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) */
48static inline void
49CMD_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
55void
56etna_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
96static unsigned
97required_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.const_count * 2;
109   size += ctx->shader.fs->uniforms.const_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+ */
125static void
126emit_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 */
156static void
157emit_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. */
213void
214etna_emit_state(struct etna_context *ctx)
215{
216   struct etna_cmd_stream *stream = ctx->stream;
217
218   /* Pre-reserve the command buffer space which we are likely to need.
219    * This must cover all the state emitted below, and the following
220    * draw command. */
221   etna_cmd_stream_reserve(stream, required_stream_size(ctx));
222
223   uint32_t dirty = ctx->dirty;
224
225   /* Pre-processing: see what caches we need to flush before making state changes. */
226   uint32_t to_flush = 0;
227   if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
228      /* Need flush COLOR when changing PE.COLOR_FORMAT.OVERWRITE. */
229#if 0
230        /* TODO*/
231        if ((ctx->gpu3d.PE_COLOR_FORMAT & VIVS_PE_COLOR_FORMAT_OVERWRITE) !=
232           (etna_blend_state(ctx->blend)->PE_COLOR_FORMAT & VIVS_PE_COLOR_FORMAT_OVERWRITE))
233#endif
234      to_flush |= VIVS_GL_FLUSH_CACHE_COLOR;
235   }
236   if (unlikely(dirty & (ETNA_DIRTY_TEXTURE_CACHES)))
237      to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE;
238   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) /* Framebuffer config changed? */
239      to_flush |= VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
240   if (DBG_ENABLED(ETNA_DBG_CFLUSH_ALL))
241      to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE | VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
242
243   if (to_flush) {
244      etna_set_state(stream, VIVS_GL_FLUSH_CACHE, to_flush);
245      etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
246   }
247
248   /* Flush TS cache before changing TS configuration. */
249   if (unlikely(dirty & ETNA_DIRTY_TS)) {
250      etna_set_state(stream, VIVS_TS_FLUSH_CACHE, VIVS_TS_FLUSH_CACHE_FLUSH);
251   }
252
253   /* Update vertex elements. This is different from any of the other states, in that
254    * a) the number of vertex elements written matters: so write only active ones
255    * b) the vertex element states must all be written: do not skip entries that stay the same */
256   if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
257      if (ctx->specs.halti >= 5) {
258         /*17800*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG0(0),
259            ctx->vertex_elements->num_elements,
260            ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG0);
261         /*17A00*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_SCALE(0),
262            ctx->vertex_elements->num_elements,
263            ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
264         /*17A80*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG1(0),
265            ctx->vertex_elements->num_elements,
266            ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG1);
267      } else {
268         /* Special case: vertex elements must always be sent in full if changed */
269         /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
270            ctx->vertex_elements->num_elements,
271            ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
272         if (ctx->specs.halti >= 2) {
273            /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
274               ctx->vertex_elements->num_elements,
275               ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
276         }
277      }
278   }
279   unsigned vs_output_count = etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex
280                           ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
281                           : ctx->shader_state.VS_OUTPUT_COUNT;
282
283   /* The following code is originally generated by gen_merge_state.py, to
284    * emit state in increasing order of address (this makes it possible to merge
285    * consecutive register updates into one SET_STATE command)
286    *
287    * There have been some manual changes, where the weaving operation is not
288    * simply bitwise or:
289    * - scissor fixp
290    * - num vertex elements
291    * - scissor handling
292    * - num samplers
293    * - texture lod
294    * - ETNA_DIRTY_TS
295    * - removed ETNA_DIRTY_BASE_SETUP statements -- these are guaranteed to not
296    * change anyway
297    * - PS / framebuffer interaction for MSAA
298    * - move update of GL_MULTI_SAMPLE_CONFIG first
299    * - add unlikely()/likely()
300    */
301   struct etna_coalesce coalesce;
302
303   etna_coalesce_start(stream, &coalesce);
304
305   /* begin only EMIT_STATE -- make sure no new etna_reserve calls are done here
306    * directly
307    *    or indirectly */
308   /* multi sample config is set first, and outside of the normal sorting
309    * order, as changing the multisample state clobbers PS.INPUT_COUNT (and
310    * possibly PS.TEMP_REGISTER_CONTROL).
311    */
312   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SAMPLE_MASK))) {
313      uint32_t val = VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_ENABLES(ctx->sample_mask);
314      val |= ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG;
315
316      /*03818*/ EMIT_STATE(GL_MULTI_SAMPLE_CONFIG, val);
317   }
318   if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
319      /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
320      /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
321   }
322   if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
323      /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
324   }
325   if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
326      if (ctx->specs.halti >= 2) { /* HALTI2+: NFE_VERTEX_STREAMS */
327         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
328            /*14600*/ EMIT_STATE_RELOC(NFE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
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               /*14640*/ EMIT_STATE(NFE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
333            }
334         }
335         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
336            if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
337               /*14680*/ EMIT_STATE(NFE_VERTEX_STREAMS_UNK14680(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_UNK14680);
338            }
339         }
340      } else if(ctx->specs.stream_count >= 1) { /* hw w/ multiple vertex streams */
341         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
342            /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
343         }
344         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
345            if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
346               /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
347            }
348         }
349      } else { /* hw w/ single vertex stream */
350         /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
351         /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
352      }
353   }
354   if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
355
356      /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT, vs_output_count);
357   }
358   if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
359      /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
360      /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
361   }
362   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
363      /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
364   }
365   if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
366      /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
367      /*00A04*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_Y, ctx->viewport.PA_VIEWPORT_SCALE_Y);
368      /*00A08*/ EMIT_STATE(PA_VIEWPORT_SCALE_Z, ctx->viewport.PA_VIEWPORT_SCALE_Z);
369      /*00A0C*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_X, ctx->viewport.PA_VIEWPORT_OFFSET_X);
370      /*00A10*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_Y, ctx->viewport.PA_VIEWPORT_OFFSET_Y);
371      /*00A14*/ EMIT_STATE(PA_VIEWPORT_OFFSET_Z, ctx->viewport.PA_VIEWPORT_OFFSET_Z);
372   }
373   if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
374      struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
375
376      /*00A18*/ EMIT_STATE(PA_LINE_WIDTH, rasterizer->PA_LINE_WIDTH);
377      /*00A1C*/ EMIT_STATE(PA_POINT_SIZE, rasterizer->PA_POINT_SIZE);
378      /*00A28*/ EMIT_STATE(PA_SYSTEM_MODE, rasterizer->PA_SYSTEM_MODE);
379   }
380   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
381      /*00A30*/ EMIT_STATE(PA_ATTRIBUTE_ELEMENT_COUNT, ctx->shader_state.PA_ATTRIBUTE_ELEMENT_COUNT);
382   }
383   if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_SHADER))) {
384      uint32_t val = etna_rasterizer_state(ctx->rasterizer)->PA_CONFIG;
385      /*00A34*/ EMIT_STATE(PA_CONFIG, val & ctx->shader_state.PA_CONFIG);
386   }
387   if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
388      struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
389      /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
390      /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
391   }
392   if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
393                         ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
394      /* this is a bit of a mess: rasterizer.scissor determines whether to use
395       * only the framebuffer scissor, or specific scissor state, and the
396       * viewport clips too so the logic spans four CSOs */
397      struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
398
399      uint32_t scissor_left =
400         MAX2(ctx->framebuffer.SE_SCISSOR_LEFT, ctx->viewport.SE_SCISSOR_LEFT);
401      uint32_t scissor_top =
402         MAX2(ctx->framebuffer.SE_SCISSOR_TOP, ctx->viewport.SE_SCISSOR_TOP);
403      uint32_t scissor_right =
404         MIN2(ctx->framebuffer.SE_SCISSOR_RIGHT, ctx->viewport.SE_SCISSOR_RIGHT);
405      uint32_t scissor_bottom =
406         MIN2(ctx->framebuffer.SE_SCISSOR_BOTTOM, ctx->viewport.SE_SCISSOR_BOTTOM);
407
408      if (rasterizer->scissor) {
409         scissor_left = MAX2(ctx->scissor.SE_SCISSOR_LEFT, scissor_left);
410         scissor_top = MAX2(ctx->scissor.SE_SCISSOR_TOP, scissor_top);
411         scissor_right = MIN2(ctx->scissor.SE_SCISSOR_RIGHT, scissor_right);
412         scissor_bottom = MIN2(ctx->scissor.SE_SCISSOR_BOTTOM, scissor_bottom);
413      }
414
415      /*00C00*/ EMIT_STATE_FIXP(SE_SCISSOR_LEFT, scissor_left);
416      /*00C04*/ EMIT_STATE_FIXP(SE_SCISSOR_TOP, scissor_top);
417      /*00C08*/ EMIT_STATE_FIXP(SE_SCISSOR_RIGHT, scissor_right);
418      /*00C0C*/ EMIT_STATE_FIXP(SE_SCISSOR_BOTTOM, scissor_bottom);
419   }
420   if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
421      struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
422
423      /*00C10*/ EMIT_STATE(SE_DEPTH_SCALE, rasterizer->SE_DEPTH_SCALE);
424      /*00C14*/ EMIT_STATE(SE_DEPTH_BIAS, rasterizer->SE_DEPTH_BIAS);
425      /*00C18*/ EMIT_STATE(SE_CONFIG, rasterizer->SE_CONFIG);
426   }
427   if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
428                         ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
429      struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
430
431      uint32_t clip_right =
432         MIN2(ctx->framebuffer.SE_CLIP_RIGHT, ctx->viewport.SE_CLIP_RIGHT);
433      uint32_t clip_bottom =
434         MIN2(ctx->framebuffer.SE_CLIP_BOTTOM, ctx->viewport.SE_CLIP_BOTTOM);
435
436      if (rasterizer->scissor) {
437         clip_right = MIN2(ctx->scissor.SE_CLIP_RIGHT, clip_right);
438         clip_bottom = MIN2(ctx->scissor.SE_CLIP_BOTTOM, clip_bottom);
439      }
440
441      /*00C20*/ EMIT_STATE_FIXP(SE_CLIP_RIGHT, clip_right);
442      /*00C24*/ EMIT_STATE_FIXP(SE_CLIP_BOTTOM, clip_bottom);
443   }
444   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
445      /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
446   }
447   if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
448      /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
449      /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
450                           ctx->framebuffer.msaa_mode
451                              ? ctx->shader_state.PS_INPUT_COUNT_MSAA
452                              : ctx->shader_state.PS_INPUT_COUNT);
453      /*0100C*/ EMIT_STATE(PS_TEMP_REGISTER_CONTROL,
454                           ctx->framebuffer.msaa_mode
455                              ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
456                              : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
457      /*01010*/ EMIT_STATE(PS_CONTROL, ctx->shader_state.PS_CONTROL);
458   }
459   if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER))) {
460      uint32_t val = etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG;
461      /*01400*/ EMIT_STATE(PE_DEPTH_CONFIG, val | ctx->framebuffer.PE_DEPTH_CONFIG);
462   }
463   if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
464      /*01404*/ EMIT_STATE(PE_DEPTH_NEAR, ctx->viewport.PE_DEPTH_NEAR);
465      /*01408*/ EMIT_STATE(PE_DEPTH_FAR, ctx->viewport.PE_DEPTH_FAR);
466   }
467   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
468      /*0140C*/ EMIT_STATE(PE_DEPTH_NORMALIZE, ctx->framebuffer.PE_DEPTH_NORMALIZE);
469
470      if (ctx->specs.pixel_pipes == 1) {
471         /*01410*/ EMIT_STATE_RELOC(PE_DEPTH_ADDR, &ctx->framebuffer.PE_DEPTH_ADDR);
472      }
473
474      /*01414*/ EMIT_STATE(PE_DEPTH_STRIDE, ctx->framebuffer.PE_DEPTH_STRIDE);
475   }
476   if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
477      uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_OP;
478      /*01418*/ EMIT_STATE(PE_STENCIL_OP, val);
479   }
480   if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_STENCIL_REF))) {
481      uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG;
482      /*0141C*/ EMIT_STATE(PE_STENCIL_CONFIG, val | ctx->stencil_ref.PE_STENCIL_CONFIG);
483   }
484   if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
485      uint32_t val = etna_zsa_state(ctx->zsa)->PE_ALPHA_OP;
486      /*01420*/ EMIT_STATE(PE_ALPHA_OP, val);
487   }
488   if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
489      /*01424*/ EMIT_STATE(PE_ALPHA_BLEND_COLOR, ctx->blend_color.PE_ALPHA_BLEND_COLOR);
490   }
491   if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
492      uint32_t val = etna_blend_state(ctx->blend)->PE_ALPHA_CONFIG;
493      /*01428*/ EMIT_STATE(PE_ALPHA_CONFIG, val);
494   }
495   if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
496      uint32_t val;
497      /* Use the components and overwrite bits in framebuffer.PE_COLOR_FORMAT
498       * as a mask to enable the bits from blend PE_COLOR_FORMAT */
499      val = ~(VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
500              VIVS_PE_COLOR_FORMAT_OVERWRITE);
501      val |= etna_blend_state(ctx->blend)->PE_COLOR_FORMAT;
502      val &= ctx->framebuffer.PE_COLOR_FORMAT;
503      /*0142C*/ EMIT_STATE(PE_COLOR_FORMAT, val);
504   }
505   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
506      if (ctx->specs.pixel_pipes == 1) {
507         /*01430*/ EMIT_STATE_RELOC(PE_COLOR_ADDR, &ctx->framebuffer.PE_COLOR_ADDR);
508         /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
509         /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
510      } else if (ctx->specs.pixel_pipes == 2) {
511         /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
512         /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
513         /*01460*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(0), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[0]);
514         /*01464*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(1), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[1]);
515         /*01480*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(0), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[0]);
516         /*01484*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(1), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[1]);
517      } else {
518         abort();
519      }
520   }
521   if (unlikely(dirty & (ETNA_DIRTY_STENCIL_REF))) {
522      /*014A0*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT, ctx->stencil_ref.PE_STENCIL_CONFIG_EXT);
523   }
524   if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
525      struct etna_blend_state *blend = etna_blend_state(ctx->blend);
526      /*014A4*/ EMIT_STATE(PE_LOGIC_OP, blend->PE_LOGIC_OP | ctx->framebuffer.PE_LOGIC_OP);
527   }
528   if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
529      struct etna_blend_state *blend = etna_blend_state(ctx->blend);
530      for (int x = 0; x < 2; ++x) {
531         /*014A8*/ EMIT_STATE(PE_DITHER(x), blend->PE_DITHER[x]);
532      }
533   }
534   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_TS))) {
535      /*01654*/ EMIT_STATE(TS_MEM_CONFIG, ctx->framebuffer.TS_MEM_CONFIG);
536      /*01658*/ EMIT_STATE_RELOC(TS_COLOR_STATUS_BASE, &ctx->framebuffer.TS_COLOR_STATUS_BASE);
537      /*0165C*/ EMIT_STATE_RELOC(TS_COLOR_SURFACE_BASE, &ctx->framebuffer.TS_COLOR_SURFACE_BASE);
538      /*01660*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE, ctx->framebuffer.TS_COLOR_CLEAR_VALUE);
539      /*01664*/ EMIT_STATE_RELOC(TS_DEPTH_STATUS_BASE, &ctx->framebuffer.TS_DEPTH_STATUS_BASE);
540      /*01668*/ EMIT_STATE_RELOC(TS_DEPTH_SURFACE_BASE, &ctx->framebuffer.TS_DEPTH_SURFACE_BASE);
541      /*0166C*/ EMIT_STATE(TS_DEPTH_CLEAR_VALUE, ctx->framebuffer.TS_DEPTH_CLEAR_VALUE);
542   }
543   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
544      /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
545   }
546   etna_coalesce_end(stream, &coalesce);
547   /* end only EMIT_STATE */
548
549   /* Emit strongly architecture-specific state */
550   if (ctx->specs.halti >= 5)
551      emit_halti5_only_state(ctx, vs_output_count);
552   else
553      emit_pre_halti5_state(ctx);
554
555   ctx->emit_texture_state(ctx);
556
557   /* Insert a FE/PE stall as changing the shader instructions (and maybe
558    * the uniforms) can corrupt the previous in-progress draw operation.
559    * Observed with amoeba on GC2000 during the right-to-left rendering
560    * of PI, and can cause GPU hangs immediately after.
561    * I summise that this is because the "new" locations at 0xc000 are not
562    * properly protected against updates as other states seem to be. Hence,
563    * we detect the "new" vertex shader instruction offset to apply this. */
564   if (ctx->dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF) && ctx->specs.vs_offset > 0x4000)
565      etna_stall(ctx->stream, SYNC_RECIPIENT_FE, SYNC_RECIPIENT_PE);
566
567   /* We need to update the uniform cache only if one of the following bits are
568    * set in ctx->dirty:
569    * - ETNA_DIRTY_SHADER
570    * - ETNA_DIRTY_CONSTBUF
571    * - uniforms_dirty_bits
572    *
573    * In case of ETNA_DIRTY_SHADER we need load all uniforms from the cache. In
574    * all
575    * other cases we can load on the changed uniforms.
576    */
577   static const uint32_t uniform_dirty_bits =
578      ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
579
580   if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
581      etna_uniforms_write(
582         ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX],
583         ctx->shader_state.VS_UNIFORMS, &ctx->shader_state.vs_uniforms_size);
584
585   if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
586      etna_uniforms_write(
587         ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT],
588         ctx->shader_state.PS_UNIFORMS, &ctx->shader_state.ps_uniforms_size);
589
590   /**** Large dynamically-sized state ****/
591   bool do_uniform_flush = ctx->specs.halti < 5;
592   if (dirty & (ETNA_DIRTY_SHADER)) {
593      /* Special case: a new shader was loaded; simply re-load all uniforms and
594       * shader code at once */
595      /* This sequence is special, do not change ordering unless necessary. According to comment
596         snippets in the Vivante kernel driver a process called "steering" goes on while programming
597         shader state. This (as I understand it) means certain unified states are "steered"
598         toward a specific shader unit (VS/PS/...) based on either explicit flags in register
599         00860, or what other state is written before "auto-steering". So this means some
600         state can legitimately be programmed multiple times.
601       */
602
603      if (ctx->specs.halti >= 5) { /* ICACHE (HALTI5) */
604         assert(ctx->shader_state.VS_INST_ADDR.bo && ctx->shader_state.PS_INST_ADDR.bo);
605         /* Set icache (VS) */
606         etna_set_state(stream, VIVS_VS_NEWRANGE_LOW, 0);
607         etna_set_state(stream, VIVS_VS_NEWRANGE_HIGH, ctx->shader_state.vs_inst_mem_size / 4);
608         assert(ctx->shader_state.VS_INST_ADDR.bo);
609         etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
610         etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
611         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
612         etna_set_state(stream, VIVS_VS_ICACHE_COUNT, ctx->shader_state.vs_inst_mem_size / 4 - 1);
613
614         /* Set icache (PS) */
615         etna_set_state(stream, VIVS_PS_NEWRANGE_LOW, 0);
616         etna_set_state(stream, VIVS_PS_NEWRANGE_HIGH, ctx->shader_state.ps_inst_mem_size / 4);
617         assert(ctx->shader_state.PS_INST_ADDR.bo);
618         etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
619         etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
620         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
621         etna_set_state(stream, VIVS_PS_ICACHE_COUNT, ctx->shader_state.ps_inst_mem_size / 4 - 1);
622
623      } else if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
624         /* ICACHE (pre-HALTI5) */
625         assert(ctx->specs.has_icache && ctx->specs.has_shader_range_registers);
626         /* Set icache (VS) */
627         etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
628         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
629               VIVS_VS_ICACHE_CONTROL_ENABLE |
630               VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
631         assert(ctx->shader_state.VS_INST_ADDR.bo);
632         etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
633
634         /* Set icache (PS) */
635         etna_set_state(stream, VIVS_PS_RANGE, (ctx->shader_state.ps_inst_mem_size / 4 - 1) << 16);
636         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
637               VIVS_VS_ICACHE_CONTROL_ENABLE |
638               VIVS_VS_ICACHE_CONTROL_FLUSH_PS);
639         assert(ctx->shader_state.PS_INST_ADDR.bo);
640         etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
641      } else {
642         /* Upload shader directly, first flushing and disabling icache if
643          * supported on this hw */
644         if (ctx->specs.has_icache) {
645            etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
646                  VIVS_VS_ICACHE_CONTROL_FLUSH_PS |
647                  VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
648         }
649         if (ctx->specs.has_shader_range_registers) {
650            etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
651            etna_set_state(stream, VIVS_PS_RANGE, ((ctx->shader_state.ps_inst_mem_size / 4 - 1 + 0x100) << 16) |
652                                        0x100);
653         }
654         etna_set_state_multi(stream, ctx->specs.vs_offset,
655                              ctx->shader_state.vs_inst_mem_size,
656                              ctx->shader_state.VS_INST_MEM);
657         etna_set_state_multi(stream, ctx->specs.ps_offset,
658                              ctx->shader_state.ps_inst_mem_size,
659                              ctx->shader_state.PS_INST_MEM);
660      }
661
662      if (ctx->specs.has_unified_uniforms) {
663         etna_set_state(stream, VIVS_VS_UNIFORM_BASE, 0);
664         etna_set_state(stream, VIVS_PS_UNIFORM_BASE, ctx->specs.max_vs_uniforms);
665      }
666
667      if (do_uniform_flush)
668         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
669      etna_set_state_multi(stream, ctx->specs.vs_uniforms_offset,
670                                     ctx->shader_state.vs_uniforms_size,
671                                     ctx->shader_state.VS_UNIFORMS);
672      if (do_uniform_flush)
673         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
674      etna_set_state_multi(stream, ctx->specs.ps_uniforms_offset,
675                                     ctx->shader_state.ps_uniforms_size,
676                                     ctx->shader_state.PS_UNIFORMS);
677
678      /* Copy uniforms to gpu3d, so that incremental updates to uniforms are
679       * possible as long as the
680       * same shader remains bound */
681      memcpy(ctx->gpu3d.VS_UNIFORMS, ctx->shader_state.VS_UNIFORMS,
682             ctx->shader_state.vs_uniforms_size * 4);
683      memcpy(ctx->gpu3d.PS_UNIFORMS, ctx->shader_state.PS_UNIFORMS,
684             ctx->shader_state.ps_uniforms_size * 4);
685
686      if (ctx->specs.halti >= 5) {
687         /* HALTI5 needs to be prompted to pre-fetch shaders */
688         etna_set_state(stream, VIVS_VS_ICACHE_PREFETCH, 0x00000000);
689         etna_set_state(stream, VIVS_PS_ICACHE_PREFETCH, 0x00000000);
690         etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
691      }
692   } else {
693      /* ideally this cache would only be flushed if there are VS uniform changes */
694      if (do_uniform_flush)
695         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
696      etna_coalesce_start(stream, &coalesce);
697      for (int x = 0; x < ctx->shader.vs->uniforms.const_count; ++x) {
698         if (ctx->gpu3d.VS_UNIFORMS[x] != ctx->shader_state.VS_UNIFORMS[x]) {
699            etna_coalsence_emit(stream, &coalesce, ctx->specs.vs_uniforms_offset + x*4, ctx->shader_state.VS_UNIFORMS[x]);
700            ctx->gpu3d.VS_UNIFORMS[x] = ctx->shader_state.VS_UNIFORMS[x];
701         }
702      }
703      etna_coalesce_end(stream, &coalesce);
704
705      /* ideally this cache would only be flushed if there are PS uniform changes */
706      if (do_uniform_flush)
707         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
708      etna_coalesce_start(stream, &coalesce);
709      for (int x = 0; x < ctx->shader.fs->uniforms.const_count; ++x) {
710         if (ctx->gpu3d.PS_UNIFORMS[x] != ctx->shader_state.PS_UNIFORMS[x]) {
711            etna_coalsence_emit(stream, &coalesce, ctx->specs.ps_uniforms_offset + x*4, ctx->shader_state.PS_UNIFORMS[x]);
712            ctx->gpu3d.PS_UNIFORMS[x] = ctx->shader_state.PS_UNIFORMS[x];
713         }
714      }
715      etna_coalesce_end(stream, &coalesce);
716   }
717/**** End of state update ****/
718#undef EMIT_STATE
719#undef EMIT_STATE_FIXP
720#undef EMIT_STATE_RELOC
721   ctx->dirty = 0;
722   ctx->dirty_sampler_views = 0;
723}
724