Unify Altivec/VSX's plog with generic implementation, and enable it!

2 files changed, 31 insertions, 58 deletions

diff --git a/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
index c5e4bede7..8f3296253 100644
--- a/Eigen/src/Core/arch/AltiVec/MathFunctions.h
+++ b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -9,13 +9,15 @@
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-/* The sin, cos, exp, and log functions of this file come from
+/* The sin, cos, and exp functions of this file come from
  * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
  */
 
 #ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
 #define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
 
+#include "../Default/GenericPacketMathFunctions.h"
+
 namespace Eigen {
 
 namespace internal {
@@ -94,62 +96,7 @@ static Packet2ul p2ul_52 = { 52, 52 };
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet4f plog<Packet4f>(const Packet4f& _x)
 {
-  Packet4f x = _x;
-
-  Packet4i emm0;
-
-  /* isvalid_mask is 0 if x < 0 or x is NaN. */
-  Packet4ui isvalid_mask = reinterpret_cast<Packet4ui>(vec_cmpge(x, p4f_ZERO));
-  Packet4ui iszero_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(x, p4f_ZERO));
-
-  x = pmax(x, p4f_min_norm_pos);  /* cut off denormalized stuff */
-  emm0 = vec_sr(reinterpret_cast<Packet4i>(x),
-                reinterpret_cast<Packet4ui>(p4i_23));
-
-  /* keep only the fractional part */
-  x = pand(x, p4f_inv_mant_mask);
-  x = por(x, p4f_half);
-
-  emm0 = psub(emm0, p4i_0x7f);
-  Packet4f e = padd(vec_ctf(emm0, 0), p4f_1);
-
-  /* part2:
-     if( x < SQRTHF ) {
-       e -= 1;
-       x = x + x - 1.0;
-     } else { x = x - 1.0; }
-  */
-  Packet4f mask = reinterpret_cast<Packet4f>(vec_cmplt(x, p4f_cephes_SQRTHF));
-  Packet4f tmp = pand(x, mask);
-  x = psub(x, p4f_1);
-  e = psub(e, pand(p4f_1, mask));
-  x = padd(x, tmp);
-
-  Packet4f x2 = pmul(x,x);
-  Packet4f x3 = pmul(x2,x);
-
-  Packet4f y, y1, y2;
-  y  = pmadd(p4f_cephes_log_p0, x, p4f_cephes_log_p1);
-  y1 = pmadd(p4f_cephes_log_p3, x, p4f_cephes_log_p4);
-  y2 = pmadd(p4f_cephes_log_p6, x, p4f_cephes_log_p7);
-  y  = pmadd(y , x, p4f_cephes_log_p2);
-  y1 = pmadd(y1, x, p4f_cephes_log_p5);
-  y2 = pmadd(y2, x, p4f_cephes_log_p8);
-  y = pmadd(y, x3, y1);
-  y = pmadd(y, x3, y2);
-  y = pmul(y, x3);
-
-  y1 = pmul(e, p4f_cephes_log_q1);
-  tmp = pmul(x2, p4f_half);
-  y = padd(y, y1);
-  x = psub(x, tmp);
-  y2 = pmul(e, p4f_cephes_log_q2);
-  x = padd(x, y);
-  x = padd(x, y2);
-  // negative arg will be NAN, 0 will be -INF
-  x = vec_sel(x, p4f_minus_inf, iszero_mask);
-  x = vec_sel(p4f_minus_nan, x, isvalid_mask);
-  return x;
+  return plog_float(_x);
 }
 
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 7f4e90f75..867aa8494 100755
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -148,7 +148,7 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasAbs  = 1,
     HasSin  = 0,
     HasCos  = 0,
-    HasLog  = 0,
+    HasLog  = 1,
     HasExp  = 1,
 #ifdef __VSX__
     HasSqrt = 1,
@@ -285,6 +285,11 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
   Packet4i v = {from, from, from, from};
   return v;
 }
+
+template<> EIGEN_STRONG_INLINE Packet4f pset1frombits<Packet4f>(unsigned int from) {
+  return reinterpret_cast<Packet4f>(pset1<Packet4i>(from));
+}
+
 template<> EIGEN_STRONG_INLINE void
 pbroadcast4<Packet4f>(const float *a,
                       Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
@@ -414,6 +419,14 @@ template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const
 }
 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
 
+template<> EIGEN_STRONG_INLINE Packet4f pcmp_le(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmple(a,b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmplt(a,b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pcmp_eq(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmpeq(a,b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan(const Packet4f& a, const Packet4f& b) {
+  Packet4f c = reinterpret_cast<Packet4f>(vec_cmpge(a,b));
+  return vec_nor(c,c);
+}
+
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
 
@@ -426,6 +439,10 @@ template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const
 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, vec_nor(b, b)); }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, vec_nor(b, b)); }
 
+template<> EIGEN_STRONG_INLINE Packet4f pselect(const Packet4f& mask, const Packet4f& a, const Packet4f& b) {
+  return vec_sel(b, a, mask);
+}
+
 template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return vec_round(a); }
 template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a) { return vec_ceil(a); }
 template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return vec_floor(a); }
@@ -550,6 +567,15 @@ template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vec_abs(a); }
 template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vec_abs(a); }
 
+template<> EIGEN_STRONG_INLINE Packet4f pshiftright_and_cast(Packet4f a, int n) {
+  return vec_ctf(vec_sr(reinterpret_cast<Packet4i>(a),
+                        reinterpret_cast<Packet4ui>(pset1<Packet4i>(n))),0);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pfrexp<Packet4f>(const Packet4f& a, Packet4f& exponent) {
+  return pfrexp_float(a,exponent);
+}
+
 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
 {
   Packet4f b, sum;