wip

sse2neon.h

@@ -566,16 +566,16 @@ typedef union ALIGN_STRUCT(16) SIMDVec {
 #define _MM_GET_DENORMALS_ZERO_MODE _sse2neon_mm_get_denormals_zero_mode
 #define _MM_SET_DENORMALS_ZERO_MODE _sse2neon_mm_set_denormals_zero_mode
 
-typedef union bit64_union {
+typedef union bit64_union_t {
     double f64;
     int64_t i64;
     uint64_t u64;
-} bit64_union;
-typedef union bit32_union {
+} bit64_union_t;
+typedef union bit32_union_t {
     float f32;
     int32_t i32;
     uint32_t u32;
-} bit32_union;
+} bit32_union_t;
 
 // Function declaration
 // SSE
@@ -3268,13 +3268,14 @@ FORCE_INLINE __m128d _mm_cmpge_pd(__m128d a, __m128d b)
     return vreinterpretq_m128d_u64(
         vcgeq_f64(vreinterpretq_f64_m128d(a), vreinterpretq_f64_m128d(b)));
 #else
-    uint64_t a0 = (uint64_t) vget_low_u64(vreinterpretq_u64_m128d(a));
-    uint64_t a1 = (uint64_t) vget_high_u64(vreinterpretq_u64_m128d(a));
-    uint64_t b0 = (uint64_t) vget_low_u64(vreinterpretq_u64_m128d(b));
-    uint64_t b1 = (uint64_t) vget_high_u64(vreinterpretq_u64_m128d(b));
-    uint64_t d[2];
-    d[0] = (*(double *) &a0) >= (*(double *) &b0) ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = (*(double *) &a1) >= (*(double *) &b1) ? ~UINT64_C(0) : UINT64_C(0);
+    bit64_union_t a0, a1, b0, b1;
+    a0.u64 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0);
+    a1.u64 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
+    b0.u64 = vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0);
+    b1.u64 = vgetq_lane_u64(vreinterpretq_u64_m128d(b), 1);
+    uint64_t d[2]; 
+    d[0] = (a0.f64) >= (b0.f64) ? ~UINT64_C(0) : UINT64_C(0);
+    d[1] = (a1.f64) >= (b1.f64) ? ~UINT64_C(0) : UINT64_C(0);
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif