aeed36981562a2c5a0358e9de81442defb4fae9c
[mesa.git] / src / mesa / drivers / dri / i965 / gen6_sf_state.c
1 /*
2 * Copyright © 2009 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 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 *
26 */
27
28 #include "brw_context.h"
29 #include "brw_state.h"
30 #include "brw_defines.h"
31 #include "brw_util.h"
32 #include "main/macros.h"
33 #include "main/fbobject.h"
34 #include "intel_batchbuffer.h"
35
36 /**
37 * Determine the appropriate attribute override value to store into the
38 * 3DSTATE_SF structure for a given fragment shader attribute. The attribute
39 * override value contains two pieces of information: the location of the
40 * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
41 * flag indicating whether to "swizzle" the attribute based on the direction
42 * the triangle is facing.
43 *
44 * If an attribute is "swizzled", then the given VUE location is used for
45 * front-facing triangles, and the VUE location that immediately follows is
46 * used for back-facing triangles. We use this to implement the mapping from
47 * gl_FrontColor/gl_BackColor to gl_Color.
48 *
49 * urb_entry_read_offset is the offset into the VUE at which the SF unit is
50 * being instructed to begin reading attribute data. It can be set to a
51 * nonzero value to prevent the SF unit from wasting time reading elements of
52 * the VUE that are not needed by the fragment shader. It is measured in
53 * 256-bit increments.
54 */
55 uint32_t
56 get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
57 int fs_attr, bool two_side_color)
58 {
59 int attr_override, slot;
60 int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
61 if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
62 /* These attributes will be overwritten by the fragment shader's
63 * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
64 * them reference the first available attribute.
65 */
66 return 0;
67 }
68
69 /* Find the VUE slot for this attribute. */
70 slot = vue_map->vert_result_to_slot[vs_attr];
71
72 /* If there was only a back color written but not front, use back
73 * as the color instead of undefined
74 */
75 if (slot == -1 && vs_attr == VERT_RESULT_COL0)
76 slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
77 if (slot == -1 && vs_attr == VERT_RESULT_COL1)
78 slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
79
80 if (slot == -1) {
81 /* This attribute does not exist in the VUE--that means that the vertex
82 * shader did not write to it. Behavior is undefined in this case, so
83 * just reference the first available attribute.
84 */
85 return 0;
86 }
87
88 /* Compute the location of the attribute relative to urb_entry_read_offset.
89 * Each increment of urb_entry_read_offset represents a 256-bit value, so
90 * it counts for two 128-bit VUE slots.
91 */
92 attr_override = slot - 2 * urb_entry_read_offset;
93 assert (attr_override >= 0 && attr_override < 32);
94
95 /* If we are doing two-sided color, and the VUE slot following this one
96 * represents a back-facing color, then we need to instruct the SF unit to
97 * do back-facing swizzling.
98 */
99 if (two_side_color) {
100 if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
101 vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
102 attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
103 else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
104 vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
105 attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
106 }
107
108 return attr_override;
109 }
110
111 static void
112 upload_sf_state(struct brw_context *brw)
113 {
114 struct intel_context *intel = &brw->intel;
115 struct gl_context *ctx = &intel->ctx;
116 uint32_t urb_entry_read_length;
117 /* BRW_NEW_FRAGMENT_PROGRAM */
118 uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
119 /* _NEW_LIGHT */
120 bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
121 uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
122 int i;
123 /* _NEW_BUFFER */
124 bool render_to_fbo = _mesa_is_user_fbo(brw->intel.ctx.DrawBuffer);
125 bool multisampled_fbo = false;
126 if (ctx->DrawBuffer->_ColorDrawBuffers[0])
127 multisampled_fbo = ctx->DrawBuffer->_ColorDrawBuffers[0]->NumSamples > 0;
128
129 int attr = 0, input_index = 0;
130 int urb_entry_read_offset = 1;
131 float point_size;
132 uint16_t attr_overrides[FRAG_ATTRIB_MAX];
133 uint32_t point_sprite_origin;
134
135 /* CACHE_NEW_VS_PROG */
136 urb_entry_read_length = ((brw->vs.prog_data->vue_map.num_slots + 1) / 2 -
137 urb_entry_read_offset);
138 if (urb_entry_read_length == 0) {
139 /* Setting the URB entry read length to 0 causes undefined behavior, so
140 * if we have no URB data to read, set it to 1.
141 */
142 urb_entry_read_length = 1;
143 }
144
145 dw1 =
146 GEN6_SF_SWIZZLE_ENABLE |
147 num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
148 urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
149 urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
150
151 dw2 = GEN6_SF_STATISTICS_ENABLE |
152 GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
153
154 dw3 = 0;
155 dw4 = 0;
156 dw16 = 0;
157 dw17 = 0;
158
159 /* _NEW_POLYGON */
160 if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
161 dw2 |= GEN6_SF_WINDING_CCW;
162
163 if (ctx->Polygon.OffsetFill)
164 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
165
166 if (ctx->Polygon.OffsetLine)
167 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
168
169 if (ctx->Polygon.OffsetPoint)
170 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
171
172 switch (ctx->Polygon.FrontMode) {
173 case GL_FILL:
174 dw2 |= GEN6_SF_FRONT_SOLID;
175 break;
176
177 case GL_LINE:
178 dw2 |= GEN6_SF_FRONT_WIREFRAME;
179 break;
180
181 case GL_POINT:
182 dw2 |= GEN6_SF_FRONT_POINT;
183 break;
184
185 default:
186 assert(0);
187 break;
188 }
189
190 switch (ctx->Polygon.BackMode) {
191 case GL_FILL:
192 dw2 |= GEN6_SF_BACK_SOLID;
193 break;
194
195 case GL_LINE:
196 dw2 |= GEN6_SF_BACK_WIREFRAME;
197 break;
198
199 case GL_POINT:
200 dw2 |= GEN6_SF_BACK_POINT;
201 break;
202
203 default:
204 assert(0);
205 break;
206 }
207
208 /* _NEW_SCISSOR */
209 if (ctx->Scissor.Enabled)
210 dw3 |= GEN6_SF_SCISSOR_ENABLE;
211
212 /* _NEW_POLYGON */
213 if (ctx->Polygon.CullFlag) {
214 switch (ctx->Polygon.CullFaceMode) {
215 case GL_FRONT:
216 dw3 |= GEN6_SF_CULL_FRONT;
217 break;
218 case GL_BACK:
219 dw3 |= GEN6_SF_CULL_BACK;
220 break;
221 case GL_FRONT_AND_BACK:
222 dw3 |= GEN6_SF_CULL_BOTH;
223 break;
224 default:
225 assert(0);
226 break;
227 }
228 } else {
229 dw3 |= GEN6_SF_CULL_NONE;
230 }
231
232 /* _NEW_LINE */
233 {
234 uint32_t line_width_u3_7 = U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7);
235 /* TODO: line width of 0 is not allowed when MSAA enabled */
236 if (line_width_u3_7 == 0)
237 line_width_u3_7 = 1;
238 dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
239 }
240 if (ctx->Line.SmoothFlag) {
241 dw3 |= GEN6_SF_LINE_AA_ENABLE;
242 dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
243 dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
244 }
245 /* _NEW_MULTISAMPLE */
246 if (multisampled_fbo && ctx->Multisample.Enabled)
247 dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
248
249 /* _NEW_PROGRAM | _NEW_POINT */
250 if (!(ctx->VertexProgram.PointSizeEnabled ||
251 ctx->Point._Attenuated))
252 dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
253
254 /* Clamp to ARB_point_parameters user limits */
255 point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
256
257 /* Clamp to the hardware limits and convert to fixed point */
258 dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
259
260 /*
261 * Window coordinates in an FBO are inverted, which means point
262 * sprite origin must be inverted, too.
263 */
264 if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
265 point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
266 } else {
267 point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
268 }
269 dw1 |= point_sprite_origin;
270
271 /* _NEW_LIGHT */
272 if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
273 dw4 |=
274 (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
275 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
276 (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
277 } else {
278 dw4 |=
279 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
280 }
281
282 /* Create the mapping from the FS inputs we produce to the VS outputs
283 * they source from.
284 */
285 for (; attr < FRAG_ATTRIB_MAX; attr++) {
286 enum glsl_interp_qualifier interp_qualifier =
287 brw->fragment_program->InterpQualifier[attr];
288 bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
289
290 if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
291 continue;
292
293 /* _NEW_POINT */
294 if (ctx->Point.PointSprite &&
295 (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
296 ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
297 dw16 |= (1 << input_index);
298 }
299
300 if (attr == FRAG_ATTRIB_PNTC)
301 dw16 |= (1 << input_index);
302
303 /* flat shading */
304 if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
305 (shade_model_flat && is_gl_Color &&
306 interp_qualifier == INTERP_QUALIFIER_NONE))
307 dw17 |= (1 << input_index);
308
309 /* The hardware can only do the overrides on 16 overrides at a
310 * time, and the other up to 16 have to be lined up so that the
311 * input index = the output index. We'll need to do some
312 * tweaking to make sure that's the case.
313 */
314 assert(input_index < 16 || attr == input_index);
315
316 /* CACHE_NEW_VS_PROG | _NEW_LIGHT | _NEW_PROGRAM */
317 attr_overrides[input_index++] =
318 get_attr_override(&brw->vs.prog_data->vue_map,
319 urb_entry_read_offset, attr,
320 ctx->VertexProgram._TwoSideEnabled);
321 }
322
323 for (; input_index < FRAG_ATTRIB_MAX; input_index++)
324 attr_overrides[input_index] = 0;
325
326 BEGIN_BATCH(20);
327 OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
328 OUT_BATCH(dw1);
329 OUT_BATCH(dw2);
330 OUT_BATCH(dw3);
331 OUT_BATCH(dw4);
332 OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant. copied from gen4 */
333 OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
334 OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
335 for (i = 0; i < 8; i++) {
336 OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
337 }
338 OUT_BATCH(dw16); /* point sprite texcoord bitmask */
339 OUT_BATCH(dw17); /* constant interp bitmask */
340 OUT_BATCH(0); /* wrapshortest enables 0-7 */
341 OUT_BATCH(0); /* wrapshortest enables 8-15 */
342 ADVANCE_BATCH();
343 }
344
345 const struct brw_tracked_state gen6_sf_state = {
346 .dirty = {
347 .mesa = (_NEW_LIGHT |
348 _NEW_PROGRAM |
349 _NEW_POLYGON |
350 _NEW_LINE |
351 _NEW_SCISSOR |
352 _NEW_BUFFERS |
353 _NEW_POINT |
354 _NEW_MULTISAMPLE),
355 .brw = (BRW_NEW_CONTEXT |
356 BRW_NEW_FRAGMENT_PROGRAM),
357 .cache = CACHE_NEW_VS_PROG
358 },
359 .emit = upload_sf_state,
360 };