Advanced Micro Devices Corporation

Intel Corporation

Jaewoong Sim

Gabriel H Loh

Alaa R Alameldeen

Christopher B Wilkerson

James Michael O'Connor

Zeshan A Chishti

Eriko Nurvitadhi

Deborah T Marr

Srikanth T Srinivasan

Vilas K Sridharan

Jonathan D Pearce

Mithuna Shamabhat Thottethodi

David B Sheffield

Andrey Ayupov

An apparatus and method for compressing floating-point values. For example, one embodiment of a processor comprises: instruction fetch circuitry to fetch instructions from a memory, the instructions including floating-point instructions; execution circuitry to execute the floating-point instructions, each floating-point instruction having one or more floating-point operands, each floating-point operand comprising an exponent value and a significand value; floating-point compression circuitry to compress a plurality of the exponent values associated with a corresponding plurality of the floating-point operands, the floating-point compression circuitry comprising: base generation circuitry to evaluate the plurality of the exponent values to generate a first base value; and delta generation circuitry to determine a difference between the plurality of exponent values and the first base value and to generate a corresponding first plurality of delta values, wherein the floating-point compression circuitry is to store the first base value and the corresponding first plurality of delta values as a plurality of compressed exponent values.

Method and system for efficient floating-point compression

An apparatus and method for efficiently processing invariant operations on a parallel execution engine. For example, one embodiment of a processor comprises: a plurality of parallel execution lanes comprising execution circuitry and registers to concurrently execute a plurality of threads; front end circuitry coupled to the plurality of parallel execution lanes, the front end circuitry to arrange the threads into parallel execution groups and schedule operations of the threads to be executed across the parallel execution lanes, wherein the front end circuitry is to dynamically evaluate one or more variables associated with the operations to determine if one or more conditionally invariant operations will be invariant across threads of a parallel execution group and/or across the parallel execution lanes; a scheduler of the front end circuitry to responsively schedule a shared thread upon a determination that a conditionally invariant operation will be invariant across threads of a parallel execution group and/or across the parallel execution lanes.

Apparatus and method for gang invariant operation optimizations using dynamic evaluation

An apparatus and method for implementing a heterogeneous memory subsystem is described. For example, one embodiment of a processor comprises: memory mapping logic to subdivide a system memory space into a plurality of memory chunks and to map the memory chunks across a first memory and a second memory, the first memory having a first set of memory access characteristics and the second memory having a second set of memory access characteristics different from the first set of memory access characteristics; and dynamic remapping logic to swap memory chunks between the first and second memories based, at least in part, on a detected frequency with which the memory chunks are accessed.

Method and apparatus for implementing a heterogeneous memory subsystem

The described embodiments include a main memory and a cache memory (or 'cache') with a cache controller that includes a mode-setting mechanism. In some embodiments, the mode-setting mechanism is configured to dynamically determine an access pattern for the main memory. Based on the determined access pattern, the mode-setting mechanism configures at least one region of the main memory in a write-back mode and configures other regions of the main memory in a write-through mode. In these embodiments, when performing a write operation in the cache memory, the cache controller determines whether a region in the main memory where the cache block is from is configured in the write-back mode or the write-through mode and then performs a corresponding write operation in the cache memory.

Dynamically configuring regions of a main memory in a write-back mode or a write-through mode

The described embodiments include a cache controller with a prediction mechanism in a cache. In the described embodiments, the prediction mechanism is configured to perform a lookup in each table in a hierarchy of lookup tables in parallel to determine if a memory request is predicted to be a hit in the cache, each table in the hierarchy comprising predictions whether memory requests to corresponding regions of a main memory will hit the cache, the corresponding regions of the main memory being smaller for tables lower in the hierarchy.

Predicting outcomes for memory requests in a cache memory

A method of managing memory includes installing a first cacheline at a first location in a cache memory and receiving a write request. In response to the write request, the first cacheline is modified in accordance with the write request and marked as dirty. Also in response to the write request, a second cacheline is installed that duplicates the first cacheline, as modified in accordance with the write request, at a second location in the cache memory.

Dirty cacheline duplication

Some embodiments include a partitioning mechanism that partitions a cache memory into sub-partitions for sub-entities. In the described embodiments, the cache memory is initially partitioned into two or more partitions for one or more corresponding entities. During a partitioning operation, the partitioning mechanism is configured to partition one or more of the partitions in the cache memory into two or more sub-partitions for one or more sub-entities of a corresponding entity. A cache controller then uses a corresponding sub-partition for memory accesses by the one or more sub-entities.

Partitioning caches for sub-entities in computing devices

A system, method and computer program product to store tag blocks in a tag buffer in order to provide early row-buffer miss detection, early page closing, and reductions in tag block transfers. A system comprises a tag buffer, a request buffer, and a memory controller. The request buffer stores a memory request having an associated tag. The memory controller compares the associated tag to a plurality of tags stored in the tag buffer and issues the memory request stored in the request buffer to either a memory cache or a main memory based on the comparison.

Memory scheduling for RAM caches based on tag caching

Some embodiments include a computing device with a control circuit that handles memory requests. The control circuit checks one or more conditions to determine when a memory request should be bypassed to a main memory instead of sending the memory request to a cache memory. When the memory request should be bypassed to a main memory, the control circuit sends the memory request to the main memory. Otherwise, the control circuit sends the memory request to the cache memory.

Bypassing memory requests to a main memory

The described embodiments include a computing device that handles memory requests. In some embodiments, when a memory request is to be sent to a cache in the computing device or to be bypassed to a next lower level of a memory hierarchy in the computing device based on expected memory request resolution times, a bypass mechanism is configured to send the memory request to the cache or bypass the memory request to the next lower level of the memory hierarchy.

Growing community of inventors

Atlanta, GA, United States of America

Jaewoong Sim