i965: Store floating point mode choice in brw_stage_prog_data.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm_state.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 */
31
32
33
34 #include "intel_fbo.h"
35 #include "brw_context.h"
36 #include "brw_state.h"
37 #include "brw_defines.h"
38 #include "brw_wm.h"
39
40 /***********************************************************************
41 * WM unit - fragment programs and rasterization
42 */
43
44 bool
45 brw_color_buffer_write_enabled(struct brw_context *brw)
46 {
47 struct gl_context *ctx = &brw->ctx;
48 /* BRW_NEW_FRAGMENT_PROGRAM */
49 const struct gl_fragment_program *fp = brw->fragment_program;
50 int i;
51
52 /* _NEW_BUFFERS */
53 for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
54 struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
55
56 /* _NEW_COLOR */
57 if (rb &&
58 (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR) ||
59 fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DATA0 + i)) &&
60 (ctx->Color.ColorMask[i][0] ||
61 ctx->Color.ColorMask[i][1] ||
62 ctx->Color.ColorMask[i][2] ||
63 ctx->Color.ColorMask[i][3])) {
64 return true;
65 }
66 }
67
68 return false;
69 }
70
71 /**
72 * Setup wm hardware state. See page 225 of Volume 2
73 */
74 static void
75 brw_upload_wm_unit(struct brw_context *brw)
76 {
77 struct gl_context *ctx = &brw->ctx;
78 /* BRW_NEW_FRAGMENT_PROGRAM */
79 const struct gl_fragment_program *fp = brw->fragment_program;
80 /* BRW_NEW_FS_PROG_DATA */
81 const struct brw_wm_prog_data *prog_data = brw->wm.prog_data;
82 struct brw_wm_unit_state *wm;
83
84 wm = brw_state_batch(brw, AUB_TRACE_WM_STATE,
85 sizeof(*wm), 32, &brw->wm.base.state_offset);
86 memset(wm, 0, sizeof(*wm));
87
88 if (prog_data->prog_offset_16) {
89 /* These two fields should be the same pre-gen6, which is why we
90 * only have one hardware field to program for both dispatch
91 * widths.
92 */
93 assert(prog_data->base.dispatch_grf_start_reg ==
94 prog_data->dispatch_grf_start_reg_16);
95 }
96
97 /* BRW_NEW_PROGRAM_CACHE | BRW_NEW_FS_PROG_DATA */
98 wm->thread0.grf_reg_count = prog_data->reg_blocks;
99 wm->wm9.grf_reg_count_2 = prog_data->reg_blocks_16;
100
101 wm->thread0.kernel_start_pointer =
102 brw_program_reloc(brw,
103 brw->wm.base.state_offset +
104 offsetof(struct brw_wm_unit_state, thread0),
105 brw->wm.base.prog_offset +
106 (wm->thread0.grf_reg_count << 1)) >> 6;
107
108 wm->wm9.kernel_start_pointer_2 =
109 brw_program_reloc(brw,
110 brw->wm.base.state_offset +
111 offsetof(struct brw_wm_unit_state, wm9),
112 brw->wm.base.prog_offset +
113 prog_data->prog_offset_16 +
114 (wm->wm9.grf_reg_count_2 << 1)) >> 6;
115
116 wm->thread1.depth_coef_urb_read_offset = 1;
117 if (prog_data->base.use_alt_mode)
118 wm->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
119 else
120 wm->thread1.floating_point_mode = BRW_FLOATING_POINT_IEEE_754;
121
122 wm->thread1.binding_table_entry_count =
123 prog_data->base.binding_table.size_bytes / 4;
124
125 if (prog_data->base.total_scratch != 0) {
126 wm->thread2.scratch_space_base_pointer =
127 brw->wm.base.scratch_bo->offset64 >> 10; /* reloc */
128 wm->thread2.per_thread_scratch_space =
129 ffs(prog_data->base.total_scratch) - 11;
130 } else {
131 wm->thread2.scratch_space_base_pointer = 0;
132 wm->thread2.per_thread_scratch_space = 0;
133 }
134
135 wm->thread3.dispatch_grf_start_reg =
136 prog_data->base.dispatch_grf_start_reg;
137 wm->thread3.urb_entry_read_length =
138 prog_data->num_varying_inputs * 2;
139 wm->thread3.urb_entry_read_offset = 0;
140 wm->thread3.const_urb_entry_read_length =
141 prog_data->base.curb_read_length;
142 /* BRW_NEW_CURBE_OFFSETS */
143 wm->thread3.const_urb_entry_read_offset = brw->curbe.wm_start * 2;
144
145 if (brw->gen == 5)
146 wm->wm4.sampler_count = 0; /* hardware requirement */
147 else {
148 wm->wm4.sampler_count = (brw->wm.base.sampler_count + 1) / 4;
149 }
150
151 if (brw->wm.base.sampler_count) {
152 /* BRW_NEW_SAMPLER_STATE_TABLE - reloc */
153 wm->wm4.sampler_state_pointer = (brw->batch.bo->offset64 +
154 brw->wm.base.sampler_offset) >> 5;
155 } else {
156 wm->wm4.sampler_state_pointer = 0;
157 }
158
159 /* BRW_NEW_FRAGMENT_PROGRAM */
160 wm->wm5.program_uses_depth = (fp->Base.InputsRead &
161 (1 << VARYING_SLOT_POS)) != 0;
162 wm->wm5.program_computes_depth = (fp->Base.OutputsWritten &
163 BITFIELD64_BIT(FRAG_RESULT_DEPTH)) != 0;
164 /* _NEW_BUFFERS
165 * Override for NULL depthbuffer case, required by the Pixel Shader Computed
166 * Depth field.
167 */
168 if (!intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH))
169 wm->wm5.program_computes_depth = 0;
170
171 /* _NEW_COLOR */
172 wm->wm5.program_uses_killpixel =
173 prog_data->uses_kill || ctx->Color.AlphaEnabled;
174
175 wm->wm5.enable_8_pix = 1;
176 if (prog_data->prog_offset_16)
177 wm->wm5.enable_16_pix = 1;
178
179 wm->wm5.max_threads = brw->max_wm_threads - 1;
180
181 /* _NEW_BUFFERS | _NEW_COLOR */
182 if (brw_color_buffer_write_enabled(brw) ||
183 wm->wm5.program_uses_killpixel ||
184 wm->wm5.program_computes_depth) {
185 wm->wm5.thread_dispatch_enable = 1;
186 }
187
188 wm->wm5.legacy_line_rast = 0;
189 wm->wm5.legacy_global_depth_bias = 0;
190 wm->wm5.early_depth_test = 1; /* never need to disable */
191 wm->wm5.line_aa_region_width = 0;
192 wm->wm5.line_endcap_aa_region_width = 1;
193
194 /* _NEW_POLYGONSTIPPLE */
195 wm->wm5.polygon_stipple = ctx->Polygon.StippleFlag;
196
197 /* _NEW_POLYGON */
198 if (ctx->Polygon.OffsetFill) {
199 wm->wm5.depth_offset = 1;
200 /* Something wierd going on with legacy_global_depth_bias,
201 * offset_constant, scaling and MRD. This value passes glean
202 * but gives some odd results elsewere (eg. the
203 * quad-offset-units test).
204 */
205 wm->global_depth_offset_constant = ctx->Polygon.OffsetUnits * 2;
206
207 /* This is the only value that passes glean:
208 */
209 wm->global_depth_offset_scale = ctx->Polygon.OffsetFactor;
210 }
211
212 /* _NEW_LINE */
213 wm->wm5.line_stipple = ctx->Line.StippleFlag;
214
215 /* BRW_NEW_STATS_WM */
216 if (unlikely(INTEL_DEBUG & DEBUG_STATS) || brw->stats_wm)
217 wm->wm4.stats_enable = 1;
218
219 /* Emit scratch space relocation */
220 if (prog_data->base.total_scratch != 0) {
221 drm_intel_bo_emit_reloc(brw->batch.bo,
222 brw->wm.base.state_offset +
223 offsetof(struct brw_wm_unit_state, thread2),
224 brw->wm.base.scratch_bo,
225 wm->thread2.per_thread_scratch_space,
226 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER);
227 }
228
229 /* Emit sampler state relocation */
230 if (brw->wm.base.sampler_count != 0) {
231 drm_intel_bo_emit_reloc(brw->batch.bo,
232 brw->wm.base.state_offset +
233 offsetof(struct brw_wm_unit_state, wm4),
234 brw->batch.bo, (brw->wm.base.sampler_offset |
235 wm->wm4.stats_enable |
236 (wm->wm4.sampler_count << 2)),
237 I915_GEM_DOMAIN_INSTRUCTION, 0);
238 }
239
240 brw->state.dirty.brw |= BRW_NEW_GEN4_UNIT_STATE;
241 }
242
243 const struct brw_tracked_state brw_wm_unit = {
244 .dirty = {
245 .mesa = _NEW_BUFFERS |
246 _NEW_COLOR |
247 _NEW_LINE |
248 _NEW_POLYGON |
249 _NEW_POLYGONSTIPPLE,
250 .brw = BRW_NEW_BATCH |
251 BRW_NEW_CURBE_OFFSETS |
252 BRW_NEW_FRAGMENT_PROGRAM |
253 BRW_NEW_FS_PROG_DATA |
254 BRW_NEW_PROGRAM_CACHE |
255 BRW_NEW_SAMPLER_STATE_TABLE |
256 BRW_NEW_STATS_WM,
257 },
258 .emit = brw_upload_wm_unit,
259 };
260