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 "intel_batchbuffer.h"
  34
  35 /**
  36  * Determine the appropriate attribute override value to store into the
  37  * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
  38  * override value contains two pieces of information: the location of the
  39  * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
  40  * flag indicating whether to "swizzle" the attribute based on the direction
  41  * the triangle is facing.
  42  *
  43  * If an attribute is "swizzled", then the given VUE location is used for
  44  * front-facing triangles, and the VUE location that immediately follows is
  45  * used for back-facing triangles.  We use this to implement the mapping from
  46  * gl_FrontColor/gl_BackColor to gl_Color.
  47  *
  48  * urb_entry_read_offset is the offset into the VUE at which the SF unit is
  49  * being instructed to begin reading attribute data.  It can be set to a
  50  * nonzero value to prevent the SF unit from wasting time reading elements of
  51  * the VUE that are not needed by the fragment shader.  It is measured in
  52  * 256-bit increments.
  53  */
  54 uint32_t
  55 get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
  56                   int fs_attr, bool two_side_color)
  57 {
  58    int attr_override, slot;
  59    int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
  60    if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
  61       /* These attributes will be overwritten by the fragment shader's
  62        * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
  63        * them reference the first available attribute.
  64        */
  65       return 0;
  66    }
  67
  68    /* Find the VUE slot for this attribute. */
  69    slot = vue_map->vert_result_to_slot[vs_attr];
  70
  71    /* If there was only a back color written but not front, use back
  72     * as the color instead of undefined
  73     */
  74    if (slot == -1 && vs_attr == VERT_RESULT_COL0)
  75       slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
  76    if (slot == -1 && vs_attr == VERT_RESULT_COL1)
  77       slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
  78
  79    if (slot == -1) {
  80       /* This attribute does not exist in the VUE--that means that the vertex
  81        * shader did not write to it.  Behavior is undefined in this case, so
  82        * just reference the first available attribute.
  83        */
  84       return 0;
  85    }
  86
  87    /* Compute the location of the attribute relative to urb_entry_read_offset.
  88     * Each increment of urb_entry_read_offset represents a 256-bit value, so
  89     * it counts for two 128-bit VUE slots.
  90     */
  91    attr_override = slot - 2 * urb_entry_read_offset;
  92    assert (attr_override >= 0 && attr_override < 32);
  93
  94    /* If we are doing two-sided color, and the VUE slot following this one
  95     * represents a back-facing color, then we need to instruct the SF unit to
  96     * do back-facing swizzling.
  97     */
  98    if (two_side_color) {
  99       if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
 100           vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
 101          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 102       else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
 103                vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
 104          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 105    }
 106
 107    return attr_override;
 108 }
 109
 110 static void
 111 upload_sf_state(struct brw_context *brw)
 112 {
 113    struct intel_context *intel = &brw->intel;
 114    struct gl_context *ctx = &intel->ctx;
 115    struct brw_vue_map vue_map;
 116    uint32_t urb_entry_read_length;
 117    /* CACHE_NEW_VS_PROG */
 118    GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
 119    /* BRW_NEW_FRAGMENT_PROGRAM */
 120    uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
 121    /* _NEW_LIGHT */
 122    bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
 123    uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
 124    int i;
 125    /* _NEW_BUFFER */
 126    bool render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 127    int attr = 0, input_index = 0;
 128    int urb_entry_read_offset = 1;
 129    float point_size;
 130    uint16_t attr_overrides[FRAG_ATTRIB_MAX];
 131    bool userclip_active;
 132
 133    /* _NEW_TRANSFORM */
 134    userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
 135
 136    brw_compute_vue_map(&vue_map, intel, userclip_active, vs_outputs_written);
 137    urb_entry_read_length = (vue_map.num_slots + 1)/2 - urb_entry_read_offset;
 138    if (urb_entry_read_length == 0) {
 139       /* Setting the URB entry read length to 0 causes undefined behavior, so
 140        * if we have no URB data to read, set it to 1.
 141        */
 142       urb_entry_read_length = 1;
 143    }
 144
 145    dw1 =
 146       GEN6_SF_SWIZZLE_ENABLE |
 147       num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
 148       urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 149       urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
 150    dw2 = GEN6_SF_VIEWPORT_TRANSFORM_ENABLE |
 151       GEN6_SF_STATISTICS_ENABLE;
 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    dw3 |= U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7) <<
 232       GEN6_SF_LINE_WIDTH_SHIFT;
 233    if (ctx->Line.SmoothFlag) {
 234       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 235       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 236       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 237    }
 238
 239    /* _NEW_POINT */
 240    if (!(ctx->VertexProgram.PointSizeEnabled ||
 241          ctx->Point._Attenuated))
 242       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 243
 244    /* Clamp to ARB_point_parameters user limits */
 245    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 246
 247    /* Clamp to the hardware limits and convert to fixed point */
 248    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 249
 250    if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
 251       dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
 252
 253    /* _NEW_LIGHT */
 254    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 255       dw4 |=
 256          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 257          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 258          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 259    } else {
 260       dw4 |=
 261          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 262    }
 263
 264    /* Create the mapping from the FS inputs we produce to the VS outputs
 265     * they source from.
 266     */
 267    for (; attr < FRAG_ATTRIB_MAX; attr++) {
 268       enum glsl_interp_qualifier interp_qualifier =
 269          brw->fragment_program->InterpQualifier[attr];
 270       bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
 271
 272       if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
 273          continue;
 274
 275       /* _NEW_POINT */
 276       if (ctx->Point.PointSprite &&
 277           (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
 278           ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
 279          dw16 |= (1 << input_index);
 280       }
 281
 282       if (attr == FRAG_ATTRIB_PNTC)
 283          dw16 |= (1 << input_index);
 284
 285       /* flat shading */
 286       if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
 287           (shade_model_flat && is_gl_Color &&
 288            interp_qualifier == INTERP_QUALIFIER_NONE))
 289          dw17 |= (1 << input_index);
 290
 291       /* The hardware can only do the overrides on 16 overrides at a
 292        * time, and the other up to 16 have to be lined up so that the
 293        * input index = the output index.  We'll need to do some
 294        * tweaking to make sure that's the case.
 295        */
 296       assert(input_index < 16 || attr == input_index);
 297
 298       /* _NEW_LIGHT | _NEW_PROGRAM */
 299       attr_overrides[input_index++] =
 300          get_attr_override(&vue_map, urb_entry_read_offset, attr,
 301                            ctx->VertexProgram._TwoSideEnabled);
 302    }
 303
 304    for (; input_index < FRAG_ATTRIB_MAX; input_index++)
 305       attr_overrides[input_index] = 0;
 306
 307    BEGIN_BATCH(20);
 308    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 309    OUT_BATCH(dw1);
 310    OUT_BATCH(dw2);
 311    OUT_BATCH(dw3);
 312    OUT_BATCH(dw4);
 313    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 314    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 315    OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
 316    for (i = 0; i < 8; i++) {
 317       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 318    }
 319    OUT_BATCH(dw16); /* point sprite texcoord bitmask */
 320    OUT_BATCH(dw17); /* constant interp bitmask */
 321    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 322    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 323    ADVANCE_BATCH();
 324 }
 325
 326 const struct brw_tracked_state gen6_sf_state = {
 327    .dirty = {
 328       .mesa  = (_NEW_LIGHT |
 329                 _NEW_PROGRAM |
 330                 _NEW_POLYGON |
 331                 _NEW_LINE |
 332                 _NEW_SCISSOR |
 333                 _NEW_BUFFERS |
 334                 _NEW_POINT |
 335                 _NEW_TRANSFORM),
 336       .brw   = (BRW_NEW_CONTEXT |
 337                 BRW_NEW_FRAGMENT_PROGRAM),
 338       .cache = CACHE_NEW_VS_PROG
 339    },
 340    .emit = upload_sf_state,
 341 };