2 * Copyright 2012 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE.
25 #include "si_build_pm4.h"
26 #include "util/u_viewport.h"
27 #include "tgsi/tgsi_scan.h"
29 #define SI_MAX_SCISSOR 16384
31 static void si_set_scissor_states(struct pipe_context
*pctx
,
33 unsigned num_scissors
,
34 const struct pipe_scissor_state
*state
)
36 struct si_context
*ctx
= (struct si_context
*)pctx
;
39 for (i
= 0; i
< num_scissors
; i
++)
40 ctx
->scissors
.states
[start_slot
+ i
] = state
[i
];
42 if (!ctx
->queued
.named
.rasterizer
||
43 !ctx
->queued
.named
.rasterizer
->scissor_enable
)
46 ctx
->scissors
.dirty_mask
|= ((1 << num_scissors
) - 1) << start_slot
;
47 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.scissors
);
50 /* Since the guard band disables clipping, we have to clip per-pixel
53 static void si_get_scissor_from_viewport(struct si_context
*ctx
,
54 const struct pipe_viewport_state
*vp
,
55 struct si_signed_scissor
*scissor
)
57 float tmp
, minx
, miny
, maxx
, maxy
;
59 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
60 minx
= -vp
->scale
[0] + vp
->translate
[0];
61 miny
= -vp
->scale
[1] + vp
->translate
[1];
62 maxx
= vp
->scale
[0] + vp
->translate
[0];
63 maxy
= vp
->scale
[1] + vp
->translate
[1];
65 /* Handle inverted viewports. */
77 /* Convert to integer and round up the max bounds. */
80 scissor
->maxx
= ceilf(maxx
);
81 scissor
->maxy
= ceilf(maxy
);
84 static void si_clamp_scissor(struct si_context
*ctx
,
85 struct pipe_scissor_state
*out
,
86 struct si_signed_scissor
*scissor
)
88 out
->minx
= CLAMP(scissor
->minx
, 0, SI_MAX_SCISSOR
);
89 out
->miny
= CLAMP(scissor
->miny
, 0, SI_MAX_SCISSOR
);
90 out
->maxx
= CLAMP(scissor
->maxx
, 0, SI_MAX_SCISSOR
);
91 out
->maxy
= CLAMP(scissor
->maxy
, 0, SI_MAX_SCISSOR
);
94 static void si_clip_scissor(struct pipe_scissor_state
*out
,
95 struct pipe_scissor_state
*clip
)
97 out
->minx
= MAX2(out
->minx
, clip
->minx
);
98 out
->miny
= MAX2(out
->miny
, clip
->miny
);
99 out
->maxx
= MIN2(out
->maxx
, clip
->maxx
);
100 out
->maxy
= MIN2(out
->maxy
, clip
->maxy
);
103 static void si_scissor_make_union(struct si_signed_scissor
*out
,
104 struct si_signed_scissor
*in
)
106 out
->minx
= MIN2(out
->minx
, in
->minx
);
107 out
->miny
= MIN2(out
->miny
, in
->miny
);
108 out
->maxx
= MAX2(out
->maxx
, in
->maxx
);
109 out
->maxy
= MAX2(out
->maxy
, in
->maxy
);
110 out
->quant_mode
= MIN2(out
->quant_mode
, in
->quant_mode
);
113 static void si_emit_one_scissor(struct si_context
*ctx
,
114 struct radeon_cmdbuf
*cs
,
115 struct si_signed_scissor
*vp_scissor
,
116 struct pipe_scissor_state
*scissor
)
118 struct pipe_scissor_state final
;
120 if (ctx
->vs_disables_clipping_viewport
) {
121 final
.minx
= final
.miny
= 0;
122 final
.maxx
= final
.maxy
= SI_MAX_SCISSOR
;
124 si_clamp_scissor(ctx
, &final
, vp_scissor
);
128 si_clip_scissor(&final
, scissor
);
130 /* Workaround for a hw bug on SI that occurs when PA_SU_HARDWARE_-
131 * SCREEN_OFFSET != 0 and any_scissor.BR_X/Y <= 0.
133 if (ctx
->chip_class
== SI
&& (final
.maxx
== 0 || final
.maxy
== 0)) {
134 radeon_emit(cs
, S_028250_TL_X(1) |
136 S_028250_WINDOW_OFFSET_DISABLE(1));
137 radeon_emit(cs
, S_028254_BR_X(1) |
142 radeon_emit(cs
, S_028250_TL_X(final
.minx
) |
143 S_028250_TL_Y(final
.miny
) |
144 S_028250_WINDOW_OFFSET_DISABLE(1));
145 radeon_emit(cs
, S_028254_BR_X(final
.maxx
) |
146 S_028254_BR_Y(final
.maxy
));
149 #define MAX_PA_SU_HARDWARE_SCREEN_OFFSET 8176
151 static void si_emit_guardband(struct si_context
*ctx
)
153 const struct si_state_rasterizer
*rs
= ctx
->queued
.named
.rasterizer
;
154 struct si_signed_scissor vp_as_scissor
;
155 struct pipe_viewport_state vp
;
156 float left
, top
, right
, bottom
, max_range
, guardband_x
, guardband_y
;
157 float discard_x
, discard_y
;
159 if (ctx
->vs_writes_viewport_index
) {
160 /* Shaders can draw to any viewport. Make a union of all
162 vp_as_scissor
= ctx
->viewports
.as_scissor
[0];
163 for (unsigned i
= 1; i
< SI_MAX_VIEWPORTS
; i
++) {
164 si_scissor_make_union(&vp_as_scissor
,
165 &ctx
->viewports
.as_scissor
[i
]);
168 vp_as_scissor
= ctx
->viewports
.as_scissor
[0];
171 /* Blits don't set the viewport state. The vertex shader determines
172 * the viewport size by scaling the coordinates, so we don't know
173 * how large the viewport is. Assume the worst case.
175 if (ctx
->vs_disables_clipping_viewport
)
176 vp_as_scissor
.quant_mode
= SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH
;
178 /* Determine the optimal hardware screen offset to center the viewport
179 * within the viewport range in order to maximize the guardband size.
181 int hw_screen_offset_x
= (vp_as_scissor
.maxx
+ vp_as_scissor
.minx
) / 2;
182 int hw_screen_offset_y
= (vp_as_scissor
.maxy
+ vp_as_scissor
.miny
) / 2;
184 /* SI-CI need to align the offset to an ubertile consisting of all SEs. */
185 const unsigned hw_screen_offset_alignment
=
186 ctx
->chip_class
>= VI
? 16 : MAX2(ctx
->screen
->se_tile_repeat
, 16);
188 /* Indexed by quantization modes */
189 static unsigned max_viewport_size
[] = {65535, 16383, 4095};
191 /* Ensure that the whole viewport stays representable in
192 * absolute coordinates.
193 * See comment in si_set_viewport_states.
195 assert(vp_as_scissor
.maxx
<= max_viewport_size
[vp_as_scissor
.quant_mode
] &&
196 vp_as_scissor
.maxy
<= max_viewport_size
[vp_as_scissor
.quant_mode
]);
198 hw_screen_offset_x
= CLAMP(hw_screen_offset_x
, 0, MAX_PA_SU_HARDWARE_SCREEN_OFFSET
);
199 hw_screen_offset_y
= CLAMP(hw_screen_offset_y
, 0, MAX_PA_SU_HARDWARE_SCREEN_OFFSET
);
201 /* Align the screen offset by dropping the low bits. */
202 hw_screen_offset_x
&= ~(hw_screen_offset_alignment
- 1);
203 hw_screen_offset_y
&= ~(hw_screen_offset_alignment
- 1);
205 /* Apply the offset to center the viewport and maximize the guardband. */
206 vp_as_scissor
.minx
-= hw_screen_offset_x
;
207 vp_as_scissor
.maxx
-= hw_screen_offset_x
;
208 vp_as_scissor
.miny
-= hw_screen_offset_y
;
209 vp_as_scissor
.maxy
-= hw_screen_offset_y
;
211 /* Reconstruct the viewport transformation from the scissor. */
212 vp
.translate
[0] = (vp_as_scissor
.minx
+ vp_as_scissor
.maxx
) / 2.0;
213 vp
.translate
[1] = (vp_as_scissor
.miny
+ vp_as_scissor
.maxy
) / 2.0;
214 vp
.scale
[0] = vp_as_scissor
.maxx
- vp
.translate
[0];
215 vp
.scale
[1] = vp_as_scissor
.maxy
- vp
.translate
[1];
217 /* Treat a 0x0 viewport as 1x1 to prevent division by zero. */
218 if (vp_as_scissor
.minx
== vp_as_scissor
.maxx
)
220 if (vp_as_scissor
.miny
== vp_as_scissor
.maxy
)
223 /* Find the biggest guard band that is inside the supported viewport
224 * range. The guard band is specified as a horizontal and vertical
225 * distance from (0,0) in clip space.
227 * This is done by applying the inverse viewport transformation
228 * on the viewport limits to get those limits in clip space.
230 * The viewport range is [-max_viewport_size/2, max_viewport_size/2].
232 assert(vp_as_scissor
.quant_mode
< ARRAY_SIZE(max_viewport_size
));
233 max_range
= max_viewport_size
[vp_as_scissor
.quant_mode
] / 2;
234 left
= (-max_range
- vp
.translate
[0]) / vp
.scale
[0];
235 right
= ( max_range
- vp
.translate
[0]) / vp
.scale
[0];
236 top
= (-max_range
- vp
.translate
[1]) / vp
.scale
[1];
237 bottom
= ( max_range
- vp
.translate
[1]) / vp
.scale
[1];
239 assert(left
<= -1 && top
<= -1 && right
>= 1 && bottom
>= 1);
241 guardband_x
= MIN2(-left
, right
);
242 guardband_y
= MIN2(-top
, bottom
);
247 if (unlikely(util_prim_is_points_or_lines(ctx
->current_rast_prim
))) {
248 /* When rendering wide points or lines, we need to be more
249 * conservative about when to discard them entirely. */
252 if (ctx
->current_rast_prim
== PIPE_PRIM_POINTS
)
253 pixels
= rs
->max_point_size
;
255 pixels
= rs
->line_width
;
257 /* Add half the point size / line width */
258 discard_x
+= pixels
/ (2.0 * vp
.scale
[0]);
259 discard_y
+= pixels
/ (2.0 * vp
.scale
[1]);
261 /* Discard primitives that would lie entirely outside the clip
263 discard_x
= MIN2(discard_x
, guardband_x
);
264 discard_y
= MIN2(discard_y
, guardband_y
);
267 /* If any of the GB registers is updated, all of them must be updated.
268 * R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, R_028BEC_PA_CL_GB_VERT_DISC_ADJ
269 * R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ
271 unsigned initial_cdw
= ctx
->gfx_cs
->current
.cdw
;
272 radeon_opt_set_context_reg4(ctx
, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ
,
273 SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ
,
274 fui(guardband_y
), fui(discard_y
),
275 fui(guardband_x
), fui(discard_x
));
276 radeon_opt_set_context_reg(ctx
, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET
,
277 SI_TRACKED_PA_SU_HARDWARE_SCREEN_OFFSET
,
278 S_028234_HW_SCREEN_OFFSET_X(hw_screen_offset_x
>> 4) |
279 S_028234_HW_SCREEN_OFFSET_Y(hw_screen_offset_y
>> 4));
280 radeon_opt_set_context_reg(ctx
, R_028BE4_PA_SU_VTX_CNTL
,
281 SI_TRACKED_PA_SU_VTX_CNTL
,
282 S_028BE4_PIX_CENTER(rs
->half_pixel_center
) |
283 S_028BE4_QUANT_MODE(V_028BE4_X_16_8_FIXED_POINT_1_256TH
+
284 vp_as_scissor
.quant_mode
));
285 if (initial_cdw
!= ctx
->gfx_cs
->current
.cdw
)
286 ctx
->context_roll_counter
++;
289 static void si_emit_scissors(struct si_context
*ctx
)
291 struct radeon_cmdbuf
*cs
= ctx
->gfx_cs
;
292 struct pipe_scissor_state
*states
= ctx
->scissors
.states
;
293 unsigned mask
= ctx
->scissors
.dirty_mask
;
294 bool scissor_enabled
= ctx
->queued
.named
.rasterizer
->scissor_enable
;
296 /* The simple case: Only 1 viewport is active. */
297 if (!ctx
->vs_writes_viewport_index
) {
298 struct si_signed_scissor
*vp
= &ctx
->viewports
.as_scissor
[0];
303 radeon_set_context_reg_seq(cs
, R_028250_PA_SC_VPORT_SCISSOR_0_TL
, 2);
304 si_emit_one_scissor(ctx
, cs
, vp
, scissor_enabled
? &states
[0] : NULL
);
305 ctx
->scissors
.dirty_mask
&= ~1; /* clear one bit */
312 u_bit_scan_consecutive_range(&mask
, &start
, &count
);
314 radeon_set_context_reg_seq(cs
, R_028250_PA_SC_VPORT_SCISSOR_0_TL
+
315 start
* 4 * 2, count
* 2);
316 for (i
= start
; i
< start
+count
; i
++) {
317 si_emit_one_scissor(ctx
, cs
, &ctx
->viewports
.as_scissor
[i
],
318 scissor_enabled
? &states
[i
] : NULL
);
321 ctx
->scissors
.dirty_mask
= 0;
324 static void si_set_viewport_states(struct pipe_context
*pctx
,
326 unsigned num_viewports
,
327 const struct pipe_viewport_state
*state
)
329 struct si_context
*ctx
= (struct si_context
*)pctx
;
333 for (i
= 0; i
< num_viewports
; i
++) {
334 unsigned index
= start_slot
+ i
;
335 struct si_signed_scissor
*scissor
= &ctx
->viewports
.as_scissor
[index
];
337 ctx
->viewports
.states
[index
] = state
[i
];
339 si_get_scissor_from_viewport(ctx
, &state
[i
], scissor
);
341 unsigned w
= scissor
->maxx
- scissor
->minx
;
342 unsigned h
= scissor
->maxy
- scissor
->miny
;
343 unsigned max_extent
= MAX2(w
, h
);
345 int max_corner
= MAX2(scissor
->maxx
, scissor
->maxy
);
347 unsigned center_x
= (scissor
->maxx
+ scissor
->minx
) / 2;
348 unsigned center_y
= (scissor
->maxy
+ scissor
->miny
) / 2;
349 unsigned max_center
= MAX2(center_x
, center_y
);
351 /* PA_SU_HARDWARE_SCREEN_OFFSET can't center viewports whose
352 * center start farther than MAX_PA_SU_HARDWARE_SCREEN_OFFSET.
353 * (for example, a 1x1 viewport in the lower right corner of
354 * 16Kx16K) Such viewports need a greater guardband, so they
355 * have to use a worse quantization mode.
357 unsigned distance_off_center
=
358 MAX2(0, (int)max_center
- MAX_PA_SU_HARDWARE_SCREEN_OFFSET
);
359 max_extent
+= distance_off_center
;
361 /* Determine the best quantization mode (subpixel precision),
362 * but also leave enough space for the guardband.
364 * Note that primitive binning requires QUANT_MODE == 16_8 on Vega10
365 * and Raven1. What we do depends on the chip:
366 * - Vega10: Never use primitive binning.
367 * - Raven1: Always use QUANT_MODE == 16_8.
369 if (ctx
->family
== CHIP_RAVEN
)
370 max_extent
= 16384; /* Use QUANT_MODE == 16_8. */
372 /* Another constraint is that all coordinates in the viewport
373 * are representable in fixed point with respect to the
376 * It means that PA_SU_HARDWARE_SCREEN_OFFSET can't be given
377 * an offset that would make the upper corner of the viewport
378 * greater than the maximum representable number post
379 * quantization, ie 2^quant_bits.
381 * This does not matter for 14.10 and 16.8 formats since the
382 * offset is already limited at 8k, but it means we can't use
383 * 12.12 if we are drawing to some pixels outside the lower
384 * 4k x 4k of the render target.
387 if (max_extent
<= 1024 && max_corner
< 4096) /* 4K scanline area for guardband */
388 scissor
->quant_mode
= SI_QUANT_MODE_12_12_FIXED_POINT_1_4096TH
;
389 else if (max_extent
<= 4096) /* 16K scanline area for guardband */
390 scissor
->quant_mode
= SI_QUANT_MODE_14_10_FIXED_POINT_1_1024TH
;
391 else /* 64K scanline area for guardband */
392 scissor
->quant_mode
= SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH
;
395 mask
= ((1 << num_viewports
) - 1) << start_slot
;
396 ctx
->viewports
.dirty_mask
|= mask
;
397 ctx
->viewports
.depth_range_dirty_mask
|= mask
;
398 ctx
->scissors
.dirty_mask
|= mask
;
399 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.viewports
);
400 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.guardband
);
401 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.scissors
);
404 static void si_emit_one_viewport(struct si_context
*ctx
,
405 struct pipe_viewport_state
*state
)
407 struct radeon_cmdbuf
*cs
= ctx
->gfx_cs
;
409 radeon_emit(cs
, fui(state
->scale
[0]));
410 radeon_emit(cs
, fui(state
->translate
[0]));
411 radeon_emit(cs
, fui(state
->scale
[1]));
412 radeon_emit(cs
, fui(state
->translate
[1]));
413 radeon_emit(cs
, fui(state
->scale
[2]));
414 radeon_emit(cs
, fui(state
->translate
[2]));
417 static void si_emit_viewports(struct si_context
*ctx
)
419 struct radeon_cmdbuf
*cs
= ctx
->gfx_cs
;
420 struct pipe_viewport_state
*states
= ctx
->viewports
.states
;
421 unsigned mask
= ctx
->viewports
.dirty_mask
;
423 /* The simple case: Only 1 viewport is active. */
424 if (!ctx
->vs_writes_viewport_index
) {
428 radeon_set_context_reg_seq(cs
, R_02843C_PA_CL_VPORT_XSCALE
, 6);
429 si_emit_one_viewport(ctx
, &states
[0]);
430 ctx
->viewports
.dirty_mask
&= ~1; /* clear one bit */
437 u_bit_scan_consecutive_range(&mask
, &start
, &count
);
439 radeon_set_context_reg_seq(cs
, R_02843C_PA_CL_VPORT_XSCALE
+
440 start
* 4 * 6, count
* 6);
441 for (i
= start
; i
< start
+count
; i
++)
442 si_emit_one_viewport(ctx
, &states
[i
]);
444 ctx
->viewports
.dirty_mask
= 0;
448 si_viewport_zmin_zmax(const struct pipe_viewport_state
*vp
, bool halfz
,
449 bool window_space_position
, float *zmin
, float *zmax
)
451 if (window_space_position
) {
456 util_viewport_zmin_zmax(vp
, halfz
, zmin
, zmax
);
459 static void si_emit_depth_ranges(struct si_context
*ctx
)
461 struct radeon_cmdbuf
*cs
= ctx
->gfx_cs
;
462 struct pipe_viewport_state
*states
= ctx
->viewports
.states
;
463 unsigned mask
= ctx
->viewports
.depth_range_dirty_mask
;
464 bool clip_halfz
= ctx
->queued
.named
.rasterizer
->clip_halfz
;
465 bool window_space
= ctx
->vs_disables_clipping_viewport
;
468 /* The simple case: Only 1 viewport is active. */
469 if (!ctx
->vs_writes_viewport_index
) {
473 si_viewport_zmin_zmax(&states
[0], clip_halfz
, window_space
,
476 radeon_set_context_reg_seq(cs
, R_0282D0_PA_SC_VPORT_ZMIN_0
, 2);
477 radeon_emit(cs
, fui(zmin
));
478 radeon_emit(cs
, fui(zmax
));
479 ctx
->viewports
.depth_range_dirty_mask
&= ~1; /* clear one bit */
486 u_bit_scan_consecutive_range(&mask
, &start
, &count
);
488 radeon_set_context_reg_seq(cs
, R_0282D0_PA_SC_VPORT_ZMIN_0
+
489 start
* 4 * 2, count
* 2);
490 for (i
= start
; i
< start
+count
; i
++) {
491 si_viewport_zmin_zmax(&states
[i
], clip_halfz
, window_space
,
493 radeon_emit(cs
, fui(zmin
));
494 radeon_emit(cs
, fui(zmax
));
497 ctx
->viewports
.depth_range_dirty_mask
= 0;
500 static void si_emit_viewport_states(struct si_context
*ctx
)
502 si_emit_viewports(ctx
);
503 si_emit_depth_ranges(ctx
);
507 * This reacts to 2 state changes:
508 * - VS.writes_viewport_index
509 * - VS output position in window space (enable/disable)
511 * Normally, we only emit 1 viewport and 1 scissor if no shader is using
512 * the VIEWPORT_INDEX output, and emitting the other viewports and scissors
513 * is delayed. When a shader with VIEWPORT_INDEX appears, this should be
514 * called to emit the rest.
516 void si_update_vs_viewport_state(struct si_context
*ctx
)
518 struct tgsi_shader_info
*info
= si_get_vs_info(ctx
);
519 bool vs_window_space
;
524 /* When the VS disables clipping and viewport transformation. */
526 info
->properties
[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION
];
528 if (ctx
->vs_disables_clipping_viewport
!= vs_window_space
) {
529 ctx
->vs_disables_clipping_viewport
= vs_window_space
;
530 ctx
->scissors
.dirty_mask
= (1 << SI_MAX_VIEWPORTS
) - 1;
531 ctx
->viewports
.depth_range_dirty_mask
= (1 << SI_MAX_VIEWPORTS
) - 1;
532 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.scissors
);
533 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.viewports
);
536 /* Viewport index handling. */
537 if (ctx
->vs_writes_viewport_index
== info
->writes_viewport_index
)
540 /* This changes how the guardband is computed. */
541 ctx
->vs_writes_viewport_index
= info
->writes_viewport_index
;
542 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.guardband
);
544 if (!ctx
->vs_writes_viewport_index
)
547 if (ctx
->scissors
.dirty_mask
)
548 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.scissors
);
550 if (ctx
->viewports
.dirty_mask
||
551 ctx
->viewports
.depth_range_dirty_mask
)
552 si_mark_atom_dirty(ctx
, &ctx
->atoms
.s
.viewports
);
555 static void si_emit_window_rectangles(struct si_context
*sctx
)
557 /* There are four clipping rectangles. Their corner coordinates are inclusive.
558 * Every pixel is assigned a number from 0 and 15 by setting bits 0-3 depending
559 * on whether the pixel is inside cliprects 0-3, respectively. For example,
560 * if a pixel is inside cliprects 0 and 1, but outside 2 and 3, it is assigned
561 * the number 3 (binary 0011).
563 * If CLIPRECT_RULE & (1 << number), the pixel is rasterized.
565 struct radeon_cmdbuf
*cs
= sctx
->gfx_cs
;
566 static const unsigned outside
[4] = {
567 /* outside rectangle 0 */
576 /* outside rectangles 0, 1 */
581 /* outside rectangles 0, 1, 2 */
584 /* outside rectangles 0, 1, 2, 3 */
587 const unsigned disabled
= 0xffff; /* all inside and outside cases */
588 unsigned num_rectangles
= sctx
->num_window_rectangles
;
589 struct pipe_scissor_state
*rects
= sctx
->window_rectangles
;
592 assert(num_rectangles
<= 4);
594 if (num_rectangles
== 0)
596 else if (sctx
->window_rectangles_include
)
597 rule
= ~outside
[num_rectangles
- 1];
599 rule
= outside
[num_rectangles
- 1];
601 radeon_opt_set_context_reg(sctx
, R_02820C_PA_SC_CLIPRECT_RULE
,
602 SI_TRACKED_PA_SC_CLIPRECT_RULE
, rule
);
603 if (num_rectangles
== 0)
606 radeon_set_context_reg_seq(cs
, R_028210_PA_SC_CLIPRECT_0_TL
,
608 for (unsigned i
= 0; i
< num_rectangles
; i
++) {
609 radeon_emit(cs
, S_028210_TL_X(rects
[i
].minx
) |
610 S_028210_TL_Y(rects
[i
].miny
));
611 radeon_emit(cs
, S_028214_BR_X(rects
[i
].maxx
) |
612 S_028214_BR_Y(rects
[i
].maxy
));
616 static void si_set_window_rectangles(struct pipe_context
*ctx
,
618 unsigned num_rectangles
,
619 const struct pipe_scissor_state
*rects
)
621 struct si_context
*sctx
= (struct si_context
*)ctx
;
623 sctx
->num_window_rectangles
= num_rectangles
;
624 sctx
->window_rectangles_include
= include
;
625 if (num_rectangles
) {
626 memcpy(sctx
->window_rectangles
, rects
,
627 sizeof(*rects
) * num_rectangles
);
630 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.window_rectangles
);
633 void si_init_viewport_functions(struct si_context
*ctx
)
635 ctx
->atoms
.s
.guardband
.emit
= si_emit_guardband
;
636 ctx
->atoms
.s
.scissors
.emit
= si_emit_scissors
;
637 ctx
->atoms
.s
.viewports
.emit
= si_emit_viewport_states
;
638 ctx
->atoms
.s
.window_rectangles
.emit
= si_emit_window_rectangles
;
640 ctx
->b
.set_scissor_states
= si_set_scissor_states
;
641 ctx
->b
.set_viewport_states
= si_set_viewport_states
;
642 ctx
->b
.set_window_rectangles
= si_set_window_rectangles
;
644 for (unsigned i
= 0; i
< 16; i
++)
645 ctx
->viewports
.as_scissor
[i
].quant_mode
= SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH
;