Ref #65. Fixed 64-bit Windows calling convention bug in cdot and zdot.

According to 64-bit Windows calling convention, the return value is in %rax instead of %xmm0 in cdot kernel.
In zdot, the caller allocates a memory space for return value and sets this memory address to the first hidden parameter. Thus, the callee (zdot) should assign the result to this memory space and return the memory address in %rax.

kernel/x86_64/zdot_sse.S

@@ -3483,6 +3483,10 @@
 	subss	 %xmm3, %xmm1
 #endif
 	unpcklps %xmm1, %xmm0
+	
+#ifdef WINDOWS_ABI
+	movq	%xmm0, %rax
+#endif
 
 	RESTOREREGISTERS
 

kernel/x86_64/zdot_sse2.S

@@ -39,14 +39,19 @@
 #define ASSEMBLER
 #include "common.h"
 
+#ifndef WINDOWS_ABI
 #define N	ARG1	/* rdi */
 #define X	ARG2	/* rsi */
 #define INCX	ARG3	/* rdx */
 #define Y	ARG4	/* rcx */
-#ifndef WINDOWS_ABI
 #define INCY	ARG5	/* r8  */
 #else
-#define INCY	%r10
+#define RESULT_ADDRESS	ARG1    /*rcx*/
+#define N	ARG2	/* rdx */
+#define X	ARG3	/* r8 */
+#define INCX	ARG4	/* r9*/
+#define Y	%r10	
+#define INCY	%r11
 #endif
 
 #include "l1param.h"
@@ -64,7 +69,8 @@
 	PROFCODE
 
 #ifdef WINDOWS_ABI
-	movq	40(%rsp), INCY
+	movq	40(%rsp), Y
+	movq	48(%rsp), INCY
 #endif
 
 	SAVEREGISTERS
@@ -1544,6 +1550,12 @@
 	subsd	 %xmm3, %xmm1
 #endif
 
+#ifdef WINDOWS_ABI
+	movq	RESULT_ADDRESS,	%rax
+	movsd	%xmm0,	(%rax)
+	movsd	%xmm1,	8(%rax)
+#endif
+
 	RESTOREREGISTERS
 	ret