One embodiment of the present invention sets forth a technique for efficiently performing voting operations within a multi-threaded parallel-processing system. A group of related parallel program threads executes within a processor core together in parallel. A new instruction, called a vote instruction, is introduced that enables a parallel program thread to post an individual vote within the context of the group of related threads and to receive the result of the vote. In this fashion, the vote instruction advantageously reduces overhead associated with inter-thread communication, thereby improving overall system performance.

Methods and apparatus are disclosed for using an index array and finite state machine for scatter/gather operations. Embodiment of apparatus may comprise: decode logic to decode a scatter/gather instruction and generate a set of micro-operations, and an index array to hold a set of indices and a corresponding set of mask elements. A finite state machine facilitates the gather operation. Address generation logic generates an address from an index of the set of indices for at least each of the corresponding mask elements having a first value. An address is accessed to load a corresponding data element if the mask element had the first value. The data element is written at an in-register position in a destination vector register according to a respective in-register position the index. Values of corresponding mask elements are changed from the first value to a second value responsive to completion of their respective loads.

Instructions and logic provide SIMD secure hashing round slice functionality. Some embodiments include a processor comprising: a decode stage to decode an instruction for a SIMD secure hashing algorithm round slice, the instruction specifying a source data operand set, a message-plus-constant operand set, a round-slice portion of the secure hashing algorithm round, and a rotator set portion of rotate settings. Processor execution units, are responsive to the decoded instruction, to perform a secure hashing round-slice set of round iterations upon the source data operand set, applying the message-plus-constant operand set and the rotator set, and store a result of the instruction in a SIMD destination register. One embodiment of the instruction specifies a hash round type as one of four MD5 round types. Other embodiments may specify a hash round type by an immediate operand as one of three SHA-1 round types or as a SHA-2 round type.

A method, system, and computer program product synchronize a group of workitems executing an instruction stream on a processor. The processor is yielded by a first workitem responsive to a synchronization instruction in the instruction stream. A first one of a plurality of program counters is updated to point to a next instruction following the synchronization instruction in the instruction stream to be executed by the first workitem. A second workitem is run on the processor after the yielding.

A method and apparatus are provided. The method includes executing a plurality of threads in a temporal dimension, executing a plurality of threads in a spatial dimension, determining a branch target address for each of the plurality of threads in the temporal dimension and the plurality of threads in the spatial dimension, and comparing each of the branch target addresses to determine a minimum branch target address, wherein the minimum branch target address is a minimum value among branch target addresses of each of the plurality of threads.

A processor having one or more processing cores is described. Each of the one or more processing cores has front end logic circuitry and a plurality of processing units. The front end logic circuitry is to fetch respective instructions of threads and decode the instructions into respective micro-code and input operand and resultant addresses of the instructions. Each of the plurality of processing units is to be assigned at least one of the threads, is coupled to said front end unit, and has a respective buffer to receive and store microcode of its assigned at least one of the threads. Each of the plurality of processing units also comprises: i) at least one set of functional units corresponding to a complete instruction set offered by the processor, the at least one set of functional units to execute its respective processing unit's received microcode; ii) registers coupled to the at least one set of functional units to store operands and resultants of the received microcode; iii) data fetch circuitry to fetch input operands for the at least one functional units' execution of the received microcode.

A graphics processing unit is disclosed, the graphics processing unit having a processor having one or more SIMD processing units, and a local data share corresponding to one of the one or more SIMD processing units, the local data share comprising one or more low latency accessible memory regions for each group of threads assigned to one or more execution wavefronts, and a global data share comprising one or more low latency memory regions for each group of threads.

A SIMD processor may be configured to determine one or more active threads from a plurality of threads, select one active thread from the one or more active threads, and perform a divergent operation on the selected active thread. The divergent operation may be a serial operation.

A processor includes a first logic to execute an instruction stream out-of-order, the instruction stream divided into a plurality of strands, the instruction stream and each strand ordered by program order (PO). The processor also includes a second logic to determine an oldest undispatched instruction in the instruction stream and store an associated PO value of the oldest undispatched instruction as an executed instruction pointer. The instruction stream includes dispatched and undispatched instructions. The processor also includes a third logic to determine a most recently retired instruction in the instruction stream and store an associated PO value of the most recently retired instruction as a retirement pointer, a fourth logic to select a range of instructions between the retirement pointer and the executed instruction pointer, and a fifth logic to identify the range of instructions as eligible for retirement.

Aspects include computation systems that can identify computation instances that are not capable of being reentrant, or are not reentrant capable on a target architecture, or are non-reentrant as a result of having a memory conflict in a particular execution situation. For example, a system can have a plurality of computation units, each with an independently schedulable SIMD vector. Computation instances can be defined by a program module, and a data element(s) that may be stored in a local cache for a particular computation unit of the plurality. Each local cache does not maintain coherency controls for such data elements. During scheduling, a scheduler can maintain a list of running (or runnable) instances, and attempt to schedule new computation instances by determining whether any new computation instance conflicts with a running instance and responsively defer scheduling. Such memory conflict checks can be conditioned on a flag or other indication of the potential for non-reentrancy.