src/mesa/drivers/dri/i965/gen8_surface_state.c

   1 /*
   2  * Copyright © 2012 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "main/blend.h"
  25 #include "main/mtypes.h"
  26 #include "main/samplerobj.h"
  27 #include "main/texformat.h"
  28 #include "main/teximage.h"
  29 #include "program/prog_parameter.h"
  30
  31 #include "intel_mipmap_tree.h"
  32 #include "intel_batchbuffer.h"
  33 #include "intel_tex.h"
  34 #include "intel_fbo.h"
  35 #include "intel_buffer_objects.h"
  36
  37 #include "brw_context.h"
  38 #include "brw_state.h"
  39 #include "brw_defines.h"
  40 #include "brw_wm.h"
  41
  42 /**
  43  * Convert an swizzle enumeration (i.e. SWIZZLE_X) to one of the Gen7.5+
  44  * "Shader Channel Select" enumerations (i.e. HSW_SCS_RED).  The mappings are
  45  *
  46  * SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W, SWIZZLE_ZERO, SWIZZLE_ONE
  47  *         0          1          2          3             4            5
  48  *         4          5          6          7             0            1
  49  *   SCS_RED, SCS_GREEN,  SCS_BLUE, SCS_ALPHA,     SCS_ZERO,     SCS_ONE
  50  *
  51  * which is simply adding 4 then modding by 8 (or anding with 7).
  52  */
  53 static unsigned
  54 swizzle_to_scs(unsigned swizzle)
  55 {
  56    return (swizzle + 4) & 7;
  57 }
  58
  59 static uint32_t
  60 surface_tiling_mode(uint32_t tiling)
  61 {
  62    switch (tiling) {
  63    case I915_TILING_X:
  64       return GEN8_SURFACE_TILING_X;
  65    case I915_TILING_Y:
  66       return GEN8_SURFACE_TILING_Y;
  67    default:
  68       return GEN8_SURFACE_TILING_NONE;
  69    }
  70 }
  71
  72 static unsigned
  73 vertical_alignment(struct intel_mipmap_tree *mt)
  74 {
  75    switch (mt->align_h) {
  76    case 4:
  77       return GEN8_SURFACE_VALIGN_4;
  78    case 8:
  79       return GEN8_SURFACE_VALIGN_8;
  80    case 16:
  81       return GEN8_SURFACE_VALIGN_16;
  82    default:
  83       unreachable("Unsupported vertical surface alignment.");
  84    }
  85 }
  86
  87 static unsigned
  88 horizontal_alignment(struct intel_mipmap_tree *mt)
  89 {
  90    switch (mt->align_w) {
  91    case 4:
  92       return GEN8_SURFACE_HALIGN_4;
  93    case 8:
  94       return GEN8_SURFACE_HALIGN_8;
  95    case 16:
  96       return GEN8_SURFACE_HALIGN_16;
  97    default:
  98       unreachable("Unsupported horizontal surface alignment.");
  99    }
 100 }
 101
 102 static uint32_t *
 103 allocate_surface_state(struct brw_context *brw, uint32_t *out_offset)
 104 {
 105    int dwords = brw->gen >= 9 ? 16 : 13;
 106    uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
 107                                     dwords * 4, 64, out_offset);
 108    memset(surf, 0, dwords * 4);
 109    return surf;
 110 }
 111
 112 static void
 113 gen8_emit_buffer_surface_state(struct brw_context *brw,
 114                                uint32_t *out_offset,
 115                                drm_intel_bo *bo,
 116                                unsigned buffer_offset,
 117                                unsigned surface_format,
 118                                unsigned buffer_size,
 119                                unsigned pitch,
 120                                bool rw)
 121 {
 122    const unsigned mocs = brw->gen >= 9 ? SKL_MOCS_WB : BDW_MOCS_WB;
 123    uint32_t *surf = allocate_surface_state(brw, out_offset);
 124
 125    surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
 126              surface_format << BRW_SURFACE_FORMAT_SHIFT |
 127              BRW_SURFACE_RC_READ_WRITE;
 128    surf[1] = SET_FIELD(mocs, GEN8_SURFACE_MOCS);
 129
 130    surf[2] = SET_FIELD((buffer_size - 1) & 0x7f, GEN7_SURFACE_WIDTH) |
 131              SET_FIELD(((buffer_size - 1) >> 7) & 0x3fff, GEN7_SURFACE_HEIGHT);
 132    if (surface_format == BRW_SURFACEFORMAT_RAW)
 133       surf[3] = SET_FIELD(((buffer_size - 1) >> 21) & 0x3ff, BRW_SURFACE_DEPTH);
 134    else
 135       surf[3] = SET_FIELD(((buffer_size - 1) >> 21) & 0x3f, BRW_SURFACE_DEPTH);
 136    surf[3] |= (pitch - 1);
 137    surf[7] = SET_FIELD(HSW_SCS_RED,   GEN7_SURFACE_SCS_R) |
 138              SET_FIELD(HSW_SCS_GREEN, GEN7_SURFACE_SCS_G) |
 139              SET_FIELD(HSW_SCS_BLUE,  GEN7_SURFACE_SCS_B) |
 140              SET_FIELD(HSW_SCS_ALPHA, GEN7_SURFACE_SCS_A);
 141    /* reloc */
 142    *((uint64_t *) &surf[8]) = (bo ? bo->offset64 : 0) + buffer_offset;
 143
 144    /* Emit relocation to surface contents. */
 145    if (bo) {
 146       drm_intel_bo_emit_reloc(brw->batch.bo, *out_offset + 8 * 4,
 147                               bo, buffer_offset, I915_GEM_DOMAIN_SAMPLER,
 148                               rw ? I915_GEM_DOMAIN_SAMPLER : 0);
 149    }
 150 }
 151
 152 static void
 153 gen8_update_texture_surface(struct gl_context *ctx,
 154                             unsigned unit,
 155                             uint32_t *surf_offset,
 156                             bool for_gather)
 157 {
 158    struct brw_context *brw = brw_context(ctx);
 159    struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
 160    struct intel_texture_object *intelObj = intel_texture_object(tObj);
 161    struct intel_mipmap_tree *mt = intelObj->mt;
 162    struct gl_texture_image *firstImage = tObj->Image[0][tObj->BaseLevel];
 163    struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit);
 164    struct intel_mipmap_tree *aux_mt = NULL;
 165    uint32_t aux_mode = 0;
 166    mesa_format format = intelObj->_Format;
 167    uint32_t mocs_wb = brw->gen >= 9 ? SKL_MOCS_WB : BDW_MOCS_WB;
 168
 169    if (tObj->Target == GL_TEXTURE_BUFFER) {
 170       brw_update_buffer_texture_surface(ctx, unit, surf_offset);
 171       return;
 172    }
 173
 174    if (tObj->StencilSampling && firstImage->_BaseFormat == GL_DEPTH_STENCIL) {
 175       mt = mt->stencil_mt;
 176       format = MESA_FORMAT_S_UINT8;
 177    }
 178
 179    unsigned tiling_mode, pitch;
 180    if (format == MESA_FORMAT_S_UINT8) {
 181       tiling_mode = GEN8_SURFACE_TILING_W;
 182       pitch = 2 * mt->pitch;
 183    } else {
 184       tiling_mode = surface_tiling_mode(mt->tiling);
 185       pitch = mt->pitch;
 186    }
 187
 188    if (mt->mcs_mt) {
 189       aux_mt = mt->mcs_mt;
 190       aux_mode = GEN8_SURFACE_AUX_MODE_MCS;
 191    }
 192
 193    /* If this is a view with restricted NumLayers, then our effective depth
 194     * is not just the miptree depth.
 195     */
 196    uint32_t effective_depth =
 197       (tObj->Immutable && tObj->Target != GL_TEXTURE_3D) ? tObj->NumLayers
 198                                                          : mt->logical_depth0;
 199
 200    uint32_t tex_format = translate_tex_format(brw, format, sampler->sRGBDecode);
 201
 202    uint32_t *surf = allocate_surface_state(brw, surf_offset);
 203
 204    surf[0] = translate_tex_target(tObj->Target) << BRW_SURFACE_TYPE_SHIFT |
 205              tex_format << BRW_SURFACE_FORMAT_SHIFT |
 206              vertical_alignment(mt) |
 207              horizontal_alignment(mt) |
 208              tiling_mode;
 209
 210    if (tObj->Target == GL_TEXTURE_CUBE_MAP ||
 211        tObj->Target == GL_TEXTURE_CUBE_MAP_ARRAY) {
 212       surf[0] |= BRW_SURFACE_CUBEFACE_ENABLES;
 213    }
 214
 215    if (_mesa_is_array_texture(tObj->Target) ||
 216        tObj->Target == GL_TEXTURE_CUBE_MAP)
 217       surf[0] |= GEN8_SURFACE_IS_ARRAY;
 218
 219    surf[1] = SET_FIELD(mocs_wb, GEN8_SURFACE_MOCS) | mt->qpitch >> 2;
 220
 221    surf[2] = SET_FIELD(mt->logical_width0 - 1, GEN7_SURFACE_WIDTH) |
 222              SET_FIELD(mt->logical_height0 - 1, GEN7_SURFACE_HEIGHT);
 223
 224    surf[3] = SET_FIELD(effective_depth - 1, BRW_SURFACE_DEPTH) | (pitch - 1);
 225
 226    surf[4] = gen7_surface_msaa_bits(mt->num_samples, mt->msaa_layout) |
 227              SET_FIELD(tObj->MinLayer, GEN7_SURFACE_MIN_ARRAY_ELEMENT) |
 228              SET_FIELD(effective_depth - 1,
 229                        GEN7_SURFACE_RENDER_TARGET_VIEW_EXTENT);
 230
 231    surf[5] = SET_FIELD(tObj->MinLevel + tObj->BaseLevel - mt->first_level,
 232                        GEN7_SURFACE_MIN_LOD) |
 233              (intelObj->_MaxLevel - tObj->BaseLevel); /* mip count */
 234
 235    if (aux_mt) {
 236       surf[6] = SET_FIELD(mt->qpitch / 4, GEN8_SURFACE_AUX_QPITCH) |
 237                 SET_FIELD((aux_mt->pitch / 128) - 1, GEN8_SURFACE_AUX_PITCH) |
 238                 aux_mode;
 239    } else {
 240       surf[6] = 0;
 241    }
 242
 243    /* Handling GL_ALPHA as a surface format override breaks 1.30+ style
 244     * texturing functions that return a float, as our code generation always
 245     * selects the .x channel (which would always be 0).
 246     */
 247    const bool alpha_depth = tObj->DepthMode == GL_ALPHA &&
 248       (firstImage->_BaseFormat == GL_DEPTH_COMPONENT ||
 249        firstImage->_BaseFormat == GL_DEPTH_STENCIL);
 250
 251    surf[7] = mt->fast_clear_color_value;
 252
 253    const int swizzle =
 254       unlikely(alpha_depth) ? SWIZZLE_XYZW : brw_get_texture_swizzle(ctx, tObj);
 255    surf[7] |=
 256       SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 0)), GEN7_SURFACE_SCS_R) |
 257       SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 1)), GEN7_SURFACE_SCS_G) |
 258       SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 2)), GEN7_SURFACE_SCS_B) |
 259       SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 3)), GEN7_SURFACE_SCS_A);
 260
 261    *((uint64_t *) &surf[8]) = mt->bo->offset64 + mt->offset; /* reloc */
 262
 263    if (aux_mt) {
 264       *((uint64_t *) &surf[10]) = aux_mt->bo->offset64;
 265       drm_intel_bo_emit_reloc(brw->batch.bo, *surf_offset + 10 * 4,
 266                               aux_mt->bo, 0,
 267                               I915_GEM_DOMAIN_SAMPLER, 0);
 268    } else {
 269       surf[10] = 0;
 270       surf[11] = 0;
 271    }
 272    surf[12] = 0;
 273
 274    /* Emit relocation to surface contents */
 275    drm_intel_bo_emit_reloc(brw->batch.bo,
 276                            *surf_offset + 8 * 4,
 277                            mt->bo,
 278                            mt->offset,
 279                            I915_GEM_DOMAIN_SAMPLER, 0);
 280 }
 281
 282 /**
 283  * Creates a null surface.
 284  *
 285  * This is used when the shader doesn't write to any color output.  An FB
 286  * write to target 0 will still be emitted, because that's how the thread is
 287  * terminated (and computed depth is returned), so we need to have the
 288  * hardware discard the target 0 color output..
 289  */
 290 static void
 291 gen8_emit_null_surface_state(struct brw_context *brw,
 292                              unsigned width,
 293                              unsigned height,
 294                              unsigned samples,
 295                              uint32_t *out_offset)
 296 {
 297    uint32_t *surf = allocate_surface_state(brw, out_offset);
 298
 299    surf[0] = BRW_SURFACE_NULL << BRW_SURFACE_TYPE_SHIFT |
 300              BRW_SURFACEFORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT |
 301              GEN8_SURFACE_TILING_Y;
 302    surf[2] = SET_FIELD(width - 1, GEN7_SURFACE_WIDTH) |
 303              SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT);
 304 }
 305
 306 /**
 307  * Sets up a surface state structure to point at the given region.
 308  * While it is only used for the front/back buffer currently, it should be
 309  * usable for further buffers when doing ARB_draw_buffer support.
 310  */
 311 static void
 312 gen8_update_renderbuffer_surface(struct brw_context *brw,
 313                                  struct gl_renderbuffer *rb,
 314                                  bool layered,
 315                                  unsigned unit)
 316 {
 317    struct gl_context *ctx = &brw->ctx;
 318    struct intel_renderbuffer *irb = intel_renderbuffer(rb);
 319    struct intel_mipmap_tree *mt = irb->mt;
 320    struct intel_mipmap_tree *aux_mt = NULL;
 321    uint32_t aux_mode = 0;
 322    unsigned width = mt->logical_width0;
 323    unsigned height = mt->logical_height0;
 324    unsigned pitch = mt->pitch;
 325    uint32_t tiling = mt->tiling;
 326    uint32_t format = 0;
 327    uint32_t surf_type;
 328    bool is_array = false;
 329    int depth = MAX2(irb->layer_count, 1);
 330    const int min_array_element = (mt->format == MESA_FORMAT_S_UINT8) ?
 331       irb->mt_layer : (irb->mt_layer / MAX2(mt->num_samples, 1));
 332    GLenum gl_target =
 333       rb->TexImage ? rb->TexImage->TexObject->Target : GL_TEXTURE_2D;
 334    uint32_t surf_index =
 335       brw->wm.prog_data->binding_table.render_target_start + unit;
 336    /* FINISHME: Use PTE MOCS on Skylake. */
 337    uint32_t mocs = brw->gen >= 9 ? SKL_MOCS_WT : BDW_MOCS_PTE;
 338
 339    intel_miptree_used_for_rendering(mt);
 340
 341    switch (gl_target) {
 342    case GL_TEXTURE_CUBE_MAP_ARRAY:
 343    case GL_TEXTURE_CUBE_MAP:
 344       surf_type = BRW_SURFACE_2D;
 345       is_array = true;
 346       depth *= 6;
 347       break;
 348    case GL_TEXTURE_3D:
 349       depth = MAX2(irb->mt->logical_depth0, 1);
 350       /* fallthrough */
 351    default:
 352       surf_type = translate_tex_target(gl_target);
 353       is_array = _mesa_tex_target_is_array(gl_target);
 354       break;
 355    }
 356
 357    /* _NEW_BUFFERS */
 358    /* Render targets can't use IMS layout. Stencil in turn gets configured as
 359     * single sampled and indexed manually by the program.
 360     */
 361    if (mt->format == MESA_FORMAT_S_UINT8) {
 362       brw_configure_w_tiled(mt, true, &width, &height, &pitch,
 363                             &tiling, &format);
 364    } else {
 365       assert(mt->msaa_layout != INTEL_MSAA_LAYOUT_IMS);
 366       assert(brw_render_target_supported(brw, rb));
 367       mesa_format rb_format = _mesa_get_render_format(ctx,
 368                                                       intel_rb_format(irb));
 369       format = brw->render_target_format[rb_format];
 370       if (unlikely(!brw->format_supported_as_render_target[rb_format]))
 371          _mesa_problem(ctx, "%s: renderbuffer format %s unsupported\n",
 372                        __func__, _mesa_get_format_name(rb_format));
 373    }
 374
 375    if (mt->mcs_mt) {
 376       aux_mt = mt->mcs_mt;
 377       aux_mode = GEN8_SURFACE_AUX_MODE_MCS;
 378    }
 379
 380    uint32_t *surf =
 381       allocate_surface_state(brw, &brw->wm.base.surf_offset[surf_index]);
 382
 383    surf[0] = (surf_type << BRW_SURFACE_TYPE_SHIFT) |
 384              (is_array ? GEN7_SURFACE_IS_ARRAY : 0) |
 385              (format << BRW_SURFACE_FORMAT_SHIFT) |
 386              vertical_alignment(mt) |
 387              horizontal_alignment(mt) |
 388              surface_tiling_mode(tiling);
 389
 390    surf[1] = SET_FIELD(mocs, GEN8_SURFACE_MOCS) | mt->qpitch >> 2;
 391
 392    surf[2] = SET_FIELD(width - 1, GEN7_SURFACE_WIDTH) |
 393              SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT);
 394
 395    surf[3] = (depth - 1) << BRW_SURFACE_DEPTH_SHIFT |
 396              (pitch - 1); /* Surface Pitch */
 397
 398    surf[4] = min_array_element << GEN7_SURFACE_MIN_ARRAY_ELEMENT_SHIFT |
 399              (depth - 1) << GEN7_SURFACE_RENDER_TARGET_VIEW_EXTENT_SHIFT;
 400
 401    if (mt->format != MESA_FORMAT_S_UINT8)
 402       surf[4] |= gen7_surface_msaa_bits(mt->num_samples, mt->msaa_layout);
 403
 404    surf[5] = irb->mt_level - irb->mt->first_level;
 405
 406    if (aux_mt) {
 407       surf[6] = SET_FIELD(mt->qpitch / 4, GEN8_SURFACE_AUX_QPITCH) |
 408                 SET_FIELD((aux_mt->pitch / 128) - 1, GEN8_SURFACE_AUX_PITCH) |
 409                 aux_mode;
 410    } else {
 411       surf[6] = 0;
 412    }
 413
 414    surf[7] = mt->fast_clear_color_value |
 415              SET_FIELD(HSW_SCS_RED,   GEN7_SURFACE_SCS_R) |
 416              SET_FIELD(HSW_SCS_GREEN, GEN7_SURFACE_SCS_G) |
 417              SET_FIELD(HSW_SCS_BLUE,  GEN7_SURFACE_SCS_B) |
 418              SET_FIELD(HSW_SCS_ALPHA, GEN7_SURFACE_SCS_A);
 419
 420    assert(mt->offset % mt->cpp == 0);
 421    *((uint64_t *) &surf[8]) = mt->bo->offset64 + mt->offset; /* reloc */
 422
 423    if (aux_mt) {
 424       *((uint64_t *) &surf[10]) = aux_mt->bo->offset64;
 425       drm_intel_bo_emit_reloc(brw->batch.bo,
 426                               brw->wm.base.surf_offset[surf_index] + 10 * 4,
 427                               aux_mt->bo, 0,
 428                               I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER);
 429    } else {
 430       surf[10] = 0;
 431       surf[11] = 0;
 432    }
 433    surf[12] = 0;
 434
 435    drm_intel_bo_emit_reloc(brw->batch.bo,
 436                            brw->wm.base.surf_offset[surf_index] + 8 * 4,
 437                            mt->bo,
 438                            mt->offset,
 439                            I915_GEM_DOMAIN_RENDER,
 440                            I915_GEM_DOMAIN_RENDER);
 441 }
 442
 443 void
 444 gen8_init_vtable_surface_functions(struct brw_context *brw)
 445 {
 446    brw->vtbl.update_texture_surface = gen8_update_texture_surface;
 447    brw->vtbl.update_renderbuffer_surface = gen8_update_renderbuffer_surface;
 448    brw->vtbl.emit_null_surface_state = gen8_emit_null_surface_state;
 449    brw->vtbl.emit_buffer_surface_state = gen8_emit_buffer_surface_state;
 450 }