Fix new generic nearest integer ops on GPU.
diff --git a/Eigen/src/Core/arch/GPU/PacketMath.h b/Eigen/src/Core/arch/GPU/PacketMath.h
index 352c8f5..6d4230a 100644
--- a/Eigen/src/Core/arch/GPU/PacketMath.h
+++ b/Eigen/src/Core/arch/GPU/PacketMath.h
@@ -74,7 +74,6 @@
HasGammaSampleDerAlpha = 1,
HasIGammac = 1,
HasBetaInc = 1,
-
HasBlend = 0
};
};
@@ -106,9 +105,7 @@
HasGammaSampleDerAlpha = 1,
HasIGammac = 1,
HasBetaInc = 1,
-
HasBlend = 0,
- HasFloor = 1,
};
};
@@ -518,6 +515,33 @@
return make_double2(floor(a.x), floor(a.y));
}
+template <>
+EIGEN_DEVICE_FUNC inline float4 pceil<float4>(const float4& a) {
+ return make_float4(ceilf(a.x), ceilf(a.y), ceilf(a.z), ceilf(a.w));
+}
+template <>
+EIGEN_DEVICE_FUNC inline double2 pceil<double2>(const double2& a) {
+ return make_double2(ceil(a.x), ceil(a.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline float4 print<float4>(const float4& a) {
+ return make_float4(rintf(a.x), rintf(a.y), rintf(a.z), rintf(a.w));
+}
+template <>
+EIGEN_DEVICE_FUNC inline double2 print<double2>(const double2& a) {
+ return make_double2(rint(a.x), rint(a.y));
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline float4 ptrunc<float4>(const float4& a) {
+ return make_float4(truncf(a.x), truncf(a.y), truncf(a.z), truncf(a.w));
+}
+template <>
+EIGEN_DEVICE_FUNC inline double2 ptrunc<double2>(const double2& a) {
+ return make_double2(trunc(a.x), trunc(a.y));
+}
+
EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<float4, 4>& kernel) {
float tmp = kernel.packet[0].y;
kernel.packet[0].y = kernel.packet[1].x;
