From 5e05eec6d6e04e044108d6f4ddf3e22ef9590790 Mon Sep 17 00:00:00 2001
From: myfreeer <myfreeer@users.noreply.github.com>
Date: Sat, 1 Apr 2023 19:03:48 +0800
Subject: [PATCH] simd128: inline _MM_TRANSPOSE4_PS

---
 include/cglm/simd/wasm.h        | 16 ----------------
 include/cglm/simd/wasm/affine.h | 11 ++++++++++-
 include/cglm/simd/wasm/mat4.h   | 12 ++++++++++--
 3 files changed, 20 insertions(+), 19 deletions(-)

diff --git a/include/cglm/simd/wasm.h b/include/cglm/simd/wasm.h
index 1567d0e..82bc4ae 100644
--- a/include/cglm/simd/wasm.h
+++ b/include/cglm/simd/wasm.h
@@ -34,22 +34,6 @@ _mm_movelh_ps(glmm_128 __a, glmm_128 __b)
   return wasm_i32x4_shuffle(__a, __b, 0, 1, 4, 5);
 }
 
-#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
-  do { \
-    glmm_128 __row0 = (row0); \
-    glmm_128 __row1 = (row1); \
-    glmm_128 __row2 = (row2); \
-    glmm_128 __row3 = (row3); \
-    glmm_128 __tmp0 = wasm_i32x4_shuffle(__row0, __row1, 0, 4, 1, 5); \
-    glmm_128 __tmp1 = wasm_i32x4_shuffle(__row0, __row1, 2, 6, 3, 7); \
-    glmm_128 __tmp2 = wasm_i32x4_shuffle(__row2, __row3, 0, 4, 1, 5); \
-    glmm_128 __tmp3 = wasm_i32x4_shuffle(__row2, __row3, 2, 6, 3, 7); \
-    (row0) = _mm_movelh_ps(__tmp0, __tmp2); \
-    (row1) = _mm_movehl_ps(__tmp2, __tmp0); \
-    (row2) = _mm_movelh_ps(__tmp1, __tmp3); \
-    (row3) = _mm_movehl_ps(__tmp3, __tmp1); \
-  } while (0)
-
 static inline
 glmm_128
 glmm_abs(glmm_128 x) {
diff --git a/include/cglm/simd/wasm/affine.h b/include/cglm/simd/wasm/affine.h
index ebf05fc..4fa2c44 100644
--- a/include/cglm/simd/wasm/affine.h
+++ b/include/cglm/simd/wasm/affine.h
@@ -92,7 +92,16 @@ glm_inv_tr_wasm(mat4 mat) {
   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);
+
+  // _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);
+  r1 = _mm_movehl_ps(x4, x2);
+  r2 = _mm_movelh_ps(x3, x5);
+  x1 = _mm_movehl_ps(x5, x3);
 
   x2 = glmm_shuff1(r3, 0, 0, 0, 0);
   x3 = glmm_shuff1(r3, 1, 1, 1, 1);
diff --git a/include/cglm/simd/wasm/mat4.h b/include/cglm/simd/wasm/mat4.h
index ea1e90e..b90daee 100644
--- a/include/cglm/simd/wasm/mat4.h
+++ b/include/cglm/simd/wasm/mat4.h
@@ -29,14 +29,22 @@ glm_mat4_scale_wasm(mat4 m, float s) {
 CGLM_INLINE
 void
 glm_mat4_transp_wasm(mat4 m, mat4 dest) {
-  glmm_128 r0, r1, r2, r3;
+  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);
+  // _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);
+  r1 = _mm_movehl_ps(tmp2, tmp0);
+  r2 = _mm_movelh_ps(tmp1, tmp3);
+  r3 = _mm_movehl_ps(tmp3, tmp1);
 
   glmm_store(dest[0], r0);
   glmm_store(dest[1], r1);