Abstract: A convolutional neural network (CNN) accelerator, including: a CNN circuit for performing a multiple-layer CNN computation, wherein the multiple layers are to receive an input feature according to an input feature map (IFM) and a weight matrix per output feature, wherein an output of a first layer provides an input for a next layer; and a mapping circuit to access a three-dimensional input matrix stored as a Z-major matrix; wherein the CNN circuit is to perform an inner-product direct convolution on the Z-major matrix, wherein the direct convolution lacks a lowering operation.

Abstract: A processing device includes a zero detection circuit to determine that an operand of a first instruction is zero and instruction conversion logic coupled with the zero detection circuit to, in response to the zero detection circuit determining that the operand is zero, convert the first instruction to a register move instruction executable by the processing device.

Abstract: An arithmetic circuitry includes a first processing circuitry, a second processing circuitry, an adder circuitry, and a saturation logic circuitry. The first processing circuitry divides one input term into blocks each of which being divided for each predetermined digit number, to make a least significant bit of each of the blocks overlap with a most significant bit of the adjacent and low-order block, and calculates a partial product of each of the blocks and the other input term based on Booth recoding in which a sign is controlled when Booth recoding values become ±0. The second processing circuitry simplifies the partial products. The adder circuitry outputs the sum of a result obtained through the simplification and an addition term. The saturation logic circuitry executes saturation processing based on a result outputted by the second processing circuitry and a result outputted by the adder circuitry.

Abstract: Techniques are described for metadata processing that can be used to encode an arbitrary number of security policies for code running on a processor. Metadata may be added to every word in the system and a metadata processing unit may be used that works in parallel with data flow to enforce an arbitrary set of policies. In one aspect, the metadata may be characterized as unbounded and software programmable to be applicable to a wide range of metadata processing policies. Techniques and policies have a wide range of uses including, for example, safety, security, and synchronization. Additionally, described are aspects and techniques in connection with metadata processing in an embodiment based on the RISC-V architecture.

Abstract: Examples of the present disclosure provide apparatuses and methods for multiple endianness compatibility. An example method comprises receiving a plurality of bytes and determining a particular endianness format of the plurality of bytes. The method can include, responsive to determining the particular endianness format is a first endianness format, reordering bits of each byte of the plurality of bytes on a bytewise basis, storing the reordered plurality of bytes in an array of memory cells, and adjusting a shift direction associated with performing a number of operations on the plurality of bytes stored in the array. The method can include, responsive to determining the particular endianness format is a second endianness format, storing the plurality of bytes in the array without reordering bits of the plurality of bytes.

Abstract: Techniques are described for metadata processing that can be used to encode an arbitrary number of security policies for code running on a processor. Metadata may be added to every word in the system and a metadata processing unit nay be used that works in parallel with data flow to enforce an arbitrary set of policies. In one aspect, the metadata may be characterized as unbounded and software programmable to be applicable to a wide range of metadata processing policies. Techniques and policies have a wide range of uses including, for example, safety, security, and synchronization. Additionally, described are aspects and techniques in connection with metadata processing in an embodiment based on the RISC-V architecture.

Abstract: One embodiment provides for a graphics processing unit to perform computations associated with a neural network, the graphics processing unit comprising compute unit including a hardware logic unit having dynamic precision fixed-point logic; a decode unit to decode an instruction for execution by the compute unit, the instruction to cause the compute unit to perform a matrix arithmetic operation on a set of dynamic fixed-point tensors; and a dynamic precision manager to dynamically adjust the precision of a compute operation performed by the compute unit during the matrix arithmetic operation, the dynamic precision manager to adjust the precision of the compute operation to prevent an arithmetic overflow.

Abstract: Examples of techniques for designing processors are described herein. In one example, a design structure can be tangibly embodied in a machine readable medium for designing, manufacturing, or testing an integrated circuit. The design structure can include a logic to determine whether a received instruction is an updating fixed point instruction or a non-updating fixed point instruction. The design structure can include a first arithmetic logic unit (ALU) to execute the received instruction if the received instruction is determined to be an updating fixed point instruction and store an update value in a general register. The design structure can include a second arithmetic logic unit (ALU) to execute the received instruction if the received instruction is determined to be a non-updating fixed point instruction.

Abstract: One embodiment of this invention provides two conditional execution auxiliary instructions directed to disparate subsets of the plural functional units. Depending on the conditional execution desired, only one of the two conditional execution auxiliary instructions may be required for a particular execute packet. Another embodiment of this invention employs only one of two possible register files for the condition registers. In a VLIW processor it may be advantageous to split the functional units into separate sets with corresponding register files. This limits the number of functional units that may simultaneously access the register files. In the preferred embodiment of this invention the functional units are divided into a scalar set which access scalar registers and a vector set which access vector registers. The data registers storing the conditions for both scalar and vector instructions are in the scalar data register file.

Abstract: Techniques are described for metadata processing that can be used to encode an arbitrary number of security policies for code running on a processor. Metadata may be added to every word in the system and a metadata processing unit may be used that works in parallel with data flow to enforce an arbitrary set of policies. In one aspect, the metadata may be characterized as unbounded and software programmable to be applicable to a wide range of metadata processing policies. Techniques and policies have a wide range of uses including, for example, safety, security, and synchronization. Additionally, described are aspects and techniques in connection with metadata processing in an embodiment based on the RISC-V architecture.

Abstract: A logic circuit and a method using thereof for zero detection of a sum of inputs without performing an addition. The logic circuit and the method using thereof perform a true mathematical zero check and a zero check for trailing N-bits. The logic circuit and the method using thereof include bitwise XOR, XNOR, and OR operations, an OR-reduction, an AND reduction, and a control signal that switches between a true mathematical zero check and a zero check for trailing N-bits. The logic circuit and the method using thereof have less timing delay than an adder or a leading zero anticipator for a zero check. The logic circuit and the method using thereof use less logic gates and therefore less area and less power are needed. The logic circuit and the method using thereof have a great advantage for the zero check of large input vectors.

Abstract: A processor includes an instruction schedule and dispatch (schedule/dispatch) unit to receive a single instruction multiple data (SIMD) instruction to perform an operation on multiple data elements stored in a storage location indicated by a first source operand. The instruction schedule/dispatch unit is to determine a first of the data elements that will not be operated to generate a result written to a destination operand based on a second source operand. The processor further includes multiple processing elements coupled to the instruction schedule/dispatch unit to process the data elements of the SIMD instruction in a vector manner, and a power management unit coupled to the instruction schedule/dispatch unit to reduce power consumption of a first of the processing elements configured to process the first data element.

Abstract: A hardware accelerator includes a bitmap processor that processes a bitmap structure for multiple list population. A population count processor processes population counts for data and aggregates the population counts. The bitmap data structure includes bitmap bit fields interleaved with aggregated population count fields.

Abstract: A logic circuit and a method using thereof for zero detection of a sum of inputs without performing an addition. The logic circuit and the method using thereof perform a true mathematical zero check and a zero check for trailing N-bits. The logic circuit and the method using thereof include bitwise XOR, XNOR, and OR operations, an OR-reduction, an AND reduction, and a control signal that switches between a true mathematical zero check and a zero check for trailing N-bits. The logic circuit and the method using thereof have less timing delay than an adder or a leading zero anticipator for a zero check. The logic circuit and the method using thereof use less logic gates and therefore less area and less power are needed. The logic circuit and the method using thereof have a great advantage for the zero check of large input vectors.

Abstract: Examples of the present disclosure provide apparatuses and methods for multiple endianness compatibility. An example method comprises receiving a plurality of bytes and determining a particular endianness format of the plurality of bytes. The method can include, responsive to determining the particular endianness format is a first endianness format, reordering bits of each byte of the plurality of bytes on a bytewise basis, storing the reordered plurality of bytes in an array of memory cells, and adjusting a shift direction associated with performing a number of operations on the plurality of bytes stored in the array. The method can include, responsive to determining the particular endianness format is a second endianness format, storing the plurality of bytes in the array without reordering bits of the plurality of bytes.

Abstract: A method validates a formula within a spreadsheet. A formula type, of an identified formula from a spreadsheet, is identified. A database of known accurate formulas, whose type matches a formula type of the identified formula within the spreadsheet, is identified and located. A determination is made as to whether the identified formula from the spreadsheet matches a known accurate formula from the database. In response to the identified formula from the spreadsheet failing to match the known accurate formula from the database, a predefined action is initiated.

Abstract: A unified computation unit for iterative multiplication and division may include an architecture having a unified integer iterative multiplication and division circuit. A method may include a device receiving a dividend and a divisor for a division operation, separating the dividend into two parts based on the determining, and evaluating whether an overflow situation exists based on the two parts. A single-cycle multiplication unit may include a multi-operand addition schema for partial products compression that implements tree-based addition methods for single-cycle multiplication operations.

Abstract: A high-power-efficiency multiplier combines a standard floating-point multiplier with a power-of-two multiplier that performs multiplications by shifting operations without the need for floating-point multiplication circuitry. By selectively steering some operands to this power-of-two multiplier, substantial power savings may be realized. In one embodiment, multiplicands may be modified to work with the power-of-two multiplier introducing low errors that may be accommodated in pixel calculations.

Abstract: Configurable specialized processing blocks, such as DSP blocks, are described that implement fixed and floating-point functionality in a single mixed architecture on a programmable device. The described architecture reduces the need to construct floating-point functions outside the configurable specialized processing block, thereby minimizing hardware cost and area. The disclosed architecture also introduces pipelining into the DSP block in order to ensure the floating-point multiplication and addition functions remain in synchronicity, thereby increasing the maximum frequency at which the DSP block can operate. Moreover, the disclosed architecture includes logic circuitry to support floating-point exception handling.

Abstract: In one embodiment, a microprocessor includes fetch logic for retrieving an instruction, decode logic configured to identify a plurality of operands and a multiply operation specified in the instruction, and execution logic configured to receive the plurality of operands and the multiply operation. The execution logic includes a first logic path configured to perform the multiply operation on the plurality of operands and output a result, and a second logic path, arranged in parallel with the first logic path, configured to output metadata associated with the result of the multiply operation.

Abstract: A system for performing public key encryption is provided. The system supports mathematical operations for a plurality of public key encryption algorithms such as Rivert, Shamir, Aldeman (RSA) and Diffie-Hellman key exchange (DH) and Elliptic Curve Cryptosystem (ECC). The system supports both prime fields and different composite binary fields.

Abstract: A predictive adder produces the result of incrementing and/or decrementing a sum of A and B by a one-bit constant of the form of the form 2k, where k is a bit position at which the sum is to be incremented or decremented. The predictive adder predicts the ripple portion of bits in the potential sum of the first operand A and the second operand B that would be toggled by incrementing or decrementing the sum A+B by the one-bit constant to generate and indication of the ripple portion of bits in the potential sum. The predictive adder uses the indication of the ripple portion of bits in the potential sum and the carry output generated by evaluating A+B to produce the results of at least one of A+B+2k and A+B?2k.

Abstract: In one embodiment, a microprocessor includes fetch logic for retrieving an instruction, decode logic configured to identify an arithmetic operation specified in the instruction, and execution logic configured to receive operands specified by the instruction. The execution logic includes a primary logic path configured to perform the arithmetic operation on such operands and a secondary parallel logic path configured to output metadata associated with the result of the arithmetic operation.

Abstract: Systems and methods for encoding floating point numbers. A system can include encoding logic which encodes invalid floating point representations as valid data. Decoding logic can be used to recognize the invalid floating point representations and map can provide the invalid floating point representations to valid data values. The decoding logic then can provide the valid data values so that operations on the valid data values can be performed in accordance with instructions received from an associated program.

Abstract: Detection of whether a result of a floating point operation is safe. Characteristics of the result are examined to determine whether the result is safe or potentially unsafe, as defined by the user. An instruction is provided to facilitate detection of safe or potentially unsafe results.

Abstract: Detection of whether a result of a floating point operation is safe. Characteristics of the result are examined to determine whether the result is safe or potentially unsafe, as defined by the user. An instruction is provided to facilitate detection of safe or potentially unsafe results.

Abstract: To perform a binary-coded decimal (BCD) calculation, a processor receives values on which the BCD calculation is to be performed. A carry resulting from the BCD calculation is stored in a flag register of the processor, and the carry stored in the flag register is used to compute a result of the BCD calculation.

Abstract: A predictive adder produces the result of incrementing and/or decrementing a sum of A and B by a one-bit constant of the form of the form 2k, where k is a bit position at which the sum is to be incremented or decremented. The predictive adder predicts the ripple portion of bits in the potential sum of the first operand A and the second operand B that would be toggled by incrementing or decrementing the sum A+B by the one-bit constant to generate and indication of the ripple portion of bits in the potential sum. The predictive adder uses the indication of the ripple portion of bits in the potential sum and the carry output generated by evaluating A+B to produce the results of at least one of A+B+2k and A+B?2k.

Abstract: The present invention relates to improved imaging devices having high dynamic range and to monitoring and automatic control systems incorporating the improved imaging devices.

Abstract: A specialized processing block for a programmable logic device includes circuitry for performing multiplications and sums thereof, as well as circuitry for performing floating point operations. The floating point circuitry preferably includes rounding and normalization circuitry. To perform mantissa multiplications, the floating point circuitry preferably relies on the aforementioned multipliers of the specialized processing block.

Abstract: Methods and arrangements to correct for double rounding errors when rounding floating point numbers to nearest away are described. Embodiments include transformations, code, state machines or other logic to perform a floating point operation on one or more floating point numbers of precision P1 in base b, producing positive result res0 of precision greater than precision P1; rounding positive result res0 to precision P1 to the nearest away, producing positive result res1; and rounding the result res1 to precision P2 to the nearest away, where P2 is narrower than P1, producing result res2. The embodiments may also include correcting res2 for double rounding errors. The correcting may include determining that res1 is midway between two consecutive floating point numbers of precision P2, the larger being res2, determining that rounding res0 to produce res1 involved rounding up, and decrementing the significand of res2 to obtain the corrected result.

Abstract: Methods and arrangements to correct for double rounding errors when rounding floating point numbers to nearest even are described. Embodiments include transformations, code, state machines or other logic to perform a floating point operation on one or more floating point numbers of precision P1 in base b, producing positive result res0 of precision greater than precision P1; rounding positive result res0 to precision P1 to the nearest even, producing positive result res1; and rounding the result res1 to precision P2 to the nearest even, where P2 is narrower than P1, producing result res2. The embodiments may also include correcting res2 for double rounding errors. The correcting may include determining that res1 is midway between two consecutive floating point numbers of precision P1, the larger (smaller) being res2, determining that rounding res0 to produce res1 involved rounding up (down), and decrementing (incrementing) the significand of res2 to obtain the corrected result res2?.

Abstract: In one embodiment, a state determiner comprises a plurality of logic circuits and a second logic circuit. Each logic circuit corresponds to a respective bit position of a result of an adder. A first logic circuit corresponds to a least significant bit of the result and is coupled to receive a least significant bit of each operand of the adder and a carry-in input to the adder. Each remaining logic circuit is coupled to receive a bit from the respective bit position of each operand and a less significant bit adjacent to the bit of each operand. Each logic circuit generates an output signal indicative of whether or not a specific result occurs in the respective bit position of the result. Coupled to receive the output signals second logic circuit is configured to generate a sum signal indicative, when asserted, that the specific result occurs.

Abstract: Modulus scaling applied a reduction techniques decreases time to perform modular arithmetic operations by avoiding shifting and multiplication operations. Modulus scaling may be applied to both integer and binary fields and the scaling multiplier factor is chosen based on a selected reduction technique for the modular arithmetic operation.

Abstract: Time to perform scalar point multiplication used for ECC is reduced by minimizing the number of shifting operations. These operations are minimized by applying modulus scaling by performing selective comparisons of points at intermediate computations based on primality of the order of an ECC group.

Abstract: The computation time to perform scalar point multiplication in an Elliptic Curve Group is reduced by modifying the Barrett Reduction technique. Computations are performed using an N-bit scaled modulus based a modulus m having k-bits to provide a scaled result, with N being greater than k. The N-bit scaled result is reduced to a k-bit result using a pre-computed N-bit scaled reduction parameter in an optimal manner avoiding shifting/aligning operations for any arbitrary values of k, N.

Abstract: Detection of whether a result of a floating point operation is safe. Characteristics of the result are examined to determine whether the result is safe or potentially unsafe, as defined by the user. An instruction is provided to facilitate detection of safe or potentially unsafe results.

Abstract: To perform a binary-coded decimal (BCD) calculation, a processor receives values on which the BCD calculation is to be performed. A carry resulting from the BCD calculation is stored in a flag register of the processor, and the carry stored in the flag register is used to compute a result of the BCD calculation.

Abstract: One or more detectors are provided for processing input in parallel with a logic component receiving the same input. Apparatus described herein include one or more logic components that are configured to perform logical operations on an input vector, and one or more detectors that are configured to receive a portion of the input vector. The detector is further configured to perform detections on this portion of the input vector in parallel with the logical operation. Methods described herein include identifying a portion of the input vector, wherein the portion of the input vector appears an output of the logic component, and analyzing the portion of the vector in parallel with a logical operation performed by the logic component.

Abstract: A semiconductor device including a first processing unit processing an input signal based on a plurality of image compression standards, a signal generation unit outputting a switching signal to the first processing unit, a first calculation unit performing an operation on the input signal in accordance with a first coefficient that is based on the switching signal, a second calculation unit performing an operation on an output of the first calculation unit in accordance with a second coefficient that is based on the switching signal, a selection unit selecting one of the output of the first calculation unit and an output of the second calculation unit based on the switching signal, and a third calculation unit selecting one or both of the input signal and the output of the first calculation unit based on the switching signal and performing a predetermined calculation on the selected signal.

Abstract: Data transfer bus charging/discharging current is reduced in a semiconductor memory device. In a data transfer device that sequentially transfers bit sequences in parallel through a plurality of buses from a transmit unit 10 to a receive unit 20, the transmit circuit 10 includes a flag generation circuit 11 and an encoding circuit 12. The flag generation circuit 11 generates a flag indicating whether bit inversion has occurred in consecutive bits in each of the bit sequences to be transferred through the buses and transmits the generated flag to the receive unit 20. The encoding circuit 12 encodes the bit sequences based on the flag, for transmission to the receive unit 20. The receive unit includes a decoding circuit 21 that decodes the bit sequences based on the bit sequences and the flag.

Abstract: A floating point flag combining or accumulating circuit includes an analysis circuit that receives a plurality of floating point operands, each having encoded status flag information, and a result assembler. The analysis circuit analyzes the plurality of floating point operands and provides an indication of one or more predetermined formats in which the plurality of floating point operands are represented. The result assembler receives the indication from the analysis circuit and assembles an accumulated result that represents a value and combines the encoded status flag information from at least two of the plurality of floating point operands.

Abstract: A system for providing a floating point square root comprises an analyzer circuit configured to determine a first status of a first floating point operand based upon data within the first floating point operand. In addition, the system comprises a results circuit coupled to the analyzer circuit. The results circuit is configured to assert a resulting floating point operand containing the square root of the first floating point operand and a resulting status embedded within the resulting floating point operand.

Abstract: A floating point unit generates results in which status information generated for an operation is encoded within the resulting operand, instead of requiring a separate floating point status register for the status information. In one embodiment, a floating point operand data structure includes a first portion having floating point operand data and a second portion having embedded status information associated with at least one status condition of the operand data. The status condition may be determined from only the embedded status information. The status condition may also be associated with at least one floating point operation that generated the operand data structure. The outcome of a conditional floating point instruction may be based on the embedded status information without regard to contents of the floating point status register.

Abstract: A method and system perform a rounding step of a floating point computation on at least one floating point operand to preserve an inexact status. Inexact status information generated from the rounding step may be encoded within the result, instead of requiring a separate floating point status register for the inexact status information. In one embodiment, inexact status information is preserved by determining whether the at least one operand is inexact. Further, an intermediate result of the floating point computation is analyzed to determine whether it is inexact. Finally, the intermediate result is rounded based on whether the at least one operand is inexact and whether the intermediate result is inexact to preserve an inexact status of the at least one operand and the intermediate result.

Abstract: It is decided whether a first source data from the memory 101 is a data which is to be subjected to arithmetic or not by a state flag detection means 150, the result of the decision is retained as a state flag, and it is decided by a condition decision means 109 whether or not the state flag satisfies a condition for performing the arithmetic. A control means 110 controls whether an ALU 100 should perform the arithmetic or not on the basis of the condition satisfaction/dissatisfaction information.

Abstract: An apparatus and method for performing fast arithmetic operations, including addition, in a pipelined circuit is described. The apparatus and method operating on a first binary number and a second binary number comprise: a first arithmetic logic unit (ALU) operating on a first lower portion of the first binary number and a second lower portion of the second binary number to produce a first result and a carry out signal; and a second ALU operating on a first upper portion of the first binary number and a second upper portion of the second binary number to produce a second result; wherein at least a portion of the pipelined circuit stalls in response to the carry out signal. Another embodiment includes memory comprising a plurality of words, each word comprising data bits and a flag bit indicating a predetermined number of the most significant data bits are all zero.

Abstract: An arithmetic unit, for example a multiply and accumulate (MAC) unit 42, for a processing engine includes a partial product reduction tree 480. The partial product reduction tree will generate carry results and provides a final output to a final adder 470 connected to the partial production reduction tree. Unbiased rounding logic 476 is provided. A carry propagation tree is responsive to the carry results for anticipating a zero on each of N least significant bits of the final adder. When zero is anticipated on each of N least significant bits of the final adder, the carry propagation tree is operable to generate an output signal 477 which is used by the unbiased rounding stage to force the (N+1)th least significant bit of the final adder to zero.

Abstract: A floating point operand testing circuit includes an analysis circuit and a result generator circuit coupled to the analysis circuit. The analysis circuit determines the status of a floating point operand based upon data within the operand. An operand buffer may supply the operand to the analysis circuit. The result generator circuit is responsive to at least one control signal and asserts a result signal if the floating point analysis circuit matches the floating point status to a predetermined format specified by the control signal. The result signal can condition the outcome of a floating point instruction. The result generator may also respond to multiple control signals asserted when testing a single operand for different formats, such as not-a-number (NaN), infinity, normalized, denormalized, invalid operation, overflow, underflow, division by zero, exact, and inexact.

Abstract: An arithmetic circuit includes an arithmetic circuit performing an arithmetic operation of a predetermined bit width in accordance with an arithmetic instruction, a holding circuit storing status information about the arithmetic operation by the arithmetic circuit and a logic circuit having a function of a logic operation. The logic circuit receives the status information stored in the holding circuit and the status information about a present result of arithmetic operation. The arithmetic circuit further includes a selector for selecting either the status information about the present result of arithmetic operation or a signal outputted form the logic circuit in accordance with a control signal based on the arithmetic instruction. The selector supplies the selected signal to the holding circuit.