src/gallium/auxiliary/vl/vl_compositor_cs.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2019 Advanced Micro Devices, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  * Authors: James Zhu <james.zhu<@amd.com>
  27  *
  28  **************************************************************************/
  29
  30 #include <assert.h>
  31
  32 #include "tgsi/tgsi_text.h"
  33 #include "vl_compositor_cs.h"
  34
  35 struct cs_viewport {
  36    float scale_x;
  37    float scale_y;
  38    struct u_rect area;
  39    int translate_x;
  40    int translate_y;
  41    float sampler0_w;
  42    float sampler0_h;
  43 };
  44
  45 const char *compute_shader_video_buffer =
  46       "COMP\n"
  47       "PROPERTY CS_FIXED_BLOCK_WIDTH 8\n"
  48       "PROPERTY CS_FIXED_BLOCK_HEIGHT 8\n"
  49       "PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
  50
  51       "DCL SV[0], THREAD_ID\n"
  52       "DCL SV[1], BLOCK_ID\n"
  53
  54       "DCL CONST[0..5]\n"
  55       "DCL SVIEW[0..2], RECT, FLOAT\n"
  56       "DCL SAMP[0..2]\n"
  57
  58       "DCL IMAGE[0], 2D, WR\n"
  59       "DCL TEMP[0..7]\n"
  60
  61       "IMM[0] UINT32 { 8, 8, 1, 0}\n"
  62       "IMM[1] FLT32 { 1.0, 2.0, 0.0, 0.0}\n"
  63
  64       "UMAD TEMP[0], SV[1], IMM[0], SV[0]\n"
  65
  66       /* Drawn area check */
  67       "USGE TEMP[1].xy, TEMP[0].xyxy, CONST[4].xyxy\n"
  68       "USLT TEMP[1].zw, TEMP[0].xyxy, CONST[4].zwzw\n"
  69       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].yyyy\n"
  70       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].zzzz\n"
  71       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].wwww\n"
  72
  73       "UIF TEMP[1]\n"
  74          /* Translate */
  75          "UADD TEMP[2].xy, TEMP[0], -CONST[5].xyxy\n"
  76          "U2F TEMP[2], TEMP[2]\n"
  77          "DIV TEMP[3], TEMP[2], IMM[1].yyyy\n"
  78
  79          /* Scale */
  80          "DIV TEMP[2], TEMP[2], CONST[3].zwzw\n"
  81          "DIV TEMP[3], TEMP[3], CONST[3].zwzw\n"
  82
  83          /* Fetch texels */
  84          "TEX_LZ TEMP[4].x, TEMP[2], SAMP[0], RECT\n"
  85          "TEX_LZ TEMP[4].y, TEMP[3], SAMP[1], RECT\n"
  86          "TEX_LZ TEMP[4].z, TEMP[3], SAMP[2], RECT\n"
  87
  88          "MOV TEMP[4].w, IMM[1].xxxx\n"
  89
  90          /* Color Space Conversion */
  91          "DP4 TEMP[7].x, CONST[0], TEMP[4]\n"
  92          "DP4 TEMP[7].y, CONST[1], TEMP[4]\n"
  93          "DP4 TEMP[7].z, CONST[2], TEMP[4]\n"
  94
  95          "MOV TEMP[5].w, TEMP[4].zzzz\n"
  96          "SLE TEMP[6].w, TEMP[5], CONST[3].xxxx\n"
  97          "SGT TEMP[5].w, TEMP[5], CONST[3].yyyy\n"
  98
  99          "MAX TEMP[7].w, TEMP[5], TEMP[6]\n"
 100
 101          "STORE IMAGE[0], TEMP[0], TEMP[7], 2D\n"
 102       "ENDIF\n"
 103
 104       "END\n";
 105
 106 const char *compute_shader_weave =
 107       "COMP\n"
 108       "PROPERTY CS_FIXED_BLOCK_WIDTH 8\n"
 109       "PROPERTY CS_FIXED_BLOCK_HEIGHT 8\n"
 110       "PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
 111
 112       "DCL SV[0], THREAD_ID\n"
 113       "DCL SV[1], BLOCK_ID\n"
 114
 115       "DCL CONST[0..5]\n"
 116       "DCL SVIEW[0..2], 2D_ARRAY, FLOAT\n"
 117       "DCL SAMP[0..2]\n"
 118
 119       "DCL IMAGE[0], 2D, WR\n"
 120       "DCL TEMP[0..15]\n"
 121
 122       "IMM[0] UINT32 { 8, 8, 1, 0}\n"
 123       "IMM[1] FLT32 { 1.0, 2.0, 0.0, 0.0}\n"
 124       "IMM[2] UINT32 { 1, 2, 4, 0}\n"
 125       "IMM[3] FLT32 { 0.25, 0.5, 0.125, 0.125}\n"
 126
 127       "UMAD TEMP[0], SV[1], IMM[0], SV[0]\n"
 128
 129       /* Drawn area check */
 130       "USGE TEMP[1].xy, TEMP[0].xyxy, CONST[4].xyxy\n"
 131       "USLT TEMP[1].zw, TEMP[0].xyxy, CONST[4].zwzw\n"
 132       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].yyyy\n"
 133       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].zzzz\n"
 134       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].wwww\n"
 135
 136       "UIF TEMP[1]\n"
 137          "MOV TEMP[2], TEMP[0]\n"
 138          /* Translate */
 139          "UADD TEMP[2].xy, TEMP[2], -CONST[5].xyxy\n"
 140
 141          /* Top Y */
 142          "U2F TEMP[2], TEMP[2]\n"
 143          "DIV TEMP[2].y, TEMP[2].yyyy, IMM[1].yyyy\n"
 144          /* Down Y */
 145          "MOV TEMP[12], TEMP[2]\n"
 146
 147          /* Top UV */
 148          "MOV TEMP[3], TEMP[2]\n"
 149          "DIV TEMP[3].xy, TEMP[3], IMM[1].yyyy\n"
 150          /* Down UV */
 151          "MOV TEMP[13], TEMP[3]\n"
 152
 153          /* Texture offset */
 154          "ADD TEMP[2].x, TEMP[2].xxxx, IMM[3].yyyy\n"
 155          "ADD TEMP[2].y, TEMP[2].yyyy, IMM[3].xxxx\n"
 156          "ADD TEMP[12].x, TEMP[12].xxxx, IMM[3].yyyy\n"
 157          "ADD TEMP[12].y, TEMP[12].yyyy, IMM[3].xxxx\n"
 158
 159          "ADD TEMP[3].x, TEMP[3].xxxx, IMM[3].xxxx\n"
 160          "ADD TEMP[3].y, TEMP[3].yyyy, IMM[3].wwww\n"
 161          "ADD TEMP[13].x, TEMP[13].xxxx, IMM[3].xxxx\n"
 162          "ADD TEMP[13].y, TEMP[13].yyyy, IMM[3].wwww\n"
 163
 164          /* Scale */
 165          "DIV TEMP[2].xy, TEMP[2], CONST[3].zwzw\n"
 166          "DIV TEMP[12].xy, TEMP[12], CONST[3].zwzw\n"
 167          "DIV TEMP[3].xy, TEMP[3], CONST[3].zwzw\n"
 168          "DIV TEMP[13].xy, TEMP[13], CONST[3].zwzw\n"
 169
 170          /* Weave offset */
 171          "ADD TEMP[2].y, TEMP[2].yyyy, IMM[3].xxxx\n"
 172          "ADD TEMP[12].y, TEMP[12].yyyy, -IMM[3].xxxx\n"
 173          "ADD TEMP[3].y, TEMP[3].yyyy, IMM[3].xxxx\n"
 174          "ADD TEMP[13].y, TEMP[13].yyyy, -IMM[3].xxxx\n"
 175
 176          /* Texture layer */
 177          "MOV TEMP[14].x, TEMP[2].yyyy\n"
 178          "MOV TEMP[14].yz, TEMP[3].yyyy\n"
 179          "ROUND TEMP[15], TEMP[14]\n"
 180          "ADD TEMP[14], TEMP[14], -TEMP[15]\n"
 181          "MOV TEMP[14], |TEMP[14]|\n"
 182          "MUL TEMP[14], TEMP[14], IMM[1].yyyy\n"
 183
 184          /* Normalize */
 185          "DIV TEMP[2].xy, TEMP[2], CONST[5].zwzw\n"
 186          "DIV TEMP[12].xy, TEMP[12], CONST[5].zwzw\n"
 187          "DIV TEMP[15].xy, CONST[5].zwzw, IMM[1].yyyy\n"
 188          "DIV TEMP[3].xy, TEMP[3], TEMP[15].xyxy\n"
 189          "DIV TEMP[13].xy, TEMP[13], TEMP[15].xyxy\n"
 190
 191          /* Fetch texels */
 192          "MOV TEMP[2].z, IMM[1].wwww\n"
 193          "MOV TEMP[3].z, IMM[1].wwww\n"
 194          "TEX_LZ TEMP[10].x, TEMP[2], SAMP[0], 2D_ARRAY\n"
 195          "TEX_LZ TEMP[10].y, TEMP[3], SAMP[1], 2D_ARRAY\n"
 196          "TEX_LZ TEMP[10].z, TEMP[3], SAMP[2], 2D_ARRAY\n"
 197
 198          "MOV TEMP[12].z, IMM[1].xxxx\n"
 199          "MOV TEMP[13].z, IMM[1].xxxx\n"
 200          "TEX_LZ TEMP[11].x, TEMP[12], SAMP[0], 2D_ARRAY\n"
 201          "TEX_LZ TEMP[11].y, TEMP[13], SAMP[1], 2D_ARRAY\n"
 202          "TEX_LZ TEMP[11].z, TEMP[13], SAMP[2], 2D_ARRAY\n"
 203
 204          "LRP TEMP[6], TEMP[14], TEMP[10], TEMP[11]\n"
 205          "MOV TEMP[6].w, IMM[1].xxxx\n"
 206
 207          /* Color Space Conversion */
 208          "DP4 TEMP[9].x, CONST[0], TEMP[6]\n"
 209          "DP4 TEMP[9].y, CONST[1], TEMP[6]\n"
 210          "DP4 TEMP[9].z, CONST[2], TEMP[6]\n"
 211
 212          "MOV TEMP[7].w, TEMP[6].zzzz\n"
 213          "SLE TEMP[8].w, TEMP[7], CONST[3].xxxx\n"
 214          "SGT TEMP[7].w, TEMP[7], CONST[3].yyyy\n"
 215
 216          "MAX TEMP[9].w, TEMP[7], TEMP[8]\n"
 217
 218          "STORE IMAGE[0], TEMP[0], TEMP[9], 2D\n"
 219       "ENDIF\n"
 220
 221       "END\n";
 222
 223 const char *compute_shader_rgba =
 224       "COMP\n"
 225       "PROPERTY CS_FIXED_BLOCK_WIDTH 8\n"
 226       "PROPERTY CS_FIXED_BLOCK_HEIGHT 8\n"
 227       "PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
 228
 229       "DCL SV[0], THREAD_ID\n"
 230       "DCL SV[1], BLOCK_ID\n"
 231
 232       "DCL CONST[0..5]\n"
 233       "DCL SVIEW[0], RECT, FLOAT\n"
 234       "DCL SAMP[0]\n"
 235
 236       "DCL IMAGE[0], 2D, WR\n"
 237       "DCL TEMP[0..3]\n"
 238
 239       "IMM[0] UINT32 { 8, 8, 1, 0}\n"
 240       "IMM[1] FLT32 { 1.0, 2.0, 0.0, 0.0}\n"
 241
 242       "UMAD TEMP[0], SV[1], IMM[0], SV[0]\n"
 243
 244       /* Drawn area check */
 245       "USGE TEMP[1].xy, TEMP[0].xyxy, CONST[4].xyxy\n"
 246       "USLT TEMP[1].zw, TEMP[0].xyxy, CONST[4].zwzw\n"
 247       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].yyyy\n"
 248       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].zzzz\n"
 249       "AND TEMP[1].x, TEMP[1].xxxx, TEMP[1].wwww\n"
 250
 251       "UIF TEMP[1]\n"
 252          /* Translate */
 253          "UADD TEMP[2].xy, TEMP[0], -CONST[5].xyxy\n"
 254          "U2F TEMP[2], TEMP[2]\n"
 255
 256          /* Scale */
 257          "DIV TEMP[2], TEMP[2], CONST[3].zwzw\n"
 258
 259          /* Fetch texels */
 260          "TEX_LZ TEMP[3], TEMP[2], SAMP[0], RECT\n"
 261
 262          "STORE IMAGE[0], TEMP[0], TEMP[3], 2D\n"
 263       "ENDIF\n"
 264
 265       "END\n";
 266
 267 static void
 268 cs_launch(struct vl_compositor *c,
 269           void                 *cs,
 270           const struct u_rect  *draw_area)
 271 {
 272    struct pipe_context *ctx = c->pipe;
 273
 274    /* Bind the image */
 275    struct pipe_image_view image = {};
 276    image.resource = c->fb_state.cbufs[0]->texture;
 277    image.shader_access = image.access = PIPE_IMAGE_ACCESS_READ_WRITE;
 278    image.format = c->fb_state.cbufs[0]->texture->format;
 279
 280    ctx->set_shader_images(c->pipe, PIPE_SHADER_COMPUTE, 0, 1, &image);
 281
 282    /* Bind compute shader */
 283    ctx->bind_compute_state(ctx, cs);
 284
 285    /* Dispatch compute */
 286    struct pipe_grid_info info = {};
 287    info.block[0] = 8;
 288    info.block[1] = 8;
 289    info.block[2] = 1;
 290    info.grid[0] = DIV_ROUND_UP(draw_area->x1, info.block[0]);
 291    info.grid[1] = DIV_ROUND_UP(draw_area->y1, info.block[1]);
 292    info.grid[2] = 1;
 293
 294    ctx->launch_grid(ctx, &info);
 295 }
 296
 297 static inline struct u_rect
 298 calc_drawn_area(struct vl_compositor_state *s,
 299                 struct vl_compositor_layer *layer)
 300 {
 301    struct vertex2f tl, br;
 302    struct u_rect result;
 303
 304    assert(s && layer);
 305
 306    tl = layer->dst.tl;
 307    br = layer->dst.br;
 308
 309    /* Scale */
 310    result.x0 = tl.x * layer->viewport.scale[0] + layer->viewport.translate[0];
 311    result.y0 = tl.y * layer->viewport.scale[1] + layer->viewport.translate[1];
 312    result.x1 = br.x * layer->viewport.scale[0] + layer->viewport.translate[0];
 313    result.y1 = br.y * layer->viewport.scale[1] + layer->viewport.translate[1];
 314
 315    /* Clip */
 316    result.x0 = MAX2(result.x0, s->scissor.minx);
 317    result.y0 = MAX2(result.y0, s->scissor.miny);
 318    result.x1 = MIN2(result.x1, s->scissor.maxx);
 319    result.y1 = MIN2(result.y1, s->scissor.maxy);
 320    return result;
 321 }
 322
 323 static bool
 324 set_viewport(struct vl_compositor_state *s,
 325              struct cs_viewport         *drawn)
 326 {
 327    struct pipe_transfer *buf_transfer;
 328
 329    assert(s && drawn);
 330
 331    void *ptr = pipe_buffer_map(s->pipe, s->shader_params,
 332                                PIPE_TRANSFER_READ | PIPE_TRANSFER_WRITE,
 333                                &buf_transfer);
 334
 335    if (!ptr)
 336      return false;
 337
 338    float *ptr_float = (float *)ptr;
 339    ptr_float += sizeof(vl_csc_matrix)/sizeof(float) + 2;
 340    *ptr_float++ = drawn->scale_x;
 341    *ptr_float++ = drawn->scale_y;
 342
 343    int *ptr_int = (int *)ptr_float;
 344    *ptr_int++ = drawn->area.x0;
 345    *ptr_int++ = drawn->area.y0;
 346    *ptr_int++ = drawn->area.x1;
 347    *ptr_int++ = drawn->area.y1;
 348    *ptr_int++ = drawn->translate_x;
 349    *ptr_int++ = drawn->translate_y;
 350
 351    ptr_float = (float *)ptr_int;
 352    *ptr_float++ = drawn->sampler0_w;
 353    *ptr_float = drawn->sampler0_h;
 354    pipe_buffer_unmap(s->pipe, buf_transfer);
 355
 356    return true;
 357 }
 358
 359 static void
 360 draw_layers(struct vl_compositor       *c,
 361             struct vl_compositor_state *s,
 362             struct u_rect              *dirty)
 363 {
 364    unsigned i;
 365    static struct cs_viewport old_drawn;
 366
 367    assert(c);
 368
 369    for (i = 0; i < VL_COMPOSITOR_MAX_LAYERS; ++i) {
 370       if (s->used_layers & (1 << i)) {
 371          struct vl_compositor_layer *layer = &s->layers[i];
 372          struct pipe_sampler_view **samplers = &layer->sampler_views[0];
 373          unsigned num_sampler_views = !samplers[1] ? 1 : !samplers[2] ? 2 : 3;
 374          struct cs_viewport drawn;
 375
 376          drawn.area = calc_drawn_area(s, layer);
 377          drawn.scale_x = layer->viewport.scale[0] /
 378                   (float)layer->sampler_views[0]->texture->width0;
 379          drawn.scale_y = drawn.scale_x;
 380          drawn.translate_x = (int)layer->viewport.translate[0];
 381          drawn.translate_y = (int)layer->viewport.translate[1];
 382          drawn.sampler0_w = (float)layer->sampler_views[0]->texture->width0;
 383          drawn.sampler0_h = (float)layer->sampler_views[0]->texture->height0;
 384
 385          if (memcmp(&drawn, &old_drawn, sizeof(struct cs_viewport))) {
 386             set_viewport(s, &drawn);
 387             old_drawn = drawn;
 388          }
 389
 390          c->pipe->bind_sampler_states(c->pipe, PIPE_SHADER_COMPUTE, 0,
 391                         num_sampler_views, layer->samplers);
 392          c->pipe->set_sampler_views(c->pipe, PIPE_SHADER_COMPUTE, 0,
 393                         num_sampler_views, samplers);
 394
 395          cs_launch(c, layer->cs, &(drawn.area));
 396
 397          if (dirty) {
 398             struct u_rect drawn = calc_drawn_area(s, layer);
 399             dirty->x0 = MIN2(drawn.x0, dirty->x0);
 400             dirty->y0 = MIN2(drawn.y0, dirty->y0);
 401             dirty->x1 = MAX2(drawn.x1, dirty->x1);
 402             dirty->y1 = MAX2(drawn.y1, dirty->y1);
 403          }
 404       }
 405    }
 406 }
 407
 408 void *
 409 vl_compositor_cs_create_shader(struct vl_compositor *c,
 410                                const char           *compute_shader_text)
 411 {
 412    assert(c && compute_shader_text);
 413
 414    struct tgsi_token tokens[1024];
 415    if (!tgsi_text_translate(compute_shader_text, tokens, ARRAY_SIZE(tokens))) {
 416       assert(0);
 417       return NULL;
 418    }
 419
 420    struct pipe_compute_state state = {};
 421    state.ir_type = PIPE_SHADER_IR_TGSI;
 422    state.prog = tokens;
 423
 424    /* create compute shader */
 425    return c->pipe->create_compute_state(c->pipe, &state);
 426 }
 427
 428 void
 429 vl_compositor_cs_render(struct vl_compositor_state *s,
 430                         struct vl_compositor       *c,
 431                         struct pipe_surface        *dst_surface,
 432                         struct u_rect              *dirty_area,
 433                         bool                        clear_dirty)
 434 {
 435    assert(c && s);
 436    assert(dst_surface);
 437
 438    c->fb_state.width = dst_surface->width;
 439    c->fb_state.height = dst_surface->height;
 440    c->fb_state.cbufs[0] = dst_surface;
 441
 442    if (!s->scissor_valid) {
 443       s->scissor.minx = 0;
 444       s->scissor.miny = 0;
 445       s->scissor.maxx = dst_surface->width;
 446       s->scissor.maxy = dst_surface->height;
 447    }
 448
 449    if (clear_dirty && dirty_area &&
 450        (dirty_area->x0 < dirty_area->x1 || dirty_area->y0 < dirty_area->y1)) {
 451
 452       c->pipe->clear_render_target(c->pipe, dst_surface, &s->clear_color,
 453                        0, 0, dst_surface->width, dst_surface->height, false);
 454       dirty_area->x0 = dirty_area->y0 = VL_COMPOSITOR_MAX_DIRTY;
 455       dirty_area->x1 = dirty_area->y1 = VL_COMPOSITOR_MIN_DIRTY;
 456    }
 457
 458    pipe_set_constant_buffer(c->pipe, PIPE_SHADER_COMPUTE, 0, s->shader_params);
 459
 460    draw_layers(c, s, dirty_area);
 461 }