src/mesa/drivers/dri/intel/intel_blit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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  **************************************************************************/
  27
  28
  29 #include "main/mtypes.h"
  30 #include "main/context.h"
  31 #include "main/enums.h"
  32 #include "main/colormac.h"
  33
  34 #include "intel_blit.h"
  35 #include "intel_buffers.h"
  36 #include "intel_context.h"
  37 #include "intel_fbo.h"
  38 #include "intel_reg.h"
  39 #include "intel_regions.h"
  40 #include "intel_batchbuffer.h"
  41 #include "intel_mipmap_tree.h"
  42
  43 #define FILE_DEBUG_FLAG DEBUG_BLIT
  44
  45 static GLuint translate_raster_op(GLenum logicop)
  46 {
  47    switch(logicop) {
  48    case GL_CLEAR: return 0x00;
  49    case GL_AND: return 0x88;
  50    case GL_AND_REVERSE: return 0x44;
  51    case GL_COPY: return 0xCC;
  52    case GL_AND_INVERTED: return 0x22;
  53    case GL_NOOP: return 0xAA;
  54    case GL_XOR: return 0x66;
  55    case GL_OR: return 0xEE;
  56    case GL_NOR: return 0x11;
  57    case GL_EQUIV: return 0x99;
  58    case GL_INVERT: return 0x55;
  59    case GL_OR_REVERSE: return 0xDD;
  60    case GL_COPY_INVERTED: return 0x33;
  61    case GL_OR_INVERTED: return 0xBB;
  62    case GL_NAND: return 0x77;
  63    case GL_SET: return 0xFF;
  64    default: return 0;
  65    }
  66 }
  67
  68 static uint32_t
  69 br13_for_cpp(int cpp)
  70 {
  71    switch (cpp) {
  72    case 4:
  73       return BR13_8888;
  74       break;
  75    case 2:
  76       return BR13_565;
  77       break;
  78    case 1:
  79       return BR13_8;
  80       break;
  81    default:
  82       assert(0);
  83       return 0;
  84    }
  85 }
  86
  87 /* Copy BitBlt
  88  */
  89 bool
  90 intelEmitCopyBlit(struct intel_context *intel,
  91                   GLuint cpp,
  92                   GLshort src_pitch,
  93                   drm_intel_bo *src_buffer,
  94                   GLuint src_offset,
  95                   uint32_t src_tiling,
  96                   GLshort dst_pitch,
  97                   drm_intel_bo *dst_buffer,
  98                   GLuint dst_offset,
  99                   uint32_t dst_tiling,
 100                   GLshort src_x, GLshort src_y,
 101                   GLshort dst_x, GLshort dst_y,
 102                   GLshort w, GLshort h,
 103                   GLenum logic_op)
 104 {
 105    GLuint CMD, BR13, pass = 0;
 106    int dst_y2 = dst_y + h;
 107    int dst_x2 = dst_x + w;
 108    drm_intel_bo *aper_array[3];
 109    BATCH_LOCALS;
 110
 111    if (dst_tiling != I915_TILING_NONE) {
 112       if (dst_offset & 4095)
 113          return false;
 114       if (dst_tiling == I915_TILING_Y)
 115          return false;
 116    }
 117    if (src_tiling != I915_TILING_NONE) {
 118       if (src_offset & 4095)
 119          return false;
 120       if (src_tiling == I915_TILING_Y)
 121          return false;
 122    }
 123
 124    /* do space check before going any further */
 125    do {
 126        aper_array[0] = intel->batch.bo;
 127        aper_array[1] = dst_buffer;
 128        aper_array[2] = src_buffer;
 129
 130        if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
 131            intel_batchbuffer_flush(intel);
 132            pass++;
 133        } else
 134            break;
 135    } while (pass < 2);
 136
 137    if (pass >= 2)
 138       return false;
 139
 140    intel_batchbuffer_require_space(intel, 8 * 4, true);
 141    DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
 142        __FUNCTION__,
 143        src_buffer, src_pitch, src_offset, src_x, src_y,
 144        dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
 145
 146    src_pitch *= cpp;
 147    dst_pitch *= cpp;
 148
 149    /* Blit pitch must be dword-aligned.  Otherwise, the hardware appears to drop
 150     * the low bits.
 151     */
 152    assert(src_pitch % 4 == 0);
 153    assert(dst_pitch % 4 == 0);
 154
 155    /* For big formats (such as floating point), do the copy using 32bpp and
 156     * multiply the coordinates.
 157     */
 158    if (cpp > 4) {
 159       assert(cpp % 4 == 0);
 160       dst_x *= cpp / 4;
 161       dst_x2 *= cpp / 4;
 162       src_x *= cpp / 4;
 163       cpp = 4;
 164    }
 165
 166    BR13 = br13_for_cpp(cpp) | translate_raster_op(logic_op) << 16;
 167
 168    switch (cpp) {
 169    case 1:
 170    case 2:
 171       CMD = XY_SRC_COPY_BLT_CMD;
 172       break;
 173    case 4:
 174       CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
 175       break;
 176    default:
 177       return false;
 178    }
 179
 180 #ifndef I915
 181    if (dst_tiling != I915_TILING_NONE) {
 182       CMD |= XY_DST_TILED;
 183       dst_pitch /= 4;
 184    }
 185    if (src_tiling != I915_TILING_NONE) {
 186       CMD |= XY_SRC_TILED;
 187       src_pitch /= 4;
 188    }
 189 #endif
 190
 191    if (dst_y2 <= dst_y || dst_x2 <= dst_x) {
 192       return true;
 193    }
 194
 195    assert(dst_x < dst_x2);
 196    assert(dst_y < dst_y2);
 197
 198    BEGIN_BATCH_BLT(8);
 199    OUT_BATCH(CMD);
 200    OUT_BATCH(BR13 | (uint16_t)dst_pitch);
 201    OUT_BATCH((dst_y << 16) | dst_x);
 202    OUT_BATCH((dst_y2 << 16) | dst_x2);
 203    OUT_RELOC_FENCED(dst_buffer,
 204                     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
 205                     dst_offset);
 206    OUT_BATCH((src_y << 16) | src_x);
 207    OUT_BATCH((uint16_t)src_pitch);
 208    OUT_RELOC_FENCED(src_buffer,
 209                     I915_GEM_DOMAIN_RENDER, 0,
 210                     src_offset);
 211    ADVANCE_BATCH();
 212
 213    intel_batchbuffer_emit_mi_flush(intel);
 214
 215    return true;
 216 }
 217
 218
 219 /**
 220  * Use blitting to clear the renderbuffers named by 'flags'.
 221  * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field
 222  * since that might include software renderbuffers or renderbuffers
 223  * which we're clearing with triangles.
 224  * \param mask  bitmask of BUFFER_BIT_* values indicating buffers to clear
 225  */
 226 GLbitfield
 227 intelClearWithBlit(struct gl_context *ctx, GLbitfield mask)
 228 {
 229    struct intel_context *intel = intel_context(ctx);
 230    struct gl_framebuffer *fb = ctx->DrawBuffer;
 231    GLuint clear_depth_value, clear_depth_mask;
 232    GLint cx, cy, cw, ch;
 233    GLbitfield fail_mask = 0;
 234    BATCH_LOCALS;
 235
 236    /*
 237     * Compute values for clearing the buffers.
 238     */
 239    clear_depth_value = 0;
 240    clear_depth_mask = 0;
 241    if (mask & BUFFER_BIT_DEPTH) {
 242       clear_depth_value = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
 243       clear_depth_mask = XY_BLT_WRITE_RGB;
 244    }
 245    if (mask & BUFFER_BIT_STENCIL) {
 246       clear_depth_value |= (ctx->Stencil.Clear & 0xff) << 24;
 247       clear_depth_mask |= XY_BLT_WRITE_ALPHA;
 248    }
 249
 250    cx = fb->_Xmin;
 251    if (fb->Name == 0)
 252       cy = ctx->DrawBuffer->Height - fb->_Ymax;
 253    else
 254       cy = fb->_Ymin;
 255    cw = fb->_Xmax - fb->_Xmin;
 256    ch = fb->_Ymax - fb->_Ymin;
 257
 258    if (cw == 0 || ch == 0)
 259       return 0;
 260
 261    /* Loop over all renderbuffers */
 262    mask &= (1 << BUFFER_COUNT) - 1;
 263    while (mask) {
 264       GLuint buf = _mesa_ffs(mask) - 1;
 265       bool is_depth_stencil = buf == BUFFER_DEPTH || buf == BUFFER_STENCIL;
 266       struct intel_renderbuffer *irb;
 267       int x1, y1, x2, y2;
 268       uint32_t clear_val;
 269       uint32_t BR13, CMD;
 270       int pitch, cpp;
 271       drm_intel_bo *aper_array[2];
 272
 273       mask &= ~(1 << buf);
 274
 275       irb = intel_get_renderbuffer(fb, buf);
 276       if (irb == NULL || irb->region == NULL || irb->region->bo == NULL) {
 277          fail_mask |= 1 << buf;
 278          continue;
 279       }
 280
 281       /* OK, clear this renderbuffer */
 282       x1 = cx + irb->draw_x;
 283       y1 = cy + irb->draw_y;
 284       x2 = cx + cw + irb->draw_x;
 285       y2 = cy + ch + irb->draw_y;
 286
 287       pitch = irb->region->pitch;
 288       cpp = irb->region->cpp;
 289
 290       DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
 291           __FUNCTION__,
 292           irb->region->bo, (pitch * cpp),
 293           x1, y1, x2 - x1, y2 - y1);
 294
 295       BR13 = 0xf0 << 16;
 296       CMD = XY_COLOR_BLT_CMD;
 297
 298       /* Setup the blit command */
 299       if (cpp == 4) {
 300          if (is_depth_stencil) {
 301             CMD |= clear_depth_mask;
 302          } else {
 303             /* clearing RGBA */
 304             CMD |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
 305          }
 306       }
 307
 308       assert(irb->region->tiling != I915_TILING_Y);
 309
 310 #ifndef I915
 311       if (irb->region->tiling != I915_TILING_NONE) {
 312          CMD |= XY_DST_TILED;
 313          pitch /= 4;
 314       }
 315 #endif
 316       BR13 |= (pitch * cpp);
 317
 318       if (is_depth_stencil) {
 319          clear_val = clear_depth_value;
 320       } else {
 321          uint8_t clear[4];
 322          GLfloat *color = ctx->Color.ClearColor.f;
 323
 324          _mesa_unclamped_float_rgba_to_ubyte(clear, color);
 325
 326          switch (irb->Base.Format) {
 327          case MESA_FORMAT_ARGB8888:
 328          case MESA_FORMAT_XRGB8888:
 329             clear_val = PACK_COLOR_8888(clear[3], clear[0],
 330                                         clear[1], clear[2]);
 331             break;
 332          case MESA_FORMAT_RGB565:
 333             clear_val = PACK_COLOR_565(clear[0], clear[1], clear[2]);
 334             break;
 335          case MESA_FORMAT_ARGB4444:
 336             clear_val = PACK_COLOR_4444(clear[3], clear[0],
 337                                         clear[1], clear[2]);
 338             break;
 339          case MESA_FORMAT_ARGB1555:
 340             clear_val = PACK_COLOR_1555(clear[3], clear[0],
 341                                         clear[1], clear[2]);
 342             break;
 343          case MESA_FORMAT_A8:
 344             clear_val = PACK_COLOR_8888(clear[3], clear[3],
 345                                         clear[3], clear[3]);
 346             break;
 347          default:
 348             fail_mask |= 1 << buf;
 349             continue;
 350          }
 351       }
 352
 353       BR13 |= br13_for_cpp(cpp);
 354
 355       assert(x1 < x2);
 356       assert(y1 < y2);
 357
 358       /* do space check before going any further */
 359       aper_array[0] = intel->batch.bo;
 360       aper_array[1] = irb->region->bo;
 361
 362       if (drm_intel_bufmgr_check_aperture_space(aper_array,
 363                                                 ARRAY_SIZE(aper_array)) != 0) {
 364          intel_batchbuffer_flush(intel);
 365       }
 366
 367       BEGIN_BATCH_BLT(6);
 368       OUT_BATCH(CMD);
 369       OUT_BATCH(BR13);
 370       OUT_BATCH((y1 << 16) | x1);
 371       OUT_BATCH((y2 << 16) | x2);
 372       OUT_RELOC_FENCED(irb->region->bo,
 373                        I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
 374                        0);
 375       OUT_BATCH(clear_val);
 376       ADVANCE_BATCH();
 377
 378       if (intel->always_flush_cache)
 379          intel_batchbuffer_emit_mi_flush(intel);
 380
 381       if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL)
 382          mask &= ~(BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL);
 383    }
 384
 385    return fail_mask;
 386 }
 387
 388 bool
 389 intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 390                                   GLuint cpp,
 391                                   GLubyte *src_bits, GLuint src_size,
 392                                   GLuint fg_color,
 393                                   GLshort dst_pitch,
 394                                   drm_intel_bo *dst_buffer,
 395                                   GLuint dst_offset,
 396                                   uint32_t dst_tiling,
 397                                   GLshort x, GLshort y,
 398                                   GLshort w, GLshort h,
 399                                   GLenum logic_op)
 400 {
 401    int dwords = ALIGN(src_size, 8) / 4;
 402    uint32_t opcode, br13, blit_cmd;
 403
 404    if (dst_tiling != I915_TILING_NONE) {
 405       if (dst_offset & 4095)
 406          return false;
 407       if (dst_tiling == I915_TILING_Y)
 408          return false;
 409    }
 410
 411    assert( logic_op - GL_CLEAR >= 0 );
 412    assert( logic_op - GL_CLEAR < 0x10 );
 413    assert(dst_pitch > 0);
 414
 415    if (w < 0 || h < 0)
 416       return true;
 417
 418    dst_pitch *= cpp;
 419
 420    DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
 421        __FUNCTION__,
 422        dst_buffer, dst_pitch, dst_offset, x, y, w, h, src_size, dwords);
 423
 424    intel_batchbuffer_require_space(intel,
 425                                    (8 * 4) +
 426                                    (3 * 4) +
 427                                    dwords * 4, true);
 428
 429    opcode = XY_SETUP_BLT_CMD;
 430    if (cpp == 4)
 431       opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
 432 #ifndef I915
 433    if (dst_tiling != I915_TILING_NONE) {
 434       opcode |= XY_DST_TILED;
 435       dst_pitch /= 4;
 436    }
 437 #endif
 438
 439    br13 = dst_pitch | (translate_raster_op(logic_op) << 16) | (1 << 29);
 440    br13 |= br13_for_cpp(cpp);
 441
 442    blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */
 443    if (dst_tiling != I915_TILING_NONE)
 444       blit_cmd |= XY_DST_TILED;
 445
 446    BEGIN_BATCH_BLT(8 + 3);
 447    OUT_BATCH(opcode);
 448    OUT_BATCH(br13);
 449    OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
 450    OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
 451    OUT_RELOC_FENCED(dst_buffer,
 452                     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
 453                     dst_offset);
 454    OUT_BATCH(0); /* bg */
 455    OUT_BATCH(fg_color); /* fg */
 456    OUT_BATCH(0); /* pattern base addr */
 457
 458    OUT_BATCH(blit_cmd | ((3 - 2) + dwords));
 459    OUT_BATCH((y << 16) | x);
 460    OUT_BATCH(((y + h) << 16) | (x + w));
 461    ADVANCE_BATCH();
 462
 463    intel_batchbuffer_data(intel, src_bits, dwords * 4, true);
 464
 465    intel_batchbuffer_emit_mi_flush(intel);
 466
 467    return true;
 468 }
 469
 470 /* We don't have a memmove-type blit like some other hardware, so we'll do a
 471  * rectangular blit covering a large space, then emit 1-scanline blit at the
 472  * end to cover the last if we need.
 473  */
 474 void
 475 intel_emit_linear_blit(struct intel_context *intel,
 476                        drm_intel_bo *dst_bo,
 477                        unsigned int dst_offset,
 478                        drm_intel_bo *src_bo,
 479                        unsigned int src_offset,
 480                        unsigned int size)
 481 {
 482    GLuint pitch, height;
 483    bool ok;
 484
 485    /* The pitch given to the GPU must be DWORD aligned, and
 486     * we want width to match pitch. Max width is (1 << 15 - 1),
 487     * rounding that down to the nearest DWORD is 1 << 15 - 4
 488     */
 489    pitch = MIN2(size, (1 << 15) - 4);
 490    height = size / pitch;
 491    ok = intelEmitCopyBlit(intel, 1,
 492                           pitch, src_bo, src_offset, I915_TILING_NONE,
 493                           pitch, dst_bo, dst_offset, I915_TILING_NONE,
 494                           0, 0, /* src x/y */
 495                           0, 0, /* dst x/y */
 496                           pitch, height, /* w, h */
 497                           GL_COPY);
 498    assert(ok);
 499
 500    src_offset += pitch * height;
 501    dst_offset += pitch * height;
 502    size -= pitch * height;
 503    assert (size < (1 << 15));
 504    assert ((size & 3) == 0); /* Pitch must be DWORD aligned */
 505    if (size != 0) {
 506       ok = intelEmitCopyBlit(intel, 1,
 507                              size, src_bo, src_offset, I915_TILING_NONE,
 508                              size, dst_bo, dst_offset, I915_TILING_NONE,
 509                              0, 0, /* src x/y */
 510                              0, 0, /* dst x/y */
 511                              size, 1, /* w, h */
 512                              GL_COPY);
 513       assert(ok);
 514    }
 515 }
 516
 517 /**
 518  * Used to initialize the alpha value of an ARGB8888 teximage after
 519  * loading it from an XRGB8888 source.
 520  *
 521  * This is very common with glCopyTexImage2D().
 522  */
 523 void
 524 intel_set_teximage_alpha_to_one(struct gl_context *ctx,
 525                                 struct intel_texture_image *intel_image)
 526 {
 527    struct intel_context *intel = intel_context(ctx);
 528    unsigned int image_x, image_y;
 529    uint32_t x1, y1, x2, y2;
 530    uint32_t BR13, CMD;
 531    int pitch, cpp;
 532    drm_intel_bo *aper_array[2];
 533    struct intel_region *region = intel_image->mt->region;
 534    int width, height, depth;
 535    BATCH_LOCALS;
 536
 537    intel_miptree_get_dimensions_for_image(&intel_image->base.Base,
 538                                           &width, &height, &depth);
 539    assert(depth == 1);
 540
 541    assert(intel_image->base.Base.TexFormat == MESA_FORMAT_ARGB8888);
 542
 543    /* get dest x/y in destination texture */
 544    intel_miptree_get_image_offset(intel_image->mt,
 545                                   intel_image->base.Base.Level,
 546                                   intel_image->base.Base.Face,
 547                                   0,
 548                                   &image_x, &image_y);
 549
 550    x1 = image_x;
 551    y1 = image_y;
 552    x2 = image_x + width;
 553    y2 = image_y + height;
 554
 555    pitch = region->pitch;
 556    cpp = region->cpp;
 557
 558    DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
 559        __FUNCTION__,
 560        intel_image->mt->region->bo, (pitch * cpp),
 561        x1, y1, x2 - x1, y2 - y1);
 562
 563    BR13 = br13_for_cpp(cpp) | 0xf0 << 16;
 564    CMD = XY_COLOR_BLT_CMD;
 565    CMD |= XY_BLT_WRITE_ALPHA;
 566
 567    assert(region->tiling != I915_TILING_Y);
 568
 569 #ifndef I915
 570    if (region->tiling != I915_TILING_NONE) {
 571       CMD |= XY_DST_TILED;
 572       pitch /= 4;
 573    }
 574 #endif
 575    BR13 |= (pitch * cpp);
 576
 577    /* do space check before going any further */
 578    aper_array[0] = intel->batch.bo;
 579    aper_array[1] = region->bo;
 580
 581    if (drm_intel_bufmgr_check_aperture_space(aper_array,
 582                                              ARRAY_SIZE(aper_array)) != 0) {
 583       intel_batchbuffer_flush(intel);
 584    }
 585
 586    BEGIN_BATCH_BLT(6);
 587    OUT_BATCH(CMD);
 588    OUT_BATCH(BR13);
 589    OUT_BATCH((y1 << 16) | x1);
 590    OUT_BATCH((y2 << 16) | x2);
 591    OUT_RELOC_FENCED(region->bo,
 592                     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
 593                     0);
 594    OUT_BATCH(0xffffffff); /* white, but only alpha gets written */
 595    ADVANCE_BATCH();
 596
 597    intel_batchbuffer_emit_mi_flush(intel);
 598 }