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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a 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, sublicense, 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
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keithw@vmware.com>
  30   */
  31
  32
  33
  34 #include "main/mtypes.h"
  35 #include "main/macros.h"
  36 #include "main/fbobject.h"
  37 #include "brw_context.h"
  38 #include "brw_state.h"
  39 #include "brw_defines.h"
  40 #include "brw_sf.h"
  41
  42 static void upload_sf_vp(struct brw_context *brw)
  43 {
  44    struct gl_context *ctx = &brw->ctx;
  45    const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
  46    struct brw_sf_viewport *sfv;
  47    GLfloat y_scale, y_bias;
  48    const bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
  49    const GLfloat *v = ctx->ViewportArray[0]._WindowMap.m;
  50
  51    sfv = brw_state_batch(brw, AUB_TRACE_SF_VP_STATE,
  52                          sizeof(*sfv), 32, &brw->sf.vp_offset);
  53    memset(sfv, 0, sizeof(*sfv));
  54
  55    if (render_to_fbo) {
  56       y_scale = 1.0;
  57       y_bias = 0;
  58    }
  59    else {
  60       y_scale = -1.0;
  61       y_bias = ctx->DrawBuffer->Height;
  62    }
  63
  64    /* _NEW_VIEWPORT */
  65
  66    sfv->viewport.m00 = v[MAT_SX];
  67    sfv->viewport.m11 = v[MAT_SY] * y_scale;
  68    sfv->viewport.m22 = v[MAT_SZ] * depth_scale;
  69    sfv->viewport.m30 = v[MAT_TX];
  70    sfv->viewport.m31 = v[MAT_TY] * y_scale + y_bias;
  71    sfv->viewport.m32 = v[MAT_TZ] * depth_scale;
  72
  73    /* _NEW_SCISSOR | _NEW_BUFFERS | _NEW_VIEWPORT
  74     * for DrawBuffer->_[XY]{min,max}
  75     */
  76
  77    /* The scissor only needs to handle the intersection of drawable
  78     * and scissor rect, since there are no longer cliprects for shared
  79     * buffers with DRI2.
  80     *
  81     * Note that the hardware's coordinates are inclusive, while Mesa's min is
  82     * inclusive but max is exclusive.
  83     */
  84
  85    if (ctx->DrawBuffer->_Xmin == ctx->DrawBuffer->_Xmax ||
  86        ctx->DrawBuffer->_Ymin == ctx->DrawBuffer->_Ymax) {
  87       /* If the scissor was out of bounds and got clamped to 0
  88        * width/height at the bounds, the subtraction of 1 from
  89        * maximums could produce a negative number and thus not clip
  90        * anything.  Instead, just provide a min > max scissor inside
  91        * the bounds, which produces the expected no rendering.
  92        */
  93       sfv->scissor.xmin = 1;
  94       sfv->scissor.xmax = 0;
  95       sfv->scissor.ymin = 1;
  96       sfv->scissor.ymax = 0;
  97    } else if (render_to_fbo) {
  98       /* texmemory: Y=0=bottom */
  99       sfv->scissor.xmin = ctx->DrawBuffer->_Xmin;
 100       sfv->scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
 101       sfv->scissor.ymin = ctx->DrawBuffer->_Ymin;
 102       sfv->scissor.ymax = ctx->DrawBuffer->_Ymax - 1;
 103    }
 104    else {
 105       /* memory: Y=0=top */
 106       sfv->scissor.xmin = ctx->DrawBuffer->_Xmin;
 107       sfv->scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
 108       sfv->scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
 109       sfv->scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
 110    }
 111
 112    brw->state.dirty.brw |= BRW_NEW_SF_VP;
 113 }
 114
 115 const struct brw_tracked_state brw_sf_vp = {
 116    .dirty = {
 117       .mesa  = _NEW_BUFFERS |
 118                _NEW_SCISSOR |
 119                _NEW_VIEWPORT,
 120       .brw   = BRW_NEW_BATCH,
 121       .cache = 0
 122    },
 123    .emit = upload_sf_vp
 124 };
 125
 126 static void upload_sf_unit( struct brw_context *brw )
 127 {
 128    struct gl_context *ctx = &brw->ctx;
 129    struct brw_sf_unit_state *sf;
 130    drm_intel_bo *bo = brw->batch.bo;
 131    int chipset_max_threads;
 132    bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 133
 134    sf = brw_state_batch(brw, AUB_TRACE_SF_STATE,
 135                         sizeof(*sf), 64, &brw->sf.state_offset);
 136
 137    memset(sf, 0, sizeof(*sf));
 138
 139    /* BRW_NEW_PROGRAM_CACHE | CACHE_NEW_SF_PROG */
 140    sf->thread0.grf_reg_count = ALIGN(brw->sf.prog_data->total_grf, 16) / 16 - 1;
 141    sf->thread0.kernel_start_pointer =
 142       brw_program_reloc(brw,
 143                         brw->sf.state_offset +
 144                         offsetof(struct brw_sf_unit_state, thread0),
 145                         brw->sf.prog_offset +
 146                         (sf->thread0.grf_reg_count << 1)) >> 6;
 147
 148    sf->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 149
 150    sf->thread3.dispatch_grf_start_reg = 3;
 151    sf->thread3.urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 152
 153    /* CACHE_NEW_SF_PROG */
 154    sf->thread3.urb_entry_read_length = brw->sf.prog_data->urb_read_length;
 155
 156    /* BRW_NEW_URB_FENCE */
 157    sf->thread4.nr_urb_entries = brw->urb.nr_sf_entries;
 158    sf->thread4.urb_entry_allocation_size = brw->urb.sfsize - 1;
 159
 160    /* Each SF thread produces 1 PUE, and there can be up to 24 (Pre-Ironlake) or
 161     * 48 (Ironlake) threads.
 162     */
 163    if (brw->gen == 5)
 164       chipset_max_threads = 48;
 165    else
 166       chipset_max_threads = 24;
 167
 168    /* BRW_NEW_URB_FENCE */
 169    sf->thread4.max_threads = MIN2(chipset_max_threads,
 170                                   brw->urb.nr_sf_entries) - 1;
 171
 172    if (unlikely(INTEL_DEBUG & DEBUG_STATS))
 173       sf->thread4.stats_enable = 1;
 174
 175    /* BRW_NEW_SF_VP */
 176    sf->sf5.sf_viewport_state_offset = (brw->batch.bo->offset64 +
 177                                        brw->sf.vp_offset) >> 5; /* reloc */
 178
 179    sf->sf5.viewport_transform = 1;
 180
 181    /* _NEW_SCISSOR */
 182    if (ctx->Scissor.EnableFlags)
 183       sf->sf6.scissor = 1;
 184
 185    /* _NEW_POLYGON */
 186    if (ctx->Polygon.FrontFace == GL_CCW)
 187       sf->sf5.front_winding = BRW_FRONTWINDING_CCW;
 188    else
 189       sf->sf5.front_winding = BRW_FRONTWINDING_CW;
 190
 191    /* _NEW_BUFFERS
 192     * The viewport is inverted for rendering to a FBO, and that inverts
 193     * polygon front/back orientation.
 194     */
 195    sf->sf5.front_winding ^= render_to_fbo;
 196
 197    /* _NEW_POLYGON */
 198    switch (ctx->Polygon.CullFlag ? ctx->Polygon.CullFaceMode : GL_NONE) {
 199    case GL_FRONT:
 200       sf->sf6.cull_mode = BRW_CULLMODE_FRONT;
 201       break;
 202    case GL_BACK:
 203       sf->sf6.cull_mode = BRW_CULLMODE_BACK;
 204       break;
 205    case GL_FRONT_AND_BACK:
 206       sf->sf6.cull_mode = BRW_CULLMODE_BOTH;
 207       break;
 208    case GL_NONE:
 209       sf->sf6.cull_mode = BRW_CULLMODE_NONE;
 210       break;
 211    default:
 212       unreachable("not reached");
 213    }
 214
 215    /* _NEW_LINE */
 216    sf->sf6.line_width =
 217       CLAMP(ctx->Line.Width, 1.0, ctx->Const.MaxLineWidth) * (1<<1);
 218
 219    sf->sf6.line_endcap_aa_region_width = 1;
 220    if (ctx->Line.SmoothFlag)
 221       sf->sf6.aa_enable = 1;
 222    else if (sf->sf6.line_width <= 0x2)
 223        sf->sf6.line_width = 0;
 224
 225    /* _NEW_BUFFERS */
 226    if (!render_to_fbo) {
 227       /* Rendering to an OpenGL window */
 228       sf->sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
 229    }
 230    else {
 231       /* If rendering to an FBO, the pixel coordinate system is
 232        * inverted with respect to the normal OpenGL coordinate
 233        * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
 234        * But this value is listed as "Reserved, but not seen as useful"
 235        * in Intel documentation (page 212, "Point Rasterization Rule",
 236        * section 7.4 "SF Pipeline State Summary", of document
 237        * "Intel® 965 Express Chipset Family and Intel® G35 Express
 238        * Chipset Graphics Controller Programmer's Reference Manual,
 239        * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
 240        * available at
 241        *     http://intellinuxgraphics.org/documentation.html
 242        * at the time of this writing).
 243        *
 244        * It does work on at least some devices, if not all;
 245        * if devices that don't support it can be identified,
 246        * the likely failure case is that points are rasterized
 247        * incorrectly, which is no worse than occurs without
 248        * the value, so we're using it here.
 249        */
 250       sf->sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
 251    }
 252    /* XXX clamp max depends on AA vs. non-AA */
 253
 254    /* _NEW_POINT */
 255    sf->sf7.sprite_point = ctx->Point.PointSprite;
 256    sf->sf7.point_size = CLAMP(rint(CLAMP(ctx->Point.Size,
 257                                          ctx->Point.MinSize,
 258                                          ctx->Point.MaxSize)), 1, 255) * (1<<3);
 259    /* _NEW_PROGRAM | _NEW_POINT */
 260    sf->sf7.use_point_size_state = !(ctx->VertexProgram.PointSizeEnabled ||
 261                                     ctx->Point._Attenuated);
 262    sf->sf7.aa_line_distance_mode = 0;
 263
 264    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
 265     * _NEW_LIGHT
 266     */
 267    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 268       sf->sf7.trifan_pv = 2;
 269       sf->sf7.linestrip_pv = 1;
 270       sf->sf7.tristrip_pv = 2;
 271    } else {
 272       sf->sf7.trifan_pv = 1;
 273       sf->sf7.linestrip_pv = 0;
 274       sf->sf7.tristrip_pv = 0;
 275    }
 276    sf->sf7.line_last_pixel_enable = 0;
 277
 278    /* Set bias for OpenGL rasterization rules:
 279     */
 280    sf->sf6.dest_org_vbias = 0x8;
 281    sf->sf6.dest_org_hbias = 0x8;
 282
 283    /* STATE_PREFETCH command description describes this state as being
 284     * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain.
 285     */
 286
 287    /* Emit SF viewport relocation */
 288    drm_intel_bo_emit_reloc(bo, (brw->sf.state_offset +
 289                                 offsetof(struct brw_sf_unit_state, sf5)),
 290                            brw->batch.bo, (brw->sf.vp_offset |
 291                                              sf->sf5.front_winding |
 292                                              (sf->sf5.viewport_transform << 1)),
 293                            I915_GEM_DOMAIN_INSTRUCTION, 0);
 294
 295    brw->state.dirty.brw |= BRW_NEW_GEN4_UNIT_STATE;
 296 }
 297
 298 const struct brw_tracked_state brw_sf_unit = {
 299    .dirty = {
 300       .mesa  = _NEW_BUFFERS |
 301                _NEW_LIGHT |
 302                _NEW_LINE |
 303                _NEW_POINT |
 304                _NEW_POLYGON |
 305                _NEW_PROGRAM |
 306                _NEW_SCISSOR,
 307       .brw   = BRW_NEW_BATCH |
 308                BRW_NEW_PROGRAM_CACHE |
 309                BRW_NEW_SF_VP |
 310                BRW_NEW_URB_FENCE,
 311       .cache = CACHE_NEW_SF_PROG,
 312    },
 313    .emit = upload_sf_unit,
 314 };