src/gallium/drivers/radeonsi/si_state_msaa.c

   1 /*
   2  * Copyright 2014 Advanced Micro Devices, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  */
  24
  25 #include "si_pipe.h"
  26 #include "sid.h"
  27 #include "radeon/r600_cs.h"
  28
  29 /* For MSAA sample positions. */
  30 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
  31         (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) |                  \
  32         (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) |         \
  33         (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) |        \
  34          (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
  35
  36 /* 2xMSAA
  37  * There are two locations (4, 4), (-4, -4). */
  38 static const uint32_t sample_locs_2x[4] = {
  39         FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
  40         FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
  41         FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
  42         FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
  43 };
  44 /* 4xMSAA
  45  * There are 4 locations: (-2, -6), (6, -2), (-6, 2), (2, 6). */
  46 static const uint32_t sample_locs_4x[4] = {
  47         FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
  48         FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
  49         FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
  50         FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
  51 };
  52
  53 /* Cayman 8xMSAA */
  54 static const uint32_t sample_locs_8x[] = {
  55         FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
  56         FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
  57         FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
  58         FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
  59         FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
  60         FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
  61         FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
  62         FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
  63 };
  64 /* Cayman 16xMSAA */
  65 static const uint32_t sample_locs_16x[] = {
  66         FILL_SREG( 1,  1, -1, -3, -3,  2,  4, -1),
  67         FILL_SREG( 1,  1, -1, -3, -3,  2,  4, -1),
  68         FILL_SREG( 1,  1, -1, -3, -3,  2,  4, -1),
  69         FILL_SREG( 1,  1, -1, -3, -3,  2,  4, -1),
  70         FILL_SREG(-5, -2,  2,  5,  5,  3,  3, -5),
  71         FILL_SREG(-5, -2,  2,  5,  5,  3,  3, -5),
  72         FILL_SREG(-5, -2,  2,  5,  5,  3,  3, -5),
  73         FILL_SREG(-5, -2,  2,  5,  5,  3,  3, -5),
  74         FILL_SREG(-2,  6,  0, -7, -4, -6, -6,  4),
  75         FILL_SREG(-2,  6,  0, -7, -4, -6, -6,  4),
  76         FILL_SREG(-2,  6,  0, -7, -4, -6, -6,  4),
  77         FILL_SREG(-2,  6,  0, -7, -4, -6, -6,  4),
  78         FILL_SREG(-8,  0,  7, -4,  6,  7, -7, -8),
  79         FILL_SREG(-8,  0,  7, -4,  6,  7, -7, -8),
  80         FILL_SREG(-8,  0,  7, -4,  6,  7, -7, -8),
  81         FILL_SREG(-8,  0,  7, -4,  6,  7, -7, -8),
  82 };
  83
  84 static void si_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
  85                                    unsigned sample_index, float *out_value)
  86 {
  87         int offset, index;
  88         struct {
  89                 int idx:4;
  90         } val;
  91
  92         switch (sample_count) {
  93         case 1:
  94         default:
  95                 out_value[0] = out_value[1] = 0.5;
  96                 break;
  97         case 2:
  98                 offset = 4 * (sample_index * 2);
  99                 val.idx = (sample_locs_2x[0] >> offset) & 0xf;
 100                 out_value[0] = (float)(val.idx + 8) / 16.0f;
 101                 val.idx = (sample_locs_2x[0] >> (offset + 4)) & 0xf;
 102                 out_value[1] = (float)(val.idx + 8) / 16.0f;
 103                 break;
 104         case 4:
 105                 offset = 4 * (sample_index * 2);
 106                 val.idx = (sample_locs_4x[0] >> offset) & 0xf;
 107                 out_value[0] = (float)(val.idx + 8) / 16.0f;
 108                 val.idx = (sample_locs_4x[0] >> (offset + 4)) & 0xf;
 109                 out_value[1] = (float)(val.idx + 8) / 16.0f;
 110                 break;
 111         case 8:
 112                 offset = 4 * (sample_index % 4 * 2);
 113                 index = (sample_index / 4) * 4;
 114                 val.idx = (sample_locs_8x[index] >> offset) & 0xf;
 115                 out_value[0] = (float)(val.idx + 8) / 16.0f;
 116                 val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf;
 117                 out_value[1] = (float)(val.idx + 8) / 16.0f;
 118                 break;
 119         case 16:
 120                 offset = 4 * (sample_index % 4 * 2);
 121                 index = (sample_index / 4) * 4;
 122                 val.idx = (sample_locs_16x[index] >> offset) & 0xf;
 123                 out_value[0] = (float)(val.idx + 8) / 16.0f;
 124                 val.idx = (sample_locs_16x[index] >> (offset + 4)) & 0xf;
 125                 out_value[1] = (float)(val.idx + 8) / 16.0f;
 126                 break;
 127         }
 128 }
 129
 130 void si_emit_sample_locations(struct radeon_winsys_cs *cs, int nr_samples)
 131 {
 132         switch (nr_samples) {
 133         default:
 134         case 1:
 135                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 0);
 136                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, 0);
 137                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, 0);
 138                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, 0);
 139                 break;
 140         case 2:
 141                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_2x[0]);
 142                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_2x[1]);
 143                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_2x[2]);
 144                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_2x[3]);
 145                 break;
 146         case 4:
 147                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_4x[0]);
 148                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_4x[1]);
 149                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_4x[2]);
 150                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_4x[3]);
 151                 break;
 152         case 8:
 153                 radeon_set_context_reg_seq(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 14);
 154                 radeon_emit(cs, sample_locs_8x[0]);
 155                 radeon_emit(cs, sample_locs_8x[4]);
 156                 radeon_emit(cs, 0);
 157                 radeon_emit(cs, 0);
 158                 radeon_emit(cs, sample_locs_8x[1]);
 159                 radeon_emit(cs, sample_locs_8x[5]);
 160                 radeon_emit(cs, 0);
 161                 radeon_emit(cs, 0);
 162                 radeon_emit(cs, sample_locs_8x[2]);
 163                 radeon_emit(cs, sample_locs_8x[6]);
 164                 radeon_emit(cs, 0);
 165                 radeon_emit(cs, 0);
 166                 radeon_emit(cs, sample_locs_8x[3]);
 167                 radeon_emit(cs, sample_locs_8x[7]);
 168                 break;
 169         case 16:
 170                 radeon_set_context_reg_seq(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 16);
 171                 radeon_emit(cs, sample_locs_16x[0]);
 172                 radeon_emit(cs, sample_locs_16x[4]);
 173                 radeon_emit(cs, sample_locs_16x[8]);
 174                 radeon_emit(cs, sample_locs_16x[12]);
 175                 radeon_emit(cs, sample_locs_16x[1]);
 176                 radeon_emit(cs, sample_locs_16x[5]);
 177                 radeon_emit(cs, sample_locs_16x[9]);
 178                 radeon_emit(cs, sample_locs_16x[13]);
 179                 radeon_emit(cs, sample_locs_16x[2]);
 180                 radeon_emit(cs, sample_locs_16x[6]);
 181                 radeon_emit(cs, sample_locs_16x[10]);
 182                 radeon_emit(cs, sample_locs_16x[14]);
 183                 radeon_emit(cs, sample_locs_16x[3]);
 184                 radeon_emit(cs, sample_locs_16x[7]);
 185                 radeon_emit(cs, sample_locs_16x[11]);
 186                 radeon_emit(cs, sample_locs_16x[15]);
 187                 break;
 188         }
 189 }
 190
 191 void si_init_msaa_functions(struct si_context *sctx)
 192 {
 193         int i;
 194
 195         sctx->b.b.get_sample_position = si_get_sample_position;
 196
 197         si_get_sample_position(&sctx->b.b, 1, 0, sctx->sample_locations_1x[0]);
 198
 199         for (i = 0; i < 2; i++)
 200                 si_get_sample_position(&sctx->b.b, 2, i, sctx->sample_locations_2x[i]);
 201         for (i = 0; i < 4; i++)
 202                 si_get_sample_position(&sctx->b.b, 4, i, sctx->sample_locations_4x[i]);
 203         for (i = 0; i < 8; i++)
 204                 si_get_sample_position(&sctx->b.b, 8, i, sctx->sample_locations_8x[i]);
 205         for (i = 0; i < 16; i++)
 206                 si_get_sample_position(&sctx->b.b, 16, i, sctx->sample_locations_16x[i]);
 207 }