GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-11 19:29:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix up PowerPC MMA flags so it builds by default.

Browse Source
This commit is contained in:
Antonio Sánchez 2022-03-16 19:16:28 +00:00
parent 518fc321cb
commit 591906477b
2 changed files with 115 additions and 108 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

212
Eigen/src/Core/arch/AltiVec/MatrixProduct.h
View File

@ -17,24 +17,28 @@
#include "MatrixProductCommon.h"
// Since LLVM doesn't support dynamic dispatching, force either always MMA or VSX
#if EIGEN_COMP_LLVM
#if !defined(EIGEN_ALTIVEC_DISABLE_MMA) && !defined(EIGEN_ALTIVEC_MMA_ONLY)
#ifdef __MMA__
#define EIGEN_ALTIVEC_MMA_ONLY
#else
#define EIGEN_ALTIVEC_DISABLE_MMA
#endif
#endif
#endif
#ifdef __has_builtin
// Check for MMA builtin support.
#if !defined(EIGEN_ALTIVEC_DISABLE_MMA) && defined(__has_builtin)
#if __has_builtin(__builtin_mma_assemble_acc)
#define ALTIVEC_MMA_SUPPORT
#define EIGEN_ALTIVEC_MMA_SUPPORT
#endif
#endif
#if defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
// Check if and how we should actually use MMA if supported.
#if defined(EIGEN_ALTIVEC_MMA_SUPPORT)
// Check if we want to enable dynamic dispatch. Not supported by LLVM.
if defined(EIGEN_ALTIVEC_ENABLE_MMA_DYNAMIC_DISPATCH) && !EIGEN_COMP_LLVM
#define EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH 1
// Otherwise, use MMA by default if available.
#elif defined(__MMA__)
#define EIGEN_ALTIVEC_MMA_ONLY 1
#endif
#endif // EIGEN_ALTIVEC_MMA_SUPPORT
#if defined(EIGEN_ALTIVEC_MMA_ONLY) || defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
#include "MatrixProductMMA.h"
#endif
@ -2481,10 +2485,10 @@ void gebp_kernel<float, float, Index, DataMapper, mr, nr, ConjugateLhs, Conjugat
const Index accCols = quad_traits<float>::size;
void (*gemm_function)(const DataMapper&, const float*, const float*, Index, Index, Index, float, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemmMMA<float, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemmMMA<float, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
}
@ -2494,7 +2498,7 @@ void gebp_kernel<float, float, Index, DataMapper, mr, nr, ConjugateLhs, Conjugat
#else
gemm_function = &Eigen::internal::gemm<float, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2520,20 +2524,20 @@ void gebp_kernel<std::complex<float>, std::complex<float>, Index, DataMapper, mr
void (*gemm_function)(const DataMapper&, const std::complex<float>*, const std::complex<float>*,
Index, Index, Index, std::complex<float>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2558,20 +2562,20 @@ void gebp_kernel<float, std::complex<float>, Index, DataMapper, mr, nr, Conjugat
const Index accCols = quad_traits<float>::size;
void (*gemm_function)(const DataMapper&, const float*, const std::complex<float>*,
Index, Index, Index, std::complex<float>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<float, std::complex<float>, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2596,20 +2600,20 @@ void gebp_kernel<std::complex<float>, float, Index, DataMapper, mr, nr, Conjugat
const Index accCols = quad_traits<float>::size;
void (*gemm_function)(const DataMapper&, const std::complex<float>*, const float*,
Index, Index, Index, std::complex<float>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<float>, float, std::complex<float>, float, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2633,10 +2637,10 @@ void gebp_kernel<double, double, Index, DataMapper, mr, nr, ConjugateLhs, Conjug
const Index accCols = quad_traits<double>::size;
void (*gemm_function)(const DataMapper&, const double*, const double*, Index, Index, Index, double, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemmMMA<double, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemmMMA<double, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
}
@ -2646,7 +2650,7 @@ void gebp_kernel<double, double, Index, DataMapper, mr, nr, ConjugateLhs, Conjug
#else
gemm_function = &Eigen::internal::gemm<double, Index, Packet, RhsPacket, DataMapper, accRows, accCols>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2671,20 +2675,20 @@ void gebp_kernel<std::complex<double>, std::complex<double>, Index, DataMapper,
const Index accCols = quad_traits<double>::size;
void (*gemm_function)(const DataMapper&, const std::complex<double>*, const std::complex<double>*,
Index, Index, Index, std::complex<double>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2709,20 +2713,20 @@ void gebp_kernel<std::complex<double>, double, Index, DataMapper, mr, nr, Conjug
const Index accCols = quad_traits<double>::size;
void (*gemm_function)(const DataMapper&, const std::complex<double>*, const double*,
Index, Index, Index, std::complex<double>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<std::complex<double>, double, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, false, true>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
template<typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
@ -2747,20 +2751,20 @@ void gebp_kernel<double, std::complex<double>, Index, DataMapper, mr, nr, Conjug
const Index accCols = quad_traits<double>::size;
void (*gemm_function)(const DataMapper&, const double*, const std::complex<double>*,
Index, Index, Index, std::complex<double>, Index, Index, Index, Index);
#ifdef EIGEN_ALTIVEC_MMA_ONLY
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#elif defined(ALTIVEC_MMA_SUPPORT) && !defined(EIGEN_ALTIVEC_DISABLE_MMA)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
#if defined(EIGEN_ALTIVEC_MMA_ONLY)
//generate with MMA only
gemm_function = &Eigen::internal::gemm_complexMMA<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#elif defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma")){
gemm_function = &Eigen::internal::gemm_complexMMA<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
else{
gemm_function = &Eigen::internal::gemm_complex<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
}
#else
gemm_function = &Eigen::internal::gemm_complex<double, std::complex<double>, std::complex<double>, double, Index, Packet, Packetc, RhsPacket, DataMapper, accRows, accCols, ConjugateLhs, ConjugateRhs, true, false>;
#endif
gemm_function(res, blockA, blockB, rows, depth, cols, alpha, strideA, strideB, offsetA, offsetB);
}
} // end namespace internal

11
Eigen/src/Core/arch/AltiVec/MatrixProductMMA.h
View File

@ -11,7 +11,9 @@
#ifndef EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
#define EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
#if !EIGEN_COMP_LLVM
// If using dynamic dispatch, set the CPU target.
#if defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
#pragma GCC push_options
#pragma GCC target("cpu=power10,htm")
#endif
@ -615,12 +617,13 @@ void gemm_complexMMA(const DataMapper& res, const LhsScalar* blockAc, const RhsS
#undef advanceRows
#undef advanceCols
#if !EIGEN_COMP_LLVM
#pragma GCC reset_options
#endif
} // end namespace internal
} // end namespace Eigen
#if defined(EIGEN_ALTIVEC_MMA_DYNAMIC_DISPATCH)
#pragma GCC pop_options
#endif
#endif // EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H