Computer Organization and Architecture Cache Memory Chapter 4

Characteristics of Computer Memory Location Capacity Unit of transfer Access method Performance Physical type Physical characteristics Organization

Location CPU Internal External Registers, Control Memory Cache – L1, L2, … Internal Main memory (RAM), consists of DRAMs External Secondary memory – hard disks Removable media – ZIP, CD-ROM, Tape

Addressable units are typically words Capacity Word size The natural unit of organization Length = # bits to store a number OR = instruction length But there are exceptions (CRAY, VAX etc.) Common word lengths: 8, 16, 32, 64 bits Addressable units are typically words But most machines allow individual bytes to be addressed N = 2A, where N = size of address space A = address length (in bits) On disks the addressable units are blocks or clusters

Unit of Transfer Internal External #of bits read or written to/from main memory at a time Determined by data bus width May be different from word or addressable unit External Usually a block which is much larger than a word

Access Methods Sequential Direct Start at the beginning and read through in order Access time depends on location of data and previous location e.g. tape Direct Individual blocks have unique address Access is by jumping to block plus sequential search Access time depends on location and previous location, but much faster than sequential access e.g. disk

Access Methods Random Associative Individual addresses identify locations exactly Access time is independent of location or previous access e.g. internal (main) memory RAM, some caches Associative Data is located by a comparison with contents of a portion of the store e.g. many cache systems

Transfer Rate (R = bits / second) Sequential memory performance: Access time (TA) Time between presenting the address and getting the valid data Memory Cycle time Time may be required for the memory to “recover” before next access Cycle time = recovery time + access time Transfer Rate (R = bits / second) Rate at which data can be moved Sequential memory performance: N = # of bits, TN = time for N bits TA = average access time

Physical Types Semiconductor Magnetic Optical Others RAM (DRAM or SRAM) Magnetic Disk & Tape Optical CD & DVD Others Bubble Hologram

Physical Characteristics Decay E.g. magnetic memory can decay - by the effect of radiation sources, closeness of magnetic media, surface wear off Volatility Volatile memory: without power, memory is erased Magnetic memory is volatile, semiconductor can be volatile or non-volatile Erasable Non-erasable memory cannot be altered E.g. ROM = Read-Only Memory Power consumption

Organization Physical arrangement of bits into words Not always obvious E.g. can be interleaved

Memory Hierarchy – multiple levels Slower access Less frequent access Lower cost Bigger capacity

Levels of Memory Hierarchy Registers L1 Cache L2 Cache Main memory Disk cache Disk Optical storage Tape The higher the level, the “closer” the memory is to the CPU – the faster the access is to it

Store related (clustered) data “closer” to the CPU Locality of Reference During the course of the execution of a program, memory references tend to cluster E.g. loops: Same instructions are used repeatedly - body of loop; subroutine/function called Data tends to be an array or table - typically stored at consecutive locations Consequently: storing related (clustered) data in faster accessible memory improves performance Store related (clustered) data “closer” to the CPU

Cache Small amount of fast memory Purpose is to exploit Locality of Reference Contains copies of portions of main memory Sits between normal main memory and CPU May be located on CPU chip or module

Cache operation - overview CPU requests a word (contents of a memory location) Check cache for this word If present (= cache hit), get from cache (fast) If not present (= cache miss), read a block from main memory, that contains this word, to cache Simultaneously, deliver word to CPU Cache includes tags to identify which block of main memory is in each cache slot

Cache Read Operation

Cache Design M = 2n / K blocks in RAM C << M blocks in cache

Typical Cache Organization

Cache Characteristics Size Mapping Function Direct Associative Set associative Replacement Algorithm LRU, FIFO, LFU, Random Write Policy Write through, write back, write once Block/Line Size Number of Caches

Size does matter Cost Speed Larger cache is more expensive Larger cache is faster But only up to a point Beyond which access slows down Because checking cache for data takes more time

Mapping Function Example Parameters RAM size = 16 MBytes (224 bytes) Byte-addressable machine (1 word = 1 byte) RAM consists of 222 (4 M) 4 byte (word) blocks 24 bit address 224=16M Cache size =64 Kbytes (216 bytes) Cache block size = 4 bytes cache consists of m =214 (16k) lines of 4 byte lines

Each block of main memory maps to only one cache line Direct Mapping Each block of main memory maps to only one cache line i.e. if a block is in cache, it must be in one specific place Address is in two parts Least Significant w bits identify unique word Most Significant s bits specify one memory block The MSBs are split into a cache line field r and a tag of s-r (most significant) Mapping Function: i = j mod m where i = cache line, j = block number m =number of lines in cache

Direct Mapping Cache Line Table Cache line Main Memory blocks held 0 0, m, 2m, 3m, … ,2s-m 1 1,m+1, 2m+1, …, 2s-m+1 m-1 m-1, 2m-1,3m-1, …, 2s-1 Note: m = 2r = 214 is the number of lines in cache s = 22 is the block identifier 2s = 222 blocks in RAM

Direct Mapping Address Structure Tag s-r Line or Slot r Word w 14 2 8 24 bit address w = 2 bit word identifier (4 byte/word block), also called offset s = 22 bit block identifier 8 bit tag (=22-14) : distinguishes blocks that map to same cache line 2s-r = 28 = 64 blocks map to the same line r =14 bit slot# or line#: identifies cache line a particular block maps to m =214 lines in cache No two blocks in the same line have the same Tag field Check contents of cache by finding Line and checking Tag

Direct Mapping Cache Organization

Direct Mapping Example

Direct Mapping Summary Address length = (s + w) bits Number of addressable units = 2s+w words or bytes Block size = line size = 2w words or bytes Number of blocks in main memory = 2s+ w/2w = 2s Number of lines in cache = m = 2r Size of tag = (s – r) bits Block j maps to line i = j mod m

Direct Mapping pros & cons Simple Inexpensive Fixed location for given block A given block always maps to the same line in cache If a program accesses 2 blocks that map to the same line repeatedly, cache misses are very high This is called thrashing

A main memory block can load into any line of cache Associative Mapping A main memory block can load into any line of cache The line is determined by a replacement algorithm Memory address is interpreted as tag and word Tag uniquely identifies block of memory Every line’s tag is examined for a match This is done in parallel for all tags Requires complex circuitry Cache searching gets expensive

Associative Mapping Address Structure Word 2 bit Tag 22 bit 22 bit tag stored with each 32 bit block of data Compare tag field with tag entry in cache to check for hit Least significant 2 bits of address identify which 16 bit word is required from 32 bit data block (offset) e.g. Address Tag Data Cache line FFFFFC FFFFFC 24682468 3FFF

Associative Mapping Example

Fully Associative Cache Organization

Associative Mapping Summary Address length = (s + w) bits Number of addressable units = 2s+w words or bytes Block size = line size = 2w words or bytes Number of blocks in main memory = 2s+ w/2w = 2s Number of lines in cache = any determined by cache size, not address length Size of tag = s bits A block can map to any line, determined by a replacement algorithm

Set Associative Mapping Combination of direct and associative mapping Cache is divided into a number of sets Each set contains a number of lines Use direct (mod) mapping to identify a set, replacement algorithm within the set Thus a block maps to any line in a given set e.g. Block B can be in any line of set i E.g. 2 lines per set Called 2-way associative mapping A given block can be in one of 2 lines in a particular set

Set Associative Mapping Address Structure Word 2 bit Tag: 9 bit Set: 13 bit Use set field to determine cache set to look in Compare tag field to see if there is a hit Word (offset) identifies word/byte within block e.g. Address Tag Data Set number 1FF 7FFC 1FF 12345678 1FFF 001 7FFC 001 11223344 1FFF

k-Way Set Associative Cache Organization

Two Way Set Associative Mapping Example

Set Associative Mapping Summary Address length = (s + w) bits Number of addressable units = 2s+w words or bytes Block size = line size = 2w words or bytes Number of blocks in main memory = 2s Number of lines in set = k Number of sets = v = 2d Number of lines in cache = kv = k * 2d Size of tag = (s – d) bits Block j maps to set i = j mod v

Replacement Algorithms: Associative & Set Associative Mapping Hardware implemented algorithms (for speed) Least Recently used (LRU) replace the block that has not had a hit the longest time First in first out (FIFO) replace block that has been in cache longest Least frequently used (LFU) replace block which has had fewest hits Random isn’t significantly inferior

Need to insure coherence of cache and main memory Write Policy Need to insure coherence of cache and main memory Must not overwrite a cache block with a main memory block unless main memory is up to date Main memory might be out of date for a number of reasons Multiple CPUs may have individual caches I/O may address main memory directly Cache blocks may also be made invalid E.g. by I/O writes to main memory

Write through All writes go to main memory as well as cache Multiple CPUs can monitor main memory traffic to keep local (to CPU) cache up to date Lots of traffic Slows down writes

Minimize memory writes Updates initially made in cache only Write back Minimize memory writes Updates initially made in cache only UPDATE bit for a cache slot is set when update occurs Portions of MM are invalid If block is to be replaced, write to main memory only if update bit is set I/O must access main memory through cache Due to invalid MM contents Note: 15% of memory references are writes

Other caches can still be out of sync Cache Coherence With multiple caches: a write policy insures only consistency between one cache and main memory Other caches can still be out of sync If the same block is present in multiple caches, a write invalidates not only main memory but the other caches, too Cache coherence: all caches are in sync (and the main memory, as well) Bus watching with write through Hardware synch Non-cacheable memory: shared memory not cached

Larger line (block) sizes and therefore fewer lines can be used Line Size Larger line (block) sizes and therefore fewer lines can be used Beneficial because hit ratio is increased But only up to a point Too large line size will replace useful existing cache content with less useful (farther words in large block) Probability of needing data in cache becomes less than probability of needing data not in cache This will reduce hit ratio There is no “universal optimum” Depends on program characteristics 8 to 32 byte block size seems to work well in most situations

Unified cache: data and instructions are cached in the same cache Unified vs. Split Cache Unified cache: data and instructions are cached in the same cache Split cache: separate caches for data and instructions Advantages of unified cache Balances possible imbalance between amount of data and instructions in a program Only one cache needs to be manufactured Advantage of split cache Eliminates contention between INS fetch and execute units Supports pipelining and speculative execution

Pentium 4 Cache 80386 – no on-chip cache 80486 – 8k cache using 16 byte lines and four way set associative organization Pentium (all versions) – two on-chip L1 caches Data & instructions Pentium 4 – 2 L1 caches (on-chip) one instruction cache and one data cache 8k bytes 64 byte lines four way set associative L2 cache – off-chip Feeding both L1 caches 256k 128 byte lines 8 way set associative

Pentium 4 Diagram (Simplified)

Pentium 4 Core Processor Fetch/Decode Unit Fetches instructions from L2 cache Decode into micro-ops Store micro-ops in L1 cache Out of order execution logic Schedules micro-ops Based on data dependence and resources May speculatively execute Execution units Execute micro-ops Data from L1 cache Results in registers Memory subsystem L2 cache and systems bus

Pentium 4 Design Reasoning Decodes instructions into RISC like micro-ops before L1 cache Micro-ops fixed length Superscalar pipelining and scheduling Pentium instructions are long and complex Performance improved by separating decoding from scheduling & pipelining (More later – ch14) Data cache is write back Can be configured to write through L1 cache controlled by 2 bits in register CD = cache disable NW = not write through 2 instructions to invalidate (flush) cache and write back, then invalidate

Power PC Cache Organization 601 – single 32kb 8 way set associative 603 – 16kb (2 x 8kb) two way set associative 604 – 32kb 610 – 64kb G3 & G4 64kb L1 cache 8 way set associative 256k, 512k or 1M L2 cache two way set associative

PowerPC G4

Comparison of Cache Sizes