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
CMD_STALL(struct etna_cmd_stream * stream,uint32_t from,uint32_t to)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
etna_stall(struct etna_cmd_stream * stream,uint32_t from,uint32_t to)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
required_stream_size(struct etna_context * ctx)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.count * 2;
109 size += ctx->shader.fs->uniforms.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
emit_halti5_only_state(struct etna_context * ctx,int vs_output_count)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 /*007C4*/ EMIT_STATE(FE_HALTI5_ID_CONFIG, ctx->shader_state.FE_HALTI5_ID_CONFIG);
136 /*00870*/ EMIT_STATE(VS_HALTI5_OUTPUT_COUNT, vs_output_count | ((vs_output_count * 0x10) << 8));
137 /*008A0*/ EMIT_STATE(VS_HALTI5_UNK008A0, 0x0001000e | ((0x110/vs_output_count) << 20));
138 for (int x = 0; x < 4; ++x) {
139 /*008E0*/ EMIT_STATE(VS_HALTI5_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
140 }
141 }
142 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
143 for (int x = 0; x < 4; ++x) {
144 /*008C0*/ EMIT_STATE(VS_HALTI5_INPUT(x), ctx->shader_state.VS_INPUT[x]);
145 }
146 }
147 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
148 /*00A90*/ EMIT_STATE(PA_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS[0]);
149 /*00A94*/ EMIT_STATE(PA_VARYING_NUM_COMPONENTS(1), ctx->shader_state.GL_VARYING_NUM_COMPONENTS[1]);
150 /*00AA8*/ EMIT_STATE(PA_VS_OUTPUT_COUNT, vs_output_count);
151 /*01080*/ EMIT_STATE(PS_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS[0]);
152 /*01084*/ EMIT_STATE(PS_VARYING_NUM_COMPONENTS(1), ctx->shader_state.GL_VARYING_NUM_COMPONENTS[1]);
153 /*03888*/ EMIT_STATE(GL_HALTI5_SH_SPECIALS, ctx->shader_state.GL_HALTI5_SH_SPECIALS);
154 }
155 etna_coalesce_end(stream, &coalesce);
156 }
157
158 /* Emit state that no longer exists on HALTI5 */
159 static void
emit_pre_halti5_state(struct etna_context * ctx)160 emit_pre_halti5_state(struct etna_context *ctx)
161 {
162 struct etna_cmd_stream *stream = ctx->stream;
163 uint32_t dirty = ctx->dirty;
164 struct etna_coalesce coalesce;
165
166 etna_coalesce_start(stream, &coalesce);
167 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
168 /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
169 }
170 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
171 for (int x = 0; x < 4; ++x) {
172 /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
173 }
174 }
175 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
176 for (int x = 0; x < 4; ++x) {
177 /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
178 }
179 }
180 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
181 /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
182 }
183 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
184 for (int x = 0; x < 10; ++x) {
185 /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
186 }
187 }
188 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
189 /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
190 for (int x = 0; x < 4; ++x) {
191 /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
192 }
193 for (int x = 0; x < 16; ++x) {
194 /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
195 }
196 }
197 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
198 /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
199 }
200 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
201 /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
202 }
203 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
204 /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS[0]);
205 for (int x = 0; x < 2; ++x) {
206 /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
207 }
208 /*03834*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS2, ctx->shader_state.GL_VARYING_NUM_COMPONENTS[1]);
209 }
210 etna_coalesce_end(stream, &coalesce);
211 }
212
213 /* Weave state before draw operation. This function merges all the compiled
214 * state blocks under the context into one device register state. Parts of
215 * this state that are changed since last call (dirty) will be uploaded as
216 * state changes in the command buffer. */
217 void
etna_emit_state(struct etna_context * ctx)218 etna_emit_state(struct etna_context *ctx)
219 {
220 struct etna_cmd_stream *stream = ctx->stream;
221 struct etna_screen *screen = ctx->screen;
222 unsigned ccw = ctx->rasterizer->front_ccw;
223
224
225 /* Pre-reserve the command buffer space which we are likely to need.
226 * This must cover all the state emitted below, and the following
227 * draw command. */
228 etna_cmd_stream_reserve(stream, required_stream_size(ctx));
229
230 uint32_t dirty = ctx->dirty;
231
232 /* Pre-processing: see what caches we need to flush before making state changes. */
233 uint32_t to_flush = 0;
234 if (unlikely(dirty & (ETNA_DIRTY_BLEND)))
235 to_flush |= VIVS_GL_FLUSH_CACHE_COLOR;
236 if (unlikely(dirty & ETNA_DIRTY_ZSA))
237 to_flush |= VIVS_GL_FLUSH_CACHE_DEPTH;
238 if (unlikely(dirty & (ETNA_DIRTY_TEXTURE_CACHES)))
239 to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE;
240 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) /* Framebuffer config changed? */
241 to_flush |= VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
242 if (DBG_ENABLED(ETNA_DBG_CFLUSH_ALL))
243 to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE | VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
244
245 if (to_flush) {
246 etna_set_state(stream, VIVS_GL_FLUSH_CACHE, to_flush);
247 etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
248 }
249
250 /* Flush TS cache before changing TS configuration. */
251 if (unlikely(dirty & ETNA_DIRTY_TS)) {
252 etna_set_state(stream, VIVS_TS_FLUSH_CACHE, VIVS_TS_FLUSH_CACHE_FLUSH);
253 }
254
255 /* Update vertex elements. This is different from any of the other states, in that
256 * a) the number of vertex elements written matters: so write only active ones
257 * b) the vertex element states must all be written: do not skip entries that stay the same */
258 if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
259 if (screen->specs.halti >= 5) {
260 /*17800*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG0(0),
261 ctx->vertex_elements->num_elements,
262 ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG0);
263 /*17A00*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_SCALE(0),
264 ctx->vertex_elements->num_elements,
265 ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
266 /*17A80*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG1(0),
267 ctx->vertex_elements->num_elements,
268 ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG1);
269 } else {
270 /* Special case: vertex elements must always be sent in full if changed */
271 /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
272 ctx->vertex_elements->num_elements,
273 ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
274 if (screen->specs.halti >= 2) {
275 /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
276 ctx->vertex_elements->num_elements,
277 ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
278 }
279 }
280 }
281 unsigned vs_output_count = etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex
282 ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
283 : ctx->shader_state.VS_OUTPUT_COUNT;
284
285 /* The following code is originally generated by gen_merge_state.py, to
286 * emit state in increasing order of address (this makes it possible to merge
287 * consecutive register updates into one SET_STATE command)
288 *
289 * There have been some manual changes, where the weaving operation is not
290 * simply bitwise or:
291 * - scissor fixp
292 * - num vertex elements
293 * - scissor handling
294 * - num samplers
295 * - texture lod
296 * - ETNA_DIRTY_TS
297 * - removed ETNA_DIRTY_BASE_SETUP statements -- these are guaranteed to not
298 * change anyway
299 * - PS / framebuffer interaction for MSAA
300 * - move update of GL_MULTI_SAMPLE_CONFIG first
301 * - add unlikely()/likely()
302 */
303 struct etna_coalesce coalesce;
304
305 etna_coalesce_start(stream, &coalesce);
306
307 /* begin only EMIT_STATE -- make sure no new etna_reserve calls are done here
308 * directly
309 * or indirectly */
310 /* multi sample config is set first, and outside of the normal sorting
311 * order, as changing the multisample state clobbers PS.INPUT_COUNT (and
312 * possibly PS.TEMP_REGISTER_CONTROL).
313 */
314 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SAMPLE_MASK))) {
315 uint32_t val = VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_ENABLES(ctx->sample_mask);
316 val |= ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG;
317
318 /*03818*/ EMIT_STATE(GL_MULTI_SAMPLE_CONFIG, val);
319 }
320 if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
321 /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
322 /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
323 }
324 if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
325 /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
326 }
327 if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
328 if (screen->specs.halti >= 2) { /* HALTI2+: NFE_VERTEX_STREAMS */
329 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
330 /*14600*/ EMIT_STATE_RELOC(NFE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
331 }
332 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
333 if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
334 /*14640*/ EMIT_STATE(NFE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
335 }
336 }
337 } else if(screen->specs.stream_count > 1) { /* hw w/ multiple vertex streams */
338 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
339 /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
340 }
341 for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
342 if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
343 /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
344 }
345 }
346 } else { /* hw w/ single vertex stream */
347 /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
348 /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
349 }
350 }
351 /* gallium has instance divisor as part of elements state */
352 if ((dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) && screen->specs.halti >= 2) {
353 for (int x = 0; x < ctx->vertex_elements->num_buffers; ++x) {
354 /*14680*/ EMIT_STATE(NFE_VERTEX_STREAMS_VERTEX_DIVISOR(x), ctx->vertex_elements->NFE_VERTEX_STREAMS_VERTEX_DIVISOR[x]);
355 }
356 }
357
358 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
359
360 /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT, vs_output_count);
361 }
362 if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
363 /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
364 /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
365 }
366 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
367 /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
368 }
369 if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
370 /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
371 /*00A04*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_Y, ctx->viewport.PA_VIEWPORT_SCALE_Y);
372 /*00A08*/ EMIT_STATE(PA_VIEWPORT_SCALE_Z, ctx->viewport.PA_VIEWPORT_SCALE_Z);
373 /*00A0C*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_X, ctx->viewport.PA_VIEWPORT_OFFSET_X);
374 /*00A10*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_Y, ctx->viewport.PA_VIEWPORT_OFFSET_Y);
375 /*00A14*/ EMIT_STATE(PA_VIEWPORT_OFFSET_Z, ctx->viewport.PA_VIEWPORT_OFFSET_Z);
376 }
377 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
378 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
379
380 /*00A18*/ EMIT_STATE(PA_LINE_WIDTH, rasterizer->PA_LINE_WIDTH);
381 /*00A1C*/ EMIT_STATE(PA_POINT_SIZE, rasterizer->PA_POINT_SIZE);
382 /*00A28*/ EMIT_STATE(PA_SYSTEM_MODE, rasterizer->PA_SYSTEM_MODE);
383 }
384 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
385 /*00A30*/ EMIT_STATE(PA_ATTRIBUTE_ELEMENT_COUNT, ctx->shader_state.PA_ATTRIBUTE_ELEMENT_COUNT);
386 }
387 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_SHADER))) {
388 uint32_t val = etna_rasterizer_state(ctx->rasterizer)->PA_CONFIG;
389 /*00A34*/ EMIT_STATE(PA_CONFIG, val & ctx->shader_state.PA_CONFIG);
390 }
391 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
392 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
393 /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
394 /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
395 }
396 if (unlikely(dirty & (ETNA_DIRTY_SCISSOR_CLIP))) {
397 /*00C00*/ EMIT_STATE_FIXP(SE_SCISSOR_LEFT, ctx->clipping.minx << 16);
398 /*00C04*/ EMIT_STATE_FIXP(SE_SCISSOR_TOP, ctx->clipping.miny << 16);
399 /*00C08*/ EMIT_STATE_FIXP(SE_SCISSOR_RIGHT, (ctx->clipping.maxx << 16) + ETNA_SE_SCISSOR_MARGIN_RIGHT);
400 /*00C0C*/ EMIT_STATE_FIXP(SE_SCISSOR_BOTTOM, (ctx->clipping.maxy << 16) + ETNA_SE_SCISSOR_MARGIN_BOTTOM);
401 }
402 if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
403 struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
404
405 /*00C10*/ EMIT_STATE(SE_DEPTH_SCALE, rasterizer->SE_DEPTH_SCALE);
406 /*00C14*/ EMIT_STATE(SE_DEPTH_BIAS, rasterizer->SE_DEPTH_BIAS);
407 /*00C18*/ EMIT_STATE(SE_CONFIG, rasterizer->SE_CONFIG);
408 }
409 if (unlikely(dirty & (ETNA_DIRTY_SCISSOR_CLIP))) {
410 /*00C20*/ EMIT_STATE_FIXP(SE_CLIP_RIGHT, (ctx->clipping.maxx << 16) + ETNA_SE_CLIP_MARGIN_RIGHT);
411 /*00C24*/ EMIT_STATE_FIXP(SE_CLIP_BOTTOM, (ctx->clipping.maxy << 16) + ETNA_SE_CLIP_MARGIN_BOTTOM);
412 }
413 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
414 /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
415 }
416 if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
417 /*00E08*/ EMIT_STATE(RA_EARLY_DEPTH, etna_zsa_state(ctx->zsa)->RA_DEPTH_CONFIG);
418 }
419 if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
420 /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
421 /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
422 ctx->framebuffer.msaa_mode
423 ? ctx->shader_state.PS_INPUT_COUNT_MSAA
424 : ctx->shader_state.PS_INPUT_COUNT);
425 /*0100C*/ EMIT_STATE(PS_TEMP_REGISTER_CONTROL,
426 ctx->framebuffer.msaa_mode
427 ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
428 : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
429 /*01010*/ EMIT_STATE(PS_CONTROL, ctx->framebuffer.PS_CONTROL);
430 /*01030*/ EMIT_STATE(PS_CONTROL_EXT, ctx->framebuffer.PS_CONTROL_EXT);
431 }
432 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SHADER))) {
433 /*01400*/ EMIT_STATE(PE_DEPTH_CONFIG, (etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG |
434 ctx->framebuffer.PE_DEPTH_CONFIG));
435 }
436 if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
437 /*01404*/ EMIT_STATE(PE_DEPTH_NEAR, ctx->viewport.PE_DEPTH_NEAR);
438 /*01408*/ EMIT_STATE(PE_DEPTH_FAR, ctx->viewport.PE_DEPTH_FAR);
439 }
440 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
441 /*0140C*/ EMIT_STATE(PE_DEPTH_NORMALIZE, ctx->framebuffer.PE_DEPTH_NORMALIZE);
442
443 if (screen->specs.pixel_pipes == 1) {
444 /*01410*/ EMIT_STATE_RELOC(PE_DEPTH_ADDR, &ctx->framebuffer.PE_DEPTH_ADDR);
445 }
446
447 /*01414*/ EMIT_STATE(PE_DEPTH_STRIDE, ctx->framebuffer.PE_DEPTH_STRIDE);
448 }
449
450 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_RASTERIZER))) {
451 uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_OP[ccw];
452 /*01418*/ EMIT_STATE(PE_STENCIL_OP, val);
453 }
454 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_STENCIL_REF | ETNA_DIRTY_RASTERIZER))) {
455 uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG[ccw];
456 /*0141C*/ EMIT_STATE(PE_STENCIL_CONFIG, val | ctx->stencil_ref.PE_STENCIL_CONFIG[ccw]);
457 }
458 if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
459 uint32_t val = etna_zsa_state(ctx->zsa)->PE_ALPHA_OP;
460 /*01420*/ EMIT_STATE(PE_ALPHA_OP, val);
461 }
462 if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
463 /*01424*/ EMIT_STATE(PE_ALPHA_BLEND_COLOR, ctx->blend_color.PE_ALPHA_BLEND_COLOR);
464 }
465 if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
466 uint32_t val = etna_blend_state(ctx->blend)->PE_ALPHA_CONFIG;
467 /*01428*/ EMIT_STATE(PE_ALPHA_CONFIG, val);
468 }
469 if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
470 uint32_t val;
471 /* Use the components and overwrite bits in framebuffer.PE_COLOR_FORMAT
472 * as a mask to enable the bits from blend PE_COLOR_FORMAT */
473 val = ~(VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
474 VIVS_PE_COLOR_FORMAT_OVERWRITE);
475 val |= etna_blend_state(ctx->blend)->PE_COLOR_FORMAT;
476 val &= ctx->framebuffer.PE_COLOR_FORMAT;
477 /*0142C*/ EMIT_STATE(PE_COLOR_FORMAT, val);
478 }
479 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
480 if (screen->specs.pixel_pipes == 1) {
481 /*01430*/ EMIT_STATE_RELOC(PE_COLOR_ADDR, &ctx->framebuffer.PE_COLOR_ADDR);
482 /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
483 /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
484 } else if (screen->specs.pixel_pipes == 2) {
485 /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
486 /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
487 /*01460*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(0), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[0]);
488 /*01464*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(1), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[1]);
489 /*01480*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(0), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[0]);
490 /*01484*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(1), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[1]);
491 } else {
492 abort();
493 }
494 }
495 if (unlikely(dirty & (ETNA_DIRTY_STENCIL_REF | ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_ZSA))) {
496 uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG_EXT;
497 /*014A0*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT, val | ctx->stencil_ref.PE_STENCIL_CONFIG_EXT[ccw]);
498 }
499 if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
500 struct etna_blend_state *blend = etna_blend_state(ctx->blend);
501 /*014A4*/ EMIT_STATE(PE_LOGIC_OP, blend->PE_LOGIC_OP | ctx->framebuffer.PE_LOGIC_OP);
502 }
503 if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
504 struct etna_blend_state *blend = etna_blend_state(ctx->blend);
505 for (int x = 0; x < 2; ++x) {
506 /*014A8*/ EMIT_STATE(PE_DITHER(x), blend->PE_DITHER[x]);
507 }
508 }
509 if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR)) &&
510 VIV_FEATURE(screen, chipMinorFeatures1, HALF_FLOAT)) {
511 /*014B0*/ EMIT_STATE(PE_ALPHA_COLOR_EXT0, ctx->blend_color.PE_ALPHA_COLOR_EXT0);
512 /*014B4*/ EMIT_STATE(PE_ALPHA_COLOR_EXT1, ctx->blend_color.PE_ALPHA_COLOR_EXT1);
513 }
514 if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_RASTERIZER))) {
515 /*014B8*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT2, etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG_EXT2[ccw]);
516 }
517 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER)) && screen->specs.halti >= 3)
518 /*014BC*/ EMIT_STATE(PE_MEM_CONFIG, ctx->framebuffer.PE_MEM_CONFIG);
519 if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_TS))) {
520 /*01654*/ EMIT_STATE(TS_MEM_CONFIG, ctx->framebuffer.TS_MEM_CONFIG);
521 /*01658*/ EMIT_STATE_RELOC(TS_COLOR_STATUS_BASE, &ctx->framebuffer.TS_COLOR_STATUS_BASE);
522 /*0165C*/ EMIT_STATE_RELOC(TS_COLOR_SURFACE_BASE, &ctx->framebuffer.TS_COLOR_SURFACE_BASE);
523 /*01660*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE, ctx->framebuffer.TS_COLOR_CLEAR_VALUE);
524 /*01664*/ EMIT_STATE_RELOC(TS_DEPTH_STATUS_BASE, &ctx->framebuffer.TS_DEPTH_STATUS_BASE);
525 /*01668*/ EMIT_STATE_RELOC(TS_DEPTH_SURFACE_BASE, &ctx->framebuffer.TS_DEPTH_SURFACE_BASE);
526 /*0166C*/ EMIT_STATE(TS_DEPTH_CLEAR_VALUE, ctx->framebuffer.TS_DEPTH_CLEAR_VALUE);
527 /*016BC*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE_EXT, ctx->framebuffer.TS_COLOR_CLEAR_VALUE_EXT);
528 }
529 if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
530 /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
531 }
532 etna_coalesce_end(stream, &coalesce);
533 /* end only EMIT_STATE */
534
535 /* Emit strongly architecture-specific state */
536 if (screen->specs.halti >= 5)
537 emit_halti5_only_state(ctx, vs_output_count);
538 else
539 emit_pre_halti5_state(ctx);
540
541 /* Beginning from Halti0 some of the new shader and sampler states are not
542 * self-synchronizing anymore. Thus we need to stall the FE on PE completion
543 * before loading the new states to avoid corrupting the state of the
544 * in-flight draw.
545 */
546 if (screen->specs.halti >= 0 &&
547 (ctx->dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF |
548 ETNA_DIRTY_SAMPLERS | ETNA_DIRTY_SAMPLER_VIEWS)))
549 etna_stall(ctx->stream, SYNC_RECIPIENT_FE, SYNC_RECIPIENT_PE);
550
551 ctx->emit_texture_state(ctx);
552
553 /* We need to update the uniform cache only if one of the following bits are
554 * set in ctx->dirty:
555 * - ETNA_DIRTY_SHADER
556 * - ETNA_DIRTY_CONSTBUF
557 * - uniforms_dirty_bits
558 *
559 * In case of ETNA_DIRTY_SHADER we need load all uniforms from the cache. In
560 * all
561 * other cases we can load on the changed uniforms.
562 */
563 static const uint32_t uniform_dirty_bits =
564 ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
565
566 /**** Large dynamically-sized state ****/
567 bool do_uniform_flush = screen->specs.halti < 5;
568 if (dirty & (ETNA_DIRTY_SHADER)) {
569 /* Special case: a new shader was loaded; simply re-load all uniforms and
570 * shader code at once */
571 /* This sequence is special, do not change ordering unless necessary. According to comment
572 snippets in the Vivante kernel driver a process called "steering" goes on while programming
573 shader state. This (as I understand it) means certain unified states are "steered"
574 toward a specific shader unit (VS/PS/...) based on either explicit flags in register
575 00860, or what other state is written before "auto-steering". So this means some
576 state can legitimately be programmed multiple times.
577 */
578
579 if (screen->specs.halti >= 5) { /* ICACHE (HALTI5) */
580 assert(ctx->shader_state.VS_INST_ADDR.bo && ctx->shader_state.PS_INST_ADDR.bo);
581 /* Set icache (VS) */
582 etna_set_state(stream, VIVS_VS_NEWRANGE_LOW, 0);
583 etna_set_state(stream, VIVS_VS_NEWRANGE_HIGH, ctx->shader_state.vs_inst_mem_size / 4);
584 assert(ctx->shader_state.VS_INST_ADDR.bo);
585 etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
586 etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
587 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
588 etna_set_state(stream, VIVS_VS_ICACHE_COUNT, ctx->shader_state.vs_inst_mem_size / 4 - 1);
589
590 /* Set icache (PS) */
591 etna_set_state(stream, VIVS_PS_NEWRANGE_LOW, 0);
592 etna_set_state(stream, VIVS_PS_NEWRANGE_HIGH, ctx->shader_state.ps_inst_mem_size / 4);
593 assert(ctx->shader_state.PS_INST_ADDR.bo);
594 etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
595 etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
596 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
597 etna_set_state(stream, VIVS_PS_ICACHE_COUNT, ctx->shader_state.ps_inst_mem_size / 4 - 1);
598
599 } else if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
600 /* ICACHE (pre-HALTI5) */
601 assert(screen->specs.has_icache && screen->specs.has_shader_range_registers);
602 /* Set icache (VS) */
603 etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
604 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
605 VIVS_VS_ICACHE_CONTROL_ENABLE |
606 VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
607 assert(ctx->shader_state.VS_INST_ADDR.bo);
608 etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
609
610 /* Set icache (PS) */
611 etna_set_state(stream, VIVS_PS_RANGE, (ctx->shader_state.ps_inst_mem_size / 4 - 1) << 16);
612 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
613 VIVS_VS_ICACHE_CONTROL_ENABLE |
614 VIVS_VS_ICACHE_CONTROL_FLUSH_PS);
615 assert(ctx->shader_state.PS_INST_ADDR.bo);
616 etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
617 } else {
618 /* Upload shader directly, first flushing and disabling icache if
619 * supported on this hw */
620 if (screen->specs.has_icache) {
621 etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
622 VIVS_VS_ICACHE_CONTROL_FLUSH_PS |
623 VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
624 }
625 if (screen->specs.has_shader_range_registers) {
626 etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
627 etna_set_state(stream, VIVS_PS_RANGE, ((ctx->shader_state.ps_inst_mem_size / 4 - 1 + 0x100) << 16) |
628 0x100);
629 }
630 etna_set_state_multi(stream, screen->specs.vs_offset,
631 ctx->shader_state.vs_inst_mem_size,
632 ctx->shader_state.VS_INST_MEM);
633 etna_set_state_multi(stream, screen->specs.ps_offset,
634 ctx->shader_state.ps_inst_mem_size,
635 ctx->shader_state.PS_INST_MEM);
636 }
637
638 if (screen->specs.has_unified_uniforms) {
639 etna_set_state(stream, VIVS_VS_UNIFORM_BASE, 0);
640 etna_set_state(stream, VIVS_PS_UNIFORM_BASE, screen->specs.max_vs_uniforms);
641 }
642
643 if (do_uniform_flush)
644 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
645
646 etna_uniforms_write(ctx, ctx->shader.vs, ctx->constant_buffer[PIPE_SHADER_VERTEX].cb);
647
648 if (do_uniform_flush)
649 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
650
651 etna_uniforms_write(ctx, ctx->shader.fs, ctx->constant_buffer[PIPE_SHADER_FRAGMENT].cb);
652
653 if (screen->specs.halti >= 5) {
654 /* HALTI5 needs to be prompted to pre-fetch shaders */
655 etna_set_state(stream, VIVS_VS_ICACHE_PREFETCH, 0x00000000);
656 etna_set_state(stream, VIVS_PS_ICACHE_PREFETCH, 0x00000000);
657 etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
658 }
659 } else {
660 /* ideally this cache would only be flushed if there are VS uniform changes */
661 if (do_uniform_flush)
662 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
663
664 if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
665 etna_uniforms_write(ctx, ctx->shader.vs, ctx->constant_buffer[PIPE_SHADER_VERTEX].cb);
666
667 /* ideally this cache would only be flushed if there are PS uniform changes */
668 if (do_uniform_flush)
669 etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
670
671 if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
672 etna_uniforms_write(ctx, ctx->shader.fs, ctx->constant_buffer[PIPE_SHADER_FRAGMENT].cb);
673 }
674 /**** End of state update ****/
675 #undef EMIT_STATE
676 #undef EMIT_STATE_FIXP
677 #undef EMIT_STATE_RELOC
678 ctx->dirty = 0;
679 ctx->dirty_sampler_views = 0;
680 }
681