diff options
author | Igor Babuschkin <igor@babuschk.in> | 2016-08-08 20:24:59 +0100 |
---|---|---|
committer | Igor Babuschkin <igor@babuschk.in> | 2016-08-08 20:24:59 +0100 |
commit | aee693ac520bc2b3980893ee7b323d8c07e0f133 (patch) | |
tree | e7a149a0bb558adeb7d0a3747fe13b94190eeee1 /Eigen/src/Core | |
parent | 72096f3bd4e1db96bae816affffcc2544a5fd005 (diff) |
Add log1p support for CUDA and half floats
Diffstat (limited to 'Eigen/src/Core')
-rw-r--r-- | Eigen/src/Core/arch/CUDA/Half.h | 3 | ||||
-rw-r--r-- | Eigen/src/Core/arch/CUDA/MathFunctions.h | 12 | ||||
-rw-r--r-- | Eigen/src/Core/arch/CUDA/PacketMathHalf.h | 16 |
3 files changed, 30 insertions, 1 deletions
diff --git a/Eigen/src/Core/arch/CUDA/Half.h b/Eigen/src/Core/arch/CUDA/Half.h index c2ff207ba..4d91420d0 100644 --- a/Eigen/src/Core/arch/CUDA/Half.h +++ b/Eigen/src/Core/arch/CUDA/Half.h @@ -391,6 +391,9 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half exp(const half& a) { EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log(const half& a) { return half(::logf(float(a))); } +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log1p(const half& a) { + return half(::log1pf(float(a))); +} EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log10(const half& a) { return half(::log10f(float(a))); } diff --git a/Eigen/src/Core/arch/CUDA/MathFunctions.h b/Eigen/src/Core/arch/CUDA/MathFunctions.h index 5b5bf4c94..0348b41db 100644 --- a/Eigen/src/Core/arch/CUDA/MathFunctions.h +++ b/Eigen/src/Core/arch/CUDA/MathFunctions.h @@ -32,6 +32,18 @@ double2 plog<double2>(const double2& a) } template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +float4 plog1p<float4>(const float4& a) +{ + return make_float4(log1pf(a.x), log1pf(a.y), log1pf(a.z), log1pf(a.w)); +} + +template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +double2 plog1p<double2>(const double2& a) +{ + return make_double2(log1p(a.x), log1p(a.y)); +} + +template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pexp<float4>(const float4& a) { return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w)); diff --git a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h index 0bf891d24..a757ea45b 100644 --- a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h +++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h @@ -34,7 +34,8 @@ template<> struct packet_traits<Eigen::half> : default_packet_traits HasSqrt = 1, HasRsqrt = 1, HasExp = 1, - HasLog = 1 + HasLog = 1, + HasLog1p = 1 }; }; @@ -273,6 +274,11 @@ half2 plog<half2>(const half2& a) { return h2log(a); } +template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +half2 plog1p<half2>(const half2& a) { + return h2log1p(a); +} + template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pexp<half2>(const half2& a) { return h2exp(a); @@ -298,6 +304,14 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 plog<half2>(const half2& return __floats2half2_rn(r1, r2); } +template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 plog1p<half2>(const half2& a) { + float a1 = __low2float(a); + float a2 = __high2float(a); + float r1 = log1pf(a1); + float r2 = log1pf(a2); + return __floats2half2_rn(r1, r2); +} + template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pexp<half2>(const half2& a) { float a1 = __low2float(a); float a2 = __high2float(a); |