etnaviv: GC7000: State changes for HALTI3..5

Update state objects to add new state, and emit function to emit new
state.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

diff --git a/src/gallium/drivers/etnaviv/etnaviv_emit.c b/src/gallium/drivers/etnaviv/etnaviv_emit.c
index 508c7b1b6aba00c7d2081e8eafb562887aa081d3..f63ec7748648b357dea9a2290f484827660fa6a2 100644 (file)
--- a/src/gallium/drivers/etnaviv/etnaviv_emit.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_emit.c
@@ -133,6 +133,91 @@ required_stream_size(struct etna_context *ctx)
    return size;
 }
 
+/* Emit state that only exists on HALTI5+ */
+static void
+emit_halti5_only_state(struct etna_context *ctx, int vs_output_count)
+{
+   struct etna_cmd_stream *stream = ctx->stream;
+   uint32_t dirty = ctx->dirty;
+   struct etna_coalesce coalesce;
+
+   etna_coalesce_start(stream, &coalesce);
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      /* Magic states (load balancing, inter-unit sync, buffers) */
+      /*00870*/ EMIT_STATE(VS_HALTI5_OUTPUT_COUNT, vs_output_count | ((vs_output_count * 0x10) << 8));
+      /*008A0*/ EMIT_STATE(VS_HALTI5_UNK008A0, 0x0001000e | ((0x110/vs_output_count) << 20));
+      for (int x = 0; x < 4; ++x) {
+         /*008E0*/ EMIT_STATE(VS_HALTI5_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
+      for (int x = 0; x < 4; ++x) {
+         /*008C0*/ EMIT_STATE(VS_HALTI5_INPUT(x), ctx->shader_state.VS_INPUT[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      /*00A90*/ EMIT_STATE(PA_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
+      /*00AA8*/ EMIT_STATE(PA_VS_OUTPUT_COUNT, vs_output_count);
+      /*01080*/ EMIT_STATE(PS_VARYING_NUM_COMPONENTS(0), ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
+      /*03888*/ EMIT_STATE(GL_HALTI5_SH_SPECIALS, ctx->shader_state.GL_HALTI5_SH_SPECIALS);
+   }
+   etna_coalesce_end(stream, &coalesce);
+}
+
+/* Emit state that no longer exists on HALTI5 */
+static void
+emit_pre_halti5_state(struct etna_context *ctx)
+{
+   struct etna_cmd_stream *stream = ctx->stream;
+   uint32_t dirty = ctx->dirty;
+   struct etna_coalesce coalesce;
+
+   etna_coalesce_start(stream, &coalesce);
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      for (int x = 0; x < 4; ++x) {
+        /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
+      for (int x = 0; x < 4; ++x) {
+        /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      for (int x = 0; x < 10; ++x) {
+         /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
+      /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
+      for (int x = 0; x < 4; ++x) {
+         /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
+      }
+      for (int x = 0; x < 16; ++x) {
+         /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
+      }
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
+      /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
+      /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
+   }
+   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
+      /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
+      for (int x = 0; x < 2; ++x) {
+         /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
+      }
+   }
+   etna_coalesce_end(stream, &coalesce);
+}
+
 /* Weave state before draw operation. This function merges all the compiled
  * state blocks under the context into one device register state. Parts of
  * this state that are changed since last call (dirty) will be uploaded as
@@ -197,16 +282,31 @@ etna_emit_state(struct etna_context *ctx)
     * a) the number of vertex elements written matters: so write only active ones
     * b) the vertex element states must all be written: do not skip entries that stay the same */
    if (dirty & (ETNA_DIRTY_VERTEX_ELEMENTS)) {
-      /* Special case: vertex elements must always be sent in full if changed */
-      /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
-         ctx->vertex_elements->num_elements,
-         ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
-      if (ctx->specs.halti >= 2) {
-         /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
+      if (ctx->specs.halti >= 5) {
+         /*17800*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG0(0),
+            ctx->vertex_elements->num_elements,
+            ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG0);
+         /*17A00*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_SCALE(0),
             ctx->vertex_elements->num_elements,
             ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
+         /*17A80*/ etna_set_state_multi(stream, VIVS_NFE_GENERIC_ATTRIB_CONFIG1(0),
+            ctx->vertex_elements->num_elements,
+            ctx->vertex_elements->NFE_GENERIC_ATTRIB_CONFIG1);
+      } else {
+         /* Special case: vertex elements must always be sent in full if changed */
+         /*00600*/ etna_set_state_multi(stream, VIVS_FE_VERTEX_ELEMENT_CONFIG(0),
+            ctx->vertex_elements->num_elements,
+            ctx->vertex_elements->FE_VERTEX_ELEMENT_CONFIG);
+         if (ctx->specs.halti >= 2) {
+            /*00780*/ etna_set_state_multi(stream, VIVS_FE_GENERIC_ATTRIB_SCALE(0),
+               ctx->vertex_elements->num_elements,
+               ctx->vertex_elements->NFE_GENERIC_ATTRIB_SCALE);
+         }
       }
    }
+   unsigned vs_output_count = etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex
+                           ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
+                           : ctx->shader_state.VS_OUTPUT_COUNT;
 
    /* The following code is originally generated by gen_merge_state.py, to
     * emit state in increasing order of address (this makes it possible to merge
@@ -247,52 +347,48 @@ etna_emit_state(struct etna_context *ctx)
       /*00644*/ EMIT_STATE_RELOC(FE_INDEX_STREAM_BASE_ADDR, &ctx->index_buffer.FE_INDEX_STREAM_BASE_ADDR);
       /*00648*/ EMIT_STATE(FE_INDEX_STREAM_CONTROL, ctx->index_buffer.FE_INDEX_STREAM_CONTROL);
    }
-   if (likely((dirty & (ETNA_DIRTY_VERTEX_BUFFERS) && ctx->specs.stream_count == 1))) {
-      /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
-      /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
-   }
    if (likely(dirty & (ETNA_DIRTY_INDEX_BUFFER))) {
       /*00674*/ EMIT_STATE(FE_PRIMITIVE_RESTART_INDEX, ctx->index_buffer.FE_PRIMITIVE_RESTART_INDEX);
    }
-   if (likely((dirty & (ETNA_DIRTY_VERTEX_BUFFERS)) && ctx->specs.stream_count > 1)) {
-      for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
-         /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
-      }
-      for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
-         if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
-            /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
+   if (likely(dirty & (ETNA_DIRTY_VERTEX_BUFFERS))) {
+      if (ctx->specs.halti >= 2) { /* HALTI2+: NFE_VERTEX_STREAMS */
+         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
+            /*14600*/ EMIT_STATE_RELOC(NFE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
+         }
+         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
+            if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
+               /*14640*/ EMIT_STATE(NFE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
+            }
+         }
+         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
+            if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
+               /*14680*/ EMIT_STATE(NFE_VERTEX_STREAMS_UNK14680(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_UNK14680);
+            }
+         }
+      } else if(ctx->specs.stream_count >= 1) { /* hw w/ multiple vertex streams */
+         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
+            /*00680*/ EMIT_STATE_RELOC(FE_VERTEX_STREAMS_BASE_ADDR(x), &ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR);
          }
+         for (int x = 0; x < ctx->vertex_buffer.count; ++x) {
+            if (ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_BASE_ADDR.bo) {
+               /*006A0*/ EMIT_STATE(FE_VERTEX_STREAMS_CONTROL(x), ctx->vertex_buffer.cvb[x].FE_VERTEX_STREAM_CONTROL);
+            }
+         }
+      } else { /* hw w/ single vertex stream */
+         /*0064C*/ EMIT_STATE_RELOC(FE_VERTEX_STREAM_BASE_ADDR, &ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_BASE_ADDR);
+         /*00650*/ EMIT_STATE(FE_VERTEX_STREAM_CONTROL, ctx->vertex_buffer.cvb[0].FE_VERTEX_STREAM_CONTROL);
       }
    }
-   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
-      /*00800*/ EMIT_STATE(VS_END_PC, ctx->shader_state.VS_END_PC);
-   }
    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_RASTERIZER))) {
-      bool point_size_per_vertex =
-         etna_rasterizer_state(ctx->rasterizer)->point_size_per_vertex;
 
-      /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT,
-                           point_size_per_vertex
-                              ? ctx->shader_state.VS_OUTPUT_COUNT_PSIZE
-                              : ctx->shader_state.VS_OUTPUT_COUNT);
+      /*00804*/ EMIT_STATE(VS_OUTPUT_COUNT, vs_output_count);
    }
    if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
       /*00808*/ EMIT_STATE(VS_INPUT_COUNT, ctx->shader_state.VS_INPUT_COUNT);
       /*0080C*/ EMIT_STATE(VS_TEMP_REGISTER_CONTROL, ctx->shader_state.VS_TEMP_REGISTER_CONTROL);
    }
-   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
-      for (int x = 0; x < 4; ++x) {
-         /*00810*/ EMIT_STATE(VS_OUTPUT(x), ctx->shader_state.VS_OUTPUT[x]);
-      }
-   }
-   if (unlikely(dirty & (ETNA_DIRTY_VERTEX_ELEMENTS | ETNA_DIRTY_SHADER))) {
-      for (int x = 0; x < 4; ++x) {
-         /*00820*/ EMIT_STATE(VS_INPUT(x), ctx->shader_state.VS_INPUT[x]);
-      }
-   }
    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
       /*00830*/ EMIT_STATE(VS_LOAD_BALANCING, ctx->shader_state.VS_LOAD_BALANCING);
-      /*00838*/ EMIT_STATE(VS_START_PC, ctx->shader_state.VS_START_PC);
    }
    if (unlikely(dirty & (ETNA_DIRTY_VIEWPORT))) {
       /*00A00*/ EMIT_STATE_FIXP(PA_VIEWPORT_SCALE_X, ctx->viewport.PA_VIEWPORT_SCALE_X);
@@ -321,11 +417,6 @@ etna_emit_state(struct etna_context *ctx)
       /*00A38*/ EMIT_STATE(PA_WIDE_LINE_WIDTH0, rasterizer->PA_LINE_WIDTH);
       /*00A3C*/ EMIT_STATE(PA_WIDE_LINE_WIDTH1, rasterizer->PA_LINE_WIDTH);
    }
-   if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
-      for (int x = 0; x < 10; ++x) {
-         /*00A40*/ EMIT_STATE(PA_SHADER_ATTRIBUTES(x), ctx->shader_state.PA_SHADER_ATTRIBUTES[x]);
-      }
-   }
    if (unlikely(dirty & (ETNA_DIRTY_SCISSOR | ETNA_DIRTY_FRAMEBUFFER |
                          ETNA_DIRTY_RASTERIZER | ETNA_DIRTY_VIEWPORT))) {
       /* this is a bit of a mess: rasterizer.scissor determines whether to use
@@ -381,17 +472,7 @@ etna_emit_state(struct etna_context *ctx)
    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
       /*00E00*/ EMIT_STATE(RA_CONTROL, ctx->shader_state.RA_CONTROL);
    }
-   if (unlikely(dirty & (ETNA_DIRTY_FRAMEBUFFER))) {
-      /*00E04*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E04, ctx->framebuffer.RA_MULTISAMPLE_UNK00E04);
-      for (int x = 0; x < 4; ++x) {
-         /*00E10*/ EMIT_STATE(RA_MULTISAMPLE_UNK00E10(x), ctx->framebuffer.RA_MULTISAMPLE_UNK00E10[x]);
-      }
-      for (int x = 0; x < 16; ++x) {
-         /*00E40*/ EMIT_STATE(RA_CENTROID_TABLE(x), ctx->framebuffer.RA_CENTROID_TABLE[x]);
-      }
-   }
    if (unlikely(dirty & (ETNA_DIRTY_SHADER | ETNA_DIRTY_FRAMEBUFFER))) {
-      /*01000*/ EMIT_STATE(PS_END_PC, ctx->shader_state.PS_END_PC);
       /*01004*/ EMIT_STATE(PS_OUTPUT_REG, ctx->shader_state.PS_OUTPUT_REG);
       /*01008*/ EMIT_STATE(PS_INPUT_COUNT,
                            ctx->framebuffer.msaa_mode
@@ -402,7 +483,6 @@ etna_emit_state(struct etna_context *ctx)
                               ? ctx->shader_state.PS_TEMP_REGISTER_CONTROL_MSAA
                               : ctx->shader_state.PS_TEMP_REGISTER_CONTROL);
       /*01010*/ EMIT_STATE(PS_CONTROL, ctx->shader_state.PS_CONTROL);
-      /*01018*/ EMIT_STATE(PS_START_PC, ctx->shader_state.PS_START_PC);
    }
    if (unlikely(dirty & (ETNA_DIRTY_ZSA | ETNA_DIRTY_FRAMEBUFFER))) {
       uint32_t val = etna_zsa_state(ctx->zsa)->PE_DEPTH_CONFIG;
@@ -583,12 +663,9 @@ etna_emit_state(struct etna_context *ctx)
          }
       }
    }
+
    if (unlikely(dirty & (ETNA_DIRTY_SHADER))) {
       /*0381C*/ EMIT_STATE(GL_VARYING_TOTAL_COMPONENTS, ctx->shader_state.GL_VARYING_TOTAL_COMPONENTS);
-      /*03820*/ EMIT_STATE(GL_VARYING_NUM_COMPONENTS, ctx->shader_state.GL_VARYING_NUM_COMPONENTS);
-      for (int x = 0; x < 2; ++x) {
-         /*03828*/ EMIT_STATE(GL_VARYING_COMPONENT_USE(x), ctx->shader_state.GL_VARYING_COMPONENT_USE[x]);
-      }
    }
    if (unlikely(ctx->specs.tex_astc && (dirty & (ETNA_DIRTY_SAMPLER_VIEWS)))) {
       for (int x = 0; x < VIVS_TE_SAMPLER__LEN; ++x) {
@@ -601,6 +678,12 @@ etna_emit_state(struct etna_context *ctx)
    etna_coalesce_end(stream, &coalesce);
    /* end only EMIT_STATE */
 
+   /* Emit strongly architecture-specific state */
+   if (ctx->specs.halti >= 5)
+      emit_halti5_only_state(ctx, vs_output_count);
+   else
+      emit_pre_halti5_state(ctx);
+
    /* Insert a FE/PE stall as changing the shader instructions (and maybe
     * the uniforms) can corrupt the previous in-progress draw operation.
     * Observed with amoeba on GC2000 during the right-to-left rendering
@@ -635,10 +718,40 @@ etna_emit_state(struct etna_context *ctx)
          ctx->shader_state.PS_UNIFORMS, &ctx->shader_state.ps_uniforms_size);
 
    /**** Large dynamically-sized state ****/
+   bool do_uniform_flush = ctx->specs.halti < 5;
    if (dirty & (ETNA_DIRTY_SHADER)) {
       /* Special case: a new shader was loaded; simply re-load all uniforms and
        * shader code at once */
-      if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
+      /* This sequence is special, do not change ordering unless necessary. According to comment
+         snippets in the Vivante kernel driver a process called "steering" goes on while programming
+         shader state. This (as I understand it) means certain unified states are "steered"
+         toward a specific shader unit (VS/PS/...) based on either explicit flags in register
+         00860, or what other state is written before "auto-steering". So this means some
+         state can legitimately be programmed multiple times.
+       */
+
+      if (ctx->specs.halti >= 5) { /* ICACHE (HALTI5) */
+         assert(ctx->shader_state.VS_INST_ADDR.bo && ctx->shader_state.PS_INST_ADDR.bo);
+         /* Set icache (VS) */
+         etna_set_state(stream, VIVS_VS_NEWRANGE_LOW, 0);
+         etna_set_state(stream, VIVS_VS_NEWRANGE_HIGH, ctx->shader_state.vs_inst_mem_size / 4);
+         assert(ctx->shader_state.VS_INST_ADDR.bo);
+         etna_set_state_reloc(stream, VIVS_VS_INST_ADDR, &ctx->shader_state.VS_INST_ADDR);
+         etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
+         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
+         etna_set_state(stream, VIVS_VS_ICACHE_COUNT, ctx->shader_state.vs_inst_mem_size / 4 - 1);
+
+         /* Set icache (PS) */
+         etna_set_state(stream, VIVS_PS_NEWRANGE_LOW, 0);
+         etna_set_state(stream, VIVS_PS_NEWRANGE_HIGH, ctx->shader_state.ps_inst_mem_size / 4);
+         assert(ctx->shader_state.PS_INST_ADDR.bo);
+         etna_set_state_reloc(stream, VIVS_PS_INST_ADDR, &ctx->shader_state.PS_INST_ADDR);
+         etna_set_state(stream, VIVS_SH_CONFIG, 0x00000002);
+         etna_set_state(stream, VIVS_VS_ICACHE_CONTROL, VIVS_VS_ICACHE_CONTROL_ENABLE);
+         etna_set_state(stream, VIVS_PS_ICACHE_COUNT, ctx->shader_state.ps_inst_mem_size / 4 - 1);
+
+      } else if (ctx->shader_state.VS_INST_ADDR.bo || ctx->shader_state.PS_INST_ADDR.bo) {
+         /* ICACHE (pre-HALTI5) */
          assert(ctx->specs.has_icache && ctx->specs.has_shader_range_registers);
          /* Set icache (VS) */
          etna_set_state(stream, VIVS_VS_RANGE, (ctx->shader_state.vs_inst_mem_size / 4 - 1) << 16);
@@ -680,11 +793,14 @@ etna_emit_state(struct etna_context *ctx)
          etna_set_state(stream, VIVS_VS_UNIFORM_BASE, 0);
          etna_set_state(stream, VIVS_PS_UNIFORM_BASE, ctx->specs.max_vs_uniforms);
       }
-      etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
+
+      if (do_uniform_flush)
+         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
       etna_set_state_multi(stream, ctx->specs.vs_uniforms_offset,
                                      ctx->shader_state.vs_uniforms_size,
                                      ctx->shader_state.VS_UNIFORMS);
-      etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
+      if (do_uniform_flush)
+         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
       etna_set_state_multi(stream, ctx->specs.ps_uniforms_offset,
                                      ctx->shader_state.ps_uniforms_size,
                                      ctx->shader_state.PS_UNIFORMS);
@@ -698,9 +814,17 @@ etna_emit_state(struct etna_context *ctx)
              ctx->shader_state.vs_uniforms_size * 4);
       memcpy(ctx->gpu3d.PS_UNIFORMS, ctx->shader_state.PS_UNIFORMS,
              ctx->shader_state.ps_uniforms_size * 4);
+
+      if (ctx->specs.halti >= 5) {
+         /* HALTI5 needs to be prompted to pre-fetch shaders */
+         etna_set_state(stream, VIVS_VS_ICACHE_PREFETCH, 0x00000000);
+         etna_set_state(stream, VIVS_PS_ICACHE_PREFETCH, 0x00000000);
+         etna_stall(stream, SYNC_RECIPIENT_RA, SYNC_RECIPIENT_PE);
+      }
    } else {
       /* ideally this cache would only be flushed if there are VS uniform changes */
-      etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
+      if (do_uniform_flush)
+         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
       etna_coalesce_start(stream, &coalesce);
       for (int x = 0; x < ctx->shader.vs->uniforms.const_count; ++x) {
          if (ctx->gpu3d.VS_UNIFORMS[x] != ctx->shader_state.VS_UNIFORMS[x]) {
@@ -711,7 +835,8 @@ etna_emit_state(struct etna_context *ctx)
       etna_coalesce_end(stream, &coalesce);
 
       /* ideally this cache would only be flushed if there are PS uniform changes */
-      etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
+      if (do_uniform_flush)
+         etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
       etna_coalesce_start(stream, &coalesce);
       for (int x = 0; x < ctx->shader.fs->uniforms.const_count; ++x) {
          if (ctx->gpu3d.PS_UNIFORMS[x] != ctx->shader_state.PS_UNIFORMS[x]) {

diff --git a/src/gallium/drivers/etnaviv/etnaviv_internal.h b/src/gallium/drivers/etnaviv/etnaviv_internal.h
index e65332b6c66f7e0b14ba828243ceed3bb3c13717..b8d21864331138db14438d7704ad2b92d649050a 100644 (file)
--- a/src/gallium/drivers/etnaviv/etnaviv_internal.h
+++ b/src/gallium/drivers/etnaviv/etnaviv_internal.h
@@ -218,12 +218,15 @@ struct compiled_framebuffer_state {
 struct compiled_vertex_elements_state {
    unsigned num_elements;
    uint32_t FE_VERTEX_ELEMENT_CONFIG[VIVS_FE_VERTEX_ELEMENT_CONFIG__LEN];
+   uint32_t NFE_GENERIC_ATTRIB_CONFIG0[VIVS_NFE_GENERIC_ATTRIB__LEN];
    uint32_t NFE_GENERIC_ATTRIB_SCALE[VIVS_NFE_GENERIC_ATTRIB__LEN];
+   uint32_t NFE_GENERIC_ATTRIB_CONFIG1[VIVS_NFE_GENERIC_ATTRIB__LEN];
 };
 
 /* Compiled context->set_vertex_buffer result */
 struct compiled_set_vertex_buffer {
    uint32_t FE_VERTEX_STREAM_CONTROL;
+   uint32_t FE_VERTEX_STREAM_UNK14680;
    struct etna_reloc FE_VERTEX_STREAM_BASE_ADDR;
 };
 
@@ -253,6 +256,7 @@ struct compiled_shader_state {
    uint32_t GL_VARYING_TOTAL_COMPONENTS;
    uint32_t GL_VARYING_NUM_COMPONENTS;
    uint32_t GL_VARYING_COMPONENT_USE[2];
+   uint32_t GL_HALTI5_SH_SPECIALS;
    unsigned vs_inst_mem_size;
    unsigned vs_uniforms_size;
    unsigned ps_inst_mem_size;

diff --git a/src/gallium/drivers/etnaviv/etnaviv_state.c b/src/gallium/drivers/etnaviv/etnaviv_state.c
index 1e0b7548a48b990139891fd5dae3ebdd8b40afdb..b859f40c29fe46b5b645fb93e7ea9644b8ff59ac 100644 (file)
--- a/src/gallium/drivers/etnaviv/etnaviv_state.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_state.c
@@ -135,7 +135,8 @@ etna_set_framebuffer_state(struct pipe_context *pctx,
          VIVS_PE_COLOR_FORMAT_FORMAT(translate_rs_format(cbuf->base.format)) |
          VIVS_PE_COLOR_FORMAT_COMPONENTS__MASK |
          VIVS_PE_COLOR_FORMAT_OVERWRITE |
-         COND(color_supertiled, VIVS_PE_COLOR_FORMAT_SUPER_TILED);
+         COND(color_supertiled, VIVS_PE_COLOR_FORMAT_SUPER_TILED) |
+         COND(color_supertiled && ctx->specs.halti >= 5, VIVS_PE_COLOR_FORMAT_SUPER_TILED_NEW);
       /* VIVS_PE_COLOR_FORMAT_COMPONENTS() and
        * VIVS_PE_COLOR_FORMAT_OVERWRITE comes from blend_state
        * but only if we set the bits above. */
@@ -211,7 +212,9 @@ etna_set_framebuffer_state(struct pipe_context *pctx,
       cs->PE_DEPTH_CONFIG =
          depth_format |
          COND(depth_supertiled, VIVS_PE_DEPTH_CONFIG_SUPER_TILED) |
-         VIVS_PE_DEPTH_CONFIG_DEPTH_MODE_Z;
+         VIVS_PE_DEPTH_CONFIG_DEPTH_MODE_Z |
+         COND(ctx->specs.halti >= 5, VIVS_PE_DEPTH_CONFIG_DISABLE_ZS) /* Needs to be enabled on GC7000, otherwise depth writes hang w/ TS - apparently it does something else now */
+         ;
       /* VIVS_PE_DEPTH_CONFIG_ONLY_DEPTH */
       /* merged with depth_stencil_alpha */
 
@@ -542,14 +545,26 @@ etna_vertex_elements_state_create(struct pipe_context *pctx,
       assert(format_type != ETNA_NO_MATCH);
       assert(normalize != ETNA_NO_MATCH);
 
-      cs->FE_VERTEX_ELEMENT_CONFIG[idx] =
-         COND(nonconsecutive, VIVS_FE_VERTEX_ELEMENT_CONFIG_NONCONSECUTIVE) |
-         format_type |
-         VIVS_FE_VERTEX_ELEMENT_CONFIG_NUM(util_format_get_nr_components(elements[idx].src_format)) |
-         normalize | VIVS_FE_VERTEX_ELEMENT_CONFIG_ENDIAN(ENDIAN_MODE_NO_SWAP) |
-         VIVS_FE_VERTEX_ELEMENT_CONFIG_STREAM(elements[idx].vertex_buffer_index) |
-         VIVS_FE_VERTEX_ELEMENT_CONFIG_START(elements[idx].src_offset) |
-         VIVS_FE_VERTEX_ELEMENT_CONFIG_END(end_offset - start_offset);
+      if (ctx->specs.halti < 5) {
+         cs->FE_VERTEX_ELEMENT_CONFIG[idx] =
+            COND(nonconsecutive, VIVS_FE_VERTEX_ELEMENT_CONFIG_NONCONSECUTIVE) |
+            format_type |
+            VIVS_FE_VERTEX_ELEMENT_CONFIG_NUM(util_format_get_nr_components(elements[idx].src_format)) |
+            normalize | VIVS_FE_VERTEX_ELEMENT_CONFIG_ENDIAN(ENDIAN_MODE_NO_SWAP) |
+            VIVS_FE_VERTEX_ELEMENT_CONFIG_STREAM(elements[idx].vertex_buffer_index) |
+            VIVS_FE_VERTEX_ELEMENT_CONFIG_START(elements[idx].src_offset) |
+            VIVS_FE_VERTEX_ELEMENT_CONFIG_END(end_offset - start_offset);
+      } else { /* HALTI5 spread vertex attrib config over two registers */
+         cs->NFE_GENERIC_ATTRIB_CONFIG0[idx] =
+            format_type |
+            VIVS_NFE_GENERIC_ATTRIB_CONFIG0_NUM(util_format_get_nr_components(elements[idx].src_format)) |
+            normalize | VIVS_NFE_GENERIC_ATTRIB_CONFIG0_ENDIAN(ENDIAN_MODE_NO_SWAP) |
+            VIVS_NFE_GENERIC_ATTRIB_CONFIG0_STREAM(elements[idx].vertex_buffer_index) |
+            VIVS_NFE_GENERIC_ATTRIB_CONFIG0_START(elements[idx].src_offset);
+         cs->NFE_GENERIC_ATTRIB_CONFIG1[idx] =
+            COND(nonconsecutive, VIVS_NFE_GENERIC_ATTRIB_CONFIG1_NONCONSECUTIVE) |
+            VIVS_NFE_GENERIC_ATTRIB_CONFIG1_END(end_offset - start_offset);
+      }
       cs->NFE_GENERIC_ATTRIB_SCALE[idx] = 0x3f800000; /* 1 for integer, 1.0 for float */
    }
 

diff --git a/src/gallium/drivers/etnaviv/etnaviv_zsa.c b/src/gallium/drivers/etnaviv/etnaviv_zsa.c
index 22c2020fe5c32b34a68498be82a8874ad5888db8..4e72cd2ffb976bfd86b37587164c349680b3202f 100644 (file)
--- a/src/gallium/drivers/etnaviv/etnaviv_zsa.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_zsa.c
@@ -98,7 +98,8 @@ etna_zsa_state_create(struct pipe_context *pctx,
                                                         : PIPE_FUNC_ALWAYS) |
       COND(so->depth.writemask, VIVS_PE_DEPTH_CONFIG_WRITE_ENABLE) |
       COND(early_z, VIVS_PE_DEPTH_CONFIG_EARLY_Z) |
-      COND(disable_zs, VIVS_PE_DEPTH_CONFIG_DISABLE_ZS);
+      /* this bit changed meaning with HALTI5: */
+      COND(disable_zs && ctx->specs.halti < 5, VIVS_PE_DEPTH_CONFIG_DISABLE_ZS);
    cs->PE_ALPHA_OP =
       COND(so->alpha.enabled, VIVS_PE_ALPHA_OP_ALPHA_TEST) |
       VIVS_PE_ALPHA_OP_ALPHA_FUNC(so->alpha.func) |