optimize affine transform matrices

* reduce some multiplications/additions for transformations matrices

include/cglm-affine-mat-sse2.h

@@ -0,0 +1,51 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_affine_mat_sse2_h
+#define cglm_affine_mat_sse2_h
+#if defined( __SSE__ ) || defined( __SSE2__ )
+
+#include "cglm-intrin.h"
+#include "cglm.h"
+
+CGLM_INLINE
+void
+glm_affine_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+  __m128 l0, l1, l2, l3, r;
+
+  l0 = _mm_load_ps(m1[0]);
+  l1 = _mm_load_ps(m1[1]);
+  l2 = _mm_load_ps(m1[2]);
+  l3 = _mm_load_ps(m1[3]);
+
+  r = _mm_load_ps(m2[0]);
+  _mm_store_ps(dest[0],
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
+                                     _mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
+                          _mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
+  r = _mm_load_ps(m2[1]);
+  _mm_store_ps(dest[1],
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
+                                     _mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
+                          _mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
+  r = _mm_load_ps(m2[2]);
+  _mm_store_ps(dest[2],
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
+                                     _mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
+                          _mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
+
+  r = _mm_load_ps(m2[3]);
+  _mm_store_ps(dest[3],
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
+                                     _mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
+                          _mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
+                                     _mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
+}
+
+#endif
+#endif /* cglm_affine_mat_sse2_h */

include/cglm-affine-mat.h

@@ -0,0 +1,59 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_affine_mat_h
+#define cglm_affine_mat_h
+
+#include "cglm.h"
+#include "cglm-mat.h"
+#include "cglm-affine-mat-sse2.h"
+#include <assert.h>
+
+CGLM_INLINE
+void
+glm_affine_mul(mat4 m1, mat4 m2, mat4 dest) {
+#if defined( __SSE__ ) || defined( __SSE2__ )
+  glm_affine_mul_sse2(m1, m2, dest);
+#else
+  float a00, a01, a02, a03,    b00, b01, b02, b03,
+        a10, a11, a12, a13,    b10, b11, b12, b13,
+        a20, a21, a22, a23,    b20, b21, b22, b23,
+        a30, a31, a32, a33,                   b33;
+
+  a00 = m1[0][0], a01 = m1[0][1], a02 = m1[0][2], a03 = m1[0][3],
+  a10 = m1[1][0], a11 = m1[1][1], a12 = m1[1][2], a13 = m1[1][3],
+  a20 = m1[2][0], a21 = m1[2][1], a22 = m1[2][2], a23 = m1[2][3],
+  a30 = m1[3][0], a31 = m1[3][1], a32 = m1[3][2], a33 = m1[3][3];
+
+  b00 = m2[0][0], b01 = m2[0][1], b02 = m2[0][2],
+  b10 = m2[1][0], b11 = m2[1][1], b12 = m2[1][2],
+  b20 = m2[2][0], b21 = m2[2][1], b22 = m2[2][2],
+  b30 = m2[3][0], b31 = m2[3][1], b32 = m2[3][2], b33 = m2[3][3];
+
+  dest[0][0] = a00 * b00 + a10 * b01 + a20 * b02;
+  dest[0][1] = a01 * b00 + a11 * b01 + a21 * b02;
+  dest[0][2] = a02 * b00 + a12 * b01 + a22 * b02;
+  dest[0][3] = a03 * b00 + a13 * b01 + a23 * b02;
+
+  dest[1][0] = a00 * b10 + a10 * b11 + a20 * b12;
+  dest[1][1] = a01 * b10 + a11 * b11 + a21 * b12;
+  dest[1][2] = a02 * b10 + a12 * b11 + a22 * b12;
+  dest[1][3] = a03 * b10 + a13 * b11 + a23 * b12;
+
+  dest[2][0] = a00 * b20 + a10 * b21 + a20 * b22;
+  dest[2][1] = a01 * b20 + a11 * b21 + a21 * b22;
+  dest[2][2] = a02 * b20 + a12 * b21 + a22 * b22;
+  dest[2][3] = a03 * b20 + a13 * b21 + a23 * b22;
+
+  dest[3][0] = a00 * b30 + a10 * b31 + a20 * b32 + a30 * b33;
+  dest[3][1] = a01 * b30 + a11 * b31 + a21 * b32 + a31 * b33;
+  dest[3][2] = a02 * b30 + a12 * b31 + a22 * b32 + a32 * b33;
+  dest[3][3] = a03 * b30 + a13 * b31 + a23 * b32 + a33 * b33;
+#endif
+}
+
+#endif /* cglm_affine_mat_h */

include/cglm-affine.h

@@ -10,6 +10,7 @@
 
 #include "cglm.h"
 #include "cglm-vec.h"
+#include "cglm-affine-mat.h"
 #include <math.h>
 
 CGLM_INLINE