Texas Instruments Corporation

Sanjive Agarwala

David A Comisky

Charles Lance Fuoco

Timothy David Anderson

Iain Craig Robertson

Christopher L Mobley

Raguram Damodaran

Patrick W Bosshart

Joel J Graber

Kyle Castille

Quang-Dieu An

Bryan David Sheffield

Duy-Loan T Le

Manjeri Krishnan

Shoban Srikrishna Jagathesan

Abhijeet Ashok Chachad

Keith Balmer

Hiep V Tran

Donald Edward Steiss

David J Hoyle

Jonathan (Son) Hung Tran

John D Keay

Hung Ong

Amarjit Singh Bhandal

Parallel pipelines are used to access a shared memory. The shared memory is accessed via a first pipeline by a processor to access cached data from the shared memory. The shared memory is accessed via a second pipeline by a memory access unit to access the shared memory. A first set of tags is maintained for use by the first pipeline to control access to the cache memory, while a second set of tags is maintained for use by the second pipeline to access the shared memory. Arbitrating for access to the cache memory for a transaction request in the first pipeline and for a transaction request in the second pipeline is performed after each pipeline has checked its respective set of tags.

Cache with multiple access pipelines

Command reordering in the hub interface unit (HIU) of Enhanced Direct Memory Access (EDMA) functions is described. Without command reordering in the EDMA, commands are issued by the HIU to the peripheral in order of issue. If the higher priority transfers are issued later by the EDMA, the previously issued lower priority transfers would block the higher priority transfers. Command reordering in the HIU causes transfers to be reordered and issued to the peripheral based on their priority. Reordering allows the EDMA and HIU to give due service to high priority transfer requests with decreased weight placed on the order in which the requests were issued.

Command re-ordering in hub interface unit based on priority

An enhanced direct memory access (EDMA) operation issues a read command to the source port to request data. The port returns the data along with response information, which contains the channel and valid byte count. The EDMA stores the read data into a write buffer and acknowledges to the source port that the EDMA can accept more data. The read response and data can come from more than one port and belong to different channels. Removing channel prioritizing according to this invention allows the EDMA to store read data in the write buffer and the EDMA then can acknowledge the port read response concurrently across all channels. This improves the EDMA inbound and outbound data flow dramatically.

Concurrent read response acknowledge enhanced direct memory access unit

A data transfer apparatus with hub and ports includes design configurable hub interface units (HIU) between the ports and corresponding external application units. The configurable HIU provides a single generic superset HIU that can be configured for specific more specialized applications during implementation as part of design synthesis. Configuration allows the super-set configurable HIU to be crafted into any one of several possible special purpose HIUs. This configuration is performed during the design phase and is not applied in field applications. Optimization aimed at eliminating functional blocks not needed in a specific design and simplifying and modifying other functional blocks allows for the efficient configuring of these other types of HIUs. Configuration of HIUs for specific needs can result in significant savings in silicon area and in power consumption.

Hardware configurable hub interface unit

The present invention provides for independent source-read and destination-write functionality for Enhanced Direct Memory Access (EDMA). Allowing source read and destination write pipelines to operate independently makes it possible for the source pipeline to issue multiple read requests and stay ahead of the destination write for fully pipelined operation. The result is that fully pipelined capability may be achieved and utilization of the full DMA bandwidth and maximum throughput performance are provided.

Independent source read and destination write enhanced DMA

The pBIST solution to memory testing is a balanced hardware-software oriented solution. pBIST hardware provides access to all memories and other such logic (e.g. register files) in pipelined logic allowing back-to-back accesses. The approach then gives the user access to this logic through CPU-like logic in which the programmer can code any algorithm to target any memory testing technique required. Because hardware inside the chip is used at-speed, the full device speed capabilities are available. CPU-like hardware can be programmed and algorithms can be developed and executed after tape-out and while testing on devices in chip form is in process.

Programmable built in self test of memory

Electrical fuses (eFuses) are applied to the task of memory performance adjustment to improve upon earlier fuse techniques by not requiring an additional processing step and expensive equipment. Standard electrical fuse (eFuse) hardware chains provide a soft test feature wherein the effect of memory slow-down can be tested prior to actually programming the fuses. Electrical fuses thus provide a very efficient non-volatile method to match the logic-memory interface through memory trimming, drastically cutting costs and cycle times involved.

Electrical fuse control of memory slowdown

A transfer request bus and transfer request bus node is described which is suitable for use in a data transfer controller processing multiple concurrent transfer requests despite the attendant collisions which result when conflicting transfer requests occur. Transfer requests are passed from an upstream transfer request node to downstream transfer request node and then to a transfer request controller with queue. At each node a local transfer request can also be inserted to be passed on to the transfer controller queue. Collisions at each transfer request node are resolved using a token passing scheme wherein a transfer request node possessing the token allows a local request to be inserted in preference to the upstream request.

Transfer request bus node for transfer controller with hub and ports

In a transfer controller with hub and ports architecture one of the data ports is an active data port. This active data port can supply its own source information, destination information and data quantity in a data transfer request. This data transfer request is serviced in a manner similar to other data transfer requests. The active data port may specify itself as the data destination in an active read. Alternatively, the active data port may specify itself as the data source in an active data write.

Active ports in a transfer controller with hub and ports

The transfer controller with hub and ports uses a write allocation counter and algorithm to control data reads from a source port. The write allocation count is the amount of data that can be consumed immediately by the write reservation station of a slow destination port and the channel data router buffers. This is used to throttle fast source port read operations to whole read bursts until space to adsorb the read data is available. This ensures that the source port response queue is not blocked with data that cannot be consumed by the channel data router and the slow destination port. This condition would otherwise block a fast source port from providing data to the other destination ports.

Write allocation counter for transfer controller with hub and ports

A transfer controller with hub and ports is viewed as a communication hub between the various locations of a global memory map. A request queue manager serves as a crucial part of the transfer controller. The request queue manager receives these data transfer request packets from plural transfer requests nodes. The request queue manager sorts transfer request packets by their priority level and stores them in the queue manager memory. The request queue manager processes dispatches transfer request packets to a free data channel based upon priority level and first-in-first-out within priority level.

Request queue manager in transfer controller with hub and ports

The configuration bus interconnection protocol provides the configuration interfaces to the memory-mapped registers throughout the digital signal processor chip. The configuration bus is a parallel set of communications protocols, but for control of peripherals rather than for data transfer. While the expanded direct memory access processor is heavily optimized for maximizing data transfers, the configuration bus protocol is made to be as simple as possible for ease of implementation and portability.

Configuration bus reconfigurable/reprogrammable interface for expanded direct memory access processor

A data transfer controller with hub and ports uses an effective channel priority processing technique and algorithm. Data transfer requests are queued in a first-in-first-out fashion at the data source ports. Each data transfer request has a priority level for execution. In effective channel priority processing the priority level assigned to a source port is the greatest priority level of any data transfer request in the corresponding first-in-first-out queue. This techniques prevents a low priority data transfer request at the output of a source port queue from blocking a higher priority data transfer request further back in the queue. Raising the priority of all data transfer requests within a source port queue enables the low priority data transfer request to complete enabling the high priority data transfer request to be reached. Thus both the low priority data transfer request and the high priority data transfer request in the queue of a single port are serviced before intermediate priority data transfer requests at the output of other source port queues.

Effective channel priority processing for transfer controller with hub and ports

This invention enables a program controlled cache state operation on a program designated address range. The program controlled cache state operation could be writeback of data cached from the program designated address range to a higher level memory or such writeback and invalidation of data cached from the program designated address range. A cache operation unit includes a base address register and a word count register loadable by the central processing unit. The program designated address range is from a base address for a number of words of the word count register. In the preferred embodiment the program controlled cache state operation begins upon loading the word count register. The cache operation unit may operate on fractional cache entries by handling misaligned first and last cycles. Alternatively, The cache operation unit may operate only on whole cache entries. The base address register increments and the word count register decrements until when the word count reaches zero.

Growing community of inventors

Richardson, TX, United States of America

Sanjive Agarwala