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 "main/framebuffer.h"
35 #include "intel_batchbuffer.h"
38 * Determine the appropriate attribute override value to store into the
39 * 3DSTATE_SF structure for a given fragment shader attribute. The attribute
40 * override value contains two pieces of information: the location of the
41 * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
42 * flag indicating whether to "swizzle" the attribute based on the direction
43 * the triangle is facing.
45 * If an attribute is "swizzled", then the given VUE location is used for
46 * front-facing triangles, and the VUE location that immediately follows is
47 * used for back-facing triangles. We use this to implement the mapping from
48 * gl_FrontColor/gl_BackColor to gl_Color.
50 * urb_entry_read_offset is the offset into the VUE at which the SF unit is
51 * being instructed to begin reading attribute data. It can be set to a
52 * nonzero value to prevent the SF unit from wasting time reading elements of
53 * the VUE that are not needed by the fragment shader. It is measured in
57 get_attr_override(const struct brw_vue_map
*vue_map
, int urb_entry_read_offset
,
58 int fs_attr
, bool two_side_color
, uint32_t *max_source_attr
)
60 /* Find the VUE slot for this attribute. */
61 int slot
= vue_map
->varying_to_slot
[fs_attr
];
63 /* Viewport and Layer are stored in the VUE header. We need to override
64 * them to zero if earlier stages didn't write them, as GL requires that
65 * they read back as zero when not explicitly set.
67 if (fs_attr
== VARYING_SLOT_VIEWPORT
|| fs_attr
== VARYING_SLOT_LAYER
) {
69 ATTRIBUTE_0_OVERRIDE_X
| ATTRIBUTE_0_OVERRIDE_W
|
70 ATTRIBUTE_CONST_0000
<< ATTRIBUTE_0_CONST_SOURCE_SHIFT
;
72 if (!(vue_map
->slots_valid
& VARYING_BIT_LAYER
))
73 override
|= ATTRIBUTE_0_OVERRIDE_Y
;
74 if (!(vue_map
->slots_valid
& VARYING_BIT_VIEWPORT
))
75 override
|= ATTRIBUTE_0_OVERRIDE_Z
;
80 /* If there was only a back color written but not front, use back
81 * as the color instead of undefined
83 if (slot
== -1 && fs_attr
== VARYING_SLOT_COL0
)
84 slot
= vue_map
->varying_to_slot
[VARYING_SLOT_BFC0
];
85 if (slot
== -1 && fs_attr
== VARYING_SLOT_COL1
)
86 slot
= vue_map
->varying_to_slot
[VARYING_SLOT_BFC1
];
89 /* This attribute does not exist in the VUE--that means that the vertex
90 * shader did not write to it. This means that either:
92 * (a) This attribute is a texture coordinate, and it is going to be
93 * replaced with point coordinates (as a consequence of a call to
94 * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
95 * hardware will ignore whatever attribute override we supply.
97 * (b) This attribute is read by the fragment shader but not written by
98 * the vertex shader, so its value is undefined. Therefore the
99 * attribute override we supply doesn't matter.
101 * (c) This attribute is gl_PrimitiveID, and it wasn't written by the
102 * previous shader stage.
104 * Note that we don't have to worry about the cases where the attribute
105 * is gl_PointCoord or is undergoing point sprite coordinate
106 * replacement, because in those cases, this function isn't called.
108 * In case (c), we need to program the attribute overrides so that the
109 * primitive ID will be stored in this slot. In every other case, the
110 * attribute override we supply doesn't matter. So just go ahead and
111 * program primitive ID in every case.
113 return (ATTRIBUTE_0_OVERRIDE_W
|
114 ATTRIBUTE_0_OVERRIDE_Z
|
115 ATTRIBUTE_0_OVERRIDE_Y
|
116 ATTRIBUTE_0_OVERRIDE_X
|
117 (ATTRIBUTE_CONST_PRIM_ID
<< ATTRIBUTE_0_CONST_SOURCE_SHIFT
));
120 /* Compute the location of the attribute relative to urb_entry_read_offset.
121 * Each increment of urb_entry_read_offset represents a 256-bit value, so
122 * it counts for two 128-bit VUE slots.
124 int source_attr
= slot
- 2 * urb_entry_read_offset
;
125 assert(source_attr
>= 0 && source_attr
< 32);
127 /* If we are doing two-sided color, and the VUE slot following this one
128 * represents a back-facing color, then we need to instruct the SF unit to
129 * do back-facing swizzling.
131 bool swizzling
= two_side_color
&&
132 ((vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL0
&&
133 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC0
) ||
134 (vue_map
->slot_to_varying
[slot
] == VARYING_SLOT_COL1
&&
135 vue_map
->slot_to_varying
[slot
+1] == VARYING_SLOT_BFC1
));
137 /* Update max_source_attr. If swizzling, the SF will read this slot + 1. */
138 if (*max_source_attr
< source_attr
+ swizzling
)
139 *max_source_attr
= source_attr
+ swizzling
;
143 (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING
<< ATTRIBUTE_SWIZZLE_SHIFT
);
151 * Create the mapping from the FS inputs we produce to the previous pipeline
152 * stage (GS or VS) outputs they source from.
155 calculate_attr_overrides(const struct brw_context
*brw
,
156 uint16_t *attr_overrides
,
157 uint32_t *point_sprite_enables
,
158 uint32_t *urb_entry_read_length
,
159 uint32_t *urb_entry_read_offset
)
161 uint32_t max_source_attr
= 0;
163 *point_sprite_enables
= 0;
165 *urb_entry_read_offset
= BRW_SF_URB_ENTRY_READ_OFFSET
;
167 /* BRW_NEW_FRAGMENT_PROGRAM
169 * If the fragment shader reads VARYING_SLOT_LAYER, then we need to pass in
170 * the full vertex header. Otherwise, we can program the SF to start
171 * reading at an offset of 1 (2 varying slots) to skip unnecessary data:
172 * - VARYING_SLOT_PSIZ and BRW_VARYING_SLOT_NDC on gen4-5
173 * - VARYING_SLOT_{PSIZ,LAYER} and VARYING_SLOT_POS on gen6+
176 bool fs_needs_vue_header
= brw
->fragment_program
->Base
.InputsRead
&
177 (VARYING_BIT_LAYER
| VARYING_BIT_VIEWPORT
);
179 *urb_entry_read_offset
= fs_needs_vue_header
? 0 : 1;
181 /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
182 * description of dw10 Point Sprite Texture Coordinate Enable:
184 * "This field must be programmed to zero when non-point primitives
187 * The SandyBridge PRM doesn't explicitly say that point sprite enables
188 * must be programmed to zero when rendering non-point primitives, but
189 * the IvyBridge PRM does, and if we don't, we get garbage.
191 * This is not required on Haswell, as the hardware ignores this state
192 * when drawing non-points -- although we do still need to be careful to
193 * correctly set the attr overrides.
196 * BRW_NEW_PRIMITIVE | BRW_NEW_GS_PROG_DATA | BRW_NEW_TES_PROG_DATA
198 bool drawing_points
= brw_is_drawing_points(brw
);
200 /* Initialize all the attr_overrides to 0. In the loop below we'll modify
201 * just the ones that correspond to inputs used by the fs.
203 memset(attr_overrides
, 0, 16*sizeof(*attr_overrides
));
205 for (int attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
206 /* BRW_NEW_FS_PROG_DATA */
207 int input_index
= brw
->wm
.prog_data
->urb_setup
[attr
];
213 bool point_sprite
= false;
214 if (drawing_points
) {
215 if (brw
->ctx
.Point
.PointSprite
&&
216 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
) &&
217 (brw
->ctx
.Point
.CoordReplace
& (1u << (attr
- VARYING_SLOT_TEX0
)))) {
221 if (attr
== VARYING_SLOT_PNTC
)
225 *point_sprite_enables
|= (1 << input_index
);
228 /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
229 uint16_t attr_override
= point_sprite
? 0 :
230 get_attr_override(&brw
->vue_map_geom_out
,
231 *urb_entry_read_offset
, attr
,
232 brw
->ctx
.VertexProgram
._TwoSideEnabled
,
235 /* The hardware can only do the overrides on 16 overrides at a
236 * time, and the other up to 16 have to be lined up so that the
237 * input index = the output index. We'll need to do some
238 * tweaking to make sure that's the case.
240 if (input_index
< 16)
241 attr_overrides
[input_index
] = attr_override
;
243 assert(attr_override
== input_index
);
246 /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
247 * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
249 * "This field should be set to the minimum length required to read the
250 * maximum source attribute. The maximum source attribute is indicated
251 * by the maximum value of the enabled Attribute # Source Attribute if
252 * Attribute Swizzle Enable is set, Number of Output Attributes-1 if
254 * read_length = ceiling((max_source_attr + 1) / 2)
256 * [errata] Corruption/Hang possible if length programmed larger than
259 * Similar text exists for Ivy Bridge.
261 *urb_entry_read_length
= ALIGN(max_source_attr
+ 1, 2) / 2;
266 upload_sf_state(struct brw_context
*brw
)
268 struct gl_context
*ctx
= &brw
->ctx
;
269 /* BRW_NEW_FS_PROG_DATA */
270 uint32_t num_outputs
= brw
->wm
.prog_data
->num_varying_inputs
;
271 uint32_t dw1
, dw2
, dw3
, dw4
;
272 uint32_t point_sprite_enables
;
275 bool render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
276 const bool multisampled_fbo
= _mesa_geometric_samples(ctx
->DrawBuffer
) > 1;
279 uint16_t attr_overrides
[16];
280 uint32_t point_sprite_origin
;
282 dw1
= GEN6_SF_SWIZZLE_ENABLE
| num_outputs
<< GEN6_SF_NUM_OUTPUTS_SHIFT
;
283 dw2
= GEN6_SF_STATISTICS_ENABLE
;
285 if (brw
->sf
.viewport_transform_enable
)
286 dw2
|= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE
;
292 if (ctx
->Polygon
._FrontBit
== render_to_fbo
)
293 dw2
|= GEN6_SF_WINDING_CCW
;
295 if (ctx
->Polygon
.OffsetFill
)
296 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID
;
298 if (ctx
->Polygon
.OffsetLine
)
299 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME
;
301 if (ctx
->Polygon
.OffsetPoint
)
302 dw2
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT
;
304 switch (ctx
->Polygon
.FrontMode
) {
306 dw2
|= GEN6_SF_FRONT_SOLID
;
310 dw2
|= GEN6_SF_FRONT_WIREFRAME
;
314 dw2
|= GEN6_SF_FRONT_POINT
;
318 unreachable("not reached");
321 switch (ctx
->Polygon
.BackMode
) {
323 dw2
|= GEN6_SF_BACK_SOLID
;
327 dw2
|= GEN6_SF_BACK_WIREFRAME
;
331 dw2
|= GEN6_SF_BACK_POINT
;
335 unreachable("not reached");
338 /* _NEW_SCISSOR | _NEW_POLYGON,
339 * BRW_NEW_GS_PROG_DATA | BRW_NEW_TES_PROG_DATA | BRW_NEW_PRIMITIVE
341 if (ctx
->Scissor
.EnableFlags
||
342 brw_is_drawing_points(brw
) || brw_is_drawing_lines(brw
))
343 dw3
|= GEN6_SF_SCISSOR_ENABLE
;
346 if (ctx
->Polygon
.CullFlag
) {
347 switch (ctx
->Polygon
.CullFaceMode
) {
349 dw3
|= GEN6_SF_CULL_FRONT
;
352 dw3
|= GEN6_SF_CULL_BACK
;
354 case GL_FRONT_AND_BACK
:
355 dw3
|= GEN6_SF_CULL_BOTH
;
358 unreachable("not reached");
361 dw3
|= GEN6_SF_CULL_NONE
;
366 uint32_t line_width_u3_7
= brw_get_line_width(brw
);
367 dw3
|= line_width_u3_7
<< GEN6_SF_LINE_WIDTH_SHIFT
;
369 if (ctx
->Line
.SmoothFlag
) {
370 dw3
|= GEN6_SF_LINE_AA_ENABLE
;
371 dw3
|= GEN6_SF_LINE_AA_MODE_TRUE
;
372 dw3
|= GEN6_SF_LINE_END_CAP_WIDTH_1_0
;
374 /* _NEW_MULTISAMPLE */
375 if (multisampled_fbo
&& ctx
->Multisample
.Enabled
)
376 dw3
|= GEN6_SF_MSRAST_ON_PATTERN
;
378 /* _NEW_PROGRAM | _NEW_POINT, BRW_NEW_VUE_MAP_GEOM_OUT */
379 if (use_state_point_size(brw
))
380 dw4
|= GEN6_SF_USE_STATE_POINT_WIDTH
;
382 /* _NEW_POINT - Clamp to ARB_point_parameters user limits */
383 point_size
= CLAMP(ctx
->Point
.Size
, ctx
->Point
.MinSize
, ctx
->Point
.MaxSize
);
385 /* Clamp to the hardware limits and convert to fixed point */
386 dw4
|= U_FIXED(CLAMP(point_size
, 0.125f
, 255.875f
), 3);
389 * Window coordinates in an FBO are inverted, which means point
390 * sprite origin must be inverted, too.
392 if ((ctx
->Point
.SpriteOrigin
== GL_LOWER_LEFT
) != render_to_fbo
) {
393 point_sprite_origin
= GEN6_SF_POINT_SPRITE_LOWERLEFT
;
395 point_sprite_origin
= GEN6_SF_POINT_SPRITE_UPPERLEFT
;
397 dw1
|= point_sprite_origin
;
400 if (ctx
->Light
.ProvokingVertex
!= GL_FIRST_VERTEX_CONVENTION
) {
402 (2 << GEN6_SF_TRI_PROVOKE_SHIFT
) |
403 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
) |
404 (1 << GEN6_SF_LINE_PROVOKE_SHIFT
);
407 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
);
410 /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
411 * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
413 uint32_t urb_entry_read_length
;
414 uint32_t urb_entry_read_offset
;
415 calculate_attr_overrides(brw
, attr_overrides
, &point_sprite_enables
,
416 &urb_entry_read_length
, &urb_entry_read_offset
);
417 dw1
|= (urb_entry_read_length
<< GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT
|
418 urb_entry_read_offset
<< GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT
);
421 OUT_BATCH(_3DSTATE_SF
<< 16 | (20 - 2));
426 OUT_BATCH_F(ctx
->Polygon
.OffsetUnits
* 2); /* constant. copied from gen4 */
427 OUT_BATCH_F(ctx
->Polygon
.OffsetFactor
); /* scale */
428 OUT_BATCH_F(ctx
->Polygon
.OffsetClamp
); /* global depth offset clamp */
429 for (i
= 0; i
< 8; i
++) {
430 OUT_BATCH(attr_overrides
[i
* 2] | attr_overrides
[i
* 2 + 1] << 16);
432 OUT_BATCH(point_sprite_enables
); /* dw16 */
433 OUT_BATCH(brw
->wm
.prog_data
->flat_inputs
);
434 OUT_BATCH(0); /* wrapshortest enables 0-7 */
435 OUT_BATCH(0); /* wrapshortest enables 8-15 */
439 const struct brw_tracked_state gen6_sf_state
= {
441 .mesa
= _NEW_BUFFERS
|
449 .brw
= BRW_NEW_BLORP
|
451 BRW_NEW_FRAGMENT_PROGRAM
|
452 BRW_NEW_FS_PROG_DATA
|
453 BRW_NEW_GS_PROG_DATA
|
455 BRW_NEW_TES_PROG_DATA
|
456 BRW_NEW_VUE_MAP_GEOM_OUT
,
458 .emit
= upload_sf_state
,