projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
freedreno/a6xx: Program state for tessellation stages
[mesa.git] / src / gallium / drivers / freedreno / a6xx / fd6_program.c
1 /*
2 * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
3 * Copyright © 2018 Google, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 * Authors:
25 * Rob Clark <robclark@freedesktop.org>
26 */
27
28 #include "pipe/p_state.h"
29 #include "util/u_string.h"
30 #include "util/u_memory.h"
31 #include "util/u_inlines.h"
32 #include "util/u_format.h"
33 #include "util/bitset.h"
34
35 #include "freedreno_program.h"
36
37 #include "fd6_program.h"
38 #include "fd6_emit.h"
39 #include "fd6_texture.h"
40 #include "fd6_format.h"
41
42 void
43 fd6_emit_shader(struct fd_ringbuffer *ring, const struct ir3_shader_variant *so)
44 {
45 enum a6xx_state_block sb = fd6_stage2shadersb(so->type);
46
47 uint32_t obj_start;
48 uint32_t instrlen;
49
50 switch (so->type) {
51 case MESA_SHADER_VERTEX:
52 obj_start = REG_A6XX_SP_VS_OBJ_START_LO;
53 instrlen = REG_A6XX_SP_VS_INSTRLEN;
54 break;
55 case MESA_SHADER_TESS_CTRL:
56 obj_start = REG_A6XX_SP_HS_OBJ_START_LO;
57 instrlen = REG_A6XX_SP_HS_INSTRLEN;
58 break;
59 case MESA_SHADER_TESS_EVAL:
60 obj_start = REG_A6XX_SP_DS_OBJ_START_LO;
61 instrlen = REG_A6XX_SP_DS_INSTRLEN;
62 break;
63 case MESA_SHADER_GEOMETRY:
64 obj_start = REG_A6XX_SP_GS_OBJ_START_LO;
65 instrlen = REG_A6XX_SP_GS_INSTRLEN;
66 break;
67 case MESA_SHADER_FRAGMENT:
68 obj_start = REG_A6XX_SP_FS_OBJ_START_LO;
69 instrlen = REG_A6XX_SP_FS_INSTRLEN;
70 break;
71 case MESA_SHADER_COMPUTE:
72 case MESA_SHADER_KERNEL:
73 obj_start = REG_A6XX_SP_CS_OBJ_START_LO;
74 instrlen = REG_A6XX_SP_CS_INSTRLEN;
75 break;
76 case MESA_SHADER_NONE:
77 unreachable("");
78 }
79
80 OUT_PKT4(ring, instrlen, 1);
81 OUT_RING(ring, so->instrlen);
82
83 OUT_PKT4(ring, obj_start, 2);
84 OUT_RELOC(ring, so->bo, 0, 0, 0);
85
86 OUT_PKT7(ring, fd6_stage2opcode(so->type), 3);
87 OUT_RING(ring, CP_LOAD_STATE6_0_DST_OFF(0) |
88 CP_LOAD_STATE6_0_STATE_TYPE(ST6_SHADER) |
89 CP_LOAD_STATE6_0_STATE_SRC(SS6_INDIRECT) |
90 CP_LOAD_STATE6_0_STATE_BLOCK(sb) |
91 CP_LOAD_STATE6_0_NUM_UNIT(so->instrlen));
92 OUT_RELOCD(ring, so->bo, 0, 0, 0);
93 }
94
95 /* Add any missing varyings needed for stream-out. Otherwise varyings not
96 * used by fragment shader will be stripped out.
97 */
98 static void
99 link_stream_out(struct ir3_shader_linkage *l, const struct ir3_shader_variant *v)
100 {
101 const struct ir3_stream_output_info *strmout = &v->shader->stream_output;
102
103 /*
104 * First, any stream-out varyings not already in linkage map (ie. also
105 * consumed by frag shader) need to be added:
106 */
107 for (unsigned i = 0; i < strmout->num_outputs; i++) {
108 const struct ir3_stream_output *out = &strmout->output[i];
109 unsigned k = out->register_index;
110 unsigned compmask =
111 (1 << (out->num_components + out->start_component)) - 1;
112 unsigned idx, nextloc = 0;
113
114 /* psize/pos need to be the last entries in linkage map, and will
115 * get added link_stream_out, so skip over them:
116 */
117 if ((v->outputs[k].slot == VARYING_SLOT_PSIZ) ||
118 (v->outputs[k].slot == VARYING_SLOT_POS))
119 continue;
120
121 for (idx = 0; idx < l->cnt; idx++) {
122 if (l->var[idx].regid == v->outputs[k].regid)
123 break;
124 nextloc = MAX2(nextloc, l->var[idx].loc + 4);
125 }
126
127 /* add if not already in linkage map: */
128 if (idx == l->cnt)
129 ir3_link_add(l, v->outputs[k].regid, compmask, nextloc);
130
131 /* expand component-mask if needed, ie streaming out all components
132 * but frag shader doesn't consume all components:
133 */
134 if (compmask & ~l->var[idx].compmask) {
135 l->var[idx].compmask |= compmask;
136 l->max_loc = MAX2(l->max_loc,
137 l->var[idx].loc + util_last_bit(l->var[idx].compmask));
138 }
139 }
140 }
141
142 static void
143 setup_stream_out(struct fd6_program_state *state, const struct ir3_shader_variant *v,
144 struct ir3_shader_linkage *l)
145 {
146 const struct ir3_stream_output_info *strmout = &v->shader->stream_output;
147 struct fd6_streamout_state *tf = &state->tf;
148
149 memset(tf, 0, sizeof(*tf));
150
151 tf->prog_count = align(l->max_loc, 2) / 2;
152
153 debug_assert(tf->prog_count < ARRAY_SIZE(tf->prog));
154
155 for (unsigned i = 0; i < strmout->num_outputs; i++) {
156 const struct ir3_stream_output *out = &strmout->output[i];
157 unsigned k = out->register_index;
158 unsigned idx;
159
160 tf->ncomp[out->output_buffer] += out->num_components;
161
162 /* linkage map sorted by order frag shader wants things, so
163 * a bit less ideal here..
164 */
165 for (idx = 0; idx < l->cnt; idx++)
166 if (l->var[idx].regid == v->outputs[k].regid)
167 break;
168
169 debug_assert(idx < l->cnt);
170
171 for (unsigned j = 0; j < out->num_components; j++) {
172 unsigned c = j + out->start_component;
173 unsigned loc = l->var[idx].loc + c;
174 unsigned off = j + out->dst_offset; /* in dwords */
175
176 if (loc & 1) {
177 tf->prog[loc/2] |= A6XX_VPC_SO_PROG_B_EN |
178 A6XX_VPC_SO_PROG_B_BUF(out->output_buffer) |
179 A6XX_VPC_SO_PROG_B_OFF(off * 4);
180 } else {
181 tf->prog[loc/2] |= A6XX_VPC_SO_PROG_A_EN |
182 A6XX_VPC_SO_PROG_A_BUF(out->output_buffer) |
183 A6XX_VPC_SO_PROG_A_OFF(off * 4);
184 }
185 }
186 }
187
188 tf->vpc_so_buf_cntl = A6XX_VPC_SO_BUF_CNTL_ENABLE |
189 COND(tf->ncomp[0] > 0, A6XX_VPC_SO_BUF_CNTL_BUF0) |
190 COND(tf->ncomp[1] > 0, A6XX_VPC_SO_BUF_CNTL_BUF1) |
191 COND(tf->ncomp[2] > 0, A6XX_VPC_SO_BUF_CNTL_BUF2) |
192 COND(tf->ncomp[3] > 0, A6XX_VPC_SO_BUF_CNTL_BUF3);
193 }
194
195 static void
196 setup_config_stateobj(struct fd_ringbuffer *ring, struct fd6_program_state *state)
197 {
198 OUT_PKT4(ring, REG_A6XX_HLSQ_UPDATE_CNTL, 1);
199 OUT_RING(ring, 0xff); /* XXX */
200
201 if (state->ds)
202 debug_assert(state->ds->constlen >= state->bs->constlen);
203 else
204 debug_assert(state->vs->constlen >= state->bs->constlen);
205
206 OUT_PKT4(ring, REG_A6XX_HLSQ_VS_CNTL, 4);
207 OUT_RING(ring, A6XX_HLSQ_VS_CNTL_CONSTLEN(align(state->vs->constlen, 4)) |
208 A6XX_HLSQ_VS_CNTL_ENABLED);
209 OUT_RING(ring, COND(state->hs,
210 A6XX_HLSQ_HS_CNTL_ENABLED |
211 A6XX_HLSQ_HS_CNTL_CONSTLEN(align(state->hs->constlen, 4))));
212 OUT_RING(ring, COND(state->ds,
213 A6XX_HLSQ_DS_CNTL_ENABLED |
214 A6XX_HLSQ_DS_CNTL_CONSTLEN(align(state->ds->constlen, 4))));
215 OUT_RING(ring, COND(state->gs,
216 A6XX_HLSQ_GS_CNTL_ENABLED |
217 A6XX_HLSQ_GS_CNTL_CONSTLEN(align(state->gs->constlen, 4))));
218 OUT_PKT4(ring, REG_A6XX_HLSQ_FS_CNTL, 1);
219 OUT_RING(ring, A6XX_HLSQ_FS_CNTL_CONSTLEN(align(state->fs->constlen, 4)) |
220 A6XX_HLSQ_FS_CNTL_ENABLED);
221
222 OUT_PKT4(ring, REG_A6XX_SP_VS_CONFIG, 1);
223 OUT_RING(ring, COND(state->vs, A6XX_SP_VS_CONFIG_ENABLED) |
224 A6XX_SP_VS_CONFIG_NIBO(state->vs->image_mapping.num_ibo) |
225 A6XX_SP_VS_CONFIG_NTEX(state->vs->num_samp) |
226 A6XX_SP_VS_CONFIG_NSAMP(state->vs->num_samp));
227
228 OUT_PKT4(ring, REG_A6XX_SP_HS_CONFIG, 1);
229 OUT_RING(ring, COND(state->hs,
230 A6XX_SP_HS_CONFIG_ENABLED |
231 A6XX_SP_HS_CONFIG_NIBO(state->hs->image_mapping.num_ibo) |
232 A6XX_SP_HS_CONFIG_NTEX(state->hs->num_samp) |
233 A6XX_SP_HS_CONFIG_NSAMP(state->hs->num_samp)));
234
235 OUT_PKT4(ring, REG_A6XX_SP_DS_CONFIG, 1);
236 OUT_RING(ring, COND(state->ds,
237 A6XX_SP_DS_CONFIG_ENABLED |
238 A6XX_SP_DS_CONFIG_NIBO(state->ds->image_mapping.num_ibo) |
239 A6XX_SP_DS_CONFIG_NTEX(state->ds->num_samp) |
240 A6XX_SP_DS_CONFIG_NSAMP(state->ds->num_samp)));
241
242 OUT_PKT4(ring, REG_A6XX_SP_GS_CONFIG, 1);
243 OUT_RING(ring, COND(state->gs,
244 A6XX_SP_GS_CONFIG_ENABLED |
245 A6XX_SP_GS_CONFIG_NIBO(state->gs->image_mapping.num_ibo) |
246 A6XX_SP_GS_CONFIG_NTEX(state->gs->num_samp) |
247 A6XX_SP_GS_CONFIG_NSAMP(state->gs->num_samp)));
248
249 OUT_PKT4(ring, REG_A6XX_SP_FS_CONFIG, 1);
250 OUT_RING(ring, COND(state->fs, A6XX_SP_FS_CONFIG_ENABLED) |
251 A6XX_SP_FS_CONFIG_NIBO(state->fs->image_mapping.num_ibo) |
252 A6XX_SP_FS_CONFIG_NTEX(state->fs->num_samp) |
253 A6XX_SP_FS_CONFIG_NSAMP(state->fs->num_samp));
254
255 OUT_PKT4(ring, REG_A6XX_SP_IBO_COUNT, 1);
256 OUT_RING(ring, state->fs->image_mapping.num_ibo);
257 }
258
259 #define VALIDREG(r) ((r) != regid(63,0))
260 #define CONDREG(r, val) COND(VALIDREG(r), (val))
261
262 static inline uint32_t
263 next_regid(uint32_t reg, uint32_t increment)
264 {
265 if (VALIDREG(reg))
266 return reg + increment;
267 else
268 return regid(63,0);
269 }
270
271 static void
272 setup_stateobj(struct fd_ringbuffer *ring, struct fd_screen *screen,
273 struct fd6_program_state *state, const struct ir3_shader_key *key,
274 bool binning_pass)
275 {
276 uint32_t pos_regid, psize_regid, color_regid[8], posz_regid;
277 uint32_t face_regid, coord_regid, zwcoord_regid, samp_id_regid;
278 uint32_t smask_in_regid, smask_regid;
279 uint32_t vertex_regid, instance_regid, layer_regid, primitive_regid;
280 uint32_t hs_invocation_regid;
281 uint32_t tess_coord_x_regid, tess_coord_y_regid, hs_patch_regid, ds_patch_regid;
282 uint32_t ij_pix_regid, ij_samp_regid, ij_cent_regid, ij_size_regid;
283 uint32_t gs_header_regid;
284 enum a3xx_threadsize fssz;
285 uint8_t psize_loc = ~0, pos_loc = ~0, layer_loc = ~0;
286 int i, j;
287
288 static const struct ir3_shader_variant dummy_fs = {0};
289 const struct ir3_shader_variant *vs = binning_pass ? state->bs : state->vs;
290 const struct ir3_shader_variant *hs = state->hs;
291 const struct ir3_shader_variant *ds = state->ds;
292 const struct ir3_shader_variant *gs = state->gs;
293 const struct ir3_shader_variant *fs = binning_pass ? &dummy_fs : state->fs;
294
295 if (binning_pass && state->ds)
296 ds = state->bs;
297 else if (binning_pass)
298 vs = state->bs;
299
300 bool sample_shading = fs->per_samp | key->sample_shading;
301
302 fssz = FOUR_QUADS;
303
304 pos_regid = ir3_find_output_regid(vs, VARYING_SLOT_POS);
305 psize_regid = ir3_find_output_regid(vs, VARYING_SLOT_PSIZ);
306 vertex_regid = ir3_find_sysval_regid(vs, SYSTEM_VALUE_VERTEX_ID);
307 instance_regid = ir3_find_sysval_regid(vs, SYSTEM_VALUE_INSTANCE_ID);
308
309 if (hs) {
310 tess_coord_x_regid = ir3_find_sysval_regid(ds, SYSTEM_VALUE_TESS_COORD);
311 tess_coord_y_regid = next_regid(tess_coord_x_regid, 1);
312 hs_patch_regid = ir3_find_sysval_regid(hs, SYSTEM_VALUE_PRIMITIVE_ID);
313 ds_patch_regid = ir3_find_sysval_regid(ds, SYSTEM_VALUE_PRIMITIVE_ID);
314 hs_invocation_regid = ir3_find_sysval_regid(hs, SYSTEM_VALUE_TCS_HEADER_IR3);
315
316 pos_regid = ir3_find_output_regid(ds, VARYING_SLOT_POS);
317 psize_regid = ir3_find_output_regid(ds, VARYING_SLOT_PSIZ);
318 } else {
319 tess_coord_x_regid = regid(63, 0);
320 tess_coord_y_regid = regid(63, 0);
321 hs_patch_regid = regid(63, 0);
322 ds_patch_regid = regid(63, 0);
323 hs_invocation_regid = regid(63, 0);
324 }
325
326 if (gs) {
327 gs_header_regid = ir3_find_sysval_regid(gs, SYSTEM_VALUE_GS_HEADER_IR3);
328 primitive_regid = ir3_find_sysval_regid(gs, SYSTEM_VALUE_PRIMITIVE_ID);
329 pos_regid = ir3_find_output_regid(gs, VARYING_SLOT_POS);
330 psize_regid = ir3_find_output_regid(gs, VARYING_SLOT_PSIZ);
331 layer_regid = ir3_find_output_regid(gs, VARYING_SLOT_LAYER);
332 } else {
333 gs_header_regid = regid(63, 0);
334 primitive_regid = regid(63, 0);
335 layer_regid = regid(63, 0);
336 }
337
338 if (fs->color0_mrt) {
339 color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
340 color_regid[4] = color_regid[5] = color_regid[6] = color_regid[7] =
341 ir3_find_output_regid(fs, FRAG_RESULT_COLOR);
342 } else {
343 color_regid[0] = ir3_find_output_regid(fs, FRAG_RESULT_DATA0);
344 color_regid[1] = ir3_find_output_regid(fs, FRAG_RESULT_DATA1);
345 color_regid[2] = ir3_find_output_regid(fs, FRAG_RESULT_DATA2);
346 color_regid[3] = ir3_find_output_regid(fs, FRAG_RESULT_DATA3);
347 color_regid[4] = ir3_find_output_regid(fs, FRAG_RESULT_DATA4);
348 color_regid[5] = ir3_find_output_regid(fs, FRAG_RESULT_DATA5);
349 color_regid[6] = ir3_find_output_regid(fs, FRAG_RESULT_DATA6);
350 color_regid[7] = ir3_find_output_regid(fs, FRAG_RESULT_DATA7);
351 }
352
353 samp_id_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_SAMPLE_ID);
354 smask_in_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_SAMPLE_MASK_IN);
355 face_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_FRONT_FACE);
356 coord_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_FRAG_COORD);
357 zwcoord_regid = next_regid(coord_regid, 2);
358 ij_pix_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_BARYCENTRIC_PIXEL);
359 ij_samp_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_BARYCENTRIC_SAMPLE);
360 ij_cent_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_BARYCENTRIC_CENTROID);
361 ij_size_regid = ir3_find_sysval_regid(fs, SYSTEM_VALUE_BARYCENTRIC_SIZE);
362 posz_regid = ir3_find_output_regid(fs, FRAG_RESULT_DEPTH);
363 smask_regid = ir3_find_output_regid(fs, FRAG_RESULT_SAMPLE_MASK);
364
365 /* If we have pre-dispatch texture fetches, then ij_pix should not
366 * be DCE'd, even if not actually used in the shader itself:
367 */
368 if (fs->num_sampler_prefetch > 0) {
369 assert(VALIDREG(ij_pix_regid));
370 /* also, it seems like ij_pix is *required* to be r0.x */
371 assert(ij_pix_regid == regid(0, 0));
372 }
373
374 /* we can't write gl_SampleMask for !msaa.. if b0 is zero then we
375 * end up masking the single sample!!
376 */
377 if (!key->msaa)
378 smask_regid = regid(63, 0);
379
380 /* we could probably divide this up into things that need to be
381 * emitted if frag-prog is dirty vs if vert-prog is dirty..
382 */
383
384 OUT_PKT4(ring, REG_A6XX_SP_HS_UNKNOWN_A833, 1);
385 OUT_RING(ring, 0x0);
386
387 OUT_PKT4(ring, REG_A6XX_SP_FS_PREFETCH_CNTL, 1 + fs->num_sampler_prefetch);
388 OUT_RING(ring, A6XX_SP_FS_PREFETCH_CNTL_COUNT(fs->num_sampler_prefetch) |
389 A6XX_SP_FS_PREFETCH_CNTL_UNK4(regid(63, 0)) |
390 0x7000); // XXX
391 for (int i = 0; i < fs->num_sampler_prefetch; i++) {
392 const struct ir3_sampler_prefetch *prefetch = &fs->sampler_prefetch[i];
393 OUT_RING(ring, A6XX_SP_FS_PREFETCH_CMD_SRC(prefetch->src) |
394 A6XX_SP_FS_PREFETCH_CMD_SAMP_ID(prefetch->samp_id) |
395 A6XX_SP_FS_PREFETCH_CMD_TEX_ID(prefetch->tex_id) |
396 A6XX_SP_FS_PREFETCH_CMD_DST(prefetch->dst) |
397 A6XX_SP_FS_PREFETCH_CMD_WRMASK(prefetch->wrmask) |
398 COND(prefetch->half_precision, A6XX_SP_FS_PREFETCH_CMD_HALF) |
399 A6XX_SP_FS_PREFETCH_CMD_CMD(prefetch->cmd));
400 }
401
402 OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A9A8, 1);
403 OUT_RING(ring, 0);
404
405 OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_AB00, 1);
406 OUT_RING(ring, 0x5);
407
408 OUT_PKT4(ring, REG_A6XX_SP_FS_OUTPUT_CNTL0, 1);
409 OUT_RING(ring, A6XX_SP_FS_OUTPUT_CNTL0_DEPTH_REGID(posz_regid) |
410 A6XX_SP_FS_OUTPUT_CNTL0_SAMPMASK_REGID(smask_regid) |
411 0xfc000000);
412
413 OUT_PKT4(ring, REG_A6XX_SP_VS_CTRL_REG0, 1);
414 OUT_RING(ring, A6XX_SP_VS_CTRL_REG0_THREADSIZE(fssz) |
415 A6XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(vs->info.max_reg + 1) |
416 A6XX_SP_VS_CTRL_REG0_MERGEDREGS |
417 A6XX_SP_VS_CTRL_REG0_BRANCHSTACK(vs->branchstack) |
418 COND(vs->need_pixlod, A6XX_SP_VS_CTRL_REG0_PIXLODENABLE));
419
420 struct ir3_shader_linkage l = {0};
421 const struct ir3_shader_variant *last_shader = fd6_last_shader(state);
422 ir3_link_shaders(&l, last_shader, fs);
423
424 BITSET_DECLARE(varbs, 128) = {0};
425 uint32_t *varmask = (uint32_t *)varbs;
426
427 for (i = 0; i < l.cnt; i++)
428 for (j = 0; j < util_last_bit(l.var[i].compmask); j++)
429 BITSET_SET(varbs, l.var[i].loc + j);
430
431 OUT_PKT4(ring, REG_A6XX_VPC_VAR_DISABLE(0), 4);
432 OUT_RING(ring, ~varmask[0]); /* VPC_VAR[0].DISABLE */
433 OUT_RING(ring, ~varmask[1]); /* VPC_VAR[1].DISABLE */
434 OUT_RING(ring, ~varmask[2]); /* VPC_VAR[2].DISABLE */
435 OUT_RING(ring, ~varmask[3]); /* VPC_VAR[3].DISABLE */
436
437 /* Add stream out outputs after computing the VPC_VAR_DISABLE bitmask. */
438 if (last_shader->shader->stream_output.num_outputs > 0)
439 link_stream_out(&l, last_shader);
440
441 if (VALIDREG(layer_regid)) {
442 layer_loc = l.max_loc;
443 ir3_link_add(&l, layer_regid, 0x1, l.max_loc);
444 }
445
446 if (VALIDREG(pos_regid)) {
447 pos_loc = l.max_loc;
448 ir3_link_add(&l, pos_regid, 0xf, l.max_loc);
449 }
450
451 if (VALIDREG(psize_regid)) {
452 psize_loc = l.max_loc;
453 ir3_link_add(&l, psize_regid, 0x1, l.max_loc);
454 }
455
456 if (last_shader->shader->stream_output.num_outputs > 0) {
457 setup_stream_out(state, last_shader, &l);
458 }
459
460 debug_assert(l.cnt < 32);
461 if (gs)
462 OUT_PKT4(ring, REG_A6XX_SP_GS_OUT_REG(0), DIV_ROUND_UP(l.cnt, 2));
463 else if (ds)
464 OUT_PKT4(ring, REG_A6XX_SP_DS_OUT_REG(0), DIV_ROUND_UP(l.cnt, 2));
465 else
466 OUT_PKT4(ring, REG_A6XX_SP_VS_OUT_REG(0), DIV_ROUND_UP(l.cnt, 2));
467
468 for (j = 0; j < l.cnt; ) {
469 uint32_t reg = 0;
470
471 reg |= A6XX_SP_VS_OUT_REG_A_REGID(l.var[j].regid);
472 reg |= A6XX_SP_VS_OUT_REG_A_COMPMASK(l.var[j].compmask);
473 j++;
474
475 reg |= A6XX_SP_VS_OUT_REG_B_REGID(l.var[j].regid);
476 reg |= A6XX_SP_VS_OUT_REG_B_COMPMASK(l.var[j].compmask);
477 j++;
478
479 OUT_RING(ring, reg);
480 }
481
482 if (gs)
483 OUT_PKT4(ring, REG_A6XX_SP_GS_VPC_DST_REG(0), DIV_ROUND_UP(l.cnt, 4));
484 else if (ds)
485 OUT_PKT4(ring, REG_A6XX_SP_DS_VPC_DST_REG(0), DIV_ROUND_UP(l.cnt, 4));
486 else
487 OUT_PKT4(ring, REG_A6XX_SP_VS_VPC_DST_REG(0), DIV_ROUND_UP(l.cnt, 4));
488
489 for (j = 0; j < l.cnt; ) {
490 uint32_t reg = 0;
491
492 reg |= A6XX_SP_VS_VPC_DST_REG_OUTLOC0(l.var[j++].loc);
493 reg |= A6XX_SP_VS_VPC_DST_REG_OUTLOC1(l.var[j++].loc);
494 reg |= A6XX_SP_VS_VPC_DST_REG_OUTLOC2(l.var[j++].loc);
495 reg |= A6XX_SP_VS_VPC_DST_REG_OUTLOC3(l.var[j++].loc);
496
497 OUT_RING(ring, reg);
498 }
499
500 fd6_emit_shader(ring, vs);
501 ir3_emit_immediates(screen, vs, ring);
502
503 if (hs) {
504 OUT_PKT4(ring, REG_A6XX_SP_HS_CTRL_REG0, 1);
505 OUT_RING(ring, A6XX_SP_HS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
506 A6XX_SP_HS_CTRL_REG0_FULLREGFOOTPRINT(hs->info.max_reg + 1) |
507 A6XX_SP_HS_CTRL_REG0_BRANCHSTACK(hs->branchstack) |
508 COND(hs->need_pixlod, A6XX_SP_HS_CTRL_REG0_PIXLODENABLE));
509
510 fd6_emit_shader(ring, hs);
511 ir3_emit_immediates(screen, hs, ring);
512 ir3_emit_link_map(screen, vs, hs, ring);
513
514 OUT_PKT4(ring, REG_A6XX_SP_DS_CTRL_REG0, 1);
515 OUT_RING(ring, A6XX_SP_DS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
516 A6XX_SP_DS_CTRL_REG0_FULLREGFOOTPRINT(ds->info.max_reg + 1) |
517 A6XX_SP_DS_CTRL_REG0_BRANCHSTACK(ds->branchstack) |
518 COND(ds->need_pixlod, A6XX_SP_DS_CTRL_REG0_PIXLODENABLE));
519
520 fd6_emit_shader(ring, ds);
521 ir3_emit_immediates(screen, ds, ring);
522 ir3_emit_link_map(screen, hs, ds, ring);
523
524 shader_info *hs_info = &hs->shader->nir->info;
525 OUT_PKT4(ring, REG_A6XX_PC_TESS_NUM_VERTEX, 1);
526 OUT_RING(ring, hs_info->tess.tcs_vertices_out);
527
528 /* Total attribute slots in HS incoming patch. */
529 OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9801, 1);
530 OUT_RING(ring, hs_info->tess.tcs_vertices_out * vs->shader->output_size / 4);
531
532 OUT_PKT4(ring, REG_A6XX_SP_HS_UNKNOWN_A831, 1);
533 OUT_RING(ring, vs->shader->output_size);
534
535 shader_info *ds_info = &ds->shader->nir->info;
536 OUT_PKT4(ring, REG_A6XX_PC_TESS_CNTL, 1);
537 uint32_t output;
538 if (ds_info->tess.point_mode)
539 output = TESS_POINTS;
540 else if (ds_info->tess.primitive_mode == GL_ISOLINES)
541 output = TESS_LINES;
542 else if (ds_info->tess.ccw)
543 output = TESS_CCW_TRIS;
544 else
545 output = TESS_CW_TRIS;
546
547 OUT_RING(ring, A6XX_PC_TESS_CNTL_SPACING(fd6_gl2spacing(ds_info->tess.spacing)) |
548 A6XX_PC_TESS_CNTL_OUTPUT(output));
549
550 /* xxx: Misc tess unknowns: */
551 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9103, 1);
552 OUT_RING(ring, 0x00ffff00);
553
554 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9106, 1);
555 OUT_RING(ring, 0x0000ffff);
556
557 OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_809D, 1);
558 OUT_RING(ring, 0x0);
559
560 OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_8002, 1);
561 OUT_RING(ring, 0x0);
562
563 OUT_PKT4(ring, REG_A6XX_VPC_PACK, 1);
564 OUT_RING(ring, A6XX_VPC_PACK_POSITIONLOC(pos_loc) |
565 A6XX_VPC_PACK_PSIZELOC(255) |
566 A6XX_VPC_PACK_STRIDE_IN_VPC(l.max_loc));
567
568 OUT_PKT4(ring, REG_A6XX_VPC_PACK_3, 1);
569 OUT_RING(ring, A6XX_VPC_PACK_3_POSITIONLOC(pos_loc) |
570 A6XX_VPC_PACK_3_PSIZELOC(psize_loc) |
571 A6XX_VPC_PACK_3_STRIDE_IN_VPC(l.max_loc));
572
573 OUT_PKT4(ring, REG_A6XX_SP_DS_PRIMITIVE_CNTL, 1);
574 OUT_RING(ring, A6XX_SP_DS_PRIMITIVE_CNTL_DSOUT(l.cnt));
575
576 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_4, 1);
577 OUT_RING(ring, A6XX_PC_PRIMITIVE_CNTL_4_STRIDE_IN_VPC(l.max_loc) |
578 CONDREG(psize_regid, 0x100));
579
580 } else {
581 OUT_PKT4(ring, REG_A6XX_SP_HS_UNKNOWN_A831, 1);
582 OUT_RING(ring, 0);
583 }
584
585 OUT_PKT4(ring, REG_A6XX_SP_PRIMITIVE_CNTL, 1);
586 OUT_RING(ring, A6XX_SP_PRIMITIVE_CNTL_VSOUT(l.cnt));
587
588 bool enable_varyings = fs->total_in > 0;
589
590 OUT_PKT4(ring, REG_A6XX_VPC_CNTL_0, 1);
591 OUT_RING(ring, A6XX_VPC_CNTL_0_NUMNONPOSVAR(fs->total_in) |
592 COND(enable_varyings, A6XX_VPC_CNTL_0_VARYING) |
593 0xff00ff00);
594
595 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_1, 1);
596 OUT_RING(ring, A6XX_PC_PRIMITIVE_CNTL_1_STRIDE_IN_VPC(l.max_loc) |
597 CONDREG(psize_regid, A6XX_PC_PRIMITIVE_CNTL_1_PSIZE));
598
599 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_3, 1);
600 OUT_RING(ring, 0);
601
602 OUT_PKT4(ring, REG_A6XX_HLSQ_CONTROL_1_REG, 5);
603 OUT_RING(ring, 0x7); /* XXX */
604 OUT_RING(ring, A6XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid) |
605 A6XX_HLSQ_CONTROL_2_REG_SAMPLEID(samp_id_regid) |
606 A6XX_HLSQ_CONTROL_2_REG_SAMPLEMASK(smask_in_regid) |
607 A6XX_HLSQ_CONTROL_2_REG_SIZE(ij_size_regid));
608 OUT_RING(ring, A6XX_HLSQ_CONTROL_3_REG_BARY_IJ_PIXEL(ij_pix_regid) |
609 A6XX_HLSQ_CONTROL_3_REG_BARY_IJ_CENTROID(ij_cent_regid) |
610 0xfc00fc00); /* XXX */
611 OUT_RING(ring, A6XX_HLSQ_CONTROL_4_REG_XYCOORDREGID(coord_regid) |
612 A6XX_HLSQ_CONTROL_4_REG_ZWCOORDREGID(zwcoord_regid) |
613 A6XX_HLSQ_CONTROL_4_REG_BARY_IJ_PIXEL_PERSAMP(ij_samp_regid) |
614 0x0000fc00); /* XXX */
615 OUT_RING(ring, 0xfc); /* XXX */
616
617 OUT_PKT4(ring, REG_A6XX_HLSQ_UNKNOWN_B980, 1);
618 OUT_RING(ring, enable_varyings ? 3 : 1);
619
620 OUT_PKT4(ring, REG_A6XX_SP_FS_CTRL_REG0, 1);
621 OUT_RING(ring, A6XX_SP_FS_CTRL_REG0_THREADSIZE(fssz) |
622 COND(enable_varyings, A6XX_SP_FS_CTRL_REG0_VARYING) |
623 COND(fs->frag_coord, A6XX_SP_FS_CTRL_REG0_VARYING) |
624 0x1000000 |
625 A6XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(fs->info.max_reg + 1) |
626 A6XX_SP_FS_CTRL_REG0_MERGEDREGS |
627 A6XX_SP_FS_CTRL_REG0_BRANCHSTACK(fs->branchstack) |
628 COND(fs->need_pixlod, A6XX_SP_FS_CTRL_REG0_PIXLODENABLE));
629
630 OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A982, 1);
631 OUT_RING(ring, 0); /* XXX */
632
633 OUT_PKT4(ring, REG_A6XX_VPC_GS_SIV_CNTL, 1);
634 OUT_RING(ring, 0x0000ffff); /* XXX */
635
636 OUT_PKT4(ring, REG_A6XX_GRAS_CNTL, 1);
637 OUT_RING(ring,
638 CONDREG(ij_pix_regid, A6XX_GRAS_CNTL_VARYING) |
639 CONDREG(ij_cent_regid, A6XX_GRAS_CNTL_CENTROID) |
640 CONDREG(ij_samp_regid, A6XX_GRAS_CNTL_PERSAMP_VARYING) |
641 COND(VALIDREG(ij_size_regid) && !sample_shading, A6XX_GRAS_CNTL_SIZE) |
642 COND(VALIDREG(ij_size_regid) && sample_shading, A6XX_GRAS_CNTL_SIZE_PERSAMP) |
643 COND(fs->frag_coord,
644 A6XX_GRAS_CNTL_SIZE |
645 A6XX_GRAS_CNTL_XCOORD |
646 A6XX_GRAS_CNTL_YCOORD |
647 A6XX_GRAS_CNTL_ZCOORD |
648 A6XX_GRAS_CNTL_WCOORD) |
649 COND(fs->frag_face, A6XX_GRAS_CNTL_SIZE));
650
651 OUT_PKT4(ring, REG_A6XX_RB_RENDER_CONTROL0, 2);
652 OUT_RING(ring,
653 CONDREG(ij_pix_regid, A6XX_RB_RENDER_CONTROL0_VARYING) |
654 CONDREG(ij_cent_regid, A6XX_RB_RENDER_CONTROL0_CENTROID) |
655 CONDREG(ij_samp_regid, A6XX_RB_RENDER_CONTROL0_PERSAMP_VARYING) |
656 COND(enable_varyings, A6XX_RB_RENDER_CONTROL0_UNK10) |
657 COND(VALIDREG(ij_size_regid) && !sample_shading, A6XX_RB_RENDER_CONTROL0_SIZE) |
658 COND(VALIDREG(ij_size_regid) && sample_shading, A6XX_RB_RENDER_CONTROL0_SIZE_PERSAMP) |
659 COND(fs->frag_coord,
660 A6XX_RB_RENDER_CONTROL0_SIZE |
661 A6XX_RB_RENDER_CONTROL0_XCOORD |
662 A6XX_RB_RENDER_CONTROL0_YCOORD |
663 A6XX_RB_RENDER_CONTROL0_ZCOORD |
664 A6XX_RB_RENDER_CONTROL0_WCOORD) |
665 COND(fs->frag_face, A6XX_RB_RENDER_CONTROL0_SIZE));
666
667 OUT_RING(ring,
668 CONDREG(smask_in_regid, A6XX_RB_RENDER_CONTROL1_SAMPLEMASK) |
669 CONDREG(samp_id_regid, A6XX_RB_RENDER_CONTROL1_SAMPLEID) |
670 CONDREG(ij_size_regid, A6XX_RB_RENDER_CONTROL1_SIZE) |
671 COND(fs->frag_face, A6XX_RB_RENDER_CONTROL1_FACENESS));
672
673 OUT_PKT4(ring, REG_A6XX_RB_SAMPLE_CNTL, 1);
674 OUT_RING(ring, COND(sample_shading, A6XX_RB_SAMPLE_CNTL_PER_SAMP_MODE));
675
676 OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_8101, 1);
677 OUT_RING(ring, COND(sample_shading, 0x6)); // XXX
678
679 OUT_PKT4(ring, REG_A6XX_GRAS_SAMPLE_CNTL, 1);
680 OUT_RING(ring, COND(sample_shading, A6XX_GRAS_SAMPLE_CNTL_PER_SAMP_MODE));
681
682 OUT_PKT4(ring, REG_A6XX_SP_FS_OUTPUT_REG(0), 8);
683 for (i = 0; i < 8; i++) {
684 OUT_RING(ring, A6XX_SP_FS_OUTPUT_REG_REGID(color_regid[i]) |
685 COND(color_regid[i] & HALF_REG_ID, A6XX_SP_FS_OUTPUT_REG_HALF_PRECISION));
686 }
687
688 OUT_PKT4(ring, REG_A6XX_VPC_PACK, 1);
689 OUT_RING(ring, A6XX_VPC_PACK_POSITIONLOC(pos_loc) |
690 A6XX_VPC_PACK_PSIZELOC(psize_loc) |
691 A6XX_VPC_PACK_STRIDE_IN_VPC(l.max_loc));
692
693 if (gs) {
694 OUT_PKT4(ring, REG_A6XX_SP_GS_CTRL_REG0, 1);
695 OUT_RING(ring, A6XX_SP_GS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
696 A6XX_SP_GS_CTRL_REG0_FULLREGFOOTPRINT(gs->info.max_reg + 1) |
697 A6XX_SP_GS_CTRL_REG0_BRANCHSTACK(gs->branchstack) |
698 COND(gs->need_pixlod, A6XX_SP_GS_CTRL_REG0_PIXLODENABLE));
699
700 fd6_emit_shader(ring, gs);
701 ir3_emit_immediates(screen, gs, ring);
702 if (ds)
703 ir3_emit_link_map(screen, ds, gs, ring);
704 else
705 ir3_emit_link_map(screen, vs, gs, ring);
706
707 OUT_PKT4(ring, REG_A6XX_VPC_PACK_GS, 1);
708 OUT_RING(ring, A6XX_VPC_PACK_GS_POSITIONLOC(pos_loc) |
709 A6XX_VPC_PACK_GS_PSIZELOC(psize_loc) |
710 A6XX_VPC_PACK_GS_STRIDE_IN_VPC(l.max_loc));
711
712 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9105, 1);
713 OUT_RING(ring, A6XX_VPC_UNKNOWN_9105_LAYERLOC(layer_loc) | 0xff00);
714
715 OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_809C, 1);
716 OUT_RING(ring, CONDREG(layer_regid, A6XX_GRAS_UNKNOWN_809C_GS_WRITES_LAYER));
717
718 uint32_t flags_regid = ir3_find_output_regid(gs, VARYING_SLOT_GS_VERTEX_FLAGS_IR3);
719
720 OUT_PKT4(ring, REG_A6XX_SP_PRIMITIVE_CNTL_GS, 1);
721 OUT_RING(ring, A6XX_SP_PRIMITIVE_CNTL_GS_GSOUT(l.cnt) |
722 A6XX_SP_PRIMITIVE_CNTL_GS_FLAGS_REGID(flags_regid));
723
724 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_2, 1);
725 OUT_RING(ring, A6XX_PC_PRIMITIVE_CNTL_2_STRIDE_IN_VPC(l.max_loc) |
726 CONDREG(psize_regid, A6XX_PC_PRIMITIVE_CNTL_2_PSIZE) |
727 CONDREG(layer_regid, A6XX_PC_PRIMITIVE_CNTL_2_LAYER) |
728 CONDREG(primitive_regid, A6XX_PC_PRIMITIVE_CNTL_2_PRIMITIVE_ID));
729
730 uint32_t output;
731 switch (gs->shader->nir->info.gs.output_primitive) {
732 case GL_POINTS:
733 output = TESS_POINTS;
734 break;
735 case GL_LINE_STRIP:
736 output = TESS_LINES;
737 break;
738 case GL_TRIANGLE_STRIP:
739 output = TESS_CW_TRIS;
740 break;
741 default:
742 unreachable("");
743 }
744 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_5, 1);
745 OUT_RING(ring,
746 A6XX_PC_PRIMITIVE_CNTL_5_GS_VERTICES_OUT(gs->shader->nir->info.gs.vertices_out - 1) |
747 A6XX_PC_PRIMITIVE_CNTL_5_GS_OUTPUT(output) |
748 A6XX_PC_PRIMITIVE_CNTL_5_GS_INVOCATIONS(gs->shader->nir->info.gs.invocations - 1));
749
750 OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_8003, 1);
751 OUT_RING(ring, 0);
752
753 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9100, 1);
754 OUT_RING(ring, 0xff);
755
756 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9102, 1);
757 OUT_RING(ring, 0xffff00);
758
759 const struct ir3_shader_variant *prev = state->ds ? state->ds : state->vs;
760
761 /* Size of per-primitive alloction in ldlw memory in vec4s. */
762 uint32_t vec4_size =
763 gs->shader->nir->info.gs.vertices_in *
764 DIV_ROUND_UP(prev->shader->output_size, 4);
765 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_6, 1);
766 OUT_RING(ring, A6XX_PC_PRIMITIVE_CNTL_6_STRIDE_IN_VPC(vec4_size));
767
768 OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9B07, 1);
769 OUT_RING(ring, 0);
770
771 OUT_PKT4(ring, REG_A6XX_SP_GS_UNKNOWN_A871, 1);
772 OUT_RING(ring, prev->shader->output_size);
773 } else {
774 OUT_PKT4(ring, REG_A6XX_PC_PRIMITIVE_CNTL_6, 1);
775 OUT_RING(ring, 0);
776 OUT_PKT4(ring, REG_A6XX_SP_GS_UNKNOWN_A871, 1);
777 OUT_RING(ring, 0);
778 }
779
780 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9101, 1);
781 OUT_RING(ring, 0xffff00);
782
783 OUT_PKT4(ring, REG_A6XX_VPC_UNKNOWN_9107, 1);
784 OUT_RING(ring, 0);
785
786
787 if (!binning_pass) {
788 /* figure out VARYING_INTERP / VARYING_PS_REPL register values: */
789 for (j = -1; (j = ir3_next_varying(fs, j)) < (int)fs->inputs_count; ) {
790 /* NOTE: varyings are packed, so if compmask is 0xb
791 * then first, third, and fourth component occupy
792 * three consecutive varying slots:
793 */
794 unsigned compmask = fs->inputs[j].compmask;
795
796 uint32_t inloc = fs->inputs[j].inloc;
797
798 if (fs->inputs[j].interpolate == INTERP_MODE_FLAT) {
799 uint32_t loc = inloc;
800
801 for (i = 0; i < 4; i++) {
802 if (compmask & (1 << i)) {
803 state->vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
804 loc++;
805 }
806 }
807 }
808 }
809 }
810
811 if (fs->instrlen)
812 fd6_emit_shader(ring, fs);
813
814 OUT_PKT4(ring, REG_A6XX_VFD_CONTROL_1, 6);
815 OUT_RING(ring, A6XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
816 A6XX_VFD_CONTROL_1_REGID4INST(instance_regid) |
817 A6XX_VFD_CONTROL_1_REGID4PRIMID(primitive_regid) |
818 0xfc000000);
819 OUT_RING(ring, A6XX_VFD_CONTROL_2_REGID_HSPATCHID(hs_patch_regid) |
820 A6XX_VFD_CONTROL_2_REGID_INVOCATIONID(hs_invocation_regid));
821 OUT_RING(ring, A6XX_VFD_CONTROL_3_REGID_DSPATCHID(ds_patch_regid) |
822 A6XX_VFD_CONTROL_3_REGID_TESSX(tess_coord_x_regid) |
823 A6XX_VFD_CONTROL_3_REGID_TESSY(tess_coord_y_regid) |
824 0xfc);
825 OUT_RING(ring, 0x000000fc); /* VFD_CONTROL_4 */
826 OUT_RING(ring, A6XX_VFD_CONTROL_5_REGID_GSHEADER(gs_header_regid) |
827 0xfc00); /* VFD_CONTROL_5 */
828 OUT_RING(ring, 0x00000000); /* VFD_CONTROL_6 */
829
830 bool fragz = fs->no_earlyz | fs->writes_pos;
831
832 OUT_PKT4(ring, REG_A6XX_RB_DEPTH_PLANE_CNTL, 1);
833 OUT_RING(ring, COND(fragz, A6XX_RB_DEPTH_PLANE_CNTL_FRAG_WRITES_Z));
834
835 OUT_PKT4(ring, REG_A6XX_GRAS_SU_DEPTH_PLANE_CNTL, 1);
836 OUT_RING(ring, COND(fragz, A6XX_GRAS_SU_DEPTH_PLANE_CNTL_FRAG_WRITES_Z));
837
838 if (!binning_pass)
839 ir3_emit_immediates(screen, fs, ring);
840 }
841
842 /* emits the program state which is not part of the stateobj because of
843 * dependency on other gl state (rasterflat or sprite-coord-replacement)
844 */
845 void
846 fd6_program_emit(struct fd_ringbuffer *ring, struct fd6_emit *emit)
847 {
848 const struct fd6_program_state *state = fd6_emit_get_prog(emit);
849
850 if (!unlikely(emit->rasterflat || emit->sprite_coord_enable)) {
851 /* fastpath: */
852 OUT_PKT4(ring, REG_A6XX_VPC_VARYING_INTERP_MODE(0), 8);
853 for (int i = 0; i < 8; i++)
854 OUT_RING(ring, state->vinterp[i]); /* VPC_VARYING_INTERP[i].MODE */
855
856 OUT_PKT4(ring, REG_A6XX_VPC_VARYING_PS_REPL_MODE(0), 8);
857 for (int i = 0; i < 8; i++)
858 OUT_RING(ring, 0x00000000); /* VPC_VARYING_PS_REPL[i] */
859 } else {
860 /* slow-path: */
861 struct ir3_shader_variant *fs = state->fs;
862 uint32_t vinterp[8], vpsrepl[8];
863
864 memset(vinterp, 0, sizeof(vinterp));
865 memset(vpsrepl, 0, sizeof(vpsrepl));
866
867 for (int j = -1; (j = ir3_next_varying(fs, j)) < (int)fs->inputs_count; ) {
868
869 /* NOTE: varyings are packed, so if compmask is 0xb
870 * then first, third, and fourth component occupy
871 * three consecutive varying slots:
872 */
873 unsigned compmask = fs->inputs[j].compmask;
874
875 uint32_t inloc = fs->inputs[j].inloc;
876
877 if ((fs->inputs[j].interpolate == INTERP_MODE_FLAT) ||
878 (fs->inputs[j].rasterflat && emit->rasterflat)) {
879 uint32_t loc = inloc;
880
881 for (int i = 0; i < 4; i++) {
882 if (compmask & (1 << i)) {
883 vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
884 loc++;
885 }
886 }
887 }
888
889 gl_varying_slot slot = fs->inputs[j].slot;
890
891 /* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
892 if (slot >= VARYING_SLOT_VAR0) {
893 unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
894 /* Replace the .xy coordinates with S/T from the point sprite. Set
895 * interpolation bits for .zw such that they become .01
896 */
897 if (emit->sprite_coord_enable & texmask) {
898 /* mask is two 2-bit fields, where:
899 * '01' -> S
900 * '10' -> T
901 * '11' -> 1 - T (flip mode)
902 */
903 unsigned mask = emit->sprite_coord_mode ? 0b1101 : 0b1001;
904 uint32_t loc = inloc;
905 if (compmask & 0x1) {
906 vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
907 loc++;
908 }
909 if (compmask & 0x2) {
910 vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
911 loc++;
912 }
913 if (compmask & 0x4) {
914 /* .z <- 0.0f */
915 vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
916 loc++;
917 }
918 if (compmask & 0x8) {
919 /* .w <- 1.0f */
920 vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
921 loc++;
922 }
923 }
924 }
925 }
926
927 OUT_PKT4(ring, REG_A6XX_VPC_VARYING_INTERP_MODE(0), 8);
928 for (int i = 0; i < 8; i++)
929 OUT_RING(ring, vinterp[i]); /* VPC_VARYING_INTERP[i].MODE */
930
931 OUT_PKT4(ring, REG_A6XX_VPC_VARYING_PS_REPL_MODE(0), 8);
932 for (int i = 0; i < 8; i++)
933 OUT_RING(ring, vpsrepl[i]); /* VPC_VARYING_PS_REPL[i] */
934 }
935 }
936
937 static struct ir3_program_state *
938 fd6_program_create(void *data, struct ir3_shader_variant *bs,
939 struct ir3_shader_variant *vs,
940 struct ir3_shader_variant *hs,
941 struct ir3_shader_variant *ds,
942 struct ir3_shader_variant *gs,
943 struct ir3_shader_variant *fs,
944 const struct ir3_shader_key *key)
945 {
946 struct fd_context *ctx = data;
947 struct fd6_program_state *state = CALLOC_STRUCT(fd6_program_state);
948
949 /* if we have streamout, use full VS in binning pass, as the
950 * binning pass VS will have outputs on other than position/psize
951 * stripped out:
952 */
953 state->bs = vs->shader->stream_output.num_outputs ? vs : bs;
954 state->vs = vs;
955 state->hs = hs;
956 state->ds = ds;
957 state->gs = gs;
958 state->fs = fs;
959 state->config_stateobj = fd_ringbuffer_new_object(ctx->pipe, 0x1000);
960 state->binning_stateobj = fd_ringbuffer_new_object(ctx->pipe, 0x1000);
961 state->stateobj = fd_ringbuffer_new_object(ctx->pipe, 0x1000);
962
963 #ifdef DEBUG
964 if (!ds) {
965 for (unsigned i = 0; i < bs->inputs_count; i++) {
966 if (vs->inputs[i].sysval)
967 continue;
968 debug_assert(bs->inputs[i].regid == vs->inputs[i].regid);
969 }
970 }
971 #endif
972
973 setup_config_stateobj(state->config_stateobj, state);
974 setup_stateobj(state->binning_stateobj, ctx->screen, state, key, true);
975 setup_stateobj(state->stateobj, ctx->screen, state, key, false);
976
977 return &state->base;
978 }
979
980 static void
981 fd6_program_destroy(void *data, struct ir3_program_state *state)
982 {
983 struct fd6_program_state *so = fd6_program_state(state);
984 fd_ringbuffer_del(so->stateobj);
985 fd_ringbuffer_del(so->binning_stateobj);
986 fd_ringbuffer_del(so->config_stateobj);
987 free(so);
988 }
989
990 static const struct ir3_cache_funcs cache_funcs = {
991 .create_state = fd6_program_create,
992 .destroy_state = fd6_program_destroy,
993 };
994
995 static void *
996 fd6_shader_state_create(struct pipe_context *pctx, const struct pipe_shader_state *cso)
997 {
998 struct fd_context *ctx = fd_context(pctx);
999 struct ir3_compiler *compiler = ctx->screen->compiler;
1000 struct ir3_shader *shader =
1001 ir3_shader_create(compiler, cso, &ctx->debug, pctx->screen);
1002 unsigned packets, size;
1003
1004 /* pre-calculate size required for userconst stateobj: */
1005 ir3_user_consts_size(&shader->ubo_state, &packets, &size);
1006
1007 /* also account for UBO addresses: */
1008 packets += 1;
1009 size += 2 * shader->const_state.num_ubos;
1010
1011 unsigned sizedwords = (4 * packets) + size;
1012 shader->ubo_state.cmdstream_size = sizedwords * 4;
1013
1014 return shader;
1015 }
1016
1017 static void
1018 fd6_shader_state_delete(struct pipe_context *pctx, void *hwcso)
1019 {
1020 struct ir3_shader *so = hwcso;
1021 struct fd_context *ctx = fd_context(pctx);
1022 ir3_cache_invalidate(fd6_context(ctx)->shader_cache, hwcso);
1023 ir3_shader_destroy(so);
1024 }
1025
1026 void
1027 fd6_prog_init(struct pipe_context *pctx)
1028 {
1029 struct fd_context *ctx = fd_context(pctx);
1030
1031 fd6_context(ctx)->shader_cache = ir3_cache_create(&cache_funcs, ctx);
1032
1033 pctx->create_vs_state = fd6_shader_state_create;
1034 pctx->delete_vs_state = fd6_shader_state_delete;
1035
1036 pctx->create_tcs_state = fd6_shader_state_create;
1037 pctx->delete_tcs_state = fd6_shader_state_delete;
1038
1039 pctx->create_tes_state = fd6_shader_state_create;
1040 pctx->delete_tes_state = fd6_shader_state_delete;
1041
1042 pctx->create_gs_state = fd6_shader_state_create;
1043 pctx->delete_gs_state = fd6_shader_state_delete;
1044
1045 pctx->create_gs_state = fd6_shader_state_create;
1046 pctx->delete_gs_state = fd6_shader_state_delete;
1047
1048 pctx->create_fs_state = fd6_shader_state_create;
1049 pctx->delete_fs_state = fd6_shader_state_delete;
1050
1051 fd_prog_init(pctx);
1052 }