Master-slave cache system for instruction and data cache memories

• US 5,551,001 A
• Filed: 06/29/1994
• Issued: 08/27/1996
• Est. Priority Date: 06/29/1994
• Status: Expired due to Term

• Current Assignee: Samsung Electronics Co. Ltd. (Samsung Group)
• Original Assignee: Exponential Technology, Inc.
• Inventors: Cohen, Earl T., Blomgren, James S., Pattin, Jay C., Tilleman, Russell W.
• Primary Examiner(s): Swann, Tod R.
• Assistant Examiner(s): PEIKARI, BEHZAD
• Application Number:
  US08/267,658
• Time in Patent Office: 790 Days
• Field of Search: 395/450, 395/455, 395/457, 395/459, 395/460, 395/462, 395/463, 395/467, 395/471, 395/473, 395/469, 395/449
• US Class Current: 711/122
• CPC Class Codes: G06F 12/0831 using a bus scheme, e.g. with bus monitoring or watching means
• CPC Class Codes: G06F 12/0848 Partitioned cache, e.g. separate instruction and operand caches
• CPC Class Codes: G06F 12/0857 by multiple requestors
• CPC Class Codes: G06F 12/0897 with two or more cache hierarchy levels with multilevel cache hierarchies G06F12/0811

Abstract

A master-slave cache system has a large, set-associative master cache, and two smaller direct-mapped slave caches, a slave instruction cache for supplying instructions to an instruction pipeline of a processor, and a slave data cache for supplying data operands to an execution pipeline of the processor. The master cache and the slave caches are tightly coupled to each other. This tight coupling allows the master cache to perform most cache management operations for the slave caches, freeing the slave caches to supply a high bandwidth of instructions and operands to the processor'"'"'s pipelines. The master cache contains tags that include valid bits for each slave, allowing the master cache to determine if a line is present and valid in either of the slave caches without interrupting the slave caches. The master cache performs all search operations required by external snooping, cache invalidation, cache data zeroing instructions, and store-to-instruction-stream detection. The master cache interrupts the slave caches only when the search reveals that a line is valid in a slave cache, the master cache causing the slave cache to invalidate the line. A store queue is shared between the master cache and the slave data cache. Store data is written from the store queue directly in to both the slave data cache and the master cache, eliminating the need for the slave data cache to write data through to the master cache. The master-slave cache system also eliminates the need for a second set of address tags for snooping and coherency operations. The master cache can be large and designed for a low miss rate, while the slave caches are designed for the high speed required by the processor'"'"'s pipelines.

Claims

1. A master-slave cache system for transferring data between a main memory and a central processing unit (CPU), the CPU having an instruction pipeline decoding instructions at a first rate, the CPU having an execution pipeline executing at a second rate, the main memory storing a plurality of operands and a plurality of instructions, the system comprising:

a master cache for storing operands and instructions, the master cache coupled to the main memory, the master cache storing a first subset of the plurality of operands and a second subset of the plurality of instructions stored in the main memory, the master cache storing a third subset of instructions and a fourth subset of operands, the third subset being a subset of the second subset, and the fourth subset being a subset of the first subset;

a slave instruction cache, coupled to the master cache and coupled to the instruction pipeline, for storing the third subset of instructions, the slave instruction cache capable of transferring instructions to the instruction pipeline at the first rate, the slave instruction cache comprising a cache that is read-only by the CPU; and

a slave data cache, coupled to the master cache and coupled to the execution pipeline, for storing the fourth subset of operands, the slave data cache capable of transferring operands to the execution pipeline at the second rate;

wherein the master cache comprises a set-associative cache, and the slave instruction cache and the slave data cache comprise direct-mapped caches;

wherein the master cache further includes means for replacing operands and instructions using a modified least-recently-used algorithm, never replacing a least-recently used operand or instruction that is present in the slave data cache or the slave instruction cache,

whereby the slave instruction cache matches the first rate required by the instruction pipeline, and the slave data cache matches the second rate required by the execution pipeline.

2. The system of claim 1 wherein the master cache has a capacity to contain at least four times a maximum number of operands and instructions that can be stored in both the slave data cache and the slave instruction cache.

3. The system of claim 1 wherein the slave data cache comprises a write-through cache, the slave data cache for writing a copy of an operand back into the master cache when the execution pipeline writes the operand to the slave data cache.

4. The system of claim 1 wherein the first subset and the second subset overlap, wherein an overlapping operand in the first subset that has an address matching an address of an instruction in the second subset indicates that the overlapping operand is being stored to an instruction stream stored in the plurality of instructions.

5. The system of claim 1 further comprising:

prefetching means, coupled to the master cache, for fetching instructions from the main memory into the master cache, the prefetching means fetching instructions having sequential addresses following a miss address from the slave instruction cache.

6. A master-slave cache system for transferring data between a main memory and a central processing unit (CPU), the CPU having an instruction pipeline decoding instructions at a first rate, the CPU having an execution pipeline executing at a second rate, the main memory storing a plurality of operands and a plurality of instructions, the system comprising:

a master cache for storing operands and instructions, the master cache coupled to the main memory, the master cache storing a first subset of the plurality of operands and a second subset of the plurality of instructions stored in the main memory, the master cache storing a third subset of instructions and a fourth subset of operands, the third subset being a subset of the second subset, and the fourth subset being a subset of the first subset;

a slave instruction cache, coupled to the master cache and coupled to the instruction pipeline, for storing the third subset of instructions, the slave instruction cache capable of transferring instructions to the instruction pipeline at the first rate;

a slave data cache, coupled to the master cache and coupled to the execution pipeline, for storing the fourth subset of operands, the slave data cache capable of transferring operands to the execution pipeline at the second rate;

a plurality of master tags, stored in the master cache, each master tag in the plurality of master tags comprising;

an address tag field, for storing a tag portion of an address of a data item, the data item being one of the plurality of operands or one of the plurality of instructions stored in main memory;

master valid indicating means for indicating that the data item is valid and present in the master cache;

slave-instruction valid indicating means for indicating that the data item is valid and present in the slave instruction cache; and

slave-data valid indicating means for indicating that the data item is valid and present in the slave data cache,

invalidation means, coupled to the master cache and to the CPU, for modifying the plurality of master tags;

the CPU including means for providing an invalidation address to the invalidation means;

the invalidation means further including;

means for signaling an invalidation hit if a tag portion of the invalidation address matches the address tag field in a matching line in the plurality of master tags;

master invalidating means for invalidating the data item in the matching line in the master cache if the invalidation hit is signaled and the master valid indicating means indicates that the data item is valid and present in the master cache;

first invalidating means for invalidating a first copy of the data item in the slave data cache if the invalidation hit is signaled and the slave-data valid indicating means indicates that the first copy of the data item is valid and present in the slave data cache; and

second invalidating means for invalidating a second copy of the data item in the slave instruction cache if the invalidation hit is signaled and the slave-instruction valid indicating means indicates that the second copy of the data item is valid and present in the slave instruction cache,

whereby the master cache contains information on valid words present in the slave instruction cache and the slave data cache and the slave instruction cache matches the first rate required by the instruction pipeline, and the slave data cache matches the second rate required by the execution pipeline and whereby all invalidation requests from the CPU are processed by the master cache.

7. The system of claim 6 further including an external snoop requester coupled to the invalidation means, the external snoop requester providing an external snoop address to the invalidation means.

8. The system of claim 7 further comprising

means for signaling a snoop hit if the tag portion of the external snoop address matches the address tag field in a snooped line in the plurality of master tags;

the master invalidating means invalidating the data item in the snooped line in the master cache if the snoop hit is signaled and the master valid indicating means indicates that the data item is valid and present in the master cache;

the first invalidating means invalidating the first copy of the data item in the slave data cache if the snoop hit is signaled and the slave-data valid indicating means indicates that the first copy of the data item is valid and present in the slave data cache; and

the second invalidating means invalidating the second copy of the data item in the slave instruction cache if the snoop hit is signaled and the slave-instruction valid indicating means indicates that the second copy of the data item is valid and present in the slave instruction cache,

whereby all snoop requests from the external snoop requester are processed by the master cache.

9. The system of claim 6 wherein the first copy of the data item in the slave data cache and the second copy of the data item in the slave instruction cache can not both be valid at the same time.

10. The system of claim 9 wherein

the first invalidating means invalidates the first copy of the data item in the slave data cache when the slave instruction cache requests that the second copy of the data item be validated and placed in the slave instruction cache;

and

the second invalidating means invalidates the second copy of the data item in the slave instruction cache when the slave data cache requests that the first copy of the data item be validated and placed in the slave data cache,

thereby keeping the slave instruction cache and the slave data cache coherent and allowing a store to the instruction stream.

11. The system of claim 10 wherein

the master cache signals a pipe-flush to the CPU when the execution pipeline writes to the data item and the slave instruction valid indicating means indicates that the second copy of the data item is valid and present in the slave instruction cache;

the instruction pipeline being flushed when the pipe-flush is signaled.

12. The system of claim 6 further comprising a store queue, the store queue further comprising:

means for receiving store data and an address of the store data from the execution pipeline;

means for writing store data to the master cache when the master cache is not busy;

means for writing store data to the slave data cache if the address of the store data matches an address of an operand in the slave data cache and the slave data cache is not busy; and

means for storing a plurality of store data and a plurality of addresses of the store data,

whereby the store queue is shared between the master cache and the slave data cache, the store queue receiving store data from the execution pipeline and writing the store data when the master cache and the slave data cache are each not busy.

13. The system of claim 12 wherein the execution pipeline uses the store queue to write store data to the slave data cache and to the master cache, eliminating a need for the slave data cache to write through to the master cache.

14. The system of claim 13 wherein invalidation and cache management operations from the CPU are placed in the store queue and written only to the master cache and not to the slave data cache, the master cache performing invalidation and cache management operations for the slave data cache.

15. The master-slave cache system of claim 6 further comprising:

selecting means for selecting a selected line in the master cache using a portion of a zeroing address;

reading means for reading the address tag field in the master tag for the selected line and allocating a new line in the master cache if the address tag field does not match a tag portion of the zeroing address;

signal means for sending an invalidation signal to the slave instruction cache if the slave-instruction valid indicating means in the master tag for the selected line indicates that the data item in the selected line is valid and present in the slave instruction cache, the slave instruction cache invalidating its copy of the selected line in response to the invalidation signal; and

zero means for successively writing a zero value to all bytes of data in the selected line in the master cache,

whereby the cache block is zeroed in the master cache without interrupting the slave instruction cache except when the copy of the selected line is in the slave instruction cache.

16. The master-slave cache system of claim 6 further comprising:

selecting means for selecting a selected line in the master cache using a portion of a zeroing address;

reading means for reading the address tag field in the master tag for the selected line and allocating a new line in the master cache if the address tag field does not match a tag portion of the zeroing address;

signal means for sending an invalidation signal to the slave data cache if the slave-data valid indicating means in the master tag for the selected line indicates that the data item in the selected line is valid and present in the slave data cache, the slave data cache invalidating its copy of the selected line in response to the invalidation signal; and

zero means for successively writing a zero value to all bytes of data in the selected line in the master cache,

whereby the cache block is zeroed in the master cache without interrupting the slave data cache except when the copy of the selected line is in the slave data cache.