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 pipe_scissor_state *rect,
  91                        unsigned minx, unsigned maxx,
  92                        unsigned miny, unsigned maxy)
  93 {
  94    rect->minx = MIN2(rect->minx, minx);
  95    rect->miny = MIN2(rect->miny, miny);
  96    rect->maxx = MAX2(rect->maxx, maxx);
  97    rect->maxy = MAX2(rect->maxy, maxy);
  98 }
  99
 100 static void
 101 lima_clear(struct pipe_context *pctx, unsigned buffers,
 102            const union pipe_color_union *color, double depth, unsigned stencil)
 103 {
 104    struct lima_context *ctx = lima_context(pctx);
 105
 106    lima_flush(ctx);
 107
 108    lima_update_submit_wb(ctx, buffers);
 109
 110    /* no need to reload if cleared */
 111    if (ctx->framebuffer.base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0)) {
 112       struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
 113       surf->reload = false;
 114    }
 115
 116    struct lima_submit *submit = lima_submit_get(ctx);
 117    struct lima_context_clear *clear = &ctx->clear;
 118    clear->buffers = buffers;
 119
 120    if (buffers & PIPE_CLEAR_COLOR0) {
 121       clear->color_8pc =
 122          ((uint32_t)float_to_ubyte(color->f[3]) << 24) |
 123          ((uint32_t)float_to_ubyte(color->f[2]) << 16) |
 124          ((uint32_t)float_to_ubyte(color->f[1]) << 8) |
 125          float_to_ubyte(color->f[0]);
 126
 127       clear->color_16pc =
 128          ((uint64_t)float_to_ushort(color->f[3]) << 48) |
 129          ((uint64_t)float_to_ushort(color->f[2]) << 32) |
 130          ((uint64_t)float_to_ushort(color->f[1]) << 16) |
 131          float_to_ushort(color->f[0]);
 132    }
 133
 134    if (buffers & PIPE_CLEAR_DEPTH)
 135       clear->depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
 136
 137    if (buffers & PIPE_CLEAR_STENCIL)
 138       clear->stencil = stencil;
 139
 140    ctx->dirty |= LIMA_CONTEXT_DIRTY_CLEAR;
 141
 142    lima_damage_rect_union(&submit->damage_rect,
 143                           0, ctx->framebuffer.base.width,
 144                           0, ctx->framebuffer.base.height);
 145 }
 146
 147 enum lima_attrib_type {
 148    LIMA_ATTRIB_FLOAT = 0x000,
 149    /* todo: find out what lives here. */
 150    LIMA_ATTRIB_I16   = 0x004,
 151    LIMA_ATTRIB_U16   = 0x005,
 152    LIMA_ATTRIB_I8    = 0x006,
 153    LIMA_ATTRIB_U8    = 0x007,
 154    LIMA_ATTRIB_I8N   = 0x008,
 155    LIMA_ATTRIB_U8N   = 0x009,
 156    LIMA_ATTRIB_I16N  = 0x00A,
 157    LIMA_ATTRIB_U16N  = 0x00B,
 158    /* todo: where is the 32 int */
 159    /* todo: find out what lives here. */
 160    LIMA_ATTRIB_FIXED = 0x101
 161 };
 162
 163 static enum lima_attrib_type
 164 lima_pipe_format_to_attrib_type(enum pipe_format format)
 165 {
 166    const struct util_format_description *desc = util_format_description(format);
 167    int i = util_format_get_first_non_void_channel(format);
 168    const struct util_format_channel_description *c = desc->channel + i;
 169
 170    switch (c->type) {
 171    case UTIL_FORMAT_TYPE_FLOAT:
 172       return LIMA_ATTRIB_FLOAT;
 173    case UTIL_FORMAT_TYPE_FIXED:
 174       return LIMA_ATTRIB_FIXED;
 175    case UTIL_FORMAT_TYPE_SIGNED:
 176       if (c->size == 8) {
 177          if (c->normalized)
 178             return LIMA_ATTRIB_I8N;
 179          else
 180             return LIMA_ATTRIB_I8;
 181       }
 182       else if (c->size == 16) {
 183          if (c->normalized)
 184             return LIMA_ATTRIB_I16N;
 185          else
 186             return LIMA_ATTRIB_I16;
 187       }
 188       break;
 189    case UTIL_FORMAT_TYPE_UNSIGNED:
 190       if (c->size == 8) {
 191          if (c->normalized)
 192             return LIMA_ATTRIB_U8N;
 193          else
 194             return LIMA_ATTRIB_U8;
 195       }
 196       else if (c->size == 16) {
 197          if (c->normalized)
 198             return LIMA_ATTRIB_U16N;
 199          else
 200             return LIMA_ATTRIB_U16;
 201       }
 202       break;
 203    }
 204
 205    return LIMA_ATTRIB_FLOAT;
 206 }
 207
 208 static void
 209 lima_pack_vs_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 210 {
 211    struct lima_submit *submit = lima_submit_get(ctx);
 212
 213    VS_CMD_BEGIN(&submit->vs_cmd_array, 24);
 214
 215    if (!info->index_size) {
 216       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1();
 217       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2();
 218    }
 219
 220    int uniform_size = ctx->vs->uniform_pending_offset + ctx->vs->constant_size + 32;
 221    VS_CMD_UNIFORMS_ADDRESS(
 222       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform),
 223       align(uniform_size, 16));
 224
 225    VS_CMD_SHADER_ADDRESS(ctx->vs->bo->va, ctx->vs->shader_size);
 226    VS_CMD_SHADER_INFO(ctx->vs->prefetch, ctx->vs->shader_size);
 227
 228    int num_outputs = ctx->vs->num_outputs;
 229    int num_attributes = ctx->vertex_elements->num_elements;
 230    VS_CMD_VARYING_ATTRIBUTE_COUNT(num_outputs, MAX2(1, num_attributes));
 231
 232    VS_CMD_UNKNOWN1();
 233
 234    VS_CMD_ATTRIBUTES_ADDRESS(
 235       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info),
 236       MAX2(1, num_attributes));
 237
 238    VS_CMD_VARYINGS_ADDRESS(
 239       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info),
 240       num_outputs);
 241
 242    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 243    VS_CMD_DRAW(num, info->index_size);
 244
 245    VS_CMD_UNKNOWN2();
 246
 247    VS_CMD_ARRAYS_SEMAPHORE_END(info->index_size);
 248
 249    VS_CMD_END();
 250 }
 251
 252 static void
 253 lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 254 {
 255    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 256    struct lima_vs_shader_state *vs = ctx->vs;
 257    unsigned minx, maxx, miny, maxy;
 258
 259    /* If it's zero scissor, we skip adding all other commands */
 260    if (lima_is_scissor_zero(ctx))
 261       return;
 262
 263    struct lima_submit *submit = lima_submit_get(ctx);
 264    PLBU_CMD_BEGIN(&submit->plbu_cmd_array, 32);
 265
 266    PLBU_CMD_VIEWPORT_LEFT(fui(ctx->viewport.left));
 267    PLBU_CMD_VIEWPORT_RIGHT(fui(ctx->viewport.right));
 268    PLBU_CMD_VIEWPORT_BOTTOM(fui(ctx->viewport.bottom));
 269    PLBU_CMD_VIEWPORT_TOP(fui(ctx->viewport.top));
 270
 271    if (!info->index_size)
 272       PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN();
 273
 274    int cf = ctx->rasterizer->base.cull_face;
 275    int ccw = ctx->rasterizer->base.front_ccw;
 276    uint32_t cull = 0;
 277    bool force_point_size = false;
 278
 279    if (cf != PIPE_FACE_NONE) {
 280       if (cf & PIPE_FACE_FRONT)
 281          cull |= ccw ? 0x00040000 : 0x00020000;
 282       if (cf & PIPE_FACE_BACK)
 283          cull |= ccw ? 0x00020000 : 0x00040000;
 284    }
 285
 286    /* Specify point size with PLBU command if shader doesn't write */
 287    if (info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1)
 288       force_point_size = true;
 289
 290    /* Specify line width with PLBU command for lines */
 291    if (info->mode > PIPE_PRIM_POINTS && info->mode < PIPE_PRIM_TRIANGLES)
 292       force_point_size = true;
 293
 294    PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, info->index_size);
 295
 296    PLBU_CMD_RSW_VERTEX_ARRAY(
 297       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw),
 298       ctx->gp_output->va);
 299
 300    /* TODO
 301     * - we should set it only for the first draw that enabled the scissor and for
 302     *   latter draw only if scissor is dirty
 303     */
 304    if (ctx->rasterizer->base.scissor) {
 305       struct pipe_scissor_state *scissor = &ctx->scissor;
 306       minx = scissor->minx;
 307       maxx = scissor->maxx;
 308       miny = scissor->miny;
 309       maxy = scissor->maxy;
 310    } else {
 311       minx = 0;
 312       maxx = fb->base.width;
 313       miny = 0;
 314       maxy = fb->base.height;
 315    }
 316
 317    minx = MAX2(minx, ctx->viewport.left);
 318    maxx = MIN2(maxx, ctx->viewport.right);
 319    miny = MAX2(miny, ctx->viewport.bottom);
 320    maxy = MIN2(maxy, ctx->viewport.top);
 321
 322    PLBU_CMD_SCISSORS(minx, maxx, miny, maxy);
 323    lima_damage_rect_union(&submit->damage_rect, minx, maxx, miny, maxy);
 324
 325    PLBU_CMD_UNKNOWN1();
 326
 327    PLBU_CMD_DEPTH_RANGE_NEAR(fui(ctx->viewport.near));
 328    PLBU_CMD_DEPTH_RANGE_FAR(fui(ctx->viewport.far));
 329
 330    if ((info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1) ||
 331        ((info->mode >= PIPE_PRIM_LINES) && (info->mode < PIPE_PRIM_TRIANGLES)))
 332    {
 333       uint32_t v = info->mode == PIPE_PRIM_POINTS ?
 334          fui(ctx->rasterizer->base.point_size) : fui(ctx->rasterizer->base.line_width);
 335       PLBU_CMD_LOW_PRIM_SIZE(v);
 336    }
 337
 338    if (info->index_size) {
 339       PLBU_CMD_INDEXED_DEST(ctx->gp_output->va);
 340       if (vs->point_size_idx != -1)
 341          PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt);
 342
 343       PLBU_CMD_INDICES(ctx->index_res->bo->va + info->start * info->index_size + ctx->index_offset);
 344    }
 345    else {
 346       /* can this make the attribute info static? */
 347       PLBU_CMD_DRAW_ARRAYS(info->mode, info->start, info->count);
 348    }
 349
 350    PLBU_CMD_ARRAYS_SEMAPHORE_END();
 351
 352    if (info->index_size)
 353       PLBU_CMD_DRAW_ELEMENTS(info->mode, ctx->min_index, info->count);
 354
 355    PLBU_CMD_END();
 356 }
 357
 358 static int
 359 lima_blend_func(enum pipe_blend_func pipe)
 360 {
 361    switch (pipe) {
 362    case PIPE_BLEND_ADD:
 363       return 2;
 364    case PIPE_BLEND_SUBTRACT:
 365       return 0;
 366    case PIPE_BLEND_REVERSE_SUBTRACT:
 367       return 1;
 368    case PIPE_BLEND_MIN:
 369       return 4;
 370    case PIPE_BLEND_MAX:
 371       return 5;
 372    }
 373    return -1;
 374 }
 375
 376 static int
 377 lima_blend_factor_has_alpha(enum pipe_blendfactor pipe)
 378 {
 379    /* Bit 4 is set if the blendfactor uses alpha */
 380    switch (pipe) {
 381    case PIPE_BLENDFACTOR_SRC_ALPHA:
 382    case PIPE_BLENDFACTOR_DST_ALPHA:
 383    case PIPE_BLENDFACTOR_CONST_ALPHA:
 384    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 385    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 386    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 387       return 1;
 388
 389    case PIPE_BLENDFACTOR_SRC_COLOR:
 390    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 391    case PIPE_BLENDFACTOR_DST_COLOR:
 392    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 393    case PIPE_BLENDFACTOR_CONST_COLOR:
 394    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 395    case PIPE_BLENDFACTOR_ZERO:
 396    case PIPE_BLENDFACTOR_ONE:
 397    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 398       return 0;
 399
 400    case PIPE_BLENDFACTOR_SRC1_COLOR:
 401    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 402    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 403    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 404       return -1; /* not supported */
 405    }
 406    return -1;
 407 }
 408
 409 static int
 410 lima_blend_factor_is_inv(enum pipe_blendfactor pipe)
 411 {
 412    /* Bit 3 is set if the blendfactor type is inverted */
 413    switch (pipe) {
 414    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 415    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 416    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 417    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 418    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 419    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 420    case PIPE_BLENDFACTOR_ONE:
 421       return 1;
 422
 423    case PIPE_BLENDFACTOR_SRC_COLOR:
 424    case PIPE_BLENDFACTOR_SRC_ALPHA:
 425    case PIPE_BLENDFACTOR_DST_COLOR:
 426    case PIPE_BLENDFACTOR_DST_ALPHA:
 427    case PIPE_BLENDFACTOR_CONST_COLOR:
 428    case PIPE_BLENDFACTOR_CONST_ALPHA:
 429    case PIPE_BLENDFACTOR_ZERO:
 430    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 431       return 0;
 432
 433    case PIPE_BLENDFACTOR_SRC1_COLOR:
 434    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 435    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 436    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 437       return -1; /* not supported */
 438    }
 439    return -1;
 440 }
 441
 442 static int
 443 lima_blend_factor(enum pipe_blendfactor pipe)
 444 {
 445    /* Bits 0-2 indicate the blendfactor type */
 446    switch (pipe) {
 447    case PIPE_BLENDFACTOR_SRC_COLOR:
 448    case PIPE_BLENDFACTOR_SRC_ALPHA:
 449    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 450    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 451       return 0;
 452
 453    case PIPE_BLENDFACTOR_DST_COLOR:
 454    case PIPE_BLENDFACTOR_DST_ALPHA:
 455    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 456    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 457       return 1;
 458
 459    case PIPE_BLENDFACTOR_CONST_COLOR:
 460    case PIPE_BLENDFACTOR_CONST_ALPHA:
 461    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 462    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 463       return 2;
 464
 465    case PIPE_BLENDFACTOR_ZERO:
 466    case PIPE_BLENDFACTOR_ONE:
 467       return 3;
 468
 469    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 470       return 4;
 471
 472    case PIPE_BLENDFACTOR_SRC1_COLOR:
 473    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 474    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 475    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 476       return -1; /* not supported */
 477    }
 478    return -1;
 479 }
 480
 481 static int
 482 lima_calculate_alpha_blend(enum pipe_blend_func rgb_func, enum pipe_blend_func alpha_func,
 483                            enum pipe_blendfactor rgb_src_factor, enum pipe_blendfactor rgb_dst_factor,
 484                            enum pipe_blendfactor alpha_src_factor, enum pipe_blendfactor alpha_dst_factor)
 485 {
 486    /* PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE has to be changed to PIPE_BLENDFACTOR_ONE
 487     * if it is set for alpha_src.
 488     */
 489    if (alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
 490       alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 491
 492    return lima_blend_func(rgb_func) |
 493       (lima_blend_func(alpha_func) << 3) |
 494
 495       (lima_blend_factor(rgb_src_factor) << 6) |
 496       (lima_blend_factor_is_inv(rgb_src_factor) << 9) |
 497       (lima_blend_factor_has_alpha(rgb_src_factor) << 10) |
 498
 499       (lima_blend_factor(rgb_dst_factor) << 11) |
 500       (lima_blend_factor_is_inv(rgb_dst_factor) << 14) |
 501       (lima_blend_factor_has_alpha(rgb_dst_factor) << 15) |
 502
 503       (lima_blend_factor(alpha_src_factor) << 16) |
 504       (lima_blend_factor_is_inv(alpha_src_factor) << 19) |
 505
 506       (lima_blend_factor(alpha_dst_factor) << 20) |
 507       (lima_blend_factor_is_inv(alpha_dst_factor) << 23) |
 508       0x0C000000; /* need to check if this is GLESv1 glAlphaFunc */
 509 }
 510
 511 static int
 512 lima_stencil_op(enum pipe_stencil_op pipe)
 513 {
 514    switch (pipe) {
 515    case PIPE_STENCIL_OP_KEEP:
 516       return 0;
 517    case PIPE_STENCIL_OP_ZERO:
 518       return 2;
 519    case PIPE_STENCIL_OP_REPLACE:
 520       return 1;
 521    case PIPE_STENCIL_OP_INCR:
 522       return 6;
 523    case PIPE_STENCIL_OP_DECR:
 524       return 7;
 525    case PIPE_STENCIL_OP_INCR_WRAP:
 526       return 4;
 527    case PIPE_STENCIL_OP_DECR_WRAP:
 528       return 5;
 529    case PIPE_STENCIL_OP_INVERT:
 530       return 3;
 531    }
 532    return -1;
 533 }
 534
 535 static unsigned
 536 lima_calculate_depth_test(struct pipe_depth_state *depth, struct pipe_rasterizer_state *rst)
 537 {
 538    int offset_scale = 0, offset_units = 0;
 539    enum pipe_compare_func func = (depth->enabled ? depth->func : PIPE_FUNC_ALWAYS);
 540
 541    offset_scale = CLAMP(rst->offset_scale * 4, -128, 127);
 542    if (offset_scale < 0)
 543       offset_scale += 0x100;
 544
 545    offset_units = CLAMP(rst->offset_units * 2, -128, 127);
 546    if (offset_units < 0)
 547       offset_units += 0x100;
 548
 549    return (depth->enabled && depth->writemask) |
 550       ((int)func << 1) |
 551       (offset_scale << 16) |
 552       (offset_units << 24) |
 553       0x30; /* find out what is this */
 554 }
 555
 556 static void
 557 lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *info)
 558 {
 559    struct lima_fs_shader_state *fs = ctx->fs;
 560    struct lima_render_state *render =
 561       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_plb_rsw,
 562                           sizeof(*render));
 563    bool early_z = true;
 564    bool pixel_kill = true;
 565
 566    /* do hw support RGBA independ blend?
 567     * PIPE_CAP_INDEP_BLEND_ENABLE
 568     *
 569     * how to handle the no cbuf only zbuf case?
 570     */
 571    struct pipe_rt_blend_state *rt = ctx->blend->base.rt;
 572    render->blend_color_bg = float_to_ubyte(ctx->blend_color.color[2]) |
 573       (float_to_ubyte(ctx->blend_color.color[1]) << 16);
 574    render->blend_color_ra = float_to_ubyte(ctx->blend_color.color[0]) |
 575       (float_to_ubyte(ctx->blend_color.color[3]) << 16);
 576
 577    if (rt->blend_enable) {
 578       render->alpha_blend = lima_calculate_alpha_blend(rt->rgb_func, rt->alpha_func,
 579          rt->rgb_src_factor, rt->rgb_dst_factor,
 580          rt->alpha_src_factor, rt->alpha_dst_factor);
 581    }
 582    else {
 583       /*
 584        * Special handling for blending disabled.
 585        * Binary driver is generating the same alpha_value,
 586        * as when we would just enable blending, without changing/setting any blend equation/params.
 587        * Normaly in this case mesa would set all rt fields (func/factor) to zero.
 588        */
 589       render->alpha_blend = lima_calculate_alpha_blend(PIPE_BLEND_ADD, PIPE_BLEND_ADD,
 590          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO,
 591          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO);
 592    }
 593
 594    render->alpha_blend |= (rt->colormask & PIPE_MASK_RGBA) << 28;
 595
 596    struct pipe_rasterizer_state *rst = &ctx->rasterizer->base;
 597    struct pipe_depth_state *depth = &ctx->zsa->base.depth;
 598    render->depth_test = lima_calculate_depth_test(depth, rst);
 599
 600    ushort far, near;
 601
 602    near = float_to_ushort(ctx->viewport.near);
 603    far = float_to_ushort(ctx->viewport.far);
 604
 605    /* Subtract epsilon from 'near' if far == near. Make sure we don't get overflow */
 606    if ((far == near) && (near != 0))
 607          near--;
 608
 609    /* overlap with plbu? any place can remove one? */
 610    render->depth_range = near | (far << 16);
 611
 612    struct pipe_stencil_state *stencil = ctx->zsa->base.stencil;
 613    struct pipe_stencil_ref *ref = &ctx->stencil_ref;
 614
 615    if (stencil[0].enabled) { /* stencil is enabled */
 616       render->stencil_front = stencil[0].func |
 617          (lima_stencil_op(stencil[0].fail_op) << 3) |
 618          (lima_stencil_op(stencil[0].zfail_op) << 6) |
 619          (lima_stencil_op(stencil[0].zpass_op) << 9) |
 620          (ref->ref_value[0] << 16) |
 621          (stencil[0].valuemask << 24);
 622       render->stencil_back = render->stencil_front;
 623       render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[0].writemask & 0xff) << 8;
 624       if (stencil[1].enabled) { /* two-side is enabled */
 625          render->stencil_back = stencil[1].func |
 626             (lima_stencil_op(stencil[1].fail_op) << 3) |
 627             (lima_stencil_op(stencil[1].zfail_op) << 6) |
 628             (lima_stencil_op(stencil[1].zpass_op) << 9) |
 629             (ref->ref_value[1] << 16) |
 630             (stencil[1].valuemask << 24);
 631          render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[1].writemask & 0xff) << 8;
 632       }
 633       /* TODO: Find out, what (render->stecil_test & 0xffff0000) is.
 634        * 0x00ff0000 is probably (float_to_ubyte(alpha->ref_value) << 16)
 635        * (render->multi_sample & 0x00000007 is probably the compare function
 636        * of glAlphaFunc then.
 637        */
 638    }
 639    else {
 640       /* Default values, when stencil is disabled:
 641        * stencil[0|1].valuemask = 0xff
 642        * stencil[0|1].func = PIPE_FUNC_ALWAYS
 643        * stencil[0|1].writemask = 0xff
 644        */
 645       render->stencil_front = 0xff000007;
 646       render->stencil_back = 0xff000007;
 647       render->stencil_test = 0x0000ffff;
 648    }
 649
 650    /* need more investigation */
 651    if (info->mode == PIPE_PRIM_POINTS)
 652       render->multi_sample = 0x0000F007;
 653    else if (info->mode < PIPE_PRIM_TRIANGLES)
 654       render->multi_sample = 0x0000F407;
 655    else
 656       render->multi_sample = 0x0000F807;
 657    if (ctx->framebuffer.base.samples)
 658       render->multi_sample |= 0x68;
 659
 660    render->shader_address =
 661       ctx->fs->bo->va | (((uint32_t *)ctx->fs->bo->map)[0] & 0x1F);
 662
 663    /* seems not needed */
 664    render->uniforms_address = 0x00000000;
 665
 666    render->textures_address = 0x00000000;
 667
 668    render->aux0 = (ctx->vs->varying_stride >> 3);
 669    render->aux1 = 0x00001000;
 670    if (ctx->blend->base.dither)
 671       render->aux1 |= 0x00002000;
 672
 673    if (fs->uses_discard) {
 674       early_z = false;
 675       pixel_kill = false;
 676    }
 677
 678    if (rt->blend_enable)
 679       pixel_kill = false;
 680
 681    if ((rt->colormask & PIPE_MASK_RGBA) != PIPE_MASK_RGBA)
 682       pixel_kill = false;
 683
 684    if (early_z)
 685       render->aux0 |= 0x300;
 686
 687    if (pixel_kill)
 688       render->aux0 |= 0x1000;
 689
 690    if (ctx->tex_stateobj.num_samplers) {
 691       render->textures_address =
 692          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_tex_desc);
 693       render->aux0 |= ctx->tex_stateobj.num_samplers << 14;
 694       render->aux0 |= 0x20;
 695    }
 696
 697    if (ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer) {
 698       render->uniforms_address =
 699          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array);
 700       uint32_t size = ctx->buffer_state[lima_ctx_buff_pp_uniform].size;
 701       uint32_t bits = 0;
 702       if (size >= 8) {
 703          bits = util_last_bit(size >> 3) - 1;
 704          bits += size & u_bit_consecutive(0, bits + 3) ? 1 : 0;
 705       }
 706       render->uniforms_address |= bits > 0xf ? 0xf : bits;
 707
 708       render->aux0 |= 0x80;
 709       render->aux1 |= 0x10000;
 710    }
 711
 712    if (ctx->vs->num_varyings) {
 713       render->varying_types = 0x00000000;
 714       render->varyings_address = ctx->gp_output->va +
 715                                  ctx->gp_output_varyings_offt;
 716       for (int i = 0, index = 0; i < ctx->vs->num_outputs; i++) {
 717          int val;
 718
 719          if (i == ctx->vs->gl_pos_idx ||
 720              i == ctx->vs->point_size_idx)
 721             continue;
 722
 723          struct lima_varying_info *v = ctx->vs->varying + i;
 724          if (v->component_size == 4)
 725             val = v->components > 2 ? 0 : 1;
 726          else
 727             val = v->components > 2 ? 2 : 3;
 728
 729          if (index < 10)
 730             render->varying_types |= val << (3 * index);
 731          else if (index == 10) {
 732             render->varying_types |= val << 30;
 733             render->varyings_address |= val >> 2;
 734          }
 735          else if (index == 11)
 736             render->varyings_address |= val << 1;
 737
 738          index++;
 739       }
 740    }
 741    else {
 742       render->varying_types = 0x00000000;
 743       render->varyings_address = 0x00000000;
 744    }
 745
 746    lima_dump_command_stream_print(
 747       render, sizeof(*render), false, "add render state at va %x\n",
 748       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 749
 750    lima_dump_rsw_command_stream_print(
 751       render, sizeof(*render), lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 752
 753 }
 754
 755 static void
 756 lima_update_gp_attribute_info(struct lima_context *ctx, const struct pipe_draw_info *info)
 757 {
 758    struct lima_submit *submit = lima_submit_get(ctx);
 759    struct lima_vertex_element_state *ve = ctx->vertex_elements;
 760    struct lima_context_vertex_buffer *vb = &ctx->vertex_buffers;
 761
 762    uint32_t *attribute =
 763       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_attribute_info,
 764                           MAX2(1, ve->num_elements) * 8);
 765
 766    int n = 0;
 767    for (int i = 0; i < ve->num_elements; i++) {
 768       struct pipe_vertex_element *pve = ve->pipe + i;
 769
 770       assert(pve->vertex_buffer_index < vb->count);
 771       assert(vb->enabled_mask & (1 << pve->vertex_buffer_index));
 772
 773       struct pipe_vertex_buffer *pvb = vb->vb + pve->vertex_buffer_index;
 774       struct lima_resource *res = lima_resource(pvb->buffer.resource);
 775
 776       lima_submit_add_bo(submit, LIMA_PIPE_GP, res->bo, LIMA_SUBMIT_BO_READ);
 777
 778       unsigned start = info->index_size ? (ctx->min_index + info->index_bias) : info->start;
 779       attribute[n++] = res->bo->va + pvb->buffer_offset + pve->src_offset
 780          + start * pvb->stride;
 781       attribute[n++] = (pvb->stride << 11) |
 782          (lima_pipe_format_to_attrib_type(pve->src_format) << 2) |
 783          (util_format_get_nr_components(pve->src_format) - 1);
 784    }
 785
 786    lima_dump_command_stream_print(
 787       attribute, n * 4, false, "update attribute info at va %x\n",
 788       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info));
 789 }
 790
 791 static void
 792 lima_update_gp_uniform(struct lima_context *ctx)
 793 {
 794    struct lima_context_constant_buffer *ccb =
 795       ctx->const_buffer + PIPE_SHADER_VERTEX;
 796    struct lima_vs_shader_state *vs = ctx->vs;
 797
 798    int size = vs->uniform_pending_offset + vs->constant_size + 32;
 799    void *vs_const_buff =
 800       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_uniform, size);
 801
 802    if (ccb->buffer)
 803       memcpy(vs_const_buff, ccb->buffer, ccb->size);
 804
 805    memcpy(vs_const_buff + vs->uniform_pending_offset,
 806           ctx->viewport.transform.scale,
 807           sizeof(ctx->viewport.transform.scale));
 808    memcpy(vs_const_buff + vs->uniform_pending_offset + 16,
 809           ctx->viewport.transform.translate,
 810           sizeof(ctx->viewport.transform.translate));
 811
 812    if (vs->constant)
 813       memcpy(vs_const_buff + vs->uniform_pending_offset + 32,
 814              vs->constant, vs->constant_size);
 815
 816    lima_dump_command_stream_print(
 817       vs_const_buff, size, true,
 818       "update gp uniform at va %x\n",
 819       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform));
 820 }
 821
 822 static void
 823 lima_update_pp_uniform(struct lima_context *ctx)
 824 {
 825    const float *const_buff = ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer;
 826    size_t const_buff_size = ctx->const_buffer[PIPE_SHADER_FRAGMENT].size / sizeof(float);
 827
 828    if (!const_buff)
 829       return;
 830
 831    uint16_t *fp16_const_buff =
 832       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform,
 833                           const_buff_size * sizeof(uint16_t));
 834
 835    uint32_t *array =
 836       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform_array, 4);
 837
 838    for (int i = 0; i < const_buff_size; i++)
 839        fp16_const_buff[i] = util_float_to_half(const_buff[i]);
 840
 841    *array = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform);
 842
 843    lima_dump_command_stream_print(
 844       fp16_const_buff, const_buff_size * 2, false, "add pp uniform data at va %x\n",
 845       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform));
 846    lima_dump_command_stream_print(
 847       array, 4, false, "add pp uniform info at va %x\n",
 848       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array));
 849 }
 850
 851 static void
 852 lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info)
 853 {
 854    struct lima_submit *submit = lima_submit_get(ctx);
 855    struct lima_screen *screen = lima_screen(ctx->base.screen);
 856    struct lima_vs_shader_state *vs = ctx->vs;
 857    uint32_t gp_output_size;
 858    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 859
 860    uint32_t *varying =
 861       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info,
 862                           vs->num_outputs * 8);
 863    int n = 0;
 864
 865    int offset = 0;
 866
 867    for (int i = 0; i < vs->num_outputs; i++) {
 868       struct lima_varying_info *v = vs->varying + i;
 869
 870       if (i == vs->gl_pos_idx ||
 871           i == vs->point_size_idx)
 872          continue;
 873
 874       int size = v->component_size * 4;
 875
 876       /* does component_size == 2 need to be 16 aligned? */
 877       if (v->component_size == 4)
 878          offset = align(offset, 16);
 879
 880       v->offset = offset;
 881       offset += size;
 882    }
 883
 884    vs->varying_stride = align(offset, 16);
 885
 886    /* gl_Position is always present, allocate space for it */
 887    gp_output_size = align(4 * 4 * num, 0x40);
 888
 889    /* Allocate space for varyings if there're any */
 890    if (vs->num_varyings) {
 891       ctx->gp_output_varyings_offt = gp_output_size;
 892       gp_output_size += align(vs->varying_stride * num, 0x40);
 893    }
 894
 895    /* Allocate space for gl_PointSize if it's there */
 896    if (vs->point_size_idx != -1) {
 897       ctx->gp_output_point_size_offt = gp_output_size;
 898       gp_output_size += 4 * num;
 899    }
 900
 901    /* gp_output can be too large for the suballocator, so create a
 902     * separate bo for it. The bo cache should prevent performance hit.
 903     */
 904    ctx->gp_output = lima_bo_create(screen, gp_output_size, 0);
 905    assert(ctx->gp_output);
 906    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->gp_output, LIMA_SUBMIT_BO_WRITE);
 907    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->gp_output, LIMA_SUBMIT_BO_READ);
 908
 909    for (int i = 0; i < vs->num_outputs; i++) {
 910       struct lima_varying_info *v = vs->varying + i;
 911
 912       if (i == vs->gl_pos_idx) {
 913          /* gl_Position */
 914          varying[n++] = ctx->gp_output->va;
 915          varying[n++] = 0x8020;
 916       } else if (i == vs->point_size_idx) {
 917          /* gl_PointSize */
 918          varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt;
 919          varying[n++] = 0x2021;
 920       } else {
 921          /* Varying */
 922          varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt +
 923                         v->offset;
 924          varying[n++] = (vs->varying_stride << 11) | (v->components - 1) |
 925             (v->component_size == 2 ? 0x0C : 0);
 926       }
 927    }
 928
 929    lima_dump_command_stream_print(
 930       varying, n * 4, false, "update varying info at va %x\n",
 931       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info));
 932 }
 933
 934 static void
 935 lima_draw_vbo_update(struct pipe_context *pctx,
 936                      const struct pipe_draw_info *info)
 937 {
 938    struct lima_context *ctx = lima_context(pctx);
 939    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 940    unsigned buffers = 0;
 941
 942    if (fb->base.zsbuf) {
 943       if (ctx->zsa->base.depth.enabled)
 944          buffers |= PIPE_CLEAR_DEPTH;
 945       if (ctx->zsa->base.stencil[0].enabled ||
 946           ctx->zsa->base.stencil[1].enabled)
 947          buffers |= PIPE_CLEAR_STENCIL;
 948    }
 949
 950    if (fb->base.nr_cbufs)
 951       buffers |= PIPE_CLEAR_COLOR0;
 952
 953    lima_update_submit_wb(ctx, buffers);
 954
 955    lima_update_gp_attribute_info(ctx, info);
 956
 957    if ((ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
 958         ctx->const_buffer[PIPE_SHADER_VERTEX].dirty) ||
 959        ctx->dirty & LIMA_CONTEXT_DIRTY_VIEWPORT ||
 960        ctx->dirty & LIMA_CONTEXT_DIRTY_SHADER_VERT) {
 961       lima_update_gp_uniform(ctx);
 962       ctx->const_buffer[PIPE_SHADER_VERTEX].dirty = false;
 963    }
 964
 965    lima_update_varying(ctx, info);
 966
 967    /* If it's zero scissor, don't build vs cmd list */
 968    if (!lima_is_scissor_zero(ctx))
 969       lima_pack_vs_cmd(ctx, info);
 970
 971    if (ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
 972        ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty) {
 973       lima_update_pp_uniform(ctx);
 974       ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty = false;
 975    }
 976
 977    lima_update_textures(ctx);
 978
 979    lima_pack_render_state(ctx, info);
 980    lima_pack_plbu_cmd(ctx, info);
 981
 982    if (ctx->gp_output) {
 983       lima_bo_unreference(ctx->gp_output); /* held by submit */
 984       ctx->gp_output = NULL;
 985    }
 986
 987    ctx->dirty = 0;
 988 }
 989
 990 static void
 991 lima_draw_vbo_indexed(struct pipe_context *pctx,
 992                       const struct pipe_draw_info *info)
 993 {
 994    struct lima_context *ctx = lima_context(pctx);
 995    struct lima_submit *submit = lima_submit_get(ctx);
 996    struct pipe_resource *indexbuf = NULL;
 997
 998    /* Mali Utgard GPU always need min/max index info for index draw,
 999     * compute it if upper layer does not do for us */
1000    if (info->max_index == ~0u)
1001       u_vbuf_get_minmax_index(pctx, info, &ctx->min_index, &ctx->max_index);
1002    else {
1003       ctx->min_index = info->min_index;
1004       ctx->max_index = info->max_index;
1005    }
1006
1007    if (info->has_user_indices) {
1008       util_upload_index_buffer(&ctx->base, info, &indexbuf, &ctx->index_offset, 0x40);
1009       ctx->index_res = lima_resource(indexbuf);
1010    }
1011    else {
1012       ctx->index_res = lima_resource(info->index.resource);
1013       ctx->index_offset = 0;
1014    }
1015
1016    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1017    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1018    lima_draw_vbo_update(pctx, info);
1019
1020    if (indexbuf)
1021       pipe_resource_reference(&indexbuf, NULL);
1022 }
1023
1024 static void
1025 lima_draw_vbo_count(struct pipe_context *pctx,
1026                     const struct pipe_draw_info *info)
1027 {
1028    static const uint32_t max_verts = 65535;
1029
1030    struct pipe_draw_info local_info = *info;
1031    unsigned start = info->start;
1032    unsigned count = info->count;
1033
1034    while (count) {
1035       unsigned this_count = count;
1036       unsigned step;
1037
1038       u_split_draw(info, max_verts, &this_count, &step);
1039
1040       local_info.start = start;
1041       local_info.count = this_count;
1042
1043       lima_draw_vbo_update(pctx, &local_info);
1044
1045       count -= step;
1046       start += step;
1047    }
1048 }
1049
1050 static void
1051 lima_draw_vbo(struct pipe_context *pctx,
1052               const struct pipe_draw_info *info)
1053 {
1054    /* check if draw mode and vertex/index count match,
1055     * otherwise gp will hang */
1056    if (!u_trim_pipe_prim(info->mode, (unsigned*)&info->count)) {
1057       debug_printf("draw mode and vertex/index count mismatch\n");
1058       return;
1059    }
1060
1061    struct lima_context *ctx = lima_context(pctx);
1062
1063    if (!ctx->vs || !ctx->fs) {
1064       debug_warn_once("no shader, skip draw\n");
1065       return;
1066    }
1067
1068    if (!lima_update_vs_state(ctx) || !lima_update_fs_state(ctx))
1069       return;
1070
1071    struct lima_submit *submit = lima_submit_get(ctx);
1072
1073    lima_dump_command_stream_print(
1074       ctx->vs->bo->map, ctx->vs->shader_size, false,
1075       "add vs at va %x\n", ctx->vs->bo->va);
1076
1077    lima_dump_command_stream_print(
1078       ctx->fs->bo->map, ctx->fs->shader_size, false,
1079       "add fs at va %x\n", ctx->fs->bo->va);
1080
1081    lima_submit_add_bo(submit, LIMA_PIPE_GP, ctx->vs->bo, LIMA_SUBMIT_BO_READ);
1082    lima_submit_add_bo(submit, LIMA_PIPE_PP, ctx->fs->bo, LIMA_SUBMIT_BO_READ);
1083
1084    if (info->index_size)
1085       lima_draw_vbo_indexed(pctx, info);
1086    else
1087       lima_draw_vbo_count(pctx, info);
1088 }
1089
1090 void
1091 lima_draw_init(struct lima_context *ctx)
1092 {
1093    ctx->base.clear = lima_clear;
1094    ctx->base.draw_vbo = lima_draw_vbo;
1095 }