simd: use new glmm_min/max in vec4 where possible

win32, simd: ensure we are on msvc when checking MSVC specific headers

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.8.2)
 project(cglm
-  VERSION 0.9.2
+  VERSION 0.9.3
   HOMEPAGE_URL https://github.com/recp/cglm
   DESCRIPTION "OpenGL Mathematics (glm) for C"
   LANGUAGES C

cglm.podspec

@@ -2,7 +2,7 @@ Pod::Spec.new do |s|
 
   # Description
   s.name         = "cglm"
-  s.version      = "0.9.1"
+  s.version      = "0.9.2"
   s.summary      = "📽 Highly Optimized Graphics Math (glm) for C"
   s.description  = <<-DESC
 cglm is math library for graphics programming for C. See the documentation or README for all features.

configure.ac

@@ -7,7 +7,7 @@
 #*****************************************************************************
 
 AC_PREREQ([2.69])
-AC_INIT([cglm], [0.9.2], [info@recp.me])
+AC_INIT([cglm], [0.9.3], [info@recp.me])
 AM_INIT_AUTOMAKE([-Wall foreign subdir-objects serial-tests])
 
 # Don't use the default cflags (-O2 -g), we set ours manually in Makefile.am.

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
+static inline float32x4_t glmm_abs(float32x4_t v)                { return vabsq_f32(v);    }
-float32x4_t
+static inline float32x4_t glmm_min(float32x4_t a, float32x4_t b) { return vminq_f32(a, b); }
-glmm_abs(float32x4_t v) {
+static inline float32x4_t glmm_max(float32x4_t a, float32x4_t b) { return vmaxq_f32(a, b); }
-  return vabsq_f32(v);
-}
 
 static inline
 float32x4_t

include/cglm/simd/intrin.h

@@ -63,7 +63,7 @@
 #endif
 
 /* ARM Neon */
-#if defined(_WIN32)
+#if defined(_WIN32) && defined(_MSC_VER)
 /* TODO: non-ARM stuff already inported, will this be better option */
 /* #  include <intrin.h> */
 

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
+static inline glmm_128 glmm_abs(glmm_128 x)             { return wasm_f32x4_abs(x);     }
-glmm_128
+static inline glmm_128 glmm_min(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmin(b, a); }
-glmm_abs(glmm_128 x) {
+static inline glmm_128 glmm_max(glmm_128 a, glmm_128 b) { return wasm_f32x4_pmax(b, a); }
-  return wasm_f32x4_abs(x);
-}
 
 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);

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) {

include/cglm/vec4.h

@@ -653,17 +653,14 @@ CGLM_INLINE
 void
 glm_vec4_maxadd(vec4 a, vec4 b, vec4 dest) {
 #if defined(__wasm__) && defined(__wasm_simd128__)
-  glmm_store(dest, wasm_f32x4_add(
+  glmm_store(dest, wasm_f32x4_add(glmm_load(dest),
-          glmm_load(dest),
+                                  glmm_max(glmm_load(a), glmm_load(b))));
-          wasm_f32x4_pmax(glmm_load(a), glmm_load(b))));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
   glmm_store(dest, _mm_add_ps(glmm_load(dest),
-                              _mm_max_ps(glmm_load(a),
+                              glmm_max(glmm_load(a), glmm_load(b))));
-                                         glmm_load(b))));
 #elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vaddq_f32(vld1q_f32(dest),
+  glmm_store(dest, vaddq_f32(glmm_load(dest),
-                            vmaxq_f32(vld1q_f32(a),
+                             glmm_max(glmm_load(a), glmm_load(b))));
-                                      vld1q_f32(b))));
 #else
   dest[0] += glm_max(a[0], b[0]);
   dest[1] += glm_max(a[1], b[1]);
@@ -685,17 +682,14 @@ CGLM_INLINE
 void
 glm_vec4_minadd(vec4 a, vec4 b, vec4 dest) {
 #if defined(__wasm__) && defined(__wasm_simd128__)
-  glmm_store(dest, wasm_f32x4_add(
+  glmm_store(dest, wasm_f32x4_add(glmm_load(dest),
-          glmm_load(dest),
+                                  glmm_min(glmm_load(a), glmm_load(b))));
-          wasm_f32x4_pmin(glmm_load(a), glmm_load(b))));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
   glmm_store(dest, _mm_add_ps(glmm_load(dest),
-                              _mm_min_ps(glmm_load(a),
+                              glmm_min(glmm_load(a), glmm_load(b))));
-                                         glmm_load(b))));
 #elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vaddq_f32(vld1q_f32(dest),
+  glmm_store(dest, vaddq_f32(glmm_load(dest),
-                            vminq_f32(vld1q_f32(a),
+                             glmm_min(glmm_load(a), glmm_load(b))));
-                                      vld1q_f32(b))));
 #else
   dest[0] += glm_min(a[0], b[0]);
   dest[1] += glm_min(a[1], b[1]);
@@ -825,17 +819,14 @@ CGLM_INLINE
 void
 glm_vec4_maxsub(vec4 a, vec4 b, vec4 dest) {
 #if defined(__wasm__) && defined(__wasm_simd128__)
-  glmm_store(dest, wasm_f32x4_sub(
+  glmm_store(dest, wasm_f32x4_sub(glmm_load(dest),
-          glmm_load(dest),
+                                  glmm_max(glmm_load(a), glmm_load(b))));
-          wasm_f32x4_pmax(glmm_load(a), glmm_load(b))));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
   glmm_store(dest, _mm_sub_ps(glmm_load(dest),
-                              _mm_max_ps(glmm_load(a),
+                              glmm_max(glmm_load(a), glmm_load(b))));
-                                         glmm_load(b))));
 #elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vsubq_f32(vld1q_f32(dest),
+  glmm_store(dest, vsubq_f32(glmm_load(dest),
-                            vmaxq_f32(vld1q_f32(a),
+                             glmm_max(glmm_load(a), glmm_load(b))));
-                                      vld1q_f32(b))));
 #else
   dest[0] -= glm_max(a[0], b[0]);
   dest[1] -= glm_max(a[1], b[1]);
@@ -857,17 +848,14 @@ CGLM_INLINE
 void
 glm_vec4_minsub(vec4 a, vec4 b, vec4 dest) {
 #if defined(__wasm__) && defined(__wasm_simd128__)
-  glmm_store(dest, wasm_f32x4_sub(
+  glmm_store(dest, wasm_f32x4_sub(glmm_load(dest),
-          glmm_load(dest),
+                                  glmm_min(glmm_load(a), glmm_load(b))));
-          wasm_f32x4_pmin(glmm_load(a), glmm_load(b))));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
   glmm_store(dest, _mm_sub_ps(glmm_load(dest),
-                              _mm_min_ps(glmm_load(a),
+                              glmm_min(glmm_load(a), glmm_load(b))));
-                                         glmm_load(b))));
 #elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vsubq_f32(vld1q_f32(dest),
+  glmm_store(dest, vsubq_f32(vld1q_f32(dest),
-                            vminq_f32(vld1q_f32(a),
+                             glmm_min(glmm_load(a), glmm_load(b))));
-                                      vld1q_f32(b))));
 #else
   dest[0] -= glm_min(a[0], b[0]);
   dest[1] -= glm_min(a[1], b[1]);
@@ -1031,12 +1019,8 @@ glm_vec4_distance2(vec4 a, vec4 b) {
 CGLM_INLINE
 void
 glm_vec4_maxv(vec4 a, vec4 b, vec4 dest) {
-#if defined(__wasm__) && defined(__wasm_simd128__)
+#if defined(CGLM_SIMD)
-  glmm_store(dest, wasm_f32x4_pmax(glmm_load(a), glmm_load(b)));
+  glmm_store(dest, glmm_max(glmm_load(a), glmm_load(b)));
-#elif defined( __SSE__ ) || defined( __SSE2__ )
-  glmm_store(dest, _mm_max_ps(glmm_load(a), glmm_load(b)));
-#elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vmaxq_f32(vld1q_f32(a), vld1q_f32(b)));
 #else
   dest[0] = glm_max(a[0], b[0]);
   dest[1] = glm_max(a[1], b[1]);
@@ -1055,12 +1039,8 @@ glm_vec4_maxv(vec4 a, vec4 b, vec4 dest) {
 CGLM_INLINE
 void
 glm_vec4_minv(vec4 a, vec4 b, vec4 dest) {
-#if defined(__wasm__) && defined(__wasm_simd128__)
+#if defined(CGLM_SIMD)
-  glmm_store(dest, wasm_f32x4_pmin(glmm_load(a), glmm_load(b)));
+  glmm_store(dest, glmm_min(glmm_load(a), glmm_load(b)));
-#elif defined( __SSE__ ) || defined( __SSE2__ )
-  glmm_store(dest, _mm_min_ps(glmm_load(a), glmm_load(b)));
-#elif defined(CGLM_NEON_FP)
-  vst1q_f32(dest, vminq_f32(vld1q_f32(a), vld1q_f32(b)));
 #else
   dest[0] = glm_min(a[0], b[0]);
   dest[1] = glm_min(a[1], b[1]);
@@ -1080,14 +1060,13 @@ CGLM_INLINE
 void
 glm_vec4_clamp(vec4 v, float minVal, float maxVal) {
 #if defined(__wasm__) && defined(__wasm_simd128__)
-  glmm_store(v, wasm_f32x4_pmin(
+  glmm_store(v, glmm_min(glmm_max(glmm_load(v), wasm_f32x4_splat(minVal)),
-          wasm_f32x4_pmax(glmm_load(v), wasm_f32x4_splat(minVal)),
+                         wasm_f32x4_splat(maxVal)));
-          wasm_f32x4_splat(maxVal)));
 #elif defined( __SSE__ ) || defined( __SSE2__ )
-  glmm_store(v, _mm_min_ps(_mm_max_ps(glmm_load(v), _mm_set1_ps(minVal)),
+  glmm_store(v, glmm_min(glmm_max(glmm_load(v), _mm_set1_ps(minVal)),
-                           _mm_set1_ps(maxVal)));
+                         _mm_set1_ps(maxVal)));
 #elif defined(CGLM_NEON_FP)
-  vst1q_f32(v, vminq_f32(vmaxq_f32(vld1q_f32(v), vdupq_n_f32(minVal)),
+  glmm_store(v, glmm_min(glmm_max(vld1q_f32(v), vdupq_n_f32(minVal)),
                          vdupq_n_f32(maxVal)));
 #else
   v[0] = glm_clamp(v[0], minVal, maxVal);

include/cglm/version.h

@@ -10,6 +10,6 @@
 
 #define CGLM_VERSION_MAJOR 0
 #define CGLM_VERSION_MINOR 9
-#define CGLM_VERSION_PATCH 2
+#define CGLM_VERSION_PATCH 3
 
 #endif /* cglm_version_h */

meson.build

@@ -1,5 +1,5 @@
 project('cglm', 'c',
-    version : '0.9.2',
+    version : '0.9.3',
     license : 'mit',
     default_options : [
         'c_std=c11',

test/include/common.h

@@ -145,7 +145,7 @@ typedef struct test_entry_t {
     } \
   } while(0);
 
-#if defined(_WIN32)
+#if defined(_WIN32) || defined(__MINGW32__) || defined(__MINGW64__) 
 # define drand48()  ((float)(rand() / (RAND_MAX + 1.0)))
 # define OK_TEXT    "ok:"
 # define FAIL_TEXT  "fail:"