i965/blorp: Do gen6 stencil offsets up-front
[mesa.git] / src / mesa / drivers / dri / i965 / gen7_wm_state.c
1 /*
2 * Copyright © 2011 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 <stdbool.h>
25 #include "brw_context.h"
26 #include "brw_state.h"
27 #include "brw_defines.h"
28 #include "brw_util.h"
29 #include "brw_wm.h"
30 #include "program/program.h"
31 #include "program/prog_parameter.h"
32 #include "program/prog_statevars.h"
33 #include "main/framebuffer.h"
34 #include "intel_batchbuffer.h"
35
36 static void
37 upload_wm_state(struct brw_context *brw)
38 {
39 struct gl_context *ctx = &brw->ctx;
40 /* BRW_NEW_FS_PROG_DATA */
41 const struct brw_wm_prog_data *prog_data = brw->wm.prog_data;
42 bool writes_depth = prog_data->computed_depth_mode != BRW_PSCDEPTH_OFF;
43 uint32_t dw1, dw2;
44
45 /* _NEW_BUFFERS */
46 const bool multisampled_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
47
48 dw1 = dw2 = 0;
49 dw1 |= GEN7_WM_STATISTICS_ENABLE;
50 dw1 |= GEN7_WM_LINE_AA_WIDTH_1_0;
51 dw1 |= GEN7_WM_LINE_END_CAP_AA_WIDTH_0_5;
52
53 /* _NEW_LINE */
54 if (ctx->Line.StippleFlag)
55 dw1 |= GEN7_WM_LINE_STIPPLE_ENABLE;
56
57 /* _NEW_POLYGON */
58 if (ctx->Polygon.StippleFlag)
59 dw1 |= GEN7_WM_POLYGON_STIPPLE_ENABLE;
60
61 if (prog_data->uses_src_depth)
62 dw1 |= GEN7_WM_USES_SOURCE_DEPTH;
63
64 if (prog_data->uses_src_w)
65 dw1 |= GEN7_WM_USES_SOURCE_W;
66
67 dw1 |= prog_data->computed_depth_mode << GEN7_WM_COMPUTED_DEPTH_MODE_SHIFT;
68 dw1 |= prog_data->barycentric_interp_modes <<
69 GEN7_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT;
70
71 /* _NEW_COLOR, _NEW_MULTISAMPLE */
72 /* Enable if the pixel shader kernel generates and outputs oMask.
73 */
74 if (prog_data->uses_kill || ctx->Color.AlphaEnabled ||
75 ctx->Multisample.SampleAlphaToCoverage ||
76 prog_data->uses_omask) {
77 dw1 |= GEN7_WM_KILL_ENABLE;
78 }
79
80 /* _NEW_BUFFERS | _NEW_COLOR */
81 if (brw_color_buffer_write_enabled(brw) || writes_depth ||
82 prog_data->has_side_effects || dw1 & GEN7_WM_KILL_ENABLE) {
83 dw1 |= GEN7_WM_DISPATCH_ENABLE;
84 }
85 if (multisampled_fbo) {
86 /* _NEW_MULTISAMPLE */
87 if (ctx->Multisample.Enabled)
88 dw1 |= GEN7_WM_MSRAST_ON_PATTERN;
89 else
90 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
91
92 if (prog_data->persample_dispatch)
93 dw2 |= GEN7_WM_MSDISPMODE_PERSAMPLE;
94 else
95 dw2 |= GEN7_WM_MSDISPMODE_PERPIXEL;
96 } else {
97 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
98 dw2 |= GEN7_WM_MSDISPMODE_PERSAMPLE;
99 }
100
101 if (prog_data->uses_sample_mask) {
102 dw1 |= GEN7_WM_USES_INPUT_COVERAGE_MASK;
103 }
104
105 /* BRW_NEW_FS_PROG_DATA */
106 if (prog_data->early_fragment_tests)
107 dw1 |= GEN7_WM_EARLY_DS_CONTROL_PREPS;
108 else if (prog_data->has_side_effects)
109 dw1 |= GEN7_WM_EARLY_DS_CONTROL_PSEXEC;
110
111 /* The "UAV access enable" bits are unnecessary on HSW because they only
112 * seem to have an effect on the HW-assisted coherency mechanism which we
113 * don't need, and the rasterization-related UAV_ONLY flag and the
114 * DISPATCH_ENABLE bit can be set independently from it.
115 * C.f. gen8_upload_ps_extra().
116 *
117 * BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_FS_PROG_DATA | _NEW_BUFFERS | _NEW_COLOR
118 */
119 if (brw->is_haswell &&
120 !(brw_color_buffer_write_enabled(brw) || writes_depth) &&
121 prog_data->has_side_effects)
122 dw2 |= HSW_WM_UAV_ONLY;
123
124 BEGIN_BATCH(3);
125 OUT_BATCH(_3DSTATE_WM << 16 | (3 - 2));
126 OUT_BATCH(dw1);
127 OUT_BATCH(dw2);
128 ADVANCE_BATCH();
129 }
130
131 const struct brw_tracked_state gen7_wm_state = {
132 .dirty = {
133 .mesa = _NEW_BUFFERS |
134 _NEW_COLOR |
135 _NEW_LINE |
136 _NEW_MULTISAMPLE |
137 _NEW_POLYGON,
138 .brw = BRW_NEW_BATCH |
139 BRW_NEW_BLORP |
140 BRW_NEW_FS_PROG_DATA,
141 },
142 .emit = upload_wm_state,
143 };
144
145 static void
146 gen7_upload_ps_state(struct brw_context *brw,
147 const struct brw_stage_state *stage_state,
148 const struct brw_wm_prog_data *prog_data,
149 bool enable_dual_src_blend, unsigned sample_mask,
150 unsigned fast_clear_op)
151 {
152 uint32_t dw2, dw4, dw5, ksp0, ksp2;
153 const int max_threads_shift = brw->is_haswell ?
154 HSW_PS_MAX_THREADS_SHIFT : IVB_PS_MAX_THREADS_SHIFT;
155
156 dw2 = dw4 = dw5 = ksp2 = 0;
157
158 const unsigned sampler_count =
159 DIV_ROUND_UP(CLAMP(stage_state->sampler_count, 0, 16), 4);
160 dw2 |= SET_FIELD(sampler_count, GEN7_PS_SAMPLER_COUNT);
161
162 dw2 |= ((prog_data->base.binding_table.size_bytes / 4) <<
163 GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT);
164
165 if (prog_data->base.use_alt_mode)
166 dw2 |= GEN7_PS_FLOATING_POINT_MODE_ALT;
167
168 /* Haswell requires the sample mask to be set in this packet as well as
169 * in 3DSTATE_SAMPLE_MASK; the values should match. */
170 /* _NEW_BUFFERS, _NEW_MULTISAMPLE */
171 if (brw->is_haswell)
172 dw4 |= SET_FIELD(sample_mask, HSW_PS_SAMPLE_MASK);
173
174 dw4 |= (brw->max_wm_threads - 1) << max_threads_shift;
175
176 if (prog_data->base.nr_params > 0)
177 dw4 |= GEN7_PS_PUSH_CONSTANT_ENABLE;
178
179 /* From the IVB PRM, volume 2 part 1, page 287:
180 * "This bit is inserted in the PS payload header and made available to
181 * the DataPort (either via the message header or via header bypass) to
182 * indicate that oMask data (one or two phases) is included in Render
183 * Target Write messages. If present, the oMask data is used to mask off
184 * samples."
185 */
186 if (prog_data->uses_omask)
187 dw4 |= GEN7_PS_OMASK_TO_RENDER_TARGET;
188
189 /* From the IVB PRM, volume 2 part 1, page 287:
190 * "If the PS kernel does not need the Position XY Offsets to
191 * compute a Position Value, then this field should be programmed
192 * to POSOFFSET_NONE."
193 * "SW Recommendation: If the PS kernel needs the Position Offsets
194 * to compute a Position XY value, this field should match Position
195 * ZW Interpolation Mode to ensure a consistent position.xyzw
196 * computation."
197 * We only require XY sample offsets. So, this recommendation doesn't
198 * look useful at the moment. We might need this in future.
199 */
200 if (prog_data->uses_pos_offset)
201 dw4 |= GEN7_PS_POSOFFSET_SAMPLE;
202 else
203 dw4 |= GEN7_PS_POSOFFSET_NONE;
204
205 /* The hardware wedges if you have this bit set but don't turn on any dual
206 * source blend factors.
207 */
208 if (enable_dual_src_blend)
209 dw4 |= GEN7_PS_DUAL_SOURCE_BLEND_ENABLE;
210
211 /* BRW_NEW_FS_PROG_DATA */
212 if (prog_data->num_varying_inputs != 0)
213 dw4 |= GEN7_PS_ATTRIBUTE_ENABLE;
214
215 dw4 |= fast_clear_op;
216
217 if (prog_data->dispatch_16)
218 dw4 |= GEN7_PS_16_DISPATCH_ENABLE;
219
220 if (prog_data->dispatch_8)
221 dw4 |= GEN7_PS_8_DISPATCH_ENABLE;
222
223 dw5 |= prog_data->base.dispatch_grf_start_reg <<
224 GEN7_PS_DISPATCH_START_GRF_SHIFT_0;
225 dw5 |= prog_data->dispatch_grf_start_reg_2 <<
226 GEN7_PS_DISPATCH_START_GRF_SHIFT_2;
227
228 ksp0 = stage_state->prog_offset;
229 ksp2 = stage_state->prog_offset + prog_data->prog_offset_2;
230
231 BEGIN_BATCH(8);
232 OUT_BATCH(_3DSTATE_PS << 16 | (8 - 2));
233 OUT_BATCH(ksp0);
234 OUT_BATCH(dw2);
235 if (prog_data->base.total_scratch) {
236 OUT_RELOC(brw->wm.base.scratch_bo,
237 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
238 ffs(stage_state->per_thread_scratch) - 11);
239 } else {
240 OUT_BATCH(0);
241 }
242 OUT_BATCH(dw4);
243 OUT_BATCH(dw5);
244 OUT_BATCH(0); /* kernel 1 pointer */
245 OUT_BATCH(ksp2);
246 ADVANCE_BATCH();
247 }
248
249 static void
250 upload_ps_state(struct brw_context *brw)
251 {
252 /* BRW_NEW_FS_PROG_DATA */
253 const struct brw_wm_prog_data *prog_data = brw->wm.prog_data;
254 const struct gl_context *ctx = &brw->ctx;
255 /* BRW_NEW_FS_PROG_DATA | _NEW_COLOR */
256 const bool enable_dual_src_blend = prog_data->dual_src_blend &&
257 (ctx->Color.BlendEnabled & 1) &&
258 ctx->Color.Blend[0]._UsesDualSrc;
259 /* _NEW_BUFFERS, _NEW_MULTISAMPLE */
260 const unsigned sample_mask =
261 brw->is_haswell ? gen6_determine_sample_mask(brw) : 0;
262
263 gen7_upload_ps_state(brw, &brw->wm.base, prog_data,
264 enable_dual_src_blend, sample_mask,
265 brw->wm.fast_clear_op);
266 }
267
268 const struct brw_tracked_state gen7_ps_state = {
269 .dirty = {
270 .mesa = _NEW_BUFFERS |
271 _NEW_COLOR |
272 _NEW_MULTISAMPLE,
273 .brw = BRW_NEW_BATCH |
274 BRW_NEW_BLORP |
275 BRW_NEW_FS_PROG_DATA,
276 },
277 .emit = upload_ps_state,
278 };