wehub-resource-sync
304 lines
11 KiB
Plaintext
304 lines
11 KiB
Plaintext
// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#ifdef PADDLE_WITH_HIP
|
|
#include "paddle/phi/backends/dynload/rocsolver.h"
|
|
#else
|
|
#include "paddle/phi/backends/dynload/cusolver.h"
|
|
#endif
|
|
|
|
#include "paddle/phi/backends/gpu/gpu_context.h"
|
|
#include "paddle/phi/core/enforce.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/impl/lu_kernel_impl.h"
|
|
#include "paddle/phi/kernels/lu_solve_kernel.h"
|
|
|
|
namespace phi {
|
|
|
|
#ifdef PADDLE_WITH_HIP
|
|
template <typename T>
|
|
void rocsolver_getrs(const solverHandle_t& handle,
|
|
rocblas_operation trans,
|
|
int n,
|
|
int nrhs,
|
|
T* a,
|
|
int lda,
|
|
int* ipiv,
|
|
T* b,
|
|
int ldb);
|
|
|
|
template <>
|
|
void rocsolver_getrs<float>(const solverHandle_t& handle,
|
|
rocblas_operation trans,
|
|
int n,
|
|
int nrhs,
|
|
float* a,
|
|
int lda,
|
|
int* ipiv,
|
|
float* b,
|
|
int ldb) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::rocsolver_sgetrs(handle, trans, n, nrhs, a, lda, ipiv, b, ldb));
|
|
}
|
|
|
|
template <>
|
|
void rocsolver_getrs<double>(const solverHandle_t& handle,
|
|
rocblas_operation trans,
|
|
int n,
|
|
int nrhs,
|
|
double* a,
|
|
int lda,
|
|
int* ipiv,
|
|
double* b,
|
|
int ldb) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::rocsolver_dgetrs(handle, trans, n, nrhs, a, lda, ipiv, b, ldb));
|
|
}
|
|
|
|
template <>
|
|
void rocsolver_getrs<dtype::complex<float>>(const solverHandle_t& handle,
|
|
rocblas_operation trans,
|
|
int n,
|
|
int nrhs,
|
|
dtype::complex<float>* a,
|
|
int lda,
|
|
int* ipiv,
|
|
dtype::complex<float>* b,
|
|
int ldb) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::rocsolver_cgetrs(handle,
|
|
trans,
|
|
n,
|
|
nrhs,
|
|
reinterpret_cast<rocblas_float_complex*>(a),
|
|
lda,
|
|
ipiv,
|
|
reinterpret_cast<rocblas_float_complex*>(b),
|
|
ldb));
|
|
}
|
|
|
|
template <>
|
|
void rocsolver_getrs<dtype::complex<double>>(const solverHandle_t& handle,
|
|
rocblas_operation trans,
|
|
int n,
|
|
int nrhs,
|
|
dtype::complex<double>* a,
|
|
int lda,
|
|
int* ipiv,
|
|
dtype::complex<double>* b,
|
|
int ldb) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::rocsolver_zgetrs(handle,
|
|
trans,
|
|
n,
|
|
nrhs,
|
|
reinterpret_cast<rocblas_double_complex*>(a),
|
|
lda,
|
|
ipiv,
|
|
reinterpret_cast<rocblas_double_complex*>(b),
|
|
ldb));
|
|
}
|
|
|
|
#else
|
|
template <typename T>
|
|
void cusolver_getrs(const solverHandle_t& handle,
|
|
cublasOperation_t trans,
|
|
int n,
|
|
int nrhs,
|
|
T* a,
|
|
int lda,
|
|
int* ipiv,
|
|
T* b,
|
|
int ldb,
|
|
int* info);
|
|
|
|
template <>
|
|
void cusolver_getrs<float>(const solverHandle_t& handle,
|
|
cublasOperation_t trans,
|
|
int n,
|
|
int nrhs,
|
|
float* a,
|
|
int lda,
|
|
int* ipiv,
|
|
float* b,
|
|
int ldb,
|
|
int* info) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(dynload::cusolverDnSgetrs(
|
|
handle, trans, n, nrhs, a, lda, ipiv, b, ldb, info));
|
|
}
|
|
|
|
template <>
|
|
void cusolver_getrs<double>(const solverHandle_t& handle,
|
|
cublasOperation_t trans,
|
|
int n,
|
|
int nrhs,
|
|
double* a,
|
|
int lda,
|
|
int* ipiv,
|
|
double* b,
|
|
int ldb,
|
|
int* info) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(dynload::cusolverDnDgetrs(
|
|
handle, trans, n, nrhs, a, lda, ipiv, b, ldb, info));
|
|
}
|
|
|
|
template <>
|
|
void cusolver_getrs<dtype::complex<float>>(const solverHandle_t& handle,
|
|
cublasOperation_t trans,
|
|
int n,
|
|
int nrhs,
|
|
dtype::complex<float>* a,
|
|
int lda,
|
|
int* ipiv,
|
|
dtype::complex<float>* b,
|
|
int ldb,
|
|
int* info) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::cusolverDnCgetrs(handle,
|
|
trans,
|
|
n,
|
|
nrhs,
|
|
reinterpret_cast<cuComplex*>(a),
|
|
lda,
|
|
ipiv,
|
|
reinterpret_cast<cuComplex*>(b),
|
|
ldb,
|
|
info));
|
|
}
|
|
|
|
template <>
|
|
void cusolver_getrs<dtype::complex<double>>(const solverHandle_t& handle,
|
|
cublasOperation_t trans,
|
|
int n,
|
|
int nrhs,
|
|
dtype::complex<double>* a,
|
|
int lda,
|
|
int* ipiv,
|
|
dtype::complex<double>* b,
|
|
int ldb,
|
|
int* info) {
|
|
PADDLE_ENFORCE_GPU_SUCCESS(
|
|
dynload::cusolverDnZgetrs(handle,
|
|
trans,
|
|
n,
|
|
nrhs,
|
|
reinterpret_cast<cuDoubleComplex*>(a),
|
|
lda,
|
|
ipiv,
|
|
reinterpret_cast<cuDoubleComplex*>(b),
|
|
ldb,
|
|
info));
|
|
}
|
|
|
|
#endif // PADDLE_WITH_HIP
|
|
|
|
template <typename T, typename Context>
|
|
void LuSolveKernel(const Context& dev_ctx,
|
|
const DenseTensor& b,
|
|
const DenseTensor& lu,
|
|
const DenseTensor& pivots,
|
|
const std::string& trans,
|
|
DenseTensor* out) {
|
|
dev_ctx.template Alloc<T>(out);
|
|
// Copy x to out since cusolverDn*getrs overwrites the input
|
|
*out = Transpose2DTo6D<Context, T>(dev_ctx, b);
|
|
DenseTensor tem_lu = Transpose2DTo6D<Context, T>(dev_ctx, lu);
|
|
// Validate input dimensions
|
|
auto x_dims = b.dims();
|
|
auto lu_dims = lu.dims();
|
|
|
|
#ifdef PADDLE_WITH_HIP
|
|
rocblas_operation trans_op;
|
|
if (trans == "N") {
|
|
trans_op = rocblas_operation_none;
|
|
} else if (trans == "T") {
|
|
trans_op = rocblas_operation_transpose;
|
|
} else if (trans == "C") {
|
|
trans_op = rocblas_operation_conjugate_transpose;
|
|
} else {
|
|
PADDLE_THROW(common::errors::InvalidArgument(
|
|
"trans must be one of ['N', 'T', 'C'], but got %s", trans));
|
|
}
|
|
#else
|
|
cublasOperation_t trans_op;
|
|
if (trans == "N") {
|
|
trans_op = CUBLAS_OP_N;
|
|
} else if (trans == "T") {
|
|
trans_op = CUBLAS_OP_T;
|
|
} else if (trans == "C") {
|
|
trans_op = CUBLAS_OP_C;
|
|
} else {
|
|
PADDLE_THROW(common::errors::InvalidArgument(
|
|
"trans must be one of ['N', 'T', 'C'], but got %s", trans));
|
|
}
|
|
#endif
|
|
int n = static_cast<int>(lu_dims[lu_dims.size() - 1]);
|
|
int nrhs = static_cast<int>(x_dims[x_dims.size() - 1]);
|
|
int lda = std::max(1, n);
|
|
int ldb = std::max(1, n);
|
|
|
|
DenseTensor info_tensor;
|
|
info_tensor.Resize({1});
|
|
dev_ctx.template Alloc<int>(&info_tensor);
|
|
int* d_info = info_tensor.data<int>();
|
|
|
|
auto outdims = out->dims();
|
|
auto outrank = outdims.size();
|
|
int batchsize = product(slice_ddim(outdims, 0, outrank - 2));
|
|
auto out_data = out->data<T>();
|
|
auto lu_data = reinterpret_cast<T*>(const_cast<T*>(tem_lu.data<T>()));
|
|
auto pivots_data =
|
|
reinterpret_cast<int*>(const_cast<int*>(pivots.data<int>()));
|
|
for (int i = 0; i < batchsize; i++) {
|
|
auto handle = dev_ctx.cusolver_dn_handle();
|
|
auto* out_data_item = &out_data[i * lda * nrhs];
|
|
auto* lu_data_item = &lu_data[i * ldb * n];
|
|
auto* pivots_data_item = &pivots_data[i * n];
|
|
#ifdef PADDLE_WITH_HIP
|
|
rocsolver_getrs<T>(handle,
|
|
trans_op,
|
|
n,
|
|
nrhs,
|
|
lu_data_item,
|
|
lda,
|
|
pivots_data_item,
|
|
out_data_item,
|
|
ldb);
|
|
#else
|
|
cusolver_getrs<T>(handle,
|
|
trans_op,
|
|
n,
|
|
nrhs,
|
|
lu_data_item,
|
|
lda,
|
|
pivots_data_item,
|
|
out_data_item,
|
|
ldb,
|
|
d_info);
|
|
#endif
|
|
}
|
|
*out = Transpose2DTo6D<Context, T>(dev_ctx, *out);
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(lu_solve,
|
|
GPU,
|
|
ALL_LAYOUT,
|
|
phi::LuSolveKernel,
|
|
float,
|
|
double,
|
|
phi::complex64,
|
|
phi::complex128) {}
|