#include <limits>
#include <c10/core/ScalarTypeToTypeMeta.h>
#include "op_plugin/OpApiInterface.h"
#include "torch_npu/csrc/framework/utils/RandomOpAdapter.h"
#include "op_plugin/utils/RandomUtil.h"
#include "op_plugin/utils/op_api_common.h"
#include "op_plugin/AclOpsInterface.h"
#include "torch_npu/csrc/core/npu/NPUGraphsUtils.h"
namespace op_api {
at::Tensor& exponential_(at::Tensor& self, double lambd, c10::optional<at::Generator> generator)
{
if (c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950 && self.scalar_type() != at::kDouble) {
TORCH_CHECK(lambd > 0.0, "npu_sim_exponential_ expects lambd > 0.0, but found lambd=",
lambd, OPS_ERROR(ErrCode::PARAM));
if (std::isinf(lambd)) {
self.zero_();
return self;
}
auto gen = at::get_generator_or_default<at_npu::NPUGeneratorImpl>(generator, at_npu::detail::getDefaultNPUGenerator());
auto counter_offset = op_plugin::utils::calc_final_counter_offset(self);
auto pair = gen->philox_engine_inputs(counter_offset);
int64_t seed = static_cast<int64_t>(pair.first);
int64_t offset = static_cast<int64_t>(pair.second);
int64_t count = self.numel();
ASCEND_LOGI("count:%lld, lambd:%lf, seed:%lld, offset:%lld", count, lambd, seed, offset);
EXEC_NPU_CMD(aclnnSimThreadExponential, self, count, lambd, seed, offset);
return self;
}
TORCH_CHECK(lambd > 0.0, "exponential_ expects lambd > 0.0, but found lambd=",
lambd, OPS_ERROR(ErrCode::PARAM));
if (std::isinf(lambd)) {
self.zero_();
return self;
}
self.uniform_(0.0, 1.0, generator);
if (self.dtype() == at::kDouble) {
self = op_api::sub_(self, at::Scalar(1.0), at::Scalar(1.0));
self = op_api::mul_(self, at::Scalar(-1.0));
self = op_api::log_(self);
self = op_api::div_(self, at::Scalar(-lambd));
auto eps = std::numeric_limits<double>::min();
self = self.add(eps);
return self;
}
self.neg_();
self.add_(1.0);
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16,
self.scalar_type(), "exponential_", [&]() {
auto eps = std::numeric_limits<scalar_t>::epsilon() / 2;
auto mask = self >= (1.0 - eps);
self.masked_fill_(mask, 1.0 - eps);
});
self.log_();
self.mul_(-1.0 / lambd);
return self;
}
}