arm, neon: neon/fma support for glm_quat_mul()

Makefile.am

@@ -111,7 +111,8 @@ cglm_simd_avx_HEADERS = include/cglm/simd/avx/mat4.h \
 cglm_simd_neondir=$(includedir)/cglm/simd/neon
 cglm_simd_neon_HEADERS = include/cglm/simd/neon/mat4.h \
                          include/cglm/simd/neon/mat2.h \
-                         include/cglm/simd/neon/affine.h
+                         include/cglm/simd/neon/affine.h \
+                         include/cglm/simd/neon/quat.h
 
 cglm_structdir=$(includedir)/cglm/struct
 cglm_struct_HEADERS = include/cglm/struct/mat4.h \

include/cglm/quat.h

@@ -63,6 +63,10 @@
 #  include "simd/sse2/quat.h"
 #endif
 
+#ifdef CGLM_NEON_FP
+#  include "simd/neon/quat.h"
+#endif
+
 CGLM_INLINE
 void
 glm_mat4_mulv(mat4 m, vec4 v, vec4 dest);
@@ -412,6 +416,8 @@ glm_quat_mul(versor p, versor q, versor dest) {
    */
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_quat_mul_sse2(p, q, dest);
+#elif defined(CGLM_NEON_FP)
+  glm_quat_mul_neon(p, q, dest);
 #else
   dest[0] = p[3] * q[0] + p[0] * q[3] + p[1] * q[2] - p[2] * q[1];
   dest[1] = p[3] * q[1] - p[0] * q[2] + p[1] * q[3] + p[2] * q[0];

include/cglm/simd/neon/quat.h

@@ -0,0 +1,56 @@
+/*
+ * 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(__ARM_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 = {-0.f, 0.f, 0.f, -0.f};
+  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 */

include/cglm/simd/sse2/quat.h

@@ -41,6 +41,5 @@ glm_quat_mul_sse2(versor p, versor q, versor dest) {
   glmm_store(dest, r);
 }
 
-
 #endif
 #endif /* cglm_quat_simd_h */

win/cglm.vcxproj

@@ -93,6 +93,7 @@
     <ClInclude Include="..\include\cglm\simd\neon\affine.h" />
     <ClInclude Include="..\include\cglm\simd\neon\mat2.h" />
     <ClInclude Include="..\include\cglm\simd\neon\mat4.h" />
+    <ClInclude Include="..\include\cglm\simd\neon\quat.h" />
     <ClInclude Include="..\include\cglm\simd\sse2\affine.h" />
     <ClInclude Include="..\include\cglm\simd\sse2\mat2.h" />
     <ClInclude Include="..\include\cglm\simd\sse2\mat3.h" />

win/cglm.vcxproj.filters

@@ -376,5 +376,8 @@
     <ClInclude Include="..\include\cglm\simd\neon\mat2.h">
       <Filter>include\cglm\simd\neon</Filter>
     </ClInclude>
+    <ClInclude Include="..\include\cglm\simd\neon\quat.h">
+      <Filter>include\cglm\simd\neon</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>