From 56903fc7076c2b25e9c3f46ab96056a0e49b44c3 Mon Sep 17 00:00:00 2001
From: Ahmet Inan <inan@aicodix.de>
Date: Tue, 13 Feb 2024 11:59:44 +0100
Subject: [PATCH] removed negz constants for consistency

---
 avx2.hh           | 10 ++++------
 neon.hh           |  5 ++---
 neon_double.hh    |  5 ++---
 neon_quadruple.hh |  5 ++---
 neon_triple.hh    |  5 ++---
 sse4_1.hh         | 10 ++++------
 6 files changed, 16 insertions(+), 24 deletions(-)

diff --git a/avx2.hh b/avx2.hh
index 9b12334..4e49303 100644
--- a/avx2.hh
+++ b/avx2.hh
@@ -602,10 +602,9 @@ template <>
 inline SIMD<float, 8> vcopysign(SIMD<float, 8> a, SIMD<float, 8> b)
 {
 	SIMD<float, 8> tmp;
-	__m256 negz = _mm256_set1_ps(-0.f);
 	tmp.m = _mm256_or_ps(
-		_mm256_andnot_ps(negz, a.m),
-		_mm256_and_ps(negz, b.m));
+		_mm256_andnot_ps(_mm256_set1_ps(-0.f), a.m),
+		_mm256_and_ps(_mm256_set1_ps(-0.f), b.m));
 	return tmp;
 }
 
@@ -613,10 +612,9 @@ template <>
 inline SIMD<double, 4> vcopysign(SIMD<double, 4> a, SIMD<double, 4> b)
 {
 	SIMD<double, 4> tmp;
-	__m256d negz = _mm256_set1_pd(-0.);
 	tmp.m = _mm256_or_pd(
-		_mm256_andnot_pd(negz, a.m),
-		_mm256_and_pd(negz, b.m));
+		_mm256_andnot_pd(_mm256_set1_pd(-0.), a.m),
+		_mm256_and_pd(_mm256_set1_pd(-0.), b.m));
 	return tmp;
 }
 
diff --git a/neon.hh b/neon.hh
index e7b235e..85d6702 100644
--- a/neon.hh
+++ b/neon.hh
@@ -526,10 +526,9 @@ template <>
 inline SIMD<float, 4> vcopysign(SIMD<float, 4> a, SIMD<float, 4> b)
 {
 	SIMD<float, 4> tmp;
-	uint32x4_t negz = (uint32x4_t)vdupq_n_f32(-0.f);
 	tmp.m = (float32x4_t)vorrq_u32(
-		vbicq_u32((uint32x4_t)a.m, negz),
-		vandq_u32((uint32x4_t)b.m, negz));
+		vbicq_u32((uint32x4_t)a.m, (uint32x4_t)vdupq_n_f32(-0.f)),
+		vandq_u32((uint32x4_t)b.m, (uint32x4_t)vdupq_n_f32(-0.f)));
 	return tmp;
 }
 
diff --git a/neon_double.hh b/neon_double.hh
index 141b95e..e5ae2fd 100644
--- a/neon_double.hh
+++ b/neon_double.hh
@@ -577,11 +577,10 @@ template <>
 inline SIMD<float, 8> vcopysign(SIMD<float, 8> a, SIMD<float, 8> b)
 {
 	SIMD<float, 8> tmp;
-	uint32x4_t negz = (uint32x4_t)vdupq_n_f32(-0.f);
 	for (int i = 0; i < 2; ++i)
 		tmp.m[i] = (float32x4_t)vorrq_u32(
-			vbicq_u32((uint32x4_t)a.m[i], negz),
-			vandq_u32((uint32x4_t)b.m[i], negz));
+			vbicq_u32((uint32x4_t)a.m[i], (uint32x4_t)vdupq_n_f32(-0.f)),
+			vandq_u32((uint32x4_t)b.m[i], (uint32x4_t)vdupq_n_f32(-0.f)));
 	return tmp;
 }
 
diff --git a/neon_quadruple.hh b/neon_quadruple.hh
index 943ed71..0585d24 100644
--- a/neon_quadruple.hh
+++ b/neon_quadruple.hh
@@ -577,11 +577,10 @@ template <>
 inline SIMD<float, 16> vcopysign(SIMD<float, 16> a, SIMD<float, 16> b)
 {
 	SIMD<float, 16> tmp;
-	uint32x4_t negz = (uint32x4_t)vdupq_n_f32(-0.f);
 	for (int i = 0; i < 4; ++i)
 		tmp.m[i] = (float32x4_t)vorrq_u32(
-			vbicq_u32((uint32x4_t)a.m[i], negz),
-			vandq_u32((uint32x4_t)b.m[i], negz));
+			vbicq_u32((uint32x4_t)a.m[i], (uint32x4_t)vdupq_n_f32(-0.f)),
+			vandq_u32((uint32x4_t)b.m[i], (uint32x4_t)vdupq_n_f32(-0.f)));
 	return tmp;
 }
 
diff --git a/neon_triple.hh b/neon_triple.hh
index f22c74b..fc7fffd 100644
--- a/neon_triple.hh
+++ b/neon_triple.hh
@@ -577,11 +577,10 @@ template <>
 inline SIMD<float, 12> vcopysign(SIMD<float, 12> a, SIMD<float, 12> b)
 {
 	SIMD<float, 12> tmp;
-	uint32x4_t negz = (uint32x4_t)vdupq_n_f32(-0.f);
 	for (int i = 0; i < 3; ++i)
 		tmp.m[i] = (float32x4_t)vorrq_u32(
-			vbicq_u32((uint32x4_t)a.m[i], negz),
-			vandq_u32((uint32x4_t)b.m[i], negz));
+			vbicq_u32((uint32x4_t)a.m[i], (uint32x4_t)vdupq_n_f32(-0.f)),
+			vandq_u32((uint32x4_t)b.m[i], (uint32x4_t)vdupq_n_f32(-0.f)));
 	return tmp;
 }
 
diff --git a/sse4_1.hh b/sse4_1.hh
index 3af6b7d..e5a1f90 100644
--- a/sse4_1.hh
+++ b/sse4_1.hh
@@ -602,10 +602,9 @@ template <>
 inline SIMD<float, 4> vcopysign(SIMD<float, 4> a, SIMD<float, 4> b)
 {
 	SIMD<float, 4> tmp;
-	__m128 negz = _mm_set1_ps(-0.f);
 	tmp.m = _mm_or_ps(
-		_mm_andnot_ps(negz, a.m),
-		_mm_and_ps(negz, b.m));
+		_mm_andnot_ps(_mm_set1_ps(-0.f), a.m),
+		_mm_and_ps(_mm_set1_ps(-0.f), b.m));
 	return tmp;
 }
 
@@ -613,10 +612,9 @@ template <>
 inline SIMD<double, 2> vcopysign(SIMD<double, 2> a, SIMD<double, 2> b)
 {
 	SIMD<double, 2> tmp;
-	__m128d negz = _mm_set1_pd(-0.);
 	tmp.m = _mm_or_pd(
-		_mm_andnot_pd(negz, a.m),
-		_mm_and_pd(negz, b.m));
+		_mm_andnot_pd(_mm_set1_pd(-0.), a.m),
+		_mm_and_pd(_mm_set1_pd(-0.), b.m));
 	return tmp;
 }