Merge pull request #42 from recp/optimizations

simd: optional shuffle configuration to save move instructions

.gitignore

@@ -61,3 +61,11 @@ docs/build/*
 win/cglm_test_*
 * copy.*
 *.o
+*.obj
+*codeanalysis.*.xml
+*codeanalysis.xml
+*.lib
+*.tlog
+win/x64
+win/x85
+win/Debug

README.md

@@ -22,6 +22,8 @@ Complete documentation: http://cglm.readthedocs.io
 - **[bugfix]** euler angles was implemented in reverse order (extrinsic) it was fixed, now they are intrinsic. Make sure that 
 you have the latest version
 - **[major change]** by starting v0.4.0, quaternions are stored as [x, y, z, w], it was [w, x, y, z] in v0.3.5 and earlier versions
+- **[api rename]** by starting v0.4.5, **glm_simd** functions are renamed to **glmm_**  
+- **[new option]** by starting v0.4.5, you can disable alignment requirement, check options in docs.  
 
 #### Note for C++ developers:
 If you don't aware about original GLM library yet, you may also want to look at:

build-deps.sh

@@ -9,19 +9,8 @@
 # check if deps are pulled
 git submodule update --init --recursive
 
-# fix glibtoolize
-
 cd $(dirname "$0")
 
-if [ "$(uname)" = "Darwin" ]; then
-  libtoolBin=$(which glibtoolize)
-  libtoolBinDir=$(dirname "${libtoolBin}")
-
-  if [ ! -f "${libtoolBinDir}/libtoolize" ]; then
-    ln -s $libtoolBin "${libtoolBinDir}/libtoolize"
-  fi
-fi
-
 # general deps: gcc make autoconf automake libtool cmake
 
 # test - cmocka

cglm.podspec

@@ -0,0 +1,28 @@
+Pod::Spec.new do |s|
+
+  # Description
+  s.name         = "cglm"
+  s.version      = "0.4.4"
+  s.summary      = "📽 Optimized OpenGL/Graphics Math (glm) for C"
+  s.description  = <<-DESC
+cglm is math library for graphics programming for C. It is similar to original glm but it is written for C instead of C++ (you can use here too). See the documentation or README for all features.
+                   DESC
+
+  s.documentation_url = "http://cglm.readthedocs.io"
+
+  # Home
+  s.homepage     = "https://github.com/recp/cglm"
+  s.license      = { :type => "MIT", :file => "LICENSE" }
+  s.author       = { "Recep Aslantas" => "recp@acm.org" }
+
+  # Sources
+  s.source               = { :git => "https://github.com/recp/cglm.git", :tag => "v#{s.version}" }
+  s.source_files         = "src", "include/cglm/**/*.h"
+  s.public_header_files  = "include", "include/cglm/**/*.h"
+  s.exclude_files        = "src/win/*", "src/dllmain.c", "src/**/*.h"
+  s.preserve_paths       = "include", "src"
+  s.header_mappings_dir  = "include"
+
+  # Linking
+  s.library = "m"
+end

configure.ac

@@ -7,7 +7,7 @@
 #*****************************************************************************
 
 AC_PREREQ([2.69])
-AC_INIT([cglm], [0.4.3], [info@recp.me])
+AC_INIT([cglm], [0.4.5], [info@recp.me])
 AM_INIT_AUTOMAKE([-Wall -Werror foreign subdir-objects])
 
 AC_CONFIG_MACRO_DIR([m4])

docs/source/conf.py

@@ -62,9 +62,9 @@ author = u'Recep Aslantas'
 # built documents.
 #
 # The short X.Y version.
-version = u'0.4.3'
+version = u'0.4.5'
 # The full version, including alpha/beta/rc tags.
-release = u'0.4.3'
+release = u'0.4.5'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

docs/source/getting_started.rst

@@ -21,17 +21,24 @@ Types:
 As you can see types don't store extra informations in favor of space.
 You can send these values e.g. matrix to OpenGL directly without casting or calling a function like *value_ptr*
 
-Aligment is Required:
+Alignment is Required:
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-**vec4** and **mat4** requires 16 byte aligment because vec4 and mat4 operations are
+**vec4** and **mat4** requires 16 byte alignment because vec4 and mat4 operations are
 vectorized by SIMD instructions (SSE/AVX).
 
+**UPDATE:**
+  By starting v0.4.5 cglm provides an option to disable alignment requirement, it is enabled as default
+
+  | Check :doc:`opt` page for more details
+
+  Also alignment is disabled for older msvc verisons as default. Now alignment is only required in Visual Studio 2017 version 15.6+ if CGLM_ALL_UNALIGNED macro is not defined.
+
 Allocations:
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 *cglm* doesn't alloc any memory on heap. So it doesn't provide any allocator.
 You must allocate memory yourself. You should alloc memory for out parameters too if you pass pointer of memory location.
-When allocating memory don't forget that **vec4** and **mat4** requires aligment.
+When allocating memory don't forget that **vec4** and **mat4** requires alignment.
 
 **NOTE:** Unaligned vec4 and unaligned mat4 operations will be supported in the future. Check todo list.
 Because you may want to multiply a CGLM matrix with external matrix.

docs/source/index.rst

@@ -40,6 +40,7 @@ Also currently only **float** type is supported for most operations.
    getting_started
    opengl
    api
+   opt
    troubleshooting
 
 Indices and tables

docs/source/opt.rst

@@ -0,0 +1,42 @@
+.. default-domain:: C
+
+Options
+===============================================================================
+
+A few options are provided via macros.
+
+Alignment Option
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+As default, cglm requires types to be aligned. Alignment requirements:
+
+vec3:   8 byte
+vec4:   16 byte
+mat4:   16 byte
+versor: 16 byte
+
+By starting **v0.4.5** cglm provides an option to disable alignment requirement.
+To enable this option define **CGLM_ALL_UNALIGNED** macro before all headers.
+You can define it in Xcode, Visual Studio (or other IDEs) or you can also prefer
+to define it in build system. If you use pre-compiled verisons then you
+have to compile cglm with **CGLM_ALL_UNALIGNED** macro.
+
+**VERY VERY IMPORTANT:** If you use cglm in multiple projects and
+ those projects are depends on each other, then
+
+ | *ALWAYS* or *NEVER USE* **CGLM_ALL_UNALIGNED** macro in linked projects
+
+ if you do not know what you are doing. Because a cglm header included
+ via 'project A' may force types to be aligned and another cglm header
+ included via 'project B' may not require alignment. In this case
+ cglm functions will read from and write to **INVALID MEMORY LOCATIONs**.
+
+ ALWAYS USE SAME CONFIGURATION / OPTION for **cglm** if you have multiple projects.
+
+ For instance if you set CGLM_ALL_UNALIGNED in a project then set it in other projects too
+
+SSE and SSE2 Shuffle Option
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+**_mm_shuffle_ps** generates **shufps** instruction even if registers are same.
+You can force it to generate **pshufd** instruction by defining
+**CGLM_USE_INT_DOMAIN** macro. As default it is not defined.

docs/source/troubleshooting.rst

@@ -15,10 +15,10 @@ makes calculations then copy the result to dest.
 
 You are responsible for allocation of **src** and **dest** parameters.
 
-Aligment:
+Alignment:
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-**vec4** and **mat4** types requires 16 byte aligment aligment.
+**vec4** and **mat4** types requires 16 byte alignment.
 These types are marked with align attribute to let compiler know about this
 requirement.
 
@@ -26,7 +26,7 @@ But since MSVC (Windows) throws the error:
 
 **"formal parameter with requested alignment of 16 won't be aligned"**
 
-The aligment attribute has been commented for MSVC
+The alignment attribute has been commented for MSVC
 
 .. code-block:: c
 
@@ -36,13 +36,16 @@ The aligment attribute has been commented for MSVC
    #  define CGLM_ALIGN(X) __attribute((aligned(X)))
    #endif.
 
-So MSVC may not know about aligment requirements when creating variables.
+So MSVC may not know about alignment requirements when creating variables.
 The interesting thing is that, if I remember correctly Visual Studio 2017
 doesn't throw the above error. So we may uncomment that line for Visual Studio 2017,
 you may do it yourself.
 
 **This MSVC issue is still in TODOs.**
 
+**UPDATE:** By starting v0.4.5 cglm provides an option to disable alignment requirement.
+Also alignment is disabled for older msvc verisons as default. Now alignment is only required in Visual Studio 2017 version 15.6+ if CGLM_ALL_UNALIGNED macro is not defined.
+
 Crashes, Invalid Memory Access:
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -61,7 +64,7 @@ Again, you may used wrong function.
 
 For instance if you use **glm_normalize()** or **glm_vec_normalize()** for **vec4**,
 it will assume that passed param is **vec3** and will normalize it for **vec3**.
-Since you need to **vec4** to be normalized in your case, you will get wrong results. 
+Since you need to **vec4** to be normalized in your case, you will get wrong results.
 
 Accessing vec4 type with vec3 functions is valid, you will not get any error, exception or crash.
 You only get wrong results if you don't know what you are doing!

include/cglm/affine-mat.h

@@ -16,6 +16,7 @@
 
 #include "common.h"
 #include "mat4.h"
+#include "mat3.h"
 
 #ifdef CGLM_SSE_FP
 #  include "simd/sse2/affine.h"

include/cglm/affine.h

@@ -34,10 +34,11 @@
 #define cglm_affine_h
 
 #include "common.h"
-#include "vec4.h"
-#include "affine-mat.h"
 #include "util.h"
+#include "vec3.h"
+#include "vec4.h"
 #include "mat4.h"
+#include "affine-mat.h"
 
 CGLM_INLINE
 void
@@ -57,19 +58,19 @@ glm_translate_to(mat4 m, vec3 v, mat4 dest) {
   mat4 t = GLM_MAT4_IDENTITY_INIT;
 
 #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_set1_ps(v[0])),
-                                     _mm_mul_ps(_mm_load_ps(t[1]),
-                                                _mm_set1_ps(v[1]))),
-                          _mm_add_ps(_mm_mul_ps(_mm_load_ps(t[2]),
-                                                _mm_set1_ps(v[2])),
-                                     _mm_load_ps(t[3]))))
+  glmm_store(dest[3],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_load(t[0]),
+                                              _mm_set1_ps(v[0])),
+                                   _mm_mul_ps(glmm_load(t[1]),
+                                              _mm_set1_ps(v[1]))),
+                        _mm_add_ps(_mm_mul_ps(glmm_load(t[2]),
+                                              _mm_set1_ps(v[2])),
+                                   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]));
+  glmm_store(dest[0], glmm_load(m[0]));
+  glmm_store(dest[1], glmm_load(m[1]));
+  glmm_store(dest[2], glmm_load(m[2]));
 #else
   vec4 v1, v2, v3;
 
@@ -96,14 +97,14 @@ CGLM_INLINE
 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_set1_ps(v[0])),
-                                     _mm_mul_ps(_mm_load_ps(m[1]),
-                                                _mm_set1_ps(v[1]))),
-                          _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
-                                                _mm_set1_ps(v[2])),
-                                     _mm_load_ps(m[3]))))
+  glmm_store(m[3],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
+                                              _mm_set1_ps(v[0])),
+                                   _mm_mul_ps(glmm_load(m[1]),
+                                              _mm_set1_ps(v[1]))),
+                        _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
+                                              _mm_set1_ps(v[2])),
+                                   glmm_load(m[3]))))
   ;
 #else
   vec4 v1, v2, v3;
@@ -128,10 +129,10 @@ CGLM_INLINE
 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_set1_ps(x)),
-                          _mm_load_ps(m[3])))
+  glmm_store(m[3],
+             _mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
+                                   _mm_set1_ps(x)),
+                        glmm_load(m[3])))
   ;
 #else
   vec4 v1;
@@ -150,10 +151,10 @@ CGLM_INLINE
 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_set1_ps(y)),
-                          _mm_load_ps(m[3])))
+  glmm_store(m[3],
+             _mm_add_ps(_mm_mul_ps(glmm_load(m[1]),
+                                   _mm_set1_ps(y)),
+                        glmm_load(m[3])))
   ;
 #else
   vec4 v1;
@@ -172,10 +173,10 @@ CGLM_INLINE
 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_set1_ps(z)),
-                          _mm_load_ps(m[3])))
+  glmm_store(m[3],
+             _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
+                                   _mm_set1_ps(z)),
+                        glmm_load(m[3])))
   ;
 #else
   vec4 v1;

include/cglm/box.h

@@ -11,6 +11,7 @@
 #include "common.h"
 #include "vec3.h"
 #include "vec4.h"
+#include "util.h"
 
 /*!
  * @brief apply transform to Axis-Aligned Bounding Box

include/cglm/common.h

@@ -14,7 +14,7 @@
 #include <math.h>
 #include <float.h>
 
-#if defined(_WIN32)
+#if defined(_MSC_VER)
 #  ifdef CGLM_DLL
 #    define CGLM_EXPORT __declspec(dllexport)
 #  else

include/cglm/frustum.h

@@ -10,6 +10,9 @@
 
 #include "common.h"
 #include "plane.h"
+#include "vec3.h"
+#include "vec4.h"
+#include "mat4.h"
 
 #define GLM_LBN 0 /* left  bottom near */
 #define GLM_LTN 1 /* left  top    near */

include/cglm/mat3.h

@@ -31,6 +31,7 @@
 #define cglm_mat3_h
 
 #include "common.h"
+#include "vec3.h"
 
 #ifdef CGLM_SSE_FP
 #  include "simd/sse2/mat3.h"

include/cglm/mat4.h

@@ -45,7 +45,8 @@
 #define cglm_mat_h
 
 #include "common.h"
-#include "quat.h"
+#include "vec4.h"
+#include "vec3.h"
 
 #ifdef CGLM_SSE_FP
 #  include "simd/sse2/mat4.h"
@@ -109,13 +110,13 @@ CGLM_INLINE
 void
 glm_mat4_copy(mat4 mat, mat4 dest) {
 #ifdef __AVX__
-  _mm256_store_ps(dest[0], _mm256_load_ps(mat[0]));
-  _mm256_store_ps(dest[2], _mm256_load_ps(mat[2]));
+  glmm_store256(dest[0], glmm_load256(mat[0]));
+  glmm_store256(dest[2], glmm_load256(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]));
+  glmm_store(dest[0], glmm_load(mat[0]));
+  glmm_store(dest[1], glmm_load(mat[1]));
+  glmm_store(dest[2], glmm_load(mat[2]));
+  glmm_store(dest[3], glmm_load(mat[3]));
 #else
   glm_mat4_ucopy(mat, dest);
 #endif

include/cglm/plane.h

@@ -9,9 +9,7 @@
 #define cglm_plane_h
 
 #include "common.h"
-#include "mat4.h"
 #include "vec4.h"
-#include "vec3.h"
 
 /*
  Plane equation:  Ax + By + Cz + D = 0;

include/cglm/project.h

@@ -8,9 +8,9 @@
 #ifndef cglm_project_h
 #define cglm_project_h
 
-#include "mat4.h"
 #include "vec3.h"
 #include "vec4.h"
+#include "mat4.h"
 
 /*!
  * @brief maps the specified viewport coordinates into specified space [1]

include/cglm/quat.h

@@ -52,6 +52,7 @@
 #define cglm_quat_h
 
 #include "common.h"
+#include "vec3.h"
 #include "vec4.h"
 #include "mat4.h"
 #include "mat3.h"
@@ -197,8 +198,8 @@ glm_quat_normalize_to(versor q, versor dest) {
   __m128 xdot, x0;
   float  dot;
 
-  x0   = _mm_load_ps(q);
-  xdot = glm_simd_dot(x0, x0);
+  x0   = glmm_load(q);
+  xdot = glmm_dot(x0, x0);
   dot  = _mm_cvtss_f32(xdot);
 
   if (dot <= 0.0f) {
@@ -206,7 +207,7 @@ glm_quat_normalize_to(versor q, versor dest) {
     return;
   }
 
-  _mm_store_ps(dest, _mm_div_ps(x0, _mm_sqrt_ps(xdot)));
+  glmm_store(dest, _mm_div_ps(x0, _mm_sqrt_ps(xdot)));
 #else
   float dot;
 

include/cglm/simd/avx/affine.h

@@ -21,11 +21,11 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
 
   __m256 y0, y1, y2, y3, y4, y5, y6, y7, y8, y9;
 
-  y0 = _mm256_load_ps(m2[0]); /* h g f e d c b a */
-  y1 = _mm256_load_ps(m2[2]); /* p o n m l k j i */
+  y0 = glmm_load256(m2[0]); /* h g f e d c b a */
+  y1 = glmm_load256(m2[2]); /* p o n m l k j i */
 
-  y2 = _mm256_load_ps(m1[0]); /* h g f e d c b a */
-  y3 = _mm256_load_ps(m1[2]); /* p o n m l k j i */
+  y2 = glmm_load256(m1[0]); /* h g f e d c b a */
+  y3 = glmm_load256(m1[2]); /* p o n m l k j i */
 
   y4 = _mm256_permute2f128_ps(y2, y2, 0b00000011); /* d c b a h g f e */
   y5 = _mm256_permute2f128_ps(y3, y3, 0b00000000); /* l k j i l k j i */
@@ -37,10 +37,10 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
   y6 = _mm256_permutevar_ps(y0, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
   y8 = _mm256_permutevar_ps(y0, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
 
-  _mm256_store_ps(dest[0],
-                  _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
-                                              _mm256_mul_ps(y4, y8)),
-                                _mm256_mul_ps(y5, y7)));
+  glmm_store256(dest[0],
+                _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
+                                            _mm256_mul_ps(y4, y8)),
+                              _mm256_mul_ps(y5, y7)));
 
 
   /* n n n n i i i i */
@@ -52,11 +52,11 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
   y8 = _mm256_permutevar_ps(y1, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
   y9 = _mm256_permutevar_ps(y1, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
 
-  _mm256_store_ps(dest[2],
-                  _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
-                                              _mm256_mul_ps(y3, y7)),
-                                _mm256_add_ps(_mm256_mul_ps(y4, y8),
-                                              _mm256_mul_ps(y5, y9))));
+  glmm_store256(dest[2],
+                _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
+                                            _mm256_mul_ps(y3, y7)),
+                              _mm256_add_ps(_mm256_mul_ps(y4, y8),
+                                            _mm256_mul_ps(y5, y9))));
 }
 
 #endif

include/cglm/simd/avx/mat4.h

@@ -21,11 +21,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
 
   __m256 y0, y1, y2, y3, y4, y5, y6, y7, y8, y9;
 
-  y0 = _mm256_load_ps(m2[0]); /* h g f e d c b a */
-  y1 = _mm256_load_ps(m2[2]); /* p o n m l k j i */
+  y0 = glmm_load256(m2[0]); /* h g f e d c b a */
+  y1 = glmm_load256(m2[2]); /* p o n m l k j i */
 
-  y2 = _mm256_load_ps(m1[0]); /* h g f e d c b a */
-  y3 = _mm256_load_ps(m1[2]); /* p o n m l k j i */
+  y2 = glmm_load256(m1[0]); /* h g f e d c b a */
+  y3 = glmm_load256(m1[2]); /* p o n m l k j i */
 
   y4 = _mm256_permute2f128_ps(y2, y2, 0b00000011); /* d c b a h g f e */
   y5 = _mm256_permute2f128_ps(y3, y3, 0b00000011); /* l k j i p o n m */
@@ -39,11 +39,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
   y8 = _mm256_permutevar_ps(y0, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
   y9 = _mm256_permutevar_ps(y0, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
 
-  _mm256_store_ps(dest[0],
-                  _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
-                                              _mm256_mul_ps(y3, y7)),
-                                _mm256_add_ps(_mm256_mul_ps(y4, y8),
-                                              _mm256_mul_ps(y5, y9))));
+  glmm_store256(dest[0],
+                _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
+                                            _mm256_mul_ps(y3, y7)),
+                              _mm256_add_ps(_mm256_mul_ps(y4, y8),
+                                            _mm256_mul_ps(y5, y9))));
 
   /* n n n n i i i i */
   /* p p p p k k k k */
@@ -54,11 +54,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
   y8 = _mm256_permutevar_ps(y1, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
   y9 = _mm256_permutevar_ps(y1, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
 
-  _mm256_store_ps(dest[2],
-                  _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
-                                              _mm256_mul_ps(y3, y7)),
-                                _mm256_add_ps(_mm256_mul_ps(y4, y8),
-                                              _mm256_mul_ps(y5, y9))));
+  glmm_store256(dest[2],
+                _mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
+                                            _mm256_mul_ps(y3, y7)),
+                              _mm256_add_ps(_mm256_mul_ps(y4, y8),
+                                            _mm256_mul_ps(y5, y9))));
 }
 
 #endif

include/cglm/simd/intrin.h

@@ -8,11 +8,19 @@
 #ifndef cglm_intrin_h
 #define cglm_intrin_h
 
-#if defined( _WIN32 )
+#if defined( _MSC_VER )
 #  if (defined(_M_AMD64) || defined(_M_X64)) || _M_IX86_FP == 2
-#    define __SSE2__
+#    ifndef __SSE2__
+#      define __SSE2__
+#    endif
 #  elif _M_IX86_FP == 1
-#    define __SSE__
+#    ifndef __SSE__
+#      define __SSE__
+#    endif
+#  endif
+/* do not use alignment for older visual studio versions */
+#  if _MSC_VER < 1913     /* Visual Studio 2017 version 15.6 */
+#    define CGLM_ALL_UNALIGNED
 #  endif
 #endif
 
@@ -20,35 +28,39 @@
 #  include <xmmintrin.h>
 #  include <emmintrin.h>
 
-/* float */
-#  define _mm_shuffle1_ps(a, z, y, x, w)                                       \
-     _mm_shuffle_ps(a, a, _MM_SHUFFLE(z, y, x, w))
+/* OPTIONAL: You may save some instructions but latency (not sure) */
+#ifdef CGLM_USE_INT_DOMAIN
+#  define glmm_shuff1(xmm, z, y, x, w)                                        \
+     _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xmm),                \
+                                        _MM_SHUFFLE(z, y, x, w)))
+#else
+#  define glmm_shuff1(xmm, z, y, x, w)                                        \
+     _mm_shuffle_ps(xmm, xmm, _MM_SHUFFLE(z, y, x, w))
+#endif
 
-#  define _mm_shuffle1_ps1(a, x)                                               \
-     _mm_shuffle_ps(a, a, _MM_SHUFFLE(x, x, x, x))
+#define glmm_shuff1x(xmm, x) glmm_shuff1(xmm, x, x, x, x)
+#define glmm_shuff2(a, b, z0, y0, x0, w0, z1, y1, x1, w1)                     \
+     glmm_shuff1(_mm_shuffle_ps(a, b, _MM_SHUFFLE(z0, y0, x0, w0)),           \
+                 z1, y1, x1, w1)
 
-#  define _mm_shuffle2_ps(a, b, z0, y0, x0, w0, z1, y1, x1, w1)                \
-     _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) {
+glmm_dot(__m128 a, __m128 b) {
   __m128 x0;
   x0 = _mm_mul_ps(a, b);
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
-  return _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1));
-}
-
-CGLM_INLINE
-__m128
-glm_simd_norm(__m128 a) {
-  return _mm_sqrt_ps(glm_simd_dot(a, a));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
+  return _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 0, 1));
 }
 
 static inline
 __m128
-glm_simd_load_v3(vec3 v) {
+glmm_norm(__m128 a) {
+  return _mm_sqrt_ps(glmm_dot(a, a));
+}
+
+static inline
+__m128
+glmm_load3(float v[3]) {
   __m128i xy;
   __m128  z;
 
@@ -60,11 +72,19 @@ glm_simd_load_v3(vec3 v) {
 
 static inline
 void
-glm_simd_store_v3(__m128 vx, vec3 v) {
+glmm_store3(__m128 vx, float v[3]) {
   _mm_storel_pi((__m64 *)&v[0], vx);
-  _mm_store_ss(&v[2], _mm_shuffle1_ps(vx, 2, 2, 2, 2));
+  _mm_store_ss(&v[2], glmm_shuff1(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 */
@@ -74,6 +94,15 @@ glm_simd_store_v3(__m128 vx, vec3 v) {
 
 #ifdef __AVX__
 #  define CGLM_AVX_FP 1
+
+#ifdef CGLM_ALL_UNALIGNED
+#  define glmm_load256(p)      _mm256_loadu_ps(p)
+#  define glmm_store256(p, a)  _mm256_storeu_ps(p, a)
+#else
+#  define glmm_load256(p)      _mm256_load_ps(p)
+#  define glmm_store256(p, a)  _mm256_store_ps(p, a)
+#endif
+
 #endif
 
 /* ARM Neon */

include/cglm/simd/sse2/affine.h

@@ -18,35 +18,35 @@ 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]);
-  _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 = glmm_load(m2[0]);
+  glmm_store(dest[0],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(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 = glmm_load(m2[1]);
+  glmm_store(dest[1],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(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 = glmm_load(m2[2]);
+  glmm_store(dest[2],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(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))));
+  r = glmm_load(m2[3]);
+  glmm_store(dest[3],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
+                                   _mm_mul_ps(glmm_shuff1x(r, 3), l3))));
 }
 
 CGLM_INLINE
@@ -55,30 +55,30 @@ 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]);
-  _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 = glmm_load(m2[0]);
+  glmm_store(dest[0],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(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 = glmm_load(m2[1]);
+  glmm_store(dest[1],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(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 = glmm_load(m2[2]);
+  glmm_store(dest[2],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_mul_ps(glmm_shuff1x(r, 2), l2)));
 
-  _mm_store_ps(dest[3], l3);
+  glmm_store(dest[3], l3);
 }
 
 CGLM_INLINE
@@ -86,25 +86,25 @@ 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]);
-  x1  = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f);
+  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);
 
-  x0 = _mm_add_ps(_mm_mul_ps(r0, _mm_shuffle1_ps(r3, 0, 0, 0, 0)),
-                  _mm_mul_ps(r1, _mm_shuffle1_ps(r3, 1, 1, 1, 1)));
-  x0 = _mm_add_ps(x0, _mm_mul_ps(r2, _mm_shuffle1_ps(r3, 2, 2, 2, 2)));
+  x0 = _mm_add_ps(_mm_mul_ps(r0, glmm_shuff1(r3, 0, 0, 0, 0)),
+                  _mm_mul_ps(r1, glmm_shuff1(r3, 1, 1, 1, 1)));
+  x0 = _mm_add_ps(x0, _mm_mul_ps(r2, glmm_shuff1(r3, 2, 2, 2, 2)));
   x0 = _mm_xor_ps(x0, _mm_set1_ps(-0.f));
 
   x0 = _mm_add_ps(x0, x1);
 
-  _mm_store_ps(mat[0], r0);
-  _mm_store_ps(mat[1], r1);
-  _mm_store_ps(mat[2], r2);
-  _mm_store_ps(mat[3], x0);
+  glmm_store(mat[0], r0);
+  glmm_store(mat[1], r1);
+  glmm_store(mat[2], r2);
+  glmm_store(mat[3], x0);
 }
 
 #endif

include/cglm/simd/sse2/mat3.h

@@ -27,27 +27,25 @@ glm_mat3_mul_sse2(mat3 m1, mat3 m2, mat3 dest) {
   r1 = _mm_loadu_ps(&m2[1][1]);
   r2 = _mm_set1_ps(m2[2][2]);
 
-  x1 = _mm_shuffle2_ps(l0, l1, 1, 0, 3, 3, 0, 3, 2, 0);
-  x2 = _mm_shuffle2_ps(l1, l2, 0, 0, 3, 2, 0, 2, 1, 0);
+  x1 = glmm_shuff2(l0, l1, 1, 0, 3, 3, 0, 3, 2, 0);
+  x2 = glmm_shuff2(l1, l2, 0, 0, 3, 2, 0, 2, 1, 0);
 
-  x0 = _mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps(l0, 0, 2, 1, 0),
-                             _mm_shuffle1_ps(r0, 3, 0, 0, 0)),
-                  _mm_mul_ps(x1,
-                             _mm_shuffle2_ps(r0, r1, 0, 0, 1, 1, 2, 0, 0, 0)));
+  x0 = _mm_add_ps(_mm_mul_ps(glmm_shuff1(l0, 0, 2, 1, 0),
+                             glmm_shuff1(r0, 3, 0, 0, 0)),
+                  _mm_mul_ps(x1, glmm_shuff2(r0, r1, 0, 0, 1, 1, 2, 0, 0, 0)));
 
   x0 = _mm_add_ps(x0,
-                  _mm_mul_ps(x2,
-                             _mm_shuffle2_ps(r0, r1, 1, 1, 2, 2, 2, 0, 0, 0)));
+                  _mm_mul_ps(x2, glmm_shuff2(r0, r1, 1, 1, 2, 2, 2, 0, 0, 0)));
 
   _mm_storeu_ps(dest[0], x0);
 
-  x0 = _mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps(l0, 1, 0, 2, 1),
+  x0 = _mm_add_ps(_mm_mul_ps(glmm_shuff1(l0, 1, 0, 2, 1),
                              _mm_shuffle_ps(r0, r1, _MM_SHUFFLE(2, 2, 3, 3))),
-                  _mm_mul_ps(_mm_shuffle1_ps(x1, 1, 0, 2, 1),
-                             _mm_shuffle1_ps(r1, 3, 3, 0, 0)));
+                  _mm_mul_ps(glmm_shuff1(x1, 1, 0, 2, 1),
+                             glmm_shuff1(r1, 3, 3, 0, 0)));
 
   x0 = _mm_add_ps(x0,
-                  _mm_mul_ps(_mm_shuffle1_ps(x2, 1, 0, 2, 1),
+                  _mm_mul_ps(glmm_shuff1(x2, 1, 0, 2, 1),
                              _mm_shuffle_ps(r1, r2, _MM_SHUFFLE(0, 0, 1, 1))));
 
   _mm_storeu_ps(&dest[1][1], x0);

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));
+  glmm_store(m[0], _mm_mul_ps(glmm_load(m[0]), x0));
+  glmm_store(m[1], _mm_mul_ps(glmm_load(m[1]), x0));
+  glmm_store(m[2], _mm_mul_ps(glmm_load(m[2]), x0));
+  glmm_store(m[3], _mm_mul_ps(glmm_load(m[3]), x0));
 }
 
 CGLM_INLINE
@@ -31,17 +31,17 @@ 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);
 
-  _mm_store_ps(dest[0], r0);
-  _mm_store_ps(dest[1], r1);
-  _mm_store_ps(dest[2], r2);
-  _mm_store_ps(dest[3], r3);
+  glmm_store(dest[0], r0);
+  glmm_store(dest[1], r1);
+  glmm_store(dest[2], r2);
+  glmm_store(dest[3], r3);
 }
 
 CGLM_INLINE
@@ -51,36 +51,36 @@ 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]);
-  _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]);
-  _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]);
-  _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 = glmm_load(m2[0]);
+  glmm_store(dest[0],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
+                                   _mm_mul_ps(glmm_shuff1x(r, 3), l3))));
+  r = glmm_load(m2[1]);
+  glmm_store(dest[1],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
+                                   _mm_mul_ps(glmm_shuff1x(r, 3), l3))));
+  r = glmm_load(m2[2]);
+  glmm_store(dest[2],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
+                                   _mm_mul_ps(glmm_shuff1x(r, 3), l3))));
 
-  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))));
+  r = glmm_load(m2[3]);
+  glmm_store(dest[3],
+             _mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
+                                   _mm_mul_ps(glmm_shuff1x(r, 1), l1)),
+                        _mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
+                                   _mm_mul_ps(glmm_shuff1x(r, 3), l3))));
 }
 
 CGLM_INLINE
@@ -88,18 +88,14 @@ 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]),
-                             _mm_shuffle1_ps1(x0, 0)),
-                  _mm_mul_ps(_mm_load_ps(m[1]),
-                             _mm_shuffle1_ps1(x0, 1)));
+  x0 = glmm_load(v);
+  x1 = _mm_add_ps(_mm_mul_ps(glmm_load(m[0]), glmm_shuff1x(x0, 0)),
+                  _mm_mul_ps(glmm_load(m[1]), glmm_shuff1x(x0, 1)));
 
-  x2 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
-                             _mm_shuffle1_ps1(x0, 2)),
-                  _mm_mul_ps(_mm_load_ps(m[3]),
-                             _mm_shuffle1_ps1(x0, 3)));
+  x2 = _mm_add_ps(_mm_mul_ps(glmm_load(m[2]), glmm_shuff1x(x0, 2)),
+                  _mm_mul_ps(glmm_load(m[3]), glmm_shuff1x(x0, 3)));
 
-  _mm_store_ps(dest, _mm_add_ps(x1, x2));
+  glmm_store(dest, _mm_add_ps(x1, x2));
 }
 
 CGLM_INLINE
@@ -108,10 +104,10 @@ 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;
@@ -119,20 +115,20 @@ glm_mat4_det_sse2(mat4 mat) {
    t[3] = i * p - m * l;
    t[4] = i * o - m * k;
    */
-  x0 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 1, 1),
-                             _mm_shuffle1_ps(r3, 2, 3, 2, 3)),
-                  _mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 1, 1),
-                             _mm_shuffle1_ps(r2, 2, 3, 2, 3)));
+  x0 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(r2, 0, 0, 1, 1),
+                             glmm_shuff1(r3, 2, 3, 2, 3)),
+                  _mm_mul_ps(glmm_shuff1(r3, 0, 0, 1, 1),
+                             glmm_shuff1(r2, 2, 3, 2, 3)));
   /*
    t[0] = k * p - o * l;
    t[0] = k * p - o * l;
    t[5] = i * n - m * j;
    t[5] = i * n - m * j;
    */
-  x1 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 2, 2),
-                             _mm_shuffle1_ps(r3, 1, 1, 3, 3)),
-                  _mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 2, 2),
-                             _mm_shuffle1_ps(r2, 1, 1, 3, 3)));
+  x1 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(r2, 0, 0, 2, 2),
+                             glmm_shuff1(r3, 1, 1, 3, 3)),
+                  _mm_mul_ps(glmm_shuff1(r3, 0, 0, 2, 2),
+                             glmm_shuff1(r2, 1, 1, 3, 3)));
 
   /*
      a * (f * t[0] - g * t[1] + h * t[2])
@@ -140,19 +136,19 @@ glm_mat4_det_sse2(mat4 mat) {
    + c * (e * t[1] - f * t[3] + h * t[5])
    - d * (e * t[2] - f * t[4] + g * t[5])
    */
-  x2 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 0, 0, 0, 1),
+  x2 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(r1, 0, 0, 0, 1),
                              _mm_shuffle_ps(x1, x0, _MM_SHUFFLE(1, 0, 0, 0))),
-                  _mm_mul_ps(_mm_shuffle1_ps(r1, 1, 1, 2, 2),
-                             _mm_shuffle1_ps(x0, 3, 2, 2, 0)));
+                  _mm_mul_ps(glmm_shuff1(r1, 1, 1, 2, 2),
+                             glmm_shuff1(x0, 3, 2, 2, 0)));
 
   x2 = _mm_add_ps(x2,
-                  _mm_mul_ps(_mm_shuffle1_ps(r1, 2, 3, 3, 3),
+                  _mm_mul_ps(glmm_shuff1(r1, 2, 3, 3, 3),
                              _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 2, 3, 1))));
   x2 = _mm_xor_ps(x2, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
 
   x0 = _mm_mul_ps(r0, x2);
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 3, 3, 1));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 3, 3, 1));
 
   return _mm_cvtss_f32(x0);
 }
@@ -166,14 +162,14 @@ 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 */
-  x2 = _mm_shuffle1_ps(x0, 0, 2, 2, 2);                  /* k o o o */
+  x1 = glmm_shuff1(x0, 1, 3, 3, 3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
   x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3));  /* h h l l */
   x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2));  /* g g k k */
 
@@ -184,7 +180,7 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
   t0 = _mm_sub_ps(_mm_mul_ps(x3, x1), _mm_mul_ps(x2, x0));
 
   x4 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(2, 1, 2, 1)); /* o n k j */
-  x4 = _mm_shuffle1_ps(x4, 0, 2, 2, 2);                 /* j n n n */
+  x4 = glmm_shuff1(x4, 0, 2, 2, 2);                     /* j n n n */
   x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
 
   /* t1[1] = j * p - n * l;
@@ -200,7 +196,7 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
   t2 = _mm_sub_ps(_mm_mul_ps(x5, x2), _mm_mul_ps(x4, x3));
 
   x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
-  x7 = _mm_shuffle2_ps(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0); /* i m m m */
+  x7 = glmm_shuff2(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0);     /* i m m m */
 
   /* t1[3] = i * p - m * l;
      t1[3] = i * p - m * l;
@@ -220,10 +216,10 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
      t3[5] = e * j - i * f; */
   t5 = _mm_sub_ps(_mm_mul_ps(x6, x4), _mm_mul_ps(x7, x5));
 
-  x0 = _mm_shuffle2_ps(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
-  x1 = _mm_shuffle2_ps(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
-  x2 = _mm_shuffle2_ps(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
-  x3 = _mm_shuffle2_ps(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
+  x0 = glmm_shuff2(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
+  x1 = glmm_shuff2(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
+  x2 = glmm_shuff2(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
+  x3 = glmm_shuff2(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
 
   /*
    dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
@@ -271,14 +267,14 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
   x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
 
   x0 = _mm_mul_ps(x0, r0);
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 0, 1));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 0, 1));
   x0 = _mm_rcp_ps(x0);
 
-  _mm_store_ps(dest[0], _mm_mul_ps(v0, x0));
-  _mm_store_ps(dest[1], _mm_mul_ps(v1, x0));
-  _mm_store_ps(dest[2], _mm_mul_ps(v2, x0));
-  _mm_store_ps(dest[3], _mm_mul_ps(v3, x0));
+  glmm_store(dest[0], _mm_mul_ps(v0, x0));
+  glmm_store(dest[1], _mm_mul_ps(v1, x0));
+  glmm_store(dest[2], _mm_mul_ps(v2, x0));
+  glmm_store(dest[3], _mm_mul_ps(v3, x0));
 }
 
 CGLM_INLINE
@@ -290,14 +286,14 @@ 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 */
-  x2 = _mm_shuffle1_ps(x0, 0, 2, 2, 2);                  /* k o o o */
+  x1 = glmm_shuff1(x0, 1, 3, 3, 3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
   x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3));  /* h h l l */
   x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2));  /* g g k k */
 
@@ -308,7 +304,7 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
   t0 = _mm_sub_ps(_mm_mul_ps(x3, x1), _mm_mul_ps(x2, x0));
 
   x4 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(2, 1, 2, 1)); /* o n k j */
-  x4 = _mm_shuffle1_ps(x4, 0, 2, 2, 2);                 /* j n n n */
+  x4 = glmm_shuff1(x4, 0, 2, 2, 2);                     /* j n n n */
   x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
 
   /* t1[1] = j * p - n * l;
@@ -324,7 +320,7 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
   t2 = _mm_sub_ps(_mm_mul_ps(x5, x2), _mm_mul_ps(x4, x3));
 
   x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
-  x7 = _mm_shuffle2_ps(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0); /* i m m m */
+  x7 = glmm_shuff2(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0);     /* i m m m */
 
   /* t1[3] = i * p - m * l;
      t1[3] = i * p - m * l;
@@ -344,10 +340,10 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
      t3[5] = e * j - i * f; */
   t5 = _mm_sub_ps(_mm_mul_ps(x6, x4), _mm_mul_ps(x7, x5));
 
-  x0 = _mm_shuffle2_ps(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
-  x1 = _mm_shuffle2_ps(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
-  x2 = _mm_shuffle2_ps(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
-  x3 = _mm_shuffle2_ps(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
+  x0 = glmm_shuff2(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
+  x1 = glmm_shuff2(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
+  x2 = glmm_shuff2(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
+  x3 = glmm_shuff2(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
 
   /*
    dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
@@ -395,14 +391,14 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
   x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
 
   x0 = _mm_mul_ps(x0, r0);
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
-  x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 0, 1));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 0, 1));
   x0 = _mm_div_ps(_mm_set1_ps(1.0f), x0);
 
-  _mm_store_ps(dest[0], _mm_mul_ps(v0, x0));
-  _mm_store_ps(dest[1], _mm_mul_ps(v1, x0));
-  _mm_store_ps(dest[2], _mm_mul_ps(v2, x0));
-  _mm_store_ps(dest[3], _mm_mul_ps(v3, x0));
+  glmm_store(dest[0], _mm_mul_ps(v0, x0));
+  glmm_store(dest[1], _mm_mul_ps(v1, x0));
+  glmm_store(dest[2], _mm_mul_ps(v2, x0));
+  glmm_store(dest[3], _mm_mul_ps(v3, x0));
 }
 
 #endif

include/cglm/simd/sse2/quat.h

@@ -24,21 +24,21 @@ 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);
+  r  = _mm_mul_ps(glmm_shuff1x(xp, 3), xq);
 
-  x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 0), _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
-  r  = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 0, 1, 2, 3)));
+  x0 = _mm_xor_ps(glmm_shuff1x(xp, 0), _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
+  r  = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 0, 1, 2, 3)));
 
-  x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 1), _mm_set_ps(-0.f, -0.f, 0.f, 0.f));
-  r  = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 1, 0, 3, 2)));
+  x0 = _mm_xor_ps(glmm_shuff1x(xp, 1), _mm_set_ps(-0.f, -0.f, 0.f, 0.f));
+  r  = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 1, 0, 3, 2)));
 
-  x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 2), _mm_set_ps(-0.f, 0.f, 0.f, -0.f));
-  r  = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 2, 3, 0, 1)));
+  x0 = _mm_xor_ps(glmm_shuff1x(xp, 2), _mm_set_ps(-0.f, 0.f, 0.f, -0.f));
+  r  = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 2, 3, 0, 1)));
 
-  _mm_store_ps(dest, r);
+  glmm_store(dest, r);
 }
 
 

include/cglm/types.h

@@ -9,23 +9,35 @@
 #define cglm_types_h
 
 #if defined(_MSC_VER)
-#  define CGLM_ALIGN(X) /* __declspec(align(X)) */
+/* do not use alignment for older visual studio versions */
+#if _MSC_VER < 1913 /*  Visual Studio 2017 version 15.6  */
+#  define CGLM_ALL_UNALIGNED
+#  define CGLM_ALIGN(X) /* no alignment */
+#else
+#  define CGLM_ALIGN(X) __declspec(align(X))
+#endif
 #else
 #  define CGLM_ALIGN(X) __attribute((aligned(X)))
 #endif
 
-typedef float vec2[2];
-typedef  CGLM_ALIGN(8) float vec3[3];
-typedef int  ivec3[3];
-typedef CGLM_ALIGN(16) float vec4[4];
+#ifndef CGLM_ALL_UNALIGNED
+#  define CGLM_ALIGN_IF(X) CGLM_ALIGN(X)
+#else
+#  define CGLM_ALIGN_IF(X) /* no alignment */
+#endif
 
-typedef vec3 mat3[3];
-typedef CGLM_ALIGN(16) vec4 mat4[4];
+typedef float                   vec2[2];
+typedef CGLM_ALIGN_IF(8)  float vec3[3];
+typedef int                    ivec3[3];
+typedef CGLM_ALIGN_IF(16) float vec4[4];
 
-typedef vec4 versor;
+typedef vec3                    mat3[3];
+typedef CGLM_ALIGN_IF(16) vec4  mat4[4];
 
-#define CGLM_PI    (float)M_PI
-#define CGLM_PI_2  (float)M_PI_2
-#define CGLM_PI_4  (float)M_PI_4
+typedef vec4                    versor;
+
+#define CGLM_PI    ((float)M_PI)
+#define CGLM_PI_2  ((float)M_PI_2)
+#define CGLM_PI_4  ((float)M_PI_4)
 
 #endif /* cglm_types_h */

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)));
+  glmm_store(d, _mm_mul_ps(glmm_load(a), glmm_load(b)));
 #else
   d[0] = a[0] * b[0];
   d[1] = a[1] * b[1];
@@ -61,7 +61,7 @@ CGLM_INLINE
 void
 glm_vec4_broadcast(float val, vec4 d) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(d, _mm_set1_ps(val));
+  glmm_store(d, _mm_set1_ps(val));
 #else
   d[0] = d[1] = d[2] = d[3] = val;
 #endif
@@ -223,14 +223,14 @@ 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);
+  x2 = glmm_shuff1x(x1, 2);
 
-  x3 = _mm_and_ps(_mm_cmpgt_ps(x0, x2), _mm_shuffle1_ps1(x1, 1));
-  x4 = _mm_and_ps(_mm_cmplt_ps(x0, x2), _mm_shuffle1_ps1(x1, 0));
+  x3 = _mm_and_ps(_mm_cmpgt_ps(x0, x2), glmm_shuff1x(x1, 1));
+  x4 = _mm_and_ps(_mm_cmplt_ps(x0, x2), glmm_shuff1x(x1, 0));
 
-  _mm_store_ps(dest, _mm_or_ps(x3, x4));
+  glmm_store(dest, _mm_or_ps(x3, x4));
 #else
   dest[0] = glm_signf(v[0]);
   dest[1] = glm_signf(v[1]);
@@ -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)));
+  glmm_store(dest, _mm_sqrt_ps(glmm_load(v)));
 #else
   dest[0] = sqrtf(v[0]);
   dest[1] = sqrtf(v[1]);

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));
+  glmm_store(dest, glmm_load(v));
 #else
   dest[0] = v[0];
   dest[1] = v[1];
@@ -129,7 +129,7 @@ CGLM_INLINE
 void
 glm_vec4_zero(vec4 v) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(v, _mm_setzero_ps());
+  glmm_store(v, _mm_setzero_ps());
 #else
   v[0] = 0.0f;
   v[1] = 0.0f;
@@ -147,7 +147,7 @@ CGLM_INLINE
 void
 glm_vec4_one(vec4 v) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
-  _mm_store_ps(v, _mm_set1_ps(1.0f));
+  glmm_store(v, _mm_set1_ps(1.0f));
 #else
   v[0] = 1.0f;
   v[1] = 1.0f;
@@ -169,9 +169,9 @@ 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_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)));
+  x0 = _mm_mul_ps(glmm_load(a), glmm_load(b));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
+  return _mm_cvtss_f32(_mm_add_ss(x0, glmm_shuff1(x0, 0, 1, 0, 1)));
 #else
   return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3];
 #endif
@@ -193,10 +193,10 @@ 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)));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
+  return _mm_cvtss_f32(_mm_add_ss(x0, glmm_shuff1(x0, 0, 1, 0, 1)));
 #else
   return v[0] * v[0] + v[1] * v[1] + v[2] * v[2] + v[3] * v[3];
 #endif
@@ -214,8 +214,8 @@ float
 glm_vec4_norm(vec4 vec) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   __m128 x0;
-  x0 = _mm_load_ps(vec);
-  return _mm_cvtss_f32(_mm_sqrt_ss(glm_simd_dot(x0, x0)));
+  x0 = glmm_load(vec);
+  return _mm_cvtss_f32(_mm_sqrt_ss(glmm_dot(x0, x0)));
 #else
   return sqrtf(glm_vec4_norm2(vec));
 #endif
@@ -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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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)));
+  glmm_store(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))));
+  glmm_store(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))));
+  glmm_store(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))));
+  glmm_store(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];
@@ -485,9 +485,9 @@ 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_set1_ps(s))));
+  glmm_store(dest, _mm_add_ps(glmm_load(dest),
+                              _mm_mul_ps(glmm_load(a),
+                                         _mm_set1_ps(s))));
 #else
   dest[0] += a[0] * s;
   dest[1] += a[1] * 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)));
+  glmm_store(v, _mm_xor_ps(glmm_load(v), _mm_set1_ps(-0.0f)));
 #else
   v[0] = -v[0];
   v[1] = -v[1];
@@ -524,8 +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_set1_ps(-0.0f)));
+  glmm_store(dest, _mm_xor_ps(glmm_load(v), _mm_set1_ps(-0.0f)));
 #else
   dest[0] = -v[0];
   dest[1] = -v[1];
@@ -571,16 +570,16 @@ glm_vec4_normalize_to(vec4 vec, vec4 dest) {
   __m128 xdot, x0;
   float  dot;
 
-  x0   = _mm_load_ps(vec);
-  xdot = glm_simd_dot(x0, x0);
+  x0   = glmm_load(vec);
+  xdot = glmm_dot(x0, x0);
   dot  = _mm_cvtss_f32(xdot);
 
   if (dot == 0.0f) {
-    _mm_store_ps(dest, _mm_setzero_ps());
+    glmm_store(dest, _mm_setzero_ps());
     return;
   }
 
-  _mm_store_ps(dest, _mm_div_ps(x0, _mm_sqrt_ps(xdot)));
+  glmm_store(dest, _mm_div_ps(x0, _mm_sqrt_ps(xdot)));
 #else
   float norm;
 
@@ -633,7 +632,7 @@ 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)));
+  glmm_store(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 +652,7 @@ 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)));
+  glmm_store(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]);
@@ -673,8 +672,8 @@ 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_set1_ps(maxVal)));
+  glmm_store(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);
   v[1] = glm_clamp(v[1], minVal, maxVal);

include/cglm/version.h

@@ -10,6 +10,6 @@
 
 #define CGLM_VERSION_MAJOR 0
 #define CGLM_VERSION_MINOR 4
-#define CGLM_VERSION_PATCH 3
+#define CGLM_VERSION_PATCH 5
 
 #endif /* cglm_version_h */