2 * Copyright © 2009 Intel Corporation
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:
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
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
24 * Eric Anholt <eric@anholt.net>
28 #include "brw_context.h"
29 #include "brw_state.h"
30 #include "brw_defines.h"
32 #include "main/macros.h"
33 #include "main/fbobject.h"
34 #include "intel_batchbuffer.h"
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.
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.
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
56 get_attr_override(const struct brw_vue_map
*vue_map
, int urb_entry_read_offset
,
57 int fs_attr
, bool two_side_color
, uint32_t *max_source_attr
)
59 /* Find the VUE slot for this attribute. */
60 int slot
= vue_map
->varying_to_slot
[fs_attr
];
62 /* If there was only a back color written but not front, use back
63 * as the color instead of undefined
65 if (slot
== -1 && fs_attr
== VARYING_SLOT_COL0
)
66 slot
= vue_map
->varying_to_slot
[VARYING_SLOT_BFC0
];
67 if (slot
== -1 && fs_attr
== VARYING_SLOT_COL1
)
68 slot
= vue_map
->varying_to_slot
[VARYING_SLOT_BFC1
];
71 /* This attribute does not exist in the VUE--that means that the vertex
72 * shader did not write to it. This means that either:
74 * (a) This attribute is a texture coordinate, and it is going to be
75 * replaced with point coordinates (as a consequence of a call to
76 * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
77 * hardware will ignore whatever attribute override we supply.
79 * (b) This attribute is read by the fragment shader but not written by
80 * the vertex shader, so its value is undefined. Therefore the
81 * attribute override we supply doesn't matter.
83 * In either case the attribute override we supply doesn't matter, so
84 * just reference the first available attribute.
89 /* Compute the location of the attribute relative to urb_entry_read_offset.
90 * Each increment of urb_entry_read_offset represents a 256-bit value, so
91 * it counts for two 128-bit VUE slots.
93 int source_attr
= slot
- 2 * urb_entry_read_offset
;
94 assert(source_attr
>= 0 && source_attr
< 32);
96 /* If we are doing two-sided color, and the VUE slot following this one
97 * represents a back-facing color, then we need to instruct the SF unit to
98 * do back-facing swizzling.
100 bool swizzling
= two_side_color
&&
101 ((vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL0
&&
102 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC0
) ||
103 (vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL1
&&
104 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC1
));
106 /* Update max_source_attr. If swizzling, the SF will read this slot + 1. */
107 if (*max_source_attr
< source_attr
+ swizzling
)
108 *max_source_attr
= source_attr
+ swizzling
;
112 (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING
<< ATTRIBUTE_SWIZZLE_SHIFT
);
120 * Create the mapping from the FS inputs we produce to the previous pipeline
121 * stage (GS or VS) outputs they source from.
124 calculate_attr_overrides(const struct brw_context
*brw
,
125 uint16_t *attr_overrides
,
126 uint32_t *point_sprite_enables
,
127 uint32_t *flat_enables
,
128 uint32_t *urb_entry_read_length
)
130 const int urb_entry_read_offset
= BRW_SF_URB_ENTRY_READ_OFFSET
;
131 uint32_t max_source_attr
= 0;
134 bool shade_model_flat
= brw
->ctx
.Light
.ShadeModel
== GL_FLAT
;
136 /* Initialize all the attr_overrides to 0. In the loop below we'll modify
137 * just the ones that correspond to inputs used by the fs.
139 memset(attr_overrides
, 0, 16*sizeof(*attr_overrides
));
141 for (int attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
142 enum glsl_interp_qualifier interp_qualifier
=
143 brw
->fragment_program
->InterpQualifier
[attr
];
144 bool is_gl_Color
= attr
== VARYING_SLOT_COL0
|| attr
== VARYING_SLOT_COL1
;
145 /* CACHE_NEW_WM_PROG */
146 int input_index
= brw
->wm
.prog_data
->urb_setup
[attr
];
152 if (brw
->ctx
.Point
.PointSprite
&&
153 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
) &&
154 brw
->ctx
.Point
.CoordReplace
[attr
- VARYING_SLOT_TEX0
]) {
155 *point_sprite_enables
|= (1 << input_index
);
158 if (attr
== VARYING_SLOT_PNTC
)
159 *point_sprite_enables
|= (1 << input_index
);
162 if (interp_qualifier
== INTERP_QUALIFIER_FLAT
||
163 (shade_model_flat
&& is_gl_Color
&&
164 interp_qualifier
== INTERP_QUALIFIER_NONE
))
165 *flat_enables
|= (1 << input_index
);
167 /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
168 uint16_t attr_override
=
169 get_attr_override(&brw
->vue_map_geom_out
,
170 urb_entry_read_offset
, attr
,
171 brw
->ctx
.VertexProgram
._TwoSideEnabled
,
174 /* The hardware can only do the overrides on 16 overrides at a
175 * time, and the other up to 16 have to be lined up so that the
176 * input index = the output index. We'll need to do some
177 * tweaking to make sure that's the case.
179 if (input_index
< 16)
180 attr_overrides
[input_index
] = attr_override
;
182 assert(attr_override
== input_index
);
185 /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
186 * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
188 * "This field should be set to the minimum length required to read the
189 * maximum source attribute. The maximum source attribute is indicated
190 * by the maximum value of the enabled Attribute # Source Attribute if
191 * Attribute Swizzle Enable is set, Number of Output Attributes-1 if
193 * read_length = ceiling((max_source_attr + 1) / 2)
195 * [errata] Corruption/Hang possible if length programmed larger than
198 * Similar text exists for Ivy Bridge.
200 *urb_entry_read_length
= ALIGN(max_source_attr
+ 1, 2) / 2;
205 upload_sf_state(struct brw_context
*brw
)
207 struct gl_context
*ctx
= &brw
->ctx
;
208 /* CACHE_NEW_WM_PROG */
209 uint32_t num_outputs
= brw
->wm
.prog_data
->num_varying_inputs
;
210 uint32_t dw1
, dw2
, dw3
, dw4
, dw16
, dw17
;
213 bool render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
214 bool multisampled_fbo
= ctx
->DrawBuffer
->Visual
.samples
> 1;
216 const int urb_entry_read_offset
= BRW_SF_URB_ENTRY_READ_OFFSET
;
218 uint16_t attr_overrides
[16];
219 uint32_t point_sprite_origin
;
221 dw1
= GEN6_SF_SWIZZLE_ENABLE
| num_outputs
<< GEN6_SF_NUM_OUTPUTS_SHIFT
;
223 dw2
= GEN6_SF_STATISTICS_ENABLE
|
224 GEN6_SF_VIEWPORT_TRANSFORM_ENABLE
;
232 if ((ctx
->Polygon
.FrontFace
== GL_CCW
) ^ render_to_fbo
)
233 dw2
|= GEN6_SF_WINDING_CCW
;
235 if (ctx
->Polygon
.OffsetFill
)
236 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID
;
238 if (ctx
->Polygon
.OffsetLine
)
239 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME
;
241 if (ctx
->Polygon
.OffsetPoint
)
242 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT
;
244 switch (ctx
->Polygon
.FrontMode
) {
246 dw2
|= GEN6_SF_FRONT_SOLID
;
250 dw2
|= GEN6_SF_FRONT_WIREFRAME
;
254 dw2
|= GEN6_SF_FRONT_POINT
;
262 switch (ctx
->Polygon
.BackMode
) {
264 dw2
|= GEN6_SF_BACK_SOLID
;
268 dw2
|= GEN6_SF_BACK_WIREFRAME
;
272 dw2
|= GEN6_SF_BACK_POINT
;
281 if (ctx
->Scissor
.Enabled
)
282 dw3
|= GEN6_SF_SCISSOR_ENABLE
;
285 if (ctx
->Polygon
.CullFlag
) {
286 switch (ctx
->Polygon
.CullFaceMode
) {
288 dw3
|= GEN6_SF_CULL_FRONT
;
291 dw3
|= GEN6_SF_CULL_BACK
;
293 case GL_FRONT_AND_BACK
:
294 dw3
|= GEN6_SF_CULL_BOTH
;
301 dw3
|= GEN6_SF_CULL_NONE
;
306 uint32_t line_width_u3_7
= U_FIXED(CLAMP(ctx
->Line
.Width
, 0.0, 7.99), 7);
307 /* TODO: line width of 0 is not allowed when MSAA enabled */
308 if (line_width_u3_7
== 0)
310 dw3
|= line_width_u3_7
<< GEN6_SF_LINE_WIDTH_SHIFT
;
312 if (ctx
->Line
.SmoothFlag
) {
313 dw3
|= GEN6_SF_LINE_AA_ENABLE
;
314 dw3
|= GEN6_SF_LINE_AA_MODE_TRUE
;
315 dw3
|= GEN6_SF_LINE_END_CAP_WIDTH_1_0
;
317 /* _NEW_MULTISAMPLE */
318 if (multisampled_fbo
&& ctx
->Multisample
.Enabled
)
319 dw3
|= GEN6_SF_MSRAST_ON_PATTERN
;
321 /* _NEW_PROGRAM | _NEW_POINT */
322 if (!(ctx
->VertexProgram
.PointSizeEnabled
||
323 ctx
->Point
._Attenuated
))
324 dw4
|= GEN6_SF_USE_STATE_POINT_WIDTH
;
326 /* Clamp to ARB_point_parameters user limits */
327 point_size
= CLAMP(ctx
->Point
.Size
, ctx
->Point
.MinSize
, ctx
->Point
.MaxSize
);
329 /* Clamp to the hardware limits and convert to fixed point */
330 dw4
|= U_FIXED(CLAMP(point_size
, 0.125, 255.875), 3);
333 * Window coordinates in an FBO are inverted, which means point
334 * sprite origin must be inverted, too.
336 if ((ctx
->Point
.SpriteOrigin
== GL_LOWER_LEFT
) != render_to_fbo
) {
337 point_sprite_origin
= GEN6_SF_POINT_SPRITE_LOWERLEFT
;
339 point_sprite_origin
= GEN6_SF_POINT_SPRITE_UPPERLEFT
;
341 dw1
|= point_sprite_origin
;
344 if (ctx
->Light
.ProvokingVertex
!= GL_FIRST_VERTEX_CONVENTION
) {
346 (2 << GEN6_SF_TRI_PROVOKE_SHIFT
) |
347 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
) |
348 (1 << GEN6_SF_LINE_PROVOKE_SHIFT
);
351 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
);
354 /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM |
357 uint32_t urb_entry_read_length
;
358 calculate_attr_overrides(brw
, attr_overrides
, &dw16
, &dw17
,
359 &urb_entry_read_length
);
360 dw1
|= (urb_entry_read_length
<< GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT
|
361 urb_entry_read_offset
<< GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT
);
364 OUT_BATCH(_3DSTATE_SF
<< 16 | (20 - 2));
369 OUT_BATCH_F(ctx
->Polygon
.OffsetUnits
* 2); /* constant. copied from gen4 */
370 OUT_BATCH_F(ctx
->Polygon
.OffsetFactor
); /* scale */
371 OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
372 for (i
= 0; i
< 8; i
++) {
373 OUT_BATCH(attr_overrides
[i
* 2] | attr_overrides
[i
* 2 + 1] << 16);
375 OUT_BATCH(dw16
); /* point sprite texcoord bitmask */
376 OUT_BATCH(dw17
); /* constant interp bitmask */
377 OUT_BATCH(0); /* wrapshortest enables 0-7 */
378 OUT_BATCH(0); /* wrapshortest enables 8-15 */
382 const struct brw_tracked_state gen6_sf_state
= {
384 .mesa
= (_NEW_LIGHT
|
392 .brw
= (BRW_NEW_CONTEXT
|
393 BRW_NEW_FRAGMENT_PROGRAM
|
394 BRW_NEW_VUE_MAP_GEOM_OUT
),
395 .cache
= CACHE_NEW_WM_PROG
397 .emit
= upload_sf_state
,