minx = MAX2(minx, ctx->viewport.left);
 314    maxx = MIN2(maxx, ctx->viewport.right);
 315    miny = MAX2(miny, ctx->viewport.bottom);
 316    maxy = MIN2(maxy, ctx->viewport.top);
 317
 318    PLBU_CMD_SCISSORS(minx, maxx, miny, maxy);
 319    lima_damage_rect_union(ctx, minx, maxx, miny, maxy);
 320
 321    PLBU_CMD_UNKNOWN1();
 322
 323    PLBU_CMD_DEPTH_RANGE_NEAR(fui(ctx->viewport.near));
 324    PLBU_CMD_DEPTH_RANGE_FAR(fui(ctx->viewport.far));
 325
 326    if ((info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1) ||
 327        ((info->mode >= PIPE_PRIM_LINES) && (info->mode < PIPE_PRIM_TRIANGLES)))
 328    {
 329       uint32_t v = info->mode == PIPE_PRIM_POINTS ?
 330          fui(ctx->rasterizer->base.point_size) : fui(ctx->rasterizer->base.line_width);
 331       PLBU_CMD_LOW_PRIM_SIZE(v);
 332    }
 333
 334    if (info->index_size) {
 335       PLBU_CMD_INDEXED_DEST(ctx->gp_output->va);
 336       if (vs->point_size_idx != -1)
 337          PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt);
 338
 339       PLBU_CMD_INDICES(ctx->index_res->bo->va + info->start * info->index_size + ctx->index_offset);
 340    }
 341    else {
 342       /* can this make the attribute info static? */
 343       PLBU_CMD_DRAW_ARRAYS(info->mode, info->start, info->count);
 344    }
 345
 346    PLBU_CMD_ARRAYS_SEMAPHORE_END();
 347
 348    if (info->index_size)
 349       PLBU_CMD_DRAW_ELEMENTS(info->mode, ctx->min_index, info->count);
 350
 351    PLBU_CMD_END();
 352 }
 353
 354 static int
 355 lima_blend_func(enum pipe_blend_func pipe)
 356 {
 357    switch (pipe) {
 358    case PIPE_BLEND_ADD:
 359       return 2;
 360    case PIPE_BLEND_SUBTRACT:
 361       return 0;
 362    case PIPE_BLEND_REVERSE_SUBTRACT:
 363       return 1;
 364    case PIPE_BLEND_MIN:
 365       return 4;
 366    case PIPE_BLEND_MAX:
 367       return 5;
 368    }
 369    return -1;
 370 }
 371
 372 static int
 373 lima_blend_factor_has_alpha(enum pipe_blendfactor pipe)
 374 {
 375    /* Bit 4 is set if the blendfactor uses alpha */
 376    switch (pipe) {
 377    case PIPE_BLENDFACTOR_SRC_ALPHA:
 378    case PIPE_BLENDFACTOR_DST_ALPHA:
 379    case PIPE_BLENDFACTOR_CONST_ALPHA:
 380    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 381    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 382    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 383       return 1;
 384
 385    case PIPE_BLENDFACTOR_SRC_COLOR:
 386    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 387    case PIPE_BLENDFACTOR_DST_COLOR:
 388    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 389    case PIPE_BLENDFACTOR_CONST_COLOR:
 390    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 391    case PIPE_BLENDFACTOR_ZERO:
 392    case PIPE_BLENDFACTOR_ONE:
 393    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 394       return 0;
 395
 396    case PIPE_BLENDFACTOR_SRC1_COLOR:
 397    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 398    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 399    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 400       return -1; /* not supported */
 401    }
 402    return -1;
 403 }
 404
 405 static int
 406 lima_blend_factor_is_inv(enum pipe_blendfactor pipe)
 407 {
 408    /* Bit 3 is set if the blendfactor type is inverted */
 409    switch (pipe) {
 410    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 411    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 412    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 413    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 414    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 415    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 416    case PIPE_BLENDFACTOR_ONE:
 417       return 1;
 418
 419    case PIPE_BLENDFACTOR_SRC_COLOR:
 420    case PIPE_BLENDFACTOR_SRC_ALPHA:
 421    case PIPE_BLENDFACTOR_DST_COLOR:
 422    case PIPE_BLENDFACTOR_DST_ALPHA:
 423    case PIPE_BLENDFACTOR_CONST_COLOR:
 424    case PIPE_BLENDFACTOR_CONST_ALPHA:
 425    case PIPE_BLENDFACTOR_ZERO:
 426    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 427       return 0;
 428
 429    case PIPE_BLENDFACTOR_SRC1_COLOR:
 430    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 431    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 432    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 433       return -1; /* not supported */
 434    }
 435    return -1;
 436 }
 437
 438 static int
 439 lima_blend_factor(enum pipe_blendfactor pipe)
 440 {
 441    /* Bits 0-2 indicate the blendfactor type */
 442    switch (pipe) {
 443    case PIPE_BLENDFACTOR_SRC_COLOR:
 444    case PIPE_BLENDFACTOR_SRC_ALPHA:
 445    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 446    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 447       return 0;
 448
 449    case PIPE_BLENDFACTOR_DST_COLOR:
 450    case PIPE_BLENDFACTOR_DST_ALPHA:
 451    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 452    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 453       return 1;
 454
 455    case PIPE_BLENDFACTOR_CONST_COLOR:
 456    case PIPE_BLENDFACTOR_CONST_ALPHA:
 457    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 458    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 459       return 2;
 460
 461    case PIPE_BLENDFACTOR_ZERO:
 462    case PIPE_BLENDFACTOR_ONE:
 463       return 3;
 464
 465    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 466       return 4;
 467
 468    case PIPE_BLENDFACTOR_SRC1_COLOR:
 469    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 470    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 471    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 472       return -1; /* not supported */
 473    }
 474    return -1;
 475 }
 476
 477 static int
 478 lima_calculate_alpha_blend(enum pipe_blend_func rgb_func, enum pipe_blend_func alpha_func,
 479                            enum pipe_blendfactor rgb_src_factor, enum pipe_blendfactor rgb_dst_factor,
 480                            enum pipe_blendfactor alpha_src_factor, enum pipe_blendfactor alpha_dst_factor)
 481 {
 482    /* PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE has to be changed to PIPE_BLENDFACTOR_ONE
 483     * if it is set for alpha_src.
 484     */
 485    if (alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
 486       alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 487
 488    return lima_blend_func(rgb_func) |
 489       (lima_blend_func(alpha_func) << 3) |
 490
 491       (lima_blend_factor(rgb_src_factor) << 6) |
 492       (lima_blend_factor_is_inv(rgb_src_factor) << 9) |
 493       (lima_blend_factor_has_alpha(rgb_src_factor) << 10) |
 494
 495       (lima_blend_factor(rgb_dst_factor) << 11) |
 496       (lima_blend_factor_is_inv(rgb_dst_factor) << 14) |
 497       (lima_blend_factor_has_alpha(rgb_dst_factor) << 15) |
 498
 499       (lima_blend_factor(alpha_src_factor) << 16) |
 500       (lima_blend_factor_is_inv(alpha_src_factor) << 19) |
 501
 502       (lima_blend_factor(alpha_dst_factor) << 20) |
 503       (lima_blend_factor_is_inv(alpha_dst_factor) << 23) |
 504       0x0C000000; /* need to check if this is GLESv1 glAlphaFunc */
 505 }
 506
 507 static int
 508 lima_stencil_op(enum pipe_stencil_op pipe)
 509 {
 510    switch (pipe) {
 511    case PIPE_STENCIL_OP_KEEP:
 512       return 0;
 513    case PIPE_STENCIL_OP_ZERO:
 514       return 2;
 515    case PIPE_STENCIL_OP_REPLACE:
 516       return 1;
 517    case PIPE_STENCIL_OP_INCR:
 518       return 6;
 519    case PIPE_STENCIL_OP_DECR:
 520       return 7;
 521    case PIPE_STENCIL_OP_INCR_WRAP:
 522       return 4;
 523    case PIPE_STENCIL_OP_DECR_WRAP:
 524       return 5;
 525    case PIPE_STENCIL_OP_INVERT:
 526       return 3;
 527    }
 528    return -1;
 529 }
 530
 531 static unsigned
 532 lima_calculate_depth_test(struct pipe_depth_state *depth, struct pipe_rasterizer_state *rst)
 533 {
 534    int offset_scale = 0, offset_units = 0;
 535    enum pipe_compare_func func = (depth->enabled ? depth->func : PIPE_FUNC_ALWAYS);
 536
 537    offset_scale = CLAMP(rst->offset_scale * 4, -128, 127);
 538    if (offset_scale < 0)
 539       offset_scale += 0x100;
 540
 541    offset_units = CLAMP(rst->offset_units * 2, -128, 127);
 542    if (offset_units < 0)
 543       offset_units += 0x100;
 544
 545    return (depth->enabled && depth->writemask) |
 546       ((int)func << 1) |
 547       (offset_scale << 16) |
 548       (offset_units << 24) |
 549       0x30; /* find out what is this */
 550 }
 551
 552 static void
 553 lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *info)
 554 {
 555    struct lima_fs_shader_state *fs = ctx->fs;
 556    struct lima_render_state *render =
 557       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_plb_rsw,
 558                           sizeof(*render));
 559    bool early_z = true;
 560    bool pixel_kill = true;
 561
 562    /* do hw support RGBA independ blend?
 563     * PIPE_CAP_INDEP_BLEND_ENABLE
 564     *
 565     * how to handle the no cbuf only zbuf case?
 566     */
 567    struct pipe_rt_blend_state *rt = ctx->blend->base.rt;
 568    render->blend_color_bg = float_to_ubyte(ctx->blend_color.color[2]) |
 569       (float_to_ubyte(ctx->blend_color.color[1]) << 16);
 570    render->blend_color_ra = float_to_ubyte(ctx->blend_color.color[0]) |
 571       (float_to_ubyte(ctx->blend_color.color[3]) << 16);
 572
 573    if (rt->blend_enable) {
 574       render->alpha_blend = lima_calculate_alpha_blend(rt->rgb_func, rt->alpha_func,
 575          rt->rgb_src_factor, rt->rgb_dst_factor,
 576          rt->alpha_src_factor, rt->alpha_dst_factor);
 577    }
 578    else {
 579       /*
 580        * Special handling for blending disabled.
 581        * Binary driver is generating the same alpha_value,
 582        * as when we would just enable blending, without changing/setting any blend equation/params.
 583        * Normaly in this case mesa would set all rt fields (func/factor) to zero.
 584        */
 585       render->alpha_blend = lima_calculate_alpha_blend(PIPE_BLEND_ADD, PIPE_BLEND_ADD,
 586          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO,
 587          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO);
 588    }
 589
 590    render->alpha_blend |= (rt->colormask & PIPE_MASK_RGBA) << 28;
 591
 592    struct pipe_rasterizer_state *rst = &ctx->rasterizer->base;
 593    struct pipe_depth_state *depth = &ctx->zsa->base.depth;
 594    render->depth_test = lima_calculate_depth_test(depth, rst);
 595
 596    ushort far, near;
 597
 598    near = float_to_ushort(ctx->viewport.near);
 599    far = float_to_ushort(ctx->viewport.far);
 600
 601    /* Subtract epsilon from 'near' if far == near. Make sure we don't get overflow */
 602    if ((far == near) && (near != 0))
 603          near--;
 604
 605    /* overlap with plbu? any place can remove one? */
 606    render->depth_range = near | (far << 16);
 607
 608    struct pipe_stencil_state *stencil = ctx->zsa->base.stencil;
 609    struct pipe_stencil_ref *ref = &ctx->stencil_ref;
 610
 611    if (stencil[0].enabled) { /* stencil is enabled */
 612       render->stencil_front = stencil[0].func |
 613          (lima_stencil_op(stencil[0].fail_op) << 3) |
 614          (lima_stencil_op(stencil[0].zfail_op) << 6) |
 615          (lima_stencil_op(stencil[0].zpass_op) << 9) |
 616          (ref->ref_value[0] << 16) |
 617          (stencil[0].valuemask << 24);
 618       render->stencil_back = render->stencil_front;
 619       render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[0].writemask & 0xff) << 8;
 620       if (stencil[1].enabled) { /* two-side is enabled */
 621          render->stencil_back = stencil[1].func |
 622             (lima_stencil_op(stencil[1].fail_op) << 3) |
 623             (lima_stencil_op(stencil[1].zfail_op) << 6) |
 624             (lima_stencil_op(stencil[1].zpass_op) << 9) |
 625             (ref->ref_value[1] << 16) |
 626             (stencil[1].valuemask << 24);
 627          render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[1].writemask & 0xff) << 8;
 628       }
 629       /* TODO: Find out, what (render->stecil_test & 0xffff0000) is.
 630        * 0x00ff0000 is probably (float_to_ubyte(alpha->ref_value) << 16)
 631        * (render->multi_sample & 0x00000007 is probably the compare function
 632        * of glAlphaFunc then.
 633        */
 634    }
 635    else {
 636       /* Default values, when stencil is disabled:
 637        * stencil[0|1].valuemask = 0xff
 638        * stencil[0|1].func = PIPE_FUNC_ALWAYS
 639        * stencil[0|1].writemask = 0xff
 640        */
 641       render->stencil_front = 0xff000007;
 642       render->stencil_back = 0xff000007;
 643       render->stencil_test = 0x0000ffff;
 644    }
 645
 646    /* need more investigation */
 647    if (info->mode == PIPE_PRIM_POINTS)
 648       render->multi_sample = 0x0000F007;
 649    else if (info->mode < PIPE_PRIM_TRIANGLES)
 650       render->multi_sample = 0x0000F407;
 651    else
 652       render->multi_sample = 0x0000F807;
 653    if (ctx->framebuffer.base.samples)
 654       render->multi_sample |= 0x68;
 655
 656    render->shader_address =
 657       ctx->fs->bo->va | (((uint32_t *)ctx->fs->bo->map)[0] & 0x1F);
 658
 659    /* seems not needed */
 660    render->uniforms_address = 0x00000000;
 661
 662    render->textures_address = 0x00000000;
 663
 664    render->aux0 = (ctx->vs->varying_stride >> 3);
 665    render->aux1 = 0x00001000;
 666    if (ctx->blend->base.dither)
 667       render->aux1 |= 0x00002000;
 668
 669    if (fs->uses_discard) {
 670       early_z = false;
 671       pixel_kill = false;
 672    }
 673
 674    if (rt->blend_enable)
 675       pixel_kill = false;
 676
 677    if ((rt->colormask & PIPE_MASK_RGBA) != PIPE_MASK_RGBA)
 678       pixel_kill = false;
 679
 680    if (early_z)
 681       render->aux0 |= 0x300;
 682
 683    if (pixel_kill)
 684       render->aux0 |= 0x1000;
 685
 686    if (ctx->tex_stateobj.num_samplers) {
 687       render->textures_address =
 688          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_tex_desc);
 689       render->aux0 |= ctx->tex_stateobj.num_samplers << 14;
 690       render->aux0 |= 0x20;
 691    }
 692
 693    if (ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer) {
 694       render->uniforms_address =
 695          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array);
 696       uint32_t size = ctx->buffer_state[lima_ctx_buff_pp_uniform].size;
 697       uint32_t bits = 0;
 698       if (size >= 8) {
 699          bits = util_last_bit(size >> 3) - 1;
 700          bits += size & u_bit_consecutive(0, bits + 3) ? 1 : 0;
 701       }
 702       render->uniforms_address |= bits > 0xf ? 0xf : bits;
 703
 704       render->aux0 |= 0x80;
 705       render->aux1 |= 0x10000;
 706    }
 707
 708    if (ctx->vs->num_varyings) {
 709       render->varying_types = 0x00000000;
 710       render->varyings_address = ctx->gp_output->va +
 711                                  ctx->gp_output_varyings_offt;
 712       for (int i = 0, index = 0; i < ctx->vs->num_outputs; i++) {
 713          int val;
 714
 715          if (i == ctx->vs->gl_pos_idx ||
 716              i == ctx->vs->point_size_idx)
 717             continue;
 718
 719          struct lima_varying_info *v = ctx->vs->varying + i;
 720          if (v->component_size == 4)
 721             val = v->components > 2 ? 0 : 1;
 722          else
 723             val = v->components > 2 ? 2 : 3;
 724
 725          if (index < 10)
 726             render->varying_types |= val << (3 * index);
 727          else if (index == 10) {
 728             render->varying_types |= val << 30;
 729             render->varyings_address |= val >> 2;
 730          }
 731          else if (index == 11)
 732             render->varyings_address |= val << 1;
 733
 734          index++;
 735       }
 736    }
 737    else {
 738       render->varying_types = 0x00000000;
 739       render->varyings_address = 0x00000000;
 740    }
 741
 742    lima_dump_command_stream_print(
 743       render, sizeof(*render), false, "add render state at va %x\n",
 744       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 745
 746    lima_dump_rsw_command_stream_print(
 747       render, sizeof(*render), lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 748
 749 }
 750
 751 static void
 752 lima_update_gp_attribute_info(struct lima_context *ctx, const struct pipe_draw_info *info)
 753 {
 754    struct lima_submit *submit = lima_submit_get(ctx);
 755    struct lima_vertex_element_state *ve = ctx->vertex_elements;
 756    struct lima_context_vertex_buffer *vb = &ctx->vertex_buffers;
 757
 758    uint32_t *attribute =
 759       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_attribute_info,
 760                           MAX2(1, ve->num_elements) * 8);
 761
 762    int n = 0;
 763    for (int i = 0; i < ve->num_elements; i++) {
 764       struct pipe_vertex_element *pve = ve->pipe + i;
 765
 766       assert(pve->vertex_buffer_index < vb->count);
 767       assert(vb->enabled_mask & (1 << pve->vertex_buffer_index));
 768
 769       struct pipe_vertex_buffer *pvb = vb->vb + pve->vertex_buffer_index;
 770       struct lima_resource *res = lima_resource(pvb->buffer.resource);
 771
 772       lima_submit_add_bo(submit, LIMA_PIPE_GP, res->bo, LIMA_SUBMIT_BO_READ);
 773
 774       unsigned start = info->index_size ? (ctx->min_index + info->index_bias) : info->start;
 775       attribute[n++] = res->bo->va + pvb->buffer_offset + pve->src_offset
 776          + start * pvb->stride;
 777       attribute[n++] = (pvb->stride << 11) |
 778          (lima_pipe_format_to_attrib_type(pve->src_format) << 2) |
 779          (util_format_get_nr_components(pve->src_format) - 1);
 780    }
 781
 782    lima_dump_command_stream_print(
 783       attribute, n * 4, false, "update attribute info at va %x\n",
 784       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info));
 785 }
 786
 787 static void
 788 lima_update_gp_uniform(struct lima_context *ctx)
 789 {
 790    struct lima_context_constant_buffer *ccb =
 791       ctx->const_buffer + PIPE_SHADER_VERTEX;
 792    struct lima_vs_shader_state *vs = ctx->vs;
 793
 794    int size = vs->uniform_pending_offset + vs->constant_size + 32;
 795    void *vs_const_buff =
 796       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_uniform, size);
 797
 798    if (ccb->buffer)
 799       memcpy(vs_const_buff, ccb->buffer, ccb->size);
 800
 801    memcpy(vs_const_buff + vs->uniform_pending_offset,
 802           ctx->viewport.transform.scale,
 803           sizeof(ctx->viewport.transform.scale));
 804    memcpy(vs_const_buff + vs->uniform_pending_offset + 16,
 805           ctx->viewport.transform.translate,
 806           sizeof(ctx->viewport.transform.translate));
 807
 808    if (vs->constant)
 809       memcpy(vs_const_buff + vs->uniform_pending_offset + 32,
 810              vs->constant, vs->constant_size);
 811
 812    lima_dump_command_stream_print(
 813       vs_const_buff, size, true,
 814       "update gp uniform at va %x\n",
 815       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform));
 816 }
 817
 818 static void
 819 lima_update_pp_uniform(struct lima_context *ctx)
 820 {
 821    const float *const_buff = ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer;
 822    size_t const_buff_size = ctx->const_buffer[PIPE_SHADER_FRAGMENT].size / sizeof(float);
 823
 824    if (!const_buff)
 825       return;
 826
 827    uint16_t *fp16_const_buff =
 828       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform,
 829                           const_buff_size * sizeof(uint16_t));
 830
 831    uint32_t *array =
 832       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform_array, 4);
 833
 834    for (int i = 0; i < const_buff_size; i++)
 835        fp16_const_buff[i] = util_float_to_half(const_buff[i]);
 836
 837    *array = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform);
 838
 839    lima_dump_command_stream_print(
 840       fp16_const_buff, const_buff_size * 2, false, "add pp uniform data at va %x\n",
 841       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform));
 842    lima_dump_command_stream_print(
 843       array, 4, false, "add pp uniform info at va %x\n",
 844       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array));
 845 }
 846
 847 static void
 848 lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info)
 849 {
 850    struct lima_submit *submit = lima_submit_get(ctx);
 851    struct lima_screen *screen = lima_screen(ctx->base.screen);
 852    struct lima_vs_shader_state *vs = ctx->vs;
 853    uint32_t gp_output_size;
 854    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 855
 856    uint32_t *varying =
 857       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info,
 858                           vs->num_outputs * 8);
 859    int n = 0;
 860
 861    int offset = 0;
 862
 863    for (int i = 0; i < vs->num_outputs; i++) {
 864       struct lima_varying_info *v = vs->varying + i;
 865
 866       if (i == vs->gl_pos_idx ||
 867           i == vs->point_size_idx)
 868          continue;
 869
 870       int size = v->component_size * 4;
 871
 872       /* does component_size == 2 need to be 16 aligned? */
 873       if (v->component_size == 4)
 874          offset = align(offset, 16);
 875
 876       v->offset = offset;
 877       offset += size;
 878    }
 879
 880    vs->varying_stride = align(offset, 16);
 881
 882    /* gl_Position is always present, allocate space for it */
 883    gp_output_size = align(4 * 4 * num, 0x40);
 884
 885    /* Allocate space for varyings if there're any */
 886    if (vs->num_varyings) {
 887       ctx->gp_output_varyings_offt = gp_output_size;
 888       gp_output_size += align(vs->varying_stride * num, 0x40);
 889    }
 890
 891    /* Allocate space for gl_PointSize if it's there */
 892    if (vs->point_size_idx != -1) {
 893       ctx->gp_output_point_size_offt = gp_output_size;
 894       gp_output_size += 4 * num;
 895    }
 896
 897    /* gp_output can be too large for the suballocator, so create a
 898     * separate bo for it. The bo cache should prevent performance hit.
 899     */
 900    ctx->gp_output = lima_bo_create(screen, gp_output_size, 0);
 901    assert(ctx->gp_output);
 902    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->gp_output, LIMA_SUBMIT_BO_WRITE);
 903    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->gp_output, LIMA_SUBMIT_BO_READ);
 904
 905    for (int i = 0; i < vs->num_outputs; i++) {
 906       struct lima_varying_info *v = vs->varying + i;
 907
 908       if (i == vs->gl_pos_idx) {
 909          /* gl_Position */
 910          varying[n++] = ctx->gp_output->va;
 911          varying[n++] = 0x8020;
 912       } else if (i == vs->point_size_idx) {
 913          /* gl_PointSize */
 914          varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt;
 915          varying[n++] = 0x2021;
 916       } else {
 917          /* Varying */
 918          varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt +
 919                         v->offset;
 920          varying[n++] = (vs->varying_stride << 11) | (v->components - 1) |
 921             (v->component_size == 2 ? 0x0C : 0);
 922       }
 923    }
 924
 925    lima_dump_command_stream_print(
 926       varying, n * 4, false, "update varying info at va %x\n",
 927       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info));
 928 }
 929
 930 static void
 931 lima_draw_vbo_update(struct pipe_context *pctx,
 932                      const struct pipe_draw_info *info)
 933 {
 934    struct lima_context *ctx = lima_context(pctx);
 935    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 936    unsigned buffers = 0;
 937
 938    if (fb->base.zsbuf) {
 939       if (ctx->zsa->base.depth.enabled)
 940          buffers |= PIPE_CLEAR_DEPTH;
 941       if (ctx->zsa->base.stencil[0].enabled ||
 942           ctx->zsa->base.stencil[1].enabled)
 943          buffers |= PIPE_CLEAR_STENCIL;
 944    }
 945
 946    if (fb->base.nr_cbufs)
 947       buffers |= PIPE_CLEAR_COLOR0;
 948
 949    lima_update_submit_wb(ctx, buffers);
 950
 951    lima_update_gp_attribute_info(ctx, info);
 952
 953    if ((ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
 954         ctx->const_buffer[PIPE_SHADER_VERTEX].dirty) ||
 955        ctx->dirty & LIMA_CONTEXT_DIRTY_VIEWPORT ||
 956        ctx->dirty & LIMA_CONTEXT_DIRTY_SHADER_VERT) {
 957       lima_update_gp_uniform(ctx);
 958       ctx->const_buffer[PIPE_SHADER_VERTEX].dirty = false;
 959    }
 960
 961    lima_update_varying(ctx, info);
 962
 963    /* If it's zero scissor, don't build vs cmd list */
 964    if (!lima_is_scissor_zero(ctx))
 965       lima_pack_vs_cmd(ctx, info);
 966
 967    if (ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
 968        ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty) {
 969       lima_update_pp_uniform(ctx);
 970       ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty = false;
 971    }
 972
 973    lima_update_textures(ctx);
 974
 975    lima_pack_render_state(ctx, info);
 976    lima_pack_plbu_cmd(ctx, info);
 977
 978    if (ctx->gp_output) {
 979       lima_bo_unreference(ctx->gp_output); /* held by submit */
 980       ctx->gp_output = NULL;
 981    }
 982
 983    ctx->dirty = 0;
 984 }
 985
 986 static void
 987 lima_draw_vbo_indexed(struct pipe_context *pctx,
 988                       const struct pipe_draw_info *info)
 989 {
 990    struct lima_context *ctx = lima_context(pctx);
 991    struct lima_submit *submit = lima_submit_get(ctx);
 992    struct pipe_resource *indexbuf = NULL;
 993
 994    /* Mali Utgard GPU always need min/max index info for index draw,
 995     * compute it if upper layer does not do for us */
 996    if (info->max_index == ~0u)
 997       u_vbuf_get_minmax_index(pctx, info, &ctx->min_index, &ctx->max_index);
 998    else {
 999       ctx->min_index = info->min_index;
1000       ctx->max_index = info->max_index;
1001    }
1002
1003    if (info->has_user_indices) {
1004       util_upload_index_buffer(&ctx->base, info, &indexbuf, &ctx->index_offset, 0x40);
1005       ctx->index_res = lima_resource(indexbuf);
1006    }
1007    else {
1008       ctx->index_res = lima_resource(info->index.resource);
1009       ctx->index_offset = 0;
1010    }
1011
1012    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1013    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1014    lima_draw_vbo_update(pctx, info);
1015
1016    if (indexbuf)
1017       pipe_resource_reference(&indexbuf, NULL);
1018 }
1019
1020 static void
1021 lima_draw_vbo_count(struct pipe_context *pctx,
1022                     const struct pipe_draw_info *info)
1023 {
1024    static const uint32_t max_verts = 65535;
1025
1026    struct pipe_draw_info local_info = *info;
1027    unsigned start = info->start;
1028    unsigned count = info->count;
1029
1030    while (count) {
1031       unsigned this_count = count;
1032       unsigned step;
1033
1034       u_split_draw(info, max_verts, &this_count, &step);
1035
1036       local_info.start = start;
1037       local_info.count = this_count;
1038
1039       lima_draw_vbo_update(pctx, &local_info);
1040
1041       count -= step;
1042       start += step;
1043    }
1044 }
1045
1046 static void
1047 lima_draw_vbo(struct pipe_context *pctx,
1048               const struct pipe_draw_info *info)
1049 {
1050    /* check if draw mode and vertex/index count match,
1051     * otherwise gp will hang */
1052    if (!u_trim_pipe_prim(info->mode, (unsigned*)&info->count)) {
1053       debug_printf("draw mode and vertex/index count mismatch\n");
1054       return;
1055    }
1056
1057    struct lima_context *ctx = lima_context(pctx);
1058
1059    if (!ctx->vs || !ctx->fs) {
1060       debug_warn_once("no shader, skip draw\n");
1061       return;
1062    }
1063
1064    if (!lima_update_vs_state(ctx) || !lima_update_fs_state(ctx))
1065       return;
1066
1067    struct lima_submit *submit = lima_submit_get(ctx);
1068
1069    lima_dump_command_stream_print(
1070       ctx->vs->bo->map, ctx->vs->shader_size, false,
1071       "add vs at va %x\n", ctx->vs->bo->va);
1072
1073    lima_dump_command_stream_print(
1074       ctx->fs->bo->map, ctx->fs->shader_size, false,
1075       "add fs at va %x\n", ctx->fs->bo->va);
1076
1077    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->vs->bo, LIMA_SUBMIT_BO_READ);
1078    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->fs->bo, LIMA_SUBMIT_BO_READ);
1079
1080    if (info->index_size)
1081       lima_draw_vbo_indexed(pctx, info);
1082    else
1083       lima_draw_vbo_count(pctx, info);
1084 }
1085
1086 void
1087 lima_draw_init(struct lima_context *ctx)
1088 {
1089    ctx->base.clear = lima_clear;
1090    ctx->base.draw_vbo = lima_draw_vbo;
1091 }