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 * (c) This attribute is gl_PrimitiveID, and it wasn't written by the
84 * previous shader stage.
86 * Note that we don't have to worry about the cases where the attribute
87 * is gl_PointCoord or is undergoing point sprite coordinate
88 * replacement, because in those cases, this function isn't called.
90 * In case (c), we need to program the attribute overrides so that the
91 * primitive ID will be stored in this slot. In every other case, the
92 * attribute override we supply doesn't matter. So just go ahead and
93 * program primitive ID in every case.
95 return (ATTRIBUTE_0_OVERRIDE_W
|
96 ATTRIBUTE_0_OVERRIDE_Z
|
97 ATTRIBUTE_0_OVERRIDE_Y
|
98 ATTRIBUTE_0_OVERRIDE_X
|
99 (ATTRIBUTE_CONST_PRIM_ID
<< ATTRIBUTE_0_CONST_SOURCE_SHIFT
));
102 /* Compute the location of the attribute relative to urb_entry_read_offset.
103 * Each increment of urb_entry_read_offset represents a 256-bit value, so
104 * it counts for two 128-bit VUE slots.
106 int source_attr
= slot
- 2 * urb_entry_read_offset
;
107 assert(source_attr
>= 0 && source_attr
< 32);
109 /* If we are doing two-sided color, and the VUE slot following this one
110 * represents a back-facing color, then we need to instruct the SF unit to
111 * do back-facing swizzling.
113 bool swizzling
= two_side_color
&&
114 ((vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL0
&&
115 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC0
) ||
116 (vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL1
&&
117 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC1
));
119 /* Update max_source_attr. If swizzling, the SF will read this slot + 1. */
120 if (*max_source_attr
< source_attr
+ swizzling
)
121 *max_source_attr
= source_attr
+ swizzling
;
125 (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING
<< ATTRIBUTE_SWIZZLE_SHIFT
);
133 is_drawing_points(const struct brw_context
*brw
)
135 /* Determine if the primitives *reaching the SF* are points */
136 if (brw
->geometry_program
) {
137 /* BRW_NEW_GEOMETRY_PROGRAM */
138 return brw
->geometry_program
->OutputType
== GL_POINTS
;
140 /* BRW_NEW_PRIMITIVE */
141 return brw
->primitive
== _3DPRIM_POINTLIST
;
147 * Create the mapping from the FS inputs we produce to the previous pipeline
148 * stage (GS or VS) outputs they source from.
151 calculate_attr_overrides(const struct brw_context
*brw
,
152 uint16_t *attr_overrides
,
153 uint32_t *point_sprite_enables
,
154 uint32_t *flat_enables
,
155 uint32_t *urb_entry_read_length
)
157 const int urb_entry_read_offset
= BRW_SF_URB_ENTRY_READ_OFFSET
;
158 uint32_t max_source_attr
= 0;
160 *point_sprite_enables
= 0;
164 bool shade_model_flat
= brw
->ctx
.Light
.ShadeModel
== GL_FLAT
;
166 /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
167 * description of dw10 Point Sprite Texture Coordinate Enable:
169 * "This field must be programmed to zero when non-point primitives
172 * The SandyBridge PRM doesn't explicitly say that point sprite enables
173 * must be programmed to zero when rendering non-point primitives, but
174 * the IvyBridge PRM does, and if we don't, we get garbage.
176 * This is not required on Haswell, as the hardware ignores this state
177 * when drawing non-points -- although we do still need to be careful to
178 * correctly set the attr overrides.
180 /* BRW_NEW_PRIMITIVE | BRW_NEW_GEOMETRY_PROGRAM */
181 bool drawing_points
= is_drawing_points(brw
);
183 /* Initialize all the attr_overrides to 0. In the loop below we'll modify
184 * just the ones that correspond to inputs used by the fs.
186 memset(attr_overrides
, 0, 16*sizeof(*attr_overrides
));
188 for (int attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
189 /* BRW_NEW_FRAGMENT_PROGRAM */
190 enum glsl_interp_qualifier interp_qualifier
=
191 brw
->fragment_program
->InterpQualifier
[attr
];
192 bool is_gl_Color
= attr
== VARYING_SLOT_COL0
|| attr
== VARYING_SLOT_COL1
;
193 /* BRW_NEW_FS_PROG_DATA */
194 int input_index
= brw
->wm
.prog_data
->urb_setup
[attr
];
200 bool point_sprite
= false;
201 if (drawing_points
) {
202 if (brw
->ctx
.Point
.PointSprite
&&
203 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
) &&
204 brw
->ctx
.Point
.CoordReplace
[attr
- VARYING_SLOT_TEX0
]) {
208 if (attr
== VARYING_SLOT_PNTC
)
212 *point_sprite_enables
|= (1 << input_index
);
216 if (interp_qualifier
== INTERP_QUALIFIER_FLAT
||
217 (shade_model_flat
&& is_gl_Color
&&
218 interp_qualifier
== INTERP_QUALIFIER_NONE
))
219 *flat_enables
|= (1 << input_index
);
221 /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
222 uint16_t attr_override
= point_sprite
? 0 :
223 get_attr_override(&brw
->vue_map_geom_out
,
224 urb_entry_read_offset
, attr
,
225 brw
->ctx
.VertexProgram
._TwoSideEnabled
,
228 /* The hardware can only do the overrides on 16 overrides at a
229 * time, and the other up to 16 have to be lined up so that the
230 * input index = the output index. We'll need to do some
231 * tweaking to make sure that's the case.
233 if (input_index
< 16)
234 attr_overrides
[input_index
] = attr_override
;
236 assert(attr_override
== input_index
);
239 /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
240 * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
242 * "This field should be set to the minimum length required to read the
243 * maximum source attribute. The maximum source attribute is indicated
244 * by the maximum value of the enabled Attribute # Source Attribute if
245 * Attribute Swizzle Enable is set, Number of Output Attributes-1 if
247 * read_length = ceiling((max_source_attr + 1) / 2)
249 * [errata] Corruption/Hang possible if length programmed larger than
252 * Similar text exists for Ivy Bridge.
254 *urb_entry_read_length
= ALIGN(max_source_attr
+ 1, 2) / 2;
259 upload_sf_state(struct brw_context
*brw
)
261 struct gl_context
*ctx
= &brw
->ctx
;
262 /* BRW_NEW_FS_PROG_DATA */
263 uint32_t num_outputs
= brw
->wm
.prog_data
->num_varying_inputs
;
264 uint32_t dw1
, dw2
, dw3
, dw4
;
265 uint32_t point_sprite_enables
;
266 uint32_t flat_enables
;
269 bool render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
270 bool multisampled_fbo
= ctx
->DrawBuffer
->Visual
.samples
> 1;
272 const int urb_entry_read_offset
= BRW_SF_URB_ENTRY_READ_OFFSET
;
274 uint16_t attr_overrides
[16];
275 uint32_t point_sprite_origin
;
277 dw1
= GEN6_SF_SWIZZLE_ENABLE
| num_outputs
<< GEN6_SF_NUM_OUTPUTS_SHIFT
;
278 dw2
= GEN6_SF_STATISTICS_ENABLE
;
280 if (brw
->sf
.viewport_transform_enable
)
281 dw2
|= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE
;
287 if ((ctx
->Polygon
.FrontFace
== GL_CCW
) ^ render_to_fbo
)
288 dw2
|= GEN6_SF_WINDING_CCW
;
290 if (ctx
->Polygon
.OffsetFill
)
291 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID
;
293 if (ctx
->Polygon
.OffsetLine
)
294 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME
;
296 if (ctx
->Polygon
.OffsetPoint
)
297 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT
;
299 switch (ctx
->Polygon
.FrontMode
) {
301 dw2
|= GEN6_SF_FRONT_SOLID
;
305 dw2
|= GEN6_SF_FRONT_WIREFRAME
;
309 dw2
|= GEN6_SF_FRONT_POINT
;
313 unreachable("not reached");
316 switch (ctx
->Polygon
.BackMode
) {
318 dw2
|= GEN6_SF_BACK_SOLID
;
322 dw2
|= GEN6_SF_BACK_WIREFRAME
;
326 dw2
|= GEN6_SF_BACK_POINT
;
330 unreachable("not reached");
334 if (ctx
->Scissor
.EnableFlags
)
335 dw3
|= GEN6_SF_SCISSOR_ENABLE
;
338 if (ctx
->Polygon
.CullFlag
) {
339 switch (ctx
->Polygon
.CullFaceMode
) {
341 dw3
|= GEN6_SF_CULL_FRONT
;
344 dw3
|= GEN6_SF_CULL_BACK
;
346 case GL_FRONT_AND_BACK
:
347 dw3
|= GEN6_SF_CULL_BOTH
;
350 unreachable("not reached");
353 dw3
|= GEN6_SF_CULL_NONE
;
358 uint32_t line_width_u3_7
=
359 U_FIXED(CLAMP(ctx
->Line
.Width
, 0.0, ctx
->Const
.MaxLineWidth
), 7);
360 /* TODO: line width of 0 is not allowed when MSAA enabled */
361 if (line_width_u3_7
== 0)
363 dw3
|= line_width_u3_7
<< GEN6_SF_LINE_WIDTH_SHIFT
;
365 if (ctx
->Line
.SmoothFlag
) {
366 dw3
|= GEN6_SF_LINE_AA_ENABLE
;
367 dw3
|= GEN6_SF_LINE_AA_MODE_TRUE
;
368 dw3
|= GEN6_SF_LINE_END_CAP_WIDTH_1_0
;
370 /* _NEW_MULTISAMPLE */
371 if (multisampled_fbo
&& ctx
->Multisample
.Enabled
)
372 dw3
|= GEN6_SF_MSRAST_ON_PATTERN
;
374 /* _NEW_PROGRAM | _NEW_POINT */
375 if (!(ctx
->VertexProgram
.PointSizeEnabled
||
376 ctx
->Point
._Attenuated
))
377 dw4
|= GEN6_SF_USE_STATE_POINT_WIDTH
;
379 /* Clamp to ARB_point_parameters user limits */
380 point_size
= CLAMP(ctx
->Point
.Size
, ctx
->Point
.MinSize
, ctx
->Point
.MaxSize
);
382 /* Clamp to the hardware limits and convert to fixed point */
383 dw4
|= U_FIXED(CLAMP(point_size
, 0.125, 255.875), 3);
386 * Window coordinates in an FBO are inverted, which means point
387 * sprite origin must be inverted, too.
389 if ((ctx
->Point
.SpriteOrigin
== GL_LOWER_LEFT
) != render_to_fbo
) {
390 point_sprite_origin
= GEN6_SF_POINT_SPRITE_LOWERLEFT
;
392 point_sprite_origin
= GEN6_SF_POINT_SPRITE_UPPERLEFT
;
394 dw1
|= point_sprite_origin
;
397 if (ctx
->Light
.ProvokingVertex
!= GL_FIRST_VERTEX_CONVENTION
) {
399 (2 << GEN6_SF_TRI_PROVOKE_SHIFT
) |
400 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
) |
401 (1 << GEN6_SF_LINE_PROVOKE_SHIFT
);
404 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
);
407 /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
408 * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
410 uint32_t urb_entry_read_length
;
411 calculate_attr_overrides(brw
, attr_overrides
, &point_sprite_enables
,
412 &flat_enables
, &urb_entry_read_length
);
413 dw1
|= (urb_entry_read_length
<< GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT
|
414 urb_entry_read_offset
<< GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT
);
417 OUT_BATCH(_3DSTATE_SF
<< 16 | (20 - 2));
422 OUT_BATCH_F(ctx
->Polygon
.OffsetUnits
* 2); /* constant. copied from gen4 */
423 OUT_BATCH_F(ctx
->Polygon
.OffsetFactor
); /* scale */
424 OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
425 for (i
= 0; i
< 8; i
++) {
426 OUT_BATCH(attr_overrides
[i
* 2] | attr_overrides
[i
* 2 + 1] << 16);
428 OUT_BATCH(point_sprite_enables
); /* dw16 */
429 OUT_BATCH(flat_enables
);
430 OUT_BATCH(0); /* wrapshortest enables 0-7 */
431 OUT_BATCH(0); /* wrapshortest enables 8-15 */
435 const struct brw_tracked_state gen6_sf_state
= {
437 .mesa
= _NEW_BUFFERS
|
445 .brw
= BRW_NEW_CONTEXT
|
446 BRW_NEW_FRAGMENT_PROGRAM
|
447 BRW_NEW_GEOMETRY_PROGRAM
|
449 BRW_NEW_VUE_MAP_GEOM_OUT
,
450 .cache
= BRW_NEW_FS_PROG_DATA
452 .emit
= upload_sf_state
,