1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3 /*
4 * Copyright (C) 2013 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 #include "util/u_viewport.h"
35
36 #include "freedreno_resource.h"
37 #include "freedreno_query_hw.h"
38
39 #include "fd3_emit.h"
40 #include "fd3_blend.h"
41 #include "fd3_context.h"
42 #include "fd3_program.h"
43 #include "fd3_rasterizer.h"
44 #include "fd3_texture.h"
45 #include "fd3_format.h"
46 #include "fd3_zsa.h"
47
48 static const enum adreno_state_block sb[] = {
49 [SHADER_VERTEX] = SB_VERT_SHADER,
50 [SHADER_FRAGMENT] = SB_FRAG_SHADER,
51 };
52
53 /* regid: base const register
54 * prsc or dwords: buffer containing constant values
55 * sizedwords: size of const value buffer
56 */
57 static void
fd3_emit_const(struct fd_ringbuffer * ring,enum shader_t type,uint32_t regid,uint32_t offset,uint32_t sizedwords,const uint32_t * dwords,struct pipe_resource * prsc)58 fd3_emit_const(struct fd_ringbuffer *ring, enum shader_t type,
59 uint32_t regid, uint32_t offset, uint32_t sizedwords,
60 const uint32_t *dwords, struct pipe_resource *prsc)
61 {
62 uint32_t i, sz;
63 enum adreno_state_src src;
64
65 debug_assert((regid % 4) == 0);
66 debug_assert((sizedwords % 4) == 0);
67
68 if (prsc) {
69 sz = 0;
70 src = SS_INDIRECT;
71 } else {
72 sz = sizedwords;
73 src = SS_DIRECT;
74 }
75
76 OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
77 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) |
78 CP_LOAD_STATE_0_STATE_SRC(src) |
79 CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
80 CP_LOAD_STATE_0_NUM_UNIT(sizedwords/2));
81 if (prsc) {
82 struct fd_bo *bo = fd_resource(prsc)->bo;
83 OUT_RELOC(ring, bo, offset,
84 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0);
85 } else {
86 OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
87 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
88 dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
89 }
90 for (i = 0; i < sz; i++) {
91 OUT_RING(ring, dwords[i]);
92 }
93 }
94
95 static void
fd3_emit_const_bo(struct fd_ringbuffer * ring,enum shader_t type,boolean write,uint32_t regid,uint32_t num,struct pipe_resource ** prscs,uint32_t * offsets)96 fd3_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
97 uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets)
98 {
99 uint32_t anum = align(num, 4);
100 uint32_t i;
101
102 debug_assert((regid % 4) == 0);
103
104 OUT_PKT3(ring, CP_LOAD_STATE, 2 + anum);
105 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) |
106 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
107 CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
108 CP_LOAD_STATE_0_NUM_UNIT(anum/2));
109 OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
110 CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
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 }
122 }
123
124 for (; i < anum; i++)
125 OUT_RING(ring, 0xffffffff);
126 }
127
128 #define VERT_TEX_OFF 0
129 #define FRAG_TEX_OFF 16
130 #define BASETABLE_SZ A3XX_MAX_MIP_LEVELS
131
132 static void
emit_textures(struct fd_context * ctx,struct fd_ringbuffer * ring,enum adreno_state_block sb,struct fd_texture_stateobj * tex)133 emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
134 enum adreno_state_block sb, struct fd_texture_stateobj *tex)
135 {
136 static const unsigned tex_off[] = {
137 [SB_VERT_TEX] = VERT_TEX_OFF,
138 [SB_FRAG_TEX] = FRAG_TEX_OFF,
139 };
140 static const enum adreno_state_block mipaddr[] = {
141 [SB_VERT_TEX] = SB_VERT_MIPADDR,
142 [SB_FRAG_TEX] = SB_FRAG_MIPADDR,
143 };
144 static const uint32_t bcolor_reg[] = {
145 [SB_VERT_TEX] = REG_A3XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
146 [SB_FRAG_TEX] = REG_A3XX_TPL1_TP_FS_BORDER_COLOR_BASE_ADDR,
147 };
148 struct fd3_context *fd3_ctx = fd3_context(ctx);
149 bool needs_border = false;
150 unsigned i, j;
151
152 if (tex->num_samplers > 0) {
153 /* output sampler state: */
154 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * tex->num_samplers));
155 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) |
156 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
157 CP_LOAD_STATE_0_STATE_BLOCK(sb) |
158 CP_LOAD_STATE_0_NUM_UNIT(tex->num_samplers));
159 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
160 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
161 for (i = 0; i < tex->num_samplers; i++) {
162 static const struct fd3_sampler_stateobj dummy_sampler = {};
163 const struct fd3_sampler_stateobj *sampler = tex->samplers[i] ?
164 fd3_sampler_stateobj(tex->samplers[i]) :
165 &dummy_sampler;
166
167 OUT_RING(ring, sampler->texsamp0);
168 OUT_RING(ring, sampler->texsamp1);
169
170 needs_border |= sampler->needs_border;
171 }
172 }
173
174 if (tex->num_textures > 0) {
175 /* emit texture state: */
176 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (4 * tex->num_textures));
177 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) |
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 fd3_pipe_sampler_view dummy_view = {};
185 const struct fd3_pipe_sampler_view *view = tex->textures[i] ?
186 fd3_pipe_sampler_view(tex->textures[i]) :
187 &dummy_view;
188 OUT_RING(ring, view->texconst0);
189 OUT_RING(ring, view->texconst1);
190 OUT_RING(ring, view->texconst2 |
191 A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));
192 OUT_RING(ring, view->texconst3);
193 }
194
195 /* emit mipaddrs: */
196 OUT_PKT3(ring, CP_LOAD_STATE, 2 + (BASETABLE_SZ * tex->num_textures));
197 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(BASETABLE_SZ * tex_off[sb]) |
198 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
199 CP_LOAD_STATE_0_STATE_BLOCK(mipaddr[sb]) |
200 CP_LOAD_STATE_0_NUM_UNIT(BASETABLE_SZ * tex->num_textures));
201 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
202 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
203 for (i = 0; i < tex->num_textures; i++) {
204 static const struct fd3_pipe_sampler_view dummy_view = {
205 .base.target = PIPE_TEXTURE_1D, /* anything !PIPE_BUFFER */
206 .base.u.tex.first_level = 1,
207 };
208 const struct fd3_pipe_sampler_view *view = tex->textures[i] ?
209 fd3_pipe_sampler_view(tex->textures[i]) :
210 &dummy_view;
211 struct fd_resource *rsc = fd_resource(view->base.texture);
212 if (rsc && rsc->base.b.target == PIPE_BUFFER) {
213 OUT_RELOC(ring, rsc->bo, view->base.u.buf.offset, 0, 0);
214 j = 1;
215 } else {
216 unsigned start = fd_sampler_first_level(&view->base);
217 unsigned end = fd_sampler_last_level(&view->base);
218
219 for (j = 0; j < (end - start + 1); j++) {
220 struct fd_resource_slice *slice =
221 fd_resource_slice(rsc, j + start);
222 OUT_RELOC(ring, rsc->bo, slice->offset, 0, 0);
223 }
224 }
225
226 /* pad the remaining entries w/ null: */
227 for (; j < BASETABLE_SZ; j++) {
228 OUT_RING(ring, 0x00000000);
229 }
230 }
231 }
232
233 if (needs_border) {
234 unsigned off;
235 void *ptr;
236
237 u_upload_alloc(fd3_ctx->border_color_uploader,
238 0, BORDER_COLOR_UPLOAD_SIZE,
239 BORDER_COLOR_UPLOAD_SIZE, &off,
240 &fd3_ctx->border_color_buf,
241 &ptr);
242
243 fd_setup_border_colors(tex, ptr, tex_off[sb]);
244
245 OUT_PKT0(ring, bcolor_reg[sb], 1);
246 OUT_RELOC(ring, fd_resource(fd3_ctx->border_color_buf)->bo, off, 0, 0);
247
248 u_upload_unmap(fd3_ctx->border_color_uploader);
249 }
250 }
251
252 /* emit texture state for mem->gmem restore operation.. eventually it would
253 * be good to get rid of this and use normal CSO/etc state for more of these
254 * special cases, but for now the compiler is not sufficient..
255 *
256 * Also, for using normal state, not quite sure how to handle the special
257 * case format (fd3_gmem_restore_format()) stuff for restoring depth/stencil.
258 */
259 void
fd3_emit_gmem_restore_tex(struct fd_ringbuffer * ring,struct pipe_surface ** psurf,int bufs)260 fd3_emit_gmem_restore_tex(struct fd_ringbuffer *ring,
261 struct pipe_surface **psurf,
262 int bufs)
263 {
264 int i, j;
265
266 /* output sampler state: */
267 OUT_PKT3(ring, CP_LOAD_STATE, 2 + 2 * bufs);
268 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(FRAG_TEX_OFF) |
269 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
270 CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
271 CP_LOAD_STATE_0_NUM_UNIT(bufs));
272 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
273 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
274 for (i = 0; i < bufs; i++) {
275 OUT_RING(ring, A3XX_TEX_SAMP_0_XY_MAG(A3XX_TEX_NEAREST) |
276 A3XX_TEX_SAMP_0_XY_MIN(A3XX_TEX_NEAREST) |
277 A3XX_TEX_SAMP_0_WRAP_S(A3XX_TEX_CLAMP_TO_EDGE) |
278 A3XX_TEX_SAMP_0_WRAP_T(A3XX_TEX_CLAMP_TO_EDGE) |
279 A3XX_TEX_SAMP_0_WRAP_R(A3XX_TEX_REPEAT));
280 OUT_RING(ring, 0x00000000);
281 }
282
283 /* emit texture state: */
284 OUT_PKT3(ring, CP_LOAD_STATE, 2 + 4 * bufs);
285 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(FRAG_TEX_OFF) |
286 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
287 CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
288 CP_LOAD_STATE_0_NUM_UNIT(bufs));
289 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
290 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
291 for (i = 0; i < bufs; i++) {
292 if (!psurf[i]) {
293 OUT_RING(ring, A3XX_TEX_CONST_0_TYPE(A3XX_TEX_2D) |
294 A3XX_TEX_CONST_0_SWIZ_X(A3XX_TEX_ONE) |
295 A3XX_TEX_CONST_0_SWIZ_Y(A3XX_TEX_ONE) |
296 A3XX_TEX_CONST_0_SWIZ_Z(A3XX_TEX_ONE) |
297 A3XX_TEX_CONST_0_SWIZ_W(A3XX_TEX_ONE));
298 OUT_RING(ring, 0x00000000);
299 OUT_RING(ring, A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));
300 OUT_RING(ring, 0x00000000);
301 continue;
302 }
303
304 struct fd_resource *rsc = fd_resource(psurf[i]->texture);
305 enum pipe_format format = fd_gmem_restore_format(psurf[i]->format);
306 /* The restore blit_zs shader expects stencil in sampler 0, and depth
307 * in sampler 1
308 */
309 if (rsc->stencil && i == 0) {
310 rsc = rsc->stencil;
311 format = fd_gmem_restore_format(rsc->base.b.format);
312 }
313
314 /* note: PIPE_BUFFER disallowed for surfaces */
315 unsigned lvl = psurf[i]->u.tex.level;
316 struct fd_resource_slice *slice = fd_resource_slice(rsc, lvl);
317
318 debug_assert(psurf[i]->u.tex.first_layer == psurf[i]->u.tex.last_layer);
319
320 OUT_RING(ring, A3XX_TEX_CONST_0_FMT(fd3_pipe2tex(format)) |
321 A3XX_TEX_CONST_0_TYPE(A3XX_TEX_2D) |
322 fd3_tex_swiz(format, PIPE_SWIZZLE_X, PIPE_SWIZZLE_Y,
323 PIPE_SWIZZLE_Z, PIPE_SWIZZLE_W));
324 OUT_RING(ring, A3XX_TEX_CONST_1_FETCHSIZE(TFETCH_DISABLE) |
325 A3XX_TEX_CONST_1_WIDTH(psurf[i]->width) |
326 A3XX_TEX_CONST_1_HEIGHT(psurf[i]->height));
327 OUT_RING(ring, A3XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp) |
328 A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));
329 OUT_RING(ring, 0x00000000);
330 }
331
332 /* emit mipaddrs: */
333 OUT_PKT3(ring, CP_LOAD_STATE, 2 + BASETABLE_SZ * bufs);
334 OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(BASETABLE_SZ * FRAG_TEX_OFF) |
335 CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
336 CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_MIPADDR) |
337 CP_LOAD_STATE_0_NUM_UNIT(BASETABLE_SZ * bufs));
338 OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
339 CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
340 for (i = 0; i < bufs; i++) {
341 if (psurf[i]) {
342 struct fd_resource *rsc = fd_resource(psurf[i]->texture);
343 /* Matches above logic for blit_zs shader */
344 if (rsc->stencil && i == 0)
345 rsc = rsc->stencil;
346 unsigned lvl = psurf[i]->u.tex.level;
347 uint32_t offset = fd_resource_offset(rsc, lvl, psurf[i]->u.tex.first_layer);
348 OUT_RELOC(ring, rsc->bo, offset, 0, 0);
349 } else {
350 OUT_RING(ring, 0x00000000);
351 }
352
353 /* pad the remaining entries w/ null: */
354 for (j = 1; j < BASETABLE_SZ; j++) {
355 OUT_RING(ring, 0x00000000);
356 }
357 }
358 }
359
360 void
fd3_emit_vertex_bufs(struct fd_ringbuffer * ring,struct fd3_emit * emit)361 fd3_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd3_emit *emit)
362 {
363 int32_t i, j, last = -1;
364 uint32_t total_in = 0;
365 const struct fd_vertex_state *vtx = emit->vtx;
366 const struct ir3_shader_variant *vp = fd3_emit_get_vp(emit);
367 unsigned vertex_regid = regid(63, 0);
368 unsigned instance_regid = regid(63, 0);
369 unsigned vtxcnt_regid = regid(63, 0);
370
371 /* Note that sysvals come *after* normal inputs: */
372 for (i = 0; i < vp->inputs_count; i++) {
373 if (!vp->inputs[i].compmask)
374 continue;
375 if (vp->inputs[i].sysval) {
376 switch(vp->inputs[i].slot) {
377 case SYSTEM_VALUE_BASE_VERTEX:
378 /* handled elsewhere */
379 break;
380 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
381 vertex_regid = vp->inputs[i].regid;
382 break;
383 case SYSTEM_VALUE_INSTANCE_ID:
384 instance_regid = vp->inputs[i].regid;
385 break;
386 case SYSTEM_VALUE_VERTEX_CNT:
387 vtxcnt_regid = vp->inputs[i].regid;
388 break;
389 default:
390 unreachable("invalid system value");
391 break;
392 }
393 } else if (i < vtx->vtx->num_elements) {
394 last = i;
395 }
396 }
397
398 for (i = 0, j = 0; i <= last; i++) {
399 assert(!vp->inputs[i].sysval);
400 if (vp->inputs[i].compmask) {
401 struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
402 const struct pipe_vertex_buffer *vb =
403 &vtx->vertexbuf.vb[elem->vertex_buffer_index];
404 struct fd_resource *rsc = fd_resource(vb->buffer);
405 enum pipe_format pfmt = elem->src_format;
406 enum a3xx_vtx_fmt fmt = fd3_pipe2vtx(pfmt);
407 bool switchnext = (i != last) ||
408 (vertex_regid != regid(63, 0)) ||
409 (instance_regid != regid(63, 0)) ||
410 (vtxcnt_regid != regid(63, 0));
411 bool isint = util_format_is_pure_integer(pfmt);
412 uint32_t fs = util_format_get_blocksize(pfmt);
413
414 debug_assert(fmt != ~0);
415
416 OUT_PKT0(ring, REG_A3XX_VFD_FETCH(j), 2);
417 OUT_RING(ring, A3XX_VFD_FETCH_INSTR_0_FETCHSIZE(fs - 1) |
418 A3XX_VFD_FETCH_INSTR_0_BUFSTRIDE(vb->stride) |
419 COND(switchnext, A3XX_VFD_FETCH_INSTR_0_SWITCHNEXT) |
420 A3XX_VFD_FETCH_INSTR_0_INDEXCODE(j) |
421 COND(elem->instance_divisor, A3XX_VFD_FETCH_INSTR_0_INSTANCED) |
422 A3XX_VFD_FETCH_INSTR_0_STEPRATE(MAX2(1, elem->instance_divisor)));
423 OUT_RELOC(ring, rsc->bo, vb->buffer_offset + elem->src_offset, 0, 0);
424
425 OUT_PKT0(ring, REG_A3XX_VFD_DECODE_INSTR(j), 1);
426 OUT_RING(ring, A3XX_VFD_DECODE_INSTR_CONSTFILL |
427 A3XX_VFD_DECODE_INSTR_WRITEMASK(vp->inputs[i].compmask) |
428 A3XX_VFD_DECODE_INSTR_FORMAT(fmt) |
429 A3XX_VFD_DECODE_INSTR_SWAP(fd3_pipe2swap(pfmt)) |
430 A3XX_VFD_DECODE_INSTR_REGID(vp->inputs[i].regid) |
431 A3XX_VFD_DECODE_INSTR_SHIFTCNT(fs) |
432 A3XX_VFD_DECODE_INSTR_LASTCOMPVALID |
433 COND(isint, A3XX_VFD_DECODE_INSTR_INT) |
434 COND(switchnext, A3XX_VFD_DECODE_INSTR_SWITCHNEXT));
435
436 total_in += vp->inputs[i].ncomp;
437 j++;
438 }
439 }
440
441 /* hw doesn't like to be configured for zero vbo's, it seems: */
442 if (last < 0) {
443 /* just recycle the shader bo, we just need to point to *something*
444 * valid:
445 */
446 struct fd_bo *dummy_vbo = vp->bo;
447 bool switchnext = (vertex_regid != regid(63, 0)) ||
448 (instance_regid != regid(63, 0)) ||
449 (vtxcnt_regid != regid(63, 0));
450
451 OUT_PKT0(ring, REG_A3XX_VFD_FETCH(0), 2);
452 OUT_RING(ring, A3XX_VFD_FETCH_INSTR_0_FETCHSIZE(0) |
453 A3XX_VFD_FETCH_INSTR_0_BUFSTRIDE(0) |
454 COND(switchnext, A3XX_VFD_FETCH_INSTR_0_SWITCHNEXT) |
455 A3XX_VFD_FETCH_INSTR_0_INDEXCODE(0) |
456 A3XX_VFD_FETCH_INSTR_0_STEPRATE(1));
457 OUT_RELOC(ring, dummy_vbo, 0, 0, 0);
458
459 OUT_PKT0(ring, REG_A3XX_VFD_DECODE_INSTR(0), 1);
460 OUT_RING(ring, A3XX_VFD_DECODE_INSTR_CONSTFILL |
461 A3XX_VFD_DECODE_INSTR_WRITEMASK(0x1) |
462 A3XX_VFD_DECODE_INSTR_FORMAT(VFMT_8_UNORM) |
463 A3XX_VFD_DECODE_INSTR_SWAP(XYZW) |
464 A3XX_VFD_DECODE_INSTR_REGID(regid(0,0)) |
465 A3XX_VFD_DECODE_INSTR_SHIFTCNT(1) |
466 A3XX_VFD_DECODE_INSTR_LASTCOMPVALID |
467 COND(switchnext, A3XX_VFD_DECODE_INSTR_SWITCHNEXT));
468
469 total_in = 1;
470 j = 1;
471 }
472
473 OUT_PKT0(ring, REG_A3XX_VFD_CONTROL_0, 2);
474 OUT_RING(ring, A3XX_VFD_CONTROL_0_TOTALATTRTOVS(total_in) |
475 A3XX_VFD_CONTROL_0_PACKETSIZE(2) |
476 A3XX_VFD_CONTROL_0_STRMDECINSTRCNT(j) |
477 A3XX_VFD_CONTROL_0_STRMFETCHINSTRCNT(j));
478 OUT_RING(ring, A3XX_VFD_CONTROL_1_MAXSTORAGE(1) | // XXX
479 A3XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
480 A3XX_VFD_CONTROL_1_REGID4INST(instance_regid));
481
482 OUT_PKT0(ring, REG_A3XX_VFD_VS_THREADING_THRESHOLD, 1);
483 OUT_RING(ring, A3XX_VFD_VS_THREADING_THRESHOLD_REGID_THRESHOLD(15) |
484 A3XX_VFD_VS_THREADING_THRESHOLD_REGID_VTXCNT(vtxcnt_regid));
485 }
486
487 void
fd3_emit_state(struct fd_context * ctx,struct fd_ringbuffer * ring,struct fd3_emit * emit)488 fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
489 struct fd3_emit *emit)
490 {
491 const struct ir3_shader_variant *vp = fd3_emit_get_vp(emit);
492 const struct ir3_shader_variant *fp = fd3_emit_get_fp(emit);
493 uint32_t dirty = emit->dirty;
494
495 emit_marker(ring, 5);
496
497 if (dirty & FD_DIRTY_SAMPLE_MASK) {
498 OUT_PKT0(ring, REG_A3XX_RB_MSAA_CONTROL, 1);
499 OUT_RING(ring, A3XX_RB_MSAA_CONTROL_DISABLE |
500 A3XX_RB_MSAA_CONTROL_SAMPLES(MSAA_ONE) |
501 A3XX_RB_MSAA_CONTROL_SAMPLE_MASK(ctx->sample_mask));
502 }
503
504 if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG | FD_DIRTY_BLEND_DUAL)) &&
505 !emit->key.binning_pass) {
506 uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_render_control |
507 fd3_blend_stateobj(ctx->blend)->rb_render_control;
508
509 val |= COND(fp->frag_face, A3XX_RB_RENDER_CONTROL_FACENESS);
510 val |= COND(fp->frag_coord, A3XX_RB_RENDER_CONTROL_XCOORD |
511 A3XX_RB_RENDER_CONTROL_YCOORD |
512 A3XX_RB_RENDER_CONTROL_ZCOORD |
513 A3XX_RB_RENDER_CONTROL_WCOORD);
514
515 /* I suppose if we needed to (which I don't *think* we need
516 * to), we could emit this for binning pass too. But we
517 * would need to keep a different patch-list for binning
518 * vs render pass.
519 */
520
521 OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
522 OUT_RINGP(ring, val, &ctx->batch->rbrc_patches);
523 }
524
525 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
526 struct fd3_zsa_stateobj *zsa = fd3_zsa_stateobj(ctx->zsa);
527 struct pipe_stencil_ref *sr = &ctx->stencil_ref;
528
529 OUT_PKT0(ring, REG_A3XX_RB_ALPHA_REF, 1);
530 OUT_RING(ring, zsa->rb_alpha_ref);
531
532 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
533 OUT_RING(ring, zsa->rb_stencil_control);
534
535 OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
536 OUT_RING(ring, zsa->rb_stencilrefmask |
537 A3XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
538 OUT_RING(ring, zsa->rb_stencilrefmask_bf |
539 A3XX_RB_STENCILREFMASK_BF_STENCILREF(sr->ref_value[1]));
540 }
541
542 if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
543 uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_depth_control;
544 if (fp->writes_pos) {
545 val |= A3XX_RB_DEPTH_CONTROL_FRAG_WRITES_Z;
546 val |= A3XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
547 }
548 if (fp->has_kill) {
549 val |= A3XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
550 }
551 if (!ctx->rasterizer->depth_clip) {
552 val |= A3XX_RB_DEPTH_CONTROL_Z_CLAMP_ENABLE;
553 }
554 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
555 OUT_RING(ring, val);
556 }
557
558 if (dirty & FD_DIRTY_RASTERIZER) {
559 struct fd3_rasterizer_stateobj *rasterizer =
560 fd3_rasterizer_stateobj(ctx->rasterizer);
561
562 OUT_PKT0(ring, REG_A3XX_GRAS_SU_MODE_CONTROL, 1);
563 OUT_RING(ring, rasterizer->gras_su_mode_control);
564
565 OUT_PKT0(ring, REG_A3XX_GRAS_SU_POINT_MINMAX, 2);
566 OUT_RING(ring, rasterizer->gras_su_point_minmax);
567 OUT_RING(ring, rasterizer->gras_su_point_size);
568
569 OUT_PKT0(ring, REG_A3XX_GRAS_SU_POLY_OFFSET_SCALE, 2);
570 OUT_RING(ring, rasterizer->gras_su_poly_offset_scale);
571 OUT_RING(ring, rasterizer->gras_su_poly_offset_offset);
572 }
573
574 if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
575 uint32_t val = fd3_rasterizer_stateobj(ctx->rasterizer)
576 ->gras_cl_clip_cntl;
577 uint8_t planes = ctx->rasterizer->clip_plane_enable;
578 val |= COND(fp->writes_pos, A3XX_GRAS_CL_CLIP_CNTL_ZCLIP_DISABLE);
579 val |= COND(fp->frag_coord, A3XX_GRAS_CL_CLIP_CNTL_ZCOORD |
580 A3XX_GRAS_CL_CLIP_CNTL_WCOORD);
581 if (!emit->key.ucp_enables)
582 val |= A3XX_GRAS_CL_CLIP_CNTL_NUM_USER_CLIP_PLANES(
583 MIN2(util_bitcount(planes), 6));
584 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
585 OUT_RING(ring, val);
586 }
587
588 if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_PROG | FD_DIRTY_UCP)) {
589 uint32_t planes = ctx->rasterizer->clip_plane_enable;
590 int count = 0;
591
592 if (emit->key.ucp_enables)
593 planes = 0;
594
595 while (planes && count < 6) {
596 int i = ffs(planes) - 1;
597
598 planes &= ~(1U << i);
599 fd_wfi(ctx->batch, ring);
600 OUT_PKT0(ring, REG_A3XX_GRAS_CL_USER_PLANE(count++), 4);
601 OUT_RING(ring, fui(ctx->ucp.ucp[i][0]));
602 OUT_RING(ring, fui(ctx->ucp.ucp[i][1]));
603 OUT_RING(ring, fui(ctx->ucp.ucp[i][2]));
604 OUT_RING(ring, fui(ctx->ucp.ucp[i][3]));
605 }
606 }
607
608 /* NOTE: since primitive_restart is not actually part of any
609 * state object, we need to make sure that we always emit
610 * PRIM_VTX_CNTL.. either that or be more clever and detect
611 * when it changes.
612 */
613 if (emit->info) {
614 const struct pipe_draw_info *info = emit->info;
615 uint32_t val = fd3_rasterizer_stateobj(ctx->rasterizer)
616 ->pc_prim_vtx_cntl;
617
618 if (!emit->key.binning_pass) {
619 uint32_t stride_in_vpc = align(fp->total_in, 4) / 4;
620 if (stride_in_vpc > 0)
621 stride_in_vpc = MAX2(stride_in_vpc, 2);
622 val |= A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(stride_in_vpc);
623 }
624
625 if (info->indexed && info->primitive_restart) {
626 val |= A3XX_PC_PRIM_VTX_CNTL_PRIMITIVE_RESTART;
627 }
628
629 val |= COND(vp->writes_psize, A3XX_PC_PRIM_VTX_CNTL_PSIZE);
630
631 OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
632 OUT_RING(ring, val);
633 }
634
635 if (dirty & (FD_DIRTY_SCISSOR | FD_DIRTY_RASTERIZER | FD_DIRTY_VIEWPORT)) {
636 struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
637 int minx = scissor->minx;
638 int miny = scissor->miny;
639 int maxx = scissor->maxx;
640 int maxy = scissor->maxy;
641
642 /* Unfortunately there is no separate depth clip disable, only an all
643 * or nothing deal. So when we disable clipping, we must handle the
644 * viewport clip via scissors.
645 */
646 if (!ctx->rasterizer->depth_clip) {
647 struct pipe_viewport_state *vp = &ctx->viewport;
648 minx = MAX2(minx, (int)floorf(vp->translate[0] - fabsf(vp->scale[0])));
649 miny = MAX2(miny, (int)floorf(vp->translate[1] - fabsf(vp->scale[1])));
650 maxx = MIN2(maxx, (int)ceilf(vp->translate[0] + fabsf(vp->scale[0])));
651 maxy = MIN2(maxy, (int)ceilf(vp->translate[1] + fabsf(vp->scale[1])));
652 }
653
654 OUT_PKT0(ring, REG_A3XX_GRAS_SC_WINDOW_SCISSOR_TL, 2);
655 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_TL_X(minx) |
656 A3XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(miny));
657 OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(maxx - 1) |
658 A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(maxy - 1));
659
660 ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, minx);
661 ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, miny);
662 ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, maxx);
663 ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, maxy);
664 }
665
666 if (dirty & FD_DIRTY_VIEWPORT) {
667 fd_wfi(ctx->batch, ring);
668 OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 6);
669 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET(ctx->viewport.translate[0] - 0.5));
670 OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE(ctx->viewport.scale[0]));
671 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YOFFSET(ctx->viewport.translate[1] - 0.5));
672 OUT_RING(ring, A3XX_GRAS_CL_VPORT_YSCALE(ctx->viewport.scale[1]));
673 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(ctx->viewport.translate[2]));
674 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(ctx->viewport.scale[2]));
675 }
676
677 if (dirty & (FD_DIRTY_VIEWPORT | FD_DIRTY_RASTERIZER | FD_DIRTY_FRAMEBUFFER)) {
678 float zmin, zmax;
679 int depth = 24;
680 if (ctx->batch->framebuffer.zsbuf) {
681 depth = util_format_get_component_bits(
682 pipe_surface_format(ctx->batch->framebuffer.zsbuf),
683 UTIL_FORMAT_COLORSPACE_ZS, 0);
684 }
685 util_viewport_zmin_zmax(&ctx->viewport, ctx->rasterizer->clip_halfz,
686 &zmin, &zmax);
687
688 OUT_PKT0(ring, REG_A3XX_RB_Z_CLAMP_MIN, 2);
689 if (depth == 32) {
690 OUT_RING(ring, (uint32_t)(zmin * 0xffffffff));
691 OUT_RING(ring, (uint32_t)(zmax * 0xffffffff));
692 } else if (depth == 16) {
693 OUT_RING(ring, (uint32_t)(zmin * 0xffff));
694 OUT_RING(ring, (uint32_t)(zmax * 0xffff));
695 } else {
696 OUT_RING(ring, (uint32_t)(zmin * 0xffffff));
697 OUT_RING(ring, (uint32_t)(zmax * 0xffffff));
698 }
699 }
700
701 if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER | FD_DIRTY_BLEND_DUAL)) {
702 struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
703 int nr_cbufs = pfb->nr_cbufs;
704 if (fd3_blend_stateobj(ctx->blend)->rb_render_control &
705 A3XX_RB_RENDER_CONTROL_DUAL_COLOR_IN_ENABLE)
706 nr_cbufs++;
707 fd3_program_emit(ring, emit, nr_cbufs, pfb->cbufs);
708 }
709
710 /* TODO we should not need this or fd_wfi() before emit_constants():
711 */
712 OUT_PKT3(ring, CP_EVENT_WRITE, 1);
713 OUT_RING(ring, HLSQ_FLUSH);
714
715 if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
716 ir3_emit_consts(vp, ring, ctx, emit->info, dirty);
717 if (!emit->key.binning_pass)
718 ir3_emit_consts(fp, ring, ctx, emit->info, dirty);
719 }
720
721 if (dirty & (FD_DIRTY_BLEND | FD_DIRTY_FRAMEBUFFER)) {
722 struct fd3_blend_stateobj *blend = fd3_blend_stateobj(ctx->blend);
723 uint32_t i;
724
725 for (i = 0; i < ARRAY_SIZE(blend->rb_mrt); i++) {
726 enum pipe_format format =
727 pipe_surface_format(ctx->batch->framebuffer.cbufs[i]);
728 const struct util_format_description *desc =
729 util_format_description(format);
730 bool is_float = util_format_is_float(format);
731 bool is_int = util_format_is_pure_integer(format);
732 bool has_alpha = util_format_has_alpha(format);
733 uint32_t control = blend->rb_mrt[i].control;
734 uint32_t blend_control = blend->rb_mrt[i].blend_control_alpha;
735
736 if (is_int) {
737 control &= (A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK |
738 A3XX_RB_MRT_CONTROL_DITHER_MODE__MASK);
739 control |= A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY);
740 }
741
742 if (format == PIPE_FORMAT_NONE)
743 control &= ~A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK;
744
745 if (has_alpha) {
746 blend_control |= blend->rb_mrt[i].blend_control_rgb;
747 } else {
748 blend_control |= blend->rb_mrt[i].blend_control_no_alpha_rgb;
749 control &= ~A3XX_RB_MRT_CONTROL_BLEND2;
750 }
751
752 if (format && util_format_get_component_bits(
753 format, UTIL_FORMAT_COLORSPACE_RGB, 0) < 8) {
754 const struct pipe_rt_blend_state *rt;
755 if (ctx->blend->independent_blend_enable)
756 rt = &ctx->blend->rt[i];
757 else
758 rt = &ctx->blend->rt[0];
759
760 if (!util_format_colormask_full(desc, rt->colormask))
761 control |= A3XX_RB_MRT_CONTROL_READ_DEST_ENABLE;
762 }
763
764 OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
765 OUT_RING(ring, control);
766
767 OUT_PKT0(ring, REG_A3XX_RB_MRT_BLEND_CONTROL(i), 1);
768 OUT_RING(ring, blend_control |
769 COND(!is_float, A3XX_RB_MRT_BLEND_CONTROL_CLAMP_ENABLE));
770 }
771 }
772
773 if (dirty & FD_DIRTY_BLEND_COLOR) {
774 struct pipe_blend_color *bcolor = &ctx->blend_color;
775 OUT_PKT0(ring, REG_A3XX_RB_BLEND_RED, 4);
776 OUT_RING(ring, A3XX_RB_BLEND_RED_UINT(bcolor->color[0] * 255.0) |
777 A3XX_RB_BLEND_RED_FLOAT(bcolor->color[0]));
778 OUT_RING(ring, A3XX_RB_BLEND_GREEN_UINT(bcolor->color[1] * 255.0) |
779 A3XX_RB_BLEND_GREEN_FLOAT(bcolor->color[1]));
780 OUT_RING(ring, A3XX_RB_BLEND_BLUE_UINT(bcolor->color[2] * 255.0) |
781 A3XX_RB_BLEND_BLUE_FLOAT(bcolor->color[2]));
782 OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(bcolor->color[3] * 255.0) |
783 A3XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]));
784 }
785
786 if (dirty & (FD_DIRTY_VERTTEX | FD_DIRTY_FRAGTEX))
787 fd_wfi(ctx->batch, ring);
788
789 if (dirty & FD_DIRTY_VERTTEX) {
790 if (vp->has_samp)
791 emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
792 else
793 dirty &= ~FD_DIRTY_VERTTEX;
794 }
795
796 if (dirty & FD_DIRTY_FRAGTEX) {
797 if (fp->has_samp)
798 emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
799 else
800 dirty &= ~FD_DIRTY_FRAGTEX;
801 }
802
803 ctx->dirty &= ~dirty;
804 }
805
806 /* emit setup at begin of new cmdstream buffer (don't rely on previous
807 * state, there could have been a context switch between ioctls):
808 */
809 void
fd3_emit_restore(struct fd_batch * batch,struct fd_ringbuffer * ring)810 fd3_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring)
811 {
812 struct fd_context *ctx = batch->ctx;
813 struct fd3_context *fd3_ctx = fd3_context(ctx);
814 int i;
815
816 if (ctx->screen->gpu_id == 320) {
817 OUT_PKT3(ring, CP_REG_RMW, 3);
818 OUT_RING(ring, REG_A3XX_RBBM_CLOCK_CTL);
819 OUT_RING(ring, 0xfffcffff);
820 OUT_RING(ring, 0x00000000);
821 }
822
823 fd_wfi(batch, ring);
824 OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
825 OUT_RING(ring, 0x00007fff);
826
827 OUT_PKT0(ring, REG_A3XX_SP_VS_PVT_MEM_PARAM_REG, 3);
828 OUT_RING(ring, 0x08000001); /* SP_VS_PVT_MEM_CTRL_REG */
829 OUT_RELOC(ring, fd3_ctx->vs_pvt_mem, 0,0,0); /* SP_VS_PVT_MEM_ADDR_REG */
830 OUT_RING(ring, 0x00000000); /* SP_VS_PVT_MEM_SIZE_REG */
831
832 OUT_PKT0(ring, REG_A3XX_SP_FS_PVT_MEM_PARAM_REG, 3);
833 OUT_RING(ring, 0x08000001); /* SP_FS_PVT_MEM_CTRL_REG */
834 OUT_RELOC(ring, fd3_ctx->fs_pvt_mem, 0,0,0); /* SP_FS_PVT_MEM_ADDR_REG */
835 OUT_RING(ring, 0x00000000); /* SP_FS_PVT_MEM_SIZE_REG */
836
837 OUT_PKT0(ring, REG_A3XX_PC_VERTEX_REUSE_BLOCK_CNTL, 1);
838 OUT_RING(ring, 0x0000000b); /* PC_VERTEX_REUSE_BLOCK_CNTL */
839
840 OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
841 OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
842 A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
843 A3XX_GRAS_SC_CONTROL_RASTER_MODE(0));
844
845 OUT_PKT0(ring, REG_A3XX_RB_MSAA_CONTROL, 2);
846 OUT_RING(ring, A3XX_RB_MSAA_CONTROL_DISABLE |
847 A3XX_RB_MSAA_CONTROL_SAMPLES(MSAA_ONE) |
848 A3XX_RB_MSAA_CONTROL_SAMPLE_MASK(0xffff));
849 OUT_RING(ring, 0x00000000); /* RB_ALPHA_REF */
850
851 OUT_PKT0(ring, REG_A3XX_GRAS_CL_GB_CLIP_ADJ, 1);
852 OUT_RING(ring, A3XX_GRAS_CL_GB_CLIP_ADJ_HORZ(0) |
853 A3XX_GRAS_CL_GB_CLIP_ADJ_VERT(0));
854
855 OUT_PKT0(ring, REG_A3XX_GRAS_TSE_DEBUG_ECO, 1);
856 OUT_RING(ring, 0x00000001); /* GRAS_TSE_DEBUG_ECO */
857
858 OUT_PKT0(ring, REG_A3XX_TPL1_TP_VS_TEX_OFFSET, 1);
859 OUT_RING(ring, A3XX_TPL1_TP_VS_TEX_OFFSET_SAMPLEROFFSET(VERT_TEX_OFF) |
860 A3XX_TPL1_TP_VS_TEX_OFFSET_MEMOBJOFFSET(VERT_TEX_OFF) |
861 A3XX_TPL1_TP_VS_TEX_OFFSET_BASETABLEPTR(BASETABLE_SZ * VERT_TEX_OFF));
862
863 OUT_PKT0(ring, REG_A3XX_TPL1_TP_FS_TEX_OFFSET, 1);
864 OUT_RING(ring, A3XX_TPL1_TP_FS_TEX_OFFSET_SAMPLEROFFSET(FRAG_TEX_OFF) |
865 A3XX_TPL1_TP_FS_TEX_OFFSET_MEMOBJOFFSET(FRAG_TEX_OFF) |
866 A3XX_TPL1_TP_FS_TEX_OFFSET_BASETABLEPTR(BASETABLE_SZ * FRAG_TEX_OFF));
867
868 OUT_PKT0(ring, REG_A3XX_VPC_VARY_CYLWRAP_ENABLE_0, 2);
869 OUT_RING(ring, 0x00000000); /* VPC_VARY_CYLWRAP_ENABLE_0 */
870 OUT_RING(ring, 0x00000000); /* VPC_VARY_CYLWRAP_ENABLE_1 */
871
872 OUT_PKT0(ring, REG_A3XX_UNKNOWN_0E43, 1);
873 OUT_RING(ring, 0x00000001); /* UNKNOWN_0E43 */
874
875 OUT_PKT0(ring, REG_A3XX_UNKNOWN_0F03, 1);
876 OUT_RING(ring, 0x00000001); /* UNKNOWN_0F03 */
877
878 OUT_PKT0(ring, REG_A3XX_UNKNOWN_0EE0, 1);
879 OUT_RING(ring, 0x00000003); /* UNKNOWN_0EE0 */
880
881 OUT_PKT0(ring, REG_A3XX_UNKNOWN_0C3D, 1);
882 OUT_RING(ring, 0x00000001); /* UNKNOWN_0C3D */
883
884 OUT_PKT0(ring, REG_A3XX_HLSQ_PERFCOUNTER0_SELECT, 1);
885 OUT_RING(ring, 0x00000000); /* HLSQ_PERFCOUNTER0_SELECT */
886
887 OUT_PKT0(ring, REG_A3XX_HLSQ_CONST_VSPRESV_RANGE_REG, 2);
888 OUT_RING(ring, A3XX_HLSQ_CONST_VSPRESV_RANGE_REG_STARTENTRY(0) |
889 A3XX_HLSQ_CONST_VSPRESV_RANGE_REG_ENDENTRY(0));
890 OUT_RING(ring, A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_STARTENTRY(0) |
891 A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_ENDENTRY(0));
892
893 fd3_emit_cache_flush(batch, ring);
894
895 OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
896 OUT_RING(ring, 0x00000000); /* GRAS_CL_CLIP_CNTL */
897
898 OUT_PKT0(ring, REG_A3XX_GRAS_SU_POINT_MINMAX, 2);
899 OUT_RING(ring, 0xffc00010); /* GRAS_SU_POINT_MINMAX */
900 OUT_RING(ring, 0x00000008); /* GRAS_SU_POINT_SIZE */
901
902 OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
903 OUT_RING(ring, 0xffffffff); /* PC_RESTART_INDEX */
904
905 OUT_PKT0(ring, REG_A3XX_RB_WINDOW_OFFSET, 1);
906 OUT_RING(ring, A3XX_RB_WINDOW_OFFSET_X(0) |
907 A3XX_RB_WINDOW_OFFSET_Y(0));
908
909 OUT_PKT0(ring, REG_A3XX_RB_BLEND_RED, 4);
910 OUT_RING(ring, A3XX_RB_BLEND_RED_UINT(0) |
911 A3XX_RB_BLEND_RED_FLOAT(0.0));
912 OUT_RING(ring, A3XX_RB_BLEND_GREEN_UINT(0) |
913 A3XX_RB_BLEND_GREEN_FLOAT(0.0));
914 OUT_RING(ring, A3XX_RB_BLEND_BLUE_UINT(0) |
915 A3XX_RB_BLEND_BLUE_FLOAT(0.0));
916 OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(0xff) |
917 A3XX_RB_BLEND_ALPHA_FLOAT(1.0));
918
919 for (i = 0; i < 6; i++) {
920 OUT_PKT0(ring, REG_A3XX_GRAS_CL_USER_PLANE(i), 4);
921 OUT_RING(ring, 0x00000000); /* GRAS_CL_USER_PLANE[i].X */
922 OUT_RING(ring, 0x00000000); /* GRAS_CL_USER_PLANE[i].Y */
923 OUT_RING(ring, 0x00000000); /* GRAS_CL_USER_PLANE[i].Z */
924 OUT_RING(ring, 0x00000000); /* GRAS_CL_USER_PLANE[i].W */
925 }
926
927 OUT_PKT0(ring, REG_A3XX_PC_VSTREAM_CONTROL, 1);
928 OUT_RING(ring, 0x00000000);
929
930 fd_event_write(batch, ring, CACHE_FLUSH);
931
932 if (is_a3xx_p0(ctx->screen)) {
933 OUT_PKT3(ring, CP_DRAW_INDX, 3);
934 OUT_RING(ring, 0x00000000);
935 OUT_RING(ring, DRAW(1, DI_SRC_SEL_AUTO_INDEX,
936 INDEX_SIZE_IGN, IGNORE_VISIBILITY, 0));
937 OUT_RING(ring, 0); /* NumIndices */
938 }
939
940 OUT_PKT3(ring, CP_NOP, 4);
941 OUT_RING(ring, 0x00000000);
942 OUT_RING(ring, 0x00000000);
943 OUT_RING(ring, 0x00000000);
944 OUT_RING(ring, 0x00000000);
945
946 fd_wfi(batch, ring);
947
948 fd_hw_query_enable(batch, ring);
949 }
950
951 static void
fd3_emit_ib(struct fd_ringbuffer * ring,struct fd_ringbuffer * target)952 fd3_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
953 {
954 __OUT_IB(ring, true, target);
955 }
956
957 void
fd3_emit_init(struct pipe_context * pctx)958 fd3_emit_init(struct pipe_context *pctx)
959 {
960 struct fd_context *ctx = fd_context(pctx);
961 ctx->emit_const = fd3_emit_const;
962 ctx->emit_const_bo = fd3_emit_const_bo;
963 ctx->emit_ib = fd3_emit_ib;
964 }
965