Abstract: A compiler is provided that determines when the use of software transactional memory (STM) primitives may be optimized with respect to a set of collectively dominating STM primitives. The compiler analysis coordinates the use of variables containing possible shadow copy pointers to allow the analysis to be performed for both direct write and buffered write STM systems. The coordination of the variables containing the possible shadow copy pointers ensures that the results of STM primitives are properly reused. The compiler analysis identifies memory accesses where STM primitives may be eliminated, combined, or substituted for lower overhead STM primitives.

Abstract: Embodiments are directed to optimizing data transfers between heterogeneous memory arenas. In one scenario, a computer system receives an indication that a data chunk is to be transferred from a first memory arena to a third memory arena, and then determines that for the data chunk to be transferred from the first memory arena to the third arena, the data chunk is to be transferred from the first memory arena to a second memory arena, and from the second memory arena to the third memory arena. The computer system divides the data chunk into smaller data portions and copies a first data portion from the first memory arena to the second memory arena. The computer system then copies the first data portion from the second memory arena to the third memory arena and copies a second data portion from the first memory arena to the second memory arena in parallel.

Abstract: A dynamic race detection system is provided that detects race conditions in code that executes concurrently in a computer system. The dynamic race detection system uses a modified software transactional memory (STM) system to detect race conditions. A compiler converts portions of the code that are not configured to operate with the STM system into pseudo STM code that operates with the STM system. The dynamic race detection system detects race conditions in response to either a pseudo STM transaction in the pseudo STM code failing to validate when executed or an actual STM transaction failing to validate when executed because of conflict with a concurrent pseudo STM transaction.

Abstract: A compile environment is provided in a computer system that allows programmers to program both CPUs and data parallel devices (e.g., GPUs) using a high level general purpose programming language that has data parallel (DP) extensions. A compilation process translates modular DP code written in the general purpose language into DP device source code in a high level DP device programming language using a set of binding descriptors for the DP device source code. A binder generates a single, self-contained DP device source code unit from the set of binding descriptors. A DP device compiler generates a DP device executable for execution on one or more data parallel devices from the DP device source code unit.

Abstract: A software transactional memory (STM) system allows the composition of traditional lock based synchronization with transactions in STM code. The STM system acquires each traditional lock the first time that a corresponding traditional lock acquire is encountered inside a transaction and defers all traditional lock releases until a top level transaction in a transaction nest commits or aborts. The STM system maintains state information associated with traditional lock operations in transactions and uses the state information to eliminate deferred traditional lock operations that are redundant. The STM system integrates with systems that implement garbage collection.

Abstract: A high level programming language provides a co-map communication operator that maps an input indexable type to an output indexable type according to a function. The function maps an index space corresponding to the output indexable type to an index space corresponding to the input indexable type. By doing so, the co-map communication operator lifts a function on an index space to a function on an indexable type to allow composability with other communication operators.

Abstract: A software transactional memory system is provided with multiple global version counters. The system assigns an affinity to one of the global version counters for each thread that executes transactions. Each thread maintains a local copy of the global version counters for use in validating read accesses of transactions. Each thread uses a corresponding affinitized global version counter to store version numbers of write accesses of executed transactions. The system adaptively changes the affinities of threads when data conflict or global version counter conflict is detected between threads.

Abstract: The present invention extends to methods, systems, and computer program products for debugging in a multiple address space environment. Embodiments of the invention include techniques for recording debug information used for translating between an abstract unified address space and multiple address spaces at a target system (e.g., a co-processor, such as, a GPU or other accelerator). A table is stored in the recorded debug information. The table includes one or more entries mapping compiler assigned IDs to address spaces. During debugging within a symbolic debugger, the recorded debug information can be used for viewing program data across multiple address spaces in a live debugging session.

Abstract: A software transactional memory system is provided with overflow handling. The system includes a global version counter with an epoch number and a version number. The system accesses the global version counter prior to and subsequent to memory accesses of transactions to validate read accesses of the transaction. The system includes mechanisms to detect global version number overflow and may allow some or all transactions to execute to completion subsequent to the global version number overflowing. The system also provides publication, privatization, and granular safety properties.

Abstract: Described herein are techniques for generating invocation stubs for a data parallel programming model so that a data parallel program written in a statically-compiled high-level programming language may be more declarative, reusable, and portable than traditional approaches. With some of the described techniques, invocation stubs are generated by a compiler and those stubs bridge a logical arrangement of data parallel computations to the actual physical arrangement of a target data parallel hardware for that data parallel computation.

Abstract: The present invention extends to methods, systems, and computer program products for changing addressing mode during code generation. Generally, embodiments of the invention use a compiler transformation to transform lower level code from one address alignment to another address alignment. The transformation can be based upon assumptions of a source programming language. Based on the assumptions, the transformation can eliminate arithmetic operations that compensate for different addressing alignment, resulting in more efficient code. Some particular embodiments use a compiler transformation to transform an Intermediate Representation (“IR”) from one-byte addressing alignment into multi-byte (e.g., four-byte) addressing alignment.

Abstract: Using cache resident transaction hardware to accelerate a software transactional memory system. The method includes identifying a plurality of atomic operations intended to be performed by a software transactional memory system as transactional operations as part of a software transaction. The method further includes selecting at least a portion of the plurality of atomic operations. The method further includes attempting to perform the portion of the plurality of atomic operations as hardware transactions using cache resident transaction hardware.

Abstract: Handling garbage collection and exceptions in hardware assisted transactions. Embodiments are practiced in a computing environment including a hardware assisted transaction system. A method includes beginning a hardware assisted transaction, raising an exception while in the hardware assisted transaction, including creating an exception object, determining that the transaction should be rolled back, and as a result of determining that the transaction should be rolled back, marshaling the exception object out of the hardware assisted transaction.

Abstract: Handling parallelism in transactions. A method includes beginning a cache resident transaction. The method further includes encountering a nested structured parallelism construct within the cache resident transaction. A determination is made as to whether the transaction would run faster serially in cache resident mode or faster parallel in software transactional memory mode for the overall transaction. In the software transactional memory mode, cache resident mode is used for one or more hierarchically lower nested transactions. The method further includes continuing the transaction in the mode determined.

Abstract: The present invention extends to methods, systems, and computer program products for optimizing execution of kernels. Embodiments of the invention include an optimization framework for optimizing runtime execution of kernels. During compilation, information about the execution properties of a kernel are identified and stored alongside the executable code for the kernel. At runtime, calling contexts access the information. The calling contexts interpret the information and optimize kernel execution based on the interpretation.

Abstract: The present invention extends to methods, systems, and computer program products for reconstructing program control flow. Embodiments include implementing or morphing a control flow graph (“CFG”) into an arbitrary loop structure to reconstruct (preserve) control flow from original source code. Loop structures can be optimized and can adhere to target platform constraints. In some embodiments, C++ source code (a first higher level format) is translated into a CFG (a lower level format). The CFG is then translated into High Level Shader Language (“HLSL”) source code (a second different higher level format) for subsequent compilation into SLSL bytecode (that can then be executed at a Graphical Processing Unit (“GPU”)). The control flow from the C++ source code is preserved in the HLSL source code.

Abstract: The present invention extends to methods, systems, and computer program products for binding executable code at runtime. Embodiments of the invention include late binding of specified aspects of code to improve execution performance. A runtime dynamically binds lower level code based on runtime information to optimize execution of a higher level algorithm. Aspects of a higher level algorithm having a requisite (e.g., higher) impact on execution performance can be targeted for late binding. Improved performance can be achieved with minimal runtime costs using late binding for aspects having the requisite execution performance impact.

Abstract: Various technologies and techniques are disclosed for transforming a sequential loop into a parallel loop for use with a transactional memory system. Open ended and/or closed ended sequential loops can be transformed to parallel loops. For example, a section of code containing an original sequential loop is analyzed to determine a fixed number of iterations for the original sequential loop. The original sequential loop is transformed into a parallel loop that can generate transactions in an amount up to the fixed number of iterations. As another example, an open ended sequential loop can be transformed into a parallel loop that generates a separate transaction containing a respective work item for each iteration of a speculation pipeline. The parallel loop is then executed using the transactional memory system, with at least some of the separate transactions being executed on different threads.

Abstract: Handling garbage collection and exceptions in hardware assisted transactions. Embodiments are practiced in a computing environment including a hardware assisted transaction system. Embodiments includes acts for writing to a card table outside of a transaction; handling garbage collection compaction occurring when a hardware transaction is active by using a common global variable and instructing one or more agents to write to the common global variable any time an operation is performed which may change an object's virtual address; acts for managing a thread-local allocation context; acts for handling exceptions while in a hardware assisted transaction.

Abstract: A high level programming language provides a map transformation that takes a data parallel algorithm and a set of one or more input indexable types as arguments. The map transformation applies the data parallel algorithm to the set of input indexable types to generate an output indexable type, and returns the output indexable type. The map transformation may be used to fuse one or more data parallel algorithms with another data parallel algorithm.