From 252bf925fc4ecd9cd939992f3651df15fc84e0c3 Mon Sep 17 00:00:00 2001
From: Recep Aslantas <info@recp.me>
Date: Tue, 8 May 2018 15:25:23 +0300
Subject: [PATCH] simd, sse2: make alignment optional for load operations

---
 include/cglm/affine.h           | 34 ++++++++---------
 include/cglm/mat4.h             |  8 ++--
 include/cglm/quat.h             |  2 +-
 include/cglm/simd/intrin.h      | 12 +++++-
 include/cglm/simd/sse2/affine.h | 38 +++++++++----------
 include/cglm/simd/sse2/mat4.h   | 66 ++++++++++++++++-----------------
 include/cglm/simd/sse2/quat.h   |  4 +-
 include/cglm/vec4-ext.h         |  6 +--
 include/cglm/vec4.h             | 58 ++++++++++++++---------------
 9 files changed, 118 insertions(+), 110 deletions(-)

diff --git a/include/cglm/affine.h b/include/cglm/affine.h
index 8124930..b200f30 100644
--- a/include/cglm/affine.h
+++ b/include/cglm/affine.h
@@ -59,18 +59,18 @@ glm_translate_to(mat4 m, vec3 v, mat4 dest) {
 
 #if defined( __SSE__ ) || defined( __SSE2__ )
   _mm_store_ps(dest[3],
-               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_load_ps(t[0]),
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_load(t[0]),
                                                 _mm_set1_ps(v[0])),
-                                     _mm_mul_ps(_mm_load_ps(t[1]),
+                                     _mm_mul_ps(glmm_load(t[1]),
                                                 _mm_set1_ps(v[1]))),
-                          _mm_add_ps(_mm_mul_ps(_mm_load_ps(t[2]),
+                          _mm_add_ps(_mm_mul_ps(glmm_load(t[2]),
                                                 _mm_set1_ps(v[2])),
-                                     _mm_load_ps(t[3]))))
+                                     glmm_load(t[3]))))
   ;
 
-  _mm_store_ps(dest[0], _mm_load_ps(m[0]));
-  _mm_store_ps(dest[1], _mm_load_ps(m[1]));
-  _mm_store_ps(dest[2], _mm_load_ps(m[2]));
+  _mm_store_ps(dest[0], glmm_load(m[0]));
+  _mm_store_ps(dest[1], glmm_load(m[1]));
+  _mm_store_ps(dest[2], glmm_load(m[2]));
 #else
   vec4 v1, v2, v3;
 
@@ -98,13 +98,13 @@ void
 glm_translate(mat4 m, vec3 v) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   _mm_store_ps(m[3],
-               _mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_load_ps(m[0]),
+               _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
                                                 _mm_set1_ps(v[0])),
-                                     _mm_mul_ps(_mm_load_ps(m[1]),
+                                     _mm_mul_ps(glmm_load(m[1]),
                                                 _mm_set1_ps(v[1]))),
-                          _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
+                          _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
                                                 _mm_set1_ps(v[2])),
-                                     _mm_load_ps(m[3]))))
+                                     glmm_load(m[3]))))
   ;
 #else
   vec4 v1, v2, v3;
@@ -130,9 +130,9 @@ void
 glm_translate_x(mat4 m, float x) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   _mm_store_ps(m[3],
-               _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[0]),
+               _mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
                                      _mm_set1_ps(x)),
-                          _mm_load_ps(m[3])))
+                          glmm_load(m[3])))
   ;
 #else
   vec4 v1;
@@ -152,9 +152,9 @@ void
 glm_translate_y(mat4 m, float y) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   _mm_store_ps(m[3],
-               _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[1]),
+               _mm_add_ps(_mm_mul_ps(glmm_load(m[1]),
                                      _mm_set1_ps(y)),
-                          _mm_load_ps(m[3])))
+                          glmm_load(m[3])))
   ;
 #else
   vec4 v1;
@@ -174,9 +174,9 @@ void
 glm_translate_z(mat4 m, float z) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   _mm_store_ps(m[3],
-               _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
+               _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
                                      _mm_set1_ps(z)),
-                          _mm_load_ps(m[3])))
+                          glmm_load(m[3])))
   ;
 #else
   vec4 v1;
diff --git a/include/cglm/mat4.h b/include/cglm/mat4.h
index d1a72cb..8cafb7c 100644
--- a/include/cglm/mat4.h
+++ b/include/cglm/mat4.h
@@ -113,10 +113,10 @@ glm_mat4_copy(mat4 mat, mat4 dest) {
   _mm256_store_ps(dest[0], _mm256_load_ps(mat[0]));
   _mm256_store_ps(dest[2], _mm256_load_ps(mat[2]));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest[0], _mm_load_ps(mat[0]));
-  _mm_store_ps(dest[1], _mm_load_ps(mat[1]));
-  _mm_store_ps(dest[2], _mm_load_ps(mat[2]));
-  _mm_store_ps(dest[3], _mm_load_ps(mat[3]));
+  _mm_store_ps(dest[0], glmm_load(mat[0]));
+  _mm_store_ps(dest[1], glmm_load(mat[1]));
+  _mm_store_ps(dest[2], glmm_load(mat[2]));
+  _mm_store_ps(dest[3], glmm_load(mat[3]));
 #else
   glm_mat4_ucopy(mat, dest);
 #endif
diff --git a/include/cglm/quat.h b/include/cglm/quat.h
index f6dc86e..6bff527 100644
--- a/include/cglm/quat.h
+++ b/include/cglm/quat.h
@@ -198,7 +198,7 @@ glm_quat_normalize_to(versor q, versor dest) {
   __m128 xdot, x0;
   float  dot;
 
-  x0   = _mm_load_ps(q);
+  x0   = glmm_load(q);
   xdot = glm_simd_dot(x0, x0);
   dot  = _mm_cvtss_f32(xdot);
 
diff --git a/include/cglm/simd/intrin.h b/include/cglm/simd/intrin.h
index cf6753f..bf1db60 100644
--- a/include/cglm/simd/intrin.h
+++ b/include/cglm/simd/intrin.h
@@ -35,7 +35,7 @@
      _mm_shuffle1_ps(_mm_shuffle_ps(a, b, _MM_SHUFFLE(z0, y0, x0, w0)),        \
                                     z1, y1, x1, w1)
 
-CGLM_INLINE
+static inline
 __m128
 glm_simd_dot(__m128 a, __m128 b) {
   __m128 x0;
@@ -44,7 +44,7 @@ glm_simd_dot(__m128 a, __m128 b) {
   return _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1));
 }
 
-CGLM_INLINE
+static inline
 __m128
 glm_simd_norm(__m128 a) {
   return _mm_sqrt_ps(glm_simd_dot(a, a));
@@ -69,6 +69,14 @@ glm_simd_store_v3(__m128 vx, vec3 v) {
   _mm_store_ss(&v[2], _mm_shuffle1_ps(vx, 2, 2, 2, 2));
 }
 
+#ifdef CGLM_ALL_UNALIGNED
+#define glmm_load(p)      _mm_loadu_ps(p)
+#define glmm_store(p, a)  _mm_storeu_ps(p, a)
+#else
+#define glmm_load(p)      _mm_load_ps(p)
+#define glmm_store(p, a)  _mm_store_ps(p, a)
+#endif
+
 #endif
 
 /* x86, x64 */
diff --git a/include/cglm/simd/sse2/affine.h b/include/cglm/simd/sse2/affine.h
index b746d0f..df8c166 100644
--- a/include/cglm/simd/sse2/affine.h
+++ b/include/cglm/simd/sse2/affine.h
@@ -18,30 +18,30 @@ glm_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]);
+  l0 = glmm_load(m1[0]);
+  l1 = glmm_load(m1[1]);
+  l2 = glmm_load(m1[2]);
+  l3 = glmm_load(m1[3]);
 
-  r = _mm_load_ps(m2[0]);
+  r = glmm_load(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]);
+  r = glmm_load(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]);
+  r = glmm_load(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]);
+  r = glmm_load(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)),
@@ -55,24 +55,24 @@ glm_mul_rot_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]);
+  l0 = glmm_load(m1[0]);
+  l1 = glmm_load(m1[1]);
+  l2 = glmm_load(m1[2]);
+  l3 = glmm_load(m1[3]);
 
-  r = _mm_load_ps(m2[0]);
+  r = glmm_load(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]);
+  r = glmm_load(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]);
+  r = glmm_load(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)),
@@ -86,10 +86,10 @@ void
 glm_inv_tr_sse2(mat4 mat) {
   __m128 r0, r1, r2, r3, x0, x1;
 
-  r0 = _mm_load_ps(mat[0]);
-  r1 = _mm_load_ps(mat[1]);
-  r2 = _mm_load_ps(mat[2]);
-  r3 = _mm_load_ps(mat[3]);
+  r0 = glmm_load(mat[0]);
+  r1 = glmm_load(mat[1]);
+  r2 = glmm_load(mat[2]);
+  r3 = glmm_load(mat[3]);
   x1  = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f);
 
   _MM_TRANSPOSE4_PS(r0, r1, r2, x1);
diff --git a/include/cglm/simd/sse2/mat4.h b/include/cglm/simd/sse2/mat4.h
index 77874a8..404b496 100644
--- a/include/cglm/simd/sse2/mat4.h
+++ b/include/cglm/simd/sse2/mat4.h
@@ -20,10 +20,10 @@ glm_mat4_scale_sse2(mat4 m, float s){
   __m128 x0;
   x0 = _mm_set1_ps(s);
 
-  _mm_store_ps(m[0], _mm_mul_ps(_mm_load_ps(m[0]), x0));
-  _mm_store_ps(m[1], _mm_mul_ps(_mm_load_ps(m[1]), x0));
-  _mm_store_ps(m[2], _mm_mul_ps(_mm_load_ps(m[2]), x0));
-  _mm_store_ps(m[3], _mm_mul_ps(_mm_load_ps(m[3]), x0));
+  _mm_store_ps(m[0], _mm_mul_ps(glmm_load(m[0]), x0));
+  _mm_store_ps(m[1], _mm_mul_ps(glmm_load(m[1]), x0));
+  _mm_store_ps(m[2], _mm_mul_ps(glmm_load(m[2]), x0));
+  _mm_store_ps(m[3], _mm_mul_ps(glmm_load(m[3]), x0));
 }
 
 CGLM_INLINE
@@ -31,10 +31,10 @@ void
 glm_mat4_transp_sse2(mat4 m, mat4 dest){
   __m128 r0, r1, r2, r3;
 
-  r0 = _mm_load_ps(m[0]);
-  r1 = _mm_load_ps(m[1]);
-  r2 = _mm_load_ps(m[2]);
-  r3 = _mm_load_ps(m[3]);
+  r0 = glmm_load(m[0]);
+  r1 = glmm_load(m[1]);
+  r2 = glmm_load(m[2]);
+  r3 = glmm_load(m[3]);
 
   _MM_TRANSPOSE4_PS(r0, r1, r2, r3);
 
@@ -51,31 +51,31 @@ glm_mat4_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
 
   __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]);
+  l0 = glmm_load(m1[0]);
+  l1 = glmm_load(m1[1]);
+  l2 = glmm_load(m1[2]);
+  l3 = glmm_load(m1[3]);
 
-  r = _mm_load_ps(m2[0]);
+  r = glmm_load(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_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
                                      _mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
-  r = _mm_load_ps(m2[1]);
+  r = glmm_load(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_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
                                      _mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
-  r = _mm_load_ps(m2[2]);
+  r = glmm_load(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_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
                                      _mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
 
-  r = _mm_load_ps(m2[3]);
+  r = glmm_load(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)),
@@ -88,15 +88,15 @@ void
 glm_mat4_mulv_sse2(mat4 m, vec4 v, vec4 dest) {
   __m128 x0, x1, x2;
 
-  x0 = _mm_load_ps(v);
-  x1 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[0]),
+  x0 = glmm_load(v);
+  x1 = _mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
                              _mm_shuffle1_ps1(x0, 0)),
-                  _mm_mul_ps(_mm_load_ps(m[1]),
+                  _mm_mul_ps(glmm_load(m[1]),
                              _mm_shuffle1_ps1(x0, 1)));
 
-  x2 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
+  x2 = _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
                              _mm_shuffle1_ps1(x0, 2)),
-                  _mm_mul_ps(_mm_load_ps(m[3]),
+                  _mm_mul_ps(glmm_load(m[3]),
                              _mm_shuffle1_ps1(x0, 3)));
 
   _mm_store_ps(dest, _mm_add_ps(x1, x2));
@@ -108,10 +108,10 @@ glm_mat4_det_sse2(mat4 mat) {
   __m128 r0, r1, r2, r3, x0, x1, x2;
 
   /* 127 <- 0, [square] det(A) = det(At) */
-  r0 = _mm_load_ps(mat[0]); /* d c b a */
-  r1 = _mm_load_ps(mat[1]); /* h g f e */
-  r2 = _mm_load_ps(mat[2]); /* l k j i */
-  r3 = _mm_load_ps(mat[3]); /* p o n m */
+  r0 = glmm_load(mat[0]); /* d c b a */
+  r1 = glmm_load(mat[1]); /* h g f e */
+  r2 = glmm_load(mat[2]); /* l k j i */
+  r3 = glmm_load(mat[3]); /* p o n m */
 
   /*
    t[1] = j * p - n * l;
@@ -166,10 +166,10 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
          x0, x1, x2, x3, x4, x5, x6, x7;
 
   /* 127 <- 0 */
-  r0 = _mm_load_ps(mat[0]); /* d c b a */
-  r1 = _mm_load_ps(mat[1]); /* h g f e */
-  r2 = _mm_load_ps(mat[2]); /* l k j i */
-  r3 = _mm_load_ps(mat[3]); /* p o n m */
+  r0 = glmm_load(mat[0]); /* d c b a */
+  r1 = glmm_load(mat[1]); /* h g f e */
+  r2 = glmm_load(mat[2]); /* l k j i */
+  r3 = glmm_load(mat[3]); /* p o n m */
 
   x0 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(3, 2, 3, 2));  /* p o l k */
   x1 = _mm_shuffle1_ps(x0, 1, 3, 3, 3);                  /* l p p p */
@@ -290,10 +290,10 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
          x0, x1, x2, x3, x4, x5, x6, x7;
 
   /* 127 <- 0 */
-  r0 = _mm_load_ps(mat[0]); /* d c b a */
-  r1 = _mm_load_ps(mat[1]); /* h g f e */
-  r2 = _mm_load_ps(mat[2]); /* l k j i */
-  r3 = _mm_load_ps(mat[3]); /* p o n m */
+  r0 = glmm_load(mat[0]); /* d c b a */
+  r1 = glmm_load(mat[1]); /* h g f e */
+  r2 = glmm_load(mat[2]); /* l k j i */
+  r3 = glmm_load(mat[3]); /* p o n m */
 
   x0 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(3, 2, 3, 2));  /* p o l k */
   x1 = _mm_shuffle1_ps(x0, 1, 3, 3, 3);                  /* l p p p */
diff --git a/include/cglm/simd/sse2/quat.h b/include/cglm/simd/sse2/quat.h
index 5dbf759..4970eff 100644
--- a/include/cglm/simd/sse2/quat.h
+++ b/include/cglm/simd/sse2/quat.h
@@ -24,8 +24,8 @@ glm_quat_mul_sse2(versor p, versor q, versor dest) {
 
   __m128 xp, xq, x0, r;
 
-  xp = _mm_load_ps(p); /* 3 2 1 0 */
-  xq = _mm_load_ps(q);
+  xp = glmm_load(p); /* 3 2 1 0 */
+  xq = glmm_load(q);
 
   r  = _mm_mul_ps(_mm_shuffle1_ps1(xp, 3), xq);
 
diff --git a/include/cglm/vec4-ext.h b/include/cglm/vec4-ext.h
index 7a6cb3d..1055ebe 100644
--- a/include/cglm/vec4-ext.h
+++ b/include/cglm/vec4-ext.h
@@ -42,7 +42,7 @@ CGLM_INLINE
 void
 glm_vec4_mulv(vec4 a, vec4 b, vec4 d) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(d, _mm_mul_ps(_mm_load_ps(a), _mm_load_ps(b)));
+  _mm_store_ps(d, _mm_mul_ps(glmm_load(a), glmm_load(b)));
 #else
   d[0] = a[0] * b[0];
   d[1] = a[1] * b[1];
@@ -223,7 +223,7 @@ glm_vec4_sign(vec4 v, vec4 dest) {
 #if defined( __SSE2__ ) || defined( __SSE2__ )
   __m128 x0, x1, x2, x3, x4;
 
-  x0 = _mm_load_ps(v);
+  x0 = glmm_load(v);
   x1 = _mm_set_ps(0.0f, 0.0f, 1.0f, -1.0f);
   x2 = _mm_shuffle1_ps1(x1, 2);
 
@@ -249,7 +249,7 @@ CGLM_INLINE
 void
 glm_vec4_sqrt(vec4 v, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_sqrt_ps(_mm_load_ps(v)));
+  _mm_store_ps(dest, _mm_sqrt_ps(glmm_load(v)));
 #else
   dest[0] = sqrtf(v[0]);
   dest[1] = sqrtf(v[1]);
diff --git a/include/cglm/vec4.h b/include/cglm/vec4.h
index 2e9ca45..b2a9b97 100644
--- a/include/cglm/vec4.h
+++ b/include/cglm/vec4.h
@@ -111,7 +111,7 @@ CGLM_INLINE
 void
 glm_vec4_copy(vec4 v, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_load_ps(v));
+  _mm_store_ps(dest, glmm_load(v));
 #else
   dest[0] = v[0];
   dest[1] = v[1];
@@ -169,7 +169,7 @@ float
 glm_vec4_dot(vec4 a, vec4 b) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   __m128 x0;
-  x0 = _mm_mul_ps(_mm_load_ps(a), _mm_load_ps(b));
+  x0 = _mm_mul_ps(glmm_load(a), glmm_load(b));
   x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
   return _mm_cvtss_f32(_mm_add_ss(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1)));
 #else
@@ -193,7 +193,7 @@ float
 glm_vec4_norm2(vec4 v) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   __m128 x0;
-  x0 = _mm_load_ps(v);
+  x0 = glmm_load(v);
   x0 = _mm_mul_ps(x0, x0);
   x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
   return _mm_cvtss_f32(_mm_add_ss(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1)));
@@ -214,7 +214,7 @@ float
 glm_vec4_norm(vec4 vec) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   __m128 x0;
-  x0 = _mm_load_ps(vec);
+  x0 = glmm_load(vec);
   return _mm_cvtss_f32(_mm_sqrt_ss(glm_simd_dot(x0, x0)));
 #else
   return sqrtf(glm_vec4_norm2(vec));
@@ -232,7 +232,7 @@ CGLM_INLINE
 void
 glm_vec4_add(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(a), _mm_load_ps(b)));
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(a), glmm_load(b)));
 #else
   dest[0] = a[0] + b[0];
   dest[1] = a[1] + b[1];
@@ -252,7 +252,7 @@ CGLM_INLINE
 void
 glm_vec4_adds(vec4 v, float s, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(v), _mm_set1_ps(s)));
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(v), _mm_set1_ps(s)));
 #else
   dest[0] = v[0] + s;
   dest[1] = v[1] + s;
@@ -272,7 +272,7 @@ CGLM_INLINE
 void
 glm_vec4_sub(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_sub_ps(_mm_load_ps(a), _mm_load_ps(b)));
+  _mm_store_ps(dest, _mm_sub_ps(glmm_load(a), glmm_load(b)));
 #else
   dest[0] = a[0] - b[0];
   dest[1] = a[1] - b[1];
@@ -292,7 +292,7 @@ CGLM_INLINE
 void
 glm_vec4_subs(vec4 v, float s, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_sub_ps(_mm_load_ps(v), _mm_set1_ps(s)));
+  _mm_store_ps(dest, _mm_sub_ps(glmm_load(v), _mm_set1_ps(s)));
 #else
   dest[0] = v[0] - s;
   dest[1] = v[1] - s;
@@ -312,7 +312,7 @@ CGLM_INLINE
 void
 glm_vec4_mul(vec4 a, vec4 b, vec4 d) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(d, _mm_mul_ps(_mm_load_ps(a), _mm_load_ps(b)));
+  _mm_store_ps(d, _mm_mul_ps(glmm_load(a), glmm_load(b)));
 #else
   d[0] = a[0] * b[0];
   d[1] = a[1] * b[1];
@@ -332,7 +332,7 @@ CGLM_INLINE
 void
 glm_vec4_scale(vec4 v, float s, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_mul_ps(_mm_load_ps(v), _mm_set1_ps(s)));
+  _mm_store_ps(dest, _mm_mul_ps(glmm_load(v), _mm_set1_ps(s)));
 #else
   dest[0] = v[0] * s;
   dest[1] = v[1] * s;
@@ -373,7 +373,7 @@ CGLM_INLINE
 void
 glm_vec4_div(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_div_ps(_mm_load_ps(a), _mm_load_ps(b)));
+  _mm_store_ps(dest, _mm_div_ps(glmm_load(a), glmm_load(b)));
 #else
   dest[0] = a[0] / b[0];
   dest[1] = a[1] / b[1];
@@ -393,7 +393,7 @@ CGLM_INLINE
 void
 glm_vec4_divs(vec4 v, float s, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_div_ps(_mm_load_ps(v), _mm_set1_ps(s)));
+  _mm_store_ps(dest, _mm_div_ps(glmm_load(v), _mm_set1_ps(s)));
 #else
   glm_vec4_scale(v, 1.0f / s, dest);
 #endif
@@ -413,9 +413,9 @@ CGLM_INLINE
 void
 glm_vec4_addadd(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(dest),
-                                _mm_add_ps(_mm_load_ps(a),
-                                           _mm_load_ps(b))));
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(dest),
+                                _mm_add_ps(glmm_load(a),
+                                           glmm_load(b))));
 #else
   dest[0] += a[0] + b[0];
   dest[1] += a[1] + b[1];
@@ -437,9 +437,9 @@ CGLM_INLINE
 void
 glm_vec4_subadd(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(dest),
-                                _mm_sub_ps(_mm_load_ps(a),
-                                           _mm_load_ps(b))));
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(dest),
+                                _mm_sub_ps(glmm_load(a),
+                                           glmm_load(b))));
 #else
   dest[0] += a[0] - b[0];
   dest[1] += a[1] - b[1];
@@ -461,9 +461,9 @@ CGLM_INLINE
 void
 glm_vec4_muladd(vec4 a, vec4 b, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(dest),
-                                _mm_mul_ps(_mm_load_ps(a),
-                                           _mm_load_ps(b))));
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(dest),
+                                _mm_mul_ps(glmm_load(a),
+                                           glmm_load(b))));
 #else
   dest[0] += a[0] * b[0];
   dest[1] += a[1] * b[1];
@@ -485,8 +485,8 @@ CGLM_INLINE
 void
 glm_vec4_muladds(vec4 a, float s, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_add_ps(_mm_load_ps(dest),
-                                _mm_mul_ps(_mm_load_ps(a),
+  _mm_store_ps(dest, _mm_add_ps(glmm_load(dest),
+                                _mm_mul_ps(glmm_load(a),
                                            _mm_set1_ps(s))));
 #else
   dest[0] += a[0] * s;
@@ -505,7 +505,7 @@ CGLM_INLINE
 void
 glm_vec4_flipsign(vec4 v) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(v, _mm_xor_ps(_mm_load_ps(v), _mm_set1_ps(-0.0f)));
+  _mm_store_ps(v, _mm_xor_ps(glmm_load(v), _mm_set1_ps(-0.0f)));
 #else
   v[0] = -v[0];
   v[1] = -v[1];
@@ -524,7 +524,7 @@ CGLM_INLINE
 void
 glm_vec4_flipsign_to(vec4 v, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_xor_ps(_mm_load_ps(v),
+  _mm_store_ps(dest, _mm_xor_ps(glmm_load(v),
                                 _mm_set1_ps(-0.0f)));
 #else
   dest[0] = -v[0];
@@ -571,7 +571,7 @@ glm_vec4_normalize_to(vec4 vec, vec4 dest) {
   __m128 xdot, x0;
   float  dot;
 
-  x0   = _mm_load_ps(vec);
+  x0   = glmm_load(vec);
   xdot = glm_simd_dot(x0, x0);
   dot  = _mm_cvtss_f32(xdot);
 
@@ -633,7 +633,7 @@ CGLM_INLINE
 void
 glm_vec4_maxv(vec4 v1, vec4 v2, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_max_ps(_mm_load_ps(v1), _mm_load_ps(v2)));
+  _mm_store_ps(dest, _mm_max_ps(glmm_load(v1), glmm_load(v2)));
 #else
   dest[0] = glm_max(v1[0], v2[0]);
   dest[1] = glm_max(v1[1], v2[1]);
@@ -653,7 +653,7 @@ CGLM_INLINE
 void
 glm_vec4_minv(vec4 v1, vec4 v2, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(dest, _mm_min_ps(_mm_load_ps(v1), _mm_load_ps(v2)));
+  _mm_store_ps(dest, _mm_min_ps(glmm_load(v1), glmm_load(v2)));
 #else
   dest[0] = glm_min(v1[0], v2[0]);
   dest[1] = glm_min(v1[1], v2[1]);
@@ -673,7 +673,7 @@ CGLM_INLINE
 void
 glm_vec4_clamp(vec4 v, float minVal, float maxVal) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(v, _mm_min_ps(_mm_max_ps(_mm_load_ps(v), _mm_set1_ps(minVal)),
+  _mm_store_ps(v, _mm_min_ps(_mm_max_ps(glmm_load(v), _mm_set1_ps(minVal)),
                              _mm_set1_ps(maxVal)));
 #else
   v[0] = glm_clamp(v[0], minVal, maxVal);