A floating-point (FP) arithmetic unit includes a first FP execution pipeline operatively coupled to a register file, the first FP execution pipeline configured to perform a first FP operation on a first FP operand provided by the register file, the first FP execution pipeline comprising a plurality of execution units; and a first normalization unit operatively coupled to the register file, and the first FP execution pipeline, the first normalization unit configured to normalize the first FP operand, wherein the first normalization unit is configured to operate in parallel with the first FP execution pipeline, and is further configured to, in response to detecting that the first FP operand is a denormal, assert a first FP execution pipeline busy flag to stall the instruction dispatch of a first subsequent FP operation, the first FP operation and the first subsequent FP operation being of one FP operation type.

Numeric operations on a series of values, each encoded as a logarithm, may be performed. A first offset is determined based on values of an initial subset of the series. Numerical operations on individual elements of the series may then be performed by performing an exponentiation of a difference between the individual elements and the first offset. A particular element in the series may be identified as being unable to be included without causing an overflow or underflow of an exponentiation or numerical operation. In this event, a second offset may be determined. The numerical operations may then proceed by adjusting a current accumulated result using the second offset and continuing to perform numerical operations on individual elements of the series by performing an exponentiation of a difference between the individual elements and the second offset. Additional offsets may be optionally determined until the numerical operation is complete.

An adder circuit inhibiting overflow is provided. A first memory, a second memory, a third memory, and a fourth memory are included. A step of supplying first data with a sign to the first memory and supplying the first data with a positive sign stored in the first memory, to the second memory; a step of supplying the first data with a negative sign stored in the second memory, to the third memory; a step of generating second data by adding the first data with a positive sign stored in the second memory and the first data with a negative sign stored in the third memory; and a step of storing the second data in the fourth memory are included. When the second data stored in the fourth memory are all second data with a positive sign or all second data with a negative sign, all the second data stored in the fourth memory are added.

A system receives and executes a sequence of tensor instructions, for example, instructions for performing a neural network computation. The system may be implemented as a multiprocessor architecture, for example, hardware for performing a neural network computation. A tensor instruction specifies a tensor computation receiving one or more input tensors for determining an output tensor. The system stores a decimal position associated with a plurality of values of a tensor. The system performs the tensor computation of a tensor instruction to determine a plurality of values of the output tensor. The system collects statistics describing the plurality of values of the output tensor and determines a decimal position for the plurality of values based on the collected statistics.

In an embodiment, a division circuit has an overflow determination circuit configured to determine whether or not a division result overflows by comparing absolute values of a dividend and a divisor, a replacement circuit configured to replace the dividend with a first value and replace the divisor with a second value when the overflow determination circuit determines that the division result overflows, and a stepwise division circuit configured to perform stepwise division on the dividend and the divisor or the first value and the second value.

An arithmetic operation device causes a convolution arithmetic unit to perform a convolution arithmetic operation between a filter and target data corresponding to a size of the filter in each of a plurality of convolution layers constituting a neural network. The arithmetic operation device includes: a bit reduction unit that reduces a bit string corresponding to a first bit number from a least significant bit of the target data and reduces a bit string corresponding to a second bit number from a least significant bit of a weight that is an element of the filter for each convolution layer; and a bit addition unit that adds a bit string corresponding to a third bit number obtained by adding the first bit number and the second bit number to a least significant bit of a convolution arithmetic operation result output from the convolution arithmetic unit by inputting the target data and the weight after being reduced by the bit reduction unit to the convolution arithmetic unit.

An adder circuit inhibiting overflow is provided. A first memory, a second memory, a third memory, and a fourth memory are included. A step of supplying first data with a sign to the first memory and supplying the first data with a positive sign stored in the first memory, to the second memory; a step of supplying the first data with a negative sign stored in the second memory, to the third memory; a step of generating second data by adding the first data with a positive sign stored in the second memory and the first data with a negative sign stored in the third memory; and a step of storing the second data in the fourth memory are included. When the second data stored in the fourth memory are all second data with a positive sign or all second data with a negative sign, all the second data stored in the fourth memory are added.

The present disclosure describes a mixed signal arithmetic logic unit configured to use a combination of analog processing elements and digital processing elements in a cohesive manner. Depending on the signals and the data received for processing, the analog processing elements and digital processing elements may be used separately, independently or in combination to optimize computational results and the performance of the mixed signal arithmetic logic unit.

Validating microprocessor instruction execution by receiving a floating-point exception selection, receiving a validation method selection, generating validation data according to the floating-point exception selection and the validation method selection by randomly generating a first tensor element value and randomly generating a second tensor element value according to the first tensor element value and the floating-point exception selection, and executing a floating-point computation according to the validation data.

A method and apparatus for handling overflow conditions resulting from arithmetic operations involving floating point numbers. An indication is stored as part of a thread's context indicating one of two possible modes for handling overflow conditions. In a first mode, a result of an arithmetic operation is set to the limit representable in the floating point format. In a second mode, a result of an arithmetic operation is set to a NaN.