Abstract: A reconfigurable matrix multiplier (RMM) system/method allowing tight or loose coupling to supervisory control processor application control logic (ACL) in a system-on-a-chip (SOC) environment is disclosed. The RMM provides for C=A*B matrix multiplication operations having A-multiplier-matrix (AMM), B-multiplicand-matrix (BMM), and C-product-matrix (CPM), as well as C=A*B+D operations in which D-summation-matrix (DSM) represents the result of a previous multiplication operation or another previously defined matrix. The RMM provides for additional CPM LOAD/STORE paths allowing overlapping of compute/data transfer operations and provides for CPM data feedback to the AMM or BMM operand inputs from a previously calculated CPM result.

Abstract: An outer product multiplier (GPM) system/method that integrates compute gating and input/output circular column rotation functions to balance time spent in compute and data transfer operations while limiting overall dynamic power dissipation is disclosed. Matrix compute gating (MCG) based on a computation decision matrix (CDM) limits the number of computations required on a per cycle basis to reduce overall matrix compute cycle power dissipation. A circular column rotation vector (CRV) automates input/output data formatting to reduce the number of data transfer operations required to achieve a given matrix computation result. Matrix function operators (MFO) utilizing these features are disclosed and include: matrix-matrix multiplication; matrix-matrix and vector-vector point-wise multiplication, addition, and assignment; matrix-vector multiplication; vector-vector inner product; matrix transpose; matrix row permute; and vector-column permute.

Abstract: An outer product multiplier (GPM) system/method that integrates compute gating and input/output circular column rotation functions to balance time spent in compute and data transfer operations while limiting overall dynamic power dissipation is disclosed. Matrix compute gating (MCG) based on a computation decision matrix (CDM) limits the number of computations required on a per cycle basis to reduce overall matrix compute cycle power dissipation. A circular column rotation vector (CRV) automates input/output data formatting to reduce the number of data transfer operations required to achieve a given matrix computation result. Matrix function operators (MFO) utilizing these features are disclosed and include: matrix-matrix multiplication; matrix-matrix and vector-vector point-wise multiplication, addition, and assignment; matrix-vector multiplication; vector-vector inner product; matrix transpose; matrix row permute; and vector-column permute.

Abstract: In one embodiment, command message generation and execution using a machine code-instruction is performed. One embodiment includes a particular machine executing a single machine-code instruction including a reference into a command-message-building data structure stored in memory. This executing the single machine-code instruction includes generating a command message and initiating communication of the command message to a hardware accelerator, including copying command information from the command-message-building data structure based on the reference into the command message. The hardware accelerator receives and executes the command message. In one embodiment, the command message is message-switched from a processor to a hardware accelerator, such as, but not limited to, a memory controller, a table lookup unit, or a prefix lookup unit. In one embodiment, a plurality of threads share the command-message-building data structure.

Abstract: A reconfigurable matrix multiplier (RMM) system/method allowing tight or loose coupling to supervisory control processor application control logic (ACL) in a system-on-a-chip (SOC) environment is disclosed. The RMM provides for C=A*B matrix multiplication operations having A-multiplier-matrix (AMM), B-multiplicand-matrix (BMM), and C-product-matrix (CPM), as well as C=A*B+D operations in which D-summation-matrix (DSM) represents the result of a previous multiplication operation or another previously defined matrix. The RMM provides for additional CPM LOAD/STORE paths allowing overlapping of compute/data transfer operations and provides for CPM data feedback to the AMM or BMM operand inputs from a previously calculated CPM result.

Abstract: A page table is a data structure used by a virtual memory system in a computer system to store the mapping between virtual addresses and physical addresses. Embodiments herein use a tree to map a virtual memory address space in the page table. The tree may be an N-ary tree where N is a power of two (e.g., 2, 4, 8, 16, etc.). The tree may include multiple levels that each correspond to a different page table size. For example, an octree includes eight different entries for each child node which may include per-thread sub-entries. Child nodes in the first level of the octree may each correspond to a 512 GiB page, while child nodes in the second level, however, may each have eight entries that correspond to a 64 GiB page. In this manner, an N-ary tree may be used to support a computing system with varying page sizes.

Abstract: Cache lines of a data cache may be assigned to a specific page type or color. In addition, the computing system may monitor when a cache line assigned to the specific page color is allocated in the cache. As each cache line assigned to a particular page color is allocated, the computing system may compare a respective index associated with each of the cache lines to determine maximum and minimum indices for that page color. These indices define a block of the cache that stores the data assigned to the page color. Thus, when the data of a page color is evicted from the cache, instead of searching the entire cache to locate the cache lines, the computing system uses the maximum and minimum indices as upper and lower bounds to reduce the portion of the cache that is searched.

Abstract: Cache lines of a data cache may be assigned to a specific page type or color. In addition, the computing system may monitor when a cache line assigned to the specific page color is allocated in the cache. As each cache line assigned to a particular page color is allocated, the computing system may compare a respective index associated with each of the cache lines to determine maximum and minimum indices for that page color. These indices define a block of the cache that stores the data assigned to the page color. Thus, when the data of a page color is evicted from the cache, instead of searching the entire cache to locate the cache lines, the computing system uses the maximum and minimum indices as upper and lower bounds to reduce the portion of the cache that is searched.

Abstract: In one embodiment, command message generation and execution using a machine code-instruction is performed. One embodiment includes a particular machine executing a single machine-code instruction including a reference into a command-message-building data structure stored in memory. This executing the single machine-code instruction includes generating a command message and initiating communication of the command message to a hardware accelerator, including copying command information from the command-message-building data structure based on the reference into the command message. The hardware accelerator receives and executes the command message. In one embodiment, the command message is message-switched from a processor to a hardware accelerator, such as, but not limited to, a memory controller, a table lookup unit, or a prefix lookup unit. In one embodiment, a plurality of threads share the command-message-building data structure.

Abstract: A page table is a data structure used by a virtual memory system in a computer system to store the mapping between virtual addresses and physical addresses. Embodiments herein use a tree to map a virtual memory address space in the page table. The tree may be an N-ary tree where N is a power of two (e.g., 2, 4, 8, 16, etc.). The tree may include multiple levels that each correspond to a different page table size. For example, an octree includes eight different entries for each child node which may include per-thread sub-entries. Child nodes in the first level of the octree may each correspond to a 512 GiB page, while child nodes in the second level, however, may each have eight entries that correspond to a 64 GiB page. In this manner, an N-ary tree may be used to support a computing system with varying page sizes.

Abstract: Cache lines of a data cache may be assigned to a specific page type or color. In addition, the computing system may monitor when a cache line assigned to the specific page color is allocated in the cache. As each cache line assigned to a particular page color is allocated, the computing system may compare a respective index associated with each of the cache lines to determine maximum and minimum indices for that page color. These indices define a block of the cache that stores the data assigned to the page color. Thus, when the data of a page color is evicted from the cache, instead of searching the entire cache to locate the cache lines, the computing system uses the maximum and minimum indices as upper and lower bounds to reduce the portion of the cache that is searched.

Abstract: In one embodiment, rule-based virtual address translation is performed for accessing data (e.g., reading and/or writing data) typically stored in different manners and/or locations among one or more memories, such as, but not limited to, in packet switching devices. A virtual address is matched against a set of predetermined rules to identify one or more storing description parameters. These storing description parameters determine in which particular memory unit(s) and/or how the data is stored. Thus, different portions of a data structure (e.g., table) can be stored in different memories and/or using different storage techniques. The virtual address is converted to a lookup address based on the identified storing description parameter(s). One or more read or write operations in one or more particular memory units is performed based on the lookup address said converted from the virtual address.

Abstract: Cache lines of a data cache may be assigned to a specific page type or color. In addition, the computing system may monitor when a cache line assigned to the specific page color is allocated in the cache. As each cache line assigned to a particular page color is allocated, the computing system may compare a respective index associated with each of the cache lines to determine maximum and minimum indices for that page color. These indices define a block of the cache that stores the data assigned to the page color. Thus, when the data of a page color is evicted from the cache, instead of searching the entire cache to locate the cache lines, the computing system uses the maximum and minimum indices as upper and lower bounds to reduce the portion of the cache that is searched.

Abstract: In one embodiment, rule-based virtual address translation is performed for accessing data (e.g., reading and/or writing data) typically stored in different manners and/or locations among one or more memories, such as, but not limited to, in packet switching devices. A virtual address is matched against a set of predetermined rules to identify one or more storing description parameters. These storing description parameters determine in which particular memory unit(s) and/or how the data is stored. Thus, different portions of a data structure (e.g., table) can be stored in different memories and/or using different storage techniques. The virtual address is converted to a lookup address based on the identified storing description parameter(s). One or more read or write operations in one or more particular memory units is performed based on the lookup address said converted from the virtual address.