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_build_pm4.h"
  26
  27 /* For MSAA sample positions. */
  28 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
  29         ((((unsigned)(s0x) & 0xf) << 0)  | (((unsigned)(s0y) & 0xf) << 4)  | \
  30          (((unsigned)(s1x) & 0xf) << 8)  | (((unsigned)(s1y) & 0xf) << 12) | \
  31          (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) | \
  32          (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
  33
  34 /* For obtaining location coordinates from registers */
  35 #define SEXT4(x)                ((int)((x) | ((x) & 0x8 ? 0xfffffff0 : 0)))
  36 #define GET_SFIELD(reg, index)  SEXT4(((reg) >> ((index) * 4)) & 0xf)
  37 #define GET_SX(reg, index)      GET_SFIELD((reg)[(index) / 4], ((index) % 4) * 2)
  38 #define GET_SY(reg, index)      GET_SFIELD((reg)[(index) / 4], ((index) % 4) * 2 + 1)
  39
  40 /* 1x MSAA */
  41 static const uint32_t sample_locs_1x =
  42         FILL_SREG( 0, 0,   0, 0,   0, 0,   0, 0); /* S1, S2, S3 fields are not used by 1x */
  43
  44 /* 2x MSAA */
  45 static const uint32_t sample_locs_2x =
  46         FILL_SREG(4, 4, -4, -4, 0, 0, 0, 0); /* S2 & S3 fields are not used by 2x MSAA */
  47
  48 /* 4xMSAA
  49  * There are 4 locations: (-2, -6), (6, -2), (-6, 2), (2, 6). */
  50 static const uint32_t sample_locs_4x =
  51         FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6);
  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(-5,  5, -7, -1, 3,  7,  7, -7),
  57 };
  58 /* Cayman 16xMSAA */
  59 static const uint32_t sample_locs_16x[] = {
  60         FILL_SREG( 1,  1, -1, -3, -3,  2,  4, -1),
  61         FILL_SREG(-5, -2,  2,  5,  5,  3,  3, -5),
  62         FILL_SREG(-2,  6,  0, -7, -4, -6, -6,  4),
  63         FILL_SREG(-8,  0,  7, -4,  6,  7, -7, -8),
  64 };
  65
  66 static void si_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
  67                                    unsigned sample_index, float *out_value)
  68 {
  69         const uint32_t *sample_locs;
  70
  71         switch (sample_count) {
  72         case 1:
  73         default:
  74                 sample_locs = &sample_locs_1x;
  75                 break;
  76         case 2:
  77                 sample_locs = &sample_locs_2x;
  78                 break;
  79         case 4:
  80                 sample_locs = &sample_locs_4x;
  81                 break;
  82         case 8:
  83                 sample_locs = sample_locs_8x;
  84                 break;
  85         case 16:
  86                 sample_locs = sample_locs_16x;
  87                 break;
  88         }
  89
  90         out_value[0] = (GET_SX(sample_locs, sample_index) + 8) / 16.0f;
  91         out_value[1] = (GET_SY(sample_locs, sample_index) + 8) / 16.0f;
  92 }
  93
  94 void si_emit_sample_locations(struct radeon_winsys_cs *cs, int nr_samples)
  95 {
  96         switch (nr_samples) {
  97         default:
  98         case 1:
  99                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 0);
 100                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, 0);
 101                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, 0);
 102                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, 0);
 103                 break;
 104         case 2:
 105                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_2x);
 106                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_2x);
 107                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_2x);
 108                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_2x);
 109                 break;
 110         case 4:
 111                 radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_4x);
 112                 radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_4x);
 113                 radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_4x);
 114                 radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_4x);
 115                 break;
 116         case 8:
 117                 radeon_set_context_reg_seq(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 14);
 118                 radeon_emit(cs, sample_locs_8x[0]);
 119                 radeon_emit(cs, sample_locs_8x[1]);
 120                 radeon_emit(cs, 0);
 121                 radeon_emit(cs, 0);
 122                 radeon_emit(cs, sample_locs_8x[0]);
 123                 radeon_emit(cs, sample_locs_8x[1]);
 124                 radeon_emit(cs, 0);
 125                 radeon_emit(cs, 0);
 126                 radeon_emit(cs, sample_locs_8x[0]);
 127                 radeon_emit(cs, sample_locs_8x[1]);
 128                 radeon_emit(cs, 0);
 129                 radeon_emit(cs, 0);
 130                 radeon_emit(cs, sample_locs_8x[0]);
 131                 radeon_emit(cs, sample_locs_8x[1]);
 132                 break;
 133         case 16:
 134                 radeon_set_context_reg_seq(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 16);
 135                 radeon_emit(cs, sample_locs_16x[0]);
 136                 radeon_emit(cs, sample_locs_16x[1]);
 137                 radeon_emit(cs, sample_locs_16x[2]);
 138                 radeon_emit(cs, sample_locs_16x[3]);
 139                 radeon_emit(cs, sample_locs_16x[0]);
 140                 radeon_emit(cs, sample_locs_16x[1]);
 141                 radeon_emit(cs, sample_locs_16x[2]);
 142                 radeon_emit(cs, sample_locs_16x[3]);
 143                 radeon_emit(cs, sample_locs_16x[0]);
 144                 radeon_emit(cs, sample_locs_16x[1]);
 145                 radeon_emit(cs, sample_locs_16x[2]);
 146                 radeon_emit(cs, sample_locs_16x[3]);
 147                 radeon_emit(cs, sample_locs_16x[0]);
 148                 radeon_emit(cs, sample_locs_16x[1]);
 149                 radeon_emit(cs, sample_locs_16x[2]);
 150                 radeon_emit(cs, sample_locs_16x[3]);
 151                 break;
 152         }
 153 }
 154
 155 void si_init_msaa_functions(struct si_context *sctx)
 156 {
 157         int i;
 158
 159         sctx->b.get_sample_position = si_get_sample_position;
 160
 161         si_get_sample_position(&sctx->b, 1, 0, sctx->sample_locations_1x[0]);
 162
 163         for (i = 0; i < 2; i++)
 164                 si_get_sample_position(&sctx->b, 2, i, sctx->sample_locations_2x[i]);
 165         for (i = 0; i < 4; i++)
 166                 si_get_sample_position(&sctx->b, 4, i, sctx->sample_locations_4x[i]);
 167         for (i = 0; i < 8; i++)
 168                 si_get_sample_position(&sctx->b, 8, i, sctx->sample_locations_8x[i]);
 169         for (i = 0; i < 16; i++)
 170                 si_get_sample_position(&sctx->b, 16, i, sctx->sample_locations_16x[i]);
 171 }