The invention comprises (i) a compilation method for automatically converting a single-threaded software program into an application-specific supercomputer, and (ii) the supercomputer system structure generated as a result of applying this method. The compilation method comprises: (a) Converting an arbitrary code fragment from the application into customized hardware whose execution is functionally equivalent to the software execution of the code fragment; and (b) Generating interfaces on the hardware and software parts of the application, which (i) Perform a software-to-hardware program state transfer at the entries of the code fragment; (ii) Perform a hardware-to-software program state transfer at the exits of the code fragment; and (iii) Maintain memory coherence between the software and hardware memories. If the resulting hardware design is large, it is divided into partitions such that each partition can fit into a single chip. Then, a single union chip is created which can realize any of the partitions.

Methods, apparatus, systems and articles of manufacture are disclosed to eliminate partial-redundant vector load operations. An example apparatus includes a node grouper to associate a vector operation with a node group, a candidate verifier to perform a dependencies test on a subset of the node group, and identify a subset of the node group as a candidate when the subset satisfies the dependencies test, and a code optimizer to determine replacement code based on a characteristic of the candidate in the node group and compare an estimated cost associated with executing the replacement code to a threshold. The example apparatus also includes a code generator to generate machine code using the replacement code when the estimated cost of executing the replacement code satisfies the threshold.

Techniques are disclosed for reordering operations of a neural network to improve runtime efficiency. In some examples, a compiler receives a description of the neural network comprising a plurality of operations. The compiler may determine which execution engine of a plurality of execution engines is to perform each of the plurality of operations. The compiler may determine an order of performance associated with the plurality of operations. The compiler may identify a runtime inefficiency based on the order of performance and a hardware usage for each of the plurality of operations. An operation may be reordered to reduce the runtime inefficiency. Instructions may be compiled based on the plurality of operations, which include the reordered operation.

An SIMD instruction generation and processing method and a related device are provided. The method may include: obtaining a length of each loop dimension of a first tensor formula; selecting, from a plurality of groups of information about a first SIMD instruction model based on the length of each loop dimension of a first tensor formula, information about a second SIMD instruction model matching the first tensor formula; generating, based on a length of at least one loop dimension of the first tensor formula and the second SIMD instruction model, a first SIMD instruction obtained after the first tensor formula is converted. The information about a second SIMD instruction model is selected from the plurality of groups of information about a first SIMD instruction model based on the length of each loop dimension of the tensor formula.

A compiler configured to configure memory nodes with a ready-to-read credit counter and a write credit counter. The ready-to-read credit counter of a particular upstream memory node initialized with as many read credits as a buffer depth of a corresponding downstream memory node. The ready-to-read credit counter configured to decrement when a buffer data unit is written by the particular upstream memory node into the corresponding downstream memory node, and to increment when the particular upstream memory node receives from the corresponding downstream memory node a read ready token. The write credit counter of the particular upstream memory node initialized with one or more write credits and configured to decrement when the particular upstream memory node begins writing the buffer data unit into the corresponding downstream memory node, and to increment when the particular upstream memory node receives from the corresponding downstream memory node a write done token.

Systems, media, and methods may identify loops of a unit of computation for performing operations associated with the loops. The system, media, and methods may receive textual program code that includes a unit of computation that comprises a loop (e.g., explicit/implicit loop). The unit of computation may be identified by an identifier (e.g., variable name within the textual program code, text string embedded in the unit of computation, and/or syntactical pattern that is unique within the unit of computation). A code portion and/or a section thereof may include an identifier referring to the unit of computation, where the code portion and the unit of computation may be at independent locations of each other. The systems, media, and methods may semantically identify a loop that corresponds to the identifier and perform operations on the textual program code using the code portion and/or section.

A compiler program causes a computer to execute optimization processing for an optimization target program. The optimization target program includes a loop including a vector store instruction and a vector load instruction for an array variable. The optimization processing includes (1) unrolling the vector store instruction and the vector load instruction in the loop by an unrolling number of times to generate a plurality of unrolled vector store instructions and a plurality of unrolled vector load instructions, and (2) scheduling to move an unrolled vector load instruction among the plurality of unrolled vector load instructions, which is located after a first unrolled vector store instruction that is located at first among the plurality of unrolled vector load instructions, before the first unrolled vector store instruction.

A system for compiling programs for execution thereof using a hierarchical processing system having two or more levels of memory hierarchy can perform memory-level-specific optimizations, without exceeding a specified maximum compilation time. To this end, the compiler system employs a polyhedral model and limits the dimensions of a polyhedral program representation that is processed by the compiler at each level using a focalization operator that temporarily reduces one or more dimensions of the polyhedral representation. Semantic correctness is provided via a defocalization operator that can restore all polyhedral dimensions that had been temporarily removed.

The present disclosure relates to a non-transitory computer-readable recording medium storing a complier that causes a computer to execute a process. The process includes generating a program. The program includes a first code that compares a first execution time from a start to an end of a loop processing when the loop processing is executed with a fixed-length SIMD instruction, with a second execution time from the start to the end of the loop processing when the loop processing is executed with a variable-length SIMD instruction, and a second code that executes the loop processing with the variable length SIMD instruction when a result of the comparison reveals that the first execution time is longer than the second execution time.

The present disclosure relates to a compiler for causing a computer to execute a process. The process includes generating a first program, wherein the first program includes a first code that determines whether a first area of a memory that a process inside a loop included in a second program refers to in a first execution time of the loop is in duplicate with a second area of the memory that the process refers to in a second execution time of the loop, a second code that executes the process in an order of the first and second execution times when it is determined that the first and the second areas are duplicate, and a third code that executes the process for the first execution time and the process for the second execution time in parallel when it is determined that the first and the second areas are not duplicate.