From dab86796a41c25dddeab825e4aab372d703d37ce Mon Sep 17 00:00:00 2001
From: Recep Aslantas <info@recp.me>
Date: Tue, 9 Jan 2024 21:35:39 +0300
Subject: [PATCH] simd: min / max helpers

---
 include/cglm/simd/arm.h  |  8 +++-----
 include/cglm/simd/wasm.h | 14 ++++++--------
 include/cglm/simd/x86.h  |  3 +++
 3 files changed, 12 insertions(+), 13 deletions(-)

diff --git a/include/cglm/simd/arm.h b/include/cglm/simd/arm.h
index e8507d7..1578390 100644
--- a/include/cglm/simd/arm.h
+++ b/include/cglm/simd/arm.h
@@ -56,11 +56,9 @@ glmm_float32x4_init(float x, float y, float z, float w) {
 #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_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
diff --git a/include/cglm/simd/wasm.h b/include/cglm/simd/wasm.h
index faaf0c0..d9a18cd 100644
--- a/include/cglm/simd/wasm.h
+++ b/include/cglm/simd/wasm.h
@@ -34,13 +34,11 @@
 #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)
+#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_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
@@ -74,7 +72,7 @@ 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] */
+  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);
 }
@@ -90,7 +88,7 @@ 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] */
+  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);
diff --git a/include/cglm/simd/x86.h b/include/cglm/simd/x86.h
index 696410d..8fd5a72 100644
--- a/include/cglm/simd/x86.h
+++ b/include/cglm/simd/x86.h
@@ -74,6 +74,9 @@ 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) {