src/mesa/drivers/dri/i965/gen6_sf_state.c

   1 /*
   2  * Copyright © 2009 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  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #include "brw_context.h"
  29 #include "brw_state.h"
  30 #include "brw_defines.h"
  31 #include "brw_util.h"
  32 #include "main/macros.h"
  33 #include "main/fbobject.h"
  34 #include "intel_batchbuffer.h"
  35
  36 /**
  37  * Determine the appropriate attribute override value to store into the
  38  * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
  39  * override value contains two pieces of information: the location of the
  40  * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
  41  * flag indicating whether to "swizzle" the attribute based on the direction
  42  * the triangle is facing.
  43  *
  44  * If an attribute is "swizzled", then the given VUE location is used for
  45  * front-facing triangles, and the VUE location that immediately follows is
  46  * used for back-facing triangles.  We use this to implement the mapping from
  47  * gl_FrontColor/gl_BackColor to gl_Color.
  48  *
  49  * urb_entry_read_offset is the offset into the VUE at which the SF unit is
  50  * being instructed to begin reading attribute data.  It can be set to a
  51  * nonzero value to prevent the SF unit from wasting time reading elements of
  52  * the VUE that are not needed by the fragment shader.  It is measured in
  53  * 256-bit increments.
  54  */
  55 uint32_t
  56 get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
  57                   int fs_attr, bool two_side_color, uint32_t *max_source_attr)
  58 {
  59    if (fs_attr == VARYING_SLOT_POS) {
  60       /* This attribute will be overwritten by the fragment shader's
  61        * interpolation code (see emit_interp() in brw_wm_fp.c), so just let it
  62        * reference the first available attribute.
  63        */
  64       return 0;
  65    }
  66
  67    /* Find the VUE slot for this attribute. */
  68    int slot = vue_map->vert_result_to_slot[fs_attr];
  69
  70    /* If there was only a back color written but not front, use back
  71     * as the color instead of undefined
  72     */
  73    if (slot == -1 && fs_attr == VARYING_SLOT_COL0)
  74       slot = vue_map->vert_result_to_slot[VARYING_SLOT_BFC0];
  75    if (slot == -1 && fs_attr == VARYING_SLOT_COL1)
  76       slot = vue_map->vert_result_to_slot[VARYING_SLOT_BFC1];
  77
  78    if (slot == -1) {
  79       /* This attribute does not exist in the VUE--that means that the vertex
  80        * shader did not write to it.  This means that either:
  81        *
  82        * (a) This attribute is a texture coordinate, and it is going to be
  83        * replaced with point coordinates (as a consequence of a call to
  84        * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
  85        * hardware will ignore whatever attribute override we supply.
  86        *
  87        * (b) This attribute is read by the fragment shader but not written by
  88        * the vertex shader, so its value is undefined.  Therefore the
  89        * attribute override we supply doesn't matter.
  90        *
  91        * In either case the attribute override we supply doesn't matter, so
  92        * just reference the first available attribute.
  93        */
  94       return 0;
  95    }
  96
  97    /* Compute the location of the attribute relative to urb_entry_read_offset.
  98     * Each increment of urb_entry_read_offset represents a 256-bit value, so
  99     * it counts for two 128-bit VUE slots.
 100     */
 101    int source_attr = slot - 2 * urb_entry_read_offset;
 102    assert(source_attr >= 0 && source_attr < 32);
 103
 104    /* If we are doing two-sided color, and the VUE slot following this one
 105     * represents a back-facing color, then we need to instruct the SF unit to
 106     * do back-facing swizzling.
 107     */
 108    bool swizzling = two_side_color &&
 109       ((vue_map->slot_to_vert_result[slot] == VARYING_SLOT_COL0 &&
 110         vue_map->slot_to_vert_result[slot+1] == VARYING_SLOT_BFC0) ||
 111        (vue_map->slot_to_vert_result[slot] == VARYING_SLOT_COL1 &&
 112         vue_map->slot_to_vert_result[slot+1] == VARYING_SLOT_BFC1));
 113
 114    /* Update max_source_attr.  If swizzling, the SF will read this slot + 1. */
 115    if (*max_source_attr < source_attr + swizzling)
 116       *max_source_attr = source_attr + swizzling;
 117
 118    if (swizzling) {
 119       return source_attr |
 120          (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 121    }
 122
 123    return source_attr;
 124 }
 125
 126 static void
 127 upload_sf_state(struct brw_context *brw)
 128 {
 129    struct intel_context *intel = &brw->intel;
 130    struct gl_context *ctx = &intel->ctx;
 131    /* BRW_NEW_FRAGMENT_PROGRAM */
 132    uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
 133    /* _NEW_LIGHT */
 134    bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
 135    uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
 136    int i;
 137    /* _NEW_BUFFER */
 138    bool render_to_fbo = _mesa_is_user_fbo(brw->intel.ctx.DrawBuffer);
 139    bool multisampled_fbo = ctx->DrawBuffer->Visual.samples > 1;
 140
 141    int attr = 0, input_index = 0;
 142    int urb_entry_read_offset = 1;
 143    float point_size;
 144    uint16_t attr_overrides[VARYING_SLOT_MAX];
 145    uint32_t point_sprite_origin;
 146
 147    dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
 148
 149    dw2 = GEN6_SF_STATISTICS_ENABLE |
 150          GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 151
 152    dw3 = 0;
 153    dw4 = 0;
 154    dw16 = 0;
 155    dw17 = 0;
 156
 157    /* _NEW_POLYGON */
 158    if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
 159       dw2 |= GEN6_SF_WINDING_CCW;
 160
 161    if (ctx->Polygon.OffsetFill)
 162        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 163
 164    if (ctx->Polygon.OffsetLine)
 165        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 166
 167    if (ctx->Polygon.OffsetPoint)
 168        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 169
 170    switch (ctx->Polygon.FrontMode) {
 171    case GL_FILL:
 172        dw2 |= GEN6_SF_FRONT_SOLID;
 173        break;
 174
 175    case GL_LINE:
 176        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 177        break;
 178
 179    case GL_POINT:
 180        dw2 |= GEN6_SF_FRONT_POINT;
 181        break;
 182
 183    default:
 184        assert(0);
 185        break;
 186    }
 187
 188    switch (ctx->Polygon.BackMode) {
 189    case GL_FILL:
 190        dw2 |= GEN6_SF_BACK_SOLID;
 191        break;
 192
 193    case GL_LINE:
 194        dw2 |= GEN6_SF_BACK_WIREFRAME;
 195        break;
 196
 197    case GL_POINT:
 198        dw2 |= GEN6_SF_BACK_POINT;
 199        break;
 200
 201    default:
 202        assert(0);
 203        break;
 204    }
 205
 206    /* _NEW_SCISSOR */
 207    if (ctx->Scissor.Enabled)
 208       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 209
 210    /* _NEW_POLYGON */
 211    if (ctx->Polygon.CullFlag) {
 212       switch (ctx->Polygon.CullFaceMode) {
 213       case GL_FRONT:
 214          dw3 |= GEN6_SF_CULL_FRONT;
 215          break;
 216       case GL_BACK:
 217          dw3 |= GEN6_SF_CULL_BACK;
 218          break;
 219       case GL_FRONT_AND_BACK:
 220          dw3 |= GEN6_SF_CULL_BOTH;
 221          break;
 222       default:
 223          assert(0);
 224          break;
 225       }
 226    } else {
 227       dw3 |= GEN6_SF_CULL_NONE;
 228    }
 229
 230    /* _NEW_LINE */
 231    {
 232       uint32_t line_width_u3_7 = U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7);
 233       /* TODO: line width of 0 is not allowed when MSAA enabled */
 234       if (line_width_u3_7 == 0)
 235          line_width_u3_7 = 1;
 236       dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
 237    }
 238    if (ctx->Line.SmoothFlag) {
 239       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 240       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 241       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 242    }
 243    /* _NEW_MULTISAMPLE */
 244    if (multisampled_fbo && ctx->Multisample.Enabled)
 245       dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
 246
 247    /* _NEW_PROGRAM | _NEW_POINT */
 248    if (!(ctx->VertexProgram.PointSizeEnabled ||
 249          ctx->Point._Attenuated))
 250       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 251
 252    /* Clamp to ARB_point_parameters user limits */
 253    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 254
 255    /* Clamp to the hardware limits and convert to fixed point */
 256    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 257
 258    /*
 259     * Window coordinates in an FBO are inverted, which means point
 260     * sprite origin must be inverted, too.
 261     */
 262    if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
 263       point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
 264    } else {
 265       point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
 266    }
 267    dw1 |= point_sprite_origin;
 268
 269    /* _NEW_LIGHT */
 270    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 271       dw4 |=
 272          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 273          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 274          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 275    } else {
 276       dw4 |=
 277          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 278    }
 279
 280    /* Create the mapping from the FS inputs we produce to the VS outputs
 281     * they source from.
 282     */
 283    uint32_t max_source_attr = 0;
 284    for (; attr < VARYING_SLOT_MAX; attr++) {
 285       enum glsl_interp_qualifier interp_qualifier =
 286          brw->fragment_program->InterpQualifier[attr];
 287       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
 288
 289       if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
 290          continue;
 291
 292       /* _NEW_POINT */
 293       if (ctx->Point.PointSprite &&
 294           (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7) &&
 295           ctx->Point.CoordReplace[attr - VARYING_SLOT_TEX0]) {
 296          dw16 |= (1 << input_index);
 297       }
 298
 299       if (attr == VARYING_SLOT_PNTC)
 300          dw16 |= (1 << input_index);
 301
 302       /* flat shading */
 303       if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
 304           (shade_model_flat && is_gl_Color &&
 305            interp_qualifier == INTERP_QUALIFIER_NONE))
 306          dw17 |= (1 << input_index);
 307
 308       /* The hardware can only do the overrides on 16 overrides at a
 309        * time, and the other up to 16 have to be lined up so that the
 310        * input index = the output index.  We'll need to do some
 311        * tweaking to make sure that's the case.
 312        */
 313       assert(input_index < 16 || attr == input_index);
 314
 315       /* CACHE_NEW_VS_PROG | _NEW_LIGHT | _NEW_PROGRAM */
 316       attr_overrides[input_index++] =
 317          get_attr_override(&brw->vs.prog_data->vue_map,
 318                            urb_entry_read_offset, attr,
 319                            ctx->VertexProgram._TwoSideEnabled,
 320                            &max_source_attr);
 321    }
 322
 323    for (; input_index < VARYING_SLOT_MAX; input_index++)
 324       attr_overrides[input_index] = 0;
 325
 326    /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
 327     * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
 328     *
 329     * "This field should be set to the minimum length required to read the
 330     *  maximum source attribute.  The maximum source attribute is indicated
 331     *  by the maximum value of the enabled Attribute # Source Attribute if
 332     *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
 333     *  enable is not set.
 334     *  read_length = ceiling((max_source_attr + 1) / 2)
 335     *
 336     *  [errata] Corruption/Hang possible if length programmed larger than
 337     *  recommended"
 338     */
 339    uint32_t urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
 340       dw1 |= urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 341              urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
 342
 343    BEGIN_BATCH(20);
 344    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 345    OUT_BATCH(dw1);
 346    OUT_BATCH(dw2);
 347    OUT_BATCH(dw3);
 348    OUT_BATCH(dw4);
 349    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 350    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 351    OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
 352    for (i = 0; i < 8; i++) {
 353       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 354    }
 355    OUT_BATCH(dw16); /* point sprite texcoord bitmask */
 356    OUT_BATCH(dw17); /* constant interp bitmask */
 357    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 358    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 359    ADVANCE_BATCH();
 360 }
 361
 362 const struct brw_tracked_state gen6_sf_state = {
 363    .dirty = {
 364       .mesa  = (_NEW_LIGHT |
 365                 _NEW_PROGRAM |
 366                 _NEW_POLYGON |
 367                 _NEW_LINE |
 368                 _NEW_SCISSOR |
 369                 _NEW_BUFFERS |
 370                 _NEW_POINT |
 371                 _NEW_MULTISAMPLE),
 372       .brw   = (BRW_NEW_CONTEXT |
 373                 BRW_NEW_FRAGMENT_PROGRAM),
 374       .cache = CACHE_NEW_VS_PROG
 375    },
 376    .emit = upload_sf_state,
 377 };