Avoid redundant checks for finals bits

[software/dgc/cdgc.git] / rt / gc / cdgc / bits.d

diff --git a/rt/gc/cdgc/bits.d b/rt/gc/cdgc/bits.d
index a2dbb50c9bcd02c8049ac35855bc8c8baaf04fb8..17b71f31d5849d5eb6bb9a46c0a270c97ca018d7 100644 (file)
--- a/rt/gc/cdgc/bits.d
+++ b/rt/gc/cdgc/bits.d
@@ -26,7 +26,9 @@
 
 module rt.gc.cdgc.bits;
 
 
 module rt.gc.cdgc.bits;
 

-import libc = rt.gc.cdgc.libc;
+import os = rt.gc.cdgc.os;
+
+import cstring = tango.stdc.string;

 
 private extern (C) void onOutOfMemoryError();
 
 
 private extern (C) void onOutOfMemoryError();
 


@@ -59,11 +61,21 @@ struct GCBits
     size_t nwords = 0;    // allocated words in data[] excluding sentinals
     size_t nbits = 0;     // number of bits in data[] excluding sentinals
 
     size_t nwords = 0;    // allocated words in data[] excluding sentinals
     size_t nbits = 0;     // number of bits in data[] excluding sentinals
 

-    void Dtor()
+    /// Get the number of bytes needed to store nbits bits
+    size_t data_size()

     {
     {

+        return (nwords + 2) * uint.sizeof; // +2 for sentinels
+    }
+
+    void Dtor(os.Vis vis = os.Vis.PRIV)
+    {
+        // Even when os.dealloc() can be called with a null pointer, the extra
+        // call might be significant. On hard GC benchmarks making the test for
+        // null here (i.e. not making the call) can reduce the GC time by
+        // almost ~5%.

         if (data)
         {
         if (data)
         {

-            libc.free(data);
+            os.dealloc(data, data_size, vis);

             data = null;
         }
     }
             data = null;
         }
     }


@@ -71,16 +83,15 @@ struct GCBits
     invariant
     {
         if (data)
     invariant
     {
         if (data)

-        {
-            assert(nwords * data[0].sizeof * 8 >= nbits);
-        }
+            assert (nwords ==
+                    ((nbits + (BITS_PER_WORD - 1)) >> BITS_SHIFT));

     }
 
     }
 

-    void alloc(size_t nbits)
+    void alloc(size_t nbits, os.Vis vis = os.Vis.PRIV)

     {
         this.nbits = nbits;
     {
         this.nbits = nbits;

-        nwords = (nbits + (BITS_PER_WORD - 1)) >> BITS_SHIFT;
-        data = cast(uint*)libc.calloc(nwords + 2, uint.sizeof);
+        this.nwords = (nbits + (BITS_PER_WORD - 1)) >> BITS_SHIFT;
+        this.data = cast(uint*) os.alloc(data_size, vis);

         if (!data)
             onOutOfMemoryError();
     }
         if (!data)
             onOutOfMemoryError();
     }


@@ -137,14 +148,13 @@ struct GCBits
             }
         }
         else
             }
         }
         else

-        {   uint result;
-
+        {

             //result = (cast(bit *)(data + 1))[i];
             //(cast(bit *)(data + 1))[i] = 0;
 
             uint* p = &data[1 + (i >> BITS_SHIFT)];
             uint  mask = (1 << (i & BITS_MASK));
             //result = (cast(bit *)(data + 1))[i];
             //(cast(bit *)(data + 1))[i] = 0;
 
             uint* p = &data[1 + (i >> BITS_SHIFT)];
             uint  mask = (1 << (i & BITS_MASK));

-            result = *p & mask;
+            uint result = *p & mask;

             *p &= ~mask;
             return result;
         }
             *p &= ~mask;
             return result;
         }


@@ -157,7 +167,7 @@ struct GCBits
             for (;d1!=dEnd;++d1)
                 *d1=0u;
         } else {
             for (;d1!=dEnd;++d1)
                 *d1=0u;
         } else {

-            libc.memset(data + 1, 0, nwords * uint.sizeof);
+            cstring.memset(data + 1, 0, nwords * uint.sizeof);

         }
     }
 
         }
     }
 


@@ -173,7 +183,7 @@ struct GCBits
             for (;d1!=dEnd;++d1,++d2)
                 *d1=*d2;
         } else {
             for (;d1!=dEnd;++d1,++d2)
                 *d1=*d2;
         } else {

-            libc.memcpy(data + 1, f.data + 1, nwords * uint.sizeof);
+            cstring.memcpy(data + 1, f.data + 1, nwords * uint.sizeof);

         }
     }
 
         }
     }
 


@@ -217,3 +227,6 @@ unittest
 
     b.Dtor();
 }
 
     b.Dtor();
 }

+
+
+// vim: set et sw=4 sts=4 :