摘要:
A mechanism for performing single-path floating-point rounding in a floating point unit is disclosed. A system of the disclosure includes a memory and a processing device communicably coupled to the memory. In one embodiment, the processing device comprises a floating point unit (FPU) to generate a plurality of status flags for a rounded value of a finite nonzero number. The plurality of status flags are generated based on the finite nonzero number without calculating the rounded value of the finite nonzero number. The plurality of status flags comprises an overflow flag and an underflow flag. The FPU determines whether a rounded value should be calculated for the finite nonzero number based on the plurality of status flags and whether the overflow flag is asserted. Upon determining that the rounded value should be calculated for the finite nonzero number based on the plurality of status flags and that the overflow flag is asserted, the FPU calculates the rounded value of the finite nonzero number based on an overflow rounding. Upon determining that the rounded value should be calculated for the finite nonzero number based on the plurality of status flags and that the overflow flag is not asserted, the FPU calculates the rounded value of the finite nonzero number based on a blended reduced precision rounding.
摘要:
In an embodiment, a fused multiply-add (FMA) circuit is configured to receive a plurality of input data values to perform an FMA instruction on the input data values. The circuit includes a multiplier unit and an adder unit coupled to an output of the multiplier unit, and a control logic to receive the input data values and to reduce switching activity and thus reduce power consumption of one or more components of the circuit based on a value of one or more of the input data values. Other embodiments are described and claimed.
摘要:
A semiconductor chip is described having a functional unit that can execute a first instruction and execute a second instruction. The first instruction is an instruction that multiplies two operands. The second instruction is an instruction that approximates a function according to C0+C1X2+C2X22. The functional unit has a multiplier circuit. The multiplier circuit has: i) a first input to receive bits of a first operand of the first instruction and receive bits of a C1 term of the second instruction; ii) a second input to receive bits of a second operand of the first instruction and receive bits of a X2 term of the second instruction.
摘要:
A processing core is described having execution unit logic circuitry having a first register to store a first vector input operand, a second register to a store a second vector input operand and a third register to store a packed data structure containing scalar input operands a, b, c. The execution unit logic circuitry further include a multiplier to perform the operation (a*(first vector input operand))+(b*(second vector operand))+c.
摘要:
A math circuit for computing an estimate of a transcendental function is described. A lookup table storage circuit has stored therein several groups of binary values, where each group of values represents a respective coefficient of a first polynomial that estimates the function to a high precision. A computing circuit uses a portion of a binary value, that is also taken from one of the groups of values, to evaluate a second polynomial that estimates the function to a low precision. Other embodiments are also described and claimed.
摘要:
A method of processing an instruction is described that includes fetching and decoding the instruction. The instruction has separate destination address, first operand source address and second operand source address components. The first operand source address identifies a location of a first mask pattern in mask register space. The second operand source address identifies a location of a second mask pattern in the mask register space. The method further includes fetching the first mask pattern from the mask register space; fetching the second mask pattern from the mask register space; merging the first and second mask patterns into a merged mask pattern; and, storing the merged mask pattern at a storage location identified by the destination address.
摘要:
A hierarchical power control system for an integrated circuit may be integrated into a clocking system that includes a global clock generator, a clock distribution network in communication with the global clock generator and a plurality of functional unit blocks each in communication with the global clock generator. The hierarchical power control system may include a first power controller provided in a communication path between the global clock generator and the clock distribution network, and a plurality of second power controllers, one provided in each communication path between the clock distribution network and a functional unit block.
摘要:
A processor including a first execution core section clocked to perform execution operations at a first clock frequency, and a second execution core section clocked to perform execution operations at a second clock frequency which is different than the first clock frequency. The second execution core section runs faster and includes a data cache and critical ALU functions, while the first execution core section includes latency-tolerant functions such as instruction fetch and decode units and non-critical ALU functions. The processor may further include an I/O ring which may be still slower than the first execution core section. Optionally, the first execution core section may include a third execution core section whose clock rate is between that of the first and second execution core sections. Clock multipliers/dividers may be used between the various sections to derive their clocks from a single source, such as the I/O clock.
摘要:
A logic structure adapted to receive pulsed active input signals produces a logical output with a very small inherent switching delay. Pull-down transistors and complementary pull-up transistors are ratioed such that the default logical output level remains close to nominal even when the logic structure sinks or sources a DC current. When the pulsed input signals are inactive, no DC current path is enabled.
摘要:
A method for compressing speech. An audio signal comprising speech is broken down into its phonetic components. These phonetic components are then converted into data elements that represent each of the phonetic components. The determination of data elements is accomplished using a predefined table that correlates phonetic sounds to data elements. The data elements representing the phonetic sounds are then stored.