256-bit entry tracker

3 files changed, 55 insertions, 10 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_binding_tables.c b/src/mesa/drivers/dri/i965/brw_binding_tables.c
index 1ea2b21d6a..447f60bced 100644
--- a/src/mesa/drivers/dri/i965/brw_binding_tables.c
+++ b/src/mesa/drivers/dri/i965/brw_binding_tables.c
@@ -426,8 +426,9 @@ gen7_reset_hw_bt_pool_offsets(struct brw_context *brw)
    brw->hw_bt_pool.next_offset = 0;
    brw->gather_pool.next_offset = 0;
 
+   // If this were contigous, you could just memset 0
    for (int i = 0; i < brw->num_offset_reset; i++)
-      *(brw->needs_offset_reset[i]) = 128;
+      *(brw->needs_offset_reset[i]) = 0;
 }
 
 const struct brw_tracked_state gen7_hw_binding_tables = {
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 0a0f306c80..a2c8af83ea 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -39,6 +39,7 @@
 #include "main/macros.h"
 #include "main/mm.h"
 #include "main/mtypes.h"
+#include "util/bitset.h"
 #include "brw_structs.h"
 #include "intel_aub.h"
 #include "program/prog_parameter.h"
@@ -374,6 +375,10 @@ struct brw_shader_program {
        */
       bool needs_update;
 
+      /* Tracking for the range that the data previously occupied in the buffer */
+      unsigned previous_start_offset;
+      unsigned previous_end_offset;
+
       unsigned size;
       unsigned elements;
       int offset;
@@ -382,6 +387,8 @@ struct brw_shader_program {
    /** Pointers to the update flag of live uniform storages */
    bool **reset_update_flags;
    unsigned nr_updates;
+
+   BITSET_DECLARE(enabled_entries, 256);
 };
 
 struct brw_gather_table
diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c
index 094c86772c..0917c70411 100644
--- a/src/mesa/drivers/dri/i965/brw_program.c
+++ b/src/mesa/drivers/dri/i965/brw_program.c
@@ -333,11 +333,11 @@ brwUniformStorageChange(struct gl_context *ctx, struct gl_shader_program *prog,
    /*       *shader_prog->reset_update_flags[i] = true; */
    /*    } */
    /* } */
-   /* printf("Try update stor: 0x%x [loc: %d] | size: %d |"  */
-   /*        "next_off: %d | bat reset? %s | prog: 0x%x\n", */
-   /*        storage, loc, shader_prog->storage_table[loc].size, */
-   /*        shader_prog->next_offset, */
-   /*        brw_state_dirty(brw, 0, BRW_NEW_BATCH) ? "YES":"NO", shader_prog); */
+   printf("Try update stor: 0x%x [loc: %d] | size: %d |"
+          "next_off: %d | bat reset? %s | prog: 0x%x\n",
+          storage, loc, shader_prog->storage_table[loc].size,
+          shader_prog->next_offset,
+          brw_state_dirty(brw, 0, BRW_NEW_BATCH) ? "YES":"NO", shader_prog);
 
    if (!shader_prog->storage_table[loc].needs_update)
       return;
@@ -347,18 +347,53 @@ brwUniformStorageChange(struct gl_context *ctx, struct gl_shader_program *prog,
     */
    if ((shader_prog->next_offset + shader_prog->storage_table[loc].size)
        > shader_prog->bo->size) {
-      //assert(0);
+      assert(0);
       shader_prog->next_offset = 0;
       // printf("                  ==-- BATCH FLUSHED --== \n");
    }
 
+   /* Check if entry is taken */
+   bool offset_updated;
+   do {
+      offset_updated = false;
+      unsigned offset = shader_prog->next_offset;
+      const int from = offset / 16;
+      const int to = (from + (shader_prog->storage_table[loc].size / 16)) - 1;
+      printf(" RING: 0x%x (%d ==> %d)\n", shader_prog->enabled_entries[0], from, to);
+      
+      if (BITSET_TEST_RANGE(shader_prog->enabled_entries, from, to)) {
+         offset += shader_prog->storage_table[loc].size;
+         offset_updated = true;
+         shader_prog->next_offset = offset;
+         // TODO: handle when offset wraps!
+         printf("  ^^ %d -> %d is TAKEN!\n", from, to);
+         assert(shader_prog->next_offset < shader_prog->bo->size);
+      } else {
+         /* Free previous occupied slot and mark current slot as in-use */
+         BITSET_CLEAR_RANGE(shader_prog->enabled_entries,
+                            shader_prog->storage_table[loc].previous_start_offset,
+                            shader_prog->storage_table[loc].previous_end_offset);
+         printf("     ** Clearing %d -> %d (OLD)\n", 
+                shader_prog->storage_table[loc].previous_start_offset,
+                shader_prog->storage_table[loc].previous_end_offset);
+         
+         shader_prog->storage_table[loc].previous_start_offset = from;
+         shader_prog->storage_table[loc].previous_end_offset = to;
+
+         BITSET_SET_RANGE(shader_prog->enabled_entries, from, to);
+         printf("     ++ Setting %d -> %d as NEW slot\n", 
+                shader_prog->storage_table[loc].previous_start_offset,
+                shader_prog->storage_table[loc].previous_end_offset);
+      }       
+   } while (offset_updated);
+
    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); */
+   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;
 
@@ -412,6 +447,8 @@ brwUniformStorageChange(struct gl_context *ctx, struct gl_shader_program *prog,
    /*    } */
    /* } */
 
+   printf(" sz: %d\n", sizeof(shader_prog->enabled_entries));
+
    shader_prog->next_offset += shader_prog->storage_table[loc].size;
 }