摘要:
A method and apparatus for supporting heterogeneous agents in on-chip busses. In one embodiment, the method includes the detection of a bus arbitration event between at least a first bus agent and a second bus agent. In one embodiment, a bus arbitration event is detected when at least the first bus agent and the second bus agent assert their respective bus request signals in a single clock cycle. Once a bus arbitration event is detected, bus ownership may be granted to both the first bus agent and the second bus agent, when the first bus agent and the second bus agent have different grant-to-valid latencies. In the embodiment, heterogeneous bus agents may coexist on a bus without requiring wasted or unused bus cycles following establishment of bus ownership. Other embodiments are described and claimed.
摘要:
Systems and methods for early fixed latency subtractive decoding are disclosed. The subtractive decoding device speculatively acknowledges a bus transaction within a fixed time period that is the same as the time period for positive decoding. Pipelining of a new bus transaction may therefore be accomplished each new time period. A bus transaction may be retried if no acknowledgement occurs within the fixed time period.
摘要:
Methods and apparatus for cache-to-cache block transfers (i.e., intervention) when the state of the transferred block is in a non-modified state (e.g., “exclusive” or “shared”) are provided. In one embodiment, a first cache holds the memory block in an “exclusive” state prior to the transfer. When a processor associated with a second cache attempts to read the block from a main memory, the first cache intervenes and supplies the block to the second cache. The state of the block in the first cache changes from “exclusive” to “shared.” In another embodiment, a processor associated with a third cache attempts to read the block from the main memory while the first cache and the second both hold the block in the “shared” state. Either the first cache or the second cache wins an arbitration and supplies the block to the third cache.
摘要:
Methods and apparatus for cache-to-cache block transfers (i.e., intervention) when the state of the transferred block is in a non-modified state and/or a modified state, without asserting a hit-modified signal line, are provided. In one example, a first cache holds the memory block prior to the transfer. When a processor associated with a second cache attempts to read the block from a main memory, the first cache intervenes and supplies the block to the second cache regardless of the state (modified or non-modified) of the cached block. In addition, an agent associated with the first cache asserts a “hit” signal line regardless of the state (modified or non-modified) of the cached block. The agent associated with the first cache does not assert a “hit-modified” signal line.
摘要:
A bus control system includes N bus agents each having a corresponding bus request delay and M bus agents each having a corresponding bus request delay. A controller determines the bus request delays of the N bus agents and the M bus agents and grants concurrent ownership of a bus to each of the N bus agents and non-concurrent ownership of the bus to each of the M bus agents based on the determination. M and N are integers greater than 1.
摘要:
Some embodiments of the invention include an address interconnect and a data interconnect to transfer data among a number of devices. The data interconnect is configured to transfer data among the devices via multiple transfer paths. A transfer of data on one transfer path is independent from a transfer of data on another transfer path. In some cases, data is concurrently transferred among more than two of the devices on at least one of the address interconnect and the data interconnect. Other embodiments are described and claimed.
摘要:
A method and apparatus for optimizing line writes in cache coherent systems. A new cache line may be allocated without loading data to fill the new cache line when a store buffer coalesces enough stores to fill the cache line. Data may be loaded to fill the line if an insufficient number of stores are coalesced to fill the entire cache line. The cache line may be allocated by initiating a read and invalidate request and asserting a back-off signal to cancel the read if there is an indication that the coalesced stores will fill the cache line.
摘要:
Systems and methods for early fixed latency subtractive decoding are disclosed. The subtractive decoding device speculatively acknowledges a bus transaction within a fixed time period that is the same as the time period for positive decoding. Pipelining of a new bus transaction may therefore be accomplished each new time period. A bus transaction may be retried if no acknowledgement occurs within the fixed time period.
摘要:
A system and method for improved cache performance is disclosed. In one embodiment, a processor with a cache having a dirty cache line subject to eviction may send the dirty cache line to an available replacement block in another processor's cache. In one embodiment, an available replacement block may contain a cache line in an invalid state. In another embodiment, an available replacement block may contain a cache line in an invalid state or in a shared state. Multiple transfers of the dirty cache line to more than one processor's cache may be inhibited using a set of accept signals and backoff signals. These accept signals may be combined to inhibit multiple processors from accepting the dirty cache line, as well as to inhibit the system memory from accepting the dirty cache line.