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 "util/u_math.h"
  45
  46 struct etna_coalesce {
  47    uint32_t start;
  48    uint32_t last_reg;
  49    uint32_t last_fixp;
  50 };
  51
  52 /* Queue a STALL command (queues 2 words) */
  53 static inline void
  54 CMD_STALL(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
  55 {
  56    etna_cmd_stream_emit(stream, VIV_FE_STALL_HEADER_OP_STALL);
  57    etna_cmd_stream_emit(stream, VIV_FE_STALL_TOKEN_FROM(from) | VIV_FE_STALL_TOKEN_TO(to));
  58 }
  59
  60 void
  61 etna_stall(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
  62 {
  63    etna_cmd_stream_reserve(stream, 4);
  64
  65    etna_emit_load_state(stream, VIVS_GL_SEMAPHORE_TOKEN >> 2, 1, 0);
  66    etna_cmd_stream_emit(stream, VIVS_GL_SEMAPHORE_TOKEN_FROM(from) | VIVS_GL_SEMAPHORE_TOKEN_TO(to));
  67
  68    if (from == SYNC_RECIPIENT_FE) {
  69       /* if the frontend is to be stalled, queue a STALL frontend command */
  70       CMD_STALL(stream, from, to);
  71    } else {
  72       /* otherwise, load the STALL token state */
  73       etna_emit_load_state(stream, VIVS_GL_STALL_TOKEN >> 2, 1, 0);
  74       etna_cmd_stream_emit(stream, VIVS_GL_STALL_TOKEN_FROM(from) | VIVS_GL_STALL_TOKEN_TO(to));
  75    }
  76 }
  77
  78 static void
  79 etna_coalesce_start(struct etna_cmd_stream *stream,
  80                     struct etna_coalesce *coalesce)
  81 {
  82    coalesce->start = etna_cmd_stream_offset(stream);
  83    coalesce->last_reg = 0;
  84    coalesce->last_fixp = 0;
  85 }
  86
  87 static void
  88 etna_coalesce_end(struct etna_cmd_stream *stream,
  89                   struct etna_coalesce *coalesce)
  90 {
  91    uint32_t end = etna_cmd_stream_offset(stream);
  92    uint32_t size = end - coalesce->start;
  93
  94    if (size) {
  95       uint32_t offset = coalesce->start - 1;
  96       uint32_t value = etna_cmd_stream_get(stream, offset);
  97
  98       value |= VIV_FE_LOAD_STATE_HEADER_COUNT(size);
  99       etna_cmd_stream_set(stream, offset, value);
 100    }
 101
 102    /* append needed padding */
 103    if (end % 2 == 1)
 104       etna_cmd_stream_emit(stream, 0xdeadbeef);
 105 }
 106
 107 static void
 108 check_coalsence(struct etna_cmd_stream *stream, struct etna_coalesce *coalesce,
 109                 uint32_t reg, uint32_t fixp)
 110 {
 111    if (coalesce->last_reg != 0) {
 112       if (((coalesce->last_reg + 4) != reg) || (coalesce->last_fixp != fixp)) {
 113          etna_coalesce_end(stream, coalesce);
 114          etna_emit_load_state(stream, reg >> 2, 0, fixp);
 115          coalesce->start = etna_cmd_stream_offset(stream);
 116       }
 117    } else {
 118       etna_emit_load_state(stream, reg >> 2, 0, fixp);
 119       coalesce->start = etna_cmd_stream_offset(stream);
 120    }
 121
 122    coalesce->last_reg = reg;
 123    coalesce->last_fixp = fixp;
 124 }
 125
 126 static inline void
 127 etna_coalsence_emit(struct etna_cmd_stream *stream,
 128                     struct etna_coalesce *coalesce, uint32_t reg,
 129                     uint32_t value)
 130 {
 131    check_coalsence(stream, coalesce, reg, 0);
 132    etna_cmd_stream_emit(stream, value);
 133 }
 134
 135 static inline void
 136 etna_coalsence_emit_fixp(struct etna_cmd_stream *stream,
 137                          struct etna_coalesce *coalesce, uint32_t reg,
 138                          uint32_t value)
 139 {
 140    check_coalsence(stream, coalesce, reg, 1);
 141    etna_cmd_stream_emit(stream, value);
 142 }
 143
 144 static inline void
 145 etna_coalsence_emit_reloc(struct etna_cmd_stream *stream,
 146                           struct etna_coalesce *coalesce, uint32_t reg,
 147                           const struct etna_reloc *r)
 148 {
 149    if (r->bo) {
 150       check_coalsence(stream, coalesce, reg, 0);
 151       etna_cmd_stream_reloc(stream, r);
 152    }
 153 }
 154
 155 #define EMIT_STATE(state_name, src_value) \
 156    etna_coalsence_emit(stream, &coalesce, VIVS_##state_name, src_value)
 157
 158 #define EMIT_STATE_FIXP(state_name, src_value) \
 159    etna_coalsence_emit_fixp(stream, &coalesce, VIVS_##state_name, src_value)
 160
 161 #define EMIT_STATE_RELOC(state_name, src_value) \
 162    etna_coalsence_emit_reloc(stream, &coalesce, VIVS_##state_name, src_value)
 163
 164 /* submit RS state, without any processing and no dependence on context
 165  * except TS if this is a source-to-destination blit. */
 166 void
 167 etna_submit_rs_state(struct etna_context *ctx,
 168                      const struct compiled_rs_state *cs)
 169 {
 170    struct etna_screen *screen = etna_screen(ctx->base.screen);
 171    struct etna_cmd_stream *stream = ctx->stream;
 172    struct etna_coalesce coalesce;
 173
 174    ctx->stats.rs_operations++;
 175
 176    if (screen->specs.pixel_pipes == 1) {
 177       etna_cmd_stream_reserve(stream, 22);
 178       etna_coalesce_start(stream, &coalesce);
 179       /* 0/1 */ EMIT_STATE(RS_CONFIG, cs->RS_CONFIG);
 180       /* 2   */ EMIT_STATE_RELOC(RS_SOURCE_ADDR, &cs->source[0]);
 181       /* 3   */ EMIT_STATE(RS_SOURCE_STRIDE, cs->RS_SOURCE_STRIDE);
 182       /* 4   */ EMIT_STATE_RELOC(RS_DEST_ADDR, &cs->dest[0]);
 183       /* 5   */ EMIT_STATE(RS_DEST_STRIDE, cs->RS_DEST_STRIDE);
 184       /* 6/7 */ EMIT_STATE(RS_WINDOW_SIZE, cs->RS_WINDOW_SIZE);
 185       /* 8/9 */ EMIT_STATE(RS_DITHER(0), cs->RS_DITHER[0]);
 186       /*10   */ EMIT_STATE(RS_DITHER(1), cs->RS_DITHER[1]);
 187       /*11 - pad */
 188       /*12/13*/ EMIT_STATE(RS_CLEAR_CONTROL, cs->RS_CLEAR_CONTROL);
 189       /*14   */ EMIT_STATE(RS_FILL_VALUE(0), cs->RS_FILL_VALUE[0]);
 190       /*15   */ EMIT_STATE(RS_FILL_VALUE(1), cs->RS_FILL_VALUE[1]);
 191       /*16   */ EMIT_STATE(RS_FILL_VALUE(2), cs->RS_FILL_VALUE[2]);
 192       /*17   */ EMIT_STATE(RS_FILL_VALUE(3), cs->RS_FILL_VALUE[3]);
 193       /*18/19*/ EMIT_STATE(RS_EXTRA_CONFIG, cs->RS_EXTRA_CONFIG);
 194       /*20/21*/ EMIT_STATE(RS_KICKER, 0xbeebbeeb);
 195       etna_coalesce_end(stream, &coalesce);
 196    } else if (screen->specs.pixel_pipes == 2) {
 197       etna_cmd_stream_reserve(stream, 34); /* worst case - both pipes multi=1 */
 198       etna_coalesce_start(stream, &coalesce);
 199       /* 0/1 */ EMIT_STATE(RS_CONFIG, cs->RS_CONFIG);
 200       /* 2/3 */ EMIT_STATE(RS_SOURCE_STRIDE, cs->RS_SOURCE_STRIDE);
 201       /* 4/5 */ EMIT_STATE(RS_DEST_STRIDE, cs->RS_DEST_STRIDE);
 202       /* 6/7 */ EMIT_STATE_RELOC(RS_PIPE_SOURCE_ADDR(0), &cs->source[0]);
 203       if (cs->RS_SOURCE_STRIDE & VIVS_RS_SOURCE_STRIDE_MULTI) {
 204          /*8 */ EMIT_STATE_RELOC(RS_PIPE_SOURCE_ADDR(1), &cs->source[1]);
 205          /*9 - pad */
 206       }
 207       /*10/11*/ EMIT_STATE_RELOC(RS_PIPE_DEST_ADDR(0), &cs->dest[0]);
 208       if (cs->RS_DEST_STRIDE & VIVS_RS_DEST_STRIDE_MULTI) {
 209          /*12*/ EMIT_STATE_RELOC(RS_PIPE_DEST_ADDR(1), &cs->dest[1]);
 210          /*13 - pad */
 211       }
 212       /*14/15*/ EMIT_STATE(RS_PIPE_OFFSET(0), cs->RS_PIPE_OFFSET[0]);
 213       /*16   */ EMIT_STATE(RS_PIPE_OFFSET(1), cs->RS_PIPE_OFFSET[1]);
 214       /*17 - pad */
 215       /*18/19*/ EMIT_STATE(RS_WINDOW_SIZE, cs->RS_WINDOW_SIZE);
 216       /*20/21*/ EMIT_STATE(RS_DITHER(0), cs->RS_DITHER[0]);
 217       /*22   */ EMIT_STATE(RS_DITHER(1), cs->RS_DITHER[1]);
 218       /*23 - pad */
 219       /*24/25*/ EMIT_STATE(RS_CLEAR_CONTROL, cs->RS_CLEAR_CONTROL);
 220       /*26   */ EMIT_STATE(RS_FILL_VALUE(0), cs->RS_FILL_VALUE[0]);
 221       /*27   */ EMIT_STATE(RS_FILL_VALUE(1), cs->RS_FILL_VALUE[1]);
 222       /*28   */ EMIT_STATE(RS_FILL_VALUE(2), cs->RS_FILL_VALUE[2]);
 223       /*29   */ EMIT_STATE(RS_FILL_VALUE(3), cs->RS_FILL_VALUE[3]);
 224       /*30/31*/ EMIT_STATE(RS_EXTRA_CONFIG, cs->RS_EXTRA_CONFIG);
 225       /*32/33*/ EMIT_STATE(RS_KICKER, 0xbeebbeeb);
 226       etna_coalesce_end(stream, &coalesce);
 227    } else {
 228       abort();
 229    }
 230 }
 231
 232 /* Create bit field that specifies which samplers are active and thus need to be
 233  * programmed
 234  * 32 bits is enough for 32 samplers. As far as I know this is the upper bound
 235  * supported on any Vivante hw
 236  * up to GC4000.
 237  */
 238 static uint32_t
 239 active_samplers_bits(struct etna_context *ctx)
 240 {
 241    return ctx->active_sampler_views & ctx->active_samplers;
 242 }
 243
 244 #define ETNA_3D_CONTEXT_SIZE  (400) /* keep this number above "Total state updates (fixed)" from gen_weave_state tool */
 245
 246 static unsigned
 247 required_stream_size(struct etna_context *ctx)
 248 {
 249    unsigned size = ETNA_3D_CONTEXT_SIZE;
 250
 251    /* stall + flush */
 252    size += 2 + 4;
 253
 254    /* vertex elements */
 255    size += ctx->vertex_elements->num_elements + 1;
 256
 257    /* uniforms - worst case (2 words per uniform load) */
 258    size += ctx->shader.vs->uniforms.const_count * 2;
 259    size += ctx->shader.fs->uniforms.const_count * 2;
 260
 261    /* shader */
 262    size += ctx->shader_state.vs_inst_mem_size + 1;
 263    size += ctx->shader_state.ps_inst_mem_size + 1;
 264
 265    /* DRAW_INDEXED_PRIMITIVES command */
 266    size += 6;
 267
 268    /* reserve for alignment etc. */
 269    size += 64;
 270
 271    return size;
 272 }
 273
 274 /* Weave state before draw operation. This function merges all the compiled
 275  * state blocks under the context into one device register state. Parts of
 276  * this state that are changed since last call (dirty) will be uploaded as
 277  * state changes in the command buffer. */
 278 void
 279 etna_emit_state(struct etna_context *ctx)
 280 {
 281    struct etna_cmd_stream *stream = ctx->stream;
 282    uint32_t active_samplers = active_samplers_bits(ctx);
 283
 284    /* Pre-reserve the command buffer space which we are likely to need.
 285     * This must cover all the state emitted below, and the following
 286     * draw command. */
 287    etna_cmd_stream_reserve(stream, required_stream_size(ctx));
 288
 289    uint32_t dirty = ctx->dirty;
 290
 291    /* Pre-processing: see what caches we need to flush before making state changes. */
 292    uint32_t to_flush = 0;
 293    if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
 294       /* Need flush COLOR when changing PE.COLOR_FORMAT.OVERWRITE. */
 295 #if 0
 296         /* TODO*/
 297         if ((ctx->gpu3d.PE_COLOR_FORMAT & VIVS_PE_COLOR_FORMAT_OVERWRITE) !=
 298            (etna_blend_state(ctx->blend)->PE_COLOR_FORMAT & VIVS_PE_COLOR_FORMAT_OVERWRITE))
 299 #endif
 300       to_flush |= VIVS_GL_FLUSH_CACHE_COLOR;
 301    }
 302    if (unlikely(dirty & (ETNA_DIRTY_TEXTURE_CACHES)))
 303       to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE;
 304    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) /* Framebuffer config changed? */
 305       to_flush |= VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
 306    if (DBG_ENABLED(ETNA_DBG_CFLUSH_ALL))
 307       to_flush |= VIVS_GL_FLUSH_CACHE_TEXTURE | VIVS_GL_FLUSH_CACHE_COLOR | VIVS_GL_FLUSH_CACHE_DEPTH;
 308
 309    if (to_flush) {
 310       etna_set_state(stream, VIVS_GL_FLUSH_CACHE, to_flush);
 311       etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
 312    }
 313
 314    /* If MULTI_SAMPLE_CONFIG.MSAA_SAMPLES changed, clobber affected shader
 315     * state to make sure it is always rewritten. */
 316    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 317       if ((ctx->gpu3d.GL_MULTI_SAMPLE_CONFIG & VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_SAMPLES__MASK) !=
 318           (ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG & VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_SAMPLES__MASK)) {
 319          /* XXX what does the GPU set these states to on MSAA samples change?
 320           * Does it do the right thing?
 321           * (increase/decrease as necessary) or something else? Just set some
 322           * invalid value until we know for
 323           * sure. */
 324          ctx->gpu3d.PS_INPUT_COUNT = 0xffffffff;
 325          ctx->gpu3d.PS_TEMP_REGISTER_CONTROL = 0xffffffff;
 326       }
 327    }
 328
 329    /* Update vertex elements. This is different from any of the other states, in that
 330     * a) the number of vertex elements written matters: so write only active ones
 331     * b) the vertex element states must all be written: do not skip entries that stay the same */
 332    if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
 333       /* Special case: vertex elements must always be sent in full if changed */
 334       /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
 335          ctx->vertex_elements->num_elements,
 336          ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
 337    }
 338
 339    /* The following code is originally generated by gen_merge_state.py, to
 340     * emit state in increasing order of address (this makes it possible to merge
 341     * consecutive register updates into one SET_STATE command)
 342     *
 343     * There have been some manual changes, where the weaving operation is not
 344     * simply bitwise or:
 345     * - scissor fixp
 346     * - num vertex elements
 347     * - scissor handling
 348     * - num samplers
 349     * - texture lod
 350     * - ETNA_DIRTY_TS
 351     * - removed ETNA_DIRTY_BASE_SETUP statements -- these are guaranteed to not
 352     * change anyway
 353     * - PS / framebuffer interaction for MSAA
 354     * - move update of GL_MULTI_SAMPLE_CONFIG first
 355     * - add unlikely()/likely()
 356     */
 357    struct etna_coalesce coalesce;
 358
 359    etna_coalesce_start(stream, &coalesce);
 360
 361    /* begin only EMIT_STATE -- make sure no new etna_reserve calls are done here
 362     * directly
 363     *    or indirectly */
 364    /* multi sample config is set first, and outside of the normal sorting
 365     * order, as changing the multisample state clobbers PS.INPUT_COUNT (and
 366     * possibly PS.TEMP_REGISTER_CONTROL).
 367     */
 368    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SAMPLE_MASK))) {
 369       uint32_t val = VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_ENABLES(ctx->sample_mask);
 370       val |= ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG;
 371
 372       /*03818*/ EMIT_STATE(GL_MULTI_SAMPLE_CONFIG, val);
 373    }
 374    if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
 375       /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
 376       /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
 377    }
 378    if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
 379       /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
 380       /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
 381    }
 382    if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
 383       /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
 384    }
 385    if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
 386       for (int x = 1; x < ctx->vertex_buffer.count; ++x) {
 387          /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
 388       }
 389       for (int x = 1; x < ctx->vertex_buffer.count; ++x) {
 390          if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
 391             /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
 392          }
 393       }
 394    }
 395    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 396       /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
 397    }
 398    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
 399       bool point_size_per_vertex =
 400          etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex;
 401
 402       /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT,
 403                            point_size_per_vertex
 404                               ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
 405                               : ctx->shader_state.VS_OUTPUT_COUNT);
 406    }
 407    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
 408       /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
 409       /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
 410    }
 411    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 412       for (int x = 0; x < 4; ++x) {
 413          /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
 414       }
 415    }
 416    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
 417       for (int x = 0; x < 4; ++x) {
 418          /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
 419       }
 420    }
 421    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 422       /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
 423       /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
 424       if (ctx->specs.has_shader_range_registers) {
 425          /*0085C*/ EMIT_STATE(VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
 426       }
 427    }
 428    if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
 429       /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
 430       /*00A04*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_Y, ctx->viewport.PA_VIEWPORT_SCALE_Y);
 431       /*00A08*/ EMIT_STATE(PA_VIEWPORT_SCALE_Z, ctx->viewport.PA_VIEWPORT_SCALE_Z);
 432       /*00A0C*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_X, ctx->viewport.PA_VIEWPORT_OFFSET_X);
 433       /*00A10*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_Y, ctx->viewport.PA_VIEWPORT_OFFSET_Y);
 434       /*00A14*/ EMIT_STATE(PA_VIEWPORT_OFFSET_Z, ctx->viewport.PA_VIEWPORT_OFFSET_Z);
 435    }
 436    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 437       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 438
 439       /*00A18*/ EMIT_STATE(PA_LINE_WIDTH, rasterizer->PA_LINE_WIDTH);
 440       /*00A1C*/ EMIT_STATE(PA_POINT_SIZE, rasterizer->PA_POINT_SIZE);
 441       /*00A28*/ EMIT_STATE(PA_SYSTEM_MODE, rasterizer->PA_SYSTEM_MODE);
 442    }
 443    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 444       /*00A30*/ EMIT_STATE(PA_ATTRIBUTE_ELEMENT_COUNT, ctx->shader_state.PA_ATTRIBUTE_ELEMENT_COUNT);
 445    }
 446    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_SHADER))) {
 447       uint32_t val = etna_rasterizer_state(ctx->rasterizer)->PA_CONFIG;
 448       /*00A34*/ EMIT_STATE(PA_CONFIG, val & ctx->shader_state.PA_CONFIG);
 449    }
 450    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 451       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 452       /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
 453       /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
 454    }
 455    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 456       for (int x = 0; x < 10; ++x) {
 457          /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
 458       }
 459    }
 460    if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
 461                          ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
 462       /* this is a bit of a mess: rasterizer.scissor determines whether to use
 463        * only the framebuffer scissor, or specific scissor state, and the
 464        * viewport clips too so the logic spans four CSOs */
 465       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 466
 467       uint32_t scissor_left =
 468          MAX2(ctx->framebuffer.SE_SCISSOR_LEFT, ctx->viewport.SE_SCISSOR_LEFT);
 469       uint32_t scissor_top =
 470          MAX2(ctx->framebuffer.SE_SCISSOR_TOP, ctx->viewport.SE_SCISSOR_TOP);
 471       uint32_t scissor_right =
 472          MIN2(ctx->framebuffer.SE_SCISSOR_RIGHT, ctx->viewport.SE_SCISSOR_RIGHT);
 473       uint32_t scissor_bottom =
 474          MIN2(ctx->framebuffer.SE_SCISSOR_BOTTOM, ctx->viewport.SE_SCISSOR_BOTTOM);
 475
 476       if (rasterizer->scissor) {
 477          scissor_left = MAX2(ctx->scissor.SE_SCISSOR_LEFT, scissor_left);
 478          scissor_top = MAX2(ctx->scissor.SE_SCISSOR_TOP, scissor_top);
 479          scissor_right = MIN2(ctx->scissor.SE_SCISSOR_RIGHT, scissor_right);
 480          scissor_bottom = MIN2(ctx->scissor.SE_SCISSOR_BOTTOM, scissor_bottom);
 481       }
 482
 483       /*00C00*/ EMIT_STATE_FIXP(SE_SCISSOR_LEFT, scissor_left);
 484       /*00C04*/ EMIT_STATE_FIXP(SE_SCISSOR_TOP, scissor_top);
 485       /*00C08*/ EMIT_STATE_FIXP(SE_SCISSOR_RIGHT, scissor_right);
 486       /*00C0C*/ EMIT_STATE_FIXP(SE_SCISSOR_BOTTOM, scissor_bottom);
 487    }
 488    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 489       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 490
 491       /*00C10*/ EMIT_STATE(SE_DEPTH_SCALE, rasterizer->SE_DEPTH_SCALE);
 492       /*00C14*/ EMIT_STATE(SE_DEPTH_BIAS, rasterizer->SE_DEPTH_BIAS);
 493       /*00C18*/ EMIT_STATE(SE_CONFIG, rasterizer->SE_CONFIG);
 494    }
 495    if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
 496                          ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
 497       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 498
 499       uint32_t clip_right =
 500          MIN2(ctx->framebuffer.SE_CLIP_RIGHT, ctx->viewport.SE_CLIP_RIGHT);
 501       uint32_t clip_bottom =
 502          MIN2(ctx->framebuffer.SE_CLIP_BOTTOM, ctx->viewport.SE_CLIP_BOTTOM);
 503
 504       if (rasterizer->scissor) {
 505          clip_right = MIN2(ctx->scissor.SE_CLIP_RIGHT, clip_right);
 506          clip_bottom = MIN2(ctx->scissor.SE_CLIP_BOTTOM, clip_bottom);
 507       }
 508
 509       /*00C20*/ EMIT_STATE_FIXP(SE_CLIP_RIGHT, clip_right);
 510       /*00C24*/ EMIT_STATE_FIXP(SE_CLIP_BOTTOM, clip_bottom);
 511    }
 512    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 513       /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
 514    }
 515    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 516       /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
 517       for (int x = 0; x < 4; ++x) {
 518          /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
 519       }
 520       for (int x = 0; x < 16; ++x) {
 521          /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
 522       }
 523    }
 524    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
 525       /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
 526       /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
 527       /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
 528                            ctx->framebuffer.msaa_mode
 529                               ? ctx->shader_state.PS_INPUT_COUNT_MSAA
 530                               : ctx->shader_state.PS_INPUT_COUNT);
 531       /*0100C*/ EMIT_STATE(PS_TEMP_REGISTER_CONTROL,
 532                            ctx->framebuffer.msaa_mode
 533                               ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
 534                               : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
 535       /*01010*/ EMIT_STATE(PS_CONTROL, ctx->shader_state.PS_CONTROL);
 536       /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
 537       if (ctx->specs.has_shader_range_registers) {
 538          /*0101C*/ EMIT_STATE(PS_RANGE, ((ctx->shader_state.ps_inst_mem_size / 4 - 1 + 0x100) << 16) |
 539                                         0x100);
 540       }
 541    }
 542    if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER))) {
 543       uint32_t val = etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG;
 544       /*01400*/ EMIT_STATE(PE_DEPTH_CONFIG, val | ctx->framebuffer.PE_DEPTH_CONFIG);
 545    }
 546    if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
 547       /*01404*/ EMIT_STATE(PE_DEPTH_NEAR, ctx->viewport.PE_DEPTH_NEAR);
 548       /*01408*/ EMIT_STATE(PE_DEPTH_FAR, ctx->viewport.PE_DEPTH_FAR);
 549    }
 550    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 551       /*0140C*/ EMIT_STATE(PE_DEPTH_NORMALIZE, ctx->framebuffer.PE_DEPTH_NORMALIZE);
 552
 553       if (ctx->specs.pixel_pipes == 1) {
 554          /*01410*/ EMIT_STATE_RELOC(PE_DEPTH_ADDR, &ctx->framebuffer.PE_DEPTH_ADDR);
 555       }
 556
 557       /*01414*/ EMIT_STATE(PE_DEPTH_STRIDE, ctx->framebuffer.PE_DEPTH_STRIDE);
 558    }
 559    if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
 560       uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_OP;
 561       /*01418*/ EMIT_STATE(PE_STENCIL_OP, val);
 562    }
 563    if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_STENCIL_REF))) {
 564       uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG;
 565       /*0141C*/ EMIT_STATE(PE_STENCIL_CONFIG, val | ctx->stencil_ref.PE_STENCIL_CONFIG);
 566    }
 567    if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
 568       uint32_t val = etna_zsa_state(ctx->zsa)->PE_ALPHA_OP;
 569       /*01420*/ EMIT_STATE(PE_ALPHA_OP, val);
 570    }
 571    if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
 572       /*01424*/ EMIT_STATE(PE_ALPHA_BLEND_COLOR, ctx->blend_color.PE_ALPHA_BLEND_COLOR);
 573    }
 574    if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
 575       uint32_t val = etna_blend_state(ctx->blend)->PE_ALPHA_CONFIG;
 576       /*01428*/ EMIT_STATE(PE_ALPHA_CONFIG, val);
 577    }
 578    if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
 579       uint32_t val;
 580       /* Use the components and overwrite bits in framebuffer.PE_COLOR_FORMAT
 581        * as a mask to enable the bits from blend PE_COLOR_FORMAT */
 582       val = ~(VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
 583               VIVS_PE_COLOR_FORMAT_OVERWRITE);
 584       val |= etna_blend_state(ctx->blend)->PE_COLOR_FORMAT;
 585       val &= ctx->framebuffer.PE_COLOR_FORMAT;
 586       /*0142C*/ EMIT_STATE(PE_COLOR_FORMAT, val);
 587    }
 588    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 589       if (ctx->specs.pixel_pipes == 1) {
 590          /*01430*/ EMIT_STATE_RELOC(PE_COLOR_ADDR, &ctx->framebuffer.PE_COLOR_ADDR);
 591          /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
 592          /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
 593       } else if (ctx->specs.pixel_pipes == 2) {
 594          /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
 595          /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
 596          /*01460*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(0), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[0]);
 597          /*01464*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(1), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[1]);
 598          /*01480*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(0), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[0]);
 599          /*01484*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(1), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[1]);
 600       } else {
 601          abort();
 602       }
 603    }
 604    if (unlikely(dirty & (ETNA_DIRTY_STENCIL_REF))) {
 605       /*014A0*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT, ctx->stencil_ref.PE_STENCIL_CONFIG_EXT);
 606    }
 607    if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
 608       struct etna_blend_state *blend = etna_blend_state(ctx->blend);
 609       /*014A4*/ EMIT_STATE(PE_LOGIC_OP, blend->PE_LOGIC_OP | ctx->framebuffer.PE_LOGIC_OP);
 610    }
 611    if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
 612       struct etna_blend_state *blend = etna_blend_state(ctx->blend);
 613       for (int x = 0; x < 2; ++x) {
 614          /*014A8*/ EMIT_STATE(PE_DITHER(x), blend->PE_DITHER[x]);
 615       }
 616    }
 617    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_TS))) {
 618       /*01654*/ EMIT_STATE(TS_MEM_CONFIG, ctx->framebuffer.TS_MEM_CONFIG);
 619       /*01658*/ EMIT_STATE_RELOC(TS_COLOR_STATUS_BASE, &ctx->framebuffer.TS_COLOR_STATUS_BASE);
 620       /*0165C*/ EMIT_STATE_RELOC(TS_COLOR_SURFACE_BASE, &ctx->framebuffer.TS_COLOR_SURFACE_BASE);
 621       /*01660*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE, ctx->framebuffer.TS_COLOR_CLEAR_VALUE);
 622       /*01664*/ EMIT_STATE_RELOC(TS_DEPTH_STATUS_BASE, &ctx->framebuffer.TS_DEPTH_STATUS_BASE);
 623       /*01668*/ EMIT_STATE_RELOC(TS_DEPTH_SURFACE_BASE, &ctx->framebuffer.TS_DEPTH_SURFACE_BASE);
 624       /*0166C*/ EMIT_STATE(TS_DEPTH_CLEAR_VALUE, ctx->framebuffer.TS_DEPTH_CLEAR_VALUE);
 625    }
 626    if (unlikely(dirty & (ETNA_DIRTY_SAMPLER_VIEWS | ETNA_DIRTY_SAMPLERS))) {
 627       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 628          uint32_t val = 0; /* 0 == sampler inactive */
 629
 630          /* set active samplers to their configuration value (determined by both
 631           * the sampler state and sampler view) */
 632          if ((1 << x) & active_samplers) {
 633             struct etna_sampler_state *ss = etna_sampler_state(ctx->sampler[x]);
 634             struct etna_sampler_view *sv = etna_sampler_view(ctx->sampler_view[x]);
 635
 636             val = (ss->TE_SAMPLER_CONFIG0 & sv->TE_SAMPLER_CONFIG0_MASK) |
 637                   sv->TE_SAMPLER_CONFIG0;
 638          }
 639
 640          /*02000*/ EMIT_STATE(TE_SAMPLER_CONFIG0(x), val);
 641       }
 642    }
 643    if (unlikely(dirty & (ETNA_DIRTY_SAMPLER_VIEWS))) {
 644       struct etna_sampler_view *sv;
 645
 646       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 647          if ((1 << x) & active_samplers) {
 648             sv = etna_sampler_view(ctx->sampler_view[x]);
 649             /*02040*/ EMIT_STATE(TE_SAMPLER_SIZE(x), sv->TE_SAMPLER_SIZE);
 650          }
 651       }
 652       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 653          if ((1 << x) & active_samplers) {
 654             sv = etna_sampler_view(ctx->sampler_view[x]);
 655             /*02080*/ EMIT_STATE(TE_SAMPLER_LOG_SIZE(x), sv->TE_SAMPLER_LOG_SIZE);
 656          }
 657       }
 658    }
 659    if (unlikely(dirty & (ETNA_DIRTY_SAMPLER_VIEWS | ETNA_DIRTY_SAMPLERS))) {
 660       struct etna_sampler_state *ss;
 661       struct etna_sampler_view *sv;
 662
 663       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 664          if ((1 << x) & active_samplers) {
 665             ss = etna_sampler_state(ctx->sampler[x]);
 666             sv = etna_sampler_view(ctx->sampler_view[x]);
 667
 668             /* min and max lod is determined both by the sampler and the view */
 669             /*020C0*/ EMIT_STATE(TE_SAMPLER_LOD_CONFIG(x),
 670                                  ss->TE_SAMPLER_LOD_CONFIG |
 671                                  VIVS_TE_SAMPLER_LOD_CONFIG_MAX(MIN2(ss->max_lod, sv->max_lod)) |
 672                                  VIVS_TE_SAMPLER_LOD_CONFIG_MIN(MAX2(ss->min_lod, sv->min_lod)));
 673          }
 674       }
 675       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 676          if ((1 << x) & active_samplers) {
 677             ss = etna_sampler_state(ctx->sampler[x]);
 678             sv = etna_sampler_view(ctx->sampler_view[x]);
 679
 680             /*021C0*/ EMIT_STATE(TE_SAMPLER_CONFIG1(x), ss->TE_SAMPLER_CONFIG1 |
 681                                                         sv->TE_SAMPLER_CONFIG1);
 682          }
 683       }
 684    }
 685    if (unlikely(dirty & (ETNA_DIRTY_SAMPLER_VIEWS))) {
 686       for (int y = 0; y < VIVS_TE_SAMPLER_LOD_ADDR__LEN; ++y) {
 687          for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
 688             if ((1 << x) & active_samplers) {
 689                struct etna_sampler_view *sv = etna_sampler_view(ctx->sampler_view[x]);
 690                /*02400*/ EMIT_STATE_RELOC(TE_SAMPLER_LOD_ADDR(x, y),&sv->TE_SAMPLER_LOD_ADDR[y]);
 691             }
 692          }
 693       }
 694    }
 695    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 696       /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
 697       /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
 698       for (int x = 0; x < 2; ++x) {
 699          /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
 700       }
 701    }
 702    etna_coalesce_end(stream, &coalesce);
 703    /* end only EMIT_STATE */
 704
 705    /* Insert a FE/PE stall as changing the shader instructions (and maybe
 706     * the uniforms) can corrupt the previous in-progress draw operation.
 707     * Observed with amoeba on GC2000 during the right-to-left rendering
 708     * of PI, and can cause GPU hangs immediately after.
 709     * I summise that this is because the "new" locations at 0xc000 are not
 710     * properly protected against updates as other states seem to be. Hence,
 711     * we detect the "new" vertex shader instruction offset to apply this. */
 712    if (ctx->dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF) && ctx->specs.vs_offset > 0x4000)
 713       etna_stall(ctx->stream, SYNC_RECIPIENT_FE, SYNC_RECIPIENT_PE);
 714
 715    /* We need to update the uniform cache only if one of the following bits are
 716     * set in ctx->dirty:
 717     * - ETNA_DIRTY_SHADER
 718     * - ETNA_DIRTY_CONSTBUF
 719     * - uniforms_dirty_bits
 720     *
 721     * In case of ETNA_DIRTY_SHADER we need load all uniforms from the cache. In
 722     * all
 723     * other cases we can load on the changed uniforms.
 724     */
 725    static const uint32_t uniform_dirty_bits =
 726       ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
 727
 728    if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
 729       etna_uniforms_write(
 730          ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX],
 731          ctx->shader_state.VS_UNIFORMS, &ctx->shader_state.vs_uniforms_size);
 732
 733    if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
 734       etna_uniforms_write(
 735          ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT],
 736          ctx->shader_state.PS_UNIFORMS, &ctx->shader_state.ps_uniforms_size);
 737
 738    /**** Large dynamically-sized state ****/
 739    if (dirty & (ETNA_DIRTY_SHADER)) {
 740       /* Special case: a new shader was loaded; simply re-load all uniforms and
 741        * shader code at once */
 742       /*04000 or 0C000*/
 743       etna_set_state_multi(stream, ctx->specs.vs_offset,
 744                            ctx->shader_state.vs_inst_mem_size,
 745                            ctx->shader_state.VS_INST_MEM);
 746       /*06000 or 0D000*/
 747       etna_set_state_multi(stream, ctx->specs.ps_offset,
 748                            ctx->shader_state.ps_inst_mem_size,
 749                            ctx->shader_state.PS_INST_MEM);
 750       /*05000*/ etna_set_state_multi(stream, VIVS_VS_UNIFORMS(0),
 751                                      ctx->shader_state.vs_uniforms_size,
 752                                      ctx->shader_state.VS_UNIFORMS);
 753       /*07000*/ etna_set_state_multi(stream, VIVS_PS_UNIFORMS(0),
 754                                      ctx->shader_state.ps_uniforms_size,
 755                                      ctx->shader_state.PS_UNIFORMS);
 756
 757       /* Copy uniforms to gpu3d, so that incremental updates to uniforms are
 758        * possible as long as the
 759        * same shader remains bound */
 760       ctx->gpu3d.vs_uniforms_size = ctx->shader_state.vs_uniforms_size;
 761       ctx->gpu3d.ps_uniforms_size = ctx->shader_state.ps_uniforms_size;
 762       memcpy(ctx->gpu3d.VS_UNIFORMS, ctx->shader_state.VS_UNIFORMS,
 763              ctx->shader_state.vs_uniforms_size * 4);
 764       memcpy(ctx->gpu3d.PS_UNIFORMS, ctx->shader_state.PS_UNIFORMS,
 765              ctx->shader_state.ps_uniforms_size * 4);
 766    } else {
 767       etna_coalesce_start(stream, &coalesce);
 768       for (int x = 0; x < ctx->shader.vs->uniforms.const_count; ++x) {
 769          if (ctx->gpu3d.VS_UNIFORMS[x] != ctx->shader_state.VS_UNIFORMS[x]) {
 770             /*05000*/ EMIT_STATE(VS_UNIFORMS(x), ctx->shader_state.VS_UNIFORMS[x]);
 771             ctx->gpu3d.VS_UNIFORMS[x] = ctx->shader_state.VS_UNIFORMS[x];
 772          }
 773       }
 774       etna_coalesce_end(stream, &coalesce);
 775
 776       etna_coalesce_start(stream, &coalesce);
 777       for (int x = 0; x < ctx->shader.fs->uniforms.const_count; ++x) {
 778          if (ctx->gpu3d.PS_UNIFORMS[x] != ctx->shader_state.PS_UNIFORMS[x]) {
 779             /*07000*/ EMIT_STATE(PS_UNIFORMS(x), ctx->shader_state.PS_UNIFORMS[x]);
 780             ctx->gpu3d.PS_UNIFORMS[x] = ctx->shader_state.PS_UNIFORMS[x];
 781          }
 782       }
 783       etna_coalesce_end(stream, &coalesce);
 784    }
 785 /**** End of state update ****/
 786 #undef EMIT_STATE
 787 #undef EMIT_STATE_FIXP
 788 #undef EMIT_STATE_RELOC
 789    ctx->dirty = 0;
 790 }