#include "op_plugin/AclOpsInterface.h"
#include "op_plugin/OpApiInterface.h"
#include "op_plugin/utils/op_api_common.h"
namespace op_api {
constexpr int MAX_DIM = 1;
static std::vector<int64_t> get_index_vector(const at::Tensor& index)
{
std::vector<int64_t> idx_vec;
if (index.dim() == 0) {
idx_vec.emplace_back(static_cast<int64_t>(index.item().to<int>()));
} else {
for (int64_t i = 0; i < index.sizes()[0]; i++) {
int64_t idx = static_cast<int64_t>(index[i].item().to<int>());
idx_vec.emplace_back(idx);
}
}
return idx_vec;
}
at::Tensor& index_fill_(at::Tensor& self, int64_t dim, const at::Tensor& index,
const at::Tensor& value)
{
TORCH_CHECK(value.dim() == 0, "Value should be a 0-dimensional tensor, but got ",
value.dim(), OPS_ERROR(ErrCode::VALUE));
TORCH_CHECK(self.is_complex() || !value.is_complex(),
"index_fill_(): Converting complex Scalar to non-complex type is not supported",
PTA_ERROR(ErrCode::NOT_SUPPORT));
static const bool is_inplace_index_fill_available = check_aclnn_kernel_available("aclnnInplaceIndexFill");
if (!is_inplace_index_fill_available ||
(c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950 && !op_plugin::utils::is_gte_cann_version_910())) {
DO_COMPATIBILITY(aclnnInplaceIndexFillTensor, acl_op::index_fill_(self, dim, index, value));
TORCH_CHECK(index.dim() <= MAX_DIM, "Index has to be a vector/scalar.", OPS_ERROR(ErrCode::TYPE));
at::Scalar value_scalar = value.item();
std::vector<int64_t> idx_vec = get_index_vector(index);
at::IntArrayRef index_array = at::IntArrayRef(idx_vec);
EXEC_NPU_CMD(aclnnInplaceIndexFillTensor, self, dim, index_array, value_scalar);
} else {
DO_COMPATIBILITY(aclnnInplaceIndexFill, acl_op::index_fill_(self, dim, index, value));
at::Scalar value_scalar = value.item();
EXEC_NPU_CMD(aclnnInplaceIndexFill, self, dim, index, value_scalar);
}
return self;
}
at::Tensor index_fill(const at::Tensor& self, int64_t dim, const at::Tensor& index,
const at::Tensor& value)
{
TORCH_CHECK(value.dim() == 0, "Value should be a 0-dimensional tensor, but got ",
value.dim(), OPS_ERROR(ErrCode::VALUE));
TORCH_CHECK(self.is_complex() || !value.is_complex(),
"index_fill(): Converting complex Scalar to non-complex type is not supported",
PTA_ERROR(ErrCode::NOT_SUPPORT));
static const bool is_index_fill_available = check_aclnn_kernel_available("aclnnIndexFill");
if (!is_index_fill_available || c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950) {
DO_COMPATIBILITY(aclnnIndexFillTensor, acl_op::index_fill(self, dim, index, value));
TORCH_CHECK(index.dim() <= MAX_DIM, "Index has to be a vector/scalar.", OPS_ERROR(ErrCode::TYPE));
at::Scalar value_scalar = value.item();
std::vector<int64_t> idx_vec = get_index_vector(index);
at::IntArrayRef index_array = at::IntArrayRef(idx_vec);
at::Tensor result = at_npu::native::OpPreparation::apply_tensor_without_format(self);
EXEC_NPU_CMD(aclnnIndexFillTensor, self, dim, index_array, value_scalar, result);
at::namedinference::propagate_names(result, self);
return result;
} else {
DO_COMPATIBILITY(aclnnIndexFill, acl_op::index_fill(self, dim, index, value));
at::Scalar value_scalar = value.item();
at::Tensor result = at_npu::native::OpPreparation::apply_tensor_without_format(self);
EXEC_NPU_CMD(aclnnIndexFill, self, dim, index, value_scalar, result);
at::namedinference::propagate_names(result, self);
return result;
}
}
at::Tensor& index_fill_(at::Tensor& self, int64_t dim, const at::Tensor& index,
const at::Scalar& value)
{
TORCH_CHECK(self.is_complex() || !value.isComplex(),
"index_fill_(): Converting complex Scalar to non-complex type is not supported",
PTA_ERROR(ErrCode::NOT_SUPPORT));
static const bool is_inplace_index_fill_available = check_aclnn_kernel_available("aclnnInplaceIndexFill");
if (!is_inplace_index_fill_available ||
(c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950 && !op_plugin::utils::is_gte_cann_version_910())) {
DO_COMPATIBILITY(aclnnInplaceIndexFillTensor, acl_op::index_fill_(self, dim, index, value));
TORCH_CHECK(index.dim() <= MAX_DIM, "Index has to be a vector/scalar.", OPS_ERROR(ErrCode::TYPE));
std::vector<int64_t> idx_vec = get_index_vector(index);
at::IntArrayRef index_array = at::IntArrayRef(idx_vec);
EXEC_NPU_CMD(aclnnInplaceIndexFillTensor, self, dim, index_array, value);
} else {
DO_COMPATIBILITY(aclnnInplaceIndexFill, acl_op::index_fill_(self, dim, index, value));
EXEC_NPU_CMD(aclnnInplaceIndexFill, self, dim, index, value);
}
return self;
}
at::Tensor index_fill(const at::Tensor& self, int64_t dim, const at::Tensor& index,
const at::Scalar& value)
{
TORCH_CHECK(self.is_complex() || !value.isComplex(),
"index_fill(): Converting complex Scalar to non-complex type is not supported",
PTA_ERROR(ErrCode::NOT_SUPPORT));
static const bool is_index_fill_available = check_aclnn_kernel_available("aclnnIndexFill");
if (!is_index_fill_available || c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950) {
DO_COMPATIBILITY(aclnnIndexFillTensor, acl_op::index_fill(self, dim, index, value));
TORCH_CHECK(index.dim() <= MAX_DIM, "Index has to be a vector/scalar.", OPS_ERROR(ErrCode::TYPE));
std::vector<int64_t> idx_vec = get_index_vector(index);
at::IntArrayRef index_array = at::IntArrayRef(idx_vec);
at::Tensor result = at_npu::native::OpPreparation::apply_tensor_without_format(self);
EXEC_NPU_CMD(aclnnIndexFillTensor, self, dim, index_array, value, result);
at::namedinference::propagate_names(result, self);
return result;
} else {
DO_COMPATIBILITY(aclnnIndexFill, acl_op::index_fill(self, dim, index, value));
at::Tensor result = at_npu::native::OpPreparation::apply_tensor_without_format(self);
EXEC_NPU_CMD(aclnnIndexFill, self, dim, index, value, result);
at::namedinference::propagate_names(result, self);
return result;
}
}
}