Sun Microsystems, Inc.

Oracle America, Inc.

Ampere Computing LLC

Christopher Patrick Nelson

John Gregory Favor

Richard Win Thaik

Matthew Ashcraft

Joseph Byron Rowlands

Seungyoon Peter Song

Ivan Pavle Radivojevic

A system and method are provided for simplifying load acquire and store release semantics that are used in reduced instruction set computing (RISC). Translating the semantics into micro-operations, or low-level instructions used to implement complex machine instructions, can avoid having to implement complicated new memory operations. Using one or more data memory barrier operations in conjunction with load and store operations can provide sufficient ordering as a data memory barrier ensures that prior instructions are performed and completed before subsequent instructions are executed.

Implementation of load acquire/store release instructions using load/store operation with DMB operation

A method and apparatus for optimizing a sequence of operations adapted for execution by a processor is disclosed to include associating with each register a symbolic expression selected from a set of possible symbolic expressions, locating an operation, if any, that is next within the sequence of operations and setting that operation to be a working operation, where the working operation has associated therewith a destination register and zero or more source registers, and processing the working operation when the working operation and any symbolic expressions of its source registers, if any, match at least one of a set of rules, where each rule specifies that the working operation must match a subset of the operation set, where each rule also specifies that the symbolic expressions, if any, of any source registers of the working operation must match a subset of the possible symbolic expressions, and where the rule also specifies a result, then posting the result as the symbolic expression of the destination register.

Symbolic renaming optimization of a trace

A method and apparatus for optimizing a sequence of operations adapted for execution by a processor is disclosed to include associating a symbolic expression with each of at least a subset of the registers, holding a set of dependency indications that specify for each particular symbolic expression which, if any, of the other symbolic expressions must be emitted as operations prior to emitting the particular symbolic expression, locating an operation, if any, that is next within the sequence of operations and setting that operation to be a working operation and processing the working operation. Processing is performed by handling the working operation by a combination of updating zero or more of the symbolic expressions and emitting zero or more of the symbolic expressions as operations, identifying which, if any, of the symbolic expressions that were updated in process were updated such that they must be emitted prior to which other, if any, of the symbolic expressions, and then updating the dependency indicators to include any such dependencies, and identifying which, if any, of the symbolic expressions that were updated in process were updated such that they no longer need to be emitted prior to which other, if any, of the symbolic expressions, and then updating the dependency indicators to remove any such dependencies.

Emit vector optimization of a trace

A method of determining a reason for a trace to be aborted includes receiving at least two incoming indications of occurrences of abort triggers stemming from the execution of at least two of the operations that are different from each other, where each of the abort triggers has an associated abort priority level, and where the trace represents multiple instructions. The method further includes prioritizing among the abort triggers for the trace based on the abort priority level of each abort trigger, where the prioritizing does not take into account a correspondence between operations and instructions and where the prioritizing selects as a pending abort reason one or more of the abort triggers that have the same abort priority level, and where that abort priority level is the highest among the abort priority levels of the abort triggers for the trace.

Abort prioritization in a trace-based processor

A method and apparatus for optimizing a sequence of operations adapted for execution by a processor is disclosed to include locating an operation, if any, that is next within the sequence of operations and setting a current operation to be that operation. The current operation is processed as follows: a) de-activating, if not already de-activated, a consumed indicator associated with the current operation; and b) when the current operation is of the producer type, then activating, if not already activated, a producer indicator associated with the current operation, and locating a first set of operations, if any, that i) are earlier in the sequence of operations than the current operation, ii) have their associated producer indicator activated, and iii) have their associated consumed indicator de-activated, and then de-activating the producer indicator associated with each operation in the first set. When the current operation is of the consumer type, then locating a second set of operations, if any, that are earlier in the sequence of operations than the current operation, and then activating, if not already activated, the consumed indicator associated with each operation in the second set.

Flag optimization of a trace

A method of handling a trace to be aborted includes receiving an indication of a trace to be aborted and an indication of an abort reason corresponding to an execution of the trace to be aborted. The trace to be aborted has a trace type associated therewith and includes a sequence of the operations, and represents a sequence of at least two of the instructions. The method further includes identifying a corrective action based at least in part on the type of the trace to be aborted and on the abort reason, not taking into account a correspondence between the at least one operation that caused the execution to be aborted and the at least one instruction that the at least one operation at least in part represents. A next trace and its trace type is determined for execution, where the determining is based on the trace to be aborted and on the corrective action.

Graceful degradation in a trace-based processor

Managing speculative execution via groups of one or more actions corresponding to atomic traces enables efficient processing of flag-related actions, as atomic traces advantageously enable single checkpoints of flag values at atomic trace boundaries. Checkpointing flags during atomic trace renaming in a processor system uses a flag checkpoint table to store a plurality of flag checkpoints, each corresponding to an atomic trace. The table is selectively accessed to provide flag information to restore speculative flags when an atomic trace is aborted. A corresponding flag checkpoint is stored when an atomic trace is renamed. An action that updates flags updates all entries in the table corresponding to younger atomic traces. If the atomic trace is aborted, then the corresponding flag checkpoint is used for restoration of flag state.

Flag management in processors enabled for speculative execution of micro-operation traces

Managing speculative execution via groups of one or more actions corresponding to atomic traces enables efficient processing of flag-related actions, as atomic traces advantageously enable single checkpoints of flag values at atomic trace boundaries. Checkpointing flags on-demand for atomic traces in a processor system uses a flag checkpoint table to store a plurality of flag checkpoints, each corresponding to an atomic trace. The table is selectively accessed to provide flag information to restore speculative flags when an atomic trace is aborted. A corresponding flag checkpoint is allocated to an invalid state when an atomic trace is renamed. An action that updates flags initializes the corresponding flag checkpoint (if invalid). If the atomic trace is aborted, then the table is searched according to program order starting with the entry corresponding to the aborted atomic trace. The first (if any) valid checkpoint found is used for flag restoration.

Managing speculative execution via groups of actions corresponding to atomic traces enables efficient processing of flag-related actions, as atomic traces advantageously enable single checkpoints of flags at trace boundaries. Flag restoration from checkpoints for trace aborts uses a flag checkpoint table to store flag checkpoints, each corresponding to an atomic trace. The table is accessed for flag restoration in response to a trace abort. In a first technique, a corresponding flag checkpoint is stored in response to trace renaming, and the flag checkpoints are updated as flags are modified. Flags are restored from the flag checkpoint corresponding to an aborted atomic trace. In a second technique, a corresponding flag checkpoint is allocated to an invalid state in response to trace renaming, and initialized on-demand when flags are first modified in accordance with the atomic trace. Flags are restored from the oldest flag checkpoint starting from an aborted atomic trace.

A method for synchronized renaming between a master processor and a coprocessor includes sending from the master processor an operation for execution by the coprocessor along with an identifier, at the coprocessor, renaming the operation for execution, including assigning a resource and associating the resource with the identifier, and at a subsequent time, sending the identifier from the master processor to the coprocessor to be used in conjunction with the execution of the renamed operation.

Coprocessor receiving renamed register identifier from master to complete an operation upon register data ready

An apparatus and a method dynamically reassign resources in a coprocessor among master processors that require service from the coprocessor. The method includes each processor, in each processor cycle, keeping track of a number of resource units required for executing operations sent to the coprocessor in that processor cycle and receiving from the coprocessor a number of resource units released during the processor cycle. When the resources need to be reassigned, the coprocessor asserts a signal to the resource yielding processor to cause it to reduce its expectation of resources to zero and ceasing sending service requests to the coprocessor. The coprocessor then moves resources from the yielding processor to the resource receiving processor. Resources are then released to both processors over time to their respective adjusted resource allocations. Such resources may be the number of operations that is allowed to be executing in the coprocessor simulataneously. It may also be the number of physical registers that the processor can use in its operations sent to the coprocessor.

Dynamic resource allocation among master processors that require service from a coprocessor

Synchronized register renaming between a master processor and a coprocessor that receives operations from the master enables efficient implementation of register renaming and operation execution in the processors. An ideal and an external register allocation map are implemented in the coprocessor. When registers are no longer allocated according to the ideal allocation map and the registers are currently allocated according to the external allocation map, the registers are deallocated in the external map and the number of freed registers is reported to the master. The master increments a free register credit count accordingly, and decrements the credit count by one for each operation issued to the coprocessor. An operation is not issued to the coprocessor unless at least a register is free according to the credit count. The master also throttles coprocessor operation issue based on a credit count corresponding to free scheduler entries available in the coprocessor.

Synchronizing master processor by stalling when tracking of coprocessor rename register resource usage count for sent instructions reaches credited apportioned number

An instruction stream having variable length instructions with embedded constants (e.g. immediate values and displacements) is translated into a stream of operations and a corresponding stream of bit fields, enabling advantageous compact storage of the embedded constants. The operations and the compact constants are optionally stored in entries in a trace cache and/or processed by execution pipelines. The compact constants are optionally formulated as a small constant field, a pointer, or both. The pointer of a particular one of the operations optionally references one of the bit fields within a window of the operations associated with the particular operation. A full-sized constant is constructed from one or more contiguous ones of the bit fields, starting with the referenced bit field, by unpacking and uncompressing information from the contiguous bit fields. An operation optionally includes a plurality of small constant fields and pointers to specify a respective plurality of constants.

Growing community of inventors

Santa Clara, CA, United States of America

Christopher Patrick Nelson