radv/radeonsi/nir: lower 64bit flrp
[mesa.git] / src / gallium / drivers / radeonsi / si_state_viewport.c
1 /*
2 * Copyright 2012 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 */
23
24 #include "si_pipe.h"
25 #include "sid.h"
26 #include "radeon/r600_cs.h"
27 #include "util/u_viewport.h"
28 #include "tgsi/tgsi_scan.h"
29
30 #define SI_MAX_SCISSOR 16384
31
32 static void si_set_scissor_states(struct pipe_context *pctx,
33 unsigned start_slot,
34 unsigned num_scissors,
35 const struct pipe_scissor_state *state)
36 {
37 struct si_context *ctx = (struct si_context *)pctx;
38 int i;
39
40 for (i = 0; i < num_scissors; i++)
41 ctx->scissors.states[start_slot + i] = state[i];
42
43 if (!ctx->queued.named.rasterizer ||
44 !ctx->queued.named.rasterizer->scissor_enable)
45 return;
46
47 ctx->scissors.dirty_mask |= ((1 << num_scissors) - 1) << start_slot;
48 si_mark_atom_dirty(ctx, &ctx->scissors.atom);
49 }
50
51 /* Since the guard band disables clipping, we have to clip per-pixel
52 * using a scissor.
53 */
54 static void si_get_scissor_from_viewport(struct si_context *ctx,
55 const struct pipe_viewport_state *vp,
56 struct si_signed_scissor *scissor)
57 {
58 float tmp, minx, miny, maxx, maxy;
59
60 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
61 minx = -vp->scale[0] + vp->translate[0];
62 miny = -vp->scale[1] + vp->translate[1];
63 maxx = vp->scale[0] + vp->translate[0];
64 maxy = vp->scale[1] + vp->translate[1];
65
66 /* Handle inverted viewports. */
67 if (minx > maxx) {
68 tmp = minx;
69 minx = maxx;
70 maxx = tmp;
71 }
72 if (miny > maxy) {
73 tmp = miny;
74 miny = maxy;
75 maxy = tmp;
76 }
77
78 /* Convert to integer and round up the max bounds. */
79 scissor->minx = minx;
80 scissor->miny = miny;
81 scissor->maxx = ceilf(maxx);
82 scissor->maxy = ceilf(maxy);
83 }
84
85 static void si_clamp_scissor(struct si_context *ctx,
86 struct pipe_scissor_state *out,
87 struct si_signed_scissor *scissor)
88 {
89 out->minx = CLAMP(scissor->minx, 0, SI_MAX_SCISSOR);
90 out->miny = CLAMP(scissor->miny, 0, SI_MAX_SCISSOR);
91 out->maxx = CLAMP(scissor->maxx, 0, SI_MAX_SCISSOR);
92 out->maxy = CLAMP(scissor->maxy, 0, SI_MAX_SCISSOR);
93 }
94
95 static void si_clip_scissor(struct pipe_scissor_state *out,
96 struct pipe_scissor_state *clip)
97 {
98 out->minx = MAX2(out->minx, clip->minx);
99 out->miny = MAX2(out->miny, clip->miny);
100 out->maxx = MIN2(out->maxx, clip->maxx);
101 out->maxy = MIN2(out->maxy, clip->maxy);
102 }
103
104 static void si_scissor_make_union(struct si_signed_scissor *out,
105 struct si_signed_scissor *in)
106 {
107 out->minx = MIN2(out->minx, in->minx);
108 out->miny = MIN2(out->miny, in->miny);
109 out->maxx = MAX2(out->maxx, in->maxx);
110 out->maxy = MAX2(out->maxy, in->maxy);
111 }
112
113 static void si_emit_one_scissor(struct si_context *ctx,
114 struct radeon_winsys_cs *cs,
115 struct si_signed_scissor *vp_scissor,
116 struct pipe_scissor_state *scissor)
117 {
118 struct pipe_scissor_state final;
119
120 if (ctx->vs_disables_clipping_viewport) {
121 final.minx = final.miny = 0;
122 final.maxx = final.maxy = SI_MAX_SCISSOR;
123 } else {
124 si_clamp_scissor(ctx, &final, vp_scissor);
125 }
126
127 if (scissor)
128 si_clip_scissor(&final, scissor);
129
130 radeon_emit(cs, S_028250_TL_X(final.minx) |
131 S_028250_TL_Y(final.miny) |
132 S_028250_WINDOW_OFFSET_DISABLE(1));
133 radeon_emit(cs, S_028254_BR_X(final.maxx) |
134 S_028254_BR_Y(final.maxy));
135 }
136
137 /* the range is [-MAX, MAX] */
138 #define GET_MAX_VIEWPORT_RANGE(rctx) (32768)
139
140 static void si_emit_guardband(struct si_context *ctx,
141 struct si_signed_scissor *vp_as_scissor)
142 {
143 struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
144 struct pipe_viewport_state vp;
145 float left, top, right, bottom, max_range, guardband_x, guardband_y;
146 float discard_x, discard_y;
147
148 /* Reconstruct the viewport transformation from the scissor. */
149 vp.translate[0] = (vp_as_scissor->minx + vp_as_scissor->maxx) / 2.0;
150 vp.translate[1] = (vp_as_scissor->miny + vp_as_scissor->maxy) / 2.0;
151 vp.scale[0] = vp_as_scissor->maxx - vp.translate[0];
152 vp.scale[1] = vp_as_scissor->maxy - vp.translate[1];
153
154 /* Treat a 0x0 viewport as 1x1 to prevent division by zero. */
155 if (vp_as_scissor->minx == vp_as_scissor->maxx)
156 vp.scale[0] = 0.5;
157 if (vp_as_scissor->miny == vp_as_scissor->maxy)
158 vp.scale[1] = 0.5;
159
160 /* Find the biggest guard band that is inside the supported viewport
161 * range. The guard band is specified as a horizontal and vertical
162 * distance from (0,0) in clip space.
163 *
164 * This is done by applying the inverse viewport transformation
165 * on the viewport limits to get those limits in clip space.
166 *
167 * Use a limit one pixel smaller to allow for some precision error.
168 */
169 max_range = GET_MAX_VIEWPORT_RANGE(ctx) - 1;
170 left = (-max_range - vp.translate[0]) / vp.scale[0];
171 right = ( max_range - vp.translate[0]) / vp.scale[0];
172 top = (-max_range - vp.translate[1]) / vp.scale[1];
173 bottom = ( max_range - vp.translate[1]) / vp.scale[1];
174
175 assert(left <= -1 && top <= -1 && right >= 1 && bottom >= 1);
176
177 guardband_x = MIN2(-left, right);
178 guardband_y = MIN2(-top, bottom);
179
180 discard_x = 1.0;
181 discard_y = 1.0;
182
183 if (unlikely(ctx->current_rast_prim < PIPE_PRIM_TRIANGLES) &&
184 ctx->queued.named.rasterizer) {
185 /* When rendering wide points or lines, we need to be more
186 * conservative about when to discard them entirely. */
187 const struct si_state_rasterizer *rs = ctx->queued.named.rasterizer;
188 float pixels;
189
190 if (ctx->current_rast_prim == PIPE_PRIM_POINTS)
191 pixels = rs->max_point_size;
192 else
193 pixels = rs->line_width;
194
195 /* Add half the point size / line width */
196 discard_x += pixels / (2.0 * vp.scale[0]);
197 discard_y += pixels / (2.0 * vp.scale[1]);
198
199 /* Discard primitives that would lie entirely outside the clip
200 * region. */
201 discard_x = MIN2(discard_x, guardband_x);
202 discard_y = MIN2(discard_y, guardband_y);
203 }
204
205 /* If any of the GB registers is updated, all of them must be updated. */
206 radeon_set_context_reg_seq(cs, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, 4);
207
208 radeon_emit(cs, fui(guardband_y)); /* R_028BE8_PA_CL_GB_VERT_CLIP_ADJ */
209 radeon_emit(cs, fui(discard_y)); /* R_028BEC_PA_CL_GB_VERT_DISC_ADJ */
210 radeon_emit(cs, fui(guardband_x)); /* R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ */
211 radeon_emit(cs, fui(discard_x)); /* R_028BF4_PA_CL_GB_HORZ_DISC_ADJ */
212 }
213
214 static void si_emit_scissors(struct r600_common_context *rctx, struct r600_atom *atom)
215 {
216 struct si_context *ctx = (struct si_context *)rctx;
217 struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
218 struct pipe_scissor_state *states = ctx->scissors.states;
219 unsigned mask = ctx->scissors.dirty_mask;
220 bool scissor_enabled = false;
221 struct si_signed_scissor max_vp_scissor;
222 int i;
223
224 if (ctx->queued.named.rasterizer)
225 scissor_enabled = ctx->queued.named.rasterizer->scissor_enable;
226
227 /* The simple case: Only 1 viewport is active. */
228 if (!ctx->vs_writes_viewport_index) {
229 struct si_signed_scissor *vp = &ctx->viewports.as_scissor[0];
230
231 if (!(mask & 1))
232 return;
233
234 radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL, 2);
235 si_emit_one_scissor(ctx, cs, vp, scissor_enabled ? &states[0] : NULL);
236 si_emit_guardband(ctx, vp);
237 ctx->scissors.dirty_mask &= ~1; /* clear one bit */
238 return;
239 }
240
241 /* Shaders can draw to any viewport. Make a union of all viewports. */
242 max_vp_scissor = ctx->viewports.as_scissor[0];
243 for (i = 1; i < SI_MAX_VIEWPORTS; i++)
244 si_scissor_make_union(&max_vp_scissor,
245 &ctx->viewports.as_scissor[i]);
246
247 while (mask) {
248 int start, count, i;
249
250 u_bit_scan_consecutive_range(&mask, &start, &count);
251
252 radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL +
253 start * 4 * 2, count * 2);
254 for (i = start; i < start+count; i++) {
255 si_emit_one_scissor(ctx, cs, &ctx->viewports.as_scissor[i],
256 scissor_enabled ? &states[i] : NULL);
257 }
258 }
259 si_emit_guardband(ctx, &max_vp_scissor);
260 ctx->scissors.dirty_mask = 0;
261 }
262
263 static void si_set_viewport_states(struct pipe_context *pctx,
264 unsigned start_slot,
265 unsigned num_viewports,
266 const struct pipe_viewport_state *state)
267 {
268 struct si_context *ctx = (struct si_context *)pctx;
269 unsigned mask;
270 int i;
271
272 for (i = 0; i < num_viewports; i++) {
273 unsigned index = start_slot + i;
274
275 ctx->viewports.states[index] = state[i];
276 si_get_scissor_from_viewport(ctx, &state[i],
277 &ctx->viewports.as_scissor[index]);
278 }
279
280 mask = ((1 << num_viewports) - 1) << start_slot;
281 ctx->viewports.dirty_mask |= mask;
282 ctx->viewports.depth_range_dirty_mask |= mask;
283 ctx->scissors.dirty_mask |= mask;
284 si_mark_atom_dirty(ctx, &ctx->viewports.atom);
285 si_mark_atom_dirty(ctx, &ctx->scissors.atom);
286 }
287
288 static void si_emit_one_viewport(struct si_context *ctx,
289 struct pipe_viewport_state *state)
290 {
291 struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
292
293 radeon_emit(cs, fui(state->scale[0]));
294 radeon_emit(cs, fui(state->translate[0]));
295 radeon_emit(cs, fui(state->scale[1]));
296 radeon_emit(cs, fui(state->translate[1]));
297 radeon_emit(cs, fui(state->scale[2]));
298 radeon_emit(cs, fui(state->translate[2]));
299 }
300
301 static void si_emit_viewports(struct si_context *ctx)
302 {
303 struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
304 struct pipe_viewport_state *states = ctx->viewports.states;
305 unsigned mask = ctx->viewports.dirty_mask;
306
307 /* The simple case: Only 1 viewport is active. */
308 if (!ctx->vs_writes_viewport_index) {
309 if (!(mask & 1))
310 return;
311
312 radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE, 6);
313 si_emit_one_viewport(ctx, &states[0]);
314 ctx->viewports.dirty_mask &= ~1; /* clear one bit */
315 return;
316 }
317
318 while (mask) {
319 int start, count, i;
320
321 u_bit_scan_consecutive_range(&mask, &start, &count);
322
323 radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE +
324 start * 4 * 6, count * 6);
325 for (i = start; i < start+count; i++)
326 si_emit_one_viewport(ctx, &states[i]);
327 }
328 ctx->viewports.dirty_mask = 0;
329 }
330
331 static inline void
332 si_viewport_zmin_zmax(const struct pipe_viewport_state *vp, bool halfz,
333 bool window_space_position, float *zmin, float *zmax)
334 {
335 if (window_space_position) {
336 *zmin = 0;
337 *zmax = 1;
338 return;
339 }
340 util_viewport_zmin_zmax(vp, halfz, zmin, zmax);
341 }
342
343 static void si_emit_depth_ranges(struct si_context *ctx)
344 {
345 struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
346 struct pipe_viewport_state *states = ctx->viewports.states;
347 unsigned mask = ctx->viewports.depth_range_dirty_mask;
348 bool clip_halfz = false;
349 bool window_space = ctx->vs_disables_clipping_viewport;
350 float zmin, zmax;
351
352 if (ctx->queued.named.rasterizer)
353 clip_halfz = ctx->queued.named.rasterizer->clip_halfz;
354
355 /* The simple case: Only 1 viewport is active. */
356 if (!ctx->vs_writes_viewport_index) {
357 if (!(mask & 1))
358 return;
359
360 si_viewport_zmin_zmax(&states[0], clip_halfz, window_space,
361 &zmin, &zmax);
362
363 radeon_set_context_reg_seq(cs, R_0282D0_PA_SC_VPORT_ZMIN_0, 2);
364 radeon_emit(cs, fui(zmin));
365 radeon_emit(cs, fui(zmax));
366 ctx->viewports.depth_range_dirty_mask &= ~1; /* clear one bit */
367 return;
368 }
369
370 while (mask) {
371 int start, count, i;
372
373 u_bit_scan_consecutive_range(&mask, &start, &count);
374
375 radeon_set_context_reg_seq(cs, R_0282D0_PA_SC_VPORT_ZMIN_0 +
376 start * 4 * 2, count * 2);
377 for (i = start; i < start+count; i++) {
378 si_viewport_zmin_zmax(&states[i], clip_halfz, window_space,
379 &zmin, &zmax);
380 radeon_emit(cs, fui(zmin));
381 radeon_emit(cs, fui(zmax));
382 }
383 }
384 ctx->viewports.depth_range_dirty_mask = 0;
385 }
386
387 static void si_emit_viewport_states(struct r600_common_context *rctx,
388 struct r600_atom *atom)
389 {
390 struct si_context *ctx = (struct si_context *)rctx;
391 si_emit_viewports(ctx);
392 si_emit_depth_ranges(ctx);
393 }
394
395 /**
396 * This reacts to 2 state changes:
397 * - VS.writes_viewport_index
398 * - VS output position in window space (enable/disable)
399 *
400 * Normally, we only emit 1 viewport and 1 scissor if no shader is using
401 * the VIEWPORT_INDEX output, and emitting the other viewports and scissors
402 * is delayed. When a shader with VIEWPORT_INDEX appears, this should be
403 * called to emit the rest.
404 */
405 void si_update_vs_viewport_state(struct si_context *ctx)
406 {
407 struct tgsi_shader_info *info = si_get_vs_info(ctx);
408 bool vs_window_space;
409
410 if (!info)
411 return;
412
413 /* When the VS disables clipping and viewport transformation. */
414 vs_window_space =
415 info->properties[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION];
416
417 if (ctx->vs_disables_clipping_viewport != vs_window_space) {
418 ctx->vs_disables_clipping_viewport = vs_window_space;
419 ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
420 ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
421 si_mark_atom_dirty(ctx, &ctx->scissors.atom);
422 si_mark_atom_dirty(ctx, &ctx->viewports.atom);
423 }
424
425 /* Viewport index handling. */
426 ctx->vs_writes_viewport_index = info->writes_viewport_index;
427 if (!ctx->vs_writes_viewport_index)
428 return;
429
430 if (ctx->scissors.dirty_mask)
431 si_mark_atom_dirty(ctx, &ctx->scissors.atom);
432
433 if (ctx->viewports.dirty_mask ||
434 ctx->viewports.depth_range_dirty_mask)
435 si_mark_atom_dirty(ctx, &ctx->viewports.atom);
436 }
437
438 void si_init_viewport_functions(struct si_context *ctx)
439 {
440 ctx->scissors.atom.emit = si_emit_scissors;
441 ctx->viewports.atom.emit = si_emit_viewport_states;
442
443 ctx->b.b.set_scissor_states = si_set_scissor_states;
444 ctx->b.b.set_viewport_states = si_set_viewport_states;
445 }