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_job.h"
  44 #include "lima_texture.h"
  45 #include "lima_util.h"
  46 #include "lima_gpu.h"
  47
  48 #include "pan_minmax_cache.h"
  49
  50 #include <drm-uapi/lima_drm.h>
  51
  52 static void
  53 lima_clip_scissor_to_viewport(struct lima_context *ctx)
  54 {
  55    struct lima_context_framebuffer *fb = &ctx->framebuffer;
  56    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
  57    int viewport_left, viewport_right, viewport_bottom, viewport_top;
  58
  59    if (ctx->rasterizer && ctx->rasterizer->base.scissor) {
  60       struct pipe_scissor_state *scissor = &ctx->scissor;
  61       cscissor->minx = scissor->minx;
  62       cscissor->maxx = scissor->maxx;
  63       cscissor->miny = scissor->miny;
  64       cscissor->maxy = scissor->maxy;
  65    } else {
  66       cscissor->minx = 0;
  67       cscissor->maxx = fb->base.width;
  68       cscissor->miny = 0;
  69       cscissor->maxy = fb->base.height;
  70    }
  71
  72    viewport_left = MAX2(ctx->viewport.left, 0);
  73    cscissor->minx = MAX2(cscissor->minx, viewport_left);
  74    viewport_right = MIN2(ctx->viewport.right, fb->base.width);
  75    cscissor->maxx = MIN2(cscissor->maxx, viewport_right);
  76    if (cscissor->minx > cscissor->maxx)
  77       cscissor->minx = cscissor->maxx;
  78
  79    viewport_bottom = MAX2(ctx->viewport.bottom, 0);
  80    cscissor->miny = MAX2(cscissor->miny, viewport_bottom);
  81    viewport_top = MIN2(ctx->viewport.top, fb->base.height);
  82    cscissor->maxy = MIN2(cscissor->maxy, viewport_top);
  83    if (cscissor->miny > cscissor->maxy)
  84       cscissor->miny = cscissor->maxy;
  85 }
  86
  87 static bool
  88 lima_is_scissor_zero(struct lima_context *ctx)
  89 {
  90    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
  91
  92    return cscissor->minx == cscissor->maxx || cscissor->miny == cscissor->maxy;
  93 }
  94
  95 static void
  96 lima_update_job_wb(struct lima_context *ctx, unsigned buffers)
  97 {
  98    struct lima_job *job = lima_job_get(ctx);
  99    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 100
 101    /* add to job when the buffer is dirty and resolve is clear (not added before) */
 102    if (fb->base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0) &&
 103        !(job->resolve & PIPE_CLEAR_COLOR0)) {
 104       struct lima_resource *res = lima_resource(fb->base.cbufs[0]->texture);
 105       lima_flush_job_accessing_bo(ctx, res->bo, true);
 106       _mesa_hash_table_insert(ctx->write_jobs, &res->base, job);
 107       lima_job_add_bo(job, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
 108    }
 109
 110    /* add to job when the buffer is dirty and resolve is clear (not added before) */
 111    if (fb->base.zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
 112        !(job->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
 113       struct lima_resource *res = lima_resource(fb->base.zsbuf->texture);
 114       lima_flush_job_accessing_bo(ctx, res->bo, true);
 115       _mesa_hash_table_insert(ctx->write_jobs, &res->base, job);
 116       lima_job_add_bo(job, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
 117    }
 118
 119    job->resolve |= buffers;
 120 }
 121
 122 static void
 123 lima_damage_rect_union(struct pipe_scissor_state *rect,
 124                        unsigned minx, unsigned maxx,
 125                        unsigned miny, unsigned maxy)
 126 {
 127    rect->minx = MIN2(rect->minx, minx);
 128    rect->miny = MIN2(rect->miny, miny);
 129    rect->maxx = MAX2(rect->maxx, maxx);
 130    rect->maxy = MAX2(rect->maxy, maxy);
 131 }
 132
 133 static void
 134 lima_clear(struct pipe_context *pctx, unsigned buffers, const struct pipe_scissor_state *scissor_state,
 135            const union pipe_color_union *color, double depth, unsigned stencil)
 136 {
 137    struct lima_context *ctx = lima_context(pctx);
 138    struct lima_job *job = lima_job_get(ctx);
 139
 140    /* flush if this job already contains any draw, otherwise multi clear can be
 141     * combined into a single job */
 142    if (lima_job_has_draw_pending(job)) {
 143       lima_do_job(job);
 144       job = lima_job_get(ctx);
 145    }
 146
 147    lima_update_job_wb(ctx, buffers);
 148
 149    /* no need to reload if cleared */
 150    if (ctx->framebuffer.base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0)) {
 151       struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
 152       surf->reload &= ~PIPE_CLEAR_COLOR0;
 153    }
 154
 155    struct lima_job_clear *clear = &job->clear;
 156    clear->buffers = buffers;
 157
 158    if (buffers & PIPE_CLEAR_COLOR0) {
 159       clear->color_8pc =
 160          ((uint32_t)float_to_ubyte(color->f[3]) << 24) |
 161          ((uint32_t)float_to_ubyte(color->f[2]) << 16) |
 162          ((uint32_t)float_to_ubyte(color->f[1]) << 8) |
 163          float_to_ubyte(color->f[0]);
 164
 165       clear->color_16pc =
 166          ((uint64_t)float_to_ushort(color->f[3]) << 48) |
 167          ((uint64_t)float_to_ushort(color->f[2]) << 32) |
 168          ((uint64_t)float_to_ushort(color->f[1]) << 16) |
 169          float_to_ushort(color->f[0]);
 170    }
 171
 172    struct lima_surface *zsbuf = lima_surface(ctx->framebuffer.base.zsbuf);
 173
 174    if (buffers & PIPE_CLEAR_DEPTH) {
 175       clear->depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
 176       if (zsbuf)
 177          zsbuf->reload &= ~PIPE_CLEAR_DEPTH;
 178    } else
 179       clear->depth = 0x00ffffff;
 180
 181    if (buffers & PIPE_CLEAR_STENCIL) {
 182       clear->stencil = stencil;
 183       if (zsbuf)
 184          zsbuf->reload &= ~PIPE_CLEAR_STENCIL;
 185    }
 186
 187    ctx->dirty |= LIMA_CONTEXT_DIRTY_CLEAR;
 188
 189    lima_damage_rect_union(&job->damage_rect,
 190                           0, ctx->framebuffer.base.width,
 191                           0, ctx->framebuffer.base.height);
 192 }
 193
 194 enum lima_attrib_type {
 195    LIMA_ATTRIB_FLOAT = 0x000,
 196    /* todo: find out what lives here. */
 197    LIMA_ATTRIB_I16   = 0x004,
 198    LIMA_ATTRIB_U16   = 0x005,
 199    LIMA_ATTRIB_I8    = 0x006,
 200    LIMA_ATTRIB_U8    = 0x007,
 201    LIMA_ATTRIB_I8N   = 0x008,
 202    LIMA_ATTRIB_U8N   = 0x009,
 203    LIMA_ATTRIB_I16N  = 0x00A,
 204    LIMA_ATTRIB_U16N  = 0x00B,
 205    /* todo: where is the 32 int */
 206    /* todo: find out what lives here. */
 207    LIMA_ATTRIB_FIXED = 0x101
 208 };
 209
 210 static enum lima_attrib_type
 211 lima_pipe_format_to_attrib_type(enum pipe_format format)
 212 {
 213    const struct util_format_description *desc = util_format_description(format);
 214    int i = util_format_get_first_non_void_channel(format);
 215    const struct util_format_channel_description *c = desc->channel + i;
 216
 217    switch (c->type) {
 218    case UTIL_FORMAT_TYPE_FLOAT:
 219       return LIMA_ATTRIB_FLOAT;
 220    case UTIL_FORMAT_TYPE_FIXED:
 221       return LIMA_ATTRIB_FIXED;
 222    case UTIL_FORMAT_TYPE_SIGNED:
 223       if (c->size == 8) {
 224          if (c->normalized)
 225             return LIMA_ATTRIB_I8N;
 226          else
 227             return LIMA_ATTRIB_I8;
 228       }
 229       else if (c->size == 16) {
 230          if (c->normalized)
 231             return LIMA_ATTRIB_I16N;
 232          else
 233             return LIMA_ATTRIB_I16;
 234       }
 235       break;
 236    case UTIL_FORMAT_TYPE_UNSIGNED:
 237       if (c->size == 8) {
 238          if (c->normalized)
 239             return LIMA_ATTRIB_U8N;
 240          else
 241             return LIMA_ATTRIB_U8;
 242       }
 243       else if (c->size == 16) {
 244          if (c->normalized)
 245             return LIMA_ATTRIB_U16N;
 246          else
 247             return LIMA_ATTRIB_U16;
 248       }
 249       break;
 250    }
 251
 252    return LIMA_ATTRIB_FLOAT;
 253 }
 254
 255 static void
 256 lima_pack_vs_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 257 {
 258    struct lima_job *job = lima_job_get(ctx);
 259
 260    VS_CMD_BEGIN(&job->vs_cmd_array, 24);
 261
 262    if (!info->index_size) {
 263       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1();
 264       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2();
 265    }
 266
 267    int uniform_size = ctx->vs->uniform_pending_offset + ctx->vs->constant_size + 32;
 268    VS_CMD_UNIFORMS_ADDRESS(
 269       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform),
 270       align(uniform_size, 16));
 271
 272    VS_CMD_SHADER_ADDRESS(ctx->vs->bo->va, ctx->vs->shader_size);
 273    VS_CMD_SHADER_INFO(ctx->vs->prefetch, ctx->vs->shader_size);
 274
 275    int num_outputs = ctx->vs->num_outputs;
 276    int num_attributes = ctx->vertex_elements->num_elements;
 277    VS_CMD_VARYING_ATTRIBUTE_COUNT(num_outputs, MAX2(1, num_attributes));
 278
 279    VS_CMD_UNKNOWN1();
 280
 281    VS_CMD_ATTRIBUTES_ADDRESS(
 282       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info),
 283       MAX2(1, num_attributes));
 284
 285    VS_CMD_VARYINGS_ADDRESS(
 286       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info),
 287       num_outputs);
 288
 289    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 290    VS_CMD_DRAW(num, info->index_size);
 291
 292    VS_CMD_UNKNOWN2();
 293
 294    VS_CMD_ARRAYS_SEMAPHORE_END(info->index_size);
 295
 296    VS_CMD_END();
 297 }
 298
 299 static void
 300 lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
 301 {
 302    struct lima_vs_shader_state *vs = ctx->vs;
 303    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
 304    struct lima_job *job = lima_job_get(ctx);
 305    PLBU_CMD_BEGIN(&job->plbu_cmd_array, 32);
 306
 307    PLBU_CMD_VIEWPORT_LEFT(fui(ctx->viewport.left));
 308    PLBU_CMD_VIEWPORT_RIGHT(fui(ctx->viewport.right));
 309    PLBU_CMD_VIEWPORT_BOTTOM(fui(ctx->viewport.bottom));
 310    PLBU_CMD_VIEWPORT_TOP(fui(ctx->viewport.top));
 311
 312    if (!info->index_size)
 313       PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN();
 314
 315    int cf = ctx->rasterizer->base.cull_face;
 316    int ccw = ctx->rasterizer->base.front_ccw;
 317    uint32_t cull = 0;
 318    bool force_point_size = false;
 319
 320    if (cf != PIPE_FACE_NONE) {
 321       if (cf & PIPE_FACE_FRONT)
 322          cull |= ccw ? 0x00040000 : 0x00020000;
 323       if (cf & PIPE_FACE_BACK)
 324          cull |= ccw ? 0x00020000 : 0x00040000;
 325    }
 326
 327    /* Specify point size with PLBU command if shader doesn't write */
 328    if (info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1)
 329       force_point_size = true;
 330
 331    /* Specify line width with PLBU command for lines */
 332    if (info->mode > PIPE_PRIM_POINTS && info->mode < PIPE_PRIM_TRIANGLES)
 333       force_point_size = true;
 334
 335    PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, info->index_size);
 336
 337    PLBU_CMD_RSW_VERTEX_ARRAY(
 338       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw),
 339       ctx->gp_output->va);
 340
 341    /* TODO
 342     * - we should set it only for the first draw that enabled the scissor and for
 343     *   latter draw only if scissor is dirty
 344     */
 345
 346    assert(cscissor->minx < cscissor->maxx && cscissor->miny < cscissor->maxy);
 347    PLBU_CMD_SCISSORS(cscissor->minx, cscissor->maxx, cscissor->miny, cscissor->maxy);
 348
 349    lima_damage_rect_union(&job->damage_rect, cscissor->minx, cscissor->maxx,
 350                           cscissor->miny, cscissor->maxy);
 351
 352    PLBU_CMD_UNKNOWN1();
 353
 354    PLBU_CMD_DEPTH_RANGE_NEAR(fui(ctx->viewport.near));
 355    PLBU_CMD_DEPTH_RANGE_FAR(fui(ctx->viewport.far));
 356
 357    if ((info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1) ||
 358        ((info->mode >= PIPE_PRIM_LINES) && (info->mode < PIPE_PRIM_TRIANGLES)))
 359    {
 360       uint32_t v = info->mode == PIPE_PRIM_POINTS ?
 361          fui(ctx->rasterizer->base.point_size) : fui(ctx->rasterizer->base.line_width);
 362       PLBU_CMD_LOW_PRIM_SIZE(v);
 363    }
 364
 365    if (info->index_size) {
 366       PLBU_CMD_INDEXED_DEST(ctx->gp_output->va);
 367       if (vs->point_size_idx != -1)
 368          PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt);
 369
 370       PLBU_CMD_INDICES(ctx->index_res->bo->va + info->start * info->index_size + ctx->index_offset);
 371    }
 372    else {
 373       /* can this make the attribute info static? */
 374       PLBU_CMD_DRAW_ARRAYS(info->mode, info->start, info->count);
 375    }
 376
 377    PLBU_CMD_ARRAYS_SEMAPHORE_END();
 378
 379    if (info->index_size)
 380       PLBU_CMD_DRAW_ELEMENTS(info->mode, ctx->min_index, info->count);
 381
 382    PLBU_CMD_END();
 383 }
 384
 385 static int
 386 lima_blend_func(enum pipe_blend_func pipe)
 387 {
 388    switch (pipe) {
 389    case PIPE_BLEND_ADD:
 390       return 2;
 391    case PIPE_BLEND_SUBTRACT:
 392       return 0;
 393    case PIPE_BLEND_REVERSE_SUBTRACT:
 394       return 1;
 395    case PIPE_BLEND_MIN:
 396       return 4;
 397    case PIPE_BLEND_MAX:
 398       return 5;
 399    }
 400    return -1;
 401 }
 402
 403 static int
 404 lima_blend_factor_has_alpha(enum pipe_blendfactor pipe)
 405 {
 406    /* Bit 4 is set if the blendfactor uses alpha */
 407    switch (pipe) {
 408    case PIPE_BLENDFACTOR_SRC_ALPHA:
 409    case PIPE_BLENDFACTOR_DST_ALPHA:
 410    case PIPE_BLENDFACTOR_CONST_ALPHA:
 411    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 412    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 413    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 414       return 1;
 415
 416    case PIPE_BLENDFACTOR_SRC_COLOR:
 417    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 418    case PIPE_BLENDFACTOR_DST_COLOR:
 419    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 420    case PIPE_BLENDFACTOR_CONST_COLOR:
 421    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 422    case PIPE_BLENDFACTOR_ZERO:
 423    case PIPE_BLENDFACTOR_ONE:
 424    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 425       return 0;
 426
 427    case PIPE_BLENDFACTOR_SRC1_COLOR:
 428    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 429    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 430    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 431       return -1; /* not supported */
 432    }
 433    return -1;
 434 }
 435
 436 static int
 437 lima_blend_factor_is_inv(enum pipe_blendfactor pipe)
 438 {
 439    /* Bit 3 is set if the blendfactor type is inverted */
 440    switch (pipe) {
 441    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 442    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 443    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 444    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 445    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 446    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 447    case PIPE_BLENDFACTOR_ONE:
 448       return 1;
 449
 450    case PIPE_BLENDFACTOR_SRC_COLOR:
 451    case PIPE_BLENDFACTOR_SRC_ALPHA:
 452    case PIPE_BLENDFACTOR_DST_COLOR:
 453    case PIPE_BLENDFACTOR_DST_ALPHA:
 454    case PIPE_BLENDFACTOR_CONST_COLOR:
 455    case PIPE_BLENDFACTOR_CONST_ALPHA:
 456    case PIPE_BLENDFACTOR_ZERO:
 457    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 458       return 0;
 459
 460    case PIPE_BLENDFACTOR_SRC1_COLOR:
 461    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 462    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 463    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 464       return -1; /* not supported */
 465    }
 466    return -1;
 467 }
 468
 469 static int
 470 lima_blend_factor(enum pipe_blendfactor pipe)
 471 {
 472    /* Bits 0-2 indicate the blendfactor type */
 473    switch (pipe) {
 474    case PIPE_BLENDFACTOR_SRC_COLOR:
 475    case PIPE_BLENDFACTOR_SRC_ALPHA:
 476    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 477    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 478       return 0;
 479
 480    case PIPE_BLENDFACTOR_DST_COLOR:
 481    case PIPE_BLENDFACTOR_DST_ALPHA:
 482    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 483    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 484       return 1;
 485
 486    case PIPE_BLENDFACTOR_CONST_COLOR:
 487    case PIPE_BLENDFACTOR_CONST_ALPHA:
 488    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 489    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 490       return 2;
 491
 492    case PIPE_BLENDFACTOR_ZERO:
 493    case PIPE_BLENDFACTOR_ONE:
 494       return 3;
 495
 496    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 497       return 4;
 498
 499    case PIPE_BLENDFACTOR_SRC1_COLOR:
 500    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 501    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 502    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 503       return -1; /* not supported */
 504    }
 505    return -1;
 506 }
 507
 508 static int
 509 lima_calculate_alpha_blend(enum pipe_blend_func rgb_func, enum pipe_blend_func alpha_func,
 510                            enum pipe_blendfactor rgb_src_factor, enum pipe_blendfactor rgb_dst_factor,
 511                            enum pipe_blendfactor alpha_src_factor, enum pipe_blendfactor alpha_dst_factor)
 512 {
 513    /* PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE has to be changed to PIPE_BLENDFACTOR_ONE
 514     * if it is set for alpha_src.
 515     */
 516    if (alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
 517       alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 518
 519    return lima_blend_func(rgb_func) |
 520       (lima_blend_func(alpha_func) << 3) |
 521
 522       (lima_blend_factor(rgb_src_factor) << 6) |
 523       (lima_blend_factor_is_inv(rgb_src_factor) << 9) |
 524       (lima_blend_factor_has_alpha(rgb_src_factor) << 10) |
 525
 526       (lima_blend_factor(rgb_dst_factor) << 11) |
 527       (lima_blend_factor_is_inv(rgb_dst_factor) << 14) |
 528       (lima_blend_factor_has_alpha(rgb_dst_factor) << 15) |
 529
 530       (lima_blend_factor(alpha_src_factor) << 16) |
 531       (lima_blend_factor_is_inv(alpha_src_factor) << 19) |
 532
 533       (lima_blend_factor(alpha_dst_factor) << 20) |
 534       (lima_blend_factor_is_inv(alpha_dst_factor) << 23) |
 535       0x0C000000; /* need to check if this is GLESv1 glAlphaFunc */
 536 }
 537
 538 static int
 539 lima_stencil_op(enum pipe_stencil_op pipe)
 540 {
 541    switch (pipe) {
 542    case PIPE_STENCIL_OP_KEEP:
 543       return 0;
 544    case PIPE_STENCIL_OP_ZERO:
 545       return 2;
 546    case PIPE_STENCIL_OP_REPLACE:
 547       return 1;
 548    case PIPE_STENCIL_OP_INCR:
 549       return 6;
 550    case PIPE_STENCIL_OP_DECR:
 551       return 7;
 552    case PIPE_STENCIL_OP_INCR_WRAP:
 553       return 4;
 554    case PIPE_STENCIL_OP_DECR_WRAP:
 555       return 5;
 556    case PIPE_STENCIL_OP_INVERT:
 557       return 3;
 558    }
 559    return -1;
 560 }
 561
 562 static unsigned
 563 lima_calculate_depth_test(struct pipe_depth_state *depth, struct pipe_rasterizer_state *rst)
 564 {
 565    int offset_scale = 0, offset_units = 0;
 566    enum pipe_compare_func func = (depth->enabled ? depth->func : PIPE_FUNC_ALWAYS);
 567
 568    offset_scale = CLAMP(rst->offset_scale * 4, -128, 127);
 569    if (offset_scale < 0)
 570       offset_scale += 0x100;
 571
 572    offset_units = CLAMP(rst->offset_units * 2, -128, 127);
 573    if (offset_units < 0)
 574       offset_units += 0x100;
 575
 576    return (depth->enabled && depth->writemask) |
 577       ((int)func << 1) |
 578       (offset_scale << 16) |
 579       (offset_units << 24) |
 580       0x30; /* find out what is this */
 581 }
 582
 583 static void
 584 lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *info)
 585 {
 586    struct lima_fs_shader_state *fs = ctx->fs;
 587    struct lima_render_state *render =
 588       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_plb_rsw,
 589                           sizeof(*render));
 590    bool early_z = true;
 591    bool pixel_kill = true;
 592
 593    /* do hw support RGBA independ blend?
 594     * PIPE_CAP_INDEP_BLEND_ENABLE
 595     *
 596     * how to handle the no cbuf only zbuf case?
 597     */
 598    struct pipe_rt_blend_state *rt = ctx->blend->base.rt;
 599    render->blend_color_bg = float_to_ubyte(ctx->blend_color.color[2]) |
 600       (float_to_ubyte(ctx->blend_color.color[1]) << 16);
 601    render->blend_color_ra = float_to_ubyte(ctx->blend_color.color[0]) |
 602       (float_to_ubyte(ctx->blend_color.color[3]) << 16);
 603
 604    if (rt->blend_enable) {
 605       render->alpha_blend = lima_calculate_alpha_blend(rt->rgb_func, rt->alpha_func,
 606          rt->rgb_src_factor, rt->rgb_dst_factor,
 607          rt->alpha_src_factor, rt->alpha_dst_factor);
 608    }
 609    else {
 610       /*
 611        * Special handling for blending disabled.
 612        * Binary driver is generating the same alpha_value,
 613        * as when we would just enable blending, without changing/setting any blend equation/params.
 614        * Normaly in this case mesa would set all rt fields (func/factor) to zero.
 615        */
 616       render->alpha_blend = lima_calculate_alpha_blend(PIPE_BLEND_ADD, PIPE_BLEND_ADD,
 617          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO,
 618          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO);
 619    }
 620
 621    render->alpha_blend |= (rt->colormask & PIPE_MASK_RGBA) << 28;
 622
 623    struct pipe_rasterizer_state *rst = &ctx->rasterizer->base;
 624    struct pipe_depth_state *depth = &ctx->zsa->base.depth;
 625    render->depth_test = lima_calculate_depth_test(depth, rst);
 626
 627    ushort far, near;
 628
 629    near = float_to_ushort(ctx->viewport.near);
 630    far = float_to_ushort(ctx->viewport.far);
 631
 632    /* Subtract epsilon from 'near' if far == near. Make sure we don't get overflow */
 633    if ((far == near) && (near != 0))
 634          near--;
 635
 636    /* overlap with plbu? any place can remove one? */
 637    render->depth_range = near | (far << 16);
 638
 639    struct pipe_stencil_state *stencil = ctx->zsa->base.stencil;
 640    struct pipe_stencil_ref *ref = &ctx->stencil_ref;
 641
 642    if (stencil[0].enabled) { /* stencil is enabled */
 643       render->stencil_front = stencil[0].func |
 644          (lima_stencil_op(stencil[0].fail_op) << 3) |
 645          (lima_stencil_op(stencil[0].zfail_op) << 6) |
 646          (lima_stencil_op(stencil[0].zpass_op) << 9) |
 647          (ref->ref_value[0] << 16) |
 648          (stencil[0].valuemask << 24);
 649       render->stencil_back = render->stencil_front;
 650       render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[0].writemask & 0xff) << 8;
 651       if (stencil[1].enabled) { /* two-side is enabled */
 652          render->stencil_back = stencil[1].func |
 653             (lima_stencil_op(stencil[1].fail_op) << 3) |
 654             (lima_stencil_op(stencil[1].zfail_op) << 6) |
 655             (lima_stencil_op(stencil[1].zpass_op) << 9) |
 656             (ref->ref_value[1] << 16) |
 657             (stencil[1].valuemask << 24);
 658          render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[1].writemask & 0xff) << 8;
 659       }
 660       /* TODO: Find out, what (render->stecil_test & 0xffff0000) is.
 661        * 0x00ff0000 is probably (float_to_ubyte(alpha->ref_value) << 16)
 662        * (render->multi_sample & 0x00000007 is probably the compare function
 663        * of glAlphaFunc then.
 664        */
 665    }
 666    else {
 667       /* Default values, when stencil is disabled:
 668        * stencil[0|1].valuemask = 0xff
 669        * stencil[0|1].func = PIPE_FUNC_ALWAYS
 670        * stencil[0|1].writemask = 0xff
 671        */
 672       render->stencil_front = 0xff000007;
 673       render->stencil_back = 0xff000007;
 674       render->stencil_test = 0x0000ffff;
 675    }
 676
 677    /* need more investigation */
 678    if (info->mode == PIPE_PRIM_POINTS)
 679       render->multi_sample = 0x0000F007;
 680    else if (info->mode < PIPE_PRIM_TRIANGLES)
 681       render->multi_sample = 0x0000F407;
 682    else
 683       render->multi_sample = 0x0000F807;
 684    if (ctx->framebuffer.base.samples)
 685       render->multi_sample |= 0x68;
 686
 687    render->shader_address =
 688       ctx->fs->bo->va | (((uint32_t *)ctx->fs->bo->map)[0] & 0x1F);
 689
 690    /* seems not needed */
 691    render->uniforms_address = 0x00000000;
 692
 693    render->textures_address = 0x00000000;
 694
 695    render->aux0 = (ctx->vs->varying_stride >> 3);
 696    render->aux1 = 0x00001000;
 697    if (ctx->blend->base.dither)
 698       render->aux1 |= 0x00002000;
 699
 700    if (fs->uses_discard) {
 701       early_z = false;
 702       pixel_kill = false;
 703    }
 704
 705    if (rt->blend_enable)
 706       pixel_kill = false;
 707
 708    if ((rt->colormask & PIPE_MASK_RGBA) != PIPE_MASK_RGBA)
 709       pixel_kill = false;
 710
 711    if (early_z)
 712       render->aux0 |= 0x300;
 713
 714    if (pixel_kill)
 715       render->aux0 |= 0x1000;
 716
 717    if (ctx->tex_stateobj.num_samplers) {
 718       render->textures_address =
 719          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_tex_desc);
 720       render->aux0 |= ctx->tex_stateobj.num_samplers << 14;
 721       render->aux0 |= 0x20;
 722    }
 723
 724    if (ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer) {
 725       render->uniforms_address =
 726          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array);
 727       uint32_t size = ctx->buffer_state[lima_ctx_buff_pp_uniform].size;
 728       uint32_t bits = 0;
 729       if (size >= 8) {
 730          bits = util_last_bit(size >> 3) - 1;
 731          bits += size & u_bit_consecutive(0, bits + 3) ? 1 : 0;
 732       }
 733       render->uniforms_address |= bits > 0xf ? 0xf : bits;
 734
 735       render->aux0 |= 0x80;
 736       render->aux1 |= 0x10000;
 737    }
 738
 739    if (ctx->vs->num_varyings) {
 740       render->varying_types = 0x00000000;
 741       render->varyings_address = ctx->gp_output->va +
 742                                  ctx->gp_output_varyings_offt;
 743       for (int i = 0, index = 0; i < ctx->vs->num_outputs; i++) {
 744          int val;
 745
 746          if (i == ctx->vs->gl_pos_idx ||
 747              i == ctx->vs->point_size_idx)
 748             continue;
 749
 750          struct lima_varying_info *v = ctx->vs->varying + i;
 751          if (v->component_size == 4)
 752             val = v->components > 2 ? 0 : 1;
 753          else
 754             val = v->components > 2 ? 2 : 3;
 755
 756          if (index < 10)
 757             render->varying_types |= val << (3 * index);
 758          else if (index == 10) {
 759             render->varying_types |= val << 30;
 760             render->varyings_address |= val >> 2;
 761          }
 762          else if (index == 11)
 763             render->varyings_address |= val << 1;
 764
 765          index++;
 766       }
 767    }
 768    else {
 769       render->varying_types = 0x00000000;
 770       render->varyings_address = 0x00000000;
 771    }
 772
 773    struct lima_job *job = lima_job_get(ctx);
 774
 775    lima_dump_command_stream_print(
 776       job->dump, render, sizeof(*render),
 777       false, "add render state at va %x\n",
 778       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 779
 780    lima_dump_rsw_command_stream_print(
 781       job->dump, render, sizeof(*render),
 782       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
 783 }
 784
 785 static void
 786 lima_update_gp_attribute_info(struct lima_context *ctx, const struct pipe_draw_info *info)
 787 {
 788    struct lima_job *job = lima_job_get(ctx);
 789    struct lima_vertex_element_state *ve = ctx->vertex_elements;
 790    struct lima_context_vertex_buffer *vb = &ctx->vertex_buffers;
 791
 792    uint32_t *attribute =
 793       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_attribute_info,
 794                           MAX2(1, ve->num_elements) * 8);
 795
 796    int n = 0;
 797    for (int i = 0; i < ve->num_elements; i++) {
 798       struct pipe_vertex_element *pve = ve->pipe + i;
 799
 800       assert(pve->vertex_buffer_index < vb->count);
 801       assert(vb->enabled_mask & (1 << pve->vertex_buffer_index));
 802
 803       struct pipe_vertex_buffer *pvb = vb->vb + pve->vertex_buffer_index;
 804       struct lima_resource *res = lima_resource(pvb->buffer.resource);
 805
 806       lima_job_add_bo(job, LIMA_PIPE_GP, res->bo, LIMA_SUBMIT_BO_READ);
 807
 808       unsigned start = info->index_size ? (ctx->min_index + info->index_bias) : info->start;
 809       attribute[n++] = res->bo->va + pvb->buffer_offset + pve->src_offset
 810          + start * pvb->stride;
 811       attribute[n++] = (pvb->stride << 11) |
 812          (lima_pipe_format_to_attrib_type(pve->src_format) << 2) |
 813          (util_format_get_nr_components(pve->src_format) - 1);
 814    }
 815
 816    lima_dump_command_stream_print(
 817       job->dump, attribute, n * 4, false, "update attribute info at va %x\n",
 818       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info));
 819 }
 820
 821 static void
 822 lima_update_gp_uniform(struct lima_context *ctx)
 823 {
 824    struct lima_context_constant_buffer *ccb =
 825       ctx->const_buffer + PIPE_SHADER_VERTEX;
 826    struct lima_vs_shader_state *vs = ctx->vs;
 827
 828    int size = vs->uniform_pending_offset + vs->constant_size + 32;
 829    void *vs_const_buff =
 830       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_uniform, size);
 831
 832    if (ccb->buffer)
 833       memcpy(vs_const_buff, ccb->buffer, ccb->size);
 834
 835    memcpy(vs_const_buff + vs->uniform_pending_offset,
 836           ctx->viewport.transform.scale,
 837           sizeof(ctx->viewport.transform.scale));
 838    memcpy(vs_const_buff + vs->uniform_pending_offset + 16,
 839           ctx->viewport.transform.translate,
 840           sizeof(ctx->viewport.transform.translate));
 841
 842    if (vs->constant)
 843       memcpy(vs_const_buff + vs->uniform_pending_offset + 32,
 844              vs->constant, vs->constant_size);
 845
 846    struct lima_job *job = lima_job_get(ctx);
 847
 848    if (lima_debug & LIMA_DEBUG_GP) {
 849       float *vs_const_buff_f = vs_const_buff;
 850       printf("gp uniforms:\n");
 851       for (int i = 0; i < (size / sizeof(float)); i++) {
 852          if ((i % 4) == 0)
 853             printf("%4d:", i / 4);
 854          printf(" %8.4f", vs_const_buff_f[i]);
 855          if ((i % 4) == 3)
 856             printf("\n");
 857       }
 858       printf("\n");
 859    }
 860
 861    lima_dump_command_stream_print(
 862       job->dump, vs_const_buff, size, true,
 863       "update gp uniform at va %x\n",
 864       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform));
 865 }
 866
 867 static void
 868 lima_update_pp_uniform(struct lima_context *ctx)
 869 {
 870    const float *const_buff = ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer;
 871    size_t const_buff_size = ctx->const_buffer[PIPE_SHADER_FRAGMENT].size / sizeof(float);
 872
 873    if (!const_buff)
 874       return;
 875
 876    uint16_t *fp16_const_buff =
 877       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform,
 878                           const_buff_size * sizeof(uint16_t));
 879
 880    uint32_t *array =
 881       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform_array, 4);
 882
 883    for (int i = 0; i < const_buff_size; i++)
 884        fp16_const_buff[i] = util_float_to_half(const_buff[i]);
 885
 886    *array = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform);
 887
 888    struct lima_job *job = lima_job_get(ctx);
 889
 890    lima_dump_command_stream_print(
 891       job->dump, fp16_const_buff, const_buff_size * 2,
 892       false, "add pp uniform data at va %x\n",
 893       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform));
 894    lima_dump_command_stream_print(
 895       job->dump, array, 4, false, "add pp uniform info at va %x\n",
 896       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array));
 897 }
 898
 899 static void
 900 lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info)
 901 {
 902    struct lima_job *job = lima_job_get(ctx);
 903    struct lima_screen *screen = lima_screen(ctx->base.screen);
 904    struct lima_vs_shader_state *vs = ctx->vs;
 905    uint32_t gp_output_size;
 906    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
 907
 908    uint32_t *varying =
 909       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info,
 910                           vs->num_outputs * 8);
 911    int n = 0;
 912
 913    int offset = 0;
 914
 915    for (int i = 0; i < vs->num_outputs; i++) {
 916       struct lima_varying_info *v = vs->varying + i;
 917
 918       if (i == vs->gl_pos_idx ||
 919           i == vs->point_size_idx)
 920          continue;
 921
 922       int size = v->component_size * 4;
 923
 924       /* does component_size == 2 need to be 16 aligned? */
 925       if (v->component_size == 4)
 926          offset = align(offset, 16);
 927
 928       v->offset = offset;
 929       offset += size;
 930    }
 931
 932    vs->varying_stride = align(offset, 16);
 933
 934    /* gl_Position is always present, allocate space for it */
 935    gp_output_size = align(4 * 4 * num, 0x40);
 936
 937    /* Allocate space for varyings if there're any */
 938    if (vs->num_varyings) {
 939       ctx->gp_output_varyings_offt = gp_output_size;
 940       gp_output_size += align(vs->varying_stride * num, 0x40);
 941    }
 942
 943    /* Allocate space for gl_PointSize if it's there */
 944    if (vs->point_size_idx != -1) {
 945       ctx->gp_output_point_size_offt = gp_output_size;
 946       gp_output_size += 4 * num;
 947    }
 948
 949    /* gp_output can be too large for the suballocator, so create a
 950     * separate bo for it. The bo cache should prevent performance hit.
 951     */
 952    ctx->gp_output = lima_bo_create(screen, gp_output_size, 0);
 953    assert(ctx->gp_output);
 954    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->gp_output, LIMA_SUBMIT_BO_WRITE);
 955    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->gp_output, LIMA_SUBMIT_BO_READ);
 956
 957    for (int i = 0; i < vs->num_outputs; i++) {
 958       struct lima_varying_info *v = vs->varying + i;
 959
 960       if (i == vs->gl_pos_idx) {
 961          /* gl_Position */
 962          varying[n++] = ctx->gp_output->va;
 963          varying[n++] = 0x8020;
 964       } else if (i == vs->point_size_idx) {
 965          /* gl_PointSize */
 966          varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt;
 967          varying[n++] = 0x2021;
 968       } else {
 969          /* Varying */
 970          varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt +
 971                         v->offset;
 972          varying[n++] = (vs->varying_stride << 11) | (v->components - 1) |
 973             (v->component_size == 2 ? 0x0C : 0);
 974       }
 975    }
 976
 977    lima_dump_command_stream_print(
 978       job->dump, varying, n * 4, false, "update varying info at va %x\n",
 979       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info));
 980 }
 981
 982 static void
 983 lima_draw_vbo_update(struct pipe_context *pctx,
 984                      const struct pipe_draw_info *info)
 985 {
 986    struct lima_context *ctx = lima_context(pctx);
 987    struct lima_context_framebuffer *fb = &ctx->framebuffer;
 988    unsigned buffers = 0;
 989
 990    if (fb->base.zsbuf) {
 991       if (ctx->zsa->base.depth.enabled)
 992          buffers |= PIPE_CLEAR_DEPTH;
 993       if (ctx->zsa->base.stencil[0].enabled ||
 994           ctx->zsa->base.stencil[1].enabled)
 995          buffers |= PIPE_CLEAR_STENCIL;
 996    }
 997
 998    if (fb->base.nr_cbufs)
 999       buffers |= PIPE_CLEAR_COLOR0;
1000
1001    lima_update_job_wb(ctx, buffers);
1002
1003    lima_update_gp_attribute_info(ctx, info);
1004
1005    if ((ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1006         ctx->const_buffer[PIPE_SHADER_VERTEX].dirty) ||
1007        ctx->dirty & LIMA_CONTEXT_DIRTY_VIEWPORT ||
1008        ctx->dirty & LIMA_CONTEXT_DIRTY_SHADER_VERT) {
1009       lima_update_gp_uniform(ctx);
1010       ctx->const_buffer[PIPE_SHADER_VERTEX].dirty = false;
1011    }
1012
1013    lima_update_varying(ctx, info);
1014
1015    lima_pack_vs_cmd(ctx, info);
1016
1017    if (ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1018        ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty) {
1019       lima_update_pp_uniform(ctx);
1020       ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty = false;
1021    }
1022
1023    lima_update_textures(ctx);
1024
1025    lima_pack_render_state(ctx, info);
1026    lima_pack_plbu_cmd(ctx, info);
1027
1028    if (ctx->gp_output) {
1029       lima_bo_unreference(ctx->gp_output); /* held by job */
1030       ctx->gp_output = NULL;
1031    }
1032
1033    ctx->dirty = 0;
1034 }
1035
1036 static void
1037 lima_draw_vbo_indexed(struct pipe_context *pctx,
1038                       const struct pipe_draw_info *info)
1039 {
1040    struct lima_context *ctx = lima_context(pctx);
1041    struct lima_job *job = lima_job_get(ctx);
1042    struct pipe_resource *indexbuf = NULL;
1043    bool needs_indices = true;
1044
1045    /* Mali Utgard GPU always need min/max index info for index draw,
1046     * compute it if upper layer does not do for us */
1047    if (info->max_index != ~0u) {
1048       ctx->min_index = info->min_index;
1049       ctx->max_index = info->max_index;
1050       needs_indices = false;
1051    }
1052
1053    if (info->has_user_indices) {
1054       util_upload_index_buffer(&ctx->base, info, &indexbuf, &ctx->index_offset, 0x40);
1055       ctx->index_res = lima_resource(indexbuf);
1056    }
1057    else {
1058       ctx->index_res = lima_resource(info->index.resource);
1059       ctx->index_offset = 0;
1060       needs_indices = !panfrost_minmax_cache_get(ctx->index_res->index_cache, info->start,
1061                                                  info->count, &ctx->min_index, &ctx->max_index);
1062    }
1063
1064    if (needs_indices) {
1065       u_vbuf_get_minmax_index(pctx, info, &ctx->min_index, &ctx->max_index);
1066       if (!info->has_user_indices)
1067          panfrost_minmax_cache_add(ctx->index_res->index_cache, info->start, info->count,
1068                                    ctx->min_index, ctx->max_index);
1069    }
1070
1071    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1072    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1073    lima_draw_vbo_update(pctx, info);
1074
1075    if (indexbuf)
1076       pipe_resource_reference(&indexbuf, NULL);
1077 }
1078
1079 static void
1080 lima_draw_vbo_count(struct pipe_context *pctx,
1081                     const struct pipe_draw_info *info)
1082 {
1083    static const uint32_t max_verts = 65535;
1084
1085    struct pipe_draw_info local_info = *info;
1086    unsigned start = info->start;
1087    unsigned count = info->count;
1088
1089    while (count) {
1090       unsigned this_count = count;
1091       unsigned step;
1092
1093       u_split_draw(info, max_verts, &this_count, &step);
1094
1095       local_info.start = start;
1096       local_info.count = this_count;
1097
1098       lima_draw_vbo_update(pctx, &local_info);
1099
1100       count -= step;
1101       start += step;
1102    }
1103 }
1104
1105 static void
1106 lima_draw_vbo(struct pipe_context *pctx,
1107               const struct pipe_draw_info *info)
1108 {
1109    /* check if draw mode and vertex/index count match,
1110     * otherwise gp will hang */
1111    if (!u_trim_pipe_prim(info->mode, (unsigned*)&info->count)) {
1112       debug_printf("draw mode and vertex/index count mismatch\n");
1113       return;
1114    }
1115
1116    struct lima_context *ctx = lima_context(pctx);
1117
1118    if (!ctx->vs || !ctx->fs) {
1119       debug_warn_once("no shader, skip draw\n");
1120       return;
1121    }
1122
1123    lima_clip_scissor_to_viewport(ctx);
1124    if (lima_is_scissor_zero(ctx))
1125       return;
1126
1127    if (!lima_update_vs_state(ctx) || !lima_update_fs_state(ctx))
1128       return;
1129
1130    struct lima_job *job = lima_job_get(ctx);
1131
1132    lima_dump_command_stream_print(
1133       job->dump, ctx->vs->bo->map, ctx->vs->shader_size, false,
1134       "add vs at va %x\n", ctx->vs->bo->va);
1135
1136    lima_dump_command_stream_print(
1137       job->dump, ctx->fs->bo->map, ctx->fs->shader_size, false,
1138       "add fs at va %x\n", ctx->fs->bo->va);
1139
1140    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->vs->bo, LIMA_SUBMIT_BO_READ);
1141    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->fs->bo, LIMA_SUBMIT_BO_READ);
1142
1143    if (info->index_size)
1144       lima_draw_vbo_indexed(pctx, info);
1145    else
1146       lima_draw_vbo_count(pctx, info);
1147 }
1148
1149 void
1150 lima_draw_init(struct lima_context *ctx)
1151 {
1152    ctx->base.clear = lima_clear;
1153    ctx->base.draw_vbo = lima_draw_vbo;
1154 }