[Mesa-dev] [PATCH 1/7] tgsi/ureg: cleanup local temporary emission

[Christian König]

From: Christian König <christian.koenig at amd.com>

Instead of emitting each temporary separately, emit them in a chunk.

Signed-off-by: Christian König <christian.koenig at amd.com>
---
 src/gallium/auxiliary/tgsi/tgsi_ureg.c |   53 ++++++++++----------------------
 1 file changed, 17 insertions(+), 36 deletions(-)

diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index 3c2a923..9303dc7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -1260,30 +1260,11 @@ emit_decl_fs(struct ureg_program *ureg,
    out[3].decl_semantic.Index = semantic_index;
 }
 
-
-static void emit_decl( struct ureg_program *ureg,
-                       unsigned file,
-                       unsigned index,
-                       boolean local )
-{
-   union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 2 );
-
-   out[0].value = 0;
-   out[0].decl.Type = TGSI_TOKEN_TYPE_DECLARATION;
-   out[0].decl.NrTokens = 2;
-   out[0].decl.File = file;
-   out[0].decl.UsageMask = TGSI_WRITEMASK_XYZW;
-   out[0].decl.Local = local;
-
-   out[1].value = 0;
-   out[1].decl_range.First = index;
-   out[1].decl_range.Last = index;
-}
-
 static void emit_decl_range( struct ureg_program *ureg,
                              unsigned file,
                              unsigned first,
-                             unsigned count )
+                             unsigned count,
+                             boolean local )
 {
    union tgsi_any_token *out = get_tokens( ureg, DOMAIN_DECL, 2 );
 
@@ -1293,6 +1274,7 @@ static void emit_decl_range( struct ureg_program *ureg,
    out[0].decl.File = file;
    out[0].decl.UsageMask = TGSI_WRITEMASK_XYZW;
    out[0].decl.Semantic = 0;
+   out[0].decl.Local = local;
 
    out[1].value = 0;
    out[1].decl_range.First = first;
@@ -1450,7 +1432,7 @@ static void emit_decls( struct ureg_program *ureg )
    if (ureg->processor == TGSI_PROCESSOR_VERTEX) {
       for (i = 0; i < UREG_MAX_INPUT; i++) {
          if (ureg->vs_inputs[i/32] & (1 << (i%32))) {
-            emit_decl_range( ureg, TGSI_FILE_INPUT, i, 1 );
+            emit_decl_range( ureg, TGSI_FILE_INPUT, i, 1, FALSE );
          }
       }
    } else if (ureg->processor == TGSI_PROCESSOR_FRAGMENT) {
@@ -1496,7 +1478,7 @@ static void emit_decls( struct ureg_program *ureg )
    for (i = 0; i < ureg->nr_samplers; i++) {
       emit_decl_range( ureg, 
                        TGSI_FILE_SAMPLER,
-                       ureg->sampler[i].Index, 1 );
+                       ureg->sampler[i].Index, 1, FALSE );
    }
 
    for (i = 0; i < ureg->nr_sampler_views; i++) {
@@ -1514,7 +1496,8 @@ static void emit_decls( struct ureg_program *ureg )
          emit_decl_range(ureg,
                          TGSI_FILE_CONSTANT,
                          ureg->const_decls.constant_range[i].first,
-                         ureg->const_decls.constant_range[i].last - ureg->const_decls.constant_range[i].first + 1);
+                         ureg->const_decls.constant_range[i].last - ureg->const_decls.constant_range[i].first + 1,
+                         FALSE);
       }
    }
 
@@ -1535,30 +1518,28 @@ static void emit_decls( struct ureg_program *ureg )
    }
 
    if (ureg->nr_temps) {
-      if (util_bitmask_get_first_index(ureg->local_temps) ==  UTIL_BITMASK_INVALID_INDEX) {
-         emit_decl_range( ureg,
-                          TGSI_FILE_TEMPORARY,
-                          0, ureg->nr_temps );
-
-      } else {
-         for (i = 0; i < ureg->nr_temps; i++) {
-            emit_decl( ureg, TGSI_FILE_TEMPORARY, i,
-                       util_bitmask_get(ureg->local_temps, i) );
-         }
+      for (i = 0; i < ureg->nr_temps;) {
+         boolean local = util_bitmask_get(ureg->local_temps, i);
+         unsigned first = i++;
+         while (i < ureg->nr_temps && local == util_bitmask_get(ureg->local_temps, i))
+            ++i;
+
+         emit_decl_range( ureg, TGSI_FILE_TEMPORARY, first,
+                          i - first, local );
       }
    }
 
    if (ureg->nr_addrs) {
       emit_decl_range( ureg,
                        TGSI_FILE_ADDRESS,
-                       0, ureg->nr_addrs );
+                       0, ureg->nr_addrs, FALSE );
    }
 
    if (ureg->nr_preds) {
       emit_decl_range(ureg,
                       TGSI_FILE_PREDICATE,
                       0,
-                      ureg->nr_preds);
+                      ureg->nr_preds, FALSE);
    }
 
    for (i = 0; i < ureg->nr_immediates; i++) {
-- 
1.7.9.5