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

   1 /*
   2  Copyright 2003 VMware, Inc.
   3  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   4  Intel funded Tungsten Graphics to
   5  develop this 3D driver.
   6
   7  Permission is hereby granted, free of charge, to any person obtaining
   8  a copy of this software and associated documentation files (the
   9  "Software"), to deal in the Software without restriction, including
  10  without limitation the rights to use, copy, modify, merge, publish,
  11  distribute, sublicense, and/or sell copies of the Software, and to
  12  permit persons to whom the Software is furnished to do so, subject to
  13  the following conditions:
  14
  15  The above copyright notice and this permission notice (including the
  16  next paragraph) shall be included in all copies or substantial
  17  portions of the Software.
  18
  19  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  20  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  22  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  23  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  24  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  25  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26
  27  **********************************************************************/
  28  /*
  29   * Authors:
  30   *   Keith Whitwell <keithw@vmware.com>
  31   */
  32
  33
  34 #include "main/api_exec.h"
  35 #include "main/context.h"
  36 #include "main/fbobject.h"
  37 #include "main/extensions.h"
  38 #include "main/imports.h"
  39 #include "main/macros.h"
  40 #include "main/points.h"
  41 #include "main/version.h"
  42 #include "main/vtxfmt.h"
  43 #include "main/texobj.h"
  44 #include "main/framebuffer.h"
  45
  46 #include "vbo/vbo_context.h"
  47
  48 #include "drivers/common/driverfuncs.h"
  49 #include "drivers/common/meta.h"
  50 #include "utils.h"
  51
  52 #include "brw_context.h"
  53 #include "brw_defines.h"
  54 #include "brw_compiler.h"
  55 #include "brw_draw.h"
  56 #include "brw_state.h"
  57
  58 #include "intel_batchbuffer.h"
  59 #include "intel_buffer_objects.h"
  60 #include "intel_buffers.h"
  61 #include "intel_fbo.h"
  62 #include "intel_mipmap_tree.h"
  63 #include "intel_pixel.h"
  64 #include "intel_image.h"
  65 #include "intel_tex.h"
  66 #include "intel_tex_obj.h"
  67
  68 #include "swrast_setup/swrast_setup.h"
  69 #include "tnl/tnl.h"
  70 #include "tnl/t_pipeline.h"
  71 #include "util/ralloc.h"
  72
  73 /***************************************
  74  * Mesa's Driver Functions
  75  ***************************************/
  76
  77 static size_t
  78 brw_query_samples_for_format(struct gl_context *ctx, GLenum target,
  79                              GLenum internalFormat, int samples[16])
  80 {
  81    struct brw_context *brw = brw_context(ctx);
  82
  83    (void) target;
  84
  85    switch (brw->gen) {
  86    case 9:
  87       samples[0] = 16;
  88       samples[1] = 8;
  89       samples[2] = 4;
  90       samples[3] = 2;
  91       return 4;
  92
  93    case 8:
  94       samples[0] = 8;
  95       samples[1] = 4;
  96       samples[2] = 2;
  97       return 3;
  98
  99    case 7:
 100       samples[0] = 8;
 101       samples[1] = 4;
 102       return 2;
 103
 104    case 6:
 105       samples[0] = 4;
 106       return 1;
 107
 108    default:
 109       assert(brw->gen < 6);
 110       samples[0] = 1;
 111       return 1;
 112    }
 113 }
 114
 115 const char *const brw_vendor_string = "Intel Open Source Technology Center";
 116
 117 const char *
 118 brw_get_renderer_string(unsigned deviceID)
 119 {
 120    const char *chipset;
 121    static char buffer[128];
 122
 123    switch (deviceID) {
 124 #undef CHIPSET
 125 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
 126 #include "pci_ids/i965_pci_ids.h"
 127    default:
 128       chipset = "Unknown Intel Chipset";
 129       break;
 130    }
 131
 132    (void) driGetRendererString(buffer, chipset, 0);
 133    return buffer;
 134 }
 135
 136 static const GLubyte *
 137 intel_get_string(struct gl_context * ctx, GLenum name)
 138 {
 139    const struct brw_context *const brw = brw_context(ctx);
 140
 141    switch (name) {
 142    case GL_VENDOR:
 143       return (GLubyte *) brw_vendor_string;
 144
 145    case GL_RENDERER:
 146       return
 147          (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID);
 148
 149    default:
 150       return NULL;
 151    }
 152 }
 153
 154 static void
 155 intel_viewport(struct gl_context *ctx)
 156 {
 157    struct brw_context *brw = brw_context(ctx);
 158    __DRIcontext *driContext = brw->driContext;
 159
 160    if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) {
 161       dri2InvalidateDrawable(driContext->driDrawablePriv);
 162       dri2InvalidateDrawable(driContext->driReadablePriv);
 163    }
 164 }
 165
 166 static void
 167 intel_update_state(struct gl_context * ctx, GLuint new_state)
 168 {
 169    struct brw_context *brw = brw_context(ctx);
 170    struct intel_texture_object *tex_obj;
 171    struct intel_renderbuffer *depth_irb;
 172
 173    if (ctx->swrast_context)
 174       _swrast_InvalidateState(ctx, new_state);
 175    _vbo_InvalidateState(ctx, new_state);
 176
 177    brw->NewGLState |= new_state;
 178
 179    _mesa_unlock_context_textures(ctx);
 180
 181    /* Resolve the depth buffer's HiZ buffer. */
 182    depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
 183    if (depth_irb)
 184       intel_renderbuffer_resolve_hiz(brw, depth_irb);
 185
 186    /* Resolve depth buffer and render cache of each enabled texture. */
 187    int maxEnabledUnit = ctx->Texture._MaxEnabledTexImageUnit;
 188    for (int i = 0; i <= maxEnabledUnit; i++) {
 189       if (!ctx->Texture.Unit[i]._Current)
 190          continue;
 191       tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current);
 192       if (!tex_obj || !tex_obj->mt)
 193          continue;
 194       intel_miptree_all_slices_resolve_depth(brw, tex_obj->mt);
 195       intel_miptree_resolve_color(brw, tex_obj->mt);
 196       brw_render_cache_set_check_flush(brw, tex_obj->mt->bo);
 197    }
 198
 199    _mesa_lock_context_textures(ctx);
 200 }
 201
 202 #define flushFront(screen)      ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
 203
 204 static void
 205 intel_flush_front(struct gl_context *ctx)
 206 {
 207    struct brw_context *brw = brw_context(ctx);
 208    __DRIcontext *driContext = brw->driContext;
 209    __DRIdrawable *driDrawable = driContext->driDrawablePriv;
 210    __DRIscreen *const screen = brw->intelScreen->driScrnPriv;
 211
 212    if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) {
 213       if (flushFront(screen) && driDrawable &&
 214           driDrawable->loaderPrivate) {
 215
 216          /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
 217           *
 218           * This potentially resolves both front and back buffer. It
 219           * is unnecessary to resolve the back, but harms nothing except
 220           * performance. And no one cares about front-buffer render
 221           * performance.
 222           */
 223          intel_resolve_for_dri2_flush(brw, driDrawable);
 224          intel_batchbuffer_flush(brw);
 225
 226          flushFront(screen)(driDrawable, driDrawable->loaderPrivate);
 227
 228          /* We set the dirty bit in intel_prepare_render() if we're
 229           * front buffer rendering once we get there.
 230           */
 231          brw->front_buffer_dirty = false;
 232       }
 233    }
 234 }
 235
 236 static void
 237 intel_glFlush(struct gl_context *ctx)
 238 {
 239    struct brw_context *brw = brw_context(ctx);
 240
 241    intel_batchbuffer_flush(brw);
 242    intel_flush_front(ctx);
 243
 244    brw->need_flush_throttle = true;
 245 }
 246
 247 static void
 248 intel_finish(struct gl_context * ctx)
 249 {
 250    struct brw_context *brw = brw_context(ctx);
 251
 252    intel_glFlush(ctx);
 253
 254    if (brw->batch.last_bo)
 255       drm_intel_bo_wait_rendering(brw->batch.last_bo);
 256 }
 257
 258 static void
 259 brw_init_driver_functions(struct brw_context *brw,
 260                           struct dd_function_table *functions)
 261 {
 262    _mesa_init_driver_functions(functions);
 263
 264    /* GLX uses DRI2 invalidate events to handle window resizing.
 265     * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
 266     * which doesn't provide a mechanism for snooping the event queues.
 267     *
 268     * So EGL still relies on viewport hacks to handle window resizing.
 269     * This should go away with DRI3000.
 270     */
 271    if (!brw->driContext->driScreenPriv->dri2.useInvalidate)
 272       functions->Viewport = intel_viewport;
 273
 274    functions->Flush = intel_glFlush;
 275    functions->Finish = intel_finish;
 276    functions->GetString = intel_get_string;
 277    functions->UpdateState = intel_update_state;
 278
 279    intelInitTextureFuncs(functions);
 280    intelInitTextureImageFuncs(functions);
 281    intelInitTextureSubImageFuncs(functions);
 282    intelInitTextureCopyImageFuncs(functions);
 283    intelInitCopyImageFuncs(functions);
 284    intelInitClearFuncs(functions);
 285    intelInitBufferFuncs(functions);
 286    intelInitPixelFuncs(functions);
 287    intelInitBufferObjectFuncs(functions);
 288    intel_init_syncobj_functions(functions);
 289    brw_init_object_purgeable_functions(functions);
 290
 291    brwInitFragProgFuncs( functions );
 292    brw_init_common_queryobj_functions(functions);
 293    if (brw->gen >= 6)
 294       gen6_init_queryobj_functions(functions);
 295    else
 296       gen4_init_queryobj_functions(functions);
 297    brw_init_compute_functions(functions);
 298    if (brw->gen >= 7)
 299       brw_init_conditional_render_functions(functions);
 300
 301    functions->QuerySamplesForFormat = brw_query_samples_for_format;
 302
 303    functions->NewTransformFeedback = brw_new_transform_feedback;
 304    functions->DeleteTransformFeedback = brw_delete_transform_feedback;
 305    functions->GetTransformFeedbackVertexCount =
 306       brw_get_transform_feedback_vertex_count;
 307    if (brw->gen >= 7) {
 308       functions->BeginTransformFeedback = gen7_begin_transform_feedback;
 309       functions->EndTransformFeedback = gen7_end_transform_feedback;
 310       functions->PauseTransformFeedback = gen7_pause_transform_feedback;
 311       functions->ResumeTransformFeedback = gen7_resume_transform_feedback;
 312    } else {
 313       functions->BeginTransformFeedback = brw_begin_transform_feedback;
 314       functions->EndTransformFeedback = brw_end_transform_feedback;
 315    }
 316
 317    if (brw->gen >= 6)
 318       functions->GetSamplePosition = gen6_get_sample_position;
 319 }
 320
 321 static void
 322 brw_initialize_context_constants(struct brw_context *brw)
 323 {
 324    struct gl_context *ctx = &brw->ctx;
 325    const struct brw_compiler *compiler = brw->intelScreen->compiler;
 326
 327    const bool stage_exists[MESA_SHADER_STAGES] = {
 328       [MESA_SHADER_VERTEX] = true,
 329       [MESA_SHADER_TESS_CTRL] = false,
 330       [MESA_SHADER_TESS_EVAL] = false,
 331       [MESA_SHADER_GEOMETRY] = brw->gen >= 6,
 332       [MESA_SHADER_FRAGMENT] = true,
 333       [MESA_SHADER_COMPUTE] = _mesa_extension_override_enables.ARB_compute_shader,
 334    };
 335
 336    unsigned num_stages = 0;
 337    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 338       if (stage_exists[i])
 339          num_stages++;
 340    }
 341
 342    unsigned max_samplers =
 343       brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16;
 344
 345    ctx->Const.MaxDualSourceDrawBuffers = 1;
 346    ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS;
 347    ctx->Const.MaxCombinedShaderOutputResources =
 348       MAX_IMAGE_UNITS + BRW_MAX_DRAW_BUFFERS;
 349
 350    ctx->Const.QueryCounterBits.Timestamp = 36;
 351
 352    ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */
 353    ctx->Const.MaxImageUnits = MAX_IMAGE_UNITS;
 354    ctx->Const.MaxRenderbufferSize = 8192;
 355    ctx->Const.MaxTextureLevels = MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS);
 356    ctx->Const.Max3DTextureLevels = 12; /* 2048 */
 357    ctx->Const.MaxCubeTextureLevels = 14; /* 8192 */
 358    ctx->Const.MaxArrayTextureLayers = brw->gen >= 7 ? 2048 : 512;
 359    ctx->Const.MaxTextureMbytes = 1536;
 360    ctx->Const.MaxTextureRectSize = 1 << 12;
 361    ctx->Const.MaxTextureMaxAnisotropy = 16.0;
 362    ctx->Const.StripTextureBorder = true;
 363    if (brw->gen >= 7)
 364       ctx->Const.MaxProgramTextureGatherComponents = 4;
 365    else if (brw->gen == 6)
 366       ctx->Const.MaxProgramTextureGatherComponents = 1;
 367
 368    ctx->Const.MaxUniformBlockSize = 65536;
 369
 370    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 371       struct gl_program_constants *prog = &ctx->Const.Program[i];
 372
 373       if (!stage_exists[i])
 374          continue;
 375
 376       prog->MaxTextureImageUnits = max_samplers;
 377
 378       prog->MaxUniformBlocks = BRW_MAX_UBO;
 379       prog->MaxCombinedUniformComponents =
 380          prog->MaxUniformComponents +
 381          ctx->Const.MaxUniformBlockSize / 4 * prog->MaxUniformBlocks;
 382
 383       prog->MaxAtomicCounters = MAX_ATOMIC_COUNTERS;
 384       prog->MaxAtomicBuffers = BRW_MAX_ABO;
 385       prog->MaxImageUniforms = compiler->scalar_stage[i] ? BRW_MAX_IMAGES : 0;
 386       prog->MaxShaderStorageBlocks = BRW_MAX_SSBO;
 387    }
 388
 389    ctx->Const.MaxTextureUnits =
 390       MIN2(ctx->Const.MaxTextureCoordUnits,
 391            ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits);
 392
 393    ctx->Const.MaxUniformBufferBindings = num_stages * BRW_MAX_UBO;
 394    ctx->Const.MaxCombinedUniformBlocks = num_stages * BRW_MAX_UBO;
 395    ctx->Const.MaxCombinedAtomicBuffers = num_stages * BRW_MAX_ABO;
 396    ctx->Const.MaxCombinedShaderStorageBlocks = num_stages * BRW_MAX_SSBO;
 397    ctx->Const.MaxShaderStorageBufferBindings = num_stages * BRW_MAX_SSBO;
 398    ctx->Const.MaxCombinedTextureImageUnits = num_stages * max_samplers;
 399    ctx->Const.MaxCombinedImageUniforms = num_stages * BRW_MAX_IMAGES;
 400
 401
 402    /* Hardware only supports a limited number of transform feedback buffers.
 403     * So we need to override the Mesa default (which is based only on software
 404     * limits).
 405     */
 406    ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS;
 407
 408    /* On Gen6, in the worst case, we use up one binding table entry per
 409     * transform feedback component (see comments above the definition of
 410     * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
 411     * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
 412     * BRW_MAX_SOL_BINDINGS.
 413     *
 414     * In "separate components" mode, we need to divide this value by
 415     * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
 416     * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
 417     */
 418    ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS;
 419    ctx->Const.MaxTransformFeedbackSeparateComponents =
 420       BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS;
 421
 422    ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true;
 423
 424    int max_samples;
 425    const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen);
 426    const int clamp_max_samples =
 427       driQueryOptioni(&brw->optionCache, "clamp_max_samples");
 428
 429    if (clamp_max_samples < 0) {
 430       max_samples = msaa_modes[0];
 431    } else {
 432       /* Select the largest supported MSAA mode that does not exceed
 433        * clamp_max_samples.
 434        */
 435       max_samples = 0;
 436       for (int i = 0; msaa_modes[i] != 0; ++i) {
 437          if (msaa_modes[i] <= clamp_max_samples) {
 438             max_samples = msaa_modes[i];
 439             break;
 440          }
 441       }
 442    }
 443
 444    ctx->Const.MaxSamples = max_samples;
 445    ctx->Const.MaxColorTextureSamples = max_samples;
 446    ctx->Const.MaxDepthTextureSamples = max_samples;
 447    ctx->Const.MaxIntegerSamples = max_samples;
 448    ctx->Const.MaxImageSamples = 0;
 449
 450    /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
 451     * to map indices of rectangular grid to sample numbers within a pixel.
 452     * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
 453     * extension implementation. For more details see the comment above
 454     * gen6_set_sample_maps() definition.
 455     */
 456    gen6_set_sample_maps(ctx);
 457
 458    ctx->Const.MinLineWidth = 1.0;
 459    ctx->Const.MinLineWidthAA = 1.0;
 460    if (brw->gen >= 6) {
 461       ctx->Const.MaxLineWidth = 7.375;
 462       ctx->Const.MaxLineWidthAA = 7.375;
 463       ctx->Const.LineWidthGranularity = 0.125;
 464    } else {
 465       ctx->Const.MaxLineWidth = 7.0;
 466       ctx->Const.MaxLineWidthAA = 7.0;
 467       ctx->Const.LineWidthGranularity = 0.5;
 468    }
 469
 470    /* For non-antialiased lines, we have to round the line width to the
 471     * nearest whole number. Make sure that we don't advertise a line
 472     * width that, when rounded, will be beyond the actual hardware
 473     * maximum.
 474     */
 475    assert(roundf(ctx->Const.MaxLineWidth) <= ctx->Const.MaxLineWidth);
 476
 477    ctx->Const.MinPointSize = 1.0;
 478    ctx->Const.MinPointSizeAA = 1.0;
 479    ctx->Const.MaxPointSize = 255.0;
 480    ctx->Const.MaxPointSizeAA = 255.0;
 481    ctx->Const.PointSizeGranularity = 1.0;
 482
 483    if (brw->gen >= 5 || brw->is_g4x)
 484       ctx->Const.MaxClipPlanes = 8;
 485
 486    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024;
 487    ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0;
 488    ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0;
 489    ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0;
 490    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0;
 491    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0;
 492    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0;
 493    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16;
 494    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256;
 495    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1;
 496    ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024;
 497    ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams =
 498       MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters,
 499            ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams);
 500
 501    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024;
 502    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024;
 503    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024;
 504    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024;
 505    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12;
 506    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256;
 507    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0;
 508    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024;
 509    ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams =
 510       MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters,
 511            ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams);
 512
 513    /* Fragment shaders use real, 32-bit twos-complement integers for all
 514     * integer types.
 515     */
 516    ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31;
 517    ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30;
 518    ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0;
 519    ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt;
 520    ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt;
 521
 522    ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMin = 31;
 523    ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMax = 30;
 524    ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.Precision = 0;
 525    ctx->Const.Program[MESA_SHADER_VERTEX].HighInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt;
 526    ctx->Const.Program[MESA_SHADER_VERTEX].MediumInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt;
 527
 528    /* Gen6 converts quads to polygon in beginning of 3D pipeline,
 529     * but we're not sure how it's actually done for vertex order,
 530     * that affect provoking vertex decision. Always use last vertex
 531     * convention for quad primitive which works as expected for now.
 532     */
 533    if (brw->gen >= 6)
 534       ctx->Const.QuadsFollowProvokingVertexConvention = false;
 535
 536    ctx->Const.NativeIntegers = true;
 537    ctx->Const.VertexID_is_zero_based = true;
 538
 539    /* Regarding the CMP instruction, the Ivybridge PRM says:
 540     *
 541     *   "For each enabled channel 0b or 1b is assigned to the appropriate flag
 542     *    bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
 543     *    0xFFFFFFFF) is assigned to dst."
 544     *
 545     * but PRMs for earlier generations say
 546     *
 547     *   "In dword format, one GRF may store up to 8 results. When the register
 548     *    is used later as a vector of Booleans, as only LSB at each channel
 549     *    contains meaning [sic] data, software should make sure all higher bits
 550     *    are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
 551     *
 552     * We select the representation of a true boolean uniform to be ~0, and fix
 553     * the results of Gen <= 5 CMP instruction's with -(result & 1).
 554     */
 555    ctx->Const.UniformBooleanTrue = ~0;
 556
 557    /* From the gen4 PRM, volume 4 page 127:
 558     *
 559     *     "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
 560     *      the base address of the first element of the surface, computed in
 561     *      software by adding the surface base address to the byte offset of
 562     *      the element in the buffer."
 563     *
 564     * However, unaligned accesses are slower, so enforce buffer alignment.
 565     */
 566    ctx->Const.UniformBufferOffsetAlignment = 16;
 567
 568    /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
 569     * that we can safely have the CPU and GPU writing the same SSBO on
 570     * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
 571     * writes, so there's no problem. For an SSBO, the GPU and the CPU can
 572     * be updating disjoint regions of the buffer simultaneously and that will
 573     * break if the regions overlap the same cacheline.
 574     */
 575    ctx->Const.ShaderStorageBufferOffsetAlignment = 64;
 576    ctx->Const.TextureBufferOffsetAlignment = 16;
 577    ctx->Const.MaxTextureBufferSize = 128 * 1024 * 1024;
 578
 579    if (brw->gen >= 6) {
 580       ctx->Const.MaxVarying = 32;
 581       ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128;
 582       ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64;
 583       ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128;
 584       ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128;
 585    }
 586
 587    /* We want the GLSL compiler to emit code that uses condition codes */
 588    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 589       ctx->Const.ShaderCompilerOptions[i] =
 590          brw->intelScreen->compiler->glsl_compiler_options[i];
 591    }
 592
 593    /* ARB_viewport_array */
 594    if (brw->gen >= 6 && ctx->API == API_OPENGL_CORE) {
 595       ctx->Const.MaxViewports = GEN6_NUM_VIEWPORTS;
 596       ctx->Const.ViewportSubpixelBits = 0;
 597
 598       /* Cast to float before negating because MaxViewportWidth is unsigned.
 599        */
 600       ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth;
 601       ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth;
 602    }
 603
 604    /* ARB_gpu_shader5 */
 605    if (brw->gen >= 7)
 606       ctx->Const.MaxVertexStreams = MIN2(4, MAX_VERTEX_STREAMS);
 607
 608    /* ARB_framebuffer_no_attachments */
 609    ctx->Const.MaxFramebufferWidth = ctx->Const.MaxViewportWidth;
 610    ctx->Const.MaxFramebufferHeight = ctx->Const.MaxViewportHeight;
 611    ctx->Const.MaxFramebufferLayers = ctx->Const.MaxArrayTextureLayers;
 612    ctx->Const.MaxFramebufferSamples = max_samples;
 613 }
 614
 615 static void
 616 brw_adjust_cs_context_constants(struct brw_context *brw)
 617 {
 618    struct gl_context *ctx = &brw->ctx;
 619
 620    /* For ES, we set these constants based on SIMD8.
 621     *
 622     * TODO: Once we can always generate SIMD16, we should update this.
 623     *
 624     * For GL, we assume we can generate a SIMD16 program, but this currently
 625     * is not always true. This allows us to run more test cases, and will be
 626     * required based on desktop GL compute shader requirements.
 627     */
 628    const int simd_size = ctx->API == API_OPENGL_CORE ? 16 : 8;
 629
 630    const uint32_t max_invocations = simd_size * brw->max_cs_threads;
 631    ctx->Const.MaxComputeWorkGroupSize[0] = max_invocations;
 632    ctx->Const.MaxComputeWorkGroupSize[1] = max_invocations;
 633    ctx->Const.MaxComputeWorkGroupSize[2] = max_invocations;
 634    ctx->Const.MaxComputeWorkGroupInvocations = max_invocations;
 635 }
 636
 637 /**
 638  * Process driconf (drirc) options, setting appropriate context flags.
 639  *
 640  * intelInitExtensions still pokes at optionCache directly, in order to
 641  * avoid advertising various extensions.  No flags are set, so it makes
 642  * sense to continue doing that there.
 643  */
 644 static void
 645 brw_process_driconf_options(struct brw_context *brw)
 646 {
 647    struct gl_context *ctx = &brw->ctx;
 648
 649    driOptionCache *options = &brw->optionCache;
 650    driParseConfigFiles(options, &brw->intelScreen->optionCache,
 651                        brw->driContext->driScreenPriv->myNum, "i965");
 652
 653    int bo_reuse_mode = driQueryOptioni(options, "bo_reuse");
 654    switch (bo_reuse_mode) {
 655    case DRI_CONF_BO_REUSE_DISABLED:
 656       break;
 657    case DRI_CONF_BO_REUSE_ALL:
 658       intel_bufmgr_gem_enable_reuse(brw->bufmgr);
 659       break;
 660    }
 661
 662    if (!driQueryOptionb(options, "hiz")) {
 663        brw->has_hiz = false;
 664        /* On gen6, you can only do separate stencil with HIZ. */
 665        if (brw->gen == 6)
 666           brw->has_separate_stencil = false;
 667    }
 668
 669    if (driQueryOptionb(options, "always_flush_batch")) {
 670       fprintf(stderr, "flushing batchbuffer before/after each draw call\n");
 671       brw->always_flush_batch = true;
 672    }
 673
 674    if (driQueryOptionb(options, "always_flush_cache")) {
 675       fprintf(stderr, "flushing GPU caches before/after each draw call\n");
 676       brw->always_flush_cache = true;
 677    }
 678
 679    if (driQueryOptionb(options, "disable_throttling")) {
 680       fprintf(stderr, "disabling flush throttling\n");
 681       brw->disable_throttling = true;
 682    }
 683
 684    brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile");
 685
 686    ctx->Const.ForceGLSLExtensionsWarn =
 687       driQueryOptionb(options, "force_glsl_extensions_warn");
 688
 689    ctx->Const.DisableGLSLLineContinuations =
 690       driQueryOptionb(options, "disable_glsl_line_continuations");
 691
 692    ctx->Const.AllowGLSLExtensionDirectiveMidShader =
 693       driQueryOptionb(options, "allow_glsl_extension_directive_midshader");
 694 }
 695
 696 GLboolean
 697 brwCreateContext(gl_api api,
 698                  const struct gl_config *mesaVis,
 699                  __DRIcontext *driContextPriv,
 700                  unsigned major_version,
 701                  unsigned minor_version,
 702                  uint32_t flags,
 703                  bool notify_reset,
 704                  unsigned *dri_ctx_error,
 705                  void *sharedContextPrivate)
 706 {
 707    __DRIscreen *sPriv = driContextPriv->driScreenPriv;
 708    struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate;
 709    struct intel_screen *screen = sPriv->driverPrivate;
 710    const struct brw_device_info *devinfo = screen->devinfo;
 711    struct dd_function_table functions;
 712
 713    /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
 714     * provides us with context reset notifications.
 715     */
 716    uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG
 717       | __DRI_CTX_FLAG_FORWARD_COMPATIBLE;
 718
 719    if (screen->has_context_reset_notification)
 720       allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS;
 721
 722    if (flags & ~allowed_flags) {
 723       *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG;
 724       return false;
 725    }
 726
 727    struct brw_context *brw = rzalloc(NULL, struct brw_context);
 728    if (!brw) {
 729       fprintf(stderr, "%s: failed to alloc context\n", __func__);
 730       *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
 731       return false;
 732    }
 733
 734    driContextPriv->driverPrivate = brw;
 735    brw->driContext = driContextPriv;
 736    brw->intelScreen = screen;
 737    brw->bufmgr = screen->bufmgr;
 738
 739    brw->gen = devinfo->gen;
 740    brw->gt = devinfo->gt;
 741    brw->is_g4x = devinfo->is_g4x;
 742    brw->is_baytrail = devinfo->is_baytrail;
 743    brw->is_haswell = devinfo->is_haswell;
 744    brw->is_cherryview = devinfo->is_cherryview;
 745    brw->is_broxton = devinfo->is_broxton;
 746    brw->has_llc = devinfo->has_llc;
 747    brw->has_hiz = devinfo->has_hiz_and_separate_stencil;
 748    brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil;
 749    brw->has_pln = devinfo->has_pln;
 750    brw->has_compr4 = devinfo->has_compr4;
 751    brw->has_surface_tile_offset = devinfo->has_surface_tile_offset;
 752    brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug;
 753    brw->needs_unlit_centroid_workaround =
 754       devinfo->needs_unlit_centroid_workaround;
 755
 756    brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil;
 757    brw->has_swizzling = screen->hw_has_swizzling;
 758
 759    brw->vs.base.stage = MESA_SHADER_VERTEX;
 760    brw->gs.base.stage = MESA_SHADER_GEOMETRY;
 761    brw->wm.base.stage = MESA_SHADER_FRAGMENT;
 762    if (brw->gen >= 8) {
 763       gen8_init_vtable_surface_functions(brw);
 764       brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz;
 765    } else if (brw->gen >= 7) {
 766       gen7_init_vtable_surface_functions(brw);
 767       brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz;
 768    } else if (brw->gen >= 6) {
 769       gen6_init_vtable_surface_functions(brw);
 770       brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz;
 771    } else {
 772       gen4_init_vtable_surface_functions(brw);
 773       brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz;
 774    }
 775
 776    brw_init_driver_functions(brw, &functions);
 777
 778    if (notify_reset)
 779       functions.GetGraphicsResetStatus = brw_get_graphics_reset_status;
 780
 781    struct gl_context *ctx = &brw->ctx;
 782
 783    if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) {
 784       *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
 785       fprintf(stderr, "%s: failed to init mesa context\n", __func__);
 786       intelDestroyContext(driContextPriv);
 787       return false;
 788    }
 789
 790    driContextSetFlags(ctx, flags);
 791
 792    /* Initialize the software rasterizer and helper modules.
 793     *
 794     * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
 795     * software fallbacks (which we have to support on legacy GL to do weird
 796     * glDrawPixels(), glBitmap(), and other functions).
 797     */
 798    if (api != API_OPENGL_CORE && api != API_OPENGLES2) {
 799       _swrast_CreateContext(ctx);
 800    }
 801
 802    _vbo_CreateContext(ctx);
 803    if (ctx->swrast_context) {
 804       _tnl_CreateContext(ctx);
 805       TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
 806       _swsetup_CreateContext(ctx);
 807
 808       /* Configure swrast to match hardware characteristics: */
 809       _swrast_allow_pixel_fog(ctx, false);
 810       _swrast_allow_vertex_fog(ctx, true);
 811    }
 812
 813    _mesa_meta_init(ctx);
 814
 815    brw_process_driconf_options(brw);
 816
 817    if (INTEL_DEBUG & DEBUG_PERF)
 818       brw->perf_debug = true;
 819
 820    brw_initialize_context_constants(brw);
 821
 822    ctx->Const.ResetStrategy = notify_reset
 823       ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB;
 824
 825    /* Reinitialize the context point state.  It depends on ctx->Const values. */
 826    _mesa_init_point(ctx);
 827
 828    intel_fbo_init(brw);
 829
 830    intel_batchbuffer_init(brw);
 831
 832    if (brw->gen >= 6) {
 833       /* Create a new hardware context.  Using a hardware context means that
 834        * our GPU state will be saved/restored on context switch, allowing us
 835        * to assume that the GPU is in the same state we left it in.
 836        *
 837        * This is required for transform feedback buffer offsets, query objects,
 838        * and also allows us to reduce how much state we have to emit.
 839        */
 840       brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr);
 841
 842       if (!brw->hw_ctx) {
 843          fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n");
 844          intelDestroyContext(driContextPriv);
 845          return false;
 846       }
 847    }
 848
 849    if (brw_init_pipe_control(brw, devinfo)) {
 850       *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
 851       intelDestroyContext(driContextPriv);
 852       return false;
 853    }
 854
 855    brw_init_state(brw);
 856
 857    intelInitExtensions(ctx);
 858
 859    brw_init_surface_formats(brw);
 860
 861    brw->max_vs_threads = devinfo->max_vs_threads;
 862    brw->max_hs_threads = devinfo->max_hs_threads;
 863    brw->max_ds_threads = devinfo->max_ds_threads;
 864    brw->max_gs_threads = devinfo->max_gs_threads;
 865    brw->max_wm_threads = devinfo->max_wm_threads;
 866    brw->max_cs_threads = devinfo->max_cs_threads;
 867    brw->urb.size = devinfo->urb.size;
 868    brw->urb.min_vs_entries = devinfo->urb.min_vs_entries;
 869    brw->urb.max_vs_entries = devinfo->urb.max_vs_entries;
 870    brw->urb.max_hs_entries = devinfo->urb.max_hs_entries;
 871    brw->urb.max_ds_entries = devinfo->urb.max_ds_entries;
 872    brw->urb.max_gs_entries = devinfo->urb.max_gs_entries;
 873
 874    brw_adjust_cs_context_constants(brw);
 875
 876    /* Estimate the size of the mappable aperture into the GTT.  There's an
 877     * ioctl to get the whole GTT size, but not one to get the mappable subset.
 878     * It turns out it's basically always 256MB, though some ancient hardware
 879     * was smaller.
 880     */
 881    uint32_t gtt_size = 256 * 1024 * 1024;
 882
 883    /* We don't want to map two objects such that a memcpy between them would
 884     * just fault one mapping in and then the other over and over forever.  So
 885     * we would need to divide the GTT size by 2.  Additionally, some GTT is
 886     * taken up by things like the framebuffer and the ringbuffer and such, so
 887     * be more conservative.
 888     */
 889    brw->max_gtt_map_object_size = gtt_size / 4;
 890
 891    if (brw->gen == 6)
 892       brw->urb.gs_present = false;
 893
 894    brw->prim_restart.in_progress = false;
 895    brw->prim_restart.enable_cut_index = false;
 896    brw->gs.enabled = false;
 897    brw->sf.viewport_transform_enable = true;
 898
 899    brw->predicate.state = BRW_PREDICATE_STATE_RENDER;
 900
 901    brw->use_resource_streamer = screen->has_resource_streamer &&
 902       (brw_env_var_as_boolean("INTEL_USE_HW_BT", false) ||
 903        brw_env_var_as_boolean("INTEL_USE_GATHER", false));
 904
 905    ctx->VertexProgram._MaintainTnlProgram = true;
 906    ctx->FragmentProgram._MaintainTexEnvProgram = true;
 907
 908    brw_draw_init( brw );
 909
 910    if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) {
 911       /* Turn on some extra GL_ARB_debug_output generation. */
 912       brw->perf_debug = true;
 913    }
 914
 915    if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0)
 916       ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB;
 917
 918    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 919       brw_init_shader_time(brw);
 920
 921    _mesa_compute_version(ctx);
 922
 923    _mesa_initialize_dispatch_tables(ctx);
 924    _mesa_initialize_vbo_vtxfmt(ctx);
 925
 926    if (ctx->Extensions.AMD_performance_monitor) {
 927       brw_init_performance_monitors(brw);
 928    }
 929
 930    vbo_use_buffer_objects(ctx);
 931    vbo_always_unmap_buffers(ctx);
 932
 933    return true;
 934 }
 935
 936 void
 937 intelDestroyContext(__DRIcontext * driContextPriv)
 938 {
 939    struct brw_context *brw =
 940       (struct brw_context *) driContextPriv->driverPrivate;
 941    struct gl_context *ctx = &brw->ctx;
 942
 943    /* Dump a final BMP in case the application doesn't call SwapBuffers */
 944    if (INTEL_DEBUG & DEBUG_AUB) {
 945       intel_batchbuffer_flush(brw);
 946       aub_dump_bmp(&brw->ctx);
 947    }
 948
 949    _mesa_meta_free(&brw->ctx);
 950    brw_meta_fast_clear_free(brw);
 951
 952    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 953       /* Force a report. */
 954       brw->shader_time.report_time = 0;
 955
 956       brw_collect_and_report_shader_time(brw);
 957       brw_destroy_shader_time(brw);
 958    }
 959
 960    brw_destroy_state(brw);
 961    brw_draw_destroy(brw);
 962
 963    drm_intel_bo_unreference(brw->curbe.curbe_bo);
 964    if (brw->vs.base.scratch_bo)
 965       drm_intel_bo_unreference(brw->vs.base.scratch_bo);
 966    if (brw->gs.base.scratch_bo)
 967       drm_intel_bo_unreference(brw->gs.base.scratch_bo);
 968    if (brw->wm.base.scratch_bo)
 969       drm_intel_bo_unreference(brw->wm.base.scratch_bo);
 970
 971    gen7_reset_hw_bt_pool_offsets(brw);
 972    drm_intel_bo_unreference(brw->hw_bt_pool.bo);
 973    brw->hw_bt_pool.bo = NULL;
 974
 975    drm_intel_gem_context_destroy(brw->hw_ctx);
 976
 977    if (ctx->swrast_context) {
 978       _swsetup_DestroyContext(&brw->ctx);
 979       _tnl_DestroyContext(&brw->ctx);
 980    }
 981    _vbo_DestroyContext(&brw->ctx);
 982
 983    if (ctx->swrast_context)
 984       _swrast_DestroyContext(&brw->ctx);
 985
 986    brw_fini_pipe_control(brw);
 987    intel_batchbuffer_free(brw);
 988
 989    drm_intel_bo_unreference(brw->throttle_batch[1]);
 990    drm_intel_bo_unreference(brw->throttle_batch[0]);
 991    brw->throttle_batch[1] = NULL;
 992    brw->throttle_batch[0] = NULL;
 993
 994    driDestroyOptionCache(&brw->optionCache);
 995
 996    /* free the Mesa context */
 997    _mesa_free_context_data(&brw->ctx);
 998
 999    ralloc_free(brw);
1000    driContextPriv->driverPrivate = NULL;
1001 }
1002
1003 GLboolean
1004 intelUnbindContext(__DRIcontext * driContextPriv)
1005 {
1006    /* Unset current context and dispath table */
1007    _mesa_make_current(NULL, NULL, NULL);
1008
1009    return true;
1010 }
1011
1012 /**
1013  * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1014  * on window system framebuffers.
1015  *
1016  * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1017  * your renderbuffer can do sRGB encode, and you can flip a switch that does
1018  * sRGB encode if the renderbuffer can handle it.  You can ask specifically
1019  * for a visual where you're guaranteed to be capable, but it turns out that
1020  * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1021  * incapable ones, because there's no difference between the two in resources
1022  * used.  Applications thus get built that accidentally rely on the default
1023  * visual choice being sRGB, so we make ours sRGB capable.  Everything sounds
1024  * great...
1025  *
1026  * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1027  * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1028  * So they removed the enable knob and made it "if the renderbuffer is sRGB
1029  * capable, do sRGB encode".  Then, for your window system renderbuffers, you
1030  * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1031  * and get no sRGB encode (assuming that both kinds of visual are available).
1032  * Thus our choice to support sRGB by default on our visuals for desktop would
1033  * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1034  *
1035  * Unfortunately, renderbuffer setup happens before a context is created.  So
1036  * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1037  * context (without an sRGB visual, though we don't have sRGB visuals exposed
1038  * yet), we go turn that back off before anyone finds out.
1039  */
1040 static void
1041 intel_gles3_srgb_workaround(struct brw_context *brw,
1042                             struct gl_framebuffer *fb)
1043 {
1044    struct gl_context *ctx = &brw->ctx;
1045
1046    if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable)
1047       return;
1048
1049    /* Some day when we support the sRGB capable bit on visuals available for
1050     * GLES, we'll need to respect that and not disable things here.
1051     */
1052    fb->Visual.sRGBCapable = false;
1053    for (int i = 0; i < BUFFER_COUNT; i++) {
1054       if (fb->Attachment[i].Renderbuffer &&
1055           fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) {
1056          fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM;
1057       }
1058    }
1059 }
1060
1061 GLboolean
1062 intelMakeCurrent(__DRIcontext * driContextPriv,
1063                  __DRIdrawable * driDrawPriv,
1064                  __DRIdrawable * driReadPriv)
1065 {
1066    struct brw_context *brw;
1067    GET_CURRENT_CONTEXT(curCtx);
1068
1069    if (driContextPriv)
1070       brw = (struct brw_context *) driContextPriv->driverPrivate;
1071    else
1072       brw = NULL;
1073
1074    /* According to the glXMakeCurrent() man page: "Pending commands to
1075     * the previous context, if any, are flushed before it is released."
1076     * But only flush if we're actually changing contexts.
1077     */
1078    if (brw_context(curCtx) && brw_context(curCtx) != brw) {
1079       _mesa_flush(curCtx);
1080    }
1081
1082    if (driContextPriv) {
1083       struct gl_context *ctx = &brw->ctx;
1084       struct gl_framebuffer *fb, *readFb;
1085
1086       if (driDrawPriv == NULL) {
1087          fb = _mesa_get_incomplete_framebuffer();
1088       } else {
1089          fb = driDrawPriv->driverPrivate;
1090          driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1;
1091       }
1092
1093       if (driReadPriv == NULL) {
1094          readFb = _mesa_get_incomplete_framebuffer();
1095       } else {
1096          readFb = driReadPriv->driverPrivate;
1097          driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1;
1098       }
1099
1100       /* The sRGB workaround changes the renderbuffer's format. We must change
1101        * the format before the renderbuffer's miptree get's allocated, otherwise
1102        * the formats of the renderbuffer and its miptree will differ.
1103        */
1104       intel_gles3_srgb_workaround(brw, fb);
1105       intel_gles3_srgb_workaround(brw, readFb);
1106
1107       /* If the context viewport hasn't been initialized, force a call out to
1108        * the loader to get buffers so we have a drawable size for the initial
1109        * viewport. */
1110       if (!brw->ctx.ViewportInitialized)
1111          intel_prepare_render(brw);
1112
1113       _mesa_make_current(ctx, fb, readFb);
1114    } else {
1115       _mesa_make_current(NULL, NULL, NULL);
1116    }
1117
1118    return true;
1119 }
1120
1121 void
1122 intel_resolve_for_dri2_flush(struct brw_context *brw,
1123                              __DRIdrawable *drawable)
1124 {
1125    if (brw->gen < 6) {
1126       /* MSAA and fast color clear are not supported, so don't waste time
1127        * checking whether a resolve is needed.
1128        */
1129       return;
1130    }
1131
1132    struct gl_framebuffer *fb = drawable->driverPrivate;
1133    struct intel_renderbuffer *rb;
1134
1135    /* Usually, only the back buffer will need to be downsampled. However,
1136     * the front buffer will also need it if the user has rendered into it.
1137     */
1138    static const gl_buffer_index buffers[2] = {
1139          BUFFER_BACK_LEFT,
1140          BUFFER_FRONT_LEFT,
1141    };
1142
1143    for (int i = 0; i < 2; ++i) {
1144       rb = intel_get_renderbuffer(fb, buffers[i]);
1145       if (rb == NULL || rb->mt == NULL)
1146          continue;
1147       if (rb->mt->num_samples <= 1)
1148          intel_miptree_resolve_color(brw, rb->mt);
1149       else
1150          intel_renderbuffer_downsample(brw, rb);
1151    }
1152 }
1153
1154 static unsigned
1155 intel_bits_per_pixel(const struct intel_renderbuffer *rb)
1156 {
1157    return _mesa_get_format_bytes(intel_rb_format(rb)) * 8;
1158 }
1159
1160 static void
1161 intel_query_dri2_buffers(struct brw_context *brw,
1162                          __DRIdrawable *drawable,
1163                          __DRIbuffer **buffers,
1164                          int *count);
1165
1166 static void
1167 intel_process_dri2_buffer(struct brw_context *brw,
1168                           __DRIdrawable *drawable,
1169                           __DRIbuffer *buffer,
1170                           struct intel_renderbuffer *rb,
1171                           const char *buffer_name);
1172
1173 static void
1174 intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable);
1175
1176 static void
1177 intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable)
1178 {
1179    struct gl_framebuffer *fb = drawable->driverPrivate;
1180    struct intel_renderbuffer *rb;
1181    __DRIbuffer *buffers = NULL;
1182    int i, count;
1183    const char *region_name;
1184
1185    /* Set this up front, so that in case our buffers get invalidated
1186     * while we're getting new buffers, we don't clobber the stamp and
1187     * thus ignore the invalidate. */
1188    drawable->lastStamp = drawable->dri2.stamp;
1189
1190    if (unlikely(INTEL_DEBUG & DEBUG_DRI))
1191       fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
1192
1193    intel_query_dri2_buffers(brw, drawable, &buffers, &count);
1194
1195    if (buffers == NULL)
1196       return;
1197
1198    for (i = 0; i < count; i++) {
1199        switch (buffers[i].attachment) {
1200        case __DRI_BUFFER_FRONT_LEFT:
1201            rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1202            region_name = "dri2 front buffer";
1203            break;
1204
1205        case __DRI_BUFFER_FAKE_FRONT_LEFT:
1206            rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1207            region_name = "dri2 fake front buffer";
1208            break;
1209
1210        case __DRI_BUFFER_BACK_LEFT:
1211            rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1212            region_name = "dri2 back buffer";
1213            break;
1214
1215        case __DRI_BUFFER_DEPTH:
1216        case __DRI_BUFFER_HIZ:
1217        case __DRI_BUFFER_DEPTH_STENCIL:
1218        case __DRI_BUFFER_STENCIL:
1219        case __DRI_BUFFER_ACCUM:
1220        default:
1221            fprintf(stderr,
1222                    "unhandled buffer attach event, attachment type %d\n",
1223                    buffers[i].attachment);
1224            return;
1225        }
1226
1227        intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name);
1228    }
1229
1230 }
1231
1232 void
1233 intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
1234 {
1235    struct brw_context *brw = context->driverPrivate;
1236    __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1237
1238    /* Set this up front, so that in case our buffers get invalidated
1239     * while we're getting new buffers, we don't clobber the stamp and
1240     * thus ignore the invalidate. */
1241    drawable->lastStamp = drawable->dri2.stamp;
1242
1243    if (unlikely(INTEL_DEBUG & DEBUG_DRI))
1244       fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
1245
1246    if (screen->image.loader)
1247       intel_update_image_buffers(brw, drawable);
1248    else
1249       intel_update_dri2_buffers(brw, drawable);
1250
1251    driUpdateFramebufferSize(&brw->ctx, drawable);
1252 }
1253
1254 /**
1255  * intel_prepare_render should be called anywhere that curent read/drawbuffer
1256  * state is required.
1257  */
1258 void
1259 intel_prepare_render(struct brw_context *brw)
1260 {
1261    struct gl_context *ctx = &brw->ctx;
1262    __DRIcontext *driContext = brw->driContext;
1263    __DRIdrawable *drawable;
1264
1265    drawable = driContext->driDrawablePriv;
1266    if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) {
1267       if (drawable->lastStamp != drawable->dri2.stamp)
1268          intel_update_renderbuffers(driContext, drawable);
1269       driContext->dri2.draw_stamp = drawable->dri2.stamp;
1270    }
1271
1272    drawable = driContext->driReadablePriv;
1273    if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) {
1274       if (drawable->lastStamp != drawable->dri2.stamp)
1275          intel_update_renderbuffers(driContext, drawable);
1276       driContext->dri2.read_stamp = drawable->dri2.stamp;
1277    }
1278
1279    /* If we're currently rendering to the front buffer, the rendering
1280     * that will happen next will probably dirty the front buffer.  So
1281     * mark it as dirty here.
1282     */
1283    if (_mesa_is_front_buffer_drawing(ctx->DrawBuffer))
1284       brw->front_buffer_dirty = true;
1285 }
1286
1287 /**
1288  * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1289  *
1290  * To determine which DRI buffers to request, examine the renderbuffers
1291  * attached to the drawable's framebuffer. Then request the buffers with
1292  * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1293  *
1294  * This is called from intel_update_renderbuffers().
1295  *
1296  * \param drawable      Drawable whose buffers are queried.
1297  * \param buffers       [out] List of buffers returned by DRI2 query.
1298  * \param buffer_count  [out] Number of buffers returned.
1299  *
1300  * \see intel_update_renderbuffers()
1301  * \see DRI2GetBuffers()
1302  * \see DRI2GetBuffersWithFormat()
1303  */
1304 static void
1305 intel_query_dri2_buffers(struct brw_context *brw,
1306                          __DRIdrawable *drawable,
1307                          __DRIbuffer **buffers,
1308                          int *buffer_count)
1309 {
1310    __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1311    struct gl_framebuffer *fb = drawable->driverPrivate;
1312    int i = 0;
1313    unsigned attachments[8];
1314
1315    struct intel_renderbuffer *front_rb;
1316    struct intel_renderbuffer *back_rb;
1317
1318    front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1319    back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1320
1321    memset(attachments, 0, sizeof(attachments));
1322    if ((_mesa_is_front_buffer_drawing(fb) ||
1323         _mesa_is_front_buffer_reading(fb) ||
1324         !back_rb) && front_rb) {
1325       /* If a fake front buffer is in use, then querying for
1326        * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1327        * the real front buffer to the fake front buffer.  So before doing the
1328        * query, we need to make sure all the pending drawing has landed in the
1329        * real front buffer.
1330        */
1331       intel_batchbuffer_flush(brw);
1332       intel_flush_front(&brw->ctx);
1333
1334       attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
1335       attachments[i++] = intel_bits_per_pixel(front_rb);
1336    } else if (front_rb && brw->front_buffer_dirty) {
1337       /* We have pending front buffer rendering, but we aren't querying for a
1338        * front buffer.  If the front buffer we have is a fake front buffer,
1339        * the X server is going to throw it away when it processes the query.
1340        * So before doing the query, make sure all the pending drawing has
1341        * landed in the real front buffer.
1342        */
1343       intel_batchbuffer_flush(brw);
1344       intel_flush_front(&brw->ctx);
1345    }
1346
1347    if (back_rb) {
1348       attachments[i++] = __DRI_BUFFER_BACK_LEFT;
1349       attachments[i++] = intel_bits_per_pixel(back_rb);
1350    }
1351
1352    assert(i <= ARRAY_SIZE(attachments));
1353
1354    *buffers = screen->dri2.loader->getBuffersWithFormat(drawable,
1355                                                         &drawable->w,
1356                                                         &drawable->h,
1357                                                         attachments, i / 2,
1358                                                         buffer_count,
1359                                                         drawable->loaderPrivate);
1360 }
1361
1362 /**
1363  * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1364  *
1365  * This is called from intel_update_renderbuffers().
1366  *
1367  * \par Note:
1368  *    DRI buffers whose attachment point is DRI2BufferStencil or
1369  *    DRI2BufferDepthStencil are handled as special cases.
1370  *
1371  * \param buffer_name is a human readable name, such as "dri2 front buffer",
1372  *        that is passed to drm_intel_bo_gem_create_from_name().
1373  *
1374  * \see intel_update_renderbuffers()
1375  */
1376 static void
1377 intel_process_dri2_buffer(struct brw_context *brw,
1378                           __DRIdrawable *drawable,
1379                           __DRIbuffer *buffer,
1380                           struct intel_renderbuffer *rb,
1381                           const char *buffer_name)
1382 {
1383    struct gl_framebuffer *fb = drawable->driverPrivate;
1384    drm_intel_bo *bo;
1385
1386    if (!rb)
1387       return;
1388
1389    unsigned num_samples = rb->Base.Base.NumSamples;
1390
1391    /* We try to avoid closing and reopening the same BO name, because the first
1392     * use of a mapping of the buffer involves a bunch of page faulting which is
1393     * moderately expensive.
1394     */
1395    struct intel_mipmap_tree *last_mt;
1396    if (num_samples == 0)
1397       last_mt = rb->mt;
1398    else
1399       last_mt = rb->singlesample_mt;
1400
1401    uint32_t old_name = 0;
1402    if (last_mt) {
1403        /* The bo already has a name because the miptree was created by a
1404         * previous call to intel_process_dri2_buffer(). If a bo already has a
1405         * name, then drm_intel_bo_flink() is a low-cost getter.  It does not
1406         * create a new name.
1407         */
1408       drm_intel_bo_flink(last_mt->bo, &old_name);
1409    }
1410
1411    if (old_name == buffer->name)
1412       return;
1413
1414    if (unlikely(INTEL_DEBUG & DEBUG_DRI)) {
1415       fprintf(stderr,
1416               "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1417               buffer->name, buffer->attachment,
1418               buffer->cpp, buffer->pitch);
1419    }
1420
1421    bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name,
1422                                           buffer->name);
1423    if (!bo) {
1424       fprintf(stderr,
1425               "Failed to open BO for returned DRI2 buffer "
1426               "(%dx%d, %s, named %d).\n"
1427               "This is likely a bug in the X Server that will lead to a "
1428               "crash soon.\n",
1429               drawable->w, drawable->h, buffer_name, buffer->name);
1430       return;
1431    }
1432
1433    intel_update_winsys_renderbuffer_miptree(brw, rb, bo,
1434                                             drawable->w, drawable->h,
1435                                             buffer->pitch);
1436
1437    if (_mesa_is_front_buffer_drawing(fb) &&
1438        (buffer->attachment == __DRI_BUFFER_FRONT_LEFT ||
1439         buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) &&
1440        rb->Base.Base.NumSamples > 1) {
1441       intel_renderbuffer_upsample(brw, rb);
1442    }
1443
1444    assert(rb->mt);
1445
1446    drm_intel_bo_unreference(bo);
1447 }
1448
1449 /**
1450  * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1451  *
1452  * To determine which DRI buffers to request, examine the renderbuffers
1453  * attached to the drawable's framebuffer. Then request the buffers from
1454  * the image loader
1455  *
1456  * This is called from intel_update_renderbuffers().
1457  *
1458  * \param drawable      Drawable whose buffers are queried.
1459  * \param buffers       [out] List of buffers returned by DRI2 query.
1460  * \param buffer_count  [out] Number of buffers returned.
1461  *
1462  * \see intel_update_renderbuffers()
1463  */
1464
1465 static void
1466 intel_update_image_buffer(struct brw_context *intel,
1467                           __DRIdrawable *drawable,
1468                           struct intel_renderbuffer *rb,
1469                           __DRIimage *buffer,
1470                           enum __DRIimageBufferMask buffer_type)
1471 {
1472    struct gl_framebuffer *fb = drawable->driverPrivate;
1473
1474    if (!rb || !buffer->bo)
1475       return;
1476
1477    unsigned num_samples = rb->Base.Base.NumSamples;
1478
1479    /* Check and see if we're already bound to the right
1480     * buffer object
1481     */
1482    struct intel_mipmap_tree *last_mt;
1483    if (num_samples == 0)
1484       last_mt = rb->mt;
1485    else
1486       last_mt = rb->singlesample_mt;
1487
1488    if (last_mt && last_mt->bo == buffer->bo)
1489       return;
1490
1491    intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo,
1492                                             buffer->width, buffer->height,
1493                                             buffer->pitch);
1494
1495    if (_mesa_is_front_buffer_drawing(fb) &&
1496        buffer_type == __DRI_IMAGE_BUFFER_FRONT &&
1497        rb->Base.Base.NumSamples > 1) {
1498       intel_renderbuffer_upsample(intel, rb);
1499    }
1500 }
1501
1502 static void
1503 intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable)
1504 {
1505    struct gl_framebuffer *fb = drawable->driverPrivate;
1506    __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1507    struct intel_renderbuffer *front_rb;
1508    struct intel_renderbuffer *back_rb;
1509    struct __DRIimageList images;
1510    unsigned int format;
1511    uint32_t buffer_mask = 0;
1512
1513    front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1514    back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1515
1516    if (back_rb)
1517       format = intel_rb_format(back_rb);
1518    else if (front_rb)
1519       format = intel_rb_format(front_rb);
1520    else
1521       return;
1522
1523    if (front_rb && (_mesa_is_front_buffer_drawing(fb) ||
1524                     _mesa_is_front_buffer_reading(fb) || !back_rb)) {
1525       buffer_mask |= __DRI_IMAGE_BUFFER_FRONT;
1526    }
1527
1528    if (back_rb)
1529       buffer_mask |= __DRI_IMAGE_BUFFER_BACK;
1530
1531    (*screen->image.loader->getBuffers) (drawable,
1532                                         driGLFormatToImageFormat(format),
1533                                         &drawable->dri2.stamp,
1534                                         drawable->loaderPrivate,
1535                                         buffer_mask,
1536                                         &images);
1537
1538    if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) {
1539       drawable->w = images.front->width;
1540       drawable->h = images.front->height;
1541       intel_update_image_buffer(brw,
1542                                 drawable,
1543                                 front_rb,
1544                                 images.front,
1545                                 __DRI_IMAGE_BUFFER_FRONT);
1546    }
1547    if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) {
1548       drawable->w = images.back->width;
1549       drawable->h = images.back->height;
1550       intel_update_image_buffer(brw,
1551                                 drawable,
1552                                 back_rb,
1553                                 images.back,
1554                                 __DRI_IMAGE_BUFFER_BACK);
1555    }
1556 }