Add missing logical packet ops for GPU and NEON.
diff --git a/Eigen/src/Core/arch/GPU/PacketMath.h b/Eigen/src/Core/arch/GPU/PacketMath.h
index ddf37b9..e3b2d56 100644
--- a/Eigen/src/Core/arch/GPU/PacketMath.h
+++ b/Eigen/src/Core/arch/GPU/PacketMath.h
@@ -100,6 +100,117 @@
return make_double2(from, from);
}
+namespace {
+
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float bitwise_and(const float& a,
+ const float& b) {
+ return __int_as_float(__float_as_int(a) & __float_as_int(b));
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double bitwise_and(const double& a,
+ const double& b) {
+ return __longlong_as_double(__double_as_longlong(a) &
+ __double_as_longlong(b));
+}
+
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float bitwise_or(const float& a,
+ const float& b) {
+ return __int_as_float(__float_as_int(a) | __float_as_int(b));
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double bitwise_or(const double& a,
+ const double& b) {
+ return __longlong_as_double(__double_as_longlong(a) |
+ __double_as_longlong(b));
+}
+
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float bitwise_xor(const float& a,
+ const float& b) {
+ return __int_as_float(__float_as_int(a) ^ __float_as_int(b));
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double bitwise_xor(const double& a,
+ const double& b) {
+ return __longlong_as_double(__double_as_longlong(a) ^
+ __double_as_longlong(b));
+}
+
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float bitwise_andnot(const float& a,
+ const float& b) {
+ return __int_as_float(__float_as_int(a) & ~__float_as_int(b));
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double bitwise_andnot(const double& a,
+ const double& b) {
+ return __longlong_as_double(__double_as_longlong(a) &
+ ~__double_as_longlong(b));
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float eq_mask(const float& a,
+ const float& b) {
+ return __int_as_float(a == b ? 0xffffffffu : 0u);
+}
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double eq_mask(const double& a,
+ const double& b) {
+ return __longlong_as_double(a == b ? 0xffffffffffffffffull : 0ull);
+}
+
+} // namespace
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pand<float4>(const float4& a,
+ const float4& b) {
+ return make_float4(bitwise_and(a.x, b.x), bitwise_and(a.y, b.y),
+ bitwise_and(a.z, b.z), bitwise_and(a.w, b.w));
+}
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pand<double2>(const double2& a,
+ const double2& b) {
+ return make_double2(bitwise_and(a.x, b.x), bitwise_and(a.y, b.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 por<float4>(const float4& a,
+ const float4& b) {
+ return make_float4(bitwise_or(a.x, b.x), bitwise_or(a.y, b.y),
+ bitwise_or(a.z, b.z), bitwise_or(a.w, b.w));
+}
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 por<double2>(const double2& a,
+ const double2& b) {
+ return make_double2(bitwise_or(a.x, b.x), bitwise_or(a.y, b.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pxor<float4>(const float4& a,
+ const float4& b) {
+ return make_float4(bitwise_xor(a.x, b.x), bitwise_xor(a.y, b.y),
+ bitwise_xor(a.z, b.z), bitwise_xor(a.w, b.w));
+}
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pxor<double2>(const double2& a,
+ const double2& b) {
+ return make_double2(bitwise_xor(a.x, b.x), bitwise_xor(a.y, b.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pandnot<float4>(const float4& a,
+ const float4& b) {
+ return make_float4(bitwise_andnot(a.x, b.x), bitwise_andnot(a.y, b.y),
+ bitwise_andnot(a.z, b.z), bitwise_andnot(a.w, b.w));
+}
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
+pandnot<double2>(const double2& a, const double2& b) {
+ return make_double2(bitwise_andnot(a.x, b.x), bitwise_andnot(a.y, b.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pcmp_eq<float4>(const float4& a,
+ const float4& b) {
+ return make_float4(eq_mask(a.x, b.x), eq_mask(a.y, b.y), eq_mask(a.z, b.z),
+ eq_mask(a.w, b.w));
+}
+template <>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
+pcmp_eq<double2>(const double2& a, const double2& b) {
+ return make_double2(eq_mask(a.x, b.x), eq_mask(a.y, b.y));
+}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 plset<float4>(const float& a) {
return make_float4(a, a+1, a+2, a+3);
