wip

DLTcollab · Jul 9, 2024 · bb5cbf0 · bb5cbf0
1 parent ad3e318
commit bb5cbf0
Showing 1 changed file with 19 additions and 19 deletions.
diff --git a/sse2neon.h b/sse2neon.h
@@ -3019,9 +3019,9 @@ FORCE_INLINE __m128d _mm_add_sd(__m128d a, __m128d b)
 #if defined(__aarch64__) || defined(_M_ARM64)
     return _mm_move_sd(a, _mm_add_pd(a, b));
 #else
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     double c[2];
     c[0] = a0 + b0;
@@ -3301,13 +3301,13 @@ FORCE_INLINE __m128d _mm_cmpge_sd(__m128d a, __m128d b)
     return _mm_move_sd(a, _mm_cmpge_pd(a, b));
 #else
     // expand "_mm_cmpge_pd()" to reduce unnecessary operations
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = a0 >= b0 ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif
@@ -3371,13 +3371,13 @@ FORCE_INLINE __m128d _mm_cmpgt_sd(__m128d a, __m128d b)
     return _mm_move_sd(a, _mm_cmpgt_pd(a, b));
 #else
     // expand "_mm_cmpge_pd()" to reduce unnecessary operations
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = a0 > b0 ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif
@@ -3416,11 +3416,11 @@ FORCE_INLINE __m128d _mm_cmple_sd(__m128d a, __m128d b)
     // expand "_mm_cmpge_pd()" to reduce unnecessary operations
     double a0, a1, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = a0 <= b0 ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif
@@ -3486,13 +3486,13 @@ FORCE_INLINE __m128d _mm_cmplt_sd(__m128d a, __m128d b)
 #if defined(__aarch64__) || defined(_M_ARM64)
     return _mm_move_sd(a, _mm_cmplt_pd(a, b));
 #else
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = a0 < b0 ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif
@@ -3682,13 +3682,13 @@ FORCE_INLINE __m128d _mm_cmpord_sd(__m128d a, __m128d b)
 #if defined(__aarch64__) || defined(_M_ARM64)
     return _mm_move_sd(a, _mm_cmpord_pd(a, b));
 #else
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = (a0 == a0 && b0 == b0) ? ~UINT64_C(0) : UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif
@@ -3729,13 +3729,13 @@ FORCE_INLINE __m128d _mm_cmpunord_sd(__m128d a, __m128d b)
 #if defined(__aarch64__) || defined(_M_ARM64)
     return _mm_move_sd(a, _mm_cmpunord_pd(a, b));
 #else
-    double a0, a1, b0;
+    double a0, b0;
     a0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 0));
-    a1 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1));
+    uint64_t a1 = vgetq_lane_u64(vreinterpretq_u64_m128d(a), 1);
     b0 = recast_f64(vgetq_lane_u64(vreinterpretq_u64_m128d(b), 0));
     uint64_t d[2];
     d[0] = (a0 == a0 && b0 == b0) ? UINT64_C(0) : ~UINT64_C(0);
-    d[1] = a1.u64;
+    d[1] = a1;
 
     return vreinterpretq_m128d_u64(vld1q_u64(d));
 #endif