From fdef58bd1aef34b8d401579fd44f2b7f3dea9c26 Mon Sep 17 00:00:00 2001
From: myfreeer <myfreeer@users.noreply.github.com>
Date: Mon, 6 Mar 2023 16:50:37 +0800
Subject: [PATCH] simd128: enable in mat3

---
 include/cglm/mat3.h           |  6 ++++++
 include/cglm/simd/wasm.h      | 17 ++++++++++++++++-
 include/cglm/simd/wasm/mat3.h |  4 ++--
 3 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/include/cglm/mat3.h b/include/cglm/mat3.h
index 0b29f97..cbfab30 100644
--- a/include/cglm/mat3.h
+++ b/include/cglm/mat3.h
@@ -42,6 +42,10 @@
 #  include "simd/sse2/mat3.h"
 #endif
 
+#ifdef CGLM_SIMD_WASM
+#  include "simd/wasm/mat3.h"
+#endif
+
 #define GLM_MAT3_IDENTITY_INIT  {{1.0f, 0.0f, 0.0f},                          \
                                  {0.0f, 1.0f, 0.0f},                          \
                                  {0.0f, 0.0f, 1.0f}}
@@ -150,6 +154,8 @@ void
 glm_mat3_mul(mat3 m1, mat3 m2, mat3 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_mat3_mul_sse2(m1, m2, dest);
+#elif defined(__wasm__) && defined(__wasm_simd128__)
+  glm_mat3_mul_wasm(m1, m2, dest);
 #else
   float a00 = m1[0][0], a01 = m1[0][1], a02 = m1[0][2],
         a10 = m1[1][0], a11 = m1[1][1], a12 = m1[1][2],
diff --git a/include/cglm/simd/wasm.h b/include/cglm/simd/wasm.h
index 87c350e..5015ec9 100644
--- a/include/cglm/simd/wasm.h
+++ b/include/cglm/simd/wasm.h
@@ -92,7 +92,22 @@ _mm_sqrt_ss(glmm_128 __a)
 static __inline__ glmm_128 __attribute__((__always_inline__, __nodebug__))
 _mm_rcp_ps(glmm_128 __a)
 {
-    return (glmm_128)wasm_f32x4_div((v128_t)wasm_f32x4_splat(1.0f), (v128_t)__a);
+    return (glmm_128)wasm_f32x4_div((glmm_128)wasm_f32x4_splat(1.0f), (glmm_128)__a);
+}
+
+static __inline__ void __attribute__((__always_inline__, __nodebug__))
+_mm_storeu_ps(float *__p, glmm_128 __a)
+{
+  struct __unaligned {
+    glmm_128 __v;
+  } __attribute__((__packed__, __may_alias__));
+  ((struct __unaligned *)__p)->__v = __a;
+}
+
+static __inline__ void __attribute__((__always_inline__, __nodebug__))
+_mm_store_ss(float *__p, glmm_128 __a)
+{
+  wasm_v128_store32_lane((void*)__p, (glmm_128)__a, 0);
 }
 
 #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
diff --git a/include/cglm/simd/wasm/mat3.h b/include/cglm/simd/wasm/mat3.h
index 6e57b8e..f4d8fed 100644
--- a/include/cglm/simd/wasm/mat3.h
+++ b/include/cglm/simd/wasm/mat3.h
@@ -27,13 +27,13 @@ glm_mat3_mul_wasm(mat3 m1, mat3 m2, mat3 dest) {
   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);
+  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 = _mm_mul_ps(x6, x7);
+  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 */