Nvidia Corporation

Karan Mehra

Peter B Holmqvist

George R Lynch

David Brian Glasco

Patrick Rene Marchand

James P Roberts

Wishwesh Anil Gandhi

Timothy John Purcell

Praveen Krishnamurthy

Steven E Molnar

Lacky V Shah

Cass Watson Everitt

Gregory Muthler

James Robertson

Peter B Holmquist

Emmett Michael Kilgariff

Rouslan Lyubomirov Dimitrov

Mark Hairgrove

David Kirk McAllister

Liang Yin

Nick Barrow-Williams

Hemayet Hossain

Jeff Pool

Chris Amsinck

Scott Phillip Cutler

Sujoyita Kaushikkar

Sami Olavi Johannes Kiminki

Daniel Glenn Robinson

Anup Holey

Alexander Michael Waterman

Jeff Smith

A processing device including a first cache is coupled to a system memory and a parallel processing unit (PPU) including a second cache. An operation to modify cache lines of the second cache associated with a first aperture of the system memory is received. A first subset of cache lines of the second cache is identified. The first subset of cache lines is associated with the first aperture of the system memory and is different from a second subset of cache lines of a second aperture of the system memory. The first subset of cache lines is modified as specified by the cache operation.

Systems and methods for aperture-specific cache operations

One embodiment of the present invention sets forth a technique for increasing available storage space within compressed blocks of memory attached to data processing chips, without requiring a proportional increase in on-chip compression status bits. A compression status bit cache provides on-chip availability of compression status bits used to determine how many bits are needed to access a potentially compressed block of memory. A backing store residing in a reserved region of attached memory provides storage for a complete set of compression status bits used to represent compression status of an arbitrarily large number of blocks residing in attached memory. Physical address remapping ('swizzling') used to distribute memory access patterns over a plurality of physical memory devices is partially replicated by the compression status bit cache to efficiently integrate allocation and access of the backing store data with other user data.

Compression status bit cache and backing store

In one embodiment of the present invention a cache unit organizes data stored in an attached memory to optimize accesses to compressed data. In operation, the cache unit introduces a layer of indirection between a physical address associated with a memory access request and groups of blocks in the attached memory. The layer of indirection&#x2014;virtual tiles&#x2014;enables the cache unit to selectively store compressed data that would conventionally be stored in separate physical tiles included in a group of blocks in a single physical tile. Because the cache unit stores compressed data associated with multiple physical tiles in a single physical tile and, more specifically, in adjacent locations within the single physical tile, the cache unit coalesces the compressed data into contiguous blocks. Subsequently, upon performing a read operation, the cache unit may retrieve the compressed data conventionally associated with separate physical tiles in a single read operation.

Organizing memory to optimize memory accesses of compressed data

A system architecture conserves memory bandwidth by including compression utility to process data transfers from the cache into external memory. The cache decompresses transfers from external memory and transfers full format data to naive clients that lack decompression capability and directly transfers compressed data to savvy clients that include decompression capability. An improved compression algorithm includes software that computes the difference between the current data word and each of a number of prior data words. Software selects the prior data word with the smallest difference as the nearest match and encodes the bit width of the difference to this data word. Software then encodes the difference between the current stride and the closest previous stride. Software combines the stride, bit width, and difference to yield final encoded data word. Software may encode the stride of one data word as a value relative to the stride of a previous data word.

Architecture and algorithms for data compression

One embodiment of the present invention sets forth a technique for processing commands received by an intermediary cache from one or more clients. The technique involves receiving a first write command from an arbiter unit, where the first write command specifies a first memory address, determining that a first cache line related to a set of cache lines included in the intermediary cache is associated with the first memory address, causing data associated with the first write command to be written into the first cache line, and marking the first cache line as dirty. The technique further involves determining whether a total number of cache lines marked as dirty in the set of cache lines is less than, equal to, or greater than a first threshold value, and: not transmitting a dirty data notification to the frame buffer logic when the total number is less than the threshold value, or transmitting a dirty data notification to the frame buffer logic when the total number is equal to or greater than the first threshold value.

Control mechanism for fine-tuned cache to backing-store synchronization

A method for managing memory traffic includes causing first data to be written to a data cache memory, where a first write request comprises a partial write and writes the first data to a first portion of the data cache memory, and further includes tracking the number of partial writes in the data cache memory. The method further includes issuing a fill request for one or more partial writes in the data cache memory if the number of partial writes in the data cache memory is greater than a predetermined first threshold.

Reducing memory traffic in DRAM ECC mode

One embodiment of the present invention sets forth a compression status cache configured to store compression information for blocks of memory stored within an external memory. A data cache unit is configured to request, in response to a cache miss, compressed data from the external memory based on compression information stored in the compression status bit cache. The compression status for active buffers is dynamically swapped into the compression status cache as needed. Different compression formats may be specified for one or more tiles within an active buffer. One advantage of the disclosed compression status cache is that a lame amount of attached memory may be allocated as compressible memory blocks, without incurring a corresponding die area cost because a portion of the compression status stored off chip in attached memory is cached in the compression status cache.

Compression status caching

A compression status bit cache provides on-chip availability of compression status bits used to determine how many bits are needed to access a potentially compressed block of memory. A backing store residing in a reserved region of attached memory provides storage for a complete set of compression status bits used to represent compression status of an arbitrarily large number of blocks residing in attached memory. Physical address remapping ('swizzling') used to distribute memory access patterns over a plurality of physical memory devices is partially replicated by the compression status bit cache to efficiently integrate allocation and access of the backing store data with other user data.

One embodiment of the present invention sets forth a technique for performing a memory access request to compressed data within a virtually mapped memory system comprising an arbitrary number of partitions. A virtual address is mapped to a linear physical address, specified by a page table entry (PTE). The PTE is configured to store compression attributes, which are used to locate compression status for a corresponding physical memory page within a compression status bit cache. The compression status bit cache operates in conjunction with a compression status bit backing store. If compression status is available from the compression status bit cache, then the memory access request proceeds using the compression status. If the compression status bit cache misses, then the miss triggers a fill operation from the backing store. After the fill completes, memory access proceeds using the newly filled compression status information.

Efficient line and page organization for compression status bit caching

One embodiment of the present invention sets forth a compression status bit cache with deterministic latency for isochronous memory clients of compressed memory. The compression status bit cache improves overall memory system performance by providing on-chip availability of compression status bits that are used to size and interpret a memory access request to compressed memory. To avoid non-deterministic latency when an isochronous memory client accesses the compression status bit cache, two design features are employed. The first design feature involves bypassing any intermediate cache when the compression status bit cache reads a new cache line in response to a cache read miss, thereby eliminating additional, potentially non-deterministic latencies outside the scope of the compression status bit cache. The second design feature involves maintaining a minimum pool of clean cache lines by opportunistically writing back dirty cache lines and, optionally, temporarily blocking non-critical requests that would dirty already clean cache lines. With clean cache lines available to be overwritten quickly, the compression status bit cache avoids incurring additional miss write back latencies.

Compression status bit cache with deterministic isochronous latency

A system and method for converting data from one format to another in a processing pipeline architecture. Data is stored in a shared cache that is coupled between one or more clients and an external memory. The shared cache provides storage that is used by multiple clients rather than being dedicated to separately convert the data format for each client. Each client may interface with the memory using a different format, such as a compressed data format. Data is converted to the format expected by the particular client as it is read from the cache and output to the client during a read operation. Bytes of a cache line may be remapped to bytes of an unpack register for output to a na&#xef;ve client, which may be configured to perform texture mapping operations. Data is converted from the client format to the memory format as it is stored into the cache during a write operation.

Method and system for converting data formats using a shared cache coupled between clients and an external memory

One embodiment of the present invention sets forth a method for implementing ECC protection in an on-chip L2 cache. When data is written to or read from an external memory, logic within the L2 cache is configured to generate ECC check bits and store the ECC check bits in the L2 cache in space typically allocated for storing byte enables. As a result, data stored in the L2 cache may be protected against bit errors without incurring the costs of providing additional storage or complex hardware for the ECC check bits.

L2 ECC implementation

One embodiment of the invention sets forth a mechanism for efficiently processing atomic operations transmitted from multiple general processing clusters to an L2 cache. A tag look-up unit tracks the availability of each cache line in the L2 cache, reserves the necessary cache lines for the atomic operations and transmits the atomic operations to an ALU for processing. The tag look-up unit also increments a reference counter associated with a reserved cache line each time an atomic operation associated with that cache line is received. This feature allows multiple atomic operations associated with the same cache line to be pipelined to the ALU. A ROP unit that includes the ALU may request additional data necessary to process an atomic operation from the L2 cache. Result data is stored in the L2 cache and may also be returned to the general processing clusters.

Growing community of inventors

Cary, NC, United States of America

Karan Mehra