etnaviv: fix two-sided stencil
[mesa.git] / src / gallium / drivers / etnaviv / etnaviv_emit.c
1 /*
2 * Copyright (c) 2014-2015 Etnaviv Project
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sub license,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the
12 * next paragraph) shall be included in all copies or substantial portions
13 * of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Wladimir J. van der Laan <laanwj@gmail.com>
25 */
26
27 #include "etnaviv_emit.h"
28
29 #include "etnaviv_blend.h"
30 #include "etnaviv_compiler.h"
31 #include "etnaviv_context.h"
32 #include "etnaviv_rasterizer.h"
33 #include "etnaviv_resource.h"
34 #include "etnaviv_rs.h"
35 #include "etnaviv_screen.h"
36 #include "etnaviv_shader.h"
37 #include "etnaviv_texture.h"
38 #include "etnaviv_translate.h"
39 #include "etnaviv_uniforms.h"
40 #include "etnaviv_util.h"
41 #include "etnaviv_zsa.h"
42 #include "hw/common.xml.h"
43 #include "hw/state.xml.h"
44 #include "hw/state_blt.xml.h"
45 #include "util/u_math.h"
46
47 /* Queue a STALL command (queues 2 words) */
48 static inline void
49 CMD_STALL(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
50 {
51 etna_cmd_stream_emit(stream, VIV_FE_STALL_HEADER_OP_STALL);
52 etna_cmd_stream_emit(stream, VIV_FE_STALL_TOKEN_FROM(from) | VIV_FE_STALL_TOKEN_TO(to));
53 }
54
55 void
56 etna_stall(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
57 {
58 bool blt = (from == SYNC_RECIPIENT_BLT) || (to == SYNC_RECIPIENT_BLT);
59 etna_cmd_stream_reserve(stream, blt ? 8 : 4);
60
61 if (blt) {
62 etna_emit_load_state(stream, VIVS_BLT_ENABLE >> 2, 1, 0);
63 etna_cmd_stream_emit(stream, 1);
64 }
65
66 /* TODO: set bit 28/29 of token after BLT COPY_BUFFER */
67 etna_emit_load_state(stream, VIVS_GL_SEMAPHORE_TOKEN >> 2, 1, 0);
68 etna_cmd_stream_emit(stream, VIVS_GL_SEMAPHORE_TOKEN_FROM(from) | VIVS_GL_SEMAPHORE_TOKEN_TO(to));
69
70 if (from == SYNC_RECIPIENT_FE) {
71 /* if the frontend is to be stalled, queue a STALL frontend command */
72 CMD_STALL(stream, from, to);
73 } else {
74 /* otherwise, load the STALL token state */
75 etna_emit_load_state(stream, VIVS_GL_STALL_TOKEN >> 2, 1, 0);
76 etna_cmd_stream_emit(stream, VIVS_GL_STALL_TOKEN_FROM(from) | VIVS_GL_STALL_TOKEN_TO(to));
77 }
78
79 if (blt) {
80 etna_emit_load_state(stream, VIVS_BLT_ENABLE >> 2, 1, 0);
81 etna_cmd_stream_emit(stream, 0);
82 }
83 }
84
85 #define EMIT_STATE(state_name, src_value) \
86 etna_coalsence_emit(stream, &coalesce, VIVS_##state_name, src_value)
87
88 #define EMIT_STATE_FIXP(state_name, src_value) \
89 etna_coalsence_emit_fixp(stream, &coalesce, VIVS_##state_name, src_value)
90
91 #define EMIT_STATE_RELOC(state_name, src_value) \
92 etna_coalsence_emit_reloc(stream, &coalesce, VIVS_##state_name, src_value)
93
94 #define ETNA_3D_CONTEXT_SIZE (400) /* keep this number above "Total state updates (fixed)" from gen_weave_state tool */
95
96 static unsigned
97 required_stream_size(struct etna_context *ctx)
98 {
99 unsigned size = ETNA_3D_CONTEXT_SIZE;
100
101 /* stall + flush */
102 size += 2 + 4;
103
104 /* vertex elements */
105 size += ctx->vertex_elements->num_elements + 1;
106
107 /* uniforms - worst case (2 words per uniform load) */
108 size += ctx->shader.vs->uniforms.imm_count * 2;
109 size += ctx->shader.fs->uniforms.imm_count * 2;
110
111 /* shader */
112 size += ctx->shader_state.vs_inst_mem_size + 1;
113 size += ctx->shader_state.ps_inst_mem_size + 1;
114
115 /* DRAW_INDEXED_PRIMITIVES command */
116 size += 6;
117
118 /* reserve for alignment etc. */
119 size += 64;
120
121 return size;
122 }
123
124 /* Emit state that only exists on HALTI5+ */
125 static void
126 emit_halti5_only_state(struct etna_context *ctx, int vs_output_count)
127 {
128 struct etna_cmd_stream *stream = ctx->stream;
129 uint32_t dirty = ctx->dirty;
130 struct etna_coalesce coalesce;
131
132 etna_coalesce_start(stream, &coalesce);
133 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
134 /* Magic states (load balancing, inter-unit sync, buffers) */
135 /*00870*/ EMIT_STATE(VS_HALTI5_OUTPUT_COUNT, vs_output_count | ((vs_output_count * 0x10) << 8));
136 /*008A0*/ EMIT_STATE(VS_HALTI5_UNK008A0, 0x0001000e | ((0x110/vs_output_count) << 20));
137 for (int x = 0; x < 4; ++x) {
138 /*008E0*/ EMIT_STATE(VS_HALTI5_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
139 }
140 }
141 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
142 for (int x = 0; x < 4; ++x) {
143 /*008C0*/ EMIT_STATE(VS_HALTI5_INPUT(x), ctx->shader_state.VS_INPUT[x]);
144 }
145 }
146 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
147 /*00A90*/ EMIT_STATE(PA_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
148 /*00AA8*/ EMIT_STATE(PA_VS_OUTPUT_COUNT, vs_output_count);
149 /*01080*/ EMIT_STATE(PS_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
150 /*03888*/ EMIT_STATE(GL_HALTI5_SH_SPECIALS, ctx->shader_state.GL_HALTI5_SH_SPECIALS);
151 }
152 etna_coalesce_end(stream, &coalesce);
153 }
154
155 /* Emit state that no longer exists on HALTI5 */
156 static void
157 emit_pre_halti5_state(struct etna_context *ctx)
158 {
159 struct etna_cmd_stream *stream = ctx->stream;
160 uint32_t dirty = ctx->dirty;
161 struct etna_coalesce coalesce;
162
163 etna_coalesce_start(stream, &coalesce);
164 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
165 /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
166 }
167 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
168 for (int x = 0; x < 4; ++x) {
169 /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
170 }
171 }
172 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
173 for (int x = 0; x < 4; ++x) {
174 /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
175 }
176 }
177 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
178 /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
179 }
180 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
181 for (int x = 0; x < 10; ++x) {
182 /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
183 }
184 }
185 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
186 /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
187 for (int x = 0; x < 4; ++x) {
188 /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
189 }
190 for (int x = 0; x < 16; ++x) {
191 /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
192 }
193 }
194 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
195 /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
196 }
197 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
198 /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
199 }
200 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
201 /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
202 for (int x = 0; x < 2; ++x) {
203 /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
204 }
205 }
206 etna_coalesce_end(stream, &coalesce);
207 }
208
209 /* Weave state before draw operation. This function merges all the compiled
210 * state blocks under the context into one device register state. Parts of
211 * this state that are changed since last call (dirty) will be uploaded as
212 * state changes in the command buffer. */
213 void
214 etna_emit_state(struct etna_context *ctx)
215 {
216 struct etna_cmd_stream *stream = ctx->stream;
217 unsigned ccw = ctx->rasterizer->front_ccw;
218
219
220 /* Pre-reserve the command buffer space which we are likely to need.
221 * This must cover all the state emitted below, and the following
222 * draw command. */
223 etna_cmd_stream_reserve(stream, required_stream_size(ctx));
224
225 uint32_t dirty = ctx->dirty;
226
227 /* Pre-processing: see what caches we need to flush before making state changes. */
228 uint32_t to_flush = 0;
229 if (unlikely(dirty & (ETNA_DIRTY_BLEND)))
230 to_flush |= VIVS_GL_FLUSH_CACHE_COLOR;
231 if (unlikely(dirty & (ETNA_DIRTY_TEXTURE_CACHES)))
232 to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE;
233 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) /* Framebuffer config changed? */
234 to_flush |= VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
235 if (DBG_ENABLED(ETNA_DBG_CFLUSH_ALL))
236 to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE | VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
237
238 if (to_flush) {
239 etna_set_state(stream, VIVS_GL_FLUSH_CACHE, to_flush);
240 etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
241 }
242
243 /* Flush TS cache before changing TS configuration. */
244 if (unlikely(dirty & ETNA_DIRTY_TS)) {
245 etna_set_state(stream, VIVS_TS_FLUSH_CACHE, VIVS_TS_FLUSH_CACHE_FLUSH);
246 }
247
248 /* Update vertex elements. This is different from any of the other states, in that
249 * a) the number of vertex elements written matters: so write only active ones
250 * b) the vertex element states must all be written: do not skip entries that stay the same */
251 if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
252 if (ctx->specs.halti >= 5) {
253 /*17800*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG0(0),
254 ctx->vertex_elements->num_elements,
255 ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG0);
256 /*17A00*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_SCALE(0),
257 ctx->vertex_elements->num_elements,
258 ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
259 /*17A80*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG1(0),
260 ctx->vertex_elements->num_elements,
261 ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG1);
262 } else {
263 /* Special case: vertex elements must always be sent in full if changed */
264 /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
265 ctx->vertex_elements->num_elements,
266 ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
267 if (ctx->specs.halti >= 2) {
268 /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
269 ctx->vertex_elements->num_elements,
270 ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
271 }
272 }
273 }
274 unsigned vs_output_count = etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex
275 ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
276 : ctx->shader_state.VS_OUTPUT_COUNT;
277
278 /* The following code is originally generated by gen_merge_state.py, to
279 * emit state in increasing order of address (this makes it possible to merge
280 * consecutive register updates into one SET_STATE command)
281 *
282 * There have been some manual changes, where the weaving operation is not
283 * simply bitwise or:
284 * - scissor fixp
285 * - num vertex elements
286 * - scissor handling
287 * - num samplers
288 * - texture lod
289 * - ETNA_DIRTY_TS
290 * - removed ETNA_DIRTY_BASE_SETUP statements -- these are guaranteed to not
291 * change anyway
292 * - PS / framebuffer interaction for MSAA
293 * - move update of GL_MULTI_SAMPLE_CONFIG first
294 * - add unlikely()/likely()
295 */
296 struct etna_coalesce coalesce;
297
298 etna_coalesce_start(stream, &coalesce);
299
300 /* begin only EMIT_STATE -- make sure no new etna_reserve calls are done here
301 * directly
302 * or indirectly */
303 /* multi sample config is set first, and outside of the normal sorting
304 * order, as changing the multisample state clobbers PS.INPUT_COUNT (and
305 * possibly PS.TEMP_REGISTER_CONTROL).
306 */
307 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SAMPLE_MASK))) {
308 uint32_t val = VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_ENABLES(ctx->sample_mask);
309 val |= ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG;
310
311 /*03818*/ EMIT_STATE(GL_MULTI_SAMPLE_CONFIG, val);
312 }
313 if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
314 /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
315 /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
316 }
317 if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
318 /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
319 }
320 if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
321 if (ctx->specs.halti >= 2) { /* HALTI2+: NFE_VERTEX_STREAMS */
322 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
323 /*14600*/ EMIT_STATE_RELOC(NFE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
324 }
325 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
326 if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
327 /*14640*/ EMIT_STATE(NFE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
328 }
329 }
330 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
331 if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
332 /*14680*/ EMIT_STATE(NFE_VERTEX_STREAMS_VERTEX_DIVISOR(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_UNK14680);
333 }
334 }
335 } else if(ctx->specs.stream_count >= 1) { /* hw w/ multiple vertex streams */
336 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
337 /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
338 }
339 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
340 if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
341 /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
342 }
343 }
344 } else { /* hw w/ single vertex stream */
345 /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
346 /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
347 }
348 }
349 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
350
351 /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT, vs_output_count);
352 }
353 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
354 /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
355 /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
356 }
357 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
358 /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
359 }
360 if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
361 /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
362 /*00A04*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_Y, ctx->viewport.PA_VIEWPORT_SCALE_Y);
363 /*00A08*/ EMIT_STATE(PA_VIEWPORT_SCALE_Z, ctx->viewport.PA_VIEWPORT_SCALE_Z);
364 /*00A0C*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_X, ctx->viewport.PA_VIEWPORT_OFFSET_X);
365 /*00A10*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_Y, ctx->viewport.PA_VIEWPORT_OFFSET_Y);
366 /*00A14*/ EMIT_STATE(PA_VIEWPORT_OFFSET_Z, ctx->viewport.PA_VIEWPORT_OFFSET_Z);
367 }
368 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
369 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
370
371 /*00A18*/ EMIT_STATE(PA_LINE_WIDTH, rasterizer->PA_LINE_WIDTH);
372 /*00A1C*/ EMIT_STATE(PA_POINT_SIZE, rasterizer->PA_POINT_SIZE);
373 /*00A28*/ EMIT_STATE(PA_SYSTEM_MODE, rasterizer->PA_SYSTEM_MODE);
374 }
375 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
376 /*00A30*/ EMIT_STATE(PA_ATTRIBUTE_ELEMENT_COUNT, ctx->shader_state.PA_ATTRIBUTE_ELEMENT_COUNT);
377 }
378 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_SHADER))) {
379 uint32_t val = etna_rasterizer_state(ctx->rasterizer)->PA_CONFIG;
380 /*00A34*/ EMIT_STATE(PA_CONFIG, val & ctx->shader_state.PA_CONFIG);
381 }
382 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
383 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
384 /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
385 /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
386 }
387 if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
388 ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
389 /* this is a bit of a mess: rasterizer.scissor determines whether to use
390 * only the framebuffer scissor, or specific scissor state, and the
391 * viewport clips too so the logic spans four CSOs */
392 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
393
394 uint32_t scissor_left =
395 MAX2(ctx->framebuffer.SE_SCISSOR_LEFT, ctx->viewport.SE_SCISSOR_LEFT);
396 uint32_t scissor_top =
397 MAX2(ctx->framebuffer.SE_SCISSOR_TOP, ctx->viewport.SE_SCISSOR_TOP);
398 uint32_t scissor_right =
399 MIN2(ctx->framebuffer.SE_SCISSOR_RIGHT, ctx->viewport.SE_SCISSOR_RIGHT);
400 uint32_t scissor_bottom =
401 MIN2(ctx->framebuffer.SE_SCISSOR_BOTTOM, ctx->viewport.SE_SCISSOR_BOTTOM);
402
403 if (rasterizer->scissor) {
404 scissor_left = MAX2(ctx->scissor.SE_SCISSOR_LEFT, scissor_left);
405 scissor_top = MAX2(ctx->scissor.SE_SCISSOR_TOP, scissor_top);
406 scissor_right = MIN2(ctx->scissor.SE_SCISSOR_RIGHT, scissor_right);
407 scissor_bottom = MIN2(ctx->scissor.SE_SCISSOR_BOTTOM, scissor_bottom);
408 }
409
410 /*00C00*/ EMIT_STATE_FIXP(SE_SCISSOR_LEFT, scissor_left);
411 /*00C04*/ EMIT_STATE_FIXP(SE_SCISSOR_TOP, scissor_top);
412 /*00C08*/ EMIT_STATE_FIXP(SE_SCISSOR_RIGHT, scissor_right);
413 /*00C0C*/ EMIT_STATE_FIXP(SE_SCISSOR_BOTTOM, scissor_bottom);
414 }
415 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
416 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
417
418 /*00C10*/ EMIT_STATE(SE_DEPTH_SCALE, rasterizer->SE_DEPTH_SCALE);
419 /*00C14*/ EMIT_STATE(SE_DEPTH_BIAS, rasterizer->SE_DEPTH_BIAS);
420 /*00C18*/ EMIT_STATE(SE_CONFIG, rasterizer->SE_CONFIG);
421 }
422 if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
423 ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
424 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
425
426 uint32_t clip_right =
427 MIN2(ctx->framebuffer.SE_CLIP_RIGHT, ctx->viewport.SE_CLIP_RIGHT);
428 uint32_t clip_bottom =
429 MIN2(ctx->framebuffer.SE_CLIP_BOTTOM, ctx->viewport.SE_CLIP_BOTTOM);
430
431 if (rasterizer->scissor) {
432 clip_right = MIN2(ctx->scissor.SE_CLIP_RIGHT, clip_right);
433 clip_bottom = MIN2(ctx->scissor.SE_CLIP_BOTTOM, clip_bottom);
434 }
435
436 /*00C20*/ EMIT_STATE_FIXP(SE_CLIP_RIGHT, clip_right);
437 /*00C24*/ EMIT_STATE_FIXP(SE_CLIP_BOTTOM, clip_bottom);
438 }
439 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
440 /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
441 }
442 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
443 /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
444 /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
445 ctx->framebuffer.msaa_mode
446 ? ctx->shader_state.PS_INPUT_COUNT_MSAA
447 : ctx->shader_state.PS_INPUT_COUNT);
448 /*0100C*/ EMIT_STATE(PS_TEMP_REGISTER_CONTROL,
449 ctx->framebuffer.msaa_mode
450 ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
451 : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
452 /*01010*/ EMIT_STATE(PS_CONTROL, ctx->shader_state.PS_CONTROL);
453 }
454 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER))) {
455 uint32_t val = etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG;
456 /*01400*/ EMIT_STATE(PE_DEPTH_CONFIG, val | ctx->framebuffer.PE_DEPTH_CONFIG);
457 }
458 if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
459 /*01404*/ EMIT_STATE(PE_DEPTH_NEAR, ctx->viewport.PE_DEPTH_NEAR);
460 /*01408*/ EMIT_STATE(PE_DEPTH_FAR, ctx->viewport.PE_DEPTH_FAR);
461 }
462 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
463 /*0140C*/ EMIT_STATE(PE_DEPTH_NORMALIZE, ctx->framebuffer.PE_DEPTH_NORMALIZE);
464
465 if (ctx->specs.pixel_pipes == 1) {
466 /*01410*/ EMIT_STATE_RELOC(PE_DEPTH_ADDR, &ctx->framebuffer.PE_DEPTH_ADDR);
467 }
468
469 /*01414*/ EMIT_STATE(PE_DEPTH_STRIDE, ctx->framebuffer.PE_DEPTH_STRIDE);
470 }
471
472 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_RASTERIZER))) {
473 uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_OP[ccw];
474 /*01418*/ EMIT_STATE(PE_STENCIL_OP, val);
475 }
476 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_STENCIL_REF | ETNA_DIRTY_RASTERIZER))) {
477 uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG[ccw];
478 /*0141C*/ EMIT_STATE(PE_STENCIL_CONFIG, val | ctx->stencil_ref.PE_STENCIL_CONFIG[ccw]);
479 }
480 if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
481 uint32_t val = etna_zsa_state(ctx->zsa)->PE_ALPHA_OP;
482 /*01420*/ EMIT_STATE(PE_ALPHA_OP, val);
483 }
484 if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
485 /*01424*/ EMIT_STATE(PE_ALPHA_BLEND_COLOR, ctx->blend_color.PE_ALPHA_BLEND_COLOR);
486 }
487 if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
488 uint32_t val = etna_blend_state(ctx->blend)->PE_ALPHA_CONFIG;
489 /*01428*/ EMIT_STATE(PE_ALPHA_CONFIG, val);
490 }
491 if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
492 uint32_t val;
493 /* Use the components and overwrite bits in framebuffer.PE_COLOR_FORMAT
494 * as a mask to enable the bits from blend PE_COLOR_FORMAT */
495 val = ~(VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
496 VIVS_PE_COLOR_FORMAT_OVERWRITE);
497 val |= etna_blend_state(ctx->blend)->PE_COLOR_FORMAT;
498 val &= ctx->framebuffer.PE_COLOR_FORMAT;
499 /*0142C*/ EMIT_STATE(PE_COLOR_FORMAT, val);
500 }
501 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
502 if (ctx->specs.pixel_pipes == 1) {
503 /*01430*/ EMIT_STATE_RELOC(PE_COLOR_ADDR, &ctx->framebuffer.PE_COLOR_ADDR);
504 /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
505 /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
506 } else if (ctx->specs.pixel_pipes == 2) {
507 /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
508 /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
509 /*01460*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(0), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[0]);
510 /*01464*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(1), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[1]);
511 /*01480*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(0), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[0]);
512 /*01484*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(1), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[1]);
513 } else {
514 abort();
515 }
516 }
517 if (unlikely(dirty & (ETNA_DIRTY_STENCIL_REF | ETNA_DIRTY_RASTERIZER))) {
518 /*014A0*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT, ctx->stencil_ref.PE_STENCIL_CONFIG_EXT[ccw]);
519 }
520 if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
521 struct etna_blend_state *blend = etna_blend_state(ctx->blend);
522 /*014A4*/ EMIT_STATE(PE_LOGIC_OP, blend->PE_LOGIC_OP | ctx->framebuffer.PE_LOGIC_OP);
523 }
524 if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
525 struct etna_blend_state *blend = etna_blend_state(ctx->blend);
526 for (int x = 0; x < 2; ++x) {
527 /*014A8*/ EMIT_STATE(PE_DITHER(x), blend->PE_DITHER[x]);
528 }
529 }
530 if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
531 /*014B0*/ EMIT_STATE(PE_ALPHA_COLOR_EXT0, ctx->blend_color.PE_ALPHA_COLOR_EXT0);
532 /*014B4*/ EMIT_STATE(PE_ALPHA_COLOR_EXT1, ctx->blend_color.PE_ALPHA_COLOR_EXT1);
533 }
534 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_RASTERIZER))) {
535 /*014B8*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT2, etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG_EXT2[ccw]);
536 }
537 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER)) && ctx->specs.halti >= 3)
538 /*014BC*/ EMIT_STATE(PE_MEM_CONFIG, ctx->framebuffer.PE_MEM_CONFIG);
539 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_TS))) {
540 /*01654*/ EMIT_STATE(TS_MEM_CONFIG, ctx->framebuffer.TS_MEM_CONFIG);
541 /*01658*/ EMIT_STATE_RELOC(TS_COLOR_STATUS_BASE, &ctx->framebuffer.TS_COLOR_STATUS_BASE);
542 /*0165C*/ EMIT_STATE_RELOC(TS_COLOR_SURFACE_BASE, &ctx->framebuffer.TS_COLOR_SURFACE_BASE);
543 /*01660*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE, ctx->framebuffer.TS_COLOR_CLEAR_VALUE);
544 /*01664*/ EMIT_STATE_RELOC(TS_DEPTH_STATUS_BASE, &ctx->framebuffer.TS_DEPTH_STATUS_BASE);
545 /*01668*/ EMIT_STATE_RELOC(TS_DEPTH_SURFACE_BASE, &ctx->framebuffer.TS_DEPTH_SURFACE_BASE);
546 /*0166C*/ EMIT_STATE(TS_DEPTH_CLEAR_VALUE, ctx->framebuffer.TS_DEPTH_CLEAR_VALUE);
547 }
548 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
549 /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
550 }
551 etna_coalesce_end(stream, &coalesce);
552 /* end only EMIT_STATE */
553
554 /* Emit strongly architecture-specific state */
555 if (ctx->specs.halti >= 5)
556 emit_halti5_only_state(ctx, vs_output_count);
557 else
558 emit_pre_halti5_state(ctx);
559
560 ctx->emit_texture_state(ctx);
561
562 /* Insert a FE/PE stall as changing the shader instructions (and maybe
563 * the uniforms) can corrupt the previous in-progress draw operation.
564 * Observed with amoeba on GC2000 during the right-to-left rendering
565 * of PI, and can cause GPU hangs immediately after.
566 * I summise that this is because the "new" locations at 0xc000 are not
567 * properly protected against updates as other states seem to be. Hence,
568 * we detect the "new" vertex shader instruction offset to apply this. */
569 if (ctx->dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF) && ctx->specs.vs_offset > 0x4000)
570 etna_stall(ctx->stream, SYNC_RECIPIENT_FE, SYNC_RECIPIENT_PE);
571
572 /* We need to update the uniform cache only if one of the following bits are
573 * set in ctx->dirty:
574 * - ETNA_DIRTY_SHADER
575 * - ETNA_DIRTY_CONSTBUF
576 * - uniforms_dirty_bits
577 *
578 * In case of ETNA_DIRTY_SHADER we need load all uniforms from the cache. In
579 * all
580 * other cases we can load on the changed uniforms.
581 */
582 static const uint32_t uniform_dirty_bits =
583 ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
584
585 /**** Large dynamically-sized state ****/
586 bool do_uniform_flush = ctx->specs.halti < 5;
587 if (dirty & (ETNA_DIRTY_SHADER)) {
588 /* Special case: a new shader was loaded; simply re-load all uniforms and
589 * shader code at once */
590 /* This sequence is special, do not change ordering unless necessary. According to comment
591 snippets in the Vivante kernel driver a process called "steering" goes on while programming
592 shader state. This (as I understand it) means certain unified states are "steered"
593 toward a specific shader unit (VS/PS/...) based on either explicit flags in register
594 00860, or what other state is written before "auto-steering". So this means some
595 state can legitimately be programmed multiple times.
596 */
597
598 if (ctx->specs.halti >= 5) { /* ICACHE (HALTI5) */
599 assert(ctx->shader_state.VS_INST_ADDR.bo && ctx->shader_state.PS_INST_ADDR.bo);
600 /* Set icache (VS) */
601 etna_set_state(stream, VIVS_VS_NEWRANGE_LOW, 0);
602 etna_set_state(stream, VIVS_VS_NEWRANGE_HIGH, ctx->shader_state.vs_inst_mem_size / 4);
603 assert(ctx->shader_state.VS_INST_ADDR.bo);
604 etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
605 etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
606 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
607 etna_set_state(stream, VIVS_VS_ICACHE_COUNT, ctx->shader_state.vs_inst_mem_size / 4 - 1);
608
609 /* Set icache (PS) */
610 etna_set_state(stream, VIVS_PS_NEWRANGE_LOW, 0);
611 etna_set_state(stream, VIVS_PS_NEWRANGE_HIGH, ctx->shader_state.ps_inst_mem_size / 4);
612 assert(ctx->shader_state.PS_INST_ADDR.bo);
613 etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
614 etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
615 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
616 etna_set_state(stream, VIVS_PS_ICACHE_COUNT, ctx->shader_state.ps_inst_mem_size / 4 - 1);
617
618 } else if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
619 /* ICACHE (pre-HALTI5) */
620 assert(ctx->specs.has_icache && ctx->specs.has_shader_range_registers);
621 /* Set icache (VS) */
622 etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
623 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
624 VIVS_VS_ICACHE_CONTROL_ENABLE |
625 VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
626 assert(ctx->shader_state.VS_INST_ADDR.bo);
627 etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
628
629 /* Set icache (PS) */
630 etna_set_state(stream, VIVS_PS_RANGE, (ctx->shader_state.ps_inst_mem_size / 4 - 1) << 16);
631 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
632 VIVS_VS_ICACHE_CONTROL_ENABLE |
633 VIVS_VS_ICACHE_CONTROL_FLUSH_PS);
634 assert(ctx->shader_state.PS_INST_ADDR.bo);
635 etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
636 } else {
637 /* Upload shader directly, first flushing and disabling icache if
638 * supported on this hw */
639 if (ctx->specs.has_icache) {
640 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
641 VIVS_VS_ICACHE_CONTROL_FLUSH_PS |
642 VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
643 }
644 if (ctx->specs.has_shader_range_registers) {
645 etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
646 etna_set_state(stream, VIVS_PS_RANGE, ((ctx->shader_state.ps_inst_mem_size / 4 - 1 + 0x100) << 16) |
647 0x100);
648 }
649 etna_set_state_multi(stream, ctx->specs.vs_offset,
650 ctx->shader_state.vs_inst_mem_size,
651 ctx->shader_state.VS_INST_MEM);
652 etna_set_state_multi(stream, ctx->specs.ps_offset,
653 ctx->shader_state.ps_inst_mem_size,
654 ctx->shader_state.PS_INST_MEM);
655 }
656
657 if (ctx->specs.has_unified_uniforms) {
658 etna_set_state(stream, VIVS_VS_UNIFORM_BASE, 0);
659 etna_set_state(stream, VIVS_PS_UNIFORM_BASE, ctx->specs.max_vs_uniforms);
660 }
661
662 if (do_uniform_flush)
663 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
664
665 etna_uniforms_write(ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX]);
666
667 if (do_uniform_flush)
668 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
669
670 etna_uniforms_write(ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT]);
671
672 if (ctx->specs.halti >= 5) {
673 /* HALTI5 needs to be prompted to pre-fetch shaders */
674 etna_set_state(stream, VIVS_VS_ICACHE_PREFETCH, 0x00000000);
675 etna_set_state(stream, VIVS_PS_ICACHE_PREFETCH, 0x00000000);
676 etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
677 }
678 } else {
679 /* ideally this cache would only be flushed if there are VS uniform changes */
680 if (do_uniform_flush)
681 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
682
683 if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
684 etna_uniforms_write(ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX]);
685
686 /* ideally this cache would only be flushed if there are PS uniform changes */
687 if (do_uniform_flush)
688 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
689
690 if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
691 etna_uniforms_write(ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT]);
692 }
693 /**** End of state update ****/
694 #undef EMIT_STATE
695 #undef EMIT_STATE_FIXP
696 #undef EMIT_STATE_RELOC
697 ctx->dirty = 0;
698 ctx->dirty_sampler_views = 0;
699 }