Abstract: According to one general aspect, an apparatus may include a floating-point multiply-accumulate unit configured to generate a floating point result by either adding or subtracting three floating point operands: an addend, a product carry, and a product sum. The floating-point multiply-accumulate unit may include a close path adder. The close path adder may include an unincremented mantissa addition circuit configured to compute an unincremented mantissa result based upon the three floating point operands. The close path adder may also include an incremented mantissa addition circuit configured to, at least partially in parallel with the mantissa addition circuit, produce an incremented mantissa result. The close path adder may further include a selection circuit configured to produce a close path result by selecting between the unincremented mantissa result and the incremented mantissa result.

Abstract: Embodiments of the present disclosure include a tininess prediction and handler engine for handling numeric underflow while streamlining the data path for handling normal range cases, thereby avoiding flushes, and reducing the complexity of a scheduler with respect to how dependent operations are handled. A preemptive tiny detection logic section can detect a potential tiny result for the function or operation that is being performed, and can produce a pessimistic tiny indicator. The tininess prediction and handler engine can further include a subnormal post-processing pipe, which can denormalize and round one or more subnormal operations while in a post-processing mode. A schedule modification logic section can reschedule in-flight operations. The schedule modification logic section can issue dependent operations optimistically assuming that a producing operation will not produce a tiny result, and so will not incur extra latency associated with fixing the tiny result in the post-processing pipe.

Abstract: Reconfiguring a register file using a rename table having a plurality of fields that indicate fracture information about a source register of an instruction for instructions which have narrow to wide dependencies.

Abstract: Embodiments of the inventive concept include a fast close path solution and circuit of a three path fused multiply-adder circuit. The fast close path circuit can include one or more compressors that can receive multiple operands and produce a result sum and a result carry. The close path circuit can include one or more leading zero anticipators (LZAs). The one or more LZAs can receive and process the result sum and the result carry. The close path circuit can include one or more adders. The one or more adders can receive and add the result sum and the result carry in parallel with the one or more LZAs processing the result sum and the result carry. Since the close path is the critical timing path, by performing the addition operations in parallel with the LZA and/or priority encode (PENC) operations, the logic depth and latency of the close path are reduced.

Abstract: Embodiments of the present disclosure include a tininess prediction and handler engine for handling numeric underflow while streamlining the data path for handling normal range cases, thereby avoiding flushes, and reducing the complexity of a scheduler with respect to how dependent operations are handled. A preemptive tiny detection logic section can detect a potential tiny result for the function or operation that is being performed, and can produce a pessimistic tiny indicator. The tininess prediction and handler engine can further include a subnormal post-processing pipe, which can denormalize and round one or more subnormal operations while in a post-processing mode. A schedule modification logic section can reschedule in-flight operations. The schedule modification logic section can issue dependent operations optimistically assuming that a producing operation will not produce a tiny result, and so will not incur extra latency associated with fixing the tiny result in the post-processing pipe.

Abstract: According to one general aspect, an apparatus may include a load/store unit, an execution unit, and a first and a second data path. The load/store unit may be configured to load/store data from/to a memory and transmit the data to/from an execution unit, wherein the data includes a plurality of elements. The execution unit may be configured to perform an operation upon the data. The load/store unit may be configured to transmit the data to/from the execution unit via either a first data path configured to communicate, without transposition, the data between the load/store unit and the execution unit, or a second data path configured to communicate, with transposition, the data between the load/store unit and the execution unit, wherein transposition includes dynamically distributing portions of the data amongst a plurality of elements according to an instruction.

Abstract: Translation of boot code read request commands from an on-board processor of a system on a chip (SoC) from a bus protocol (e.g., advanced high-performance bus (AHB) protocol) into a sequence of commands understandable by a serial interface of the SoC to read boot code from an off-board (e.g., flash or other non-volatile) memory device. The serial interface of the memory device may include a relatively low pin count (e.g., 5 pins) and boot code of the memory device may be modified after tape-out of the SoC free of necessitating a subsequent tape-out of the SoC.

Abstract: According to one general aspect, an apparatus may include a load/store unit, an execution unit, and a first and a second data path. The load/store unit may be configured to load/store data from/to a memory and transmit the data to/from an execution unit, wherein the data includes a plurality of elements. The execution unit may be configured to perform an operation upon the data. The load/store unit may be configured to transmit the data to/from the execution unit via either a first data path configured to communicate, without transposition, the data between the load/store unit and the execution unit, or a second data path configured to communicate, with transposition, the data between the load/store unit and the execution unit, wherein transposition includes dynamically distributing portions of the data amongst a plurality of elements according to an instruction.

Abstract: Reconfiguring a register file using a rename table having a plurality of fields that indicate fracture information about a source register of an instruction for instructions which have narrow to wide dependencies.

Abstract: A technique of operating a processor includes determining whether a floating point unit (FPU) of the processor is to operate in a full-bit mode or a reduced-bit mode. An instruction is fetched and the instruction is decoded into a single operation, when the full-bit mode is indicated, or multiple operations, when the reduced-bit mode is indicated.

Abstract: A technique of operating a processor includes determining whether a floating point unit (FPU) of the processor is to operate in a full-bit mode or a reduced-bit mode. An instruction is fetched and the instruction is decoded into one or more full-bit operations, when the full-bit mode is indicated, or one or more reduced-bit operations, when the reduced-bit mode is indicated.

Abstract: A technique of operating a processor includes determining whether a floating point unit (FPU) of the processor is to operate in a full-bit mode or a reduced-bit mode. An instruction is fetched and the instruction is decoded into a single operation, when the full-bit mode is indicated, or multiple operations, when the reduced-bit mode is indicated.

Abstract: A functional unit of a processor may be configured to operate on instructions as either a single, wide functional unit or as multiple, independent narrower units. For example, an execution unit may be scheduled to execute an instruction as a single double-wide execution unit or as two independently schedulable single-wide execution units. Functional unit portions may be independently schedulable for execution of instructions operating on a first data type (e.g. SISD instructions). For single-wide instructions, functional unit portions may be scheduled independently. An issue lock mechanism may lock functional unit portions together so that they form a single multi-wide functional unit. For certain multi-wide instructions (e.g. certain SIMD instructions), an instruction operating on a multi-wide or vector data type may be scheduled so that the full multi-wide operation is performed concurrently by functional unit portions locked together as a one wide functional unit.

Abstract: A functional unit of a processor may be configured to operate on instructions as either a single, wide functional unit or as multiple, independent narrower units. For example, an execution unit may be scheduled to execute an instruction as a single double-wide execution unit or as two independently schedulable single-wide execution units. Functional unit portions may be independently schedulable for execution of instructions operating on a first data type (e.g. SISD instructions). For single-wide instructions, functional unit portions may be scheduled independently. An issue lock mechanism may lock functional unit portions together so that they form a single multi-wide functional unit. For certain multi-wide instructions (e.g. certain SIMD instructions), an instruction operating on a multi-wide or vector data type may be scheduled so that the full multi-wide operation is performed concurrently by functional unit portions locked together as a one wide functional unit.

Abstract: A superscalar microprocessor appends a tag value to each floating point number. The tag value indicates whether the corresponding floating point number is a normal floating point number or a special floating point number. Additionally, the tag value indicates the type of special floating point number represented by the corresponding floating point number. The tag value is stored with the floating point number in a register file of the floating point unit. Tag values are also generated for floating point numbers read from memory. When a floating point core of a floating point unit receives operands from either the register file or memory, the floating point core examines the tag values to determine whether each operand is a normal floating point number or a special floating point number. If either operand is a special floating point number, the floating point core determines the type of special floating point number and applies any applicable special rules.

Abstract: A superscalar microprocessor includes a combination floating point and multimedia unit. The floating point and multimedia unit includes one set of registers. The multimedia core and floating point core share the one set of registers. Each register as a type field associated with the register. The type field identifies whether the associated register contains valid data and whether the data is of multimedia type or floating point type. If the register stores floating point type data, the type field further indicates which of a plurality of floating point types the register stores such as: zero, infinity and normal. The floating point core relies on the type field to identify special floating point numbers such as zero and infinity. To ensure predictable results when a floating point instruction is executed subsequent to a multimedia instruction, a retyping algorithm retypes registers typed as multimedia type when the first floating point instruction subsequent to a multimedia instruction is executed.