reduced stack usage on windows to 16K

kernel/x86_64/dgemm_kernel_4x4_haswell.S

@@ -80,10 +80,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef WINDOWS_ABI
 
 #define STACKSIZE 96
+#define L_BUFFER_SIZE 256*8*12+4096
 
 #else
 
 #define STACKSIZE 256
+#define L_BUFFER_SIZE 128*8*12+4096
 
 #define OLD_A		40 + STACKSIZE(%rsp)
 #define OLD_B		48 + STACKSIZE(%rsp)
@@ -93,7 +95,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#define L_BUFFER_SIZE 512*8*12
 
 #define Ndiv12	 24(%rsp)
 #define Nmod12	 32(%rsp)

param.h

@@ -1228,7 +1228,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_P 256
 
 #define SGEMM_DEFAULT_Q 384
+#ifdef WINDOWS_ABI
+#define DGEMM_DEFAULT_Q 128
+#else
 #define DGEMM_DEFAULT_Q 256
+#endif
 #define CGEMM_DEFAULT_Q 192 
 #define ZGEMM_DEFAULT_Q 128