wip

20 files changed, 3430 insertions, 0 deletions

diff --git a/include/cglm/simd/arm.h b/include/cglm/simd/arm.h
new file mode 100644
index 0000000..9f51742
--- /dev/null
+++ b/include/cglm/simd/arm.h
@@ -0,0 +1,206 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_simd_arm_h
+#define cglm_simd_arm_h
+#include "intrin.h"
+#ifdef CGLM_SIMD_ARM
+
+#if defined(_M_ARM64) || defined(_M_HYBRID_X86_ARM64) || defined(_M_ARM64EC) || defined(__aarch64__)
+# define CGLM_ARM64 1
+#else
+# define CGLM_ARM64 0
+#endif
+
+#define glmm_load(p)      vld1q_f32(p)
+#define glmm_store(p, a)  vst1q_f32(p, a)
+
+#define glmm_set1(x)      vdupq_n_f32(x)
+#define glmm_set1_ptr(x)  vdupq_n_f32(*x)
+#define glmm_set1_rval(x) vdupq_n_f32(x)
+#define glmm_128          float32x4_t
+
+#define glmm_splat_x(x) vdupq_lane_f32(vget_low_f32(x),  0)
+#define glmm_splat_y(x) vdupq_lane_f32(vget_low_f32(x),  1)
+#define glmm_splat_z(x) vdupq_lane_f32(vget_high_f32(x), 0)
+#define glmm_splat_w(x) vdupq_lane_f32(vget_high_f32(x), 1)
+
+#define glmm_xor(a, b)                                                        \
+  vreinterpretq_f32_s32(veorq_s32(vreinterpretq_s32_f32(a),                   \
+                                  vreinterpretq_s32_f32(b)))
+
+#define glmm_swplane(v) vextq_f32(v, v, 2)
+#define glmm_low(x)     vget_low_f32(x)
+#define glmm_high(x)    vget_high_f32(x)
+
+#define glmm_combine_ll(x, y) vcombine_f32(vget_low_f32(x),  vget_low_f32(y))
+#define glmm_combine_hl(x, y) vcombine_f32(vget_high_f32(x), vget_low_f32(y))
+#define glmm_combine_lh(x, y) vcombine_f32(vget_low_f32(x),  vget_high_f32(y))
+#define glmm_combine_hh(x, y) vcombine_f32(vget_high_f32(x), vget_high_f32(y))
+
+#if defined(_WIN32) && defined(_MSC_VER)
+/* #  define glmm_float32x4_init(x, y, z, w) { .n128_f32 = { x, y, z, w } } */
+CGLM_INLINE
+float32x4_t
+glmm_float32x4_init(float x, float y, float z, float w) {
+  CGLM_ALIGN(16) float v[4] = {x, y, z, w};
+  return vld1q_f32(v);
+}
+#else
+#  define glmm_float32x4_init(x, y, z, w) { x, y, z, w }
+#endif
+
+#define glmm_float32x4_SIGNMASK_PNPN glmm_float32x4_init( 0.f, -0.f,  0.f, -0.f)
+#define glmm_float32x4_SIGNMASK_NPNP glmm_float32x4_init(-0.f,  0.f, -0.f,  0.f)
+#define glmm_float32x4_SIGNMASK_NPPN glmm_float32x4_init(-0.f,  0.f,  0.f, -0.f)
+
+static inline float32x4_t glmm_abs(float32x4_t v)                { return vabsq_f32(v);    }
+static inline float32x4_t glmm_min(float32x4_t a, float32x4_t b) { return vminq_f32(a, b); }
+static inline float32x4_t glmm_max(float32x4_t a, float32x4_t b) { return vmaxq_f32(a, b); }
+
+static inline
+float32x4_t
+glmm_vhadd(float32x4_t v) {
+#if CGLM_ARM64
+  float32x4_t p;
+  p = vpaddq_f32(v, v); /* [a+b, c+d, a+b, c+d] */
+  return vpaddq_f32(p, p); /* [t, t, t, t] */;
+#else
+  return vaddq_f32(vaddq_f32(glmm_splat_x(v), glmm_splat_y(v)),
+                   vaddq_f32(glmm_splat_z(v), glmm_splat_w(v)));
+#endif
+  /* TODO: measure speed of this compare to above */
+  /* return vdupq_n_f32(vaddvq_f32(v)); */
+
+  /*
+  return vaddq_f32(vaddq_f32(glmm_splat_x(v), glmm_splat_y(v)),
+                   vaddq_f32(glmm_splat_z(v), glmm_splat_w(v)));
+   */
+  /*
+   this seems slower:
+   v = vaddq_f32(v, vrev64q_f32(v));
+   return vaddq_f32(v, vcombine_f32(vget_high_f32(v), vget_low_f32(v)));
+   */
+}
+
+static inline
+float
+glmm_hadd(float32x4_t v) {
+#if CGLM_ARM64
+  return vaddvq_f32(v);
+#else
+  v = vaddq_f32(v, vrev64q_f32(v));
+  v = vaddq_f32(v, vcombine_f32(vget_high_f32(v), vget_low_f32(v)));
+  return vgetq_lane_f32(v, 0);
+#endif
+}
+
+static inline
+float
+glmm_hmin(float32x4_t v) {
+  float32x2_t t;
+  t = vpmin_f32(vget_low_f32(v), vget_high_f32(v));
+  t = vpmin_f32(t, t);
+  return vget_lane_f32(t, 0);
+}
+
+static inline
+float
+glmm_hmax(float32x4_t v) {
+  float32x2_t t;
+  t = vpmax_f32(vget_low_f32(v), vget_high_f32(v));
+  t = vpmax_f32(t, t);
+  return vget_lane_f32(t, 0);
+}
+
+static inline
+float
+glmm_dot(float32x4_t a, float32x4_t b) {
+  return glmm_hadd(vmulq_f32(a, b));
+}
+
+static inline
+float32x4_t
+glmm_vdot(float32x4_t a, float32x4_t b) {
+  return glmm_vhadd(vmulq_f32(a, b));
+}
+
+static inline
+float
+glmm_norm(float32x4_t a) {
+  return sqrtf(glmm_dot(a, a));
+}
+
+static inline
+float
+glmm_norm2(float32x4_t a) {
+  return glmm_dot(a, a);
+}
+
+static inline
+float
+glmm_norm_one(float32x4_t a) {
+  return glmm_hadd(glmm_abs(a));
+}
+
+static inline
+float
+glmm_norm_inf(float32x4_t a) {
+  return glmm_hmax(glmm_abs(a));
+}
+
+static inline
+float32x4_t
+glmm_div(float32x4_t a, float32x4_t b) {
+#if CGLM_ARM64
+  return vdivq_f32(a, b);
+#else
+  /* 2 iterations of Newton-Raphson refinement of reciprocal */
+  float32x4_t r0, r1;
+  r0 = vrecpeq_f32(b);
+  r1 = vrecpsq_f32(r0, b);
+  r0 = vmulq_f32(r1, r0);
+  r1 = vrecpsq_f32(r0, b);
+  r0 = vmulq_f32(r1, r0);
+  return vmulq_f32(a, r0);
+#endif
+}
+
+static inline
+float32x4_t
+glmm_fmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
+#if CGLM_ARM64
+  return vfmaq_f32(c, a, b); /* why vfmaq_f32 is slower than vmlaq_f32 ??? */
+#else
+  return vmlaq_f32(c, a, b);
+#endif
+}
+
+static inline
+float32x4_t
+glmm_fnmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
+#if CGLM_ARM64
+  return vfmsq_f32(c, a, b);
+#else
+  return vmlsq_f32(c, a, b);
+#endif
+}
+
+static inline
+float32x4_t
+glmm_fmsub(float32x4_t a, float32x4_t b, float32x4_t c) {
+  return glmm_fmadd(a, b, vnegq_f32(c));
+}
+
+static inline
+float32x4_t
+glmm_fnmsub(float32x4_t a, float32x4_t b, float32x4_t c) {
+  return vsubq_f32(vdupq_n_f32(0.0f), glmm_fmadd(a, b, c));
+}
+
+#endif
+#endif /* cglm_simd_arm_h */
diff --git a/include/cglm/simd/avx/affine.h b/include/cglm/simd/avx/affine.h
new file mode 100644
index 0000000..b02ff0c
--- /dev/null
+++ b/include/cglm/simd/avx/affine.h
@@ -0,0 +1,66 @@
+/*
+ * 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_avx_h
+#define cglm_affine_mat_avx_h
+#ifdef __AVX__
+
+#include "../../common.h"
+#include "../intrin.h"
+
+#include <immintrin.h>
+
+CGLM_INLINE
+void
+glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  __m256 y0, y1, y2, y3, y4, y5, y6, y7, y8, y9;
+
+  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 = 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 */
+
+  /* 0x03: 0b00000011 */
+  y4 = _mm256_permute2f128_ps(y2, y2, 0x03); /* d c b a h g f e */
+  y5 = _mm256_permute2f128_ps(y3, y3, 0x03); /* l k j i p o n m */
+
+  /* f f f f a a a a */
+  /* h h h h c c c c */
+  /* e e e e b b b b */
+  /* g g g g d d d d */
+  y6 = _mm256_permutevar_ps(y0, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
+  y7 = _mm256_permutevar_ps(y0, _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2));
+  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));
+
+  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 */
+  /* m m m m j j j j */
+  /* o o o o l l l l */
+  y6 = _mm256_permutevar_ps(y1, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
+  y7 = _mm256_permutevar_ps(y1, _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2));
+  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));
+
+  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
+#endif /* cglm_affine_mat_avx_h */
diff --git a/include/cglm/simd/avx/mat4.h b/include/cglm/simd/avx/mat4.h
new file mode 100644
index 0000000..33771c2
--- /dev/null
+++ b/include/cglm/simd/avx/mat4.h
@@ -0,0 +1,115 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat_simd_avx_h
+#define cglm_mat_simd_avx_h
+#ifdef __AVX__
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat4_scale_avx(mat4 m, float s) {
+  __m256 y0, y1, y2, y3, y4;
+
+  y0 = glmm_load256(m[0]);            /* h g f e d c b a */
+  y1 = glmm_load256(m[2]);            /* p o n m l k j i */
+
+  y2 = _mm256_broadcast_ss(&s);
+
+  y3 = _mm256_mul_ps(y0, y2);
+  y4 = _mm256_mul_ps(y1, y2);
+
+  glmm_store256(m[0], y3);
+  glmm_store256(m[2], y4);
+}
+
+/* TODO: this must be tested and compared to SSE version, may be slower!!! */
+CGLM_INLINE
+void
+glm_mat4_transp_avx(mat4 m, mat4 dest) {
+  __m256 y0, y1, y2, y3;
+
+  y0 = glmm_load256(m[0]);                   /* h g f e d c b a */
+  y1 = glmm_load256(m[2]);                   /* p o n m l k j i */
+
+  y2 = _mm256_unpacklo_ps(y0, y1);           /* n f m e j b i a */
+  y3 = _mm256_unpackhi_ps(y0, y1);           /* p h o g l d k c */
+  
+  y0 = _mm256_permute2f128_ps(y2, y3, 0x20); /* l d k c j b i a */
+  y1 = _mm256_permute2f128_ps(y2, y3, 0x31); /* p h o g n f m e */
+
+  y2 = _mm256_unpacklo_ps(y0, y1);           /* o k g c m i e a */
+  y3 = _mm256_unpackhi_ps(y0, y1);           /* p l h d n j f b */
+
+  y0 = _mm256_permute2f128_ps(y2, y3, 0x20); /* n j f b m i e a */
+  y1 = _mm256_permute2f128_ps(y2, y3, 0x31); /* p l h d o k g c */
+
+  glmm_store256(dest[0], y0);
+  glmm_store256(dest[2], y1);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  __m256  y0, y1, y2, y3, y4, y5, y6, y7, y8, y9, y10, y11, y12, y13;
+  __m256i yi0, yi1, yi2, yi3;
+
+  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 = 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 */
+
+  /* 0x03: 0b00000011 */
+  y4 = _mm256_permute2f128_ps(y2, y2, 0x03); /* d c b a h g f e */
+  y5 = _mm256_permute2f128_ps(y3, y3, 0x03); /* l k j i p o n m */
+
+  yi0 = _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0);
+  yi1 = _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2);
+  yi2 = _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1);
+  yi3 = _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3);
+  
+  /* f f f f a a a a */
+  /* h h h h c c c c */
+  /* e e e e b b b b */
+  /* g g g g d d d d */
+  y6 = _mm256_permutevar_ps(y0, yi0);
+  y7 = _mm256_permutevar_ps(y0, yi1);
+  y8 = _mm256_permutevar_ps(y0, yi2);
+  y9 = _mm256_permutevar_ps(y0, yi3);
+
+  /* n n n n i i i i */
+  /* p p p p k k k k */
+  /* m m m m j j j j */
+  /* o o o o l l l l */
+  y10 = _mm256_permutevar_ps(y1, yi0);
+  y11 = _mm256_permutevar_ps(y1, yi1);
+  y12 = _mm256_permutevar_ps(y1, yi2);
+  y13 = _mm256_permutevar_ps(y1, yi3);
+
+  y0 = _mm256_mul_ps(y2, y6);
+  y1 = _mm256_mul_ps(y2, y10);
+
+  y0 = glmm256_fmadd(y3, y7, y0);
+  y1 = glmm256_fmadd(y3, y11, y1);
+
+  y0 = glmm256_fmadd(y4, y8, y0);
+  y1 = glmm256_fmadd(y4, y12, y1);
+
+  y0 = glmm256_fmadd(y5, y9, y0);
+  y1 = glmm256_fmadd(y5, y13, y1);
+
+  glmm_store256(dest[0], y0);
+  glmm_store256(dest[2], y1);
+}
+
+#endif
+#endif /* cglm_mat_simd_avx_h */
diff --git a/include/cglm/simd/intrin.h b/include/cglm/simd/intrin.h
new file mode 100644
index 0000000..c477f34
--- /dev/null
+++ b/include/cglm/simd/intrin.h
@@ -0,0 +1,153 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_intrin_h
+#define cglm_intrin_h
+
+#if defined(_MSC_VER) && !defined(_M_ARM64EC)
+#  if (defined(_M_AMD64) || defined(_M_X64)) || _M_IX86_FP == 2
+#    ifndef __SSE__
+#      define __SSE__
+#    endif
+#    ifndef __SSE2__
+#      define __SSE2__
+#    endif
+#  elif _M_IX86_FP == 1
+#    ifndef __SSE__
+#      define __SSE__
+#    endif
+#  endif
+/* do not use alignment for older visual studio versions */
+/* also ARM32 also causes similar error, disable it for now on ARM32 too */
+#  if _MSC_VER < 1913 || _M_ARM     /* Visual Studio 2017 version 15.6 */
+#    define CGLM_ALL_UNALIGNED
+#  endif
+#endif
+
+#ifdef __AVX__
+#  include <immintrin.h>
+#  define CGLM_AVX_FP 1
+#    ifndef __SSE2__
+#      define __SSE2__
+#    endif
+#    ifndef __SSE3__
+#      define __SSE3__
+#    endif
+#    ifndef __SSE4__
+#      define __SSE4__
+#    endif
+#    ifndef __SSE4_1__
+#      define __SSE4_1__
+#    endif
+#    ifndef __SSE4_2__
+#      define __SSE4_2__
+#    endif
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+#if defined(__SSE__)
+#  include <xmmintrin.h>
+#  define CGLM_SSE_FP 1
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+#if defined(__SSE2__)
+#  include <emmintrin.h>
+#  define CGLM_SSE2_FP 1
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+#if defined(__SSE3__)
+#  include <pmmintrin.h>
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+#if defined(__SSE4_1__)
+#  include <smmintrin.h>
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+#if defined(__SSE4_2__)
+#  include <nmmintrin.h>
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
+/* ARM Neon */
+#if defined(_WIN32) && defined(_MSC_VER)
+/* TODO: non-ARM stuff already inported, will this be better option */
+/* #  include <intrin.h> */
+
+#  if defined(_M_ARM64) || defined(_M_HYBRID_X86_ARM64) || defined(_M_ARM64EC)
+#    include <arm64intr.h>
+#    include <arm64_neon.h>
+#    ifndef CGLM_NEON_FP
+#      define CGLM_NEON_FP  1
+#    endif
+#    ifndef CGLM_SIMD_ARM
+#      define CGLM_SIMD_ARM
+#    endif
+#  elif defined(_M_ARM)
+#    include <armintr.h>
+#    include <arm_neon.h>
+#    ifndef CGLM_NEON_FP
+#      define CGLM_NEON_FP 1
+#    endif
+#    ifndef CGLM_SIMD_ARM
+#      define CGLM_SIMD_ARM
+#    endif
+#  endif
+
+#else /* non-windows */
+#  if defined(__ARM_NEON) || defined(__ARM_NEON__)
+#    include <arm_neon.h>
+#    if defined(__ARM_NEON_FP) || defined(__ARM_FP)
+#      define CGLM_NEON_FP 1
+#    endif
+#    ifndef CGLM_SIMD_ARM
+#      define CGLM_SIMD_ARM
+#    endif
+#  endif
+#endif
+
+/* WebAssembly */
+#if defined(__wasm__) && defined(__wasm_simd128__)
+#  ifndef CGLM_SIMD_WASM
+#    define CGLM_SIMD_WASM
+#  endif
+#endif
+
+#if defined(CGLM_SIMD_x86) || defined(CGLM_SIMD_ARM) || defined(CGLM_SIMD_WASM)
+#  ifndef CGLM_SIMD
+#    define CGLM_SIMD
+#  endif
+#endif
+
+#if defined(CGLM_SIMD_x86) && !defined(CGLM_SIMD_WASM)
+#  include "x86.h"
+#endif
+
+#if defined(CGLM_SIMD_ARM)
+#  include "arm.h"
+#endif
+
+#if defined(CGLM_SIMD_WASM)
+#  include "wasm.h"
+#endif
+
+#endif /* cglm_intrin_h */
diff --git a/include/cglm/simd/neon/affine.h b/include/cglm/simd/neon/affine.h
new file mode 100644
index 0000000..b0a65a6
--- /dev/null
+++ b/include/cglm/simd/neon/affine.h
@@ -0,0 +1,121 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_affine_neon_h
+#define cglm_affine_neon_h
+#if defined(CGLM_NEON_FP)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mul_neon(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = vmulq_f32(glmm_splat_x(r0), l);
+  v1 = vmulq_f32(glmm_splat_x(r1), l);
+  v2 = vmulq_f32(glmm_splat_x(r2), l);
+  v3 = vmulq_f32(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  v3 = glmm_fmadd(glmm_splat_w(r3), glmm_load(m1[3]), v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mul_rot_neon(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, v0, v1, v2;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+
+  v0 = vmulq_f32(glmm_splat_x(r0), l);
+  v1 = vmulq_f32(glmm_splat_x(r1), l);
+  v2 = vmulq_f32(glmm_splat_x(r2), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], glmm_load(m1[3]));
+}
+
+CGLM_INLINE
+void
+glm_inv_tr_neon(mat4 mat) {
+  float32x4x4_t vmat;
+  glmm_128      r0, r1, r2, x0;
+
+  vmat = vld4q_f32(mat[0]);
+  r0   = vmat.val[0];
+  r1   = vmat.val[1];
+  r2   = vmat.val[2];
+
+  x0 = glmm_fmadd(r0, glmm_splat_w(r0),
+                  glmm_fmadd(r1, glmm_splat_w(r1),
+                             vmulq_f32(r2, glmm_splat_w(r2))));
+  x0 = vnegq_f32(x0);
+
+  glmm_store(mat[0], r0);
+  glmm_store(mat[1], r1);
+  glmm_store(mat[2], r2);
+  glmm_store(mat[3], x0);
+  
+  mat[0][3] = 0.0f;
+  mat[1][3] = 0.0f;
+  mat[2][3] = 0.0f;
+  mat[3][3] = 1.0f;
+
+  /* TODO: ?
+  zo   = vget_high_f32(r3);
+  vst1_lane_f32(&mat[0][3], zo, 0);
+  vst1_lane_f32(&mat[1][3], zo, 0);
+  vst1_lane_f32(&mat[2][3], zo, 0);
+  vst1_lane_f32(&mat[3][3], zo, 1);
+  */
+}
+
+#endif
+#endif /* cglm_affine_neon_h */
diff --git a/include/cglm/simd/neon/mat2.h b/include/cglm/simd/neon/mat2.h
new file mode 100644
index 0000000..7d0d9eb
--- /dev/null
+++ b/include/cglm/simd/neon/mat2.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat2_neon_h
+#define cglm_mat2_neon_h
+#if defined(CGLM_NEON_FP)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat2_mul_neon(mat2 m1, mat2 m2, mat2 dest) {
+  float32x4x2_t a1;
+  glmm_128 x0,  x1, x2;
+  float32x2_t   dc, ba;
+
+  x1 = glmm_load(m1[0]); /* d c b a */
+  x2 = glmm_load(m2[0]); /* h g f e */
+  
+  dc = vget_high_f32(x1);
+  ba = vget_low_f32(x1);
+
+  /* g g e e, h h f f */
+  a1 = vtrnq_f32(x2, x2);
+  
+  /*
+   dest[0][0] = a * e + c * f;
+   dest[0][1] = b * e + d * f;
+   dest[1][0] = a * g + c * h;
+   dest[1][1] = b * g + d * h;
+   */
+  x0 = glmm_fmadd(vcombine_f32(ba, ba), a1.val[0],
+                  vmulq_f32(vcombine_f32(dc, dc), a1.val[1]));
+
+  glmm_store(dest[0], x0);
+}
+
+#endif
+#endif /* cglm_mat2_neon_h */
diff --git a/include/cglm/simd/neon/mat4.h b/include/cglm/simd/neon/mat4.h
new file mode 100644
index 0000000..6cf9811
--- /dev/null
+++ b/include/cglm/simd/neon/mat4.h
@@ -0,0 +1,468 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat4_neon_h
+#define cglm_mat4_neon_h
+#if defined(CGLM_NEON_FP)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat4_scale_neon(mat4 m, float s) {
+  float32x4_t v0;
+  
+  v0 = vdupq_n_f32(s);
+
+  vst1q_f32(m[0], vmulq_f32(vld1q_f32(m[0]), v0));
+  vst1q_f32(m[1], vmulq_f32(vld1q_f32(m[1]), v0));
+  vst1q_f32(m[2], vmulq_f32(vld1q_f32(m[2]), v0));
+  vst1q_f32(m[3], vmulq_f32(vld1q_f32(m[3]), v0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_transp_neon(mat4 m, mat4 dest) {
+  float32x4x4_t vmat;
+  
+  vmat = vld4q_f32(m[0]);
+
+  vst1q_f32(dest[0], vmat.val[0]);
+  vst1q_f32(dest[1], vmat.val[1]);
+  vst1q_f32(dest[2], vmat.val[2]);
+  vst1q_f32(dest[3], vmat.val[3]);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mul_neon(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = vmulq_f32(glmm_splat_x(r0), l);
+  v1 = vmulq_f32(glmm_splat_x(r1), l);
+  v2 = vmulq_f32(glmm_splat_x(r2), l);
+  v3 = vmulq_f32(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  l  = glmm_load(m1[3]);
+  v0 = glmm_fmadd(glmm_splat_w(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_w(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_w(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mulv_neon(mat4 m, vec4 v, vec4 dest) {
+  float32x4_t l0, l1, l2, l3;
+  float32x2_t vlo, vhi;
+  
+  l0  = vld1q_f32(m[0]);
+  l1  = vld1q_f32(m[1]);
+  l2  = vld1q_f32(m[2]);
+  l3  = vld1q_f32(m[3]);
+
+  vlo = vld1_f32(&v[0]);
+  vhi = vld1_f32(&v[2]);
+
+  l0  = vmulq_lane_f32(l0, vlo, 0);
+  l0  = vmlaq_lane_f32(l0, l1, vlo, 1);
+  l0  = vmlaq_lane_f32(l0, l2, vhi, 0);
+  l0  = vmlaq_lane_f32(l0, l3, vhi, 1);
+
+  vst1q_f32(dest, l0);
+}
+
+CGLM_INLINE
+float
+glm_mat4_det_neon(mat4 mat) {
+  float32x4_t   r0, r1, r2, r3, x0, x1, x2;
+  float32x2_t   ij, op, mn, kl, nn, mm, jj, ii, gh, ef, t12, t34;
+  float32x4x2_t a1;
+  float32x4_t   x3 = glmm_float32x4_SIGNMASK_PNPN;
+
+  /* 127 <- 0, [square] det(A) = det(At) */
+  r0 = glmm_load(mat[0]);              /* d c b a */
+  r1 = vrev64q_f32(glmm_load(mat[1])); /* g h e f */
+  r2 = vrev64q_f32(glmm_load(mat[2])); /* l k i j */
+  r3 = vrev64q_f32(glmm_load(mat[3])); /* o p m n */
+
+  gh = vget_high_f32(r1);
+  ef = vget_low_f32(r1);
+  kl = vget_high_f32(r2);
+  ij = vget_low_f32(r2);
+  op = vget_high_f32(r3);
+  mn = vget_low_f32(r3);
+  mm = vdup_lane_f32(mn, 1);
+  nn = vdup_lane_f32(mn, 0);
+  ii = vdup_lane_f32(ij, 1);
+  jj = vdup_lane_f32(ij, 0);
+  
+  /*
+   t[1] = j * p - n * l;
+   t[2] = j * o - n * k;
+   t[3] = i * p - m * l;
+   t[4] = i * o - m * k;
+   */
+  x0 = glmm_fnmadd(vcombine_f32(kl, kl), vcombine_f32(nn, mm),
+                   vmulq_f32(vcombine_f32(op, op), vcombine_f32(jj, ii)));
+
+  t12 = vget_low_f32(x0);
+  t34 = vget_high_f32(x0);
+  
+  /* 1 3 1 3 2 4 2 4 */
+  a1 = vuzpq_f32(x0, x0);
+  
+  /*
+   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 = glmm_fnmadd(vcombine_f32(vdup_lane_f32(kl, 0), jj),
+                   vcombine_f32(vdup_lane_f32(op, 1), mm),
+                   vmulq_f32(vcombine_f32(vdup_lane_f32(op, 0), nn),
+                             vcombine_f32(vdup_lane_f32(kl, 1), ii)));
+
+  /*
+     a * (f * t[0] - g * t[1] + h * t[2])
+   - b * (e * t[0] - g * t[3] + h * t[4])
+   + c * (e * t[1] - f * t[3] + h * t[5])
+   - d * (e * t[2] - f * t[4] + g * t[5])
+   */
+  x2 = glmm_fnmadd(vcombine_f32(vdup_lane_f32(gh, 1), vdup_lane_f32(ef, 0)),
+                   vcombine_f32(vget_low_f32(a1.val[0]), t34),
+                   vmulq_f32(vcombine_f32(ef, vdup_lane_f32(ef, 1)),
+                             vcombine_f32(vget_low_f32(x1), t12)));
+
+  x2 = glmm_fmadd(vcombine_f32(vdup_lane_f32(gh, 0), gh),
+                  vcombine_f32(vget_low_f32(a1.val[1]), vget_high_f32(x1)), x2);
+
+  x2 = glmm_xor(x2, x3);
+
+  return glmm_hadd(vmulq_f32(x2, r0));
+}
+
+/* old one */
+#if 0
+CGLM_INLINE
+void
+glm_mat4_inv_neon(mat4 mat, mat4 dest) {
+  float32x4_t   r0, r1, r2, r3,
+                v0, v1, v2, v3,
+                t0, t1, t2, t3, t4, t5,
+                x0, x1, x2, x3, x4, x5, x6, x7, x8;
+  float32x4x2_t a1;
+  float32x2_t   lp, ko, hg, jn, im, fe, ae, bf, cg, dh;
+  float32x4_t   x9 = glmm_float32x4_SIGNMASK_NPNP;
+
+  x8 = vrev64q_f32(x9);
+
+  /* 127 <- 0 */
+  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 */
+  
+  /* l p k o, j n i m */
+  a1  = vzipq_f32(r3, r2);
+  
+  jn  = vget_high_f32(a1.val[0]);
+  im  = vget_low_f32(a1.val[0]);
+  lp  = vget_high_f32(a1.val[1]);
+  ko  = vget_low_f32(a1.val[1]);
+  hg  = vget_high_f32(r1);
+
+  x1  = vcombine_f32(vdup_lane_f32(lp, 0), lp);                   /* l p p p */
+  x2  = vcombine_f32(vdup_lane_f32(ko, 0), ko);                   /* k o o o */
+  x0  = vcombine_f32(vdup_lane_f32(lp, 1), vdup_lane_f32(hg, 1)); /* h h l l */
+  x3  = vcombine_f32(vdup_lane_f32(ko, 1), vdup_lane_f32(hg, 0)); /* g g k k */
+  
+  /* t1[0] = k * p - o * l;
+     t1[0] = k * p - o * l;
+     t2[0] = g * p - o * h;
+     t3[0] = g * l - k * h; */
+  t0 = glmm_fnmadd(x2, x0, vmulq_f32(x3, x1));
+
+  fe = vget_low_f32(r1);
+  x4 = vcombine_f32(vdup_lane_f32(jn, 0), jn);                   /* j n n n */
+  x5 = vcombine_f32(vdup_lane_f32(jn, 1), vdup_lane_f32(fe, 1)); /* f f j j */
+  
+  /* t1[1] = j * p - n * l;
+     t1[1] = j * p - n * l;
+     t2[1] = f * p - n * h;
+     t3[1] = f * l - j * h; */
+   t1 = glmm_fnmadd(x4, x0, vmulq_f32(x5, x1));
+  
+  /* t1[2] = j * o - n * k
+     t1[2] = j * o - n * k;
+     t2[2] = f * o - n * g;
+     t3[2] = f * k - j * g; */
+  t2 = glmm_fnmadd(x4, x3, vmulq_f32(x5, x2));
+  
+  x6 = vcombine_f32(vdup_lane_f32(im, 1), vdup_lane_f32(fe, 0)); /* e e i i */
+  x7 = vcombine_f32(vdup_lane_f32(im, 0), im);                   /* i m m m */
+  
+  /* t1[3] = i * p - m * l;
+     t1[3] = i * p - m * l;
+     t2[3] = e * p - m * h;
+     t3[3] = e * l - i * h; */
+  t3 = glmm_fnmadd(x7, x0, vmulq_f32(x6, x1));
+  
+  /* t1[4] = i * o - m * k;
+     t1[4] = i * o - m * k;
+     t2[4] = e * o - m * g;
+     t3[4] = e * k - i * g; */
+  t4 = glmm_fnmadd(x7, x3, vmulq_f32(x6, x2));
+  
+  /* t1[5] = i * n - m * j;
+     t1[5] = i * n - m * j;
+     t2[5] = e * n - m * f;
+     t3[5] = e * j - i * f; */
+  t5 = glmm_fnmadd(x7, x5, vmulq_f32(x6, x4));
+  
+  /* h d f b, g c e a */
+  a1 = vtrnq_f32(r0, r1);
+  
+  x4 = vrev64q_f32(a1.val[0]); /* c g a e */
+  x5 = vrev64q_f32(a1.val[1]); /* d h b f */
+
+  ae = vget_low_f32(x4);
+  cg = vget_high_f32(x4);
+  bf = vget_low_f32(x5);
+  dh = vget_high_f32(x5);
+  
+  x0 = vcombine_f32(ae, vdup_lane_f32(ae, 1)); /* a a a e */
+  x1 = vcombine_f32(bf, vdup_lane_f32(bf, 1)); /* b b b f */
+  x2 = vcombine_f32(cg, vdup_lane_f32(cg, 1)); /* c c c g */
+  x3 = vcombine_f32(dh, vdup_lane_f32(dh, 1)); /* d d d h */
+  
+  /*
+   dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
+   dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
+   dest[0][2] =  b * t2[0] - c * t2[1] + d * t2[2];
+   dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
+  v0 = glmm_xor(glmm_fmadd(x3, t2, glmm_fnmadd(x2, t1, vmulq_f32(x1, t0))), x8);
+  
+  /*
+   dest[2][0] =  e * t1[1] - f * t1[3] + h * t1[5];
+   dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
+   dest[2][2] =  a * t2[1] - b * t2[3] + d * t2[5];
+   dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
+  v2 = glmm_xor(glmm_fmadd(x3, t5, glmm_fnmadd(x1, t3, vmulq_f32(x0, t1))), x8);
+
+  /*
+   dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
+   dest[1][1] =  a * t1[0] - c * t1[3] + d * t1[4];
+   dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
+   dest[1][3] =  a * t3[0] - c * t3[3] + d * t3[4]; */
+  v1 = glmm_xor(glmm_fmadd(x3, t4, glmm_fnmadd(x2, t3, vmulq_f32(x0, t0))), x9);
+  
+  /*
+   dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
+   dest[3][1] =  a * t1[2] - b * t1[4] + c * t1[5];
+   dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
+   dest[3][3] =  a * t3[2] - b * t3[4] + c * t3[5]; */
+  v3 = glmm_xor(glmm_fmadd(x2, t5, glmm_fnmadd(x1, t4, vmulq_f32(x0, t2))), x9);
+
+  /* determinant */
+  x0 = vcombine_f32(vget_low_f32(vzipq_f32(v0, v1).val[0]),
+                    vget_low_f32(vzipq_f32(v2, v3).val[0]));
+
+  /*
+  x0 = glmm_div(glmm_set1_rval(1.0f), glmm_vhadd(vmulq_f32(x0, r0)));
+
+  glmm_store(dest[0], vmulq_f32(v0, x0));
+  glmm_store(dest[1], vmulq_f32(v1, x0));
+  glmm_store(dest[2], vmulq_f32(v2, x0));
+  glmm_store(dest[3], vmulq_f32(v3, x0));
+  */
+
+  x0 = glmm_vhadd(vmulq_f32(x0, r0));
+
+  glmm_store(dest[0], glmm_div(v0, x0));
+  glmm_store(dest[1], glmm_div(v1, x0));
+  glmm_store(dest[2], glmm_div(v2, x0));
+  glmm_store(dest[3], glmm_div(v3, x0));
+}
+#endif
+
+CGLM_INLINE
+void
+glm_mat4_inv_neon(mat4 mat, mat4 dest) {
+  float32x4_t   r0, r1, r2, r3,
+                v0, v1, v2, v3, v4, v5,
+                t0, t1, t2;
+  float32x4x2_t a0, a1, a2, a3, a4;
+  float32x4_t   s1 = glmm_float32x4_SIGNMASK_PNPN, s2;
+
+#if !CGLM_ARM64
+  float32x2_t   l0, l1;
+#endif
+
+  s2 = vrev64q_f32(s1);
+
+  /* 127 <- 0 */
+  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 */
+
+  a1 = vzipq_f32(r0, r2);                  /* l d k c, j b i a */
+  a2 = vzipq_f32(r1, r3);                  /* p h o g, n f m e */
+  a3 = vzipq_f32(a2.val[0], a1.val[0]);    /* j n b f, i m a e */
+  a4 = vzipq_f32(a2.val[1], a1.val[1]);    /* l p d h, k o c g */
+
+  v0 = vextq_f32(a1.val[0], a1.val[1], 2); /* k c j b */
+  v1 = vextq_f32(a2.val[0], a2.val[1], 2); /* o g n f */
+  v2 = vextq_f32(a1.val[1], a2.val[0], 2); /* m e l d */
+  v3 = vextq_f32(a2.val[1], a1.val[0], 2); /* i a p h */
+  v4 = vextq_f32(v1, v2, 2);               /* l d o g */
+  v5 = vextq_f32(v0, v3, 2);               /* p h k c */
+
+  /* c2 = c * h - g * d   c12 = a * g - c * e   c8  = a * f - b * e
+     c1 = k * p - o * l   c11 = i * o - k * m   c7  = i * n - j * m
+     c4 = h * a - d * e   c6  = b * h - d * f   c10 = b * g - c * f
+     c3 = p * i - l * m   c5  = j * p - l * n   c9  = j * o - k * n */
+  t0 = vmulq_f32(v5, v3);
+  t1 = vmulq_f32(a1.val[0], a2.val[1]);
+  t2 = vmulq_f32(a1.val[0], v1);
+
+  t0 = glmm_fnmadd(v4, v2, t0);
+  t1 = glmm_fnmadd(a1.val[1], a2.val[0], t1);
+  t2 = glmm_fnmadd(v0, a2.val[0], t2);
+
+  t0 = vrev64q_f32(t0);
+  t1 = vrev64q_f32(t1);
+  t2 = vrev64q_f32(t2);
+
+  /* det */
+  v0 = vrev64q_f32(t2);
+  v1 = vextq_f32(t1, t1, 2);
+  v0 = vmulq_f32(t0, v0);
+  v1 = vrev64q_f32(v1);
+  v1 = vmulq_f32(v1, t1);
+
+  /* c3 * c10 + c4 * c9 + c1 * c8 + c2 * c7 */
+#if CGLM_ARM64
+  v0 = vpaddq_f32(v0, v0);
+  v0 = vpaddq_f32(v0, v0);
+#else
+  l0 = vget_low_f32(v0);
+  l1 = vget_high_f32(v0);
+
+  l0 = vpadd_f32(l0, l0); /* [a+b, a+b] */ 
+  l1 = vpadd_f32(l1, l1); /* [c+d, c+d] */ 
+  l0 = vadd_f32(l0, l1);  /* [sum, sum] */ 
+
+  v0 = vcombine_f32(l0, l0); 
+#endif
+
+  /* c5 * c12 + c6 * c11 */
+#if CGLM_ARM64
+  v1 = vpaddq_f32(v1, v1);
+#else
+  l0 = vget_low_f32(v1);
+  l1 = vget_high_f32(v1);
+
+  l0 = vpadd_f32(l0, l0); /* [a+b, a+b] */ 
+  l1 = vpadd_f32(l1, l1); /* [c+d, c+d] */ 
+
+  v1 = vcombine_f32(l0, l1);
+#endif
+
+  v0 = vsubq_f32(v0, v1);    /* det */
+
+  /* inv div */
+  v1 = vdupq_n_f32(1.0f);
+  v0 = glmm_div(v1, v0);     /* inv div */
+
+  /* multiply t0,t1,t2 by idt to reduce 1mul below: 2eor+4mul vs 3mul+4eor */
+  t0 = vmulq_f32(t0, v0);
+  t1 = vmulq_f32(t1, v0);
+  t2 = vmulq_f32(t2, v0);
+
+  a0 = vzipq_f32(t0, t0);    /* c4  c4  c3 c3, c2  c2  c1  c1  */
+  a1 = vzipq_f32(t1, t1);    /* c6  c6  c5 c5, c12 c12 c11 c11 */
+  a2 = vzipq_f32(t2, t2);    /* c10 c10 c9 c9, c8  c8  c7  c7  */
+
+  /* result */
+
+  /* dest[0][0] = (f * c1  - g * c5  + h * c9)  * idt;
+     dest[0][1] = (b * c1  - c * c5  + d * c9)  * ndt;
+     dest[0][2] = (n * c2  - o * c6  + p * c10) * idt;
+     dest[0][3] = (j * c2  - k * c6  + l * c10) * ndt;
+
+     dest[1][0] = (e * c1  - g * c3  + h * c11) * ndt;
+     dest[1][1] = (a * c1  - c * c3  + d * c11) * idt;
+     dest[1][2] = (m * c2  - o * c4  + p * c12) * ndt;
+     dest[1][3] = (i * c2  - k * c4  + l * c12) * idt;
+
+     dest[2][0] = (e * c5  - f * c3  + h * c7)  * idt;
+     dest[2][1] = (a * c5  - b * c3  + d * c7)  * ndt;
+     dest[2][2] = (m * c6  - n * c4  + p * c8)  * idt;
+     dest[2][3] = (i * c6  - j * c4  + l * c8)  * ndt;
+
+     dest[3][0] = (e * c9  - f * c11 + g * c7)  * ndt;
+     dest[3][1] = (a * c9  - b * c11 + c * c7)  * idt;
+     dest[3][2] = (m * c10 - n * c12 + o * c8)  * ndt;
+     dest[3][3] = (i * c10 - j * c12 + k * c8)  * idt; */
+
+  r0 = vmulq_f32(a3.val[1], a0.val[0]);
+  r1 = vmulq_f32(a3.val[0], a0.val[0]);
+  r2 = vmulq_f32(a3.val[0], a1.val[1]);
+  r3 = vmulq_f32(a3.val[0], a2.val[1]);
+
+  r0 = glmm_fnmadd(a4.val[0], a1.val[1], r0);
+  r1 = glmm_fnmadd(a4.val[0], a0.val[1], r1);
+  r2 = glmm_fnmadd(a3.val[1], a0.val[1], r2);
+  r3 = glmm_fnmadd(a3.val[1], a1.val[0], r3);
+
+  r0 = glmm_fmadd(a4.val[1], a2.val[1], r0);
+  r1 = glmm_fmadd(a4.val[1], a1.val[0], r1);
+  r2 = glmm_fmadd(a4.val[1], a2.val[0], r2);
+  r3 = glmm_fmadd(a4.val[0], a2.val[0], r3);
+
+  /* 4xor may be fastart then 4mul, see above  */
+  r0 = glmm_xor(r0, s1);
+  r1 = glmm_xor(r1, s2);
+  r2 = glmm_xor(r2, s1);
+  r3 = glmm_xor(r3, s2);
+
+  glmm_store(dest[0], r0);
+  glmm_store(dest[1], r1);
+  glmm_store(dest[2], r2);
+  glmm_store(dest[3], r3);
+}
+
+#endif
+#endif /* cglm_mat4_neon_h */
diff --git a/include/cglm/simd/neon/quat.h b/include/cglm/simd/neon/quat.h
new file mode 100644
index 0000000..55dc1da
--- /dev/null
+++ b/include/cglm/simd/neon/quat.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_quat_neon_h
+#define cglm_quat_neon_h
+#if defined(CGLM_NEON_FP)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_quat_mul_neon(versor p, versor q, versor dest) {
+  /*
+   + (a1 b2 + b1 a2 + c1 d2 − d1 c2)i
+   + (a1 c2 − b1 d2 + c1 a2 + d1 b2)j
+   + (a1 d2 + b1 c2 − c1 b2 + d1 a2)k
+     a1 a2 − b1 b2 − c1 c2 − d1 d2
+   */
+
+  glmm_128 xp, xq, xqr, r, x, y, z, s2, s3;
+  glmm_128 s1 = glmm_float32x4_SIGNMASK_NPPN;
+
+  float32x2_t   qh, ql;
+  
+  xp  = glmm_load(p); /* 3 2 1 0 */
+  xq  = glmm_load(q);
+
+  r   = vmulq_f32(glmm_splat_w(xp), xq);
+  x   = glmm_splat_x(xp);
+  y   = glmm_splat_y(xp);
+  z   = glmm_splat_z(xp);
+
+  ql  = vget_high_f32(s1);
+  s3  = vcombine_f32(ql, ql);
+  s2  = vzipq_f32(s3, s3).val[0];
+
+  xqr = vrev64q_f32(xq);
+  qh  = vget_high_f32(xqr);
+  ql  = vget_low_f32(xqr);
+
+  r = glmm_fmadd(glmm_xor(x, s3), vcombine_f32(qh, ql), r);
+  
+  r = glmm_fmadd(glmm_xor(y, s2), vcombine_f32(vget_high_f32(xq),
+                                               vget_low_f32(xq)), r);
+  
+  r = glmm_fmadd(glmm_xor(z, s1), vcombine_f32(ql, qh), r);
+
+  glmm_store(dest, r);
+}
+
+#endif
+#endif /* cglm_quat_neon_h */
diff --git a/include/cglm/simd/sse2/affine.h b/include/cglm/simd/sse2/affine.h
new file mode 100644
index 0000000..0619995
--- /dev/null
+++ b/include/cglm/simd/sse2/affine.h
@@ -0,0 +1,115 @@
+/*
+ * 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 "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = _mm_mul_ps(glmm_splat_x(r0), l);
+  v1 = _mm_mul_ps(glmm_splat_x(r1), l);
+  v2 = _mm_mul_ps(glmm_splat_x(r2), l);
+  v3 = _mm_mul_ps(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  l  = glmm_load(m1[3]);
+  v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mul_rot_sse2(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, v0, v1, v2;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+
+  v0 = _mm_mul_ps(glmm_splat_x(r0), l);
+  v1 = _mm_mul_ps(glmm_splat_x(r1), l);
+  v2 = _mm_mul_ps(glmm_splat_x(r2), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], glmm_load(m1[3]));
+}
+
+CGLM_INLINE
+void
+glm_inv_tr_sse2(mat4 mat) {
+  __m128 r0, r1, r2, r3, x0, x1, x2, x3, x4, x5;
+
+  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);
+
+  x2 = glmm_shuff1(r3, 0, 0, 0, 0);
+  x3 = glmm_shuff1(r3, 1, 1, 1, 1);
+  x4 = glmm_shuff1(r3, 2, 2, 2, 2);
+  x5 = glmm_float32x4_SIGNMASK_NEG;
+
+  x0 = glmm_fmadd(r0, x2, glmm_fmadd(r1, x3, _mm_mul_ps(r2, x4)));
+  x0 = _mm_xor_ps(x0, x5);
+
+  x0 = _mm_add_ps(x0, x1);
+
+  glmm_store(mat[0], r0);
+  glmm_store(mat[1], r1);
+  glmm_store(mat[2], r2);
+  glmm_store(mat[3], x0);
+}
+
+#endif
+#endif /* cglm_affine_mat_sse2_h */
diff --git a/include/cglm/simd/sse2/mat2.h b/include/cglm/simd/sse2/mat2.h
new file mode 100644
index 0000000..31b3a29
--- /dev/null
+++ b/include/cglm/simd/sse2/mat2.h
@@ -0,0 +1,48 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat2_sse_h
+#define cglm_mat2_sse_h
+#if defined( __SSE__ ) || defined( __SSE2__ )
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat2_mul_sse2(mat2 m1, mat2 m2, mat2 dest) {
+  __m128 x0, x1, x2, x3, x4;
+
+  x1 = glmm_load(m1[0]); /* d c b a */
+  x2 = glmm_load(m2[0]); /* h g f e */
+
+  x3 = glmm_shuff1(x2, 2, 2, 0, 0);
+  x4 = glmm_shuff1(x2, 3, 3, 1, 1);
+  x0 = _mm_movelh_ps(x1, x1);
+  x2 = _mm_movehl_ps(x1, x1);
+
+  /*
+   dest[0][0] = a * e + c * f;
+   dest[0][1] = b * e + d * f;
+   dest[1][0] = a * g + c * h;
+   dest[1][1] = b * g + d * h;
+   */
+  x0 = glmm_fmadd(x0, x3, _mm_mul_ps(x2, x4));
+
+  glmm_store(dest[0], x0);
+}
+
+CGLM_INLINE
+void
+glm_mat2_transp_sse2(mat2 m, mat2 dest) {
+  /* d c b a */
+  /* d b c a */
+  glmm_store(dest[0], glmm_shuff1(glmm_load(m[0]), 3, 1, 2, 0));
+}
+
+#endif
+#endif /* cglm_mat2_sse_h */
diff --git a/include/cglm/simd/sse2/mat3.h b/include/cglm/simd/sse2/mat3.h
new file mode 100644
index 0000000..f07320c
--- /dev/null
+++ b/include/cglm/simd/sse2/mat3.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat3_sse_h
+#define cglm_mat3_sse_h
+#if defined( __SSE__ ) || defined( __SSE2__ )
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat3_mul_sse2(mat3 m1, mat3 m2, mat3 dest) {
+  __m128 l0, l1, l2, r0, r1, r2, x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+  
+  l0 = _mm_loadu_ps(m1[0]);
+  l1 = _mm_loadu_ps(&m1[1][1]);
+
+  r0 = _mm_loadu_ps(m2[0]);
+  r1 = _mm_loadu_ps(&m2[1][1]);
+
+  x8 = glmm_shuff1(l0, 0, 2, 1, 0);                     /* a00 a02 a01 a00 */
+  x1 = glmm_shuff1(r0, 3, 0, 0, 0);                     /* b10 b00 b00 b00 */
+  x2 = _mm_shuffle_ps(l0, l1, _MM_SHUFFLE(1, 0, 3, 3)); /* a12 a11 a10 a10 */
+  x3 = _mm_shuffle_ps(r0, r1, _MM_SHUFFLE(2, 0, 3, 1)); /* b20 b11 b10 b01 */
+  x0 = _mm_mul_ps(x8, x1);
+
+  x6 = glmm_shuff1(l0, 1, 0, 2, 1);                     /* a01 a00 a02 a01 */
+  x7 = glmm_shuff1(x3, 3, 3, 1, 1);                     /* b20 b20 b10 b10 */
+  l2 = _mm_load_ss(&m1[2][2]);
+  r2 = _mm_load_ss(&m2[2][2]);
+  x1 = _mm_mul_ps(x6, x7);
+  l2 = glmm_shuff1(l2, 0, 0, 1, 0);                     /* a22 a22 0.f a22 */
+  r2 = glmm_shuff1(r2, 0, 0, 1, 0);                     /* b22 b22 0.f b22 */
+
+  x4 = glmm_shuff1(x2, 0, 3, 2, 0);                     /* a10 a12 a11 a10 */
+  x5 = glmm_shuff1(x2, 2, 0, 3, 2);                     /* a11 a10 a12 a11 */
+  x6 = glmm_shuff1(x3, 2, 0, 0, 0);                     /* b11 b01 b01 b01 */
+  x2 = glmm_shuff1(r1, 3, 3, 0, 0);                     /* b21 b21 b11 b11 */
+
+  x8 = _mm_unpackhi_ps(x8, x4);                         /* a10 a00 a12 a02 */
+  x9 = _mm_unpackhi_ps(x7, x2);                         /* b21 b20 b21 b20 */
+
+  x0 = glmm_fmadd(x4, x6, x0);
+  x1 = glmm_fmadd(x5, x2, x1);
+
+  x2 = _mm_movehl_ps(l2, l1);                           /* a22 a22 a21 a20 */
+  x3 = glmm_shuff1(x2, 0, 2, 1, 0);                     /* a20 a22 a21 a20 */
+  x2 = glmm_shuff1(x2, 1, 0, 2, 1);                     /* a21 a20 a22 a21 */
+  x4 = _mm_shuffle_ps(r0, r1, _MM_SHUFFLE(1, 1, 2, 2)); /* b12 b12 b02 b02 */
+  
+  x5 = glmm_shuff1(x4, 3, 0, 0, 0);                     /* b12 b02 b02 b02 */
+  x4 = _mm_movehl_ps(r2, x4);                           /* b22 b22 b12 b12 */
+  x0 = glmm_fmadd(x3, x5, x0);
+  x1 = glmm_fmadd(x2, x4, x1);
+
+  /*
+   Dot Product : dest[2][2] =  a02 * b20 +
+                               a12 * b21 +
+                               a22 * b22 +
+                               0   * 00                                    */
+  x2 = _mm_movelh_ps(x8, l2);                           /* 0.f a22 a12 a02 */
+  x3 = _mm_movelh_ps(x9, r2);                           /* 0.f b22 b21 b20 */
+  x2 = glmm_vdots(x2, x3);
+
+  _mm_storeu_ps(&dest[0][0], x0);
+  _mm_storeu_ps(&dest[1][1], x1);
+  _mm_store_ss (&dest[2][2], x2);
+}
+
+#endif
+#endif /* cglm_mat3_sse_h */
diff --git a/include/cglm/simd/sse2/mat4.h b/include/cglm/simd/sse2/mat4.h
new file mode 100644
index 0000000..2127e72
--- /dev/null
+++ b/include/cglm/simd/sse2/mat4.h
@@ -0,0 +1,573 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat_sse_h
+#define cglm_mat_sse_h
+#if defined( __SSE__ ) || defined( __SSE2__ )
+
+#include "../../common.h"
+#include "../intrin.h"
+
+#define glm_mat4_inv_precise_sse2(mat, dest) glm_mat4_inv_sse2(mat, dest)
+
+CGLM_INLINE
+void
+glm_mat4_scale_sse2(mat4 m, float s) {
+  __m128 x0;
+  x0 = glmm_set1(s);
+
+  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
+void
+glm_mat4_transp_sse2(mat4 m, mat4 dest) {
+  __m128 r0, r1, r2, r3;
+
+  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);
+
+  glmm_store(dest[0], r0);
+  glmm_store(dest[1], r1);
+  glmm_store(dest[2], r2);
+  glmm_store(dest[3], r3);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = _mm_mul_ps(glmm_splat_x(r0), l);
+  v1 = _mm_mul_ps(glmm_splat_x(r1), l);
+  v2 = _mm_mul_ps(glmm_splat_x(r2), l);
+  v3 = _mm_mul_ps(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  l  = glmm_load(m1[3]);
+  v0 = glmm_fmadd(glmm_splat_w(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_w(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_w(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mulv_sse2(mat4 m, vec4 v, vec4 dest) {
+  __m128 x0, x1, m0, m1, m2, m3, v0, v1, v2, v3;
+
+  m0 = glmm_load(m[0]);
+  m1 = glmm_load(m[1]);
+  m2 = glmm_load(m[2]);
+  m3 = glmm_load(m[3]);
+
+  x0 = glmm_load(v);
+  v0 = glmm_splat_x(x0);
+  v1 = glmm_splat_y(x0);
+  v2 = glmm_splat_z(x0);
+  v3 = glmm_splat_w(x0);
+
+  x1 = _mm_mul_ps(m3, v3);
+  x1 = glmm_fmadd(m2, v2, x1);
+  x1 = glmm_fmadd(m1, v1, x1);
+  x1 = glmm_fmadd(m0, v0, x1);
+
+  glmm_store(dest, x1);
+}
+
+CGLM_INLINE
+float
+glm_mat4_det_sse2(mat4 mat) {
+  __m128 r0, r1, r2, r3, x0, x1, x2;
+
+  /* 127 <- 0, [square] det(A) = det(At) */
+  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;
+   t[2] = j * o - n * k;
+   t[3] = i * p - m * l;
+   t[4] = i * o - m * k;
+   */
+  x0 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 1, 1), glmm_shuff1(r2, 2, 3, 2, 3),
+                   _mm_mul_ps(glmm_shuff1(r2, 0, 0, 1, 1),
+                              glmm_shuff1(r3, 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 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 2, 2), glmm_shuff1(r2, 1, 1, 3, 3),
+                   _mm_mul_ps(glmm_shuff1(r2, 0, 0, 2, 2),
+                              glmm_shuff1(r3, 1, 1, 3, 3)));
+
+  /*
+     a * (f * t[0] - g * t[1] + h * t[2])
+   - b * (e * t[0] - g * t[3] + h * t[4])
+   + c * (e * t[1] - f * t[3] + h * t[5])
+   - d * (e * t[2] - f * t[4] + g * t[5])
+   */
+  x2 = glmm_fnmadd(glmm_shuff1(r1, 1, 1, 2, 2), glmm_shuff1(x0, 3, 2, 2, 0),
+                   _mm_mul_ps(glmm_shuff1(r1, 0, 0, 0, 1),
+                              _mm_shuffle_ps(x1, x0, _MM_SHUFFLE(1, 0, 0, 0))));
+  x2 = glmm_fmadd(glmm_shuff1(r1, 2, 3, 3, 3),
+                  _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 2, 3, 1)),
+                  x2);
+  
+  x2 = _mm_xor_ps(x2, glmm_float32x4_SIGNMASK_NPNP);
+  
+  return glmm_hadd(_mm_mul_ps(x2, r0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
+  __m128 r0, r1, r2, r3,
+         v0, v1, v2, v3,
+         t0, t1, t2, t3, t4, t5,
+         x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+
+  /* x8 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f); */
+  x8 = glmm_float32x4_SIGNMASK_NPNP;
+  x9 = glmm_shuff1(x8, 2, 1, 2, 1);
+
+  /* 127 <- 0 */
+  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_movehl_ps(r3, r2);                            /* p o l k */
+  x3 = _mm_movelh_ps(r2, r3);                            /* n m j i */
+  x1 = glmm_shuff1(x0, 1, 3, 3 ,3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
+  x4 = glmm_shuff1(x3, 1, 3, 3, 3);                      /* j n n n */
+  x7 = glmm_shuff1(x3, 0, 2, 2, 2);                      /* i m m m */
+
+  x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0));  /* e e i i */
+  x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1));  /* f f j j */
+  x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2));  /* g g k k */
+  x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3));  /* h h l l */
+  
+  t0 = _mm_mul_ps(x3, x1);
+  t1 = _mm_mul_ps(x5, x1);
+  t2 = _mm_mul_ps(x5, x2);
+  t3 = _mm_mul_ps(x6, x1);
+  t4 = _mm_mul_ps(x6, x2);
+  t5 = _mm_mul_ps(x6, x4);
+  
+  /* t1[0] = k * p - o * l;
+     t1[0] = k * p - o * l;
+     t2[0] = g * p - o * h;
+     t3[0] = g * l - k * h; */
+  t0 = glmm_fnmadd(x2, x0, t0);
+  
+  /* t1[1] = j * p - n * l;
+     t1[1] = j * p - n * l;
+     t2[1] = f * p - n * h;
+     t3[1] = f * l - j * h; */
+   t1 = glmm_fnmadd(x4, x0, t1);
+  
+  /* t1[2] = j * o - n * k
+     t1[2] = j * o - n * k;
+     t2[2] = f * o - n * g;
+     t3[2] = f * k - j * g; */
+  t2 = glmm_fnmadd(x4, x3, t2);
+  
+  /* t1[3] = i * p - m * l;
+     t1[3] = i * p - m * l;
+     t2[3] = e * p - m * h;
+     t3[3] = e * l - i * h; */
+  t3 = glmm_fnmadd(x7, x0, t3);
+  
+  /* t1[4] = i * o - m * k;
+     t1[4] = i * o - m * k;
+     t2[4] = e * o - m * g;
+     t3[4] = e * k - i * g; */
+  t4 = glmm_fnmadd(x7, x3, t4);
+  
+  /* t1[5] = i * n - m * j;
+     t1[5] = i * n - m * j;
+     t2[5] = e * n - m * f;
+     t3[5] = e * j - i * f; */
+  t5 = glmm_fnmadd(x7, x5, t5);
+  
+  x4 = _mm_movelh_ps(r0, r1);        /* f e b a */
+  x5 = _mm_movehl_ps(r1, r0);        /* h g d c */
+  
+  x0 = glmm_shuff1(x4, 0, 0, 0, 2);  /* a a a e */
+  x1 = glmm_shuff1(x4, 1, 1, 1, 3);  /* b b b f */
+  x2 = glmm_shuff1(x5, 0, 0, 0, 2);  /* c c c g */
+  x3 = glmm_shuff1(x5, 1, 1, 1, 3);  /* d d d h */
+  
+  v2 = _mm_mul_ps(x0, t1);
+  v1 = _mm_mul_ps(x0, t0);
+  v3 = _mm_mul_ps(x0, t2);
+  v0 = _mm_mul_ps(x1, t0);
+  
+  v2 = glmm_fnmadd(x1, t3, v2);
+  v3 = glmm_fnmadd(x1, t4, v3);
+  v0 = glmm_fnmadd(x2, t1, v0);
+  v1 = glmm_fnmadd(x2, t3, v1);
+  
+  v3 = glmm_fmadd(x2, t5, v3);
+  v0 = glmm_fmadd(x3, t2, v0);
+  v2 = glmm_fmadd(x3, t5, v2);
+  v1 = glmm_fmadd(x3, t4, v1);
+
+  /*
+   dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
+   dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
+   dest[0][2] =  b * t2[0] - c * t2[1] + d * t2[2];
+   dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
+  v0 = _mm_xor_ps(v0, x8);
+  
+  /*
+   dest[2][0] =  e * t1[1] - f * t1[3] + h * t1[5];
+   dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
+   dest[2][2] =  a * t2[1] - b * t2[3] + d * t2[5];
+   dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
+  v2 = _mm_xor_ps(v2, x8);
+
+  /*
+   dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
+   dest[1][1] =  a * t1[0] - c * t1[3] + d * t1[4];
+   dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
+   dest[1][3] =  a * t3[0] - c * t3[3] + d * t3[4]; */
+  v1 = _mm_xor_ps(v1, x9);
+
+  /*
+   dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
+   dest[3][1] =  a * t1[2] - b * t1[4] + c * t1[5];
+   dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
+   dest[3][3] =  a * t3[2] - b * t3[4] + c * t3[5]; */
+  v3 = _mm_xor_ps(v3, x9);
+
+  /* determinant */
+  x0 = _mm_shuffle_ps(v0, v1, _MM_SHUFFLE(0, 0, 0, 0));
+  x1 = _mm_shuffle_ps(v2, v3, _MM_SHUFFLE(0, 0, 0, 0));
+  x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
+
+  x0 = _mm_rcp_ps(glmm_vhadd(_mm_mul_ps(x0, r0)));
+
+  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));
+}
+
+/* old one */
+#if 0
+CGLM_INLINE
+void
+glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
+  __m128 r0, r1, r2, r3,
+         v0, v1, v2, v3,
+         t0, t1, t2, t3, t4, t5,
+         x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+
+  /* x8 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f); */
+  x8 = glmm_float32x4_SIGNMASK_NPNP;
+  x9 = glmm_shuff1(x8, 2, 1, 2, 1);
+
+  /* 127 <- 0 */
+  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_movehl_ps(r3, r2);                            /* p o l k */
+  x3 = _mm_movelh_ps(r2, r3);                            /* n m j i */
+  x1 = glmm_shuff1(x0, 1, 3, 3 ,3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
+  x4 = glmm_shuff1(x3, 1, 3, 3, 3);                      /* j n n n */
+  x7 = glmm_shuff1(x3, 0, 2, 2, 2);                      /* i m m m */
+
+  x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0));  /* e e i i */
+  x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1));  /* f f j j */
+  x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2));  /* g g k k */
+  x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3));  /* h h l l */
+  
+  t0 = _mm_mul_ps(x3, x1);
+  t1 = _mm_mul_ps(x5, x1);
+  t2 = _mm_mul_ps(x5, x2);
+  t3 = _mm_mul_ps(x6, x1);
+  t4 = _mm_mul_ps(x6, x2);
+  t5 = _mm_mul_ps(x6, x4);
+  
+  /* t1[0] = k * p - o * l;
+     t1[0] = k * p - o * l;
+     t2[0] = g * p - o * h;
+     t3[0] = g * l - k * h; */
+  t0 = glmm_fnmadd(x2, x0, t0);
+  
+  /* t1[1] = j * p - n * l;
+     t1[1] = j * p - n * l;
+     t2[1] = f * p - n * h;
+     t3[1] = f * l - j * h; */
+   t1 = glmm_fnmadd(x4, x0, t1);
+  
+  /* t1[2] = j * o - n * k
+     t1[2] = j * o - n * k;
+     t2[2] = f * o - n * g;
+     t3[2] = f * k - j * g; */
+  t2 = glmm_fnmadd(x4, x3, t2);
+  
+  /* t1[3] = i * p - m * l;
+     t1[3] = i * p - m * l;
+     t2[3] = e * p - m * h;
+     t3[3] = e * l - i * h; */
+  t3 = glmm_fnmadd(x7, x0, t3);
+  
+  /* t1[4] = i * o - m * k;
+     t1[4] = i * o - m * k;
+     t2[4] = e * o - m * g;
+     t3[4] = e * k - i * g; */
+  t4 = glmm_fnmadd(x7, x3, t4);
+  
+  /* t1[5] = i * n - m * j;
+     t1[5] = i * n - m * j;
+     t2[5] = e * n - m * f;
+     t3[5] = e * j - i * f; */
+  t5 = glmm_fnmadd(x7, x5, t5);
+  
+  x4 = _mm_movelh_ps(r0, r1);        /* f e b a */
+  x5 = _mm_movehl_ps(r1, r0);        /* h g d c */
+  
+  x0 = glmm_shuff1(x4, 0, 0, 0, 2);  /* a a a e */
+  x1 = glmm_shuff1(x4, 1, 1, 1, 3);  /* b b b f */
+  x2 = glmm_shuff1(x5, 0, 0, 0, 2);  /* c c c g */
+  x3 = glmm_shuff1(x5, 1, 1, 1, 3);  /* d d d h */
+  
+  v2 = _mm_mul_ps(x0, t1);
+  v1 = _mm_mul_ps(x0, t0);
+  v3 = _mm_mul_ps(x0, t2);
+  v0 = _mm_mul_ps(x1, t0);
+  
+  v2 = glmm_fnmadd(x1, t3, v2);
+  v3 = glmm_fnmadd(x1, t4, v3);
+  v0 = glmm_fnmadd(x2, t1, v0);
+  v1 = glmm_fnmadd(x2, t3, v1);
+  
+  v3 = glmm_fmadd(x2, t5, v3);
+  v0 = glmm_fmadd(x3, t2, v0);
+  v2 = glmm_fmadd(x3, t5, v2);
+  v1 = glmm_fmadd(x3, t4, v1);
+
+  /*
+   dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
+   dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
+   dest[0][2] =  b * t2[0] - c * t2[1] + d * t2[2];
+   dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
+  v0 = _mm_xor_ps(v0, x8);
+  
+  /*
+   dest[2][0] =  e * t1[1] - f * t1[3] + h * t1[5];
+   dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
+   dest[2][2] =  a * t2[1] - b * t2[3] + d * t2[5];
+   dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
+  v2 = _mm_xor_ps(v2, x8);
+
+  /*
+   dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
+   dest[1][1] =  a * t1[0] - c * t1[3] + d * t1[4];
+   dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
+   dest[1][3] =  a * t3[0] - c * t3[3] + d * t3[4]; */
+  v1 = _mm_xor_ps(v1, x9);
+
+  /*
+   dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
+   dest[3][1] =  a * t1[2] - b * t1[4] + c * t1[5];
+   dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
+   dest[3][3] =  a * t3[2] - b * t3[4] + c * t3[5]; */
+  v3 = _mm_xor_ps(v3, x9);
+
+  /* determinant */
+  x0 = _mm_shuffle_ps(v0, v1, _MM_SHUFFLE(0, 0, 0, 0));
+  x1 = _mm_shuffle_ps(v2, v3, _MM_SHUFFLE(0, 0, 0, 0));
+  x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
+
+  x0 = _mm_div_ps(glmm_set1(1.0f), glmm_vhadd(_mm_mul_ps(x0, r0)));
+
+  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
+
+CGLM_INLINE
+void
+glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
+  __m128 r0, r1, r2, r3, s1, s2,
+         v0, v1, v2, v3, v4, v5,
+         t0, t1, t2,
+         x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13;
+
+  /* s1 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f); */
+  s1 = glmm_float32x4_SIGNMASK_NPNP;
+  s2 = glmm_shuff1(s1, 2, 1, 2, 1);
+
+  /* 127 <- 0 */
+  r1 = glmm_load(mat[1]); /* h g f e */
+  r0 = glmm_load(mat[0]); /* d c b a */
+  r3 = glmm_load(mat[3]); /* p o n m */
+  r2 = glmm_load(mat[2]); /* l k j i */
+  
+  x4  = _mm_unpackhi_ps(r0, r2); /* l d k c */
+  x5  = _mm_unpacklo_ps(r0, r2); /* j b i a */
+  x6  = _mm_unpackhi_ps(r1, r3); /* p h o g */
+  x7  = _mm_unpacklo_ps(r1, r3); /* n f m e */
+
+  x0  = _mm_unpackhi_ps(x7, x5); /* j n b f */
+  x1  = _mm_unpacklo_ps(x7, x5); /* i m a e */
+  x2  = _mm_unpackhi_ps(x6, x4); /* l p d h */
+  x3  = _mm_unpacklo_ps(x6, x4); /* k o c g */
+
+  /* c2 = c * h - d * g   c12 = a * g - c * e    c8  = a * f - b * e
+     c1 = k * p - l * o   c11 = i * o - k * m    c7  = i * n - j * m
+     c4 = a * h - d * e   c6  = b * h - d * f    c10 = b * g - c * f
+     c3 = i * p - l * m   c5  = j * p - l * n    c9  = j * o - k * n */
+
+  x8  = _mm_shuffle_ps(x0, x3, _MM_SHUFFLE(3, 1, 3, 1)); /* k c j b */
+  x9  = _mm_shuffle_ps(x0, x3, _MM_SHUFFLE(2, 0, 2, 0)); /* o g n f */
+
+  x10 = glmm_shuff1(x2, 2, 0, 2, 0);                     /* p h p h */
+  x11 = glmm_shuff1(x2, 3, 1, 3, 1);                     /* l d l d */
+
+#if 0 /* TODO measure both */
+  x12 = _mm_shuffle_ps(x4, x5, _MM_SHUFFLE(1, 0, 1, 0)); /* i a k c */
+  x13 = _mm_shuffle_ps(x6, x7, _MM_SHUFFLE(1, 0, 1, 0)); /* m e o g */
+#else
+  x12 = _mm_movelh_ps(x4, x5);                           /* i a k c */
+  x13 = _mm_movelh_ps(x6, x7);                           /* m e o g */
+#endif
+  
+  t0 = _mm_mul_ps(x12, x10);
+  t1 = _mm_mul_ps(x5, x6);
+  t2 = _mm_mul_ps(x5, x9);
+
+  t0 = glmm_fnmadd(x11, x13, t0);
+  t1 = glmm_fnmadd(x4, x7, t1);
+  t2 = glmm_fnmadd(x8, x7, t2);
+
+  /* det */
+  /* v0: c3 * c10 + c4 * c9 + c1 * c8 + c2 * c7 */
+  /* v1: c5 * c12 + c6 * c11 */
+
+  v5 = glmm_set1_rval(1.0f);
+  v0 = glmm_shuff1(t2, 2, 3, 0, 1);
+  v1 = glmm_shuff1(t1, 0, 1, 2, 3);
+  v0 = _mm_mul_ps(t0, v0);
+  v1 = _mm_mul_ps(t1, v1);  
+  v2 = glmm_shuff1(v1, 1, 0, 0, 1);
+  v3 = glmm_shuff1(v0, 0, 1, 2, 3);
+  v1 = _mm_add_ps(v1, v2);
+  v0 = _mm_add_ps(v0, v3);
+  v2 = glmm_shuff1(v0, 1, 0, 0, 1);
+  v0 = _mm_add_ps(v0, v2); 
+
+  v0 = _mm_sub_ps(v0, v1); /* det */
+  v0 = _mm_div_ps(v5, v0); /* idt */
+
+  /* multiply t0,t1,t2 by idt to reduce 1mul below: 2eor+4mul vs 3mul+4eor */
+  t0 = _mm_mul_ps(t0, v0);
+  t1 = _mm_mul_ps(t1, v0);
+  t2 = _mm_mul_ps(t2, v0);
+
+  v0 = glmm_shuff1(t0, 0, 0, 1, 1); /* c2  c2  c1  c1  */
+  v1 = glmm_shuff1(t0, 2, 2, 3, 3); /* c4  c4  c3 c3   */
+  v2 = glmm_shuff1(t1, 0, 0, 1, 1); /* c12 c12 c11 c11 */
+  v3 = glmm_shuff1(t1, 2, 2, 3, 3); /* c6  c6  c5 c5   */
+  v4 = glmm_shuff1(t2, 0, 0, 1, 1); /* c8  c8  c7  c7  */
+  v5 = glmm_shuff1(t2, 2, 2, 3, 3); /* c10 c10 c9 c9   */
+
+  /* result */
+
+  /* dest[0][0] = (f * c1  - g * c5  + h * c9)  * idt;
+     dest[0][1] = (b * c1  - c * c5  + d * c9)  * ndt;
+     dest[0][2] = (n * c2  - o * c6  + p * c10) * idt;
+     dest[0][3] = (j * c2  - k * c6  + l * c10) * ndt;
+  
+     dest[1][0] = (e * c1  - g * c3  + h * c11) * ndt;
+     dest[1][1] = (a * c1  - c * c3  + d * c11) * idt;
+     dest[1][2] = (m * c2  - o * c4  + p * c12) * ndt;
+     dest[1][3] = (i * c2  - k * c4  + l * c12) * idt;
+  
+     dest[2][0] = (e * c5  - f * c3  + h * c7)  * idt;
+     dest[2][1] = (a * c5  - b * c3  + d * c7)  * ndt;
+     dest[2][2] = (m * c6  - n * c4  + p * c8)  * idt;
+     dest[2][3] = (i * c6  - j * c4  + l * c8)  * ndt;
+  
+     dest[3][0] = (e * c9  - f * c11 + g * c7)  * ndt;
+     dest[3][1] = (a * c9  - b * c11 + c * c7)  * idt;
+     dest[3][2] = (m * c10 - n * c12 + o * c8)  * ndt;
+     dest[3][3] = (i * c10 - j * c12 + k * c8)  * idt; */
+
+  r0 = _mm_mul_ps(x0, v0);
+  r1 = _mm_mul_ps(x1, v0);
+  r2 = _mm_mul_ps(x1, v3);
+  r3 = _mm_mul_ps(x1, v5);
+
+  r0 = glmm_fnmadd(x3, v3, r0);
+  r1 = glmm_fnmadd(x3, v1, r1);
+  r2 = glmm_fnmadd(x0, v1, r2);
+  r3 = glmm_fnmadd(x0, v2, r3);
+
+  r0 = glmm_fmadd(x2, v5, r0);
+  r1 = glmm_fmadd(x2, v2, r1);
+  r2 = glmm_fmadd(x2, v4, r2);
+  r3 = glmm_fmadd(x3, v4, r3);
+
+  /* 4xor may be fastart then 4mul, see above  */
+  r0 = _mm_xor_ps(r0, s1);
+  r1 = _mm_xor_ps(r1, s2);
+  r2 = _mm_xor_ps(r2, s1);
+  r3 = _mm_xor_ps(r3, s2);
+
+  glmm_store(dest[0], r0);
+  glmm_store(dest[1], r1);
+  glmm_store(dest[2], r2);
+  glmm_store(dest[3], r3);
+}
+#endif
+#endif /* cglm_mat_sse_h */
diff --git a/include/cglm/simd/sse2/quat.h b/include/cglm/simd/sse2/quat.h
new file mode 100644
index 0000000..def0fe2
--- /dev/null
+++ b/include/cglm/simd/sse2/quat.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_quat_simd_h
+#define cglm_quat_simd_h
+#if defined( __SSE__ ) || defined( __SSE2__ )
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_quat_mul_sse2(versor p, versor q, versor dest) {
+  /*
+   + (a1 b2 + b1 a2 + c1 d2 − d1 c2)i
+   + (a1 c2 − b1 d2 + c1 a2 + d1 b2)j
+   + (a1 d2 + b1 c2 − c1 b2 + d1 a2)k
+     a1 a2 − b1 b2 − c1 c2 − d1 d2
+   */
+
+  __m128 xp, xq, x1, x2, x3, r, x, y, z;
+
+  xp = glmm_load(p); /* 3 2 1 0 */
+  xq = glmm_load(q);
+  x1 = glmm_float32x4_SIGNMASK_NPNP; /* TODO: _mm_set1_ss() + shuff ? */
+  r  = _mm_mul_ps(glmm_splat_w(xp), xq);
+
+  x2 = _mm_unpackhi_ps(x1, x1);
+  x3 = glmm_shuff1(x1, 3, 2, 0, 1);
+  x  = glmm_splat_x(xp);
+  y  = glmm_splat_y(xp);
+  z  = glmm_splat_z(xp);
+
+  x  = _mm_xor_ps(x, x1);
+  y  = _mm_xor_ps(y, x2);
+  z  = _mm_xor_ps(z, x3);
+  
+  x1 = glmm_shuff1(xq, 0, 1, 2, 3);
+  x2 = glmm_shuff1(xq, 1, 0, 3, 2);
+  x3 = glmm_shuff1(xq, 2, 3, 0, 1);
+  
+  r  = glmm_fmadd(x, x1, r);
+  r  = glmm_fmadd(y, x2, r);
+  r  = glmm_fmadd(z, x3, r);
+
+  glmm_store(dest, r);
+}
+
+#endif
+#endif /* cglm_quat_simd_h */
diff --git a/include/cglm/simd/wasm.h b/include/cglm/simd/wasm.h
new file mode 100644
index 0000000..2ced51f
--- /dev/null
+++ b/include/cglm/simd/wasm.h
@@ -0,0 +1,198 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_simd_wasm_h
+#define cglm_simd_wasm_h
+#include "intrin.h"
+#ifdef CGLM_SIMD_WASM
+#include <wasm_simd128.h>
+
+#define glmm_load(p)      wasm_v128_load(p)
+#define glmm_store(p, a)  wasm_v128_store(p, (a))
+
+#define glmm_set1(x)      wasm_f32x4_splat(x)
+#define glmm_set1_ptr(x)  wasm_f32x4_splat(*x)
+#define glmm_set1_rval(x) wasm_f32x4_splat(x)
+#define glmm_128          v128_t
+
+#define glmm_shuff1(xmm, z, y, x, w) wasm_i32x4_shuffle(xmm, xmm, w, x, y, z)
+
+#define glmm_splat(x, lane) glmm_shuff1(x, lane, lane, lane, lane)
+
+#define glmm_splat_x(x) glmm_splat(x, 0)
+#define glmm_splat_y(x) glmm_splat(x, 1)
+#define glmm_splat_z(x) glmm_splat(x, 2)
+#define glmm_splat_w(x) glmm_splat(x, 3)
+
+#define GLMM_NEGZEROf 0x80000000 /*  0x80000000 ---> -0.0f  */
+
+/* _mm_set_ps(X, Y, Z, W); */
+#define GLMM__SIGNMASKf(X, Y, Z, W) wasm_i32x4_const(X, Y, Z, W)
+
+#define glmm_float32x4_SIGNMASK_PNPN GLMM__SIGNMASKf(0, GLMM_NEGZEROf, 0, GLMM_NEGZEROf)
+#define glmm_float32x4_SIGNMASK_NPNP GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, GLMM_NEGZEROf, 0)
+#define glmm_float32x4_SIGNMASK_NPPN GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, 0, GLMM_NEGZEROf)
+#define glmm_float32x4_SIGNMASK_NEG  wasm_i32x4_const_splat(GLMM_NEGZEROf)
+
+static inline glmm_128 glmm_abs(glmm_128 x)             { return wasm_f32x4_abs(x);     }
+static inline glmm_128 glmm_min(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmin(b, a); }
+static inline glmm_128 glmm_max(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmax(b, a); }
+
+static inline
+glmm_128
+glmm_vhadd(glmm_128 v) {
+  glmm_128 x0;
+  x0 = wasm_f32x4_add(v,  glmm_shuff1(v, 0, 1, 2, 3));
+  x0 = wasm_f32x4_add(x0, glmm_shuff1(x0, 1, 0, 0, 1));
+  return x0;
+}
+
+static inline
+glmm_128
+glmm_vhadds(glmm_128 v) {
+  glmm_128 shuf, sums;
+  shuf = glmm_shuff1(v, 2, 3, 0, 1);
+  sums = wasm_f32x4_add(v, shuf);
+  /* shuf = _mm_movehl_ps(shuf, sums); */
+  shuf = wasm_i32x4_shuffle(shuf, sums, 6, 7, 2, 3);
+  sums = wasm_i32x4_shuffle(sums, wasm_f32x4_add(sums, shuf), 4, 1, 2, 3);
+  return sums;
+}
+
+static inline
+float
+glmm_hadd(glmm_128 v) {
+  return wasm_f32x4_extract_lane(glmm_vhadds(v), 0);
+}
+
+static inline
+glmm_128
+glmm_vhmin(glmm_128 v) {
+  glmm_128 x0, x1, x2;
+  x0 = glmm_shuff1(v, 2, 3, 2, 3);     /* [2, 3, 2, 3] */
+  x1 = wasm_f32x4_pmin(x0, v);         /* [0|2, 1|3, 2|2, 3|3] */
+  x2 = glmm_splat(x1, 1);              /* [1|3, 1|3, 1|3, 1|3] */
+  return wasm_f32x4_pmin(x1, x2);
+}
+
+static inline
+float
+glmm_hmin(glmm_128 v) {
+  return wasm_f32x4_extract_lane(glmm_vhmin(v), 0);
+}
+
+static inline
+glmm_128
+glmm_vhmax(glmm_128 v) {
+  glmm_128 x0, x1, x2;
+  x0 = glmm_shuff1(v, 2, 3, 2, 3);     /* [2, 3, 2, 3] */
+  x1 = wasm_f32x4_pmax(x0, v);         /* [0|2, 1|3, 2|2, 3|3] */
+  x2 = glmm_splat(x1, 1);              /* [1|3, 1|3, 1|3, 1|3] */
+  /* _mm_max_ss */
+  return wasm_i32x4_shuffle(x1, wasm_f32x4_pmax(x1, x2), 4, 1, 2, 3);
+}
+
+static inline
+float
+glmm_hmax(glmm_128 v) {
+  return wasm_f32x4_extract_lane(glmm_vhmax(v), 0);
+}
+
+static inline
+glmm_128
+glmm_vdots(glmm_128 a, glmm_128 b) {
+  return glmm_vhadds(wasm_f32x4_mul(a, b));
+}
+
+static inline
+glmm_128
+glmm_vdot(glmm_128 a, glmm_128 b) {
+  glmm_128 x0;
+  x0 = wasm_f32x4_mul(a, b);
+  x0 = wasm_f32x4_add(x0, glmm_shuff1(x0, 1, 0, 3, 2));
+  return wasm_f32x4_add(x0, glmm_shuff1(x0, 0, 1, 0, 1));
+}
+
+static inline
+float
+glmm_dot(glmm_128 a, glmm_128 b) {
+  return wasm_f32x4_extract_lane(glmm_vdots(a, b), 0);
+}
+
+static inline
+float
+glmm_norm(glmm_128 a) {
+  glmm_128 x0;
+  x0 = glmm_vhadds(wasm_f32x4_mul(a, a));
+  return wasm_f32x4_extract_lane(
+          wasm_i32x4_shuffle(x0, wasm_f32x4_sqrt(x0),4, 1, 2, 3), 0);
+}
+
+static inline
+float
+glmm_norm2(glmm_128 a) {
+  return wasm_f32x4_extract_lane(glmm_vhadds(wasm_f32x4_mul(a, a)), 0);
+}
+
+static inline
+float
+glmm_norm_one(glmm_128 a) {
+  return wasm_f32x4_extract_lane(glmm_vhadds(glmm_abs(a)), 0);
+}
+
+static inline
+float
+glmm_norm_inf(glmm_128 a) {
+  return wasm_f32x4_extract_lane(glmm_vhmax(glmm_abs(a)), 0);
+}
+
+static inline
+glmm_128
+glmm_load3(float v[3]) {
+  glmm_128 xy = wasm_v128_load64_zero(v);
+  return wasm_f32x4_replace_lane(xy, 2, v[2]);
+}
+
+static inline
+void
+glmm_store3(float v[3], glmm_128 vx) {
+  wasm_v128_store64_lane(v, vx, 0);
+  wasm_v128_store32_lane(&v[2], vx, 2);
+}
+
+static inline
+glmm_128
+glmm_div(glmm_128 a, glmm_128 b) {
+  return wasm_f32x4_div(a, b);
+}
+
+static inline
+glmm_128
+glmm_fmadd(glmm_128 a, glmm_128 b, glmm_128 c) {
+  return wasm_f32x4_add(c, wasm_f32x4_mul(a, b));
+}
+
+static inline
+glmm_128
+glmm_fnmadd(glmm_128 a, glmm_128 b, glmm_128 c) {
+  return wasm_f32x4_sub(c, wasm_f32x4_mul(a, b));
+}
+
+static inline
+glmm_128
+glmm_fmsub(glmm_128 a, glmm_128 b, glmm_128 c) {
+  return wasm_f32x4_sub(wasm_f32x4_mul(a, b), c);
+}
+
+static inline
+glmm_128
+glmm_fnmsub(glmm_128 a, glmm_128 b, glmm_128 c) {
+  return wasm_f32x4_neg(wasm_f32x4_add(wasm_f32x4_mul(a, b), c));
+}
+
+#endif
+#endif /* cglm_simd_wasm_h */
diff --git a/include/cglm/simd/wasm/affine.h b/include/cglm/simd/wasm/affine.h
new file mode 100644
index 0000000..80b98fb
--- /dev/null
+++ b/include/cglm/simd/wasm/affine.h
@@ -0,0 +1,127 @@
+/*
+ * 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_wasm_h
+#define cglm_affine_mat_wasm_h
+#if defined(__wasm__) && defined(__wasm_simd128__)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mul_wasm(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = wasm_f32x4_mul(glmm_splat_x(r0), l);
+  v1 = wasm_f32x4_mul(glmm_splat_x(r1), l);
+  v2 = wasm_f32x4_mul(glmm_splat_x(r2), l);
+  v3 = wasm_f32x4_mul(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  l  = glmm_load(m1[3]);
+  v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mul_rot_wasm(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, v0, v1, v2;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+
+  v0 = wasm_f32x4_mul(glmm_splat_x(r0), l);
+  v1 = wasm_f32x4_mul(glmm_splat_x(r1), l);
+  v2 = wasm_f32x4_mul(glmm_splat_x(r2), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], glmm_load(m1[3]));
+}
+
+CGLM_INLINE
+void
+glm_inv_tr_wasm(mat4 mat) {
+  glmm_128 r0, r1, r2, r3, x0, x1, x2, x3, x4, x5;
+
+  r0 = glmm_load(mat[0]);
+  r1 = glmm_load(mat[1]);
+  r2 = glmm_load(mat[2]);
+  r3 = glmm_load(mat[3]);
+  x1 = wasm_f32x4_const(0.0f, 0.0f, 0.0f, 1.0f);
+
+  /* _MM_TRANSPOSE4_PS(r0, r1, r2, x1); */
+  x2 = wasm_i32x4_shuffle(r0, r1, 0, 4, 1, 5);
+  x3 = wasm_i32x4_shuffle(r0, r1, 2, 6, 3, 7);
+  x4 = wasm_i32x4_shuffle(r2, x1, 0, 4, 1, 5);
+  x5 = wasm_i32x4_shuffle(r2, x1, 2, 6, 3, 7);
+  /* r0 = _mm_movelh_ps(x2, x4); */
+  r0 = wasm_i32x4_shuffle(x2, x4, 0, 1, 4, 5);
+  /* r1 = _mm_movehl_ps(x4, x2); */
+  r1 = wasm_i32x4_shuffle(x4, x2, 6, 7, 2, 3);
+  /* r2 = _mm_movelh_ps(x3, x5); */
+  r2 = wasm_i32x4_shuffle(x3, x5, 0, 1, 4, 5);
+  /* x1 = _mm_movehl_ps(x5, x3); */
+  x1 = wasm_i32x4_shuffle(x5, x3, 6, 7, 2, 3);
+
+  x2 = glmm_shuff1(r3, 0, 0, 0, 0);
+  x3 = glmm_shuff1(r3, 1, 1, 1, 1);
+  x4 = glmm_shuff1(r3, 2, 2, 2, 2);
+
+  x0 = glmm_fmadd(r0, x2,
+                  glmm_fmadd(r1, x3, wasm_f32x4_mul(r2, x4)));
+  x0 = wasm_f32x4_neg(x0);
+
+  x0 = wasm_f32x4_add(x0, x1);
+
+  glmm_store(mat[0], r0);
+  glmm_store(mat[1], r1);
+  glmm_store(mat[2], r2);
+  glmm_store(mat[3], x0);
+}
+
+#endif
+#endif /* cglm_affine_mat_wasm_h */
diff --git a/include/cglm/simd/wasm/mat2.h b/include/cglm/simd/wasm/mat2.h
new file mode 100644
index 0000000..80ce0fb
--- /dev/null
+++ b/include/cglm/simd/wasm/mat2.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat2_wasm_h
+#define cglm_mat2_wasm_h
+#if defined(__wasm__) && defined(__wasm_simd128__)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat2_mul_wasm(mat2 m1, mat2 m2, mat2 dest) {
+  glmm_128 x0, x1, x2, x3, x4;
+
+  x1 = glmm_load(m1[0]); /* d c b a */
+  x2 = glmm_load(m2[0]); /* h g f e */
+
+  x3 = glmm_shuff1(x2, 2, 2, 0, 0);
+  x4 = glmm_shuff1(x2, 3, 3, 1, 1);
+  /* x0 = _mm_movelh_ps(x1, x1); */
+  x0 = wasm_i32x4_shuffle(x1, x1, 0, 1, 4, 5);
+  /* x2 = _mm_movehl_ps(x1, x1); */
+  x2 = wasm_i32x4_shuffle(x1, x1, 6, 7, 2, 3);
+
+  /*
+   dest[0][0] = a * e + c * f;
+   dest[0][1] = b * e + d * f;
+   dest[1][0] = a * g + c * h;
+   dest[1][1] = b * g + d * h;
+   */
+  x0 = glmm_fmadd(x0, x3, wasm_f32x4_mul(x2, x4));
+
+  glmm_store(dest[0], x0);
+}
+
+CGLM_INLINE
+void
+glm_mat2_transp_wasm(mat2 m, mat2 dest) {
+  /* d c b a */
+  /* d b c a */
+  glmm_store(dest[0], glmm_shuff1(glmm_load(m[0]), 3, 1, 2, 0));
+}
+
+#endif
+#endif /* cglm_mat2_wasm_h */
diff --git a/include/cglm/simd/wasm/mat3.h b/include/cglm/simd/wasm/mat3.h
new file mode 100644
index 0000000..dfe192d
--- /dev/null
+++ b/include/cglm/simd/wasm/mat3.h
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat3_wasm_h
+#define cglm_mat3_wasm_h
+#if defined(__wasm__) && defined(__wasm_simd128__)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_mat3_mul_wasm(mat3 m1, mat3 m2, mat3 dest) {
+  glmm_128 l0, l1, l2, r0, r1, r2, x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+  
+  l0 = wasm_v128_load(m1[0]);
+  l1 = wasm_v128_load(&m1[1][1]);
+
+  r0 = wasm_v128_load(m2[0]);
+  r1 = wasm_v128_load(&m2[1][1]);
+
+  x8 = glmm_shuff1(l0, 0, 2, 1, 0);                     /* a00 a02 a01 a00 */
+  x1 = glmm_shuff1(r0, 3, 0, 0, 0);                     /* b10 b00 b00 b00 */
+  x2 = wasm_i32x4_shuffle(l0, l1, 3, 3, 4, 5);          /* a12 a11 a10 a10 */
+  x3 = wasm_i32x4_shuffle(r0, r1, 1, 3, 4, 6);          /* b20 b11 b10 b01 */
+  x0 = wasm_f32x4_mul(x8, x1);
+
+  x6 = glmm_shuff1(l0, 1, 0, 2, 1);                     /* a01 a00 a02 a01 */
+  x7 = glmm_shuff1(x3, 3, 3, 1, 1);                     /* b20 b20 b10 b10 */
+  l2 = wasm_v128_load32_zero(&m1[2][2]);
+  r2 = wasm_v128_load32_zero(&m2[2][2]);
+  x1 = wasm_f32x4_mul(x6, x7);
+  l2 = glmm_shuff1(l2, 0, 0, 1, 0);                     /* a22 a22 0.f a22 */
+  r2 = glmm_shuff1(r2, 0, 0, 1, 0);                     /* b22 b22 0.f b22 */
+
+  x4 = glmm_shuff1(x2, 0, 3, 2, 0);                     /* a10 a12 a11 a10 */
+  x5 = glmm_shuff1(x2, 2, 0, 3, 2);                     /* a11 a10 a12 a11 */
+  x6 = glmm_shuff1(x3, 2, 0, 0, 0);                     /* b11 b01 b01 b01 */
+  x2 = glmm_shuff1(r1, 3, 3, 0, 0);                     /* b21 b21 b11 b11 */
+
+  /* x8 = _mm_unpackhi_ps(x8, x4); */
+  /* x9 = _mm_unpackhi_ps(x7, x2); */
+  x8 = wasm_i32x4_shuffle(x8, x4, 2, 6, 3, 7);          /* a10 a00 a12 a02 */
+  x9 = wasm_i32x4_shuffle(x7, x2, 2, 6, 3, 7);          /* b21 b20 b21 b20 */
+
+  x0 = glmm_fmadd(x4, x6, x0);
+  x1 = glmm_fmadd(x5, x2, x1);
+
+  /* x2 = _mm_movehl_ps(l2, l1); */
+  x2 = wasm_i32x4_shuffle(l2, l1, 6, 7, 2, 3);          /* a22 a22 a21 a20 */
+  x3 = glmm_shuff1(x2, 0, 2, 1, 0);                     /* a20 a22 a21 a20 */
+  x2 = glmm_shuff1(x2, 1, 0, 2, 1);                     /* a21 a20 a22 a21 */
+  x4 = wasm_i32x4_shuffle(r0, r1, 2, 2, 5, 5);          /* b12 b12 b02 b02 */
+  
+  x5 = glmm_shuff1(x4, 3, 0, 0, 0);                     /* b12 b02 b02 b02 */
+  /* x4 = _mm_movehl_ps(r2, x4); */
+  x4 = wasm_i32x4_shuffle(r2, x4, 6, 7, 2, 3);          /* b22 b22 b12 b12 */
+  x0 = glmm_fmadd(x3, x5, x0);
+  x1 = glmm_fmadd(x2, x4, x1);
+
+  /*
+   Dot Product : dest[2][2] =  a02 * b20 +
+                               a12 * b21 +
+                               a22 * b22 +
+                               0   * 00                                    */
+  /* x2 = _mm_movelh_ps(x8, l2); */
+  /* x3 = _mm_movelh_ps(x9, r2); */
+  x2 = wasm_i32x4_shuffle(x8, l2, 0, 1, 4, 5);           /* 0.f a22 a12 a02 */
+  x3 = wasm_i32x4_shuffle(x9, r2, 0, 1, 4, 5);           /* 0.f b22 b21 b20 */
+  x2 = glmm_vdots(x2, x3);
+
+  /* _mm_storeu_ps(&dest[0][0], x0); */
+  wasm_v128_store(&dest[0][0], x0);
+  /* _mm_storeu_ps(&dest[1][1], x1); */
+  wasm_v128_store(&dest[1][1], x1);
+  /* _mm_store_ss (&dest[2][2], x2); */
+  wasm_v128_store32_lane(&dest[2][2], x2, 0);
+}
+
+#endif
+#endif /* cglm_mat3_wasm_h */
diff --git a/include/cglm/simd/wasm/mat4.h b/include/cglm/simd/wasm/mat4.h
new file mode 100644
index 0000000..79ed688
--- /dev/null
+++ b/include/cglm/simd/wasm/mat4.h
@@ -0,0 +1,454 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_mat_wasm_h
+#define cglm_mat_wasm_h
+#if defined(__wasm__) && defined(__wasm_simd128__)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+#define glm_mat4_inv_precise_wasm(mat, dest) glm_mat4_inv_wasm(mat, dest)
+
+CGLM_INLINE
+void
+glm_mat4_scale_wasm(mat4 m, float s) {
+  glmm_128 x0;
+  x0 = wasm_f32x4_splat(s);
+
+  glmm_store(m[0], wasm_f32x4_mul(glmm_load(m[0]), x0));
+  glmm_store(m[1], wasm_f32x4_mul(glmm_load(m[1]), x0));
+  glmm_store(m[2], wasm_f32x4_mul(glmm_load(m[2]), x0));
+  glmm_store(m[3], wasm_f32x4_mul(glmm_load(m[3]), x0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_transp_wasm(mat4 m, mat4 dest) {
+  glmm_128 r0, r1, r2, r3, tmp0, tmp1, tmp2, tmp3;
+
+  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); */
+  tmp0 = wasm_i32x4_shuffle(r0, r1, 0, 4, 1, 5);
+  tmp1 = wasm_i32x4_shuffle(r0, r1, 2, 6, 3, 7);
+  tmp2 = wasm_i32x4_shuffle(r2, r3, 0, 4, 1, 5);
+  tmp3 = wasm_i32x4_shuffle(r2, r3, 2, 6, 3, 7);
+  /* r0 = _mm_movelh_ps(tmp0, tmp2); */
+  r0 = wasm_i32x4_shuffle(tmp0, tmp2, 0, 1, 4, 5);
+  /* r1 = _mm_movehl_ps(tmp2, tmp0); */
+  r1 = wasm_i32x4_shuffle(tmp2, tmp0, 6, 7, 2, 3);
+  /* r2 = _mm_movelh_ps(tmp1, tmp3); */
+  r2 = wasm_i32x4_shuffle(tmp1, tmp3, 0, 1, 4, 5);
+  /* r3 = _mm_movehl_ps(tmp3, tmp1); */
+  r3 = wasm_i32x4_shuffle(tmp3, tmp1, 6, 7, 2, 3);
+
+  glmm_store(dest[0], r0);
+  glmm_store(dest[1], r1);
+  glmm_store(dest[2], r2);
+  glmm_store(dest[3], r3);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mul_wasm(mat4 m1, mat4 m2, mat4 dest) {
+  /* D = R * L (Column-Major) */
+
+  glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
+
+  l  = glmm_load(m1[0]);
+  r0 = glmm_load(m2[0]);
+  r1 = glmm_load(m2[1]);
+  r2 = glmm_load(m2[2]);
+  r3 = glmm_load(m2[3]);
+
+  v0 = wasm_f32x4_mul(glmm_splat_x(r0), l);
+  v1 = wasm_f32x4_mul(glmm_splat_x(r1), l);
+  v2 = wasm_f32x4_mul(glmm_splat_x(r2), l);
+  v3 = wasm_f32x4_mul(glmm_splat_x(r3), l);
+
+  l  = glmm_load(m1[1]);
+  v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
+
+  l  = glmm_load(m1[2]);
+  v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
+
+  l  = glmm_load(m1[3]);
+  v0 = glmm_fmadd(glmm_splat_w(r0), l, v0);
+  v1 = glmm_fmadd(glmm_splat_w(r1), l, v1);
+  v2 = glmm_fmadd(glmm_splat_w(r2), l, v2);
+  v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
+
+  glmm_store(dest[0], v0);
+  glmm_store(dest[1], v1);
+  glmm_store(dest[2], v2);
+  glmm_store(dest[3], v3);
+}
+
+CGLM_INLINE
+void
+glm_mat4_mulv_wasm(mat4 m, vec4 v, vec4 dest) {
+  glmm_128 x0, x1, m0, m1, m2, m3, v0, v1, v2, v3;
+
+  m0 = glmm_load(m[0]);
+  m1 = glmm_load(m[1]);
+  m2 = glmm_load(m[2]);
+  m3 = glmm_load(m[3]);
+
+  x0 = glmm_load(v);
+  v0 = glmm_splat_x(x0);
+  v1 = glmm_splat_y(x0);
+  v2 = glmm_splat_z(x0);
+  v3 = glmm_splat_w(x0);
+
+  x1 = wasm_f32x4_mul(m3, v3);
+  x1 = glmm_fmadd(m2, v2, x1);
+  x1 = glmm_fmadd(m1, v1, x1);
+  x1 = glmm_fmadd(m0, v0, x1);
+
+  glmm_store(dest, x1);
+}
+
+CGLM_INLINE
+float
+glm_mat4_det_wasm(mat4 mat) {
+  glmm_128 r0, r1, r2, r3, x0, x1, x2;
+
+  /* 127 <- 0, [square] det(A) = det(At) */
+  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;
+   t[2] = j * o - n * k;
+   t[3] = i * p - m * l;
+   t[4] = i * o - m * k;
+   */
+  x0 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 1, 1), glmm_shuff1(r2, 2, 3, 2, 3),
+                   wasm_f32x4_mul(glmm_shuff1(r2, 0, 0, 1, 1),
+                              glmm_shuff1(r3, 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 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 2, 2), glmm_shuff1(r2, 1, 1, 3, 3),
+                   wasm_f32x4_mul(glmm_shuff1(r2, 0, 0, 2, 2),
+                              glmm_shuff1(r3, 1, 1, 3, 3)));
+
+  /*
+     a * (f * t[0] - g * t[1] + h * t[2])
+   - b * (e * t[0] - g * t[3] + h * t[4])
+   + c * (e * t[1] - f * t[3] + h * t[5])
+   - d * (e * t[2] - f * t[4] + g * t[5])
+   */
+  x2 = glmm_fnmadd(glmm_shuff1(r1, 1, 1, 2, 2), glmm_shuff1(x0, 3, 2, 2, 0),
+                   wasm_f32x4_mul(glmm_shuff1(r1, 0, 0, 0, 1),
+                              wasm_i32x4_shuffle(x1, x0, 0, 0, 4, 5)));
+  x2 = glmm_fmadd(glmm_shuff1(r1, 2, 3, 3, 3),
+                  wasm_i32x4_shuffle(x0, x1, 1, 3, 6, 6),
+                  x2);
+  /* x2 = wasm_v128_xor(x2, wasm_f32x4_const(0.f, -0.f, 0.f, -0.f)); */
+  x2 = wasm_v128_xor(x2, glmm_float32x4_SIGNMASK_PNPN);
+  
+  return glmm_hadd(wasm_f32x4_mul(x2, r0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_inv_fast_wasm(mat4 mat, mat4 dest) {
+  glmm_128 r0, r1, r2, r3,
+         v0, v1, v2, v3,
+         t0, t1, t2, t3, t4, t5,
+         x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+
+  /* x8 = wasm_f32x4_const(0.f, -0.f, 0.f, -0.f); */
+  x8 = glmm_float32x4_SIGNMASK_PNPN;
+  x9 = glmm_shuff1(x8, 2, 1, 2, 1);
+
+  /* 127 <- 0 */
+  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_movehl_ps(r3, r2); */
+  x0 = wasm_i32x4_shuffle(r3, r2, 6, 7, 2, 3);           /* p o l k */
+  /* x3 = _mm_movelh_ps(r2, r3); */
+  x3 = wasm_i32x4_shuffle(r2, r3, 0, 1, 4, 5);           /* n m j i */
+  x1 = glmm_shuff1(x0, 1, 3, 3 ,3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
+  x4 = glmm_shuff1(x3, 1, 3, 3, 3);                      /* j n n n */
+  x7 = glmm_shuff1(x3, 0, 2, 2, 2);                      /* i m m m */
+
+  x6 = wasm_i32x4_shuffle(r2, r1, 0, 0, 4, 4);           /* e e i i */
+  x5 = wasm_i32x4_shuffle(r2, r1, 1, 1, 5, 5);           /* f f j j */
+  x3 = wasm_i32x4_shuffle(r2, r1, 2, 2, 6, 6);           /* g g k k */
+  x0 = wasm_i32x4_shuffle(r2, r1, 3, 3, 7, 7);           /* h h l l */
+  
+  t0 = wasm_f32x4_mul(x3, x1);
+  t1 = wasm_f32x4_mul(x5, x1);
+  t2 = wasm_f32x4_mul(x5, x2);
+  t3 = wasm_f32x4_mul(x6, x1);
+  t4 = wasm_f32x4_mul(x6, x2);
+  t5 = wasm_f32x4_mul(x6, x4);
+  
+  /* t1[0] = k * p - o * l;
+     t1[0] = k * p - o * l;
+     t2[0] = g * p - o * h;
+     t3[0] = g * l - k * h; */
+  t0 = glmm_fnmadd(x2, x0, t0);
+  
+  /* t1[1] = j * p - n * l;
+     t1[1] = j * p - n * l;
+     t2[1] = f * p - n * h;
+     t3[1] = f * l - j * h; */
+   t1 = glmm_fnmadd(x4, x0, t1);
+  
+  /* t1[2] = j * o - n * k
+     t1[2] = j * o - n * k;
+     t2[2] = f * o - n * g;
+     t3[2] = f * k - j * g; */
+  t2 = glmm_fnmadd(x4, x3, t2);
+  
+  /* t1[3] = i * p - m * l;
+     t1[3] = i * p - m * l;
+     t2[3] = e * p - m * h;
+     t3[3] = e * l - i * h; */
+  t3 = glmm_fnmadd(x7, x0, t3);
+  
+  /* t1[4] = i * o - m * k;
+     t1[4] = i * o - m * k;
+     t2[4] = e * o - m * g;
+     t3[4] = e * k - i * g; */
+  t4 = glmm_fnmadd(x7, x3, t4);
+  
+  /* t1[5] = i * n - m * j;
+     t1[5] = i * n - m * j;
+     t2[5] = e * n - m * f;
+     t3[5] = e * j - i * f; */
+  t5 = glmm_fnmadd(x7, x5, t5);
+  /* x4 = _mm_movelh_ps(r0, r1); */
+  x4 = wasm_i32x4_shuffle(r0, r1, 0, 1, 4, 5);           /* f e b a */
+  /* x5 = _mm_movehl_ps(r1, r0); */
+  x5 = wasm_i32x4_shuffle(r1, r0, 6, 7, 2, 3);           /* h g d c */
+  
+  x0 = glmm_shuff1(x4, 0, 0, 0, 2);                      /* a a a e */
+  x1 = glmm_shuff1(x4, 1, 1, 1, 3);                      /* b b b f */
+  x2 = glmm_shuff1(x5, 0, 0, 0, 2);                      /* c c c g */
+  x3 = glmm_shuff1(x5, 1, 1, 1, 3);                      /* d d d h */
+  
+  v2 = wasm_f32x4_mul(x0, t1);
+  v1 = wasm_f32x4_mul(x0, t0);
+  v3 = wasm_f32x4_mul(x0, t2);
+  v0 = wasm_f32x4_mul(x1, t0);
+  
+  v2 = glmm_fnmadd(x1, t3, v2);
+  v3 = glmm_fnmadd(x1, t4, v3);
+  v0 = glmm_fnmadd(x2, t1, v0);
+  v1 = glmm_fnmadd(x2, t3, v1);
+  
+  v3 = glmm_fmadd(x2, t5, v3);
+  v0 = glmm_fmadd(x3, t2, v0);
+  v2 = glmm_fmadd(x3, t5, v2);
+  v1 = glmm_fmadd(x3, t4, v1);
+
+  /*
+   dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
+   dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
+   dest[0][2] =  b * t2[0] - c * t2[1] + d * t2[2];
+   dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
+  v0 = wasm_v128_xor(v0, x8);
+  
+  /*
+   dest[2][0] =  e * t1[1] - f * t1[3] + h * t1[5];
+   dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
+   dest[2][2] =  a * t2[1] - b * t2[3] + d * t2[5];
+   dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
+  v2 = wasm_v128_xor(v2, x8);
+
+  /*
+   dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
+   dest[1][1] =  a * t1[0] - c * t1[3] + d * t1[4];
+   dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
+   dest[1][3] =  a * t3[0] - c * t3[3] + d * t3[4]; */
+  v1 = wasm_v128_xor(v1, x9);
+
+  /*
+   dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
+   dest[3][1] =  a * t1[2] - b * t1[4] + c * t1[5];
+   dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
+   dest[3][3] =  a * t3[2] - b * t3[4] + c * t3[5]; */
+  v3 = wasm_v128_xor(v3, x9);
+
+  /* determinant */
+  x0 = wasm_i32x4_shuffle(v0, v1, 0, 0, 4, 4);
+  x1 = wasm_i32x4_shuffle(v2, v3, 0, 0, 4, 4);
+  x0 = wasm_i32x4_shuffle(x0, x1, 0, 2, 4, 6);
+
+  /* x0 = _mm_rcp_ps(glmm_vhadd(wasm_f32x4_mul(x0, r0))); */
+  x0 = wasm_f32x4_div(wasm_f32x4_const_splat(1.0f),
+                      glmm_vhadd(wasm_f32x4_mul(x0, r0)));
+
+  glmm_store(dest[0], wasm_f32x4_mul(v0, x0));
+  glmm_store(dest[1], wasm_f32x4_mul(v1, x0));
+  glmm_store(dest[2], wasm_f32x4_mul(v2, x0));
+  glmm_store(dest[3], wasm_f32x4_mul(v3, x0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_inv_wasm(mat4 mat, mat4 dest) {
+  glmm_128 r0, r1, r2, r3,
+         v0, v1, v2, v3,
+         t0, t1, t2, t3, t4, t5,
+         x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
+
+  /* x8 = wasm_f32x4_const(0.f, -0.f, 0.f, -0.f); */
+  x8 = glmm_float32x4_SIGNMASK_PNPN;
+  x9 = glmm_shuff1(x8, 2, 1, 2, 1);
+
+  /* 127 <- 0 */
+  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_movehl_ps(r3, r2); */
+  x0 = wasm_i32x4_shuffle(r3, r2, 6, 7, 2, 3);           /* p o l k */
+  /* x3 = _mm_movelh_ps(r2, r3); */
+  x3 = wasm_i32x4_shuffle(r2, r3, 0, 1, 4, 5);           /* n m j i */
+  x1 = glmm_shuff1(x0, 1, 3, 3 ,3);                      /* l p p p */
+  x2 = glmm_shuff1(x0, 0, 2, 2, 2);                      /* k o o o */
+  x4 = glmm_shuff1(x3, 1, 3, 3, 3);                      /* j n n n */
+  x7 = glmm_shuff1(x3, 0, 2, 2, 2);                      /* i m m m */
+
+  x6 = wasm_i32x4_shuffle(r2, r1, 0, 0, 4, 4);           /* e e i i */
+  x5 = wasm_i32x4_shuffle(r2, r1, 1, 1, 5, 5);           /* f f j j */
+  x3 = wasm_i32x4_shuffle(r2, r1, 2, 2, 6, 6);           /* g g k k */
+  x0 = wasm_i32x4_shuffle(r2, r1, 3, 3, 7, 7);           /* h h l l */
+  
+  t0 = wasm_f32x4_mul(x3, x1);
+  t1 = wasm_f32x4_mul(x5, x1);
+  t2 = wasm_f32x4_mul(x5, x2);
+  t3 = wasm_f32x4_mul(x6, x1);
+  t4 = wasm_f32x4_mul(x6, x2);
+  t5 = wasm_f32x4_mul(x6, x4);
+  
+  /* t1[0] = k * p - o * l;
+     t1[0] = k * p - o * l;
+     t2[0] = g * p - o * h;
+     t3[0] = g * l - k * h; */
+  t0 = glmm_fnmadd(x2, x0, t0);
+  
+  /* t1[1] = j * p - n * l;
+     t1[1] = j * p - n * l;
+     t2[1] = f * p - n * h;
+     t3[1] = f * l - j * h; */
+   t1 = glmm_fnmadd(x4, x0, t1);
+  
+  /* t1[2] = j * o - n * k
+     t1[2] = j * o - n * k;
+     t2[2] = f * o - n * g;
+     t3[2] = f * k - j * g; */
+  t2 = glmm_fnmadd(x4, x3, t2);
+  
+  /* t1[3] = i * p - m * l;
+     t1[3] = i * p - m * l;
+     t2[3] = e * p - m * h;
+     t3[3] = e * l - i * h; */
+  t3 = glmm_fnmadd(x7, x0, t3);
+  
+  /* t1[4] = i * o - m * k;
+     t1[4] = i * o - m * k;
+     t2[4] = e * o - m * g;
+     t3[4] = e * k - i * g; */
+  t4 = glmm_fnmadd(x7, x3, t4);
+  
+  /* t1[5] = i * n - m * j;
+     t1[5] = i * n - m * j;
+     t2[5] = e * n - m * f;
+     t3[5] = e * j - i * f; */
+  t5 = glmm_fnmadd(x7, x5, t5);
+  /* x4 = _mm_movelh_ps(r0, r1); */
+  x4 = wasm_i32x4_shuffle(r0, r1, 0, 1, 4, 5);           /* f e b a */
+  /* x5 = _mm_movehl_ps(r1, r0); */
+  x5 = wasm_i32x4_shuffle(r1, r0, 6, 7, 2, 3);           /* h g d c */
+  
+  x0 = glmm_shuff1(x4, 0, 0, 0, 2);                      /* a a a e */
+  x1 = glmm_shuff1(x4, 1, 1, 1, 3);                      /* b b b f */
+  x2 = glmm_shuff1(x5, 0, 0, 0, 2);                      /* c c c g */
+  x3 = glmm_shuff1(x5, 1, 1, 1, 3);                      /* d d d h */
+  
+  v2 = wasm_f32x4_mul(x0, t1);
+  v1 = wasm_f32x4_mul(x0, t0);
+  v3 = wasm_f32x4_mul(x0, t2);
+  v0 = wasm_f32x4_mul(x1, t0);
+  
+  v2 = glmm_fnmadd(x1, t3, v2);
+  v3 = glmm_fnmadd(x1, t4, v3);
+  v0 = glmm_fnmadd(x2, t1, v0);
+  v1 = glmm_fnmadd(x2, t3, v1);
+  
+  v3 = glmm_fmadd(x2, t5, v3);
+  v0 = glmm_fmadd(x3, t2, v0);
+  v2 = glmm_fmadd(x3, t5, v2);
+  v1 = glmm_fmadd(x3, t4, v1);
+
+  /*
+   dest[0][0] =  f * t1[0] - g * t1[1] + h * t1[2];
+   dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
+   dest[0][2] =  b * t2[0] - c * t2[1] + d * t2[2];
+   dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
+  v0 = wasm_v128_xor(v0, x8);
+  
+  /*
+   dest[2][0] =  e * t1[1] - f * t1[3] + h * t1[5];
+   dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
+   dest[2][2] =  a * t2[1] - b * t2[3] + d * t2[5];
+   dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
+  v2 = wasm_v128_xor(v2, x8);
+
+  /*
+   dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
+   dest[1][1] =  a * t1[0] - c * t1[3] + d * t1[4];
+   dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
+   dest[1][3] =  a * t3[0] - c * t3[3] + d * t3[4]; */
+  v1 = wasm_v128_xor(v1, x9);
+
+  /*
+   dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
+   dest[3][1] =  a * t1[2] - b * t1[4] + c * t1[5];
+   dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
+   dest[3][3] =  a * t3[2] - b * t3[4] + c * t3[5]; */
+  v3 = wasm_v128_xor(v3, x9);
+
+  /* determinant */
+  x0 = wasm_i32x4_shuffle(v0, v1, 0, 0, 4, 4);
+  x1 = wasm_i32x4_shuffle(v2, v3, 0, 0, 4, 4);
+  x0 = wasm_i32x4_shuffle(x0, x1, 0, 2, 4, 6);
+
+  x0 = wasm_f32x4_div(wasm_f32x4_splat(1.0f), glmm_vhadd(wasm_f32x4_mul(x0, r0)));
+
+  glmm_store(dest[0], wasm_f32x4_mul(v0, x0));
+  glmm_store(dest[1], wasm_f32x4_mul(v1, x0));
+  glmm_store(dest[2], wasm_f32x4_mul(v2, x0));
+  glmm_store(dest[3], wasm_f32x4_mul(v3, x0));
+}
+
+#endif
+#endif /* cglm_mat_wasm_h */
diff --git a/include/cglm/simd/wasm/quat.h b/include/cglm/simd/wasm/quat.h
new file mode 100644
index 0000000..8d72546
--- /dev/null
+++ b/include/cglm/simd/wasm/quat.h
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_quat_wasm_h
+#define cglm_quat_wasm_h
+#if defined(__wasm__) && defined(__wasm_simd128__)
+
+#include "../../common.h"
+#include "../intrin.h"
+
+CGLM_INLINE
+void
+glm_quat_mul_wasm(versor p, versor q, versor dest) {
+  /*
+   + (a1 b2 + b1 a2 + c1 d2 − d1 c2)i
+   + (a1 c2 − b1 d2 + c1 a2 + d1 b2)j
+   + (a1 d2 + b1 c2 − c1 b2 + d1 a2)k
+     a1 a2 − b1 b2 − c1 c2 − d1 d2
+   */
+
+  glmm_128 xp, xq, x1, x2, x3, r, x, y, z;
+
+  xp = glmm_load(p); /* 3 2 1 0 */
+  xq = glmm_load(q);
+  /* x1 = wasm_f32x4_const(0.f, -0.f, 0.f, -0.f); */
+  x1 = glmm_float32x4_SIGNMASK_PNPN; /* TODO: _mm_set1_ss() + shuff ? */
+  r  = wasm_f32x4_mul(glmm_splat_w(xp), xq);
+  /* x2 = _mm_unpackhi_ps(x1, x1); */
+  x2 = wasm_i32x4_shuffle(x1, x1, 2, 6, 3, 7);
+  x3 = glmm_shuff1(x1, 3, 2, 0, 1);
+  x  = glmm_splat_x(xp);
+  y  = glmm_splat_y(xp);
+  z  = glmm_splat_z(xp);
+
+  x  = wasm_v128_xor(x, x1);
+  y  = wasm_v128_xor(y, x2);
+  z  = wasm_v128_xor(z, x3);
+  
+  x1 = glmm_shuff1(xq, 0, 1, 2, 3);
+  x2 = glmm_shuff1(xq, 1, 0, 3, 2);
+  x3 = glmm_shuff1(xq, 2, 3, 0, 1);
+  
+  r  = glmm_fmadd(x, x1, r);
+  r  = glmm_fmadd(y, x2, r);
+  r  = glmm_fmadd(z, x3, r);
+
+  glmm_store(dest, r);
+}
+
+#endif
+#endif /* cglm_quat_wasm_h */
diff --git a/include/cglm/simd/x86.h b/include/cglm/simd/x86.h
new file mode 100644
index 0000000..2410d0f
--- /dev/null
+++ b/include/cglm/simd/x86.h
@@ -0,0 +1,365 @@
+/*
+ * Copyright (c), Recep Aslantas.
+ *
+ * MIT License (MIT), http://opensource.org/licenses/MIT
+ * Full license can be found in the LICENSE file
+ */
+
+#ifndef cglm_simd_x86_h
+#define cglm_simd_x86_h
+#include "intrin.h"
+#ifdef CGLM_SIMD_x86
+
+#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
+
+#define glmm_128     __m128
+
+#ifdef __AVX__
+#  define glmm_shuff1(xmm, z, y, x, w)                                        \
+     _mm_permute_ps((xmm), _MM_SHUFFLE(z, y, x, w))
+#else
+#  if !defined(CGLM_NO_INT_DOMAIN) && defined(__SSE2__)
+#    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
+#endif
+
+#define glmm_splat(x, lane) glmm_shuff1(x, lane, lane, lane, lane)
+
+#ifdef __AVX__
+#  define glmm_set1(x)      _mm_broadcast_ss(&x)
+#  define glmm_set1_ptr(x)  _mm_broadcast_ss(x)
+#  define glmm_set1_rval(x) _mm_set1_ps(x)
+#  ifdef __AVX2__
+#    define glmm_splat_x(x) _mm_broadcastss_ps(x)
+#  else
+#    define glmm_splat_x(x) _mm_permute_ps(x, _MM_SHUFFLE(0, 0, 0, 0))
+#  endif
+#  define glmm_splat_y(x)   _mm_permute_ps(x, _MM_SHUFFLE(1, 1, 1, 1))
+#  define glmm_splat_z(x)   _mm_permute_ps(x, _MM_SHUFFLE(2, 2, 2, 2))
+#  define glmm_splat_w(x)   _mm_permute_ps(x, _MM_SHUFFLE(3, 3, 3, 3))
+#else
+#  define glmm_set1(x)      _mm_set1_ps(x)
+#  define glmm_set1_ptr(x)  _mm_set1_ps(*x)
+#  define glmm_set1_rval(x) _mm_set1_ps(x)
+
+#  define glmm_splat_x(x)   glmm_splat(x, 0)
+#  define glmm_splat_y(x)   glmm_splat(x, 1)
+#  define glmm_splat_z(x)   glmm_splat(x, 2)
+#  define glmm_splat_w(x)   glmm_splat(x, 3)
+#endif
+
+#ifdef __AVX__
+#  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
+
+/* Note that `0x80000000` corresponds to `INT_MIN` for a 32-bit int. */
+
+#if defined(__SSE2__)
+#  define GLMM_NEGZEROf ((int)0x80000000) /*  0x80000000 ---> -0.0f  */
+#  define GLMM_POSZEROf ((int)0x00000000) /*  0x00000000 ---> +0.0f  */
+#else
+#  ifdef CGLM_FAST_MATH
+     union { int i; float f; } static GLMM_NEGZEROf_TU = { .i = (int)0x80000000 };
+#    define GLMM_NEGZEROf GLMM_NEGZEROf_TU.f
+#    define GLMM_POSZEROf 0.0f
+#  else
+#    define GLMM_NEGZEROf -0.0f
+#    define GLMM_POSZEROf  0.0f
+#  endif
+#endif
+
+#if defined(__SSE2__)
+#  define GLMM__SIGNMASKf(X, Y, Z, W)                                         \
+   _mm_castsi128_ps(_mm_set_epi32(X, Y, Z, W))
+  /* _mm_set_ps(X, Y, Z, W); */
+#else
+#  define GLMM__SIGNMASKf(X, Y, Z, W)  _mm_set_ps(X, Y, Z, W)
+#endif
+
+#define glmm_float32x4_SIGNMASK_PNPN GLMM__SIGNMASKf(GLMM_POSZEROf, GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_NEGZEROf)
+#define glmm_float32x4_SIGNMASK_NPNP GLMM__SIGNMASKf(GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_NEGZEROf, GLMM_POSZEROf)
+#define glmm_float32x4_SIGNMASK_NPPN GLMM__SIGNMASKf(GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_POSZEROf, GLMM_NEGZEROf)
+
+/* fasth math prevents -0.0f to work */
+#if defined(__SSE2__)
+#  define glmm_float32x4_SIGNMASK_NEG _mm_castsi128_ps(_mm_set1_epi32(GLMM_NEGZEROf)) /* _mm_set1_ps(-0.0f) */
+#else
+#  define glmm_float32x4_SIGNMASK_NEG glmm_set1(GLMM_NEGZEROf)
+#endif
+
+#define glmm_float32x8_SIGNMASK_NEG _mm256_castsi256_ps(_mm256_set1_epi32(GLMM_NEGZEROf))
+
+static inline
+__m128
+glmm_abs(__m128 x) {
+  return _mm_andnot_ps(glmm_float32x4_SIGNMASK_NEG, x);
+}
+
+static inline __m128 glmm_min(__m128 a, __m128 b) { return _mm_min_ps(a, b); }
+static inline __m128 glmm_max(__m128 a, __m128 b) { return _mm_max_ps(a, b); }
+
+static inline
+__m128
+glmm_vhadd(__m128 v) {
+  __m128 x0;
+  x0 = _mm_add_ps(v,  glmm_shuff1(v, 0, 1, 2, 3));
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 0, 1));
+  return x0;
+}
+
+static inline
+__m128
+glmm_vhadds(__m128 v) {
+#if defined(__SSE3__)
+  __m128 shuf, sums;
+  shuf = _mm_movehdup_ps(v);
+  sums = _mm_add_ps(v, shuf);
+  shuf = _mm_movehl_ps(shuf, sums);
+  sums = _mm_add_ss(sums, shuf);
+  return sums;
+#else
+  __m128 shuf, sums;
+  shuf = glmm_shuff1(v, 2, 3, 0, 1);
+  sums = _mm_add_ps(v, shuf);
+  shuf = _mm_movehl_ps(shuf, sums);
+  sums = _mm_add_ss(sums, shuf);
+  return sums;
+#endif
+}
+
+static inline
+float
+glmm_hadd(__m128 v) {
+  return _mm_cvtss_f32(glmm_vhadds(v));
+}
+
+static inline
+__m128
+glmm_vhmin(__m128 v) {
+  __m128 x0, x1, x2;
+  x0 = _mm_movehl_ps(v, v);     /* [2, 3, 2, 3] */
+  x1 = _mm_min_ps(x0, v);       /* [0|2, 1|3, 2|2, 3|3] */
+  x2 = glmm_splat(x1, 1);       /* [1|3, 1|3, 1|3, 1|3] */
+  return _mm_min_ss(x1, x2);
+}
+
+static inline
+float
+glmm_hmin(__m128 v) {
+  return _mm_cvtss_f32(glmm_vhmin(v));
+}
+
+static inline
+__m128
+glmm_vhmax(__m128 v) {
+  __m128 x0, x1, x2;
+  x0 = _mm_movehl_ps(v, v);     /* [2, 3, 2, 3] */
+  x1 = _mm_max_ps(x0, v);       /* [0|2, 1|3, 2|2, 3|3] */
+  x2 = glmm_splat(x1, 1);       /* [1|3, 1|3, 1|3, 1|3] */
+  return _mm_max_ss(x1, x2);
+}
+
+static inline
+float
+glmm_hmax(__m128 v) {
+  return _mm_cvtss_f32(glmm_vhmax(v));
+}
+
+static inline
+__m128
+glmm_vdots(__m128 a, __m128 b) {
+#if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT)
+  return _mm_dp_ps(a, b, 0xFF);
+#elif defined(__SSE3__) && defined(CGLM_SSE3_DOT)
+  __m128 x0, x1;
+  x0 = _mm_mul_ps(a, b);
+  x1 = _mm_hadd_ps(x0, x0);
+  return _mm_hadd_ps(x1, x1);
+#else
+  return glmm_vhadds(_mm_mul_ps(a, b));
+#endif
+}
+
+static inline
+__m128
+glmm_vdot(__m128 a, __m128 b) {
+#if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT)
+  return _mm_dp_ps(a, b, 0xFF);
+#elif defined(__SSE3__) && defined(CGLM_SSE3_DOT)
+  __m128 x0, x1;
+  x0 = _mm_mul_ps(a, b);
+  x1 = _mm_hadd_ps(x0, x0);
+  return _mm_hadd_ps(x1, x1);
+#else
+  __m128 x0;
+  x0 = _mm_mul_ps(a, b);
+  x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
+  return _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 0, 1));
+#endif
+}
+
+static inline
+float
+glmm_dot(__m128 a, __m128 b) {
+  return _mm_cvtss_f32(glmm_vdots(a, b));
+}
+
+static inline
+float
+glmm_norm(__m128 a) {
+  return _mm_cvtss_f32(_mm_sqrt_ss(glmm_vhadds(_mm_mul_ps(a, a))));
+}
+
+static inline
+float
+glmm_norm2(__m128 a) {
+  return _mm_cvtss_f32(glmm_vhadds(_mm_mul_ps(a, a)));
+}
+
+static inline
+float
+glmm_norm_one(__m128 a) {
+  return _mm_cvtss_f32(glmm_vhadds(glmm_abs(a)));
+}
+
+static inline
+float
+glmm_norm_inf(__m128 a) {
+  return _mm_cvtss_f32(glmm_vhmax(glmm_abs(a)));
+}
+
+#if defined(__SSE2__)
+static inline
+__m128
+glmm_load3(float v[3]) {
+  __m128i xy;
+  __m128  z;
+
+  xy = _mm_loadl_epi64(CGLM_CASTPTR_ASSUME_ALIGNED(v, const __m128i));
+  z  = _mm_load_ss(&v[2]);
+
+  return _mm_movelh_ps(_mm_castsi128_ps(xy), z);
+}
+
+static inline
+void
+glmm_store3(float v[3], __m128 vx) {
+  _mm_storel_pi(CGLM_CASTPTR_ASSUME_ALIGNED(v, __m64), vx);
+  _mm_store_ss(&v[2], glmm_shuff1(vx, 2, 2, 2, 2));
+}
+#endif
+
+static inline
+__m128
+glmm_div(__m128 a, __m128 b) {
+  return _mm_div_ps(a, b);
+}
+
+/* enable FMA macro for MSVC? */
+#if defined(_MSC_VER) && !defined(__FMA__) && defined(__AVX2__)
+#  define __FMA__ 1
+#endif
+
+static inline
+__m128
+glmm_fmadd(__m128 a, __m128 b, __m128 c) {
+#ifdef __FMA__
+  return _mm_fmadd_ps(a, b, c);
+#else
+  return _mm_add_ps(c, _mm_mul_ps(a, b));
+#endif
+}
+
+static inline
+__m128
+glmm_fnmadd(__m128 a, __m128 b, __m128 c) {
+#ifdef __FMA__
+  return _mm_fnmadd_ps(a, b, c);
+#else
+  return _mm_sub_ps(c, _mm_mul_ps(a, b));
+#endif
+}
+
+static inline
+__m128
+glmm_fmsub(__m128 a, __m128 b, __m128 c) {
+#ifdef __FMA__
+  return _mm_fmsub_ps(a, b, c);
+#else
+  return _mm_sub_ps(_mm_mul_ps(a, b), c);
+#endif
+}
+
+static inline
+__m128
+glmm_fnmsub(__m128 a, __m128 b, __m128 c) {
+#ifdef __FMA__
+  return _mm_fnmsub_ps(a, b, c);
+#else
+  return _mm_xor_ps(_mm_add_ps(_mm_mul_ps(a, b), c),
+                    glmm_float32x4_SIGNMASK_NEG);
+#endif
+}
+
+#if defined(__AVX__)
+static inline
+__m256
+glmm256_fmadd(__m256 a, __m256 b, __m256 c) {
+#ifdef __FMA__
+  return _mm256_fmadd_ps(a, b, c);
+#else
+  return _mm256_add_ps(c, _mm256_mul_ps(a, b));
+#endif
+}
+
+static inline
+__m256
+glmm256_fnmadd(__m256 a, __m256 b, __m256 c) {
+#ifdef __FMA__
+  return _mm256_fnmadd_ps(a, b, c);
+#else
+  return _mm256_sub_ps(c, _mm256_mul_ps(a, b));
+#endif
+}
+
+static inline
+__m256
+glmm256_fmsub(__m256 a, __m256 b, __m256 c) {
+#ifdef __FMA__
+  return _mm256_fmsub_ps(a, b, c);
+#else
+  return _mm256_sub_ps(_mm256_mul_ps(a, b), c);
+#endif
+}
+
+static inline
+__m256
+glmm256_fnmsub(__m256 a, __m256 b, __m256 c) {
+#ifdef __FMA__
+  return _mm256_fmsub_ps(a, b, c);
+#else
+  return _mm256_xor_ps(_mm256_sub_ps(_mm256_mul_ps(a, b), c),
+                       glmm_float32x8_SIGNMASK_NEG);
+#endif
+}
+#endif
+
+#endif
+#endif /* cglm_simd_x86_h */