projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
freedreno: update generated headers
[mesa.git] / src / gallium / drivers / freedreno / a5xx / fd5_emit.c
1 /*
2 * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
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 * 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:
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 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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Rob Clark <robclark@freedesktop.org>
25 */
26
27 #include "pipe/p_state.h"
28 #include "util/u_string.h"
29 #include "util/u_memory.h"
30 #include "util/u_helpers.h"
31 #include "util/u_format.h"
32 #include "util/u_viewport.h"
33
34 #include "freedreno_resource.h"
35 #include "freedreno_query_hw.h"
36
37 #include "fd5_emit.h"
38 #include "fd5_blend.h"
39 #include "fd5_context.h"
40 #include "fd5_program.h"
41 #include "fd5_rasterizer.h"
42 #include "fd5_texture.h"
43 #include "fd5_format.h"
44 #include "fd5_zsa.h"
45
46 static const enum adreno_state_block sb[] = {
47 [SHADER_VERTEX] = SB_VERT_SHADER,
48 [SHADER_FRAGMENT] = SB_FRAG_SHADER,
49 };
50
51 /* regid: base const register
52 * prsc or dwords: buffer containing constant values
53 * sizedwords: size of const value buffer
54 */
55 static void
56 fd5_emit_const(struct fd_ringbuffer *ring, enum shader_t type,
57 uint32_t regid, uint32_t offset, uint32_t sizedwords,
58 const uint32_t *dwords, struct pipe_resource *prsc)
59 {
60 uint32_t i, sz;
61 enum adreno_state_src src;
62
63 debug_assert((regid % 4) == 0);
64 debug_assert((sizedwords % 4) == 0);
65
66 if (prsc) {
67 sz = 0;
68 src = 0x2; // TODO ??
69 } else {
70 sz = sizedwords;
71 src = SS_DIRECT;
72 }
73
74 OUT_PKT7(ring, CP_LOAD_STATE, 3 + sz);
75 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
76 CP_LOAD_STATE_0_STATE_SRC(src) |
77 CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
78 CP_LOAD_STATE_0_NUM_UNIT(sizedwords/4));
79 if (prsc) {
80 struct fd_bo *bo = fd_resource(prsc)->bo;
81 OUT_RELOC(ring, bo, offset,
82 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0);
83 } else {
84 OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
85 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
86 OUT_RING(ring, CP_LOAD_STATE_2_EXT_SRC_ADDR_HI(0));
87 dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
88 }
89 for (i = 0; i < sz; i++) {
90 OUT_RING(ring, dwords[i]);
91 }
92 }
93
94 static void
95 fd5_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
96 uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets)
97 {
98 uint32_t anum = align(num, 2);
99 uint32_t i;
100
101 debug_assert((regid % 4) == 0);
102
103 OUT_PKT7(ring, CP_LOAD_STATE, 3 + (2 * anum));
104 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
105 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
106 CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
107 CP_LOAD_STATE_0_NUM_UNIT(anum/2));
108 OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
109 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
110 OUT_RING(ring, CP_LOAD_STATE_2_EXT_SRC_ADDR_HI(0));
111
112 for (i = 0; i < num; i++) {
113 if (prscs[i]) {
114 if (write) {
115 OUT_RELOCW(ring, fd_resource(prscs[i])->bo, offsets[i], 0, 0);
116 } else {
117 OUT_RELOC(ring, fd_resource(prscs[i])->bo, offsets[i], 0, 0);
118 }
119 } else {
120 OUT_RING(ring, 0xbad00000 | (i << 16));
121 OUT_RING(ring, 0xbad00000 | (i << 16));
122 }
123 }
124
125 for (; i < anum; i++) {
126 OUT_RING(ring, 0xffffffff);
127 OUT_RING(ring, 0xffffffff);
128 }
129 }
130
131 /* Border color layout is diff from a4xx/a5xx.. if it turns out to be
132 * the same as a6xx then move this somewhere common ;-)
133 *
134 * Entry layout looks like (total size, 0x60 bytes):
135 *
136 * offset | description
137 * -------+-------------
138 * 0x00 | fp32[0]
139 * | fp32[1]
140 * | fp32[2]
141 * | fp32[3]
142 * 0x10 | uint16[0]
143 * | uint16[1]
144 * | uint16[2]
145 * | uint16[3]
146 * 0x18 | int16[0]
147 * | int16[1]
148 * | int16[2]
149 * | int16[3]
150 * 0x20 | fp16[0]
151 * | fp16[1]
152 * | fp16[2]
153 * | fp16[3]
154 * 0x28 | ?? maybe padding ??
155 * 0x30 | uint8[0]
156 * | uint8[1]
157 * | uint8[2]
158 * | uint8[3]
159 * 0x34 | int8[0]
160 * | int8[1]
161 * | int8[2]
162 * | int8[3]
163 * 0x38 | ?? maybe padding ??
164 *
165 * Some uncertainty, because not clear that this actually works properly
166 * with blob, so who knows..
167 */
168
169 struct PACKED bcolor_entry {
170 uint32_t fp32[4];
171 uint16_t ui16[4];
172 int16_t si16[4];
173 uint16_t fp16[4];
174 uint8_t __pad0[8];
175 uint8_t ui8[4];
176 int8_t si8[4];
177 uint8_t __pad1[40];
178 };
179
180 #define FD5_BORDER_COLOR_SIZE 0x60
181 #define FD5_BORDER_COLOR_UPLOAD_SIZE (2 * PIPE_MAX_SAMPLERS * FD5_BORDER_COLOR_SIZE)
182 #define FD5_BORDER_COLOR_OFFSET 8 /* TODO probably should be dynamic */
183
184 static void
185 setup_border_colors(struct fd_texture_stateobj *tex, struct bcolor_entry *entries)
186 {
187 unsigned i, j;
188
189 debug_assert(tex->num_samplers < FD5_BORDER_COLOR_OFFSET); // TODO
190
191 for (i = 0; i < tex->num_samplers; i++) {
192 struct bcolor_entry *e = &entries[i];
193 struct pipe_sampler_state *sampler = tex->samplers[i];
194 union pipe_color_union *bc;
195
196 if (!sampler)
197 continue;
198
199 bc = &sampler->border_color;
200
201 /*
202 * XXX HACK ALERT XXX
203 *
204 * The border colors need to be swizzled in a particular
205 * format-dependent order. Even though samplers don't know about
206 * formats, we can assume that with a GL state tracker, there's a
207 * 1:1 correspondence between sampler and texture. Take advantage
208 * of that knowledge.
209 */
210 if ((i >= tex->num_textures) || !tex->textures[i])
211 continue;
212
213 const struct util_format_description *desc =
214 util_format_description(tex->textures[i]->format);
215
216 for (j = 0; j < 4; j++) {
217 int c = desc->swizzle[j];
218
219 if (c >= 4)
220 continue;
221
222 if (desc->channel[c].pure_integer) {
223 float f = bc->i[c];
224
225 e->fp32[j] = fui(f);
226 e->fp16[j] = util_float_to_half(f);
227 e->ui16[j] = bc->ui[c];
228 e->si16[j] = bc->i[c];
229 e->ui8[j] = bc->ui[c];
230 e->si8[j] = bc->i[c];
231 } else {
232 float f = bc->f[c];
233
234 e->fp32[j] = fui(f);
235 e->fp16[j] = util_float_to_half(f);
236 e->ui16[j] = f * 65535.0;
237 e->si16[j] = f * 32767.5;
238 e->ui8[j] = f * 255.0;
239 e->si8[j] = f * 128.0;
240 }
241 }
242
243 #ifdef DEBUG
244 memset(&e->__pad0, 0, sizeof(e->__pad0));
245 memset(&e->__pad1, 0, sizeof(e->__pad1));
246 #endif
247 }
248 }
249
250 static void
251 emit_border_color(struct fd_context *ctx, struct fd_ringbuffer *ring)
252 {
253 struct fd5_context *fd5_ctx = fd5_context(ctx);
254 struct bcolor_entry *entries;
255 unsigned off;
256 void *ptr;
257
258 STATIC_ASSERT(sizeof(struct bcolor_entry) == FD5_BORDER_COLOR_SIZE);
259
260 u_upload_alloc(fd5_ctx->border_color_uploader,
261 0, FD5_BORDER_COLOR_UPLOAD_SIZE,
262 FD5_BORDER_COLOR_UPLOAD_SIZE, &off,
263 &fd5_ctx->border_color_buf,
264 &ptr);
265
266 entries = ptr;
267
268 setup_border_colors(&ctx->verttex, &entries[0]);
269 setup_border_colors(&ctx->fragtex, &entries[ctx->verttex.num_samplers]);
270
271 OUT_PKT4(ring, REG_A5XX_TPL1_TP_BORDER_COLOR_BASE_ADDR_LO, 2);
272 OUT_RELOC(ring, fd_resource(fd5_ctx->border_color_buf)->bo, off, 0, 0);
273
274 u_upload_unmap(fd5_ctx->border_color_uploader);
275 }
276
277 static bool
278 emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
279 enum adreno_state_block sb, struct fd_texture_stateobj *tex)
280 {
281 bool needs_border = false;
282 unsigned bcolor_offset = (sb == SB_FRAG_TEX) ? ctx->verttex.num_samplers : 0;
283 unsigned i;
284
285 if (tex->num_samplers > 0) {
286 /* output sampler state: */
287 OUT_PKT7(ring, CP_LOAD_STATE, 3 + (4 * tex->num_samplers));
288 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
289 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
290 CP_LOAD_STATE_0_STATE_BLOCK(sb) |
291 CP_LOAD_STATE_0_NUM_UNIT(tex->num_samplers));
292 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
293 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
294 OUT_RING(ring, CP_LOAD_STATE_2_EXT_SRC_ADDR_HI(0));
295 for (i = 0; i < tex->num_samplers; i++) {
296 static const struct fd5_sampler_stateobj dummy_sampler = {};
297 const struct fd5_sampler_stateobj *sampler = tex->samplers[i] ?
298 fd5_sampler_stateobj(tex->samplers[i]) :
299 &dummy_sampler;
300 OUT_RING(ring, sampler->texsamp0);
301 OUT_RING(ring, sampler->texsamp1);
302 OUT_RING(ring, sampler->texsamp2 |
303 A5XX_TEX_SAMP_2_BCOLOR_OFFSET(bcolor_offset));
304 OUT_RING(ring, sampler->texsamp3);
305
306 needs_border |= sampler->needs_border;
307 }
308 }
309
310 if (tex->num_textures > 0) {
311 unsigned num_textures = tex->num_textures;
312
313 /* emit texture state: */
314 OUT_PKT7(ring, CP_LOAD_STATE, 3 + (12 * num_textures));
315 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
316 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
317 CP_LOAD_STATE_0_STATE_BLOCK(sb) |
318 CP_LOAD_STATE_0_NUM_UNIT(num_textures));
319 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
320 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
321 OUT_RING(ring, CP_LOAD_STATE_2_EXT_SRC_ADDR_HI(0));
322 for (i = 0; i < tex->num_textures; i++) {
323 static const struct fd5_pipe_sampler_view dummy_view = {};
324 const struct fd5_pipe_sampler_view *view = tex->textures[i] ?
325 fd5_pipe_sampler_view(tex->textures[i]) :
326 &dummy_view;
327
328 OUT_RING(ring, view->texconst0);
329 OUT_RING(ring, view->texconst1);
330 OUT_RING(ring, view->texconst2);
331 OUT_RING(ring, view->texconst3);
332 if (view->base.texture) {
333 struct fd_resource *rsc = fd_resource(view->base.texture);
334 OUT_RELOC(ring, rsc->bo, view->offset,
335 (uint64_t)view->texconst5 << 32, 0);
336 } else {
337 OUT_RING(ring, 0x00000000);
338 OUT_RING(ring, view->texconst5);
339 }
340 OUT_RING(ring, view->texconst6);
341 OUT_RING(ring, view->texconst7);
342 OUT_RING(ring, view->texconst8);
343 OUT_RING(ring, view->texconst9);
344 OUT_RING(ring, view->texconst10);
345 OUT_RING(ring, view->texconst11);
346 }
347 }
348
349 return needs_border;
350 }
351
352 void
353 fd5_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd5_emit *emit)
354 {
355 int32_t i, j;
356 const struct fd_vertex_state *vtx = emit->vtx;
357 const struct ir3_shader_variant *vp = fd5_emit_get_vp(emit);
358
359 for (i = 0, j = 0; i <= vp->inputs_count; i++) {
360 if (vp->inputs[i].sysval)
361 continue;
362 if (vp->inputs[i].compmask) {
363 struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
364 const struct pipe_vertex_buffer *vb =
365 &vtx->vertexbuf.vb[elem->vertex_buffer_index];
366 struct fd_resource *rsc = fd_resource(vb->buffer);
367 enum pipe_format pfmt = elem->src_format;
368 enum a5xx_vtx_fmt fmt = fd5_pipe2vtx(pfmt);
369 uint32_t off = vb->buffer_offset + elem->src_offset;
370 uint32_t size = fd_bo_size(rsc->bo) - off;
371 debug_assert(fmt != ~0);
372
373 OUT_PKT4(ring, REG_A5XX_VFD_FETCH(j), 4);
374 OUT_RELOC(ring, rsc->bo, off, 0, 0);
375 OUT_RING(ring, size); /* VFD_FETCH[j].SIZE */
376 OUT_RING(ring, vb->stride); /* VFD_FETCH[j].STRIDE */
377
378 OUT_PKT4(ring, REG_A5XX_VFD_DECODE(j), 2);
379 OUT_RING(ring, A5XX_VFD_DECODE_INSTR_IDX(j) |
380 A5XX_VFD_DECODE_INSTR_FORMAT(fmt) |
381 0xc0000000); // XXX
382 OUT_RING(ring, MAX2(1, elem->instance_divisor)); /* VFD_DECODE[j].STEP_RATE */
383
384 OUT_PKT4(ring, REG_A5XX_VFD_DEST_CNTL(j), 1);
385 OUT_RING(ring, A5XX_VFD_DEST_CNTL_INSTR_WRITEMASK(vp->inputs[i].compmask) |
386 A5XX_VFD_DEST_CNTL_INSTR_REGID(vp->inputs[i].regid));
387
388 j++;
389 }
390 }
391
392 OUT_PKT4(ring, REG_A5XX_VFD_CONTROL_0, 1);
393 OUT_RING(ring, A5XX_VFD_CONTROL_0_VTXCNT(j));
394 }
395
396 void
397 fd5_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
398 struct fd5_emit *emit)
399 {
400 const struct ir3_shader_variant *vp = fd5_emit_get_vp(emit);
401 const struct ir3_shader_variant *fp = fd5_emit_get_fp(emit);
402 uint32_t dirty = emit->dirty;
403 bool needs_border = false;
404
405 emit_marker5(ring, 5);
406
407 if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
408 struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
409 unsigned char mrt_comp[A5XX_MAX_RENDER_TARGETS] = {0};
410
411 for (unsigned i = 0; i < A5XX_MAX_RENDER_TARGETS; i++) {
412 mrt_comp[i] = ((i < pfb->nr_cbufs) && pfb->cbufs[i]) ? 0xf : 0;
413 }
414
415 OUT_PKT4(ring, REG_A5XX_RB_RENDER_COMPONENTS, 1);
416 OUT_RING(ring, A5XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
417 A5XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
418 A5XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
419 A5XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
420 A5XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
421 A5XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
422 A5XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
423 A5XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
424 }
425
426 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_FRAMEBUFFER)) {
427 struct fd5_zsa_stateobj *zsa = fd5_zsa_stateobj(ctx->zsa);
428 struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
429 uint32_t rb_alpha_control = zsa->rb_alpha_control;
430
431 if (util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0])))
432 rb_alpha_control &= ~A5XX_RB_ALPHA_CONTROL_ALPHA_TEST;
433
434 OUT_PKT4(ring, REG_A5XX_RB_ALPHA_CONTROL, 1);
435 OUT_RING(ring, rb_alpha_control);
436
437 OUT_PKT4(ring, REG_A5XX_RB_STENCIL_CONTROL, 1);
438 OUT_RING(ring, zsa->rb_stencil_control);
439 }
440
441 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
442 struct fd5_zsa_stateobj *zsa = fd5_zsa_stateobj(ctx->zsa);
443 struct pipe_stencil_ref *sr = &ctx->stencil_ref;
444
445 OUT_PKT4(ring, REG_A5XX_RB_STENCILREFMASK, 1);
446 OUT_RING(ring, zsa->rb_stencilrefmask |
447 A5XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
448 }
449
450 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
451 struct fd5_zsa_stateobj *zsa = fd5_zsa_stateobj(ctx->zsa);
452 bool fragz = fp->has_kill | fp->writes_pos;
453
454 OUT_PKT4(ring, REG_A5XX_RB_DEPTH_CNTL, 1);
455 OUT_RING(ring, zsa->rb_depth_cntl);
456
457 OUT_PKT4(ring, REG_A5XX_RB_DEPTH_PLANE_CNTL, 1);
458 OUT_RING(ring, COND(fragz, A5XX_RB_DEPTH_PLANE_CNTL_FRAG_WRITES_Z));
459
460 OUT_PKT4(ring, REG_A5XX_GRAS_SU_DEPTH_PLANE_CNTL, 1);
461 OUT_RING(ring, COND(fragz, A5XX_GRAS_SU_DEPTH_PLANE_CNTL_FRAG_WRITES_Z));
462 }
463
464 if (dirty & FD_DIRTY_RASTERIZER) {
465 struct fd5_rasterizer_stateobj *rasterizer =
466 fd5_rasterizer_stateobj(ctx->rasterizer);
467
468 OUT_PKT4(ring, REG_A5XX_GRAS_SU_CNTL, 1);
469 OUT_RING(ring, rasterizer->gras_su_cntl);
470
471 OUT_PKT4(ring, REG_A5XX_GRAS_SU_POINT_MINMAX, 2);
472 OUT_RING(ring, rasterizer->gras_su_point_minmax);
473 OUT_RING(ring, rasterizer->gras_su_point_size);
474
475 OUT_PKT4(ring, REG_A5XX_GRAS_SU_POLY_OFFSET_SCALE, 3);
476 OUT_RING(ring, rasterizer->gras_su_poly_offset_scale);
477 OUT_RING(ring, rasterizer->gras_su_poly_offset_offset);
478 OUT_RING(ring, rasterizer->gras_su_poly_offset_clamp);
479 }
480
481 /* NOTE: since primitive_restart is not actually part of any
482 * state object, we need to make sure that we always emit
483 * PRIM_VTX_CNTL.. either that or be more clever and detect
484 * when it changes.
485 */
486 if (emit->info) {
487 struct fd5_rasterizer_stateobj *rast =
488 fd5_rasterizer_stateobj(ctx->rasterizer);
489 uint32_t val = rast->pc_prim_vtx_cntl;
490
491 val |= COND(vp->writes_psize, A5XX_PC_PRIM_VTX_CNTL_PSIZE);
492
493 OUT_PKT4(ring, REG_A5XX_PC_PRIM_VTX_CNTL, 1);
494 OUT_RING(ring, val);
495 }
496
497 if (dirty & FD_DIRTY_SCISSOR) {
498 struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
499
500 OUT_PKT4(ring, REG_A5XX_GRAS_SC_SCREEN_SCISSOR_TL_0, 2);
501 OUT_RING(ring, A5XX_GRAS_SC_SCREEN_SCISSOR_TL_0_X(scissor->minx) |
502 A5XX_GRAS_SC_SCREEN_SCISSOR_TL_0_Y(scissor->miny));
503 OUT_RING(ring, A5XX_GRAS_SC_SCREEN_SCISSOR_TL_0_X(scissor->maxx - 1) |
504 A5XX_GRAS_SC_SCREEN_SCISSOR_TL_0_Y(scissor->maxy - 1));
505
506 OUT_PKT4(ring, REG_A5XX_GRAS_SC_VIEWPORT_SCISSOR_TL_0, 2);
507 OUT_RING(ring, A5XX_GRAS_SC_VIEWPORT_SCISSOR_TL_0_X(scissor->minx) |
508 A5XX_GRAS_SC_VIEWPORT_SCISSOR_TL_0_Y(scissor->miny));
509 OUT_RING(ring, A5XX_GRAS_SC_VIEWPORT_SCISSOR_TL_0_X(scissor->maxx - 1) |
510 A5XX_GRAS_SC_VIEWPORT_SCISSOR_TL_0_Y(scissor->maxy - 1));
511
512 ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
513 ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
514 ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
515 ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
516 }
517
518 if (dirty & FD_DIRTY_VIEWPORT) {
519 fd_wfi(ctx->batch, ring);
520 OUT_PKT4(ring, REG_A5XX_GRAS_CL_VPORT_XOFFSET_0, 6);
521 OUT_RING(ring, A5XX_GRAS_CL_VPORT_XOFFSET_0(ctx->viewport.translate[0]));
522 OUT_RING(ring, A5XX_GRAS_CL_VPORT_XSCALE_0(ctx->viewport.scale[0]));
523 OUT_RING(ring, A5XX_GRAS_CL_VPORT_YOFFSET_0(ctx->viewport.translate[1]));
524 OUT_RING(ring, A5XX_GRAS_CL_VPORT_YSCALE_0(ctx->viewport.scale[1]));
525 OUT_RING(ring, A5XX_GRAS_CL_VPORT_ZOFFSET_0(ctx->viewport.translate[2]));
526 OUT_RING(ring, A5XX_GRAS_CL_VPORT_ZSCALE_0(ctx->viewport.scale[2]));
527 }
528
529 if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER)) {
530 struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
531 unsigned n = pfb->nr_cbufs;
532 /* if we have depth/stencil, we need at least on MRT: */
533 if (pfb->zsbuf)
534 n = MAX2(1, n);
535 fd5_program_emit(ring, emit, n, pfb->cbufs);
536 }
537
538 if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
539 ir3_emit_consts(vp, ring, ctx, emit->info, dirty);
540 if (!emit->key.binning_pass)
541 ir3_emit_consts(fp, ring, ctx, emit->info, dirty);
542 }
543
544 if ((dirty & FD_DIRTY_BLEND)) {
545 struct fd5_blend_stateobj *blend = fd5_blend_stateobj(ctx->blend);
546 uint32_t i;
547
548 for (i = 0; i < A5XX_MAX_RENDER_TARGETS; i++) {
549 enum pipe_format format = pipe_surface_format(
550 ctx->batch->framebuffer.cbufs[i]);
551 bool is_int = util_format_is_pure_integer(format);
552 bool has_alpha = util_format_has_alpha(format);
553 uint32_t control = blend->rb_mrt[i].control;
554 uint32_t blend_control = blend->rb_mrt[i].blend_control_alpha;
555
556 if (is_int) {
557 control &= A5XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK;
558 // control |= A5XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY);
559 }
560
561 if (has_alpha) {
562 blend_control |= blend->rb_mrt[i].blend_control_rgb;
563 } else {
564 blend_control |= blend->rb_mrt[i].blend_control_no_alpha_rgb;
565 control &= ~A5XX_RB_MRT_CONTROL_BLEND2;
566 }
567
568 OUT_PKT4(ring, REG_A5XX_RB_MRT_CONTROL(i), 1);
569 OUT_RING(ring, control);
570
571 OUT_PKT4(ring, REG_A5XX_RB_MRT_BLEND_CONTROL(i), 1);
572 OUT_RING(ring, blend_control);
573 }
574
575 OUT_PKT4(ring, REG_A5XX_RB_BLEND_CNTL, 1);
576 OUT_RING(ring, blend->rb_blend_cntl |
577 A5XX_RB_BLEND_CNTL_SAMPLE_MASK(0xffff));
578
579 OUT_PKT4(ring, REG_A5XX_SP_BLEND_CNTL, 1);
580 OUT_RING(ring, 0x00000100);
581 }
582
583 if (dirty & FD_DIRTY_BLEND_COLOR) {
584 struct pipe_blend_color *bcolor = &ctx->blend_color;
585
586 OUT_PKT4(ring, REG_A5XX_RB_BLEND_RED, 8);
587 OUT_RING(ring, A5XX_RB_BLEND_RED_FLOAT(bcolor->color[0]) |
588 A5XX_RB_BLEND_RED_UINT(bcolor->color[0] * 0xff) |
589 A5XX_RB_BLEND_RED_SINT(bcolor->color[0] * 0x7f));
590 OUT_RING(ring, A5XX_RB_BLEND_RED_F32(bcolor->color[0]));
591 OUT_RING(ring, A5XX_RB_BLEND_GREEN_FLOAT(bcolor->color[1]) |
592 A5XX_RB_BLEND_GREEN_UINT(bcolor->color[1] * 0xff) |
593 A5XX_RB_BLEND_GREEN_SINT(bcolor->color[1] * 0x7f));
594 OUT_RING(ring, A5XX_RB_BLEND_RED_F32(bcolor->color[1]));
595 OUT_RING(ring, A5XX_RB_BLEND_BLUE_FLOAT(bcolor->color[2]) |
596 A5XX_RB_BLEND_BLUE_UINT(bcolor->color[2] * 0xff) |
597 A5XX_RB_BLEND_BLUE_SINT(bcolor->color[2] * 0x7f));
598 OUT_RING(ring, A5XX_RB_BLEND_BLUE_F32(bcolor->color[2]));
599 OUT_RING(ring, A5XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]) |
600 A5XX_RB_BLEND_ALPHA_UINT(bcolor->color[3] * 0xff) |
601 A5XX_RB_BLEND_ALPHA_SINT(bcolor->color[3] * 0x7f));
602 OUT_RING(ring, A5XX_RB_BLEND_ALPHA_F32(bcolor->color[3]));
603 }
604
605 if (dirty & FD_DIRTY_VERTTEX) {
606 if (vp->has_samp) {
607 needs_border |= emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
608 OUT_PKT4(ring, REG_A5XX_TPL1_VS_TEX_COUNT, 1);
609 OUT_RING(ring, ctx->verttex.num_textures);
610 } else {
611 dirty &= ~FD_DIRTY_VERTTEX;
612 }
613 }
614
615 if (dirty & FD_DIRTY_FRAGTEX) {
616 if (fp->has_samp) {
617 needs_border |= emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
618 OUT_PKT4(ring, REG_A5XX_TPL1_FS_TEX_COUNT, 1);
619 OUT_RING(ring, ctx->fragtex.num_textures);
620 } else {
621 dirty &= ~FD_DIRTY_FRAGTEX;
622 }
623 }
624
625 if (needs_border)
626 emit_border_color(ctx, ring);
627
628 ctx->dirty &= ~dirty;
629 }
630
631 /* emit setup at begin of new cmdstream buffer (don't rely on previous
632 * state, there could have been a context switch between ioctls):
633 */
634 void
635 fd5_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring)
636 {
637 struct fd_context *ctx = batch->ctx;
638
639 fd5_set_render_mode(ctx, ring, BYPASS);
640 fd5_cache_flush(batch, ring);
641
642 OUT_PKT4(ring, REG_A5XX_HLSQ_UPDATE_CNTL, 1);
643 OUT_RING(ring, 0xfffff);
644
645 /*
646 t7 opcode: CP_PERFCOUNTER_ACTION (50) (4 dwords)
647 0000000500024048: 70d08003 00000000 001c5000 00000005
648 t7 opcode: CP_PERFCOUNTER_ACTION (50) (4 dwords)
649 0000000500024058: 70d08003 00000010 001c7000 00000005
650
651 t7 opcode: CP_WAIT_FOR_IDLE (26) (1 dwords)
652 0000000500024068: 70268000
653 */
654
655 OUT_PKT4(ring, REG_A5XX_PC_RESTART_INDEX, 1);
656 OUT_RING(ring, 0xffffffff);
657
658 OUT_PKT4(ring, REG_A5XX_PC_RASTER_CNTL, 1);
659 OUT_RING(ring, 0x00000012);
660
661 OUT_PKT4(ring, REG_A5XX_GRAS_LRZ_CNTL, 1);
662 OUT_RING(ring, 0x00000000);
663
664 OUT_PKT4(ring, REG_A5XX_GRAS_SU_POINT_MINMAX, 2);
665 OUT_RING(ring, A5XX_GRAS_SU_POINT_MINMAX_MIN(1.0) |
666 A5XX_GRAS_SU_POINT_MINMAX_MAX(4092.0));
667 OUT_RING(ring, A5XX_GRAS_SU_POINT_SIZE(0.5));
668
669 OUT_PKT4(ring, REG_A5XX_GRAS_SU_CONSERVATIVE_RAS_CNTL, 1);
670 OUT_RING(ring, 0x00000000); /* GRAS_SU_CONSERVATIVE_RAS_CNTL */
671
672 OUT_PKT4(ring, REG_A5XX_GRAS_SC_SCREEN_SCISSOR_CNTL, 1);
673 OUT_RING(ring, 0x00000000); /* GRAS_SC_SCREEN_SCISSOR_CNTL */
674
675 OUT_PKT4(ring, REG_A5XX_SP_VS_CONFIG_MAX_CONST, 1);
676 OUT_RING(ring, 0); /* SP_VS_CONFIG_MAX_CONST */
677
678 OUT_PKT4(ring, REG_A5XX_SP_FS_CONFIG_MAX_CONST, 1);
679 OUT_RING(ring, 0); /* SP_FS_CONFIG_MAX_CONST */
680
681 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E292, 2);
682 OUT_RING(ring, 0x00000000); /* UNKNOWN_E292 */
683 OUT_RING(ring, 0x00000000); /* UNKNOWN_E293 */
684
685 OUT_PKT4(ring, REG_A5XX_RB_MODE_CNTL, 1);
686 OUT_RING(ring, 0x00000044); /* RB_MODE_CNTL */
687
688 OUT_PKT4(ring, REG_A5XX_RB_DBG_ECO_CNTL, 1);
689 OUT_RING(ring, 0x00100000); /* RB_DBG_ECO_CNTL */
690
691 OUT_PKT4(ring, REG_A5XX_VFD_MODE_CNTL, 1);
692 OUT_RING(ring, 0x00000000); /* VFD_MODE_CNTL */
693
694 OUT_PKT4(ring, REG_A5XX_PC_MODE_CNTL, 1);
695 OUT_RING(ring, 0x0000001f); /* PC_MODE_CNTL */
696
697 OUT_PKT4(ring, REG_A5XX_SP_MODE_CNTL, 1);
698 OUT_RING(ring, 0x0000001e); /* SP_MODE_CNTL */
699
700 OUT_PKT4(ring, REG_A5XX_SP_DBG_ECO_CNTL, 1);
701 OUT_RING(ring, 0x40000800); /* SP_DBG_ECO_CNTL */
702
703 OUT_PKT4(ring, REG_A5XX_TPL1_MODE_CNTL, 1);
704 OUT_RING(ring, 0x00000544); /* TPL1_MODE_CNTL */
705
706 OUT_PKT4(ring, REG_A5XX_HLSQ_TIMEOUT_THRESHOLD_0, 2);
707 OUT_RING(ring, 0x00000080); /* HLSQ_TIMEOUT_THRESHOLD_0 */
708 OUT_RING(ring, 0x00000000); /* HLSQ_TIMEOUT_THRESHOLD_1 */
709
710 OUT_PKT4(ring, REG_A5XX_VPC_DBG_ECO_CNTL, 1);
711 OUT_RING(ring, 0x00000400); /* VPC_DBG_ECO_CNTL */
712
713 OUT_PKT4(ring, REG_A5XX_HLSQ_MODE_CNTL, 1);
714 OUT_RING(ring, 0x00000001); /* HLSQ_MODE_CNTL */
715
716 OUT_PKT4(ring, REG_A5XX_VPC_MODE_CNTL, 1);
717 OUT_RING(ring, 0x00000000); /* VPC_MODE_CNTL */
718
719 /* we don't use this yet.. probably best to disable.. */
720 OUT_PKT7(ring, CP_SET_DRAW_STATE, 3);
721 OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) |
722 CP_SET_DRAW_STATE__0_DISABLE_ALL_GROUPS |
723 CP_SET_DRAW_STATE__0_GROUP_ID(0));
724 OUT_RING(ring, CP_SET_DRAW_STATE__1_ADDR_LO(0));
725 OUT_RING(ring, CP_SET_DRAW_STATE__2_ADDR_HI(0));
726
727 /* other regs not used (yet?) and always seem to have same value: */
728 OUT_PKT4(ring, REG_A5XX_GRAS_CL_CNTL, 1);
729 OUT_RING(ring, 0x00000080); /* GRAS_CL_CNTL */
730
731 OUT_PKT4(ring, REG_A5XX_GRAS_SU_CONSERVATIVE_RAS_CNTL, 1);
732 OUT_RING(ring, 0x00000000); /* GRAS_SU_CONSERVATIVE_RAS_CNTL */
733
734 OUT_PKT4(ring, REG_A5XX_GRAS_SC_BIN_CNTL, 1);
735 OUT_RING(ring, 0x00000000); /* GRAS_SC_BIN_CNTL */
736
737 OUT_PKT4(ring, REG_A5XX_GRAS_SC_BIN_CNTL, 1);
738 OUT_RING(ring, 0x00000000); /* GRAS_SC_BIN_CNTL */
739
740 OUT_PKT4(ring, REG_A5XX_VPC_FS_PRIMITIVEID_CNTL, 1);
741 OUT_RING(ring, 0x000000ff); /* VPC_FS_PRIMITIVEID_CNTL */
742
743 OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
744 OUT_RING(ring, A5XX_VPC_SO_OVERRIDE_SO_DISABLE);
745
746 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_BASE_LO(0), 3);
747 OUT_RING(ring, 0x00000000); /* VPC_SO_BUFFER_BASE_LO_0 */
748 OUT_RING(ring, 0x00000000); /* VPC_SO_BUFFER_BASE_HI_0 */
749 OUT_RING(ring, 0x00000000); /* VPC_SO_BUFFER_SIZE_0 */
750
751 OUT_PKT4(ring, REG_A5XX_VPC_SO_FLUSH_BASE_LO(0), 2);
752 OUT_RING(ring, 0x00000000); /* VPC_SO_FLUSH_BASE_LO_0 */
753 OUT_RING(ring, 0x00000000); /* VPC_SO_FLUSH_BASE_HI_0 */
754
755 OUT_PKT4(ring, REG_A5XX_PC_GS_PARAM, 1);
756 OUT_RING(ring, 0x00000000); /* PC_GS_PARAM */
757
758 OUT_PKT4(ring, REG_A5XX_PC_HS_PARAM, 1);
759 OUT_RING(ring, 0x00000000); /* PC_HS_PARAM */
760
761 OUT_PKT4(ring, REG_A5XX_TPL1_TP_FS_ROTATION_CNTL, 1);
762 OUT_RING(ring, 0x00000000); /* TPL1_TP_FS_ROTATION_CNTL */
763
764 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E001, 1);
765 OUT_RING(ring, 0x00000000); /* UNKNOWN_E001 */
766
767 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E004, 1);
768 OUT_RING(ring, 0x00000000); /* UNKNOWN_E004 */
769
770 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E093, 1);
771 OUT_RING(ring, 0x00000000); /* UNKNOWN_E093 */
772
773 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E1C7, 1);
774 OUT_RING(ring, 0x00000000); /* UNKNOWN_E1C7 */
775
776 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E29A, 1);
777 OUT_RING(ring, 0x00ffff00); /* UNKNOWN_E29A */
778
779 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUF_CNTL, 1);
780 OUT_RING(ring, 0x00000000); /* VPC_SO_BUF_CNTL */
781
782 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_OFFSET(0), 1);
783 OUT_RING(ring, 0x00000000); /* UNKNOWN_E2AB */
784
785 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E389, 1);
786 OUT_RING(ring, 0x00000000); /* UNKNOWN_E389 */
787
788 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E38D, 1);
789 OUT_RING(ring, 0x00000000); /* UNKNOWN_E38D */
790
791 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E5AB, 1);
792 OUT_RING(ring, 0x00000000); /* UNKNOWN_E5AB */
793
794 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E5C2, 1);
795 OUT_RING(ring, 0x00000000); /* UNKNOWN_E5C2 */
796
797 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_BASE_LO(1), 3);
798 OUT_RING(ring, 0x00000000);
799 OUT_RING(ring, 0x00000000);
800 OUT_RING(ring, 0x00000000);
801
802 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_OFFSET(1), 6);
803 OUT_RING(ring, 0x00000000);
804 OUT_RING(ring, 0x00000000);
805 OUT_RING(ring, 0x00000000);
806 OUT_RING(ring, 0x00000000);
807 OUT_RING(ring, 0x00000000);
808 OUT_RING(ring, 0x00000000);
809
810 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_OFFSET(2), 6);
811 OUT_RING(ring, 0x00000000);
812 OUT_RING(ring, 0x00000000);
813 OUT_RING(ring, 0x00000000);
814 OUT_RING(ring, 0x00000000);
815 OUT_RING(ring, 0x00000000);
816 OUT_RING(ring, 0x00000000);
817
818 OUT_PKT4(ring, REG_A5XX_VPC_SO_BUFFER_OFFSET(3), 3);
819 OUT_RING(ring, 0x00000000);
820 OUT_RING(ring, 0x00000000);
821 OUT_RING(ring, 0x00000000);
822
823 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E5DB, 1);
824 OUT_RING(ring, 0x00000000);
825
826 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E600, 1);
827 OUT_RING(ring, 0x00000000);
828
829 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E640, 1);
830 OUT_RING(ring, 0x00000000);
831
832 OUT_PKT4(ring, REG_A5XX_TPL1_VS_TEX_COUNT, 4);
833 OUT_RING(ring, 0x00000000);
834 OUT_RING(ring, 0x00000000);
835 OUT_RING(ring, 0x00000000);
836 OUT_RING(ring, 0x00000000);
837
838 OUT_PKT4(ring, REG_A5XX_TPL1_FS_TEX_COUNT, 2);
839 OUT_RING(ring, 0x00000000);
840 OUT_RING(ring, 0x00000000);
841
842 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7C0, 3);
843 OUT_RING(ring, 0x00000000);
844 OUT_RING(ring, 0x00000000);
845 OUT_RING(ring, 0x00000000);
846
847 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7C5, 3);
848 OUT_RING(ring, 0x00000000);
849 OUT_RING(ring, 0x00000000);
850 OUT_RING(ring, 0x00000000);
851
852 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7CA, 3);
853 OUT_RING(ring, 0x00000000);
854 OUT_RING(ring, 0x00000000);
855 OUT_RING(ring, 0x00000000);
856
857 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7CF, 3);
858 OUT_RING(ring, 0x00000000);
859 OUT_RING(ring, 0x00000000);
860 OUT_RING(ring, 0x00000000);
861
862 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7D4, 3);
863 OUT_RING(ring, 0x00000000);
864 OUT_RING(ring, 0x00000000);
865 OUT_RING(ring, 0x00000000);
866
867 OUT_PKT4(ring, REG_A5XX_UNKNOWN_E7D9, 3);
868 OUT_RING(ring, 0x00000000);
869 OUT_RING(ring, 0x00000000);
870 OUT_RING(ring, 0x00000000);
871
872 // TODO hacks.. these should not be hardcoded:
873 OUT_PKT4(ring, REG_A5XX_GRAS_SC_CNTL, 1);
874 OUT_RING(ring, 0x00000008); /* GRAS_SC_CNTL */
875
876 fd_hw_query_enable(batch, ring);
877 }
878
879 static void
880 fd5_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
881 {
882 __OUT_IB5(ring, target);
883 }
884
885 void
886 fd5_emit_init(struct pipe_context *pctx)
887 {
888 struct fd_context *ctx = fd_context(pctx);
889 ctx->emit_const = fd5_emit_const;
890 ctx->emit_const_bo = fd5_emit_const_bo;
891 ctx->emit_ib = fd5_emit_ib;
892 }