Don't wait until the ringbuffer is full to reset the offsets. Reset it right away on batch flushcontiguous_uniforms2

6 files changed, 91 insertions, 57 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 16f7c42401..394ac17cd6 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -348,7 +348,10 @@ struct brw_shader_program {
    drm_intel_bo *bo; /**< Actual uniform backing store */
    unsigned next_offset;
 
-   /* Mapping to uniform storage locations in this shader program */
+   /** If true, the head pointer should be reset to zero at next use.*/
+   bool zero_offsets;
+
+   /** Mapping to uniform storage locations in this shader program */
    struct storage_data {
       struct brw_gather_table *gather_table;
 
@@ -374,6 +377,9 @@ struct brw_shader_program {
       unsigned size;
       unsigned offset;
    } *storage_table;
+
+   /** Pointers to the update flag of live uniform storages */
+   bool **reset_update_flags;
 };
 
 struct brw_gather_table
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 481fa63290..92ee521e6a 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -1936,12 +1936,14 @@ fs_visitor::assign_constant_locations()
    struct brw_shader_program *prog = (struct brw_shader_program *) shader_prog;
    for (unsigned i = 0; i < this->nr_gather_table; i++) {
       int const_idx = this->ubo_gather_table[i].reg;
+      int loc = this->ubo_gather_table[i].uniform_location;
+
       if (push_constant_loc[const_idx] == -1) {
-         printf("Skip GT[%d]\n", i);
+         printf("Skip GT[%d] | loc: %d\n", i,
+                prog->storage_table[loc].base_index);
          continue;
       }
 
-      int loc = this->ubo_gather_table[i].uniform_location;
       prog->storage_table[loc].gather_table = stage_prog_data->gather_table;
 
       if (!prog->storage_table[loc].live_array_indices) {
diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c
index c5211b30c1..d60374ad5b 100644
--- a/src/mesa/drivers/dri/i965/brw_program.c
+++ b/src/mesa/drivers/dri/i965/brw_program.c
@@ -46,6 +46,7 @@
 #include "brw_shader.h"
 #include "brw_nir.h"
 #include "brw_wm.h"
+#include "brw_state.h"
 #include "intel_batchbuffer.h"
 
 static unsigned
@@ -319,55 +320,72 @@ brwUniformStorageChange(struct gl_context *ctx, struct gl_shader_program *prog,
 
    int loc = storage->remap_location;
 
-   printf(" Try update stor: 0x%x |  next_off: %d\n",
-          storage, shader_prog->next_offset
-      );
-   if (shader_prog->storage_table[loc].needs_update &&
-       shader_prog->storage_table[loc].live_array_nr > 0) {
-
-      if ((shader_prog->next_offset + shader_prog->storage_table[loc].size)
-          > shader_prog->bo->size) {
-         /* Our uniform ringbuffer has finally wrapped */
-         intel_batchbuffer_flush(brw);
-         shader_prog->next_offset = 0;
+   /* Always clear out the ring whenever the batch flushes */
+   if (!brw_state_dirty(brw, 0, BRW_NEW_BATCH)) {
+      shader_prog->zero_offsets = false;
+   } else if (!shader_prog->zero_offsets) {
+      printf("                   --------RESETTING to Zero0000000 nrp: %d-----\n",
+             prog->NumUniformStorage); 
+
+      shader_prog->zero_offsets = true;
+      shader_prog->next_offset = 0;
+      for (int i = 0; i < prog->NumUniformStorage; i++) {
+         *shader_prog->reset_update_flags[i] = true;
       }
+   }
+   printf("Try update stor: 0x%x [loc: %d] |  next_off: %d | bat reset? %s\n",
+          storage, loc, shader_prog->next_offset,
+          shader_prog->zero_offsets ? "YES":"NO");
 
-      storage->driver_storage[0].data =
-        shader_prog->bo->virtual + shader_prog->next_offset;
-
-      /* Update the current offset within the buffer */
-      printf("         ++ Update stor: 0x%x prev offset: %d to %d\n",
-             storage, shader_prog->storage_table[loc].offset,
-             shader_prog->next_offset);
-      shader_prog->storage_table[loc].offset = shader_prog->next_offset;
-      shader_prog->storage_table[loc].needs_update = false;
-
-      /* Update the gather table to reflect the new offsets */
-      struct brw_gather_table *gt = shader_prog->storage_table[loc].gather_table;
-
-      printf(" Active indices: 0x%lx | livearr: %d\n",0,
-             shader_prog->storage_table[loc].live_array_nr);
-      unsigned base_size = shader_prog->storage_table[loc].size /
-         MAX2(storage->array_elements, 1);
-      /** Update active entries in the gather table. If the uniform
-       *  storage is an array, the gather table might have been repacked. Only
-       *  update the corresponding offsets in the gather table that are
-       *  marked live.
-       */
-      for (int i = 0; i < shader_prog->storage_table[loc].live_array_nr; i++) {
-         uint16_t gather_idx = shader_prog->storage_table[loc]
-            .live_array_indices[i].gather_index;
-         uint16_t array_idx = shader_prog->storage_table[loc]
-            .live_array_indices[i].array_index;
-         gt[gather_idx].const_offset =
-            shader_prog->storage_table[loc].offset + (base_size * array_idx);
-
-         printf(" GT[%d] = %d | new offset: %d | base: %d\n",
-                gather_idx, array_idx,
-                gt[gather_idx].const_offset, 0);
-      }
-      shader_prog->next_offset += shader_prog->storage_table[loc].size;
+   if (!shader_prog->storage_table[loc].needs_update)
+      return;
+
+   /* Unlikely, but do an explicit flush when the ringbuffer is full enough that
+    * the uniform offset starts to wrap around.
+    */
+   if ((shader_prog->next_offset + shader_prog->storage_table[loc].size)
+       > shader_prog->bo->size) {
+      intel_batchbuffer_flush(brw);
+      shader_prog->next_offset = 0;
+      printf("                  ==-- BATCH FLUSHED --== \n");
+   }
+
+   storage->driver_storage[0].data =
+      shader_prog->bo->virtual + shader_prog->next_offset;
+
+   /* Update the current offset within the buffer */
+   printf("         ++ Update stor: 0x%x prev offset: %d to Current: %d\n",
+          storage, shader_prog->storage_table[loc].offset,
+          shader_prog->next_offset);
+   shader_prog->storage_table[loc].offset = shader_prog->next_offset;
+   shader_prog->storage_table[loc].needs_update = false;
+
+   /* Update the gather table to reflect the new offsets */
+   struct brw_gather_table *gt = shader_prog->storage_table[loc].gather_table;
+
+   printf("    =Active indices: 0x%lx | livearr: %d\n",0,
+          shader_prog->storage_table[loc].live_array_nr);
+   unsigned base_size = shader_prog->storage_table[loc].size /
+      MAX2(storage->array_elements, 1);
+
+   /** Update active entries in the gather table. If the uniform
+    *  storage is an array, the gather table might have been repacked. Only
+    *  update the corresponding offsets in the gather table that are
+    *  marked live.
+    */
+   for (int i = 0; i < shader_prog->storage_table[loc].live_array_nr; i++) {
+      uint16_t gather_idx = shader_prog->storage_table[loc]
+         .live_array_indices[i].gather_index;
+      uint16_t array_idx = shader_prog->storage_table[loc]
+         .live_array_indices[i].array_index;
+      gt[gather_idx].const_offset =
+         shader_prog->storage_table[loc].offset + (base_size * array_idx);
+
+      printf("     GT[%d] = %d | new offset: %d | base: %d\n",
+             gather_idx, array_idx,
+             gt[gather_idx].const_offset, 0);
    }
+   shader_prog->next_offset += shader_prog->storage_table[loc].size;
 }
 
 void brwInitFragProgFuncs( struct dd_function_table *functions )
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 37fd108270..936cff7ce5 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -356,7 +356,10 @@ void brw_associate_uniform_storage(struct gl_context *ctx,
 
    prog->storage_table = rzalloc_array(prog, brw_shader_program::storage_data,
                                        shader_prog->NumUniformRemapTable);
+   prog->reset_update_flags = rzalloc_array(prog, bool *,
+                                            shader_prog->NumUniformRemapTable);
 
+   int flags = 0;
    for (unsigned u = 0; u < shader_prog->NumUniformStorage; u++) {
       struct gl_uniform_storage *storage = &shader_prog->UniformStorage[u];
 
@@ -422,8 +425,10 @@ void brw_associate_uniform_storage(struct gl_context *ctx,
       prog->storage_table[loc].offset = prog->next_offset;
       // prog->next_offset += size;
 
-      printf(" Attach Storage size: %d  dmul: %d cols: %d els: %d | nextoff:%d\n", 
-             size, dmul, columns, MAX2(storage->array_elements, 1),
+      prog->reset_update_flags[flags++] = &prog->storage_table[loc].needs_update;
+      printf(" Attach Storage size: %d [loc:%d] "
+             "dmul: %d cols: %d els: %d | nextoff:%d\n", 
+             size, loc, dmul, columns, MAX2(storage->array_elements, 1),
              prog->next_offset);
     }
 }
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
index 9460544c0c..4a4e4674f9 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
@@ -218,12 +218,13 @@ vec4_visitor::nir_setup_uniform(nir_variable *var)
           assert(uniforms < uniform_array_size);
           uniform_vector_size[uniforms] = storage->type->vector_elements;
 
-          printf("UNI[%d]\n", uniforms);
+          printf("UNI[%d] size: %d\n", uniforms, uniform_vector_size[uniforms]);
           int i;
           for (i = 0; i < uniform_vector_size[uniforms]; i++) {
              int idx = uniforms * 4 + i;
              stage_prog_data->param[idx] = components;
-                printf("     i:%d uni:%d p[%d]\n", i, uniforms, idx);
+             printf("     i:%d uni:%d p[%d] loc: %d\n", i, 
+                    uniforms, idx, loc);
              components++;
           }
           for (; i < 4; i++) {
diff --git a/src/mesa/drivers/dri/i965/gen7_vs_state.c b/src/mesa/drivers/dri/i965/gen7_vs_state.c
index 0dd929e1e4..dbba510df3 100644
--- a/src/mesa/drivers/dri/i965/gen7_vs_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_vs_state.c
@@ -44,7 +44,13 @@ gen7_submit_gather_table(struct brw_context* brw,
    cb_valid |= (prog_data->nr_ubo_params > 0) ?
       (2 << (BRW_UBO_GATHER_INDEX_APPEND + prog_data->max_ubo_const_block)) - 1 : 0;
 
+   printf(" Gen %s GT for stage_prog: 0x%x | gt: 0x%x | num: %d\n",
+          _mesa_shader_stage_to_string(stage_state->stage),
+          prog_data,
+          prog_data->gather_table,
+          prog_data->nr_gather_table);
    assert(cb_valid < 0xffff);
+   assert(prog_data->nr_gather_table > 0 && prog_data->nr_gather_table < 128);
 
    BEGIN_BATCH(gather_dwords);
    OUT_BATCH(gather_opcode << 16 | (gather_dwords - 2));
@@ -52,10 +58,6 @@ gen7_submit_gather_table(struct brw_context* brw,
              SET_FIELD(BRW_UNIFORM_GATHER_INDEX_START / 16,
                        BRW_GATHER_BINDING_TABLE_BLOCK));
    OUT_BATCH(stage_state->push_const_offset);
-   printf(" Gen %s GT for stage_prog: 0x%x | gt: 0x%x\n",
-          _mesa_shader_stage_to_string(stage_state->stage),
-          prog_data,
-          prog_data->gather_table);
    struct brw_shader_program* prog = prog_data->program;
    for (int i = 0; i < prog_data->nr_gather_table; i++) {
       /* Which bo are we referring to? The uniform constant buffer or