i965/fs: Set default access mode to Align1 for all instructions in the generator.
[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 const bool active_fs_has_side_effects =
82 _mesa_active_fragment_shader_has_side_effects(&brw->ctx);
83 if (brw_color_buffer_write_enabled(brw) || writes_depth ||
84 active_fs_has_side_effects || dw1 & GEN7_WM_KILL_ENABLE) {
85 dw1 |= GEN7_WM_DISPATCH_ENABLE;
86 }
87 if (multisampled_fbo) {
88 /* _NEW_MULTISAMPLE */
89 if (ctx->Multisample.Enabled)
90 dw1 |= GEN7_WM_MSRAST_ON_PATTERN;
91 else
92 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
93
94 if (prog_data->persample_dispatch)
95 dw2 |= GEN7_WM_MSDISPMODE_PERSAMPLE;
96 else
97 dw2 |= GEN7_WM_MSDISPMODE_PERPIXEL;
98 } else {
99 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
100 dw2 |= GEN7_WM_MSDISPMODE_PERSAMPLE;
101 }
102
103 if (prog_data->uses_sample_mask) {
104 dw1 |= GEN7_WM_USES_INPUT_COVERAGE_MASK;
105 }
106
107 /* BRW_NEW_FS_PROG_DATA */
108 if (prog_data->early_fragment_tests)
109 dw1 |= GEN7_WM_EARLY_DS_CONTROL_PREPS;
110 else if (active_fs_has_side_effects)
111 dw1 |= GEN7_WM_EARLY_DS_CONTROL_PSEXEC;
112
113 /* The "UAV access enable" bits are unnecessary on HSW because they only
114 * seem to have an effect on the HW-assisted coherency mechanism which we
115 * don't need, and the rasterization-related UAV_ONLY flag and the
116 * DISPATCH_ENABLE bit can be set independently from it.
117 * C.f. gen8_upload_ps_extra().
118 *
119 * BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_FS_PROG_DATA | _NEW_BUFFERS | _NEW_COLOR
120 */
121 if (brw->is_haswell &&
122 !(brw_color_buffer_write_enabled(brw) || writes_depth) &&
123 active_fs_has_side_effects)
124 dw2 |= HSW_WM_UAV_ONLY;
125
126 BEGIN_BATCH(3);
127 OUT_BATCH(_3DSTATE_WM << 16 | (3 - 2));
128 OUT_BATCH(dw1);
129 OUT_BATCH(dw2);
130 ADVANCE_BATCH();
131 }
132
133 const struct brw_tracked_state gen7_wm_state = {
134 .dirty = {
135 .mesa = _NEW_BUFFERS |
136 _NEW_COLOR |
137 _NEW_LINE |
138 _NEW_MULTISAMPLE |
139 _NEW_POLYGON,
140 .brw = BRW_NEW_BATCH |
141 BRW_NEW_BLORP |
142 BRW_NEW_FS_PROG_DATA,
143 },
144 .emit = upload_wm_state,
145 };
146
147 static void
148 gen7_upload_ps_state(struct brw_context *brw,
149 const struct brw_stage_state *stage_state,
150 const struct brw_wm_prog_data *prog_data,
151 bool enable_dual_src_blend, unsigned sample_mask,
152 unsigned fast_clear_op)
153 {
154 uint32_t dw2, dw4, dw5, ksp0, ksp2;
155 const int max_threads_shift = brw->is_haswell ?
156 HSW_PS_MAX_THREADS_SHIFT : IVB_PS_MAX_THREADS_SHIFT;
157
158 dw2 = dw4 = dw5 = ksp2 = 0;
159
160 const unsigned sampler_count =
161 DIV_ROUND_UP(CLAMP(stage_state->sampler_count, 0, 16), 4);
162 dw2 |= SET_FIELD(sampler_count, GEN7_PS_SAMPLER_COUNT);
163
164 dw2 |= ((prog_data->base.binding_table.size_bytes / 4) <<
165 GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT);
166
167 if (prog_data->base.use_alt_mode)
168 dw2 |= GEN7_PS_FLOATING_POINT_MODE_ALT;
169
170 /* Haswell requires the sample mask to be set in this packet as well as
171 * in 3DSTATE_SAMPLE_MASK; the values should match. */
172 /* _NEW_BUFFERS, _NEW_MULTISAMPLE */
173 if (brw->is_haswell)
174 dw4 |= SET_FIELD(sample_mask, HSW_PS_SAMPLE_MASK);
175
176 dw4 |= (brw->max_wm_threads - 1) << max_threads_shift;
177
178 if (prog_data->base.nr_params > 0)
179 dw4 |= GEN7_PS_PUSH_CONSTANT_ENABLE;
180
181 /* From the IVB PRM, volume 2 part 1, page 287:
182 * "This bit is inserted in the PS payload header and made available to
183 * the DataPort (either via the message header or via header bypass) to
184 * indicate that oMask data (one or two phases) is included in Render
185 * Target Write messages. If present, the oMask data is used to mask off
186 * samples."
187 */
188 if (prog_data->uses_omask)
189 dw4 |= GEN7_PS_OMASK_TO_RENDER_TARGET;
190
191 /* From the IVB PRM, volume 2 part 1, page 287:
192 * "If the PS kernel does not need the Position XY Offsets to
193 * compute a Position Value, then this field should be programmed
194 * to POSOFFSET_NONE."
195 * "SW Recommendation: If the PS kernel needs the Position Offsets
196 * to compute a Position XY value, this field should match Position
197 * ZW Interpolation Mode to ensure a consistent position.xyzw
198 * computation."
199 * We only require XY sample offsets. So, this recommendation doesn't
200 * look useful at the moment. We might need this in future.
201 */
202 if (prog_data->uses_pos_offset)
203 dw4 |= GEN7_PS_POSOFFSET_SAMPLE;
204 else
205 dw4 |= GEN7_PS_POSOFFSET_NONE;
206
207 /* The hardware wedges if you have this bit set but don't turn on any dual
208 * source blend factors.
209 */
210 if (enable_dual_src_blend)
211 dw4 |= GEN7_PS_DUAL_SOURCE_BLEND_ENABLE;
212
213 /* BRW_NEW_FS_PROG_DATA */
214 if (prog_data->num_varying_inputs != 0)
215 dw4 |= GEN7_PS_ATTRIBUTE_ENABLE;
216
217 dw4 |= fast_clear_op;
218
219 if (prog_data->dispatch_16)
220 dw4 |= GEN7_PS_16_DISPATCH_ENABLE;
221
222 if (prog_data->dispatch_8)
223 dw4 |= GEN7_PS_8_DISPATCH_ENABLE;
224
225 dw5 |= prog_data->base.dispatch_grf_start_reg <<
226 GEN7_PS_DISPATCH_START_GRF_SHIFT_0;
227 dw5 |= prog_data->dispatch_grf_start_reg_2 <<
228 GEN7_PS_DISPATCH_START_GRF_SHIFT_2;
229
230 ksp0 = stage_state->prog_offset;
231 ksp2 = stage_state->prog_offset + prog_data->prog_offset_2;
232
233 BEGIN_BATCH(8);
234 OUT_BATCH(_3DSTATE_PS << 16 | (8 - 2));
235 OUT_BATCH(ksp0);
236 OUT_BATCH(dw2);
237 if (prog_data->base.total_scratch) {
238 OUT_RELOC(brw->wm.base.scratch_bo,
239 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
240 ffs(prog_data->base.total_scratch) - 11);
241 } else {
242 OUT_BATCH(0);
243 }
244 OUT_BATCH(dw4);
245 OUT_BATCH(dw5);
246 OUT_BATCH(0); /* kernel 1 pointer */
247 OUT_BATCH(ksp2);
248 ADVANCE_BATCH();
249 }
250
251 static void
252 upload_ps_state(struct brw_context *brw)
253 {
254 /* BRW_NEW_FS_PROG_DATA */
255 const struct brw_wm_prog_data *prog_data = brw->wm.prog_data;
256 const struct gl_context *ctx = &brw->ctx;
257 /* BRW_NEW_FS_PROG_DATA | _NEW_COLOR */
258 const bool enable_dual_src_blend = prog_data->dual_src_blend &&
259 (ctx->Color.BlendEnabled & 1) &&
260 ctx->Color.Blend[0]._UsesDualSrc;
261 /* _NEW_BUFFERS, _NEW_MULTISAMPLE */
262 const unsigned sample_mask =
263 brw->is_haswell ? gen6_determine_sample_mask(brw) : 0;
264
265 gen7_upload_ps_state(brw, &brw->wm.base, prog_data,
266 enable_dual_src_blend, sample_mask,
267 brw->wm.fast_clear_op);
268 }
269
270 const struct brw_tracked_state gen7_ps_state = {
271 .dirty = {
272 .mesa = _NEW_BUFFERS |
273 _NEW_COLOR |
274 _NEW_MULTISAMPLE,
275 .brw = BRW_NEW_BATCH |
276 BRW_NEW_BLORP |
277 BRW_NEW_FS_PROG_DATA,
278 },
279 .emit = upload_ps_state,
280 };