Intel Corporation

Michael A Fetterman

Glenn J Hinton

David B Papworth

Andrew F Glew

Robert P Colwell

Robert W Martell

Ashwani Kumar Gupta

Sunil R Shenoy

Reynold V D'Sa

Bradley D Hoyt

Subramanian Natarajan

Darrell D Boggs

Haitham H Akkary

Shantanu R Gupta

Rohit A Vidwans

Atiq A Bajwa

James L Schwartz

Bret L Toll

Stephan Jourdan

Per H Hammarlund

Paul D Madland

Kris G Konigsfeld

Jeffrey M Abramson

John Alan Miller

Joel C Huang

Narayan Hegde

Glenn I Hinton

Steven J Griffith

Scott Dion Rodgers

Avinash Sodani

Ronak Singhal

Lawrence O Smith

Michael Cornaby

Sebastien Hily

Frederick J Pollack

Kamla P Huck

James S Griffith

Michael M Hancock

Grace C Chen

Stephen M Coward

Atig A Bajwa

Prem Pahlajrai

Beth Schultz

A method and apparatus for combining micro-operations to process immediate data. The immediate data may be wider than the immediate data storage capacity of a micro-operation. A first micro-operation is issued to process a first portion of the immediate data, which can be processed within the immediate data storage capacity of a micro-operation. A second micro-operation is issued to process a second portion of the immediate data, which can be processed within the immediate data storage capacity of a micro-operation. Execution of the first and second micro-operations and optionally of a third micro-operation serves to reconstruct the immediate data comprising the first portion and the second portion of the immediate data.

Method and apparatus for combining micro-operations to process immediate data

A method is disclosed. The method includes scheduling a load operation at least twice the size of a maximum access supported by a memory device, dividing the load operation into a plurality of separate load operation segments having a size equivalent to the maximum access supported by the memory device, and performing each of the plurality of load operation segments. A further method is disclosed where a temporary register is used to minimize the number of memory accesses to support unaligned accesses.

Load mechanism

In one embodiment, a method for flow optimization and prediction for vector streaming single instruction, multiple data (SIMD) extension (VSSE) memory operations is disclosed. The method comprises generating an optimized micro-operation (&#x3bc;op) flow for an instruction to operate on a vector if the instruction is predicted to be unmasked and unit-stride, the instruction to access elements in memory, and accessing via the optimized &#x3bc;op flow two or more of the elements at the same time without determining masks of the two or more elements. Other embodiments are also described.

Flow optimization and prediction for VSSE memory operations

A method and apparatus for representing an address in canonical form. The address is received and an error indicator is computed according to whether the address is received in a correct canonical form. The error indicator is stored together with a portion of the address, the portion being less than the entire address. The error indicator, together with the portion of the address stored, represent the address received.

Method and apparatus for representation of an address in canonical form

Maximum throughput or &ldquo;back-to-back&rdquo; scheduling of dependent instructions in a pipelined processor is achieved by maximizing the efficiency in which the processor determines the availability of the source operands of a dependent instruction and provides those operands to an execution unit executing the dependent instruction. These two operations are implemented through a number of mechanisms. One mechanism for determining the availability of source operands, and hence the readiness of a dependent instruction for dispatch to an available execution unit, relies on the early setting of a source valid bit during allocation when a source operand is a retired or immediate value. This allows the ready logic of a reservation station to begin scheduling the instruction for dispatch.

Method and apparatus for pipeline streamlining where resources are immediate or certainly retired

Maximum throughput or 'back-to-back' scheduling of dependent instructions in a pipelined processor is achieved by maximizing the efficiency in which the processor determines the availability of the source operands of a dependent instruction and provides those operands to an execution unit executing the dependent instruction. These two operations are implemented through number of mechanisms. One mechanism for determining the availability of source operands, and hence the readiness of a dependent instruction for dispatch to an available execution unit, relies on the prospective determination of the availability of a source operand before the operand itself is actually computed as a result of the execution of another instruction. Storage addresses of the source operands of an instruction are stored in a content addressable memory (CAM). Before an instruction is executed and its result data written back, the storage location address of the result is provided to the CAM and associatively compared with the source operand addresses stored therein. A CAM match and its accompanying match bit indicate that the result of the instruction to be executed will provide a source operand to the dependent instruction waiting in the reservation station. Using a bypass mechanism, if the operand is computed after dispatch of the dependent instruction, then the source operand is provided directly from the execution unit computing the source operand to a source operand input of the execution unit executing the dependent instruction.

Method and apparatus for maximum throughput scheduling of dependent

A processor is disclosed comprising a front end circuit that fetches a series of instructions according to a program sequence determined by at least one branch prediction, a register renaming circuit that allocates execution resources to each instruction, and an execution circuit that executes each instruction in the instruction stream. The execution circuit causes the front end circuit to refetch the series of instructions if a branch misprediction is detected. A stall signal disables the register renaming circuit until the execution circuit commits the branch result to an architectural state according to the program sequence.

Processor that redirects an instruction fetch pipeline immediately upon

A mechanism and method for renaming flags within a register alias table ('RAT') to increase processor parallelism and also providing and using flag masks associated with individual instructions. In order to reduce the amount of data dependencies between instructions that are concurrently processed, the flags used by these instructions are renamed. In general, a RAT unit provides register renaming to provide a larger physical register set than would ordinarily be available within a given macroarchitecture's logical register set (such as the Intel architecture or PowerPC or Alpha designs, for instance) to eliminate false data dependencies between instructions that reduce overall superscalar processing performance for the microprocessor. The renamed flag registers contain several flag bits and various flag bits may be updated or read by different instructions. Also, static and dynamic flag masks are associated with particular instructions and indicate which flags are capable of being updated by a particular instruction and also indicate which flags are actually updated by the instruction. Static flag masks are used in flag renaming and dynamic flag masks are used at retirement. The invention also discovers cases in which a flag register is required that is a superset of the previously renamed flag register portion.

Flag renaming and flag masks within register alias table

A method and circuitry for coordinating exceptions in a processor. The processor generates a result data value and an exception data value for each instruction wherein the exception data value specifies whether the corresponding instruction causes an exception. The processor commits the result data values to an architectural state of the processor in the sequential program order, and fetches an exception handler to processes the exception if the exception is indicated by one of the exception data values. The processor fetches an asynchronous event handler to processes an asynchronous event if the asynchronous event is detected while the result data values are committed to the architectural state of the processor.

Exception handling in a processor that performs speculative out-of-order

A mechanism for efficiently overlapping multiple operand types is used in a microprocessor which includes a plurality of execution units and a mechanism to provide operations, which include one or more operands, to the plurality of execution units. Each of the plurality of execution units interprets the one or more operands as different types of operands, and the mechanism to provide operations overlaps the different types of operands.

Mechanism for efficiently overlapping multiple operand types in a

A processor for executing a plurality of instructions. The processor comprises a plurality of logical segment registers, wherein the logical segment registers define an architectural state for memory segmentation of the processor. A plurality of physical segment registers are coupled to the logical segment registers. The processor further comprises an issue cluster that issues the instructions and that maps the logical segment registers, specified by the operations, to the physical segment registers to provide segment register renaming in the processor.

Segment register renaming in an out of order processor

A Branch Target Buffer Circuit in a computer processor that predicts branch instructions with a stream of computer instructions is disclosed. The Branch Target Buffer Circuit uses a Branch Target Buffer Cache that stores branch information about previously executed branch instructions. The branch information stored in the Branch Target Buffer Cache is addressed by the last byte of each branch instruction. When an Instruction Fetch Unit in the computer processor fetches a block of instructions it sends the Branch Target Buffer Circuit an instruction pointer. Based on the instruction pointer, the Branch Target Buffer Circuit looks in the Branch Target Buffer Cache to see if any of the instructions in the block being fetched is a branch instruction. When the Branch Target Buffer Circuit finds an upcoming branch instruction in the Branch Target Buffer Cache, the Branch Target Buffer Circuit informs an Instruction Fetch Unit about the upcoming branch instruction.

Method and apparatus for implementing a set-associative branch target

This invention overcomes the address size backward compatibility problem by first subtracting the segment base address from the linear destination address of a branch instruction to generate a virtual destination address. It is assumed that the branch instruction destination address is n bits long with m most significant bits. It is desired to provide backward compatibility in the n-bit processor for branch instruction code written for processors utilizing instruction address fields of size (n-m) bits. After obtaining the virtual address, if any of the m most significant bits are non-zero, then those m bits are set to zero to thereby generate a corrected virtual address. If such a compatibility correction is necessary, then a clear signal is asserted to flush all state of the processor that resulted from instructions being fetched after the branch instruction was fetched. The corrected virtual address is added back to the segment base address to generate a corrected linear address. The next instruction is fetched at the corrected linear address.

Method and apparatus for providing address-size backward compatibility

Method and apparatus for implementing a branch target buffer in CISC

A method and apparatus for handling events, such as those which occur in a processor. An event vector is formed by combining event type information indicating a type of event in the processor and mode information indicating an operating mode of the processor. A microcode event handler vector is generated therefrom, for example, by referencing a lookup table. The microcode event handler vector is then used for invoking a microcode event handler to handle occurrence of these events in the processor. By the formation of an event vector, and the microcode event handler vector, execution performance is increased due to avoiding conditional branching within the processor, such as modem high performance architectures, including those which execute instructions in and out-of-order.

Method and apparatus for generating event handler vectors based on both

Growing community of inventors

Hillsboro, OR, United States of America

Michael A Fetterman