Merge pull request #59 from jonathanplatzer/avx

Fix alignment issue when using AVX

include/cglm/affine-mat.h

@@ -151,7 +151,7 @@ glm_inv_tr(mat4 mat) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_inv_tr_sse2(mat);
 #else
-  CGLM_ALIGN(16) mat3 r;
+  CGLM_ALIGN_MAT mat3 r;
   CGLM_ALIGN(16) vec3 t;
 
   /* rotate */

include/cglm/affine.h

@@ -244,7 +244,7 @@ glm_scale_uni(mat4 m, float s) {
 CGLM_INLINE
 void
 glm_rotate_x(mat4 m, float angle, mat4 dest) {
-  CGLM_ALIGN(16) mat4 t = GLM_MAT4_IDENTITY_INIT;
+  CGLM_ALIGN_MAT mat4 t = GLM_MAT4_IDENTITY_INIT;
   float c, s;
 
   c = cosf(angle);
@@ -269,7 +269,7 @@ glm_rotate_x(mat4 m, float angle, mat4 dest) {
 CGLM_INLINE
 void
 glm_rotate_y(mat4 m, float angle, mat4 dest) {
-  CGLM_ALIGN(16) mat4 t = GLM_MAT4_IDENTITY_INIT;
+  CGLM_ALIGN_MAT mat4 t = GLM_MAT4_IDENTITY_INIT;
   float c, s;
 
   c = cosf(angle);
@@ -294,7 +294,7 @@ glm_rotate_y(mat4 m, float angle, mat4 dest) {
 CGLM_INLINE
 void
 glm_rotate_z(mat4 m, float angle, mat4 dest) {
-  CGLM_ALIGN(16) mat4 t = GLM_MAT4_IDENTITY_INIT;
+  CGLM_ALIGN_MAT mat4 t = GLM_MAT4_IDENTITY_INIT;
   float c, s;
 
   c = cosf(angle);
@@ -351,7 +351,7 @@ glm_rotate_make(mat4 m, float angle, vec3 axis) {
 CGLM_INLINE
 void
 glm_rotate(mat4 m, float angle, vec3 axis) {
-  CGLM_ALIGN(16) mat4 rot;
+  CGLM_ALIGN_MAT mat4 rot;
   glm_rotate_make(rot, angle, axis);
   glm_mul_rot(m, rot, m);
 }

include/cglm/bezier.h

@@ -1,124 +0,0 @@
-/*
- * Copyright (c), Recep Aslantas.
- *
- * MIT License (MIT), http://opensource.org/licenses/MIT
- * Full license can be found in the LICENSE file
- */
-
-#ifndef cglm_bezier_h
-#define cglm_bezier_h
-
-#include "common.h"
-
-/*!
- * @brief cubic bezier interpolation
- *
- * @param[in]  t    time between 0 and 1
- * @param[in]  p0   begin point
- * @param[in]  c0   control point 1
- * @param[in]  c1   control point 2
- * @param[in]  p1   end point
- *
- * @return B(s)
- */
-CGLM_INLINE
-float
-glm_bezier_cubic(float t, float p0, float c0, float c1, float p1) {
-  float s, ss, tt;
-
-  s  = 1.0f - t;
-  ss = s * s;
-  tt = t * t;
-
-  return p0 * ss * s + 3.0f * c0 * t * ss + 3.0f * c1 * tt * s + p1 * tt * t;
-}
-
-/*!
- * @brief cubic bezier interpolation for vec3
- *
- * @param[in]  t    time between 0 and 1
- * @param[in]  p0   begin point
- * @param[in]  c0   control point 1
- * @param[in]  c1   control point 2
- * @param[in]  p1   end point
- * @param[in]  dest destination
- */
-CGLM_INLINE
-void
-glm_bezier_cubicv3(float t, vec3 p0, vec3 c0, vec3 c1, vec3 p1, vec3 dest) {
-  vec3  Bs, tmp0, tmp1, tmp2;
-  float s, ss, tt;
-
-  s  = 1.0f - t;
-  ss = s * s;
-  tt = t * t;
-
-  glm_vec_scale(p0, ss * s, Bs);
-  glm_vec_scale(c0, 3.0f * t * ss, tmp0);
-  glm_vec_scale(c1, 3.0f * tt * s, tmp1);
-  glm_vec_scale(p1, tt * t, tmp2);
-
-  glm_vec_add(Bs, tmp0, Bs);
-  glm_vec_add(Bs, tmp1, Bs);
-  glm_vec_add(Bs, tmp2, Bs);
-
-  glm_vec_copy(Bs, dest);
-}
-
-/*!
- * @brief cubic hermite interpolation
- *
- * @param[in]  t    time between 0 and 1
- * @param[in]  p0   begin point
- * @param[in]  t0   tangent 1
- * @param[in]  t1   tangent 2
- * @param[in]  p1   end point
- *
- * @return B(s)
- */
-CGLM_INLINE
-float
-glm_hermite_cubic(float t, float p0, float t0, float t1, float p1) {
-  float tt, ttt;
-
-  tt  = t * t;
-  ttt = tt * t;
-
-  return p0 * (2.0f * ttt - 3.0f * tt + 1)
-          + t0 * (ttt - 2.0f * tt + t)
-          + p1 * (3.0f * tt - 2.0f * ttt)
-          - t1 * (ttt - tt);
-}
-
-/*!
- * @brief cubic hermite interpolation for vec3
- *
- * @param[in]  t    time between 0 and 1
- * @param[in]  p0   begin point
- * @param[in]  t0   tangent 1
- * @param[in]  t1   tangent 2
- * @param[in]  p1   end point
- * @param[in]  dest destination
- */
-CGLM_INLINE
-void
-glm_hermite_cubicv3(float t, vec3 p0, vec3 t0, vec3 t1, vec3 p1, vec3 dest) {
-  vec3  Hs, tmp0, tmp1, tmp2;
-  float tt, ttt;
-
-  tt  = t * t;
-  ttt = tt * t;
-
-  glm_vec_scale(p0, 2.0f * ttt - 3.0f * tt + 1, Hs);
-  glm_vec_scale(t0, ttt - 2.0f * tt + t, tmp0);
-  glm_vec_scale(p1, 3.0f * tt - 2.0f * ttt, tmp1);
-  glm_vec_scale(t1, ttt - tt, tmp2);
-
-  glm_vec_add(Hs, tmp0, Hs);
-  glm_vec_add(Hs, tmp1, Hs);
-  glm_vec_add(Hs, tmp2, Hs);
-
-  glm_vec_copy(Hs, dest);
-}
-
-#endif /* cglm_bezier_h */

include/cglm/cglm.h

@@ -26,6 +26,5 @@
 #include "project.h"
 #include "sphere.h"
 #include "ease.h"
-#include "bezier.h"
 
 #endif /* cglm_h */

include/cglm/mat3.h

@@ -81,7 +81,7 @@ glm_mat3_copy(mat3 mat, mat3 dest) {
 CGLM_INLINE
 void
 glm_mat3_identity(mat3 mat) {
-  CGLM_ALIGN(16) mat3 t = GLM_MAT3_IDENTITY_INIT;
+  CGLM_ALIGN_MAT mat3 t = GLM_MAT3_IDENTITY_INIT;
   glm_mat3_copy(t, mat);
 }
 
@@ -155,7 +155,7 @@ glm_mat3_transpose_to(mat3 m, mat3 dest) {
 CGLM_INLINE
 void
 glm_mat3_transpose(mat3 m) {
-  CGLM_ALIGN(16) mat3 tmp;
+  CGLM_ALIGN_MAT mat3 tmp;
 
   tmp[0][1] = m[1][0];
   tmp[0][2] = m[2][0];

include/cglm/mat4.h

@@ -139,7 +139,7 @@ glm_mat4_copy(mat4 mat, mat4 dest) {
 CGLM_INLINE
 void
 glm_mat4_identity(mat4 mat) {
-  CGLM_ALIGN(16) mat4 t = GLM_MAT4_IDENTITY_INIT;
+  CGLM_ALIGN_MAT mat4 t = GLM_MAT4_IDENTITY_INIT;
   glm_mat4_copy(t, mat);
 }
 

include/cglm/quat.h

@@ -742,7 +742,7 @@ glm_quat_rotatev(versor q, vec3 v, vec3 dest) {
 CGLM_INLINE
 void
 glm_quat_rotate(mat4 m, versor q, mat4 dest) {
-  CGLM_ALIGN(16) mat4 rot;
+  CGLM_ALIGN_MAT mat4 rot;
   glm_quat_mat4(q, rot);
   glm_mul_rot(m, rot, dest);
 }

include/cglm/types.h

@@ -26,13 +26,25 @@
 #  define CGLM_ALIGN_IF(X) /* no alignment */
 #endif
 
+#ifdef __AVX__
+#  define CGLM_ALIGN_MAT CGLM_ALIGN(32)
+#else
+#  define CGLM_ALIGN_MAT CGLM_ALIGN(16)
+#endif
+
 typedef float                   vec2[2];
 typedef CGLM_ALIGN_IF(8)  float vec3[3];
 typedef int                    ivec3[3];
 typedef CGLM_ALIGN_IF(16) float vec4[4];
 
+#ifdef __AVX__
+typedef CGLM_ALIGN_IF(32) vec3  mat3[3];
+typedef CGLM_ALIGN_IF(32) vec4  mat4[4];
+#else
 typedef                   vec3  mat3[3];
 typedef CGLM_ALIGN_IF(16) vec4  mat4[4];
+#endif
+
 
 typedef vec4                    versor;