From 32a477ef07141ef9cd1055615960a11c449b0716 Mon Sep 17 00:00:00 2001
From: Recep Aslantas <info@recp.me>
Date: Sun, 31 Mar 2024 23:35:15 +0300
Subject: [PATCH] separate SSE and SSE2

---
 include/cglm/simd/intrin.h | 14 ++++++++++++--
 include/cglm/simd/x86.h    | 23 +++++++++++++++++------
 2 files changed, 29 insertions(+), 8 deletions(-)

diff --git a/include/cglm/simd/intrin.h b/include/cglm/simd/intrin.h
index 137df65..11c46e5 100644
--- a/include/cglm/simd/intrin.h
+++ b/include/cglm/simd/intrin.h
@@ -10,6 +10,9 @@
 
 #if defined( _MSC_VER )
 #  if (defined(_M_AMD64) || defined(_M_X64)) || _M_IX86_FP == 2
+#    ifndef __SSE__
+#      define __SSE__
+#    endif
 #    ifndef __SSE2__
 #      define __SSE2__
 #    endif
@@ -24,15 +27,22 @@
 #  endif
 #endif
 
-#if defined( __SSE__ ) || defined( __SSE2__ )
+#if defined(__SSE__)
 #  include <xmmintrin.h>
-#  include <emmintrin.h>
 #  define CGLM_SSE_FP 1
 #  ifndef CGLM_SIMD_x86
 #    define CGLM_SIMD_x86
 #  endif
 #endif
 
+#if defined(__SSE2__)
+#  include <emmintrin.h>
+#  define CGLM_SSE2_FP 1
+#  ifndef CGLM_SIMD_x86
+#    define CGLM_SIMD_x86
+#  endif
+#endif
+
 #if defined(__SSE3__)
 #  include <pmmintrin.h>
 #  ifndef CGLM_SIMD_x86
diff --git a/include/cglm/simd/x86.h b/include/cglm/simd/x86.h
index 8fd5a72..eb227e7 100644
--- a/include/cglm/simd/x86.h
+++ b/include/cglm/simd/x86.h
@@ -21,7 +21,7 @@
 #define glmm_set1(x) _mm_set1_ps(x)
 #define glmm_128     __m128
 
-#ifdef CGLM_USE_INT_DOMAIN
+#if defined(CGLM_USE_INT_DOMAIN) && defined(__SSE2__)
 #  define glmm_shuff1(xmm, z, y, x, w)                                        \
      _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xmm),                \
                                         _MM_SHUFFLE(z, y, x, w)))
@@ -56,16 +56,27 @@
 
 /* Note that `0x80000000` corresponds to `INT_MIN` for a 32-bit int. */
 #define GLMM_NEGZEROf ((int)0x80000000) /*  0x80000000 ---> -0.0f  */
+#define GLMM_POSZEROf ((int)0x00000000) /*  0x00000000 ---> +0.0f  */
 
-#define GLMM__SIGNMASKf(X, Y, Z, W)                                           \
+#if defined(__SSE2__)
+#  define GLMM__SIGNMASKf(X, Y, Z, W)                                         \
    _mm_castsi128_ps(_mm_set_epi32(X, Y, Z, W))
   /* _mm_set_ps(X, Y, Z, W); */
+#else
+#  define GLMM__SIGNMASKf(X, Y, Z, W)  _mm_set_ps(X, Y, Z, W)
+#endif
 
-#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_PNPN GLMM__SIGNMASKf(GLMM_POSZEROf, GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_NEGZEROf)
+#define glmm_float32x4_SIGNMASK_NPNP GLMM__SIGNMASKf(GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_NEGZEROf, GLMM_POSZEROf)
+#define glmm_float32x4_SIGNMASK_NPPN GLMM__SIGNMASKf(GLMM_NEGZEROf, GLMM_POSZEROf, GLMM_POSZEROf, GLMM_NEGZEROf)
+
+/* fasth math prevents -0.0f to work */
+#if defined(__SSE2__)
+#  define glmm_float32x4_SIGNMASK_NEG _mm_castsi128_ps(_mm_set1_epi32(GLMM_NEGZEROf)) /* _mm_set1_ps(-0.0f) */
+#else
+#  define glmm_float32x4_SIGNMASK_NEG _mm_set1_ps(GLMM_NEGZEROf)
+#endif
 
-#define glmm_float32x4_SIGNMASK_NEG _mm_castsi128_ps(_mm_set1_epi32(GLMM_NEGZEROf)) /* _mm_set1_ps(-0.0f) */
 #define glmm_float32x8_SIGNMASK_NEG _mm256_castsi256_ps(_mm256_set1_epi32(GLMM_NEGZEROf))
 
 static inline