1 /*
2 * Copyright 2014 Advanced Micro Devices, Inc.
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors: Marek Olšák <maraeo@gmail.com>
24 *
25 */
26
27 #include "r600_cs.h"
28
29 /* 2xMSAA
30 * There are two locations (4, 4), (-4, -4). */
31 const uint32_t eg_sample_locs_2x[4] = {
32 FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
33 FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
34 FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
35 FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
36 };
37 const unsigned eg_max_dist_2x = 4;
38 /* 4xMSAA
39 * There are 4 locations: (-2, 6), (6, -2), (-6, 2), (2, 6). */
40 const uint32_t eg_sample_locs_4x[4] = {
41 FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
42 FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
43 FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
44 FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
45 };
46 const unsigned eg_max_dist_4x = 6;
47
48 /* Cayman 8xMSAA */
49 static const uint32_t cm_sample_locs_8x[] = {
50 FILL_SREG( 1, -3, -1, 3, 5, 1, -3, -5),
51 FILL_SREG( 1, -3, -1, 3, 5, 1, -3, -5),
52 FILL_SREG( 1, -3, -1, 3, 5, 1, -3, -5),
53 FILL_SREG( 1, -3, -1, 3, 5, 1, -3, -5),
54 FILL_SREG(-5, 5, -7, -1, 3, 7, 7, -7),
55 FILL_SREG(-5, 5, -7, -1, 3, 7, 7, -7),
56 FILL_SREG(-5, 5, -7, -1, 3, 7, 7, -7),
57 FILL_SREG(-5, 5, -7, -1, 3, 7, 7, -7),
58 };
59 static const unsigned cm_max_dist_8x = 8;
60 /* Cayman 16xMSAA */
61 static const uint32_t cm_sample_locs_16x[] = {
62 FILL_SREG( 1, 1, -1, -3, -3, 2, 4, -1),
63 FILL_SREG( 1, 1, -1, -3, -3, 2, 4, -1),
64 FILL_SREG( 1, 1, -1, -3, -3, 2, 4, -1),
65 FILL_SREG( 1, 1, -1, -3, -3, 2, 4, -1),
66 FILL_SREG(-5, -2, 2, 5, 5, 3, 3, -5),
67 FILL_SREG(-5, -2, 2, 5, 5, 3, 3, -5),
68 FILL_SREG(-5, -2, 2, 5, 5, 3, 3, -5),
69 FILL_SREG(-5, -2, 2, 5, 5, 3, 3, -5),
70 FILL_SREG(-2, 6, 0, -7, -4, -6, -6, 4),
71 FILL_SREG(-2, 6, 0, -7, -4, -6, -6, 4),
72 FILL_SREG(-2, 6, 0, -7, -4, -6, -6, 4),
73 FILL_SREG(-2, 6, 0, -7, -4, -6, -6, 4),
74 FILL_SREG(-8, 0, 7, -4, 6, 7, -7, -8),
75 FILL_SREG(-8, 0, 7, -4, 6, 7, -7, -8),
76 FILL_SREG(-8, 0, 7, -4, 6, 7, -7, -8),
77 FILL_SREG(-8, 0, 7, -4, 6, 7, -7, -8),
78 };
79 static const unsigned cm_max_dist_16x = 8;
80
cayman_get_sample_position(struct pipe_context * ctx,unsigned sample_count,unsigned sample_index,float * out_value)81 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
82 unsigned sample_index, float *out_value)
83 {
84 int offset, index;
85 struct {
86 int idx:4;
87 } val;
88 switch (sample_count) {
89 case 1:
90 default:
91 out_value[0] = out_value[1] = 0.5;
92 break;
93 case 2:
94 offset = 4 * (sample_index * 2);
95 val.idx = (eg_sample_locs_2x[0] >> offset) & 0xf;
96 out_value[0] = (float)(val.idx + 8) / 16.0f;
97 val.idx = (eg_sample_locs_2x[0] >> (offset + 4)) & 0xf;
98 out_value[1] = (float)(val.idx + 8) / 16.0f;
99 break;
100 case 4:
101 offset = 4 * (sample_index * 2);
102 val.idx = (eg_sample_locs_4x[0] >> offset) & 0xf;
103 out_value[0] = (float)(val.idx + 8) / 16.0f;
104 val.idx = (eg_sample_locs_4x[0] >> (offset + 4)) & 0xf;
105 out_value[1] = (float)(val.idx + 8) / 16.0f;
106 break;
107 case 8:
108 offset = 4 * (sample_index % 4 * 2);
109 index = (sample_index / 4) * 4;
110 val.idx = (cm_sample_locs_8x[index] >> offset) & 0xf;
111 out_value[0] = (float)(val.idx + 8) / 16.0f;
112 val.idx = (cm_sample_locs_8x[index] >> (offset + 4)) & 0xf;
113 out_value[1] = (float)(val.idx + 8) / 16.0f;
114 break;
115 case 16:
116 offset = 4 * (sample_index % 4 * 2);
117 index = (sample_index / 4) * 4;
118 val.idx = (cm_sample_locs_16x[index] >> offset) & 0xf;
119 out_value[0] = (float)(val.idx + 8) / 16.0f;
120 val.idx = (cm_sample_locs_16x[index] >> (offset + 4)) & 0xf;
121 out_value[1] = (float)(val.idx + 8) / 16.0f;
122 break;
123 }
124 }
125
cayman_init_msaa(struct pipe_context * ctx)126 void cayman_init_msaa(struct pipe_context *ctx)
127 {
128 struct r600_common_context *rctx = (struct r600_common_context*)ctx;
129 int i;
130
131 cayman_get_sample_position(ctx, 1, 0, rctx->sample_locations_1x[0]);
132
133 for (i = 0; i < 2; i++)
134 cayman_get_sample_position(ctx, 2, i, rctx->sample_locations_2x[i]);
135 for (i = 0; i < 4; i++)
136 cayman_get_sample_position(ctx, 4, i, rctx->sample_locations_4x[i]);
137 for (i = 0; i < 8; i++)
138 cayman_get_sample_position(ctx, 8, i, rctx->sample_locations_8x[i]);
139 for (i = 0; i < 16; i++)
140 cayman_get_sample_position(ctx, 16, i, rctx->sample_locations_16x[i]);
141 }
142
cayman_emit_msaa_sample_locs(struct radeon_winsys_cs * cs,int nr_samples)143 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples)
144 {
145 switch (nr_samples) {
146 default:
147 case 1:
148 radeon_set_context_reg(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 0);
149 radeon_set_context_reg(cs, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, 0);
150 radeon_set_context_reg(cs, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, 0);
151 radeon_set_context_reg(cs, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, 0);
152 break;
153 case 2:
154 radeon_set_context_reg(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, eg_sample_locs_2x[0]);
155 radeon_set_context_reg(cs, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, eg_sample_locs_2x[1]);
156 radeon_set_context_reg(cs, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, eg_sample_locs_2x[2]);
157 radeon_set_context_reg(cs, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, eg_sample_locs_2x[3]);
158 break;
159 case 4:
160 radeon_set_context_reg(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, eg_sample_locs_4x[0]);
161 radeon_set_context_reg(cs, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, eg_sample_locs_4x[1]);
162 radeon_set_context_reg(cs, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, eg_sample_locs_4x[2]);
163 radeon_set_context_reg(cs, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, eg_sample_locs_4x[3]);
164 break;
165 case 8:
166 radeon_set_context_reg_seq(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 14);
167 radeon_emit(cs, cm_sample_locs_8x[0]);
168 radeon_emit(cs, cm_sample_locs_8x[4]);
169 radeon_emit(cs, 0);
170 radeon_emit(cs, 0);
171 radeon_emit(cs, cm_sample_locs_8x[1]);
172 radeon_emit(cs, cm_sample_locs_8x[5]);
173 radeon_emit(cs, 0);
174 radeon_emit(cs, 0);
175 radeon_emit(cs, cm_sample_locs_8x[2]);
176 radeon_emit(cs, cm_sample_locs_8x[6]);
177 radeon_emit(cs, 0);
178 radeon_emit(cs, 0);
179 radeon_emit(cs, cm_sample_locs_8x[3]);
180 radeon_emit(cs, cm_sample_locs_8x[7]);
181 break;
182 case 16:
183 radeon_set_context_reg_seq(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 16);
184 radeon_emit(cs, cm_sample_locs_16x[0]);
185 radeon_emit(cs, cm_sample_locs_16x[4]);
186 radeon_emit(cs, cm_sample_locs_16x[8]);
187 radeon_emit(cs, cm_sample_locs_16x[12]);
188 radeon_emit(cs, cm_sample_locs_16x[1]);
189 radeon_emit(cs, cm_sample_locs_16x[5]);
190 radeon_emit(cs, cm_sample_locs_16x[9]);
191 radeon_emit(cs, cm_sample_locs_16x[13]);
192 radeon_emit(cs, cm_sample_locs_16x[2]);
193 radeon_emit(cs, cm_sample_locs_16x[6]);
194 radeon_emit(cs, cm_sample_locs_16x[10]);
195 radeon_emit(cs, cm_sample_locs_16x[14]);
196 radeon_emit(cs, cm_sample_locs_16x[3]);
197 radeon_emit(cs, cm_sample_locs_16x[7]);
198 radeon_emit(cs, cm_sample_locs_16x[11]);
199 radeon_emit(cs, cm_sample_locs_16x[15]);
200 break;
201 }
202 }
203
cayman_emit_msaa_config(struct radeon_winsys_cs * cs,int nr_samples,int ps_iter_samples,int overrast_samples,unsigned sc_mode_cntl_1)204 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
205 int ps_iter_samples, int overrast_samples,
206 unsigned sc_mode_cntl_1)
207 {
208 int setup_samples = nr_samples > 1 ? nr_samples :
209 overrast_samples > 1 ? overrast_samples : 0;
210 /* Required by OpenGL line rasterization.
211 *
212 * TODO: We should also enable perpendicular endcaps for AA lines,
213 * but that requires implementing line stippling in the pixel
214 * shader. SC can only do line stippling with axis-aligned
215 * endcaps.
216 */
217 unsigned sc_line_cntl = S_028BDC_DX10_DIAMOND_TEST_ENA(1);
218
219 if (setup_samples > 1) {
220 /* indexed by log2(nr_samples) */
221 unsigned max_dist[] = {
222 0,
223 eg_max_dist_2x,
224 eg_max_dist_4x,
225 cm_max_dist_8x,
226 cm_max_dist_16x
227 };
228 unsigned log_samples = util_logbase2(setup_samples);
229 unsigned log_ps_iter_samples =
230 util_logbase2(util_next_power_of_two(ps_iter_samples));
231
232 radeon_set_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
233 radeon_emit(cs, sc_line_cntl |
234 S_028BDC_EXPAND_LINE_WIDTH(1)); /* CM_R_028BDC_PA_SC_LINE_CNTL */
235 radeon_emit(cs, S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
236 S_028BE0_MAX_SAMPLE_DIST(max_dist[log_samples]) |
237 S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples)); /* CM_R_028BE0_PA_SC_AA_CONFIG */
238
239 if (nr_samples > 1) {
240 radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
241 S_028804_MAX_ANCHOR_SAMPLES(log_samples) |
242 S_028804_PS_ITER_SAMPLES(log_ps_iter_samples) |
243 S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
244 S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples) |
245 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
246 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
247 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
248 EG_S_028A4C_PS_ITER_SAMPLE(ps_iter_samples > 1) |
249 sc_mode_cntl_1);
250 } else if (overrast_samples > 1) {
251 radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
252 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
253 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1) |
254 S_028804_OVERRASTERIZATION_AMOUNT(log_samples));
255 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
256 sc_mode_cntl_1);
257 }
258 } else {
259 radeon_set_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
260 radeon_emit(cs, sc_line_cntl); /* CM_R_028BDC_PA_SC_LINE_CNTL */
261 radeon_emit(cs, 0); /* CM_R_028BE0_PA_SC_AA_CONFIG */
262
263 radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
264 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
265 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
266 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
267 sc_mode_cntl_1);
268 }
269 }
270