Rename memory.op.zero into fill

- Rename memory.op.zero to fill, which accepts one template parameter: one
byte to fill the memory with.
- Fix bug on x86_64: RAX (register keeping the value to fill with) isn't set if
the pointer was already aligned.

source/tanya/memory/arch/x86_64.d

@@ -12,6 +12,8 @@
  */
 module tanya.memory.arch.x86_64;
 
+import tanya.memory.op;
+
 version (D_InlineAsm_X86_64):
 
 pragma(inline, true)
@@ -80,8 +82,22 @@ pure nothrow @system @nogc
     }
 }
 
+private enum const(char[]) MovArrayPointer(string Destination)()
+{
+    string asmCode = "asm pure nothrow @nogc { mov ";
+    version (Windows)
+    {
+        asmCode ~= Destination ~ ", [ RCX + 8 ];";
+    }
+    else
+    {
+        asmCode ~= Destination ~ ", RSI;";
+    }
+    return asmCode ~ "}";
+}
+
 pragma(inline, true)
-package (tanya.memory) void zero(void[] memory)
+package (tanya.memory) void fill(ubyte Byte)(void[] memory)
 pure nothrow @system @nogc
 {
     asm pure nothrow @nogc
@@ -94,7 +110,6 @@ pure nothrow @system @nogc
          * RCX - array.
          */
         mov       R8,           [ RCX ];
-        mov       R9,           [ RCX + 8 ];
     }
     else asm pure nothrow @nogc
     {
@@ -103,53 +118,55 @@ pure nothrow @system @nogc
          * RDI - length.
          */
         mov       R8,           RDI;
-        mov       R9,           RSI;
     }
+    mixin(MovArrayPointer!"R9");
+
     asm pure nothrow @nogc
     {
         // Check for zero length.
         test      R8,           R8;
         jz        end;
-
-        // Set to 0.
-        pxor      XMM0,         XMM0;
-
+    }
+    // Set 128- and 64-bit registers to values we want to fill with.
+    static if (Byte == 0)
+    {
+        asm pure nothrow @nogc
+        {
+            xor  RAX,  RAX;
+            pxor XMM0, XMM0;
+        }
+    }
+    else
+    {
+        enum ulong FilledBytes = FilledBytes!Byte;
+        asm pure nothrow @nogc
+        {
+            mov     RAX,  FilledBytes;
+            movq    XMM0, RAX;
+            movlhps XMM0, XMM0;
+        }
+    }
+    asm pure nothrow @nogc
+    {
         // Check if the pointer is aligned to a 16-byte boundary.
         and       R9,           -0x10;
     }
     // Compute the number of misaligned bytes.
-    version (Windows) asm pure nothrow @nogc
-    {
-        mov       RAX,          [ RCX + 8 ];
-    }
-    else asm pure nothrow @nogc
-    {
-        mov       RAX,          RSI;
-    }
+    mixin(MovArrayPointer!"R10");
     asm pure nothrow @nogc
     {
-        sub       RAX,          R9;
+        sub       R10,          R9;
 
-        test      RAX,          RAX;
+        test      R10,          R10;
         jz aligned;
 
         // Get the number of bytes to be written until we are aligned.
         mov       RDX,          0x10;
-        sub       RDX,          RAX;
-    }
-    version (Windows) asm pure nothrow @nogc
-    {
-        mov       R9,           [ RCX + 8 ];
-    }
-    else asm pure nothrow @nogc
-    {
-        mov       R9,           RSI;
+        sub       RDX,          R10;
     }
+    mixin(MovArrayPointer!"R9");
     asm pure nothrow @nogc
     {
-        // Set RAX to zero, so we can set bytes and dwords.
-        xor       RAX,          RAX;
-
     naligned:
         mov       [ R9 ],       AL; // Write a byte.
 

source/tanya/memory/op.d

@@ -113,17 +113,33 @@ private pure nothrow @safe @nogc unittest
     }
 }
 
+/*
+ * size_t value each of which bytes is set to `Byte`.
+ */
+package template FilledBytes(ubyte Byte, ubyte I = 0)
+{
+    static if (I == size_t.sizeof)
+    {
+        enum size_t FilledBytes = Byte;
+    }
+    else
+    {
+        enum size_t FilledBytes = (FilledBytes!(Byte, I + 1) << 8) | Byte;
+    }
+}
+
 /**
- * Fills $(D_PARAM memory) with zero-valued bytes.
+ * Fills $(D_PARAM memory) with single $(D_PARAM Byte)s.
  *
  * Param:
+ *  Byte   = The value to fill $(D_PARAM memory) with.
  *  memory = Memory block.
  */
-void zero(void[] memory) pure nothrow @trusted @nogc
+void fill(ubyte Byte = 0)(void[] memory) pure nothrow @trusted @nogc
 {
     version (D_InlineAsm_X86_64)
     {
-        tanya.memory.arch.x86_64.zero(memory);
+        tanya.memory.arch.x86_64.fill!Byte(memory);
     }
     else // Naive implementation.
     {
@@ -133,7 +149,7 @@ void zero(void[] memory) pure nothrow @trusted @nogc
         // Align.
         while (((cast(size_t) vp) & alignmentMask) != 0)
         {
-            *vp++ = 0;
+            *vp++ = Byte;
             --n;
         }
 
@@ -141,7 +157,7 @@ void zero(void[] memory) pure nothrow @trusted @nogc
         auto sp = cast(size_t*) vp;
         while (n / size_t.sizeof > 0)
         {
-            *sp++ = 0;
+            *sp++ = FilledBytes!Byte;
             n -= size_t.sizeof;
         }
 
@@ -149,7 +165,7 @@ void zero(void[] memory) pure nothrow @trusted @nogc
         vp = cast(ubyte*) sp;
         while (n--)
         {
-            *vp = 0;
+            *vp = Byte;
             ++vp;
         }
     }
@@ -159,14 +175,14 @@ void zero(void[] memory) pure nothrow @trusted @nogc
 pure nothrow @safe @nogc unittest
 {
     ubyte[9] memory = [1, 2, 3, 4, 5, 6, 7, 8, 9];
-    memory.zero();
+    memory.fill!0();
     foreach (ubyte v; memory)
     {
         assert(v == 0);
     }
 }
 
-// Stress test. Checks that `zero` can handle unaligned pointers and different
+// Stress test. Checks that `fill` can handle unaligned pointers and different
 // lengths.
 pure nothrow @safe @nogc private unittest
 {
@@ -178,10 +194,15 @@ pure nothrow @safe @nogc private unittest
         {
             v = i;
         }
-        zero(memory[j .. $]);
+        fill(memory[j .. $]);
         foreach (ubyte v; memory[j .. $])
         {
             assert(v == 0);
         }
+        fill!1(memory[j .. $]);
+        foreach (ubyte v; memory[j .. $])
+        {
+            assert(v == 1);
+        }
     }
 }