projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
freedreno/a4xx: fix dst_alpha blend for RGBX render targets
[mesa.git] / src / gallium / drivers / freedreno / a4xx / fd4_emit.c
1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3 /*
4 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 *
25 * Authors:
26 * Rob Clark <robclark@freedesktop.org>
27 */
28
29 #include "pipe/p_state.h"
30 #include "util/u_string.h"
31 #include "util/u_memory.h"
32 #include "util/u_helpers.h"
33 #include "util/u_format.h"
34
35 #include "freedreno_resource.h"
36
37 #include "fd4_emit.h"
38 #include "fd4_blend.h"
39 #include "fd4_context.h"
40 #include "fd4_program.h"
41 #include "fd4_rasterizer.h"
42 #include "fd4_texture.h"
43 #include "fd4_format.h"
44 #include "fd4_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 void
56 fd4_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_PKT3(ring, CP_LOAD_STATE, 2 + 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 dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
87 }
88 for (i = 0; i < sz; i++) {
89 OUT_RING(ring, dwords[i]);
90 }
91 }
92
93 static void
94 fd4_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
95 uint32_t regid, uint32_t num, struct fd_bo **bos, uint32_t *offsets)
96 {
97 uint32_t i;
98
99 debug_assert((regid % 4) == 0);
100 debug_assert((num % 4) == 0);
101
102 OUT_PKT3(ring, CP_LOAD_STATE, 2 + num);
103 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
104 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
105 CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
106 CP_LOAD_STATE_0_NUM_UNIT(num/4));
107 OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
108 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
109
110 for (i = 0; i < num; i++) {
111 if (bos[i]) {
112 if (write) {
113 OUT_RELOCW(ring, bos[i], offsets[i], 0, 0);
114 } else {
115 OUT_RELOC(ring, bos[i], offsets[i], 0, 0);
116 }
117 } else {
118 OUT_RING(ring, 0xbad00000 | (i << 16));
119 }
120 }
121 }
122
123 static void
124 emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
125 enum adreno_state_block sb, struct fd_texture_stateobj *tex)
126 {
127 static const uint32_t bcolor_reg[] = {
128 [SB_VERT_TEX] = REG_A4XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
129 [SB_FRAG_TEX] = REG_A4XX_TPL1_TP_FS_BORDER_COLOR_BASE_ADDR,
130 };
131 struct fd4_context *fd4_ctx = fd4_context(ctx);
132 unsigned i, off;
133 void *ptr;
134
135 u_upload_alloc(fd4_ctx->border_color_uploader,
136 0, 2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE, &off,
137 &fd4_ctx->border_color_buf,
138 &ptr);
139
140 fd_setup_border_colors(tex, ptr, 0);
141
142 if (tex->num_samplers > 0) {
143 int num_samplers;
144
145 /* not sure if this is an a420.0 workaround, but we seem
146 * to need to emit these in pairs.. emit a final dummy
147 * entry if odd # of samplers:
148 */
149 num_samplers = align(tex->num_samplers, 2);
150
151 /* output sampler state: */
152 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * num_samplers));
153 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
154 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
155 CP_LOAD_STATE_0_STATE_BLOCK(sb) |
156 CP_LOAD_STATE_0_NUM_UNIT(num_samplers));
157 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
158 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
159 for (i = 0; i < tex->num_samplers; i++) {
160 static const struct fd4_sampler_stateobj dummy_sampler = {};
161 const struct fd4_sampler_stateobj *sampler = tex->samplers[i] ?
162 fd4_sampler_stateobj(tex->samplers[i]) :
163 &dummy_sampler;
164 OUT_RING(ring, sampler->texsamp0);
165 OUT_RING(ring, sampler->texsamp1);
166 }
167
168 for (; i < num_samplers; i++) {
169 OUT_RING(ring, 0x00000000);
170 OUT_RING(ring, 0x00000000);
171 }
172 }
173
174 if (tex->num_textures > 0) {
175 /* emit texture state: */
176 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * tex->num_textures));
177 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
178 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
179 CP_LOAD_STATE_0_STATE_BLOCK(sb) |
180 CP_LOAD_STATE_0_NUM_UNIT(tex->num_textures));
181 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
182 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
183 for (i = 0; i < tex->num_textures; i++) {
184 static const struct fd4_pipe_sampler_view dummy_view = {};
185 const struct fd4_pipe_sampler_view *view = tex->textures[i] ?
186 fd4_pipe_sampler_view(tex->textures[i]) :
187 &dummy_view;
188 unsigned start = fd_sampler_first_level(&view->base);
189
190 OUT_RING(ring, view->texconst0);
191 OUT_RING(ring, view->texconst1);
192 OUT_RING(ring, view->texconst2);
193 OUT_RING(ring, view->texconst3);
194 if (view->base.texture) {
195 struct fd_resource *rsc = fd_resource(view->base.texture);
196 uint32_t offset = fd_resource_offset(rsc, start, 0);
197 OUT_RELOC(ring, rsc->bo, offset, view->textconst4, 0);
198 } else {
199 OUT_RING(ring, 0x00000000);
200 }
201 OUT_RING(ring, 0x00000000);
202 OUT_RING(ring, 0x00000000);
203 OUT_RING(ring, 0x00000000);
204 }
205 }
206
207 OUT_PKT0(ring, bcolor_reg[sb], 1);
208 OUT_RELOC(ring, fd_resource(fd4_ctx->border_color_buf)->bo, off, 0, 0);
209
210 u_upload_unmap(fd4_ctx->border_color_uploader);
211 }
212
213 /* emit texture state for mem->gmem restore operation.. eventually it would
214 * be good to get rid of this and use normal CSO/etc state for more of these
215 * special cases..
216 */
217 void
218 fd4_emit_gmem_restore_tex(struct fd_ringbuffer *ring, unsigned nr_bufs,
219 struct pipe_surface **bufs)
220 {
221 unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS];
222 int i;
223
224 for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
225 mrt_comp[i] = (i < nr_bufs) ? 0xf : 0;
226 }
227
228 /* output sampler state: */
229 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * nr_bufs));
230 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
231 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
232 CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
233 CP_LOAD_STATE_0_NUM_UNIT(nr_bufs));
234 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
235 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
236 for (i = 0; i < nr_bufs; i++) {
237 OUT_RING(ring, A4XX_TEX_SAMP_0_XY_MAG(A4XX_TEX_NEAREST) |
238 A4XX_TEX_SAMP_0_XY_MIN(A4XX_TEX_NEAREST) |
239 A4XX_TEX_SAMP_0_WRAP_S(A4XX_TEX_CLAMP_TO_EDGE) |
240 A4XX_TEX_SAMP_0_WRAP_T(A4XX_TEX_CLAMP_TO_EDGE) |
241 A4XX_TEX_SAMP_0_WRAP_R(A4XX_TEX_REPEAT));
242 OUT_RING(ring, 0x00000000);
243 }
244
245 /* emit texture state: */
246 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * nr_bufs));
247 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
248 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
249 CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
250 CP_LOAD_STATE_0_NUM_UNIT(nr_bufs));
251 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
252 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
253 for (i = 0; i < nr_bufs; i++) {
254 if (bufs[i]) {
255 struct fd_resource *rsc = fd_resource(bufs[i]->texture);
256 /* note: PIPE_BUFFER disallowed for surfaces */
257 unsigned lvl = bufs[i]->u.tex.level;
258 struct fd_resource_slice *slice = fd_resource_slice(rsc, lvl);
259 uint32_t offset = fd_resource_offset(rsc, lvl, bufs[i]->u.tex.first_layer);
260 enum pipe_format format = fd4_gmem_restore_format(bufs[i]->format);
261
262 /* The restore blit_zs shader expects stencil in sampler 0,
263 * and depth in sampler 1
264 */
265 if (rsc->stencil && (i == 0)) {
266 rsc = rsc->stencil;
267 format = fd4_gmem_restore_format(rsc->base.b.format);
268 }
269
270 /* z32 restore is accomplished using depth write. If there is
271 * no stencil component (ie. PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
272 * then no render target:
273 *
274 * (The same applies for z32_s8x24, since for stencil sampler
275 * state the above 'if' will replace 'format' with s8)
276 */
277 if ((format == PIPE_FORMAT_Z32_FLOAT) ||
278 (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT))
279 mrt_comp[i] = 0;
280
281 debug_assert(bufs[i]->u.tex.first_layer == bufs[i]->u.tex.last_layer);
282
283 OUT_RING(ring, A4XX_TEX_CONST_0_FMT(fd4_pipe2tex(format)) |
284 A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
285 fd4_tex_swiz(format, PIPE_SWIZZLE_RED, PIPE_SWIZZLE_GREEN,
286 PIPE_SWIZZLE_BLUE, PIPE_SWIZZLE_ALPHA));
287 OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(bufs[i]->width) |
288 A4XX_TEX_CONST_1_HEIGHT(bufs[i]->height));
289 OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp));
290 OUT_RING(ring, 0x00000000);
291 OUT_RELOC(ring, rsc->bo, offset, 0, 0);
292 OUT_RING(ring, 0x00000000);
293 OUT_RING(ring, 0x00000000);
294 OUT_RING(ring, 0x00000000);
295 } else {
296 OUT_RING(ring, A4XX_TEX_CONST_0_FMT(0) |
297 A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
298 A4XX_TEX_CONST_0_SWIZ_X(A4XX_TEX_ONE) |
299 A4XX_TEX_CONST_0_SWIZ_Y(A4XX_TEX_ONE) |
300 A4XX_TEX_CONST_0_SWIZ_Z(A4XX_TEX_ONE) |
301 A4XX_TEX_CONST_0_SWIZ_W(A4XX_TEX_ONE));
302 OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(0) |
303 A4XX_TEX_CONST_1_HEIGHT(0));
304 OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(0));
305 OUT_RING(ring, 0x00000000);
306 OUT_RING(ring, 0x00000000);
307 OUT_RING(ring, 0x00000000);
308 OUT_RING(ring, 0x00000000);
309 OUT_RING(ring, 0x00000000);
310 }
311 }
312
313 OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
314 OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
315 A4XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
316 A4XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
317 A4XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
318 A4XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
319 A4XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
320 A4XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
321 A4XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
322 }
323
324 void
325 fd4_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd4_emit *emit)
326 {
327 int32_t i, j, last = -1;
328 uint32_t total_in = 0;
329 const struct fd_vertex_state *vtx = emit->vtx;
330 struct ir3_shader_variant *vp = fd4_emit_get_vp(emit);
331 unsigned vertex_regid = regid(63, 0);
332 unsigned instance_regid = regid(63, 0);
333 unsigned vtxcnt_regid = regid(63, 0);
334
335 /* Note that sysvals come *after* normal inputs: */
336 for (i = 0; i < vp->inputs_count; i++) {
337 if (!vp->inputs[i].compmask)
338 continue;
339 if (vp->inputs[i].sysval) {
340 switch(vp->inputs[i].slot) {
341 case SYSTEM_VALUE_BASE_VERTEX:
342 /* handled elsewhere */
343 break;
344 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
345 vertex_regid = vp->inputs[i].regid;
346 break;
347 case SYSTEM_VALUE_INSTANCE_ID:
348 instance_regid = vp->inputs[i].regid;
349 break;
350 case SYSTEM_VALUE_VERTEX_CNT:
351 vtxcnt_regid = vp->inputs[i].regid;
352 break;
353 default:
354 unreachable("invalid system value");
355 break;
356 }
357 } else if (i < vtx->vtx->num_elements) {
358 last = i;
359 }
360 }
361
362 for (i = 0, j = 0; i <= last; i++) {
363 assert(!vp->inputs[i].sysval);
364 if (vp->inputs[i].compmask) {
365 struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
366 const struct pipe_vertex_buffer *vb =
367 &vtx->vertexbuf.vb[elem->vertex_buffer_index];
368 struct fd_resource *rsc = fd_resource(vb->buffer);
369 enum pipe_format pfmt = elem->src_format;
370 enum a4xx_vtx_fmt fmt = fd4_pipe2vtx(pfmt);
371 bool switchnext = (i != last) ||
372 (vertex_regid != regid(63, 0)) ||
373 (instance_regid != regid(63, 0)) ||
374 (vtxcnt_regid != regid(63, 0));
375 bool isint = util_format_is_pure_integer(pfmt);
376 uint32_t fs = util_format_get_blocksize(pfmt);
377 uint32_t off = vb->buffer_offset + elem->src_offset;
378 uint32_t size = fd_bo_size(rsc->bo) - off;
379 debug_assert(fmt != ~0);
380
381 OUT_PKT0(ring, REG_A4XX_VFD_FETCH(j), 4);
382 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_0_FETCHSIZE(fs - 1) |
383 A4XX_VFD_FETCH_INSTR_0_BUFSTRIDE(vb->stride) |
384 COND(elem->instance_divisor, A4XX_VFD_FETCH_INSTR_0_INSTANCED) |
385 COND(switchnext, A4XX_VFD_FETCH_INSTR_0_SWITCHNEXT));
386 OUT_RELOC(ring, rsc->bo, off, 0, 0);
387 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_2_SIZE(size));
388 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_3_STEPRATE(MAX2(1, elem->instance_divisor)));
389
390 OUT_PKT0(ring, REG_A4XX_VFD_DECODE_INSTR(j), 1);
391 OUT_RING(ring, A4XX_VFD_DECODE_INSTR_CONSTFILL |
392 A4XX_VFD_DECODE_INSTR_WRITEMASK(vp->inputs[i].compmask) |
393 A4XX_VFD_DECODE_INSTR_FORMAT(fmt) |
394 A4XX_VFD_DECODE_INSTR_SWAP(fd4_pipe2swap(pfmt)) |
395 A4XX_VFD_DECODE_INSTR_REGID(vp->inputs[i].regid) |
396 A4XX_VFD_DECODE_INSTR_SHIFTCNT(fs) |
397 A4XX_VFD_DECODE_INSTR_LASTCOMPVALID |
398 COND(isint, A4XX_VFD_DECODE_INSTR_INT) |
399 COND(switchnext, A4XX_VFD_DECODE_INSTR_SWITCHNEXT));
400
401 total_in += vp->inputs[i].ncomp;
402 j++;
403 }
404 }
405
406 /* hw doesn't like to be configured for zero vbo's, it seems: */
407 if (last < 0) {
408 /* just recycle the shader bo, we just need to point to *something*
409 * valid:
410 */
411 struct fd_bo *dummy_vbo = vp->bo;
412 bool switchnext = (vertex_regid != regid(63, 0)) ||
413 (instance_regid != regid(63, 0)) ||
414 (vtxcnt_regid != regid(63, 0));
415
416 OUT_PKT0(ring, REG_A4XX_VFD_FETCH(0), 4);
417 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_0_FETCHSIZE(0) |
418 A4XX_VFD_FETCH_INSTR_0_BUFSTRIDE(0) |
419 COND(switchnext, A4XX_VFD_FETCH_INSTR_0_SWITCHNEXT));
420 OUT_RELOC(ring, dummy_vbo, 0, 0, 0);
421 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_2_SIZE(1));
422 OUT_RING(ring, A4XX_VFD_FETCH_INSTR_3_STEPRATE(1));
423
424 OUT_PKT0(ring, REG_A4XX_VFD_DECODE_INSTR(0), 1);
425 OUT_RING(ring, A4XX_VFD_DECODE_INSTR_CONSTFILL |
426 A4XX_VFD_DECODE_INSTR_WRITEMASK(0x1) |
427 A4XX_VFD_DECODE_INSTR_FORMAT(VFMT4_8_UNORM) |
428 A4XX_VFD_DECODE_INSTR_SWAP(XYZW) |
429 A4XX_VFD_DECODE_INSTR_REGID(regid(0,0)) |
430 A4XX_VFD_DECODE_INSTR_SHIFTCNT(1) |
431 A4XX_VFD_DECODE_INSTR_LASTCOMPVALID |
432 COND(switchnext, A4XX_VFD_DECODE_INSTR_SWITCHNEXT));
433
434 total_in = 1;
435 j = 1;
436 }
437
438 OUT_PKT0(ring, REG_A4XX_VFD_CONTROL_0, 5);
439 OUT_RING(ring, A4XX_VFD_CONTROL_0_TOTALATTRTOVS(total_in) |
440 0xa0000 | /* XXX */
441 A4XX_VFD_CONTROL_0_STRMDECINSTRCNT(j) |
442 A4XX_VFD_CONTROL_0_STRMFETCHINSTRCNT(j));
443 OUT_RING(ring, A4XX_VFD_CONTROL_1_MAXSTORAGE(129) | // XXX
444 A4XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
445 A4XX_VFD_CONTROL_1_REGID4INST(instance_regid));
446 OUT_RING(ring, 0x00000000); /* XXX VFD_CONTROL_2 */
447 OUT_RING(ring, A4XX_VFD_CONTROL_3_REGID_VTXCNT(vtxcnt_regid));
448 OUT_RING(ring, 0x00000000); /* XXX VFD_CONTROL_4 */
449
450 /* cache invalidate, otherwise vertex fetch could see
451 * stale vbo contents:
452 */
453 OUT_PKT0(ring, REG_A4XX_UCHE_INVALIDATE0, 2);
454 OUT_RING(ring, 0x00000000);
455 OUT_RING(ring, 0x00000012);
456 }
457
458 void
459 fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
460 struct fd4_emit *emit)
461 {
462 struct ir3_shader_variant *vp = fd4_emit_get_vp(emit);
463 struct ir3_shader_variant *fp = fd4_emit_get_fp(emit);
464 uint32_t dirty = emit->dirty;
465
466 emit_marker(ring, 5);
467
468 if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
469 struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
470 unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
471
472 for (unsigned i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
473 mrt_comp[i] = ((i < pfb->nr_cbufs) && pfb->cbufs[i]) ? 0xf : 0;
474 }
475
476 OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
477 OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
478 A4XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
479 A4XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
480 A4XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
481 A4XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
482 A4XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
483 A4XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
484 A4XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
485 }
486
487 if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) && !emit->key.binning_pass) {
488 uint32_t val = fd4_zsa_stateobj(ctx->zsa)->rb_render_control;
489
490 /* I suppose if we needed to (which I don't *think* we need
491 * to), we could emit this for binning pass too. But we
492 * would need to keep a different patch-list for binning
493 * vs render pass.
494 */
495
496 OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
497 OUT_RINGP(ring, val, &fd4_context(ctx)->rbrc_patches);
498 }
499
500 if (dirty & FD_DIRTY_ZSA) {
501 struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
502
503 OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
504 OUT_RING(ring, zsa->rb_alpha_control);
505
506 OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
507 OUT_RING(ring, zsa->rb_stencil_control);
508 OUT_RING(ring, zsa->rb_stencil_control2);
509 }
510
511 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
512 struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
513 struct pipe_stencil_ref *sr = &ctx->stencil_ref;
514
515 OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
516 OUT_RING(ring, zsa->rb_stencilrefmask |
517 A4XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
518 OUT_RING(ring, zsa->rb_stencilrefmask_bf |
519 A4XX_RB_STENCILREFMASK_BF_STENCILREF(sr->ref_value[1]));
520 }
521
522 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) {
523 struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
524 bool fragz = fp->has_kill | fp->writes_pos;
525
526 OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
527 OUT_RING(ring, zsa->rb_depth_control |
528 COND(fragz, A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE));
529
530 /* maybe this register/bitfield needs a better name.. this
531 * appears to be just disabling early-z
532 */
533 OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
534 OUT_RING(ring, zsa->gras_alpha_control |
535 COND(fragz, A4XX_GRAS_ALPHA_CONTROL_ALPHA_TEST_ENABLE));
536 }
537
538 if (dirty & FD_DIRTY_RASTERIZER) {
539 struct fd4_rasterizer_stateobj *rasterizer =
540 fd4_rasterizer_stateobj(ctx->rasterizer);
541
542 OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
543 OUT_RING(ring, rasterizer->gras_su_mode_control |
544 A4XX_GRAS_SU_MODE_CONTROL_RENDERING_PASS);
545
546 OUT_PKT0(ring, REG_A4XX_GRAS_SU_POINT_MINMAX, 2);
547 OUT_RING(ring, rasterizer->gras_su_point_minmax);
548 OUT_RING(ring, rasterizer->gras_su_point_size);
549
550 OUT_PKT0(ring, REG_A4XX_GRAS_SU_POLY_OFFSET_SCALE, 2);
551 OUT_RING(ring, rasterizer->gras_su_poly_offset_scale);
552 OUT_RING(ring, rasterizer->gras_su_poly_offset_offset);
553
554 OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
555 OUT_RING(ring, rasterizer->gras_cl_clip_cntl);
556 }
557
558 /* NOTE: since primitive_restart is not actually part of any
559 * state object, we need to make sure that we always emit
560 * PRIM_VTX_CNTL.. either that or be more clever and detect
561 * when it changes.
562 */
563 if (emit->info) {
564 const struct pipe_draw_info *info = emit->info;
565 uint32_t val = fd4_rasterizer_stateobj(ctx->rasterizer)
566 ->pc_prim_vtx_cntl;
567
568 if (info->indexed && info->primitive_restart)
569 val |= A4XX_PC_PRIM_VTX_CNTL_PRIMITIVE_RESTART;
570
571 val |= COND(vp->writes_psize, A4XX_PC_PRIM_VTX_CNTL_PSIZE);
572
573 if (fp->total_in > 0) {
574 uint32_t varout = align(fp->total_in, 16) / 16;
575 if (varout > 1)
576 varout = align(varout, 2);
577 val |= A4XX_PC_PRIM_VTX_CNTL_VAROUT(varout);
578 }
579
580 OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 2);
581 OUT_RING(ring, val);
582 OUT_RING(ring, 0x12); /* XXX UNKNOWN_21C5 */
583 }
584
585 if (dirty & FD_DIRTY_SCISSOR) {
586 struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
587
588 OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
589 OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(scissor->maxx - 1) |
590 A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(scissor->maxy - 1));
591 OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(scissor->minx) |
592 A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(scissor->miny));
593
594 ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, scissor->minx);
595 ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, scissor->miny);
596 ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, scissor->maxx);
597 ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, scissor->maxy);
598 }
599
600 if (dirty & FD_DIRTY_VIEWPORT) {
601 fd_wfi(ctx, ring);
602 OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
603 OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0(ctx->viewport.translate[0]));
604 OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0(ctx->viewport.scale[0]));
605 OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0(ctx->viewport.translate[1]));
606 OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(ctx->viewport.scale[1]));
607 OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(ctx->viewport.translate[2]));
608 OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(ctx->viewport.scale[2]));
609 }
610
611 if (dirty & FD_DIRTY_PROG) {
612 struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
613 fd4_program_emit(ring, emit, pfb->nr_cbufs, pfb->cbufs);
614 }
615
616 if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
617 ir3_emit_consts(vp, ring, emit->info, dirty);
618 if (!emit->key.binning_pass)
619 ir3_emit_consts(fp, ring, emit->info, dirty);
620 /* mark clean after emitting consts: */
621 ctx->prog.dirty = 0;
622 }
623
624 if ((dirty & FD_DIRTY_BLEND)) {
625 struct fd4_blend_stateobj *blend = fd4_blend_stateobj(ctx->blend);
626 uint32_t i;
627
628 for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
629 enum pipe_format format = pipe_surface_format(
630 ctx->framebuffer.cbufs[i]);
631 bool has_alpha = util_format_has_alpha(format);
632 uint32_t control = blend->rb_mrt[i].control;
633 uint32_t blend_control = blend->rb_mrt[i].blend_control_alpha;
634
635 if (has_alpha) {
636 blend_control |= blend->rb_mrt[i].blend_control_rgb;
637 } else {
638 blend_control |= blend->rb_mrt[i].blend_control_no_alpha_rgb;
639 control &= ~A4XX_RB_MRT_CONTROL_BLEND2;
640 }
641
642 OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
643 OUT_RING(ring, control);
644
645 OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
646 OUT_RING(ring, blend_control);
647 }
648
649 OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
650 OUT_RING(ring, blend->rb_fs_output |
651 A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
652 }
653
654 if (dirty & FD_DIRTY_BLEND_COLOR) {
655 struct pipe_blend_color *bcolor = &ctx->blend_color;
656 OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 8);
657 OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(bcolor->color[0] * 65535.0) |
658 A4XX_RB_BLEND_RED_FLOAT(bcolor->color[0]));
659 OUT_RING(ring, A4XX_RB_BLEND_RED_F32(bcolor->color[0]));
660 OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(bcolor->color[1] * 65535.0) |
661 A4XX_RB_BLEND_GREEN_FLOAT(bcolor->color[1]));
662 OUT_RING(ring, A4XX_RB_BLEND_GREEN_F32(bcolor->color[1]));
663 OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(bcolor->color[2] * 65535.0) |
664 A4XX_RB_BLEND_BLUE_FLOAT(bcolor->color[2]));
665 OUT_RING(ring, A4XX_RB_BLEND_BLUE_F32(bcolor->color[2]));
666 OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(bcolor->color[3] * 65535.0) |
667 A4XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]));
668 OUT_RING(ring, A4XX_RB_BLEND_ALPHA_F32(bcolor->color[3]));
669 }
670
671 if (dirty & FD_DIRTY_VERTTEX) {
672 if (vp->has_samp)
673 emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
674 else
675 dirty &= ~FD_DIRTY_VERTTEX;
676 }
677
678 if (dirty & FD_DIRTY_FRAGTEX) {
679 if (fp->has_samp)
680 emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
681 else
682 dirty &= ~FD_DIRTY_FRAGTEX;
683 }
684
685 ctx->dirty &= ~dirty;
686 }
687
688 /* emit setup at begin of new cmdstream buffer (don't rely on previous
689 * state, there could have been a context switch between ioctls):
690 */
691 void
692 fd4_emit_restore(struct fd_context *ctx)
693 {
694 struct fd4_context *fd4_ctx = fd4_context(ctx);
695 struct fd_ringbuffer *ring = ctx->ring;
696
697 OUT_PKT0(ring, REG_A4XX_RBBM_PERFCTR_CTL, 1);
698 OUT_RING(ring, 0x00000001);
699
700 OUT_PKT0(ring, REG_A4XX_GRAS_DEBUG_ECO_CONTROL, 1);
701 OUT_RING(ring, 0x00000000);
702
703 OUT_PKT0(ring, REG_A4XX_SP_MODE_CONTROL, 1);
704 OUT_RING(ring, 0x00000006);
705
706 OUT_PKT0(ring, REG_A4XX_TPL1_TP_MODE_CONTROL, 1);
707 OUT_RING(ring, 0x0000003a);
708
709 OUT_PKT0(ring, REG_A4XX_UNKNOWN_0D01, 1);
710 OUT_RING(ring, 0x00000001);
711
712 OUT_PKT0(ring, REG_A4XX_UNKNOWN_0E42, 1);
713 OUT_RING(ring, 0x00000000);
714
715 OUT_PKT0(ring, REG_A4XX_UCHE_CACHE_WAYS_VFD, 1);
716 OUT_RING(ring, 0x00000007);
717
718 OUT_PKT0(ring, REG_A4XX_UCHE_CACHE_MODE_CONTROL, 1);
719 OUT_RING(ring, 0x00000000);
720
721 OUT_PKT0(ring, REG_A4XX_UCHE_INVALIDATE0, 2);
722 OUT_RING(ring, 0x00000000);
723 OUT_RING(ring, 0x00000012);
724
725 OUT_PKT0(ring, REG_A4XX_HLSQ_MODE_CONTROL, 1);
726 OUT_RING(ring, 0x00000000);
727
728 OUT_PKT0(ring, REG_A4XX_UNKNOWN_0CC5, 1);
729 OUT_RING(ring, 0x00000006);
730
731 OUT_PKT0(ring, REG_A4XX_UNKNOWN_0CC6, 1);
732 OUT_RING(ring, 0x00000000);
733
734 OUT_PKT0(ring, REG_A4XX_UNKNOWN_0EC2, 1);
735 OUT_RING(ring, 0x00040000);
736
737 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2001, 1);
738 OUT_RING(ring, 0x00000000);
739
740 OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
741 OUT_RING(ring, 0x00001000);
742
743 OUT_PKT0(ring, REG_A4XX_UNKNOWN_20EF, 1);
744 OUT_RING(ring, 0x00000000);
745
746 OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 4);
747 OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(0) |
748 A4XX_RB_BLEND_RED_FLOAT(0.0));
749 OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(0) |
750 A4XX_RB_BLEND_GREEN_FLOAT(0.0));
751 OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(0) |
752 A4XX_RB_BLEND_BLUE_FLOAT(0.0));
753 OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(0x7fff) |
754 A4XX_RB_BLEND_ALPHA_FLOAT(1.0));
755
756 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2152, 1);
757 OUT_RING(ring, 0x00000000);
758
759 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2153, 1);
760 OUT_RING(ring, 0x00000000);
761
762 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2154, 1);
763 OUT_RING(ring, 0x00000000);
764
765 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2155, 1);
766 OUT_RING(ring, 0x00000000);
767
768 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2156, 1);
769 OUT_RING(ring, 0x00000000);
770
771 OUT_PKT0(ring, REG_A4XX_UNKNOWN_2157, 1);
772 OUT_RING(ring, 0x00000000);
773
774 OUT_PKT0(ring, REG_A4XX_UNKNOWN_21C3, 1);
775 OUT_RING(ring, 0x0000001d);
776
777 OUT_PKT0(ring, REG_A4XX_PC_GS_PARAM, 1);
778 OUT_RING(ring, 0x00000000);
779
780 OUT_PKT0(ring, REG_A4XX_UNKNOWN_21E6, 1);
781 OUT_RING(ring, 0x00000001);
782
783 OUT_PKT0(ring, REG_A4XX_PC_HS_PARAM, 1);
784 OUT_RING(ring, 0x00000000);
785
786 OUT_PKT0(ring, REG_A4XX_UNKNOWN_22D7, 1);
787 OUT_RING(ring, 0x00000000);
788
789 OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_OFFSET, 1);
790 OUT_RING(ring, 0x00000000);
791
792 OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_COUNT, 1);
793 OUT_RING(ring, A4XX_TPL1_TP_TEX_COUNT_VS(16) |
794 A4XX_TPL1_TP_TEX_COUNT_HS(0) |
795 A4XX_TPL1_TP_TEX_COUNT_DS(0) |
796 A4XX_TPL1_TP_TEX_COUNT_GS(0));
797
798 OUT_PKT0(ring, REG_A4XX_TPL1_TP_FS_TEX_COUNT, 1);
799 OUT_RING(ring, 16);
800
801 /* we don't use this yet.. probably best to disable.. */
802 OUT_PKT3(ring, CP_SET_DRAW_STATE, 2);
803 OUT_RING(ring, CP_SET_DRAW_STATE_0_COUNT(0) |
804 CP_SET_DRAW_STATE_0_DISABLE_ALL_GROUPS |
805 CP_SET_DRAW_STATE_0_GROUP_ID(0));
806 OUT_RING(ring, CP_SET_DRAW_STATE_1_ADDR(0));
807
808 OUT_PKT0(ring, REG_A4XX_SP_VS_PVT_MEM_PARAM, 2);
809 OUT_RING(ring, 0x08000001); /* SP_VS_PVT_MEM_PARAM */
810 OUT_RELOC(ring, fd4_ctx->vs_pvt_mem, 0,0,0); /* SP_VS_PVT_MEM_ADDR */
811
812 OUT_PKT0(ring, REG_A4XX_SP_FS_PVT_MEM_PARAM, 2);
813 OUT_RING(ring, 0x08000001); /* SP_FS_PVT_MEM_PARAM */
814 OUT_RELOC(ring, fd4_ctx->fs_pvt_mem, 0,0,0); /* SP_FS_PVT_MEM_ADDR */
815
816 OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
817 OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
818 A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
819 A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
820 A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
821
822 OUT_PKT0(ring, REG_A4XX_RB_MSAA_CONTROL, 1);
823 OUT_RING(ring, A4XX_RB_MSAA_CONTROL_DISABLE |
824 A4XX_RB_MSAA_CONTROL_SAMPLES(MSAA_ONE));
825
826 OUT_PKT0(ring, REG_A4XX_GRAS_CL_GB_CLIP_ADJ, 1);
827 OUT_RING(ring, A4XX_GRAS_CL_GB_CLIP_ADJ_HORZ(0) |
828 A4XX_GRAS_CL_GB_CLIP_ADJ_VERT(0));
829
830 OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
831 OUT_RING(ring, A4XX_RB_ALPHA_CONTROL_ALPHA_TEST_FUNC(FUNC_ALWAYS));
832
833 OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
834 OUT_RING(ring, A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
835
836 OUT_PKT0(ring, REG_A4XX_GRAS_CLEAR_CNTL, 1);
837 OUT_RING(ring, A4XX_GRAS_CLEAR_CNTL_NOT_FASTCLEAR);
838
839 OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
840 OUT_RING(ring, 0x0);
841
842 ctx->needs_rb_fbd = true;
843 }
844
845 void
846 fd4_emit_init(struct pipe_context *pctx)
847 {
848 struct fd_context *ctx = fd_context(pctx);
849 ctx->emit_const = fd4_emit_const;
850 ctx->emit_const_bo = fd4_emit_const_bo;
851 }