src/mesa/drivers/dri/i965/gen6_multisample_state.c

   1 /*
   2  * Copyright © 2012 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "intel_batchbuffer.h"
  25
  26 #include "brw_context.h"
  27 #include "brw_defines.h"
  28
  29 /* Sample positions:
  30  *   2 6 a e
  31  * 2   0
  32  * 6       1
  33  * a 2
  34  * e     3
  35  */
  36 static uint32_t
  37 sample_positions_4x[] = { 0xae2ae662 };
  38 /* Sample positions are based on a solution to the "8 queens" puzzle.
  39  * Rationale: in a solution to the 8 queens puzzle, no two queens share
  40  * a row, column, or diagonal.  This is a desirable property for samples
  41  * in a multisampling pattern, because it ensures that the samples are
  42  * relatively uniformly distributed through the pixel.
  43  *
  44  * There are several solutions to the 8 queens puzzle (see
  45  * http://en.wikipedia.org/wiki/Eight_queens_puzzle).  This solution was
  46  * chosen because it has a queen close to the center; this should
  47  * improve the accuracy of centroid interpolation, since the hardware
  48  * implements centroid interpolation by choosing the centermost sample
  49  * that overlaps with the primitive being drawn.
  50  *
  51  * Note: from the Ivy Bridge PRM, Vol2 Part1 p304 (3DSTATE_MULTISAMPLE:
  52  * Programming Notes):
  53  *
  54  *     "When programming the sample offsets (for NUMSAMPLES_4 or _8 and
  55  *     MSRASTMODE_xxx_PATTERN), the order of the samples 0 to 3 (or 7
  56  *     for 8X) must have monotonically increasing distance from the
  57  *     pixel center. This is required to get the correct centroid
  58  *     computation in the device."
  59  *
  60  * Sample positions:
  61  *   1 3 5 7 9 b d f
  62  * 1     5
  63  * 3           2
  64  * 5               6
  65  * 7 4
  66  * 9       0
  67  * b             3
  68  * d         1
  69  * f   7
  70  */
  71 static uint32_t
  72 sample_positions_8x[] = { 0xdbb39d79, 0x3ff55117 };
  73
  74
  75 void
  76 gen6_get_sample_position(struct gl_context *ctx,
  77                          struct gl_framebuffer *fb,
  78                          GLuint index, GLfloat *result)
  79 {
  80    switch (fb->Visual.samples) {
  81    case 1:
  82       result[0] = result[1] = 0.5f;
  83       break;
  84    case 4: {
  85       uint8_t val = (uint8_t)(sample_positions_4x[0] >> (8*index));
  86       result[0] = ((val >> 4) & 0xf) / 16.0f;
  87       result[1] = (val & 0xf) / 16.0f;
  88       break;
  89    }
  90    case 8: {
  91       uint8_t val = (uint8_t)(sample_positions_8x[index>>2] >> (8*(index & 3)));
  92       result[0] = ((val >> 4) & 0xf) / 16.0f;
  93       result[1] = (val & 0xf) / 16.0f;
  94       break;
  95    }
  96    default:
  97       assert(!"Not implemented");
  98    }
  99 }
 100
 101 /**
 102  * 3DSTATE_MULTISAMPLE
 103  */
 104 void
 105 gen6_emit_3dstate_multisample(struct brw_context *brw,
 106                               unsigned num_samples)
 107 {
 108    struct intel_context *intel = &brw->intel;
 109
 110    uint32_t number_of_multisamples = 0;
 111    uint32_t sample_positions_3210 = 0;
 112    uint32_t sample_positions_7654 = 0;
 113
 114    switch (num_samples) {
 115    case 0:
 116    case 1:
 117       number_of_multisamples = MS_NUMSAMPLES_1;
 118       break;
 119    case 4:
 120       number_of_multisamples = MS_NUMSAMPLES_4;
 121       sample_positions_3210 = sample_positions_4x[0];
 122       break;
 123    case 8:
 124       number_of_multisamples = MS_NUMSAMPLES_8;
 125       sample_positions_3210 = sample_positions_8x[0];
 126       sample_positions_7654 = sample_positions_8x[1];
 127       break;
 128    default:
 129       assert(!"Unrecognized num_samples in gen6_emit_3dstate_multisample");
 130       break;
 131    }
 132
 133    int len = intel->gen >= 7 ? 4 : 3;
 134    BEGIN_BATCH(len);
 135    OUT_BATCH(_3DSTATE_MULTISAMPLE << 16 | (len - 2));
 136    OUT_BATCH(MS_PIXEL_LOCATION_CENTER | number_of_multisamples);
 137    OUT_BATCH(sample_positions_3210);
 138    if (intel->gen >= 7)
 139       OUT_BATCH(sample_positions_7654);
 140    ADVANCE_BATCH();
 141 }
 142
 143
 144 /**
 145  * 3DSTATE_SAMPLE_MASK
 146  */
 147 void
 148 gen6_emit_3dstate_sample_mask(struct brw_context *brw,
 149                               unsigned num_samples, float coverage,
 150                               bool coverage_invert, unsigned sample_mask)
 151 {
 152    struct intel_context *intel = &brw->intel;
 153
 154    BEGIN_BATCH(2);
 155    OUT_BATCH(_3DSTATE_SAMPLE_MASK << 16 | (2 - 2));
 156    if (num_samples > 1) {
 157       int coverage_int = (int) (num_samples * coverage + 0.5);
 158       uint32_t coverage_bits = (1 << coverage_int) - 1;
 159       if (coverage_invert)
 160          coverage_bits ^= (1 << num_samples) - 1;
 161       OUT_BATCH(coverage_bits & sample_mask);
 162    } else {
 163       OUT_BATCH(1);
 164    }
 165    ADVANCE_BATCH();
 166 }
 167
 168
 169 static void upload_multisample_state(struct brw_context *brw)
 170 {
 171    struct intel_context *intel = &brw->intel;
 172    struct gl_context *ctx = &intel->ctx;
 173    float coverage = 1.0;
 174    float coverage_invert = false;
 175    unsigned sample_mask = ~0u;
 176
 177    /* _NEW_BUFFERS */
 178    unsigned num_samples = ctx->DrawBuffer->Visual.samples;
 179
 180    /* _NEW_MULTISAMPLE */
 181    if (ctx->Multisample._Enabled) {
 182       if (ctx->Multisample.SampleCoverage) {
 183          coverage = ctx->Multisample.SampleCoverageValue;
 184          coverage_invert = ctx->Multisample.SampleCoverageInvert;
 185       }
 186       if (ctx->Multisample.SampleMask) {
 187          sample_mask = ctx->Multisample.SampleMaskValue;
 188       }
 189    }
 190
 191    /* 3DSTATE_MULTISAMPLE is nonpipelined. */
 192    intel_emit_post_sync_nonzero_flush(intel);
 193
 194    gen6_emit_3dstate_multisample(brw, num_samples);
 195    gen6_emit_3dstate_sample_mask(brw, num_samples, coverage,
 196          coverage_invert, sample_mask);
 197 }
 198
 199
 200 const struct brw_tracked_state gen6_multisample_state = {
 201    .dirty = {
 202       .mesa = _NEW_BUFFERS |
 203               _NEW_MULTISAMPLE,
 204       .brw = BRW_NEW_CONTEXT,
 205       .cache = 0
 206    },
 207    .emit = upload_multisample_state
 208 };