Abstract: A method and apparatus for utilizing hardware mechanisms of a transactional memory system is herein described. Various embodiments relate to software-based filtering of operations from read and write barriers and read isolation barriers during transactional execution. Other embodiments relate to software-implemented read barrier processing to accelerate strong atomicity. Other embodiments are also described and claimed.

Abstract: A method and apparatus for enabling a Software Transactional Memory (STM) with precompiled binaries is herein described. Upon encountering an access operation in a transaction, an annotation field associated with a memory location referenced by the access is checked. In response to the memory location representing a previous similar access within the transaction, the access is performed without access barriers. However, if the annotation field is in a default state representing no previous access during a pendancy of the transaction, then a mode of the processor is determined. If the processor mode is in implicit mode, an access handler/barrier is asynchronously executed. Conversely, in an explicit mode, a flag is set instead of asynchronously executing the handler. In addition, during compilation convert explicit and convert implicit instructions are inserted to intelligently convert modes for precompiled and newly compiled binaries.

Abstract: A computer system may comprise a computer platform and input-output devices. The computer platform may include a plurality of heterogeneous processors comprising a central processing unit (CPU) and a graphics processing unit) GPU, for example. The GPU may be coupled to a GPU compiler and a GPU linker/loader and the CPU may be coupled to a CPU compiler and a CPU linker/loader. The user may create a shared object in an object oriented language and the shared object may include virtual functions. The shared object may be fine grain partitioned between the heterogeneous processors. The GPU compiler may allocate the shared object to the CPU and may create a first and a second enabling path to allow the GPU to invoke virtual functions of the shared object. Thus, the shared object that may include virtual functions may be shared seamlessly between the CPU and the GPU.

Abstract: A method and apparatus for providing efficient strong atomicity is herein described. Optimized strong operations may be inserted at non-transactional read accesses to provide efficient strong atomicity. A global transaction value is copied at a beginning of a non-transactional function to a local transaction value; essentially creating a local timestamp of the global transaction value. At a non-transactional memory access within the function, a counter value or version value is compared to the LTV to see if a transaction has started updating memory locations, or specifically the memory location accessed. If memory locations have not been updated by a transaction, execution is accelerated by avoiding a full set of slowpath strong atomic operations to ensure validity of data accessed. In contrast, the slowpath operations may be executed to resolve contention between a transactional and non-transaction access contending for the same memory location.

Abstract: Embodiments of the invention provide language support for CPU-GPU platforms. In one embodiment, code can be flexibly executed on both the CPU and GPU. CPU code can offload a kernel to the GPU. That kernel may in turn call preexisting libraries on the CPU, or make other calls into CPU functions. This allows an application to be built without requiring the entire call chain to be recompiled. Additionally, in one embodiment data may be shared seamlessly between CPU and GPU. This includes sharing objects that may have virtual functions. Embodiments thus ensure the right virtual function gets invoked on the CPU or the GPU if a virtual function is called by either the CPU or GPU.

Abstract: Methods and apparatus to provide unbounded transactional memory systems are described. In one embodiment, an operation corresponding to a software transactional memory (STM) access may be executed if a preceding hardware transactional memory (HTM) access operation fails.

Abstract: Embodiments of the invention provide a programming model for CPU-GPU platforms. In particular, embodiments of the invention provide a uniform programming model for both integrated and discrete devices. The model also works uniformly for multiple GPU cards and hybrid GPU systems (discrete and integrated). This allows software vendors to write a single application stack and target it to all the different platforms. Additionally, embodiments of the invention provide a shared memory model between the CPU and GPU. Instead of sharing the entire virtual address space, only a part of the virtual address space needs to be shared. This allows efficient implementation in both discrete and integrated settings.

Abstract: A method and apparatus for accelerating lookups in an address based table is herein described. When an address and value pair is added to an address based table, the value is privately stored in the address to allow for quick and efficient local access to the value. In response to the private store, a cache line holding the value is transitioned to a private state, to ensure the value is not made globally visible. Upon eviction of the privately held cache line, the information is not written-back to ensure locality of the value. In one embodiment, the address based table includes a transactional write buffer to hold addresses, which correspond to tentatively updated values during a transaction. Accesses to the tentative values during the transaction may be accelerated through use of annotation bits and private stores as discussed herein. Upon commit of the transaction, the values are copied to the location to make the updates globally visible.

Abstract: A technique for using memory attributes to relay information to a program or other agent. More particularly, embodiments of the invention relate to using memory attribute bits to check various memory properties in an efficient manner.

Abstract: A method and apparatus for designating and handling irrevocable transactions is herein described. In response to detecting an irrevocable event, such as an I/O operation, a user-defined irrevocable designation, and a dynamic failure profile, a transaction is designated as irrevocable. In response to designating a transaction as irrevocable, Single Owner Read Locks (SORLs) are acquired for previous and subsequent reads in the irrevocably designated transaction to ensure the transaction is able to complete without modification to locations read from, while permitting remote resources to load from those locations to continue execution.

Abstract: A method, apparatus, and system are provided for performing compare and exchange operations using a sleep-wakeup mechanism. According to one embodiment, an instruction at a processor is executed to help acquire a lock on behalf of the processor. If the lock is unavailable to be acquired by the processor, the instruction is put to sleep until an event has occurred.

Abstract: A method and apparatus for unified concurrency control in a Software Transactional Memory (STM) is herein described. A transaction record associated with a memory address referenced by a transactional memory access operation includes optimistic and pessimistic concurrency control fields. Access barriers and other transactional operations/functions are utilized to maintain both fields of the transaction record, appropriately. Consequently, concurrent execution of optimistic and pessimistic transactions is enabled.

Abstract: Embodiments of the invention provide a programming model for CPU-GPU platforms. In particular, embodiments of the invention provide a uniform programming model for both integrated and discrete devices. The model also works uniformly for multiple GPU cards and hybrid GPU systems (discrete and integrated). This allows software vendors to write a single application stack and target it to all the different platforms. Additionally, embodiments of the invention provide a shared memory model between the CPU and GPU. Instead of sharing the entire virtual address space, only a part of the virtual address space needs to be shared. This allows efficient implementation in both discrete and integrated settings.

Abstract: A method, apparatus, and system are provided for performing compare and exchange operations using a sleep-wakeup mechanism. According to one embodiment, an instruction at a processor is executed to help acquire a lock on behalf of the processor. If the lock is unavailable to be acquired by the processor, the instruction is put to sleep until an event has occurred.

Abstract: A method and apparatus for accelerating transactional execution. Barriers associated with shared memory lines referenced by memory accesses within a transaction are only invoked/executed the first time the shared memory lines are accessed within a transaction. Hardware support, such as a transaction field/transaction bits, are provided to determine if an access is the first access to a shared memory line during a pendancy of a transaction. Additionally, in an aggressive operational mode version numbers representing versions of elements stored in shared memory lines are not stored and validated upon commitment to save on validation costs. Moreover, even in a cautious mode, that stores version numbers to enable validation, validation costs may not be incurred, if eviction of accessed shared memory lines do not occur during execution of the transaction.

Abstract: A page table entry dirty bit system may be utilized to record dirty information for a software distributed shared memory system. In some embodiments, this may improve performance without substantially increasing overhead because the dirty bit recording system is already available in certain processors. By providing extra bits, coherence can be obtained with respect to all the other uses of the existing page table entry dirty bits.

Abstract: Embodiments of the invention provide language support for CPU-GPU platforms. In one embodiment, code can be flexibly executed on both the CPU and GPU. CPU code can offload a kernel to the GPU. That kernel may in turn call preexisting libraries on the CPU, or make other calls into CPU functions. This allows an application to be built without requiring the entire call chain to be recompiled. Additionally, in one embodiment data may be shared seamlessly between CPU and GPU. This includes sharing objects that may have virtual functions. Embodiments thus ensure the right virtual function gets invoked on the CPU or the GPU if a virtual function is called by either the CPU or GPU.

Abstract: Performing non-transactional escape actions within a hardware based transactional memory system. A method includes at a hardware thread on a processor beginning a hardware based transaction for the thread. Without committing or aborting the transaction, the method further includes suspending the hardware based transaction and performing one or more operations for the thread, non-transactionally and not affected by: transaction monitoring and buffering for the transaction, an abort for the transaction, or a commit for the transaction. After performing one or more operations for the thread, non-transactionally, the method further includes resuming the transaction and performing additional operations transactionally. After performing the additional operations, the method further includes either committing or aborting the transaction.

Abstract: In one embodiment, the present invention includes a method for accessing a shared memory associated with a reader-writer lock according to a first concurrency mode, dynamically changing from the first concurrency mode to a second concurrency mode, and accessing the shared memory according to the second concurrency mode. In this way, concurrency modes can be adaptively changed based on system conditions. Other embodiments are described and claimed.

Abstract: A computing platform may include heterogeneous processors (e.g., CPU and a GPU) to support sharing of virtual functions between such processors. In one embodiment, a CPU side vtable pointer used to access a shared object from the CPU 110 may be used to determine a GPU vtable if a GPU-side table exists. In other embodiment, a shared non-coherent region, which may not maintain data consistency, may be created within the shared virtual memory. The CPU and the GPU side data stored within the shared non-coherent region may have a same address as seen from the CPU and the GPU side. However, the contents of the CPU-side data may be different from that of GPU-side data as shared virtual memory may not maintain coherency during the run-time. In one embodiment, the vptr may be modified to point to the CPU vtable and GPU vtable stored in the shared virtual memory.