git: 1b3f665f8f94 - main - math/eigen3: fix runtime on powerpc64
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Wed, 10 Aug 2022 00:32:09 UTC
The branch main has been updated by pkubaj:
URL: https://cgit.FreeBSD.org/ports/commit/?id=1b3f665f8f94be33397d2da714466eb796d5dfef
commit 1b3f665f8f94be33397d2da714466eb796d5dfef
Author: Piotr Kubaj <pkubaj@FreeBSD.org>
AuthorDate: 2022-08-10 00:30:00 +0000
Commit: Piotr Kubaj <pkubaj@FreeBSD.org>
CommitDate: 2022-08-10 00:32:04 +0000
math/eigen3: fix runtime on powerpc64
Merge https://gitlab.com/libeigen/eigen/-/merge_requests/1028 and https://gitlab.com/libeigen/eigen/-/merge_requests/1025
to fix build of reverse dependencies.
---
math/eigen3/Makefile | 1 +
math/eigen3/files/patch-Eigen_Core | 11 ++
.../patch-Eigen_src_Core_arch_AltiVec_Complex.h | 18 ++++
...tch-Eigen_src_Core_arch_AltiVec_MathFunctions.h | 47 +++++++++
.../patch-Eigen_src_Core_arch_AltiVec_PacketMath.h | 117 +++++++++++++++++++++
5 files changed, 194 insertions(+)
diff --git a/math/eigen3/Makefile b/math/eigen3/Makefile
index 537bb9e9f654..5cb289e1c513 100644
--- a/math/eigen3/Makefile
+++ b/math/eigen3/Makefile
@@ -1,5 +1,6 @@
PORTNAME= eigen
DISTVERSION= 3.4.0
+PORTREVISION= 1
CATEGORIES= math
MASTER_SITES= https://gitlab.com/libeigen/eigen/-/archive/${DISTVERSION}/
diff --git a/math/eigen3/files/patch-Eigen_Core b/math/eigen3/files/patch-Eigen_Core
new file mode 100644
index 000000000000..e132efbc152d
--- /dev/null
+++ b/math/eigen3/files/patch-Eigen_Core
@@ -0,0 +1,11 @@
+--- Eigen/Core.orig 2021-08-18 20:41:58 UTC
++++ Eigen/Core
+@@ -346,7 +346,7 @@ using std::ptrdiff_t;
+ #include "src/Core/CoreIterators.h"
+ #include "src/Core/ConditionEstimator.h"
+
+-#if defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
++#if defined(EIGEN_VECTORIZE_VSX)
+ #include "src/Core/arch/AltiVec/MatrixProduct.h"
+ #elif defined EIGEN_VECTORIZE_NEON
+ #include "src/Core/arch/NEON/GeneralBlockPanelKernel.h"
diff --git a/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_Complex.h b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_Complex.h
new file mode 100644
index 000000000000..5aa71fed0310
--- /dev/null
+++ b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_Complex.h
@@ -0,0 +1,18 @@
+--- Eigen/src/Core/arch/AltiVec/Complex.h.orig 2021-08-18 20:41:58 UTC
++++ Eigen/src/Core/arch/AltiVec/Complex.h
+@@ -100,6 +100,7 @@ template<> struct packet_traits<std::complex<float> >
+ HasAbs2 = 0,
+ HasMin = 0,
+ HasMax = 0,
++ HasSqrt = 1,
+ #ifdef __VSX__
+ HasBlend = 1,
+ #endif
+@@ -320,6 +321,7 @@ template<> struct packet_traits<std::complex<double> >
+ HasAbs2 = 0,
+ HasMin = 0,
+ HasMax = 0,
++ HasSqrt = 1,
+ HasSetLinear = 0
+ };
+ };
diff --git a/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_MathFunctions.h b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_MathFunctions.h
new file mode 100644
index 000000000000..a27c2bc166c1
--- /dev/null
+++ b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_MathFunctions.h
@@ -0,0 +1,47 @@
+--- Eigen/src/Core/arch/AltiVec/MathFunctions.h.orig 2021-08-18 20:41:58 UTC
++++ Eigen/src/Core/arch/AltiVec/MathFunctions.h
+@@ -40,16 +40,14 @@ Packet4f pcos<Packet4f>(const Packet4f& _x)
+ return pcos_float(_x);
+ }
+
++#ifdef __VSX__
+ #ifndef EIGEN_COMP_CLANG
+ template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+ Packet4f prsqrt<Packet4f>(const Packet4f& x)
+ {
+ return vec_rsqrt(x);
+ }
+-#endif
+
+-#ifdef __VSX__
+-#ifndef EIGEN_COMP_CLANG
+ template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+ Packet2d prsqrt<Packet2d>(const Packet2d& x)
+ {
+@@ -73,6 +71,26 @@ template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEF
+ Packet2d pexp<Packet2d>(const Packet2d& _x)
+ {
+ return pexp_double(_x);
++}
++
++template<> EIGEN_STRONG_INLINE Packet8bf psqrt<Packet8bf> (const Packet8bf& a){
++ BF16_TO_F32_UNARY_OP_WRAPPER(psqrt<Packet4f>, a);
++}
++
++#ifndef EIGEN_COMP_CLANG
++template<> EIGEN_STRONG_INLINE Packet8bf prsqrt<Packet8bf> (const Packet8bf& a){
++ BF16_TO_F32_UNARY_OP_WRAPPER(prsqrt<Packet4f>, a);
++}
++#endif
++#else
++template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
++Packet4f psqrt<Packet4f>(const Packet4f& x)
++{
++ Packet4f a;
++ for (Index i = 0; i < packet_traits<float>::size; i++) {
++ a[i] = numext::sqrt(x[i]);
++ }
++ return a;
+ }
+ #endif
+
diff --git a/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_PacketMath.h b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_PacketMath.h
new file mode 100644
index 000000000000..db517c61b677
--- /dev/null
+++ b/math/eigen3/files/patch-Eigen_src_Core_arch_AltiVec_PacketMath.h
@@ -0,0 +1,117 @@
+--- Eigen/src/Core/arch/AltiVec/PacketMath.h.orig 2021-08-18 20:41:58 UTC
++++ Eigen/src/Core/arch/AltiVec/PacketMath.h
+@@ -175,16 +175,19 @@ struct packet_traits<float> : default_packet_traits {
+ #else
+ HasRsqrt = 0,
+ #endif
++ HasTanh = EIGEN_FAST_MATH,
++ HasErf = EIGEN_FAST_MATH,
++ HasRint = 1,
+ #else
+ HasSqrt = 0,
+ HasRsqrt = 0,
+- HasTanh = EIGEN_FAST_MATH,
+- HasErf = EIGEN_FAST_MATH,
++ HasTanh = 0,
++ HasErf = 0,
++ HasRint = 0,
+ #endif
+ HasRound = 1,
+ HasFloor = 1,
+ HasCeil = 1,
+- HasRint = 1,
+ HasNegate = 1,
+ HasBlend = 1
+ };
+@@ -217,16 +220,17 @@ struct packet_traits<bfloat16> : default_packet_traits
+ #else
+ HasRsqrt = 0,
+ #endif
++ HasRint = 1,
+ #else
+ HasSqrt = 0,
+ HasRsqrt = 0,
+- HasTanh = EIGEN_FAST_MATH,
+- HasErf = EIGEN_FAST_MATH,
++ HasRint = 0,
+ #endif
++ HasTanh = 0,
++ HasErf = 0,
+ HasRound = 1,
+ HasFloor = 1,
+ HasCeil = 1,
+- HasRint = 1,
+ HasNegate = 1,
+ HasBlend = 1
+ };
+@@ -872,19 +876,29 @@ template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan
+ return vec_nor(c,c);
+ }
+
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet4i pcmp_le(const Packet4i& a, const Packet4i& b) { return reinterpret_cast<Packet4i>(vec_cmple(a,b)); }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet4i pcmp_lt(const Packet4i& a, const Packet4i& b) { return reinterpret_cast<Packet4i>(vec_cmplt(a,b)); }
+ template<> EIGEN_STRONG_INLINE Packet4i pcmp_eq(const Packet4i& a, const Packet4i& b) { return reinterpret_cast<Packet4i>(vec_cmpeq(a,b)); }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet8s pcmp_le(const Packet8s& a, const Packet8s& b) { return reinterpret_cast<Packet8s>(vec_cmple(a,b)); }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet8s pcmp_lt(const Packet8s& a, const Packet8s& b) { return reinterpret_cast<Packet8s>(vec_cmplt(a,b)); }
+ template<> EIGEN_STRONG_INLINE Packet8s pcmp_eq(const Packet8s& a, const Packet8s& b) { return reinterpret_cast<Packet8s>(vec_cmpeq(a,b)); }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet8us pcmp_le(const Packet8us& a, const Packet8us& b) { return reinterpret_cast<Packet8us>(vec_cmple(a,b)); }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet8us pcmp_lt(const Packet8us& a, const Packet8us& b) { return reinterpret_cast<Packet8us>(vec_cmplt(a,b)); }
+ template<> EIGEN_STRONG_INLINE Packet8us pcmp_eq(const Packet8us& a, const Packet8us& b) { return reinterpret_cast<Packet8us>(vec_cmpeq(a,b)); }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet16c pcmp_le(const Packet16c& a, const Packet16c& b) { return reinterpret_cast<Packet16c>(vec_cmple(a,b)); }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet16c pcmp_lt(const Packet16c& a, const Packet16c& b) { return reinterpret_cast<Packet16c>(vec_cmplt(a,b)); }
+ template<> EIGEN_STRONG_INLINE Packet16c pcmp_eq(const Packet16c& a, const Packet16c& b) { return reinterpret_cast<Packet16c>(vec_cmpeq(a,b)); }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet16uc pcmp_le(const Packet16uc& a, const Packet16uc& b) { return reinterpret_cast<Packet16uc>(vec_cmple(a,b)); }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet16uc pcmp_lt(const Packet16uc& a, const Packet16uc& b) { return reinterpret_cast<Packet16uc>(vec_cmplt(a,b)); }
+ template<> EIGEN_STRONG_INLINE Packet16uc pcmp_eq(const Packet16uc& a, const Packet16uc& b) { return reinterpret_cast<Packet16uc>(vec_cmpeq(a,b)); }
+
+@@ -937,6 +951,7 @@ template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4
+ }
+ 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); }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet4f print<Packet4f>(const Packet4f& a)
+ {
+ Packet4f res;
+@@ -947,6 +962,7 @@ template<> EIGEN_STRONG_INLINE Packet4f print<Packet4f
+
+ return res;
+ }
++#endif
+
+ template<typename Packet> EIGEN_STRONG_INLINE Packet ploadu_common(const __UNPACK_TYPE__(Packet)* from)
+ {
+@@ -1341,12 +1357,6 @@ template<> EIGEN_STRONG_INLINE Packet8bf psub<Packet8b
+ BF16_TO_F32_BINARY_OP_WRAPPER(psub<Packet4f>, a, b);
+ }
+
+-template<> EIGEN_STRONG_INLINE Packet8bf psqrt<Packet8bf> (const Packet8bf& a){
+- BF16_TO_F32_UNARY_OP_WRAPPER(vec_sqrt, a);
+-}
+-template<> EIGEN_STRONG_INLINE Packet8bf prsqrt<Packet8bf> (const Packet8bf& a){
+- BF16_TO_F32_UNARY_OP_WRAPPER(prsqrt<Packet4f>, a);
+-}
+ template<> EIGEN_STRONG_INLINE Packet8bf pexp<Packet8bf> (const Packet8bf& a){
+ BF16_TO_F32_UNARY_OP_WRAPPER(pexp_float, a);
+ }
+@@ -1390,9 +1400,11 @@ template<> EIGEN_STRONG_INLINE Packet8bf pceil<Packet8
+ template<> EIGEN_STRONG_INLINE Packet8bf pround<Packet8bf> (const Packet8bf& a){
+ BF16_TO_F32_UNARY_OP_WRAPPER(pround<Packet4f>, a);
+ }
++#ifdef __VSX__
+ template<> EIGEN_STRONG_INLINE Packet8bf print<Packet8bf> (const Packet8bf& a){
+ BF16_TO_F32_UNARY_OP_WRAPPER(print<Packet4f>, a);
+ }
++#endif
+ template<> EIGEN_STRONG_INLINE Packet8bf pmadd(const Packet8bf& a, const Packet8bf& b, const Packet8bf& c) {
+ Packet4f a_even = Bf16ToF32Even(a);
+ Packet4f a_odd = Bf16ToF32Odd(a);