@@ -83,11 +83,11 @@
 {
 	float (*left)[4] = matrix->_values, (*right)[4] = self->_values;
 	float result[4][4], (*resultPtr)[4] = result;
 
 	__asm__ __volatile__ (
-	    "xorw	%%cx, %%cx\n"
+	    "movl	$4, %%ecx\n\t"
 	    "\n\t"
 	    "0:\n\t"
 	    "movd	(%2), %%mm0\n\t"
 	    "punpckldq	16(%2), %%mm0\n\t"
 	    "pfmul	(%1), %%mm0\n\t"
@@ -94,28 +94,47 @@
 	    "movd	32(%2), %%mm1\n\t"
 	    "punpckldq  48(%2), %%mm1\n\t"
 	    "pfmul	8(%1), %%mm1\n\t"
 	    "pfacc	%%mm1, %%mm0\n\t"
 	    "pfacc	%%mm0, %%mm0\n\t"
-	    "movd	%%mm0, (%0)\n"
-	    "\n\t"
-	    "add	$4, %0\n\t"
-	    "add	$4, %2\n\t"
-	    "incb	%%cl\n\t"
-	    "cmpb	$4, %%cl\n\t"
-	    "jb		0b\n"
-	    "\n\t"
+	    "movd	%%mm0, (%0)\n\t"
+	    "movd	4(%2), %%mm0\n\t"
+	    "punpckldq	20(%2), %%mm0\n\t"
+	    "pfmul	(%1), %%mm0\n\t"
+	    "movd	36(%2), %%mm1\n\t"
+	    "punpckldq  52(%2), %%mm1\n\t"
+	    "pfmul	8(%1), %%mm1\n\t"
+	    "pfacc	%%mm1, %%mm0\n\t"
+	    "pfacc	%%mm0, %%mm0\n\t"
+	    "movd	%%mm0, 4(%0)\n\t"
+	    "movd	8(%2), %%mm0\n\t"
+	    "punpckldq	24(%2), %%mm0\n\t"
+	    "pfmul	(%1), %%mm0\n\t"
+	    "movd	40(%2), %%mm1\n\t"
+	    "punpckldq  56(%2), %%mm1\n\t"
+	    "pfmul	8(%1), %%mm1\n\t"
+	    "pfacc	%%mm1, %%mm0\n\t"
+	    "pfacc	%%mm0, %%mm0\n\t"
+	    "movd	%%mm0, 8(%0)\n\t"
+	    "movd	12(%2), %%mm0\n\t"
+	    "punpckldq	28(%2), %%mm0\n\t"
+	    "pfmul	(%1), %%mm0\n\t"
+	    "movd	44(%2), %%mm1\n\t"
+	    "punpckldq  60(%2), %%mm1\n\t"
+	    "pfmul	8(%1), %%mm1\n\t"
+	    "pfacc	%%mm1, %%mm0\n\t"
+	    "pfacc	%%mm0, %%mm0\n\t"
+	    "movd	%%mm0, 12(%0)\n"
+	    "\n\t"
+	    "add	$16, %0\n\t"
 	    "add	$16, %1\n\t"
-	    "sub	$16, %2\n\t"
-	    "xorb	%%cl, %%cl\n\t"
-	    "incb	%%ch\n\t"
-	    "cmpb	$4, %%ch\n\t"
-	    "jb		0b\n"
+	    "decl	%%ecx\n\t"
+	    "jnz	0b\n"
 	    "\n\t"
 	    "femms"
 	    : "+r"(resultPtr), "+r"(left), "+r"(right)
-	    :: "cx", "mm0", "mm1", "memory"
+	    :: "ecx", "mm0", "mm1", "memory"
 	);
 
 	memcpy(self->_values, result, 16 * sizeof(float));
 }