src/gallium/drivers/etnaviv/etnaviv_emit.c

   1 /*
   2  * Copyright (c) 2014-2015 Etnaviv Project
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sub license,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the
  12  * next paragraph) shall be included in all copies or substantial portions
  13  * of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Wladimir J. van der Laan <laanwj@gmail.com>
  25  */
  26
  27 #include "etnaviv_emit.h"
  28
  29 #include "etnaviv_blend.h"
  30 #include "etnaviv_compiler.h"
  31 #include "etnaviv_context.h"
  32 #include "etnaviv_rasterizer.h"
  33 #include "etnaviv_resource.h"
  34 #include "etnaviv_rs.h"
  35 #include "etnaviv_screen.h"
  36 #include "etnaviv_shader.h"
  37 #include "etnaviv_texture.h"
  38 #include "etnaviv_translate.h"
  39 #include "etnaviv_uniforms.h"
  40 #include "etnaviv_util.h"
  41 #include "etnaviv_zsa.h"
  42 #include "hw/common.xml.h"
  43 #include "hw/state.xml.h"
  44 #include "hw/state_blt.xml.h"
  45 #include "util/u_math.h"
  46
  47 /* Queue a STALL command (queues 2 words) */
  48 static inline void
  49 CMD_STALL(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
  50 {
  51    etna_cmd_stream_emit(stream, VIV_FE_STALL_HEADER_OP_STALL);
  52    etna_cmd_stream_emit(stream, VIV_FE_STALL_TOKEN_FROM(from) | VIV_FE_STALL_TOKEN_TO(to));
  53 }
  54
  55 void
  56 etna_stall(struct etna_cmd_stream *stream, uint32_t from, uint32_t to)
  57 {
  58    bool blt = (from == SYNC_RECIPIENT_BLT) || (to == SYNC_RECIPIENT_BLT);
  59    etna_cmd_stream_reserve(stream, blt ? 8 : 4);
  60
  61    if (blt) {
  62       etna_emit_load_state(stream, VIVS_BLT_ENABLE >> 2, 1, 0);
  63       etna_cmd_stream_emit(stream, 1);
  64    }
  65
  66    /* TODO: set bit 28/29 of token after BLT COPY_BUFFER */
  67    etna_emit_load_state(stream, VIVS_GL_SEMAPHORE_TOKEN >> 2, 1, 0);
  68    etna_cmd_stream_emit(stream, VIVS_GL_SEMAPHORE_TOKEN_FROM(from) | VIVS_GL_SEMAPHORE_TOKEN_TO(to));
  69
  70    if (from == SYNC_RECIPIENT_FE) {
  71       /* if the frontend is to be stalled, queue a STALL frontend command */
  72       CMD_STALL(stream, from, to);
  73    } else {
  74       /* otherwise, load the STALL token state */
  75       etna_emit_load_state(stream, VIVS_GL_STALL_TOKEN >> 2, 1, 0);
  76       etna_cmd_stream_emit(stream, VIVS_GL_STALL_TOKEN_FROM(from) | VIVS_GL_STALL_TOKEN_TO(to));
  77    }
  78
  79    if (blt) {
  80       etna_emit_load_state(stream, VIVS_BLT_ENABLE >> 2, 1, 0);
  81       etna_cmd_stream_emit(stream, 0);
  82    }
  83 }
  84
  85 #define EMIT_STATE(state_name, src_value) \
  86    etna_coalsence_emit(stream, &coalesce, VIVS_##state_name, src_value)
  87
  88 #define EMIT_STATE_FIXP(state_name, src_value) \
  89    etna_coalsence_emit_fixp(stream, &coalesce, VIVS_##state_name, src_value)
  90
  91 #define EMIT_STATE_RELOC(state_name, src_value) \
  92    etna_coalsence_emit_reloc(stream, &coalesce, VIVS_##state_name, src_value)
  93
  94 #define ETNA_3D_CONTEXT_SIZE  (400) /* keep this number above "Total state updates (fixed)" from gen_weave_state tool */
  95
  96 static unsigned
  97 required_stream_size(struct etna_context *ctx)
  98 {
  99    unsigned size = ETNA_3D_CONTEXT_SIZE;
 100
 101    /* stall + flush */
 102    size += 2 + 4;
 103
 104    /* vertex elements */
 105    size += ctx->vertex_elements->num_elements + 1;
 106
 107    /* uniforms - worst case (2 words per uniform load) */
 108    size += ctx->shader.vs->uniforms.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+ */
 125 static void
 126 emit_halti5_only_state(struct etna_context *ctx, int vs_output_count)
 127 {
 128    struct etna_cmd_stream *stream = ctx->stream;
 129    uint32_t dirty = ctx->dirty;
 130    struct etna_coalesce coalesce;
 131
 132    etna_coalesce_start(stream, &coalesce);
 133    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 134       /* Magic states (load balancing, inter-unit sync, buffers) */
 135       /*00870*/ EMIT_STATE(VS_HALTI5_OUTPUT_COUNT, vs_output_count | ((vs_output_count * 0x10) << 8));
 136       /*008A0*/ EMIT_STATE(VS_HALTI5_UNK008A0, 0x0001000e | ((0x110/vs_output_count) << 20));
 137       for (int x = 0; x < 4; ++x) {
 138          /*008E0*/ EMIT_STATE(VS_HALTI5_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
 139       }
 140    }
 141    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
 142       for (int x = 0; x < 4; ++x) {
 143          /*008C0*/ EMIT_STATE(VS_HALTI5_INPUT(x), ctx->shader_state.VS_INPUT[x]);
 144       }
 145    }
 146    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 147       /*00A90*/ EMIT_STATE(PA_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
 148       /*00AA8*/ EMIT_STATE(PA_VS_OUTPUT_COUNT, vs_output_count);
 149       /*01080*/ EMIT_STATE(PS_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
 150       /*03888*/ EMIT_STATE(GL_HALTI5_SH_SPECIALS, ctx->shader_state.GL_HALTI5_SH_SPECIALS);
 151    }
 152    etna_coalesce_end(stream, &coalesce);
 153 }
 154
 155 /* Emit state that no longer exists on HALTI5 */
 156 static void
 157 emit_pre_halti5_state(struct etna_context *ctx)
 158 {
 159    struct etna_cmd_stream *stream = ctx->stream;
 160    uint32_t dirty = ctx->dirty;
 161    struct etna_coalesce coalesce;
 162
 163    etna_coalesce_start(stream, &coalesce);
 164    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 165       /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
 166    }
 167    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 168       for (int x = 0; x < 4; ++x) {
 169         /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
 170       }
 171    }
 172    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
 173       for (int x = 0; x < 4; ++x) {
 174         /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
 175       }
 176    }
 177    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 178       /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
 179    }
 180    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 181       for (int x = 0; x < 10; ++x) {
 182          /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
 183       }
 184    }
 185    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 186       /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
 187       for (int x = 0; x < 4; ++x) {
 188          /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
 189       }
 190       for (int x = 0; x < 16; ++x) {
 191          /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
 192       }
 193    }
 194    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
 195       /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
 196    }
 197    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
 198       /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
 199    }
 200    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 201       /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
 202       for (int x = 0; x < 2; ++x) {
 203          /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
 204       }
 205    }
 206    etna_coalesce_end(stream, &coalesce);
 207 }
 208
 209 /* Weave state before draw operation. This function merges all the compiled
 210  * state blocks under the context into one device register state. Parts of
 211  * this state that are changed since last call (dirty) will be uploaded as
 212  * state changes in the command buffer. */
 213 void
 214 etna_emit_state(struct etna_context *ctx)
 215 {
 216    struct etna_cmd_stream *stream = ctx->stream;
 217
 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    /* If MULTI_SAMPLE_CONFIG.MSAA_SAMPLES changed, clobber affected shader
 254     * state to make sure it is always rewritten. */
 255    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 256       if ((ctx->gpu3d.GL_MULTI_SAMPLE_CONFIG & VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_SAMPLES__MASK) !=
 257           (ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG & VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_SAMPLES__MASK)) {
 258          /* XXX what does the GPU set these states to on MSAA samples change?
 259           * Does it do the right thing?
 260           * (increase/decrease as necessary) or something else? Just set some
 261           * invalid value until we know for
 262           * sure. */
 263          ctx->gpu3d.PS_INPUT_COUNT = 0xffffffff;
 264          ctx->gpu3d.PS_TEMP_REGISTER_CONTROL = 0xffffffff;
 265       }
 266    }
 267
 268    /* Update vertex elements. This is different from any of the other states, in that
 269     * a) the number of vertex elements written matters: so write only active ones
 270     * b) the vertex element states must all be written: do not skip entries that stay the same */
 271    if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
 272       if (ctx->specs.halti >= 5) {
 273          /*17800*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG0(0),
 274             ctx->vertex_elements->num_elements,
 275             ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG0);
 276          /*17A00*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_SCALE(0),
 277             ctx->vertex_elements->num_elements,
 278             ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
 279          /*17A80*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG1(0),
 280             ctx->vertex_elements->num_elements,
 281             ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG1);
 282       } else {
 283          /* Special case: vertex elements must always be sent in full if changed */
 284          /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
 285             ctx->vertex_elements->num_elements,
 286             ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
 287          if (ctx->specs.halti >= 2) {
 288             /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
 289                ctx->vertex_elements->num_elements,
 290                ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
 291          }
 292       }
 293    }
 294    unsigned vs_output_count = etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex
 295                            ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
 296                            : ctx->shader_state.VS_OUTPUT_COUNT;
 297
 298    /* The following code is originally generated by gen_merge_state.py, to
 299     * emit state in increasing order of address (this makes it possible to merge
 300     * consecutive register updates into one SET_STATE command)
 301     *
 302     * There have been some manual changes, where the weaving operation is not
 303     * simply bitwise or:
 304     * - scissor fixp
 305     * - num vertex elements
 306     * - scissor handling
 307     * - num samplers
 308     * - texture lod
 309     * - ETNA_DIRTY_TS
 310     * - removed ETNA_DIRTY_BASE_SETUP statements -- these are guaranteed to not
 311     * change anyway
 312     * - PS / framebuffer interaction for MSAA
 313     * - move update of GL_MULTI_SAMPLE_CONFIG first
 314     * - add unlikely()/likely()
 315     */
 316    struct etna_coalesce coalesce;
 317
 318    etna_coalesce_start(stream, &coalesce);
 319
 320    /* begin only EMIT_STATE -- make sure no new etna_reserve calls are done here
 321     * directly
 322     *    or indirectly */
 323    /* multi sample config is set first, and outside of the normal sorting
 324     * order, as changing the multisample state clobbers PS.INPUT_COUNT (and
 325     * possibly PS.TEMP_REGISTER_CONTROL).
 326     */
 327    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_SAMPLE_MASK))) {
 328       uint32_t val = VIVS_GL_MULTI_SAMPLE_CONFIG_MSAA_ENABLES(ctx->sample_mask);
 329       val |= ctx->framebuffer.GL_MULTI_SAMPLE_CONFIG;
 330
 331       /*03818*/ EMIT_STATE(GL_MULTI_SAMPLE_CONFIG, val);
 332    }
 333    if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
 334       /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
 335       /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
 336    }
 337    if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
 338       /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
 339    }
 340    if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
 341       if (ctx->specs.halti >= 2) { /* HALTI2+: NFE_VERTEX_STREAMS */
 342          for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
 343             /*14600*/ EMIT_STATE_RELOC(NFE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
 344          }
 345          for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
 346             if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
 347                /*14640*/ EMIT_STATE(NFE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
 348             }
 349          }
 350          for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
 351             if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
 352                /*14680*/ EMIT_STATE(NFE_VERTEX_STREAMS_UNK14680(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_UNK14680);
 353             }
 354          }
 355       } else if(ctx->specs.stream_count >= 1) { /* hw w/ multiple vertex streams */
 356          for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
 357             /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
 358          }
 359          for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
 360             if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
 361                /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
 362             }
 363          }
 364       } else { /* hw w/ single vertex stream */
 365          /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
 366          /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
 367       }
 368    }
 369    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
 370
 371       /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT, vs_output_count);
 372    }
 373    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
 374       /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
 375       /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
 376    }
 377    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 378       /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
 379    }
 380    if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
 381       /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
 382       /*00A04*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_Y, ctx->viewport.PA_VIEWPORT_SCALE_Y);
 383       /*00A08*/ EMIT_STATE(PA_VIEWPORT_SCALE_Z, ctx->viewport.PA_VIEWPORT_SCALE_Z);
 384       /*00A0C*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_X, ctx->viewport.PA_VIEWPORT_OFFSET_X);
 385       /*00A10*/ EMIT_STATE_FIXP(PA_VIEWPORT_OFFSET_Y, ctx->viewport.PA_VIEWPORT_OFFSET_Y);
 386       /*00A14*/ EMIT_STATE(PA_VIEWPORT_OFFSET_Z, ctx->viewport.PA_VIEWPORT_OFFSET_Z);
 387    }
 388    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 389       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 390
 391       /*00A18*/ EMIT_STATE(PA_LINE_WIDTH, rasterizer->PA_LINE_WIDTH);
 392       /*00A1C*/ EMIT_STATE(PA_POINT_SIZE, rasterizer->PA_POINT_SIZE);
 393       /*00A28*/ EMIT_STATE(PA_SYSTEM_MODE, rasterizer->PA_SYSTEM_MODE);
 394    }
 395    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 396       /*00A30*/ EMIT_STATE(PA_ATTRIBUTE_ELEMENT_COUNT, ctx->shader_state.PA_ATTRIBUTE_ELEMENT_COUNT);
 397    }
 398    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_SHADER))) {
 399       uint32_t val = etna_rasterizer_state(ctx->rasterizer)->PA_CONFIG;
 400       /*00A34*/ EMIT_STATE(PA_CONFIG, val & ctx->shader_state.PA_CONFIG);
 401    }
 402    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 403       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 404       /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
 405       /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
 406    }
 407    if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
 408                          ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
 409       /* this is a bit of a mess: rasterizer.scissor determines whether to use
 410        * only the framebuffer scissor, or specific scissor state, and the
 411        * viewport clips too so the logic spans four CSOs */
 412       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 413
 414       uint32_t scissor_left =
 415          MAX2(ctx->framebuffer.SE_SCISSOR_LEFT, ctx->viewport.SE_SCISSOR_LEFT);
 416       uint32_t scissor_top =
 417          MAX2(ctx->framebuffer.SE_SCISSOR_TOP, ctx->viewport.SE_SCISSOR_TOP);
 418       uint32_t scissor_right =
 419          MIN2(ctx->framebuffer.SE_SCISSOR_RIGHT, ctx->viewport.SE_SCISSOR_RIGHT);
 420       uint32_t scissor_bottom =
 421          MIN2(ctx->framebuffer.SE_SCISSOR_BOTTOM, ctx->viewport.SE_SCISSOR_BOTTOM);
 422
 423       if (rasterizer->scissor) {
 424          scissor_left = MAX2(ctx->scissor.SE_SCISSOR_LEFT, scissor_left);
 425          scissor_top = MAX2(ctx->scissor.SE_SCISSOR_TOP, scissor_top);
 426          scissor_right = MIN2(ctx->scissor.SE_SCISSOR_RIGHT, scissor_right);
 427          scissor_bottom = MIN2(ctx->scissor.SE_SCISSOR_BOTTOM, scissor_bottom);
 428       }
 429
 430       /*00C00*/ EMIT_STATE_FIXP(SE_SCISSOR_LEFT, scissor_left);
 431       /*00C04*/ EMIT_STATE_FIXP(SE_SCISSOR_TOP, scissor_top);
 432       /*00C08*/ EMIT_STATE_FIXP(SE_SCISSOR_RIGHT, scissor_right);
 433       /*00C0C*/ EMIT_STATE_FIXP(SE_SCISSOR_BOTTOM, scissor_bottom);
 434    }
 435    if (unlikely(dirty & (ETNA_DIRTY_RASTERIZER))) {
 436       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 437
 438       /*00C10*/ EMIT_STATE(SE_DEPTH_SCALE, rasterizer->SE_DEPTH_SCALE);
 439       /*00C14*/ EMIT_STATE(SE_DEPTH_BIAS, rasterizer->SE_DEPTH_BIAS);
 440       /*00C18*/ EMIT_STATE(SE_CONFIG, rasterizer->SE_CONFIG);
 441    }
 442    if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
 443                          ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
 444       struct etna_rasterizer_state *rasterizer = etna_rasterizer_state(ctx->rasterizer);
 445
 446       uint32_t clip_right =
 447          MIN2(ctx->framebuffer.SE_CLIP_RIGHT, ctx->viewport.SE_CLIP_RIGHT);
 448       uint32_t clip_bottom =
 449          MIN2(ctx->framebuffer.SE_CLIP_BOTTOM, ctx->viewport.SE_CLIP_BOTTOM);
 450
 451       if (rasterizer->scissor) {
 452          clip_right = MIN2(ctx->scissor.SE_CLIP_RIGHT, clip_right);
 453          clip_bottom = MIN2(ctx->scissor.SE_CLIP_BOTTOM, clip_bottom);
 454       }
 455
 456       /*00C20*/ EMIT_STATE_FIXP(SE_CLIP_RIGHT, clip_right);
 457       /*00C24*/ EMIT_STATE_FIXP(SE_CLIP_BOTTOM, clip_bottom);
 458    }
 459    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 460       /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
 461    }
 462    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
 463       /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
 464       /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
 465                            ctx->framebuffer.msaa_mode
 466                               ? ctx->shader_state.PS_INPUT_COUNT_MSAA
 467                               : ctx->shader_state.PS_INPUT_COUNT);
 468       /*0100C*/ EMIT_STATE(PS_TEMP_REGISTER_CONTROL,
 469                            ctx->framebuffer.msaa_mode
 470                               ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
 471                               : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
 472       /*01010*/ EMIT_STATE(PS_CONTROL, ctx->shader_state.PS_CONTROL);
 473    }
 474    if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER))) {
 475       uint32_t val = etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG;
 476       /*01400*/ EMIT_STATE(PE_DEPTH_CONFIG, val | ctx->framebuffer.PE_DEPTH_CONFIG);
 477    }
 478    if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
 479       /*01404*/ EMIT_STATE(PE_DEPTH_NEAR, ctx->viewport.PE_DEPTH_NEAR);
 480       /*01408*/ EMIT_STATE(PE_DEPTH_FAR, ctx->viewport.PE_DEPTH_FAR);
 481    }
 482    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 483       /*0140C*/ EMIT_STATE(PE_DEPTH_NORMALIZE, ctx->framebuffer.PE_DEPTH_NORMALIZE);
 484
 485       if (ctx->specs.pixel_pipes == 1) {
 486          /*01410*/ EMIT_STATE_RELOC(PE_DEPTH_ADDR, &ctx->framebuffer.PE_DEPTH_ADDR);
 487       }
 488
 489       /*01414*/ EMIT_STATE(PE_DEPTH_STRIDE, ctx->framebuffer.PE_DEPTH_STRIDE);
 490    }
 491    if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
 492       uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_OP;
 493       /*01418*/ EMIT_STATE(PE_STENCIL_OP, val);
 494    }
 495    if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_STENCIL_REF))) {
 496       uint32_t val = etna_zsa_state(ctx->zsa)->PE_STENCIL_CONFIG;
 497       /*0141C*/ EMIT_STATE(PE_STENCIL_CONFIG, val | ctx->stencil_ref.PE_STENCIL_CONFIG);
 498    }
 499    if (unlikely(dirty & (ETNA_DIRTY_ZSA))) {
 500       uint32_t val = etna_zsa_state(ctx->zsa)->PE_ALPHA_OP;
 501       /*01420*/ EMIT_STATE(PE_ALPHA_OP, val);
 502    }
 503    if (unlikely(dirty & (ETNA_DIRTY_BLEND_COLOR))) {
 504       /*01424*/ EMIT_STATE(PE_ALPHA_BLEND_COLOR, ctx->blend_color.PE_ALPHA_BLEND_COLOR);
 505    }
 506    if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
 507       uint32_t val = etna_blend_state(ctx->blend)->PE_ALPHA_CONFIG;
 508       /*01428*/ EMIT_STATE(PE_ALPHA_CONFIG, val);
 509    }
 510    if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
 511       uint32_t val;
 512       /* Use the components and overwrite bits in framebuffer.PE_COLOR_FORMAT
 513        * as a mask to enable the bits from blend PE_COLOR_FORMAT */
 514       val = ~(VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
 515               VIVS_PE_COLOR_FORMAT_OVERWRITE);
 516       val |= etna_blend_state(ctx->blend)->PE_COLOR_FORMAT;
 517       val &= ctx->framebuffer.PE_COLOR_FORMAT;
 518       /*0142C*/ EMIT_STATE(PE_COLOR_FORMAT, val);
 519    }
 520    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
 521       if (ctx->specs.pixel_pipes == 1) {
 522          /*01430*/ EMIT_STATE_RELOC(PE_COLOR_ADDR, &ctx->framebuffer.PE_COLOR_ADDR);
 523          /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
 524          /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
 525       } else if (ctx->specs.pixel_pipes == 2) {
 526          /*01434*/ EMIT_STATE(PE_COLOR_STRIDE, ctx->framebuffer.PE_COLOR_STRIDE);
 527          /*01454*/ EMIT_STATE(PE_HDEPTH_CONTROL, ctx->framebuffer.PE_HDEPTH_CONTROL);
 528          /*01460*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(0), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[0]);
 529          /*01464*/ EMIT_STATE_RELOC(PE_PIPE_COLOR_ADDR(1), &ctx->framebuffer.PE_PIPE_COLOR_ADDR[1]);
 530          /*01480*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(0), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[0]);
 531          /*01484*/ EMIT_STATE_RELOC(PE_PIPE_DEPTH_ADDR(1), &ctx->framebuffer.PE_PIPE_DEPTH_ADDR[1]);
 532       } else {
 533          abort();
 534       }
 535    }
 536    if (unlikely(dirty & (ETNA_DIRTY_STENCIL_REF))) {
 537       /*014A0*/ EMIT_STATE(PE_STENCIL_CONFIG_EXT, ctx->stencil_ref.PE_STENCIL_CONFIG_EXT);
 538    }
 539    if (unlikely(dirty & (ETNA_DIRTY_BLEND | ETNA_DIRTY_FRAMEBUFFER))) {
 540       struct etna_blend_state *blend = etna_blend_state(ctx->blend);
 541       /*014A4*/ EMIT_STATE(PE_LOGIC_OP, blend->PE_LOGIC_OP | ctx->framebuffer.PE_LOGIC_OP);
 542    }
 543    if (unlikely(dirty & (ETNA_DIRTY_BLEND))) {
 544       struct etna_blend_state *blend = etna_blend_state(ctx->blend);
 545       for (int x = 0; x < 2; ++x) {
 546          /*014A8*/ EMIT_STATE(PE_DITHER(x), blend->PE_DITHER[x]);
 547       }
 548    }
 549    if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER | ETNA_DIRTY_TS))) {
 550       /*01654*/ EMIT_STATE(TS_MEM_CONFIG, ctx->framebuffer.TS_MEM_CONFIG);
 551       /*01658*/ EMIT_STATE_RELOC(TS_COLOR_STATUS_BASE, &ctx->framebuffer.TS_COLOR_STATUS_BASE);
 552       /*0165C*/ EMIT_STATE_RELOC(TS_COLOR_SURFACE_BASE, &ctx->framebuffer.TS_COLOR_SURFACE_BASE);
 553       /*01660*/ EMIT_STATE(TS_COLOR_CLEAR_VALUE, ctx->framebuffer.TS_COLOR_CLEAR_VALUE);
 554       /*01664*/ EMIT_STATE_RELOC(TS_DEPTH_STATUS_BASE, &ctx->framebuffer.TS_DEPTH_STATUS_BASE);
 555       /*01668*/ EMIT_STATE_RELOC(TS_DEPTH_SURFACE_BASE, &ctx->framebuffer.TS_DEPTH_SURFACE_BASE);
 556       /*0166C*/ EMIT_STATE(TS_DEPTH_CLEAR_VALUE, ctx->framebuffer.TS_DEPTH_CLEAR_VALUE);
 557    }
 558    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
 559       /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
 560    }
 561    etna_coalesce_end(stream, &coalesce);
 562    /* end only EMIT_STATE */
 563
 564    /* Emit strongly architecture-specific state */
 565    if (ctx->specs.halti >= 5)
 566       emit_halti5_only_state(ctx, vs_output_count);
 567    else
 568       emit_pre_halti5_state(ctx);
 569
 570    ctx->emit_texture_state(ctx);
 571
 572    /* Insert a FE/PE stall as changing the shader instructions (and maybe
 573     * the uniforms) can corrupt the previous in-progress draw operation.
 574     * Observed with amoeba on GC2000 during the right-to-left rendering
 575     * of PI, and can cause GPU hangs immediately after.
 576     * I summise that this is because the "new" locations at 0xc000 are not
 577     * properly protected against updates as other states seem to be. Hence,
 578     * we detect the "new" vertex shader instruction offset to apply this. */
 579    if (ctx->dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF) && ctx->specs.vs_offset > 0x4000)
 580       etna_stall(ctx->stream, SYNC_RECIPIENT_FE, SYNC_RECIPIENT_PE);
 581
 582    /* We need to update the uniform cache only if one of the following bits are
 583     * set in ctx->dirty:
 584     * - ETNA_DIRTY_SHADER
 585     * - ETNA_DIRTY_CONSTBUF
 586     * - uniforms_dirty_bits
 587     *
 588     * In case of ETNA_DIRTY_SHADER we need load all uniforms from the cache. In
 589     * all
 590     * other cases we can load on the changed uniforms.
 591     */
 592    static const uint32_t uniform_dirty_bits =
 593       ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
 594
 595    if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
 596       etna_uniforms_write(
 597          ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX],
 598          ctx->shader_state.VS_UNIFORMS, &ctx->shader_state.vs_uniforms_size);
 599
 600    if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
 601       etna_uniforms_write(
 602          ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT],
 603          ctx->shader_state.PS_UNIFORMS, &ctx->shader_state.ps_uniforms_size);
 604
 605    /**** Large dynamically-sized state ****/
 606    bool do_uniform_flush = ctx->specs.halti < 5;
 607    if (dirty & (ETNA_DIRTY_SHADER)) {
 608       /* Special case: a new shader was loaded; simply re-load all uniforms and
 609        * shader code at once */
 610       /* This sequence is special, do not change ordering unless necessary. According to comment
 611          snippets in the Vivante kernel driver a process called "steering" goes on while programming
 612          shader state. This (as I understand it) means certain unified states are "steered"
 613          toward a specific shader unit (VS/PS/...) based on either explicit flags in register
 614          00860, or what other state is written before "auto-steering". So this means some
 615          state can legitimately be programmed multiple times.
 616        */
 617
 618       if (ctx->specs.halti >= 5) { /* ICACHE (HALTI5) */
 619          assert(ctx->shader_state.VS_INST_ADDR.bo && ctx->shader_state.PS_INST_ADDR.bo);
 620          /* Set icache (VS) */
 621          etna_set_state(stream, VIVS_VS_NEWRANGE_LOW, 0);
 622          etna_set_state(stream, VIVS_VS_NEWRANGE_HIGH, ctx->shader_state.vs_inst_mem_size / 4);
 623          assert(ctx->shader_state.VS_INST_ADDR.bo);
 624          etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
 625          etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
 626          etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
 627          etna_set_state(stream, VIVS_VS_ICACHE_COUNT, ctx->shader_state.vs_inst_mem_size / 4 - 1);
 628
 629          /* Set icache (PS) */
 630          etna_set_state(stream, VIVS_PS_NEWRANGE_LOW, 0);
 631          etna_set_state(stream, VIVS_PS_NEWRANGE_HIGH, ctx->shader_state.ps_inst_mem_size / 4);
 632          assert(ctx->shader_state.PS_INST_ADDR.bo);
 633          etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
 634          etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
 635          etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
 636          etna_set_state(stream, VIVS_PS_ICACHE_COUNT, ctx->shader_state.ps_inst_mem_size / 4 - 1);
 637
 638       } else if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
 639          /* ICACHE (pre-HALTI5) */
 640          assert(ctx->specs.has_icache && ctx->specs.has_shader_range_registers);
 641          /* Set icache (VS) */
 642          etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
 643          etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
 644                VIVS_VS_ICACHE_CONTROL_ENABLE |
 645                VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
 646          assert(ctx->shader_state.VS_INST_ADDR.bo);
 647          etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
 648
 649          /* Set icache (PS) */
 650          etna_set_state(stream, VIVS_PS_RANGE, (ctx->shader_state.ps_inst_mem_size / 4 - 1) << 16);
 651          etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
 652                VIVS_VS_ICACHE_CONTROL_ENABLE |
 653                VIVS_VS_ICACHE_CONTROL_FLUSH_PS);
 654          assert(ctx->shader_state.PS_INST_ADDR.bo);
 655          etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
 656       } else {
 657          /* Upload shader directly, first flushing and disabling icache if
 658           * supported on this hw */
 659          if (ctx->specs.has_icache) {
 660             etna_set_state(stream, VIVS_VS_ICACHE_CONTROL,
 661                   VIVS_VS_ICACHE_CONTROL_FLUSH_PS |
 662                   VIVS_VS_ICACHE_CONTROL_FLUSH_VS);
 663          }
 664          if (ctx->specs.has_shader_range_registers) {
 665             etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
 666             etna_set_state(stream, VIVS_PS_RANGE, ((ctx->shader_state.ps_inst_mem_size / 4 - 1 + 0x100) << 16) |
 667                                         0x100);
 668          }
 669          etna_set_state_multi(stream, ctx->specs.vs_offset,
 670                               ctx->shader_state.vs_inst_mem_size,
 671                               ctx->shader_state.VS_INST_MEM);
 672          etna_set_state_multi(stream, ctx->specs.ps_offset,
 673                               ctx->shader_state.ps_inst_mem_size,
 674                               ctx->shader_state.PS_INST_MEM);
 675       }
 676
 677       if (ctx->specs.has_unified_uniforms) {
 678          etna_set_state(stream, VIVS_VS_UNIFORM_BASE, 0);
 679          etna_set_state(stream, VIVS_PS_UNIFORM_BASE, ctx->specs.max_vs_uniforms);
 680       }
 681
 682       if (do_uniform_flush)
 683          etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
 684       etna_set_state_multi(stream, ctx->specs.vs_uniforms_offset,
 685                                      ctx->shader_state.vs_uniforms_size,
 686                                      ctx->shader_state.VS_UNIFORMS);
 687       if (do_uniform_flush)
 688          etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
 689       etna_set_state_multi(stream, ctx->specs.ps_uniforms_offset,
 690                                      ctx->shader_state.ps_uniforms_size,
 691                                      ctx->shader_state.PS_UNIFORMS);
 692
 693       /* Copy uniforms to gpu3d, so that incremental updates to uniforms are
 694        * possible as long as the
 695        * same shader remains bound */
 696       ctx->gpu3d.vs_uniforms_size = ctx->shader_state.vs_uniforms_size;
 697       ctx->gpu3d.ps_uniforms_size = ctx->shader_state.ps_uniforms_size;
 698       memcpy(ctx->gpu3d.VS_UNIFORMS, ctx->shader_state.VS_UNIFORMS,
 699              ctx->shader_state.vs_uniforms_size * 4);
 700       memcpy(ctx->gpu3d.PS_UNIFORMS, ctx->shader_state.PS_UNIFORMS,
 701              ctx->shader_state.ps_uniforms_size * 4);
 702
 703       if (ctx->specs.halti >= 5) {
 704          /* HALTI5 needs to be prompted to pre-fetch shaders */
 705          etna_set_state(stream, VIVS_VS_ICACHE_PREFETCH, 0x00000000);
 706          etna_set_state(stream, VIVS_PS_ICACHE_PREFETCH, 0x00000000);
 707          etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
 708       }
 709    } else {
 710       /* ideally this cache would only be flushed if there are VS uniform changes */
 711       if (do_uniform_flush)
 712          etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
 713       etna_coalesce_start(stream, &coalesce);
 714       for (int x = 0; x < ctx->shader.vs->uniforms.const_count; ++x) {
 715          if (ctx->gpu3d.VS_UNIFORMS[x] != ctx->shader_state.VS_UNIFORMS[x]) {
 716             etna_coalsence_emit(stream, &coalesce, ctx->specs.vs_uniforms_offset + x*4, ctx->shader_state.VS_UNIFORMS[x]);
 717             ctx->gpu3d.VS_UNIFORMS[x] = ctx->shader_state.VS_UNIFORMS[x];
 718          }
 719       }
 720       etna_coalesce_end(stream, &coalesce);
 721
 722       /* ideally this cache would only be flushed if there are PS uniform changes */
 723       if (do_uniform_flush)
 724          etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
 725       etna_coalesce_start(stream, &coalesce);
 726       for (int x = 0; x < ctx->shader.fs->uniforms.const_count; ++x) {
 727          if (ctx->gpu3d.PS_UNIFORMS[x] != ctx->shader_state.PS_UNIFORMS[x]) {
 728             etna_coalsence_emit(stream, &coalesce, ctx->specs.ps_uniforms_offset + x*4, ctx->shader_state.PS_UNIFORMS[x]);
 729             ctx->gpu3d.PS_UNIFORMS[x] = ctx->shader_state.PS_UNIFORMS[x];
 730          }
 731       }
 732       etna_coalesce_end(stream, &coalesce);
 733    }
 734 /**** End of state update ****/
 735 #undef EMIT_STATE
 736 #undef EMIT_STATE_FIXP
 737 #undef EMIT_STATE_RELOC
 738    ctx->dirty = 0;
 739    ctx->dirty_sampler_views = 0;
 740 }