1.3 The Pentium Processor 1.3.1 On-Chip Caches The Pentium processor implements two internal caches for a total integrated cache size of 16 Kbytes: an 8 Kbyte data cache and a separate 8 Kbyte code cache. These caches are transparent to application software to maintain compatibility with previous Intel Architecture generations. The data cache fully supports the MESI (modified/exclusive/shared/invalid) writeback cache consistency protocol. The code cache is inherently write protected to prevent code from being inadvertently corrupted, and as a consequence supports a subset of the MESI protocol, the S(shared) and I (invalid) states. The caches have been designed for maximum flexibility and performance. The data cache is configurable as writeback or writethrough on a line-by-line basis. Memory areas can be defined as non-cacheable by software and external hardware. Cache writeback and invalidations can be initiated by hardware or software. Protocols for cache consistency and line replacement are implemented in hardware, easing system design. 1.3.2 Cache Organization On the Pentium processor, each of the caches are 8 Kbytes in size and each is organized as a 2-way set associative cache. There are 128 sets in each cache, each set containing 2 lines (each line has its own tag address). Each cache line is 32 bytes wide. In the Pentium processor, replacement in both the data and instruction caches is handled by the LRU mechanism which requires one bit per set in each of the caches. The data cache consists of eight banks interleaved on 4-byte boundaries. The data cache can be accessed simultaneously from both pipes, as long as the references are to different cache banks. A conceptual diagram of the organization of the data and code caches is shown in Figure 2-8. Note that the data cache supports the MESI writeback cache consistency protocol which requires 2 state bits, while the code cache supports the S and I state only and therefore requires only one state bit.
Figure 1-3 Conceptual Organization of Code and Data Caches
1.3.3 Cache Structure The instruction and data caches can be accessed simultaneously. The instruction cache can provide up to 32 bytes of raw opcodes and the data cache can provide data for two data references all in the same clock. This capability is implemented partially through the tag structure. The tags in the data cache are triple ported. One of the ports is dedicated to snooping while the other two are used to lookup two independent addresses corresponding to data references from each of the pipelines. The instruction cache tags of the Pentium processor are also triple ported. Again, one port is dedicated to support snooping and the other two ports facilitate split line accesses (simultaneously accessing upper half of one line and lower half of the next line). The storage array in the data cache is single ported but interleaved on 4-byte boundaries to be able to provide data for two simultaneous accesses to the same cache line. Each of the caches are parity protected. In the instruction cache, there are parity bits on a quarter line basis and there is one parity bit for each tag. The data cache contains one parity bit for each tag and a parity bit per byte of data. Each of the caches are accessed with physical addresses and each cache has its own TLB (translation lookaside buffer) to translate linear addresses to physical addresses. The TLBs associated with the instruction cache are single ported whereas the data cache TLBs are fully dual ported to be able to translate two independent linear addresses for two data references simultaneously. The tag and data arrays of the TLBs are parity protected with a parity bit associated with each of the tag and data entries in the TLBs. The data cache of the Pentium processor has a 4-way set associative, 64-entry TLB for 4-Kbyte pages and a separate 4-way set associative, 8-entry TLB to support 4-Mbyte pages. The code cache has one 4-way set associative, 32-entry TLB for 4Kbyte pages and 4-Mbyte pages which are cached in 4-Kbyte increments.
Replacement in the TLBs is handled by a pseudo LRU mechanism (similar to the Intel486 CPU) that requires 3 bits per set.
1.4 The Pentium Pro /Pentium II/Pentium III 1.4.1 The Pentium pro The Pentium Pro Processor on-chip level one (L1) caches consist of one 8-Kbyte four-way set associative instruction cache unit with a cache line length of 32 bytes and one 8-Kbyte two-way set associative data cache unit. Not all misses in the L1 cache expose the full memory latency. The level two (L2) cache masks the full latency caused by an L1 cache miss. The minimum delay for a L1 and L2 cache miss is between 11 and 14 cycles based on DRAM page hit or miss. The data cache can be accessed simultaneously by a load instruction and a store instruction, as long as the references are to different cache banks.
Figure 1.4 The Pentium Pro, II, III Processor Micro-Architecture with Advanced TransferCache Enhancement, The first and second level caches
1.4.2 The Pentium II /Pentium III The on-chip cache subsystem of Pentium II and Pentium III processors consists of two 16-Kbyte four-way set associative caches with a cache line length of 32 bytes. The caches employ a write-back mechanism and a pseudo-LRU (least recently used) replacement algorithm. The data cache consists of eight banks interleaved on four-byte boundaries.Level two (L2) caches have been off chip but in the same package. They are 128K or more in size. L2 latencies are in the range of 4 to 10 cycles. An L2 miss initiates a transaction across the bus to memory chips. Such
an access requires on the order of at least 11 additional bus cycles, assuming a DRAM page hit. A DRAM page miss incurs another three bus cycles. Each bus cycle equals several processor cycles, for example, one bus cycle for a 100 MHz bus is equal to four processor cycles on a 400 MHz processor. The speed of the bus and sizes of L2 caches are implementation dependent, however. Check the specifications of a given system to understand the precise characteristics of the L2 cache.
Figure 1-5 The Intel NetBurst Micro-Architecture, the First Level, the Second Level Caches and Trace Cache
1.5 The Pentium 4 Processor The Intel Pentium 4 processor is the latest IA-32 processor, and the first based on the Intel NetBurst micro-architecture ( Figure1.5). The Intel NetBurst microarchitecture can support up to three levels of on-chip cache. Only two levels of onchip caches are implemented in the Pentium 4 processor, but there brings a new concept: Trace Caches. The level nearest to the execution core of the processor, the first level, contains separate caches for instructions and data: a first-level data cache and the trace cache, which is an advanced first-level instruction cache. All other levels of caches are shared. The levels in the cache hierarchy are not inclusive, that is, the fact that a line is in level i does not imply that it is also in level i+1. All caches use a pseudo-LRU (least recently used) replacement algorithm. 1.5.1 Execution Trace Cache The execution trace cache (TC) is the primary instruction cache in the Intel NetBurst micro-architecture. The TC stores decoded IA-32 instructions, or µops. This removes decoding costs on frequently-executed code, such as template restrictions and the extra latency to decode instructions upon a branch misprediction.
In the Pentium 4 processor implementation, the TC can hold up to 12K µops and can deliver up to three µops per cycle. The TC does not hold all of the µops that need to be executed in the execution core. In some situations, the execution core may need to execute a microcode flow, instead of the µop traces that are stored in the trace cache.The Pentium 4 processor is optimized so that most frequently-executed IA-32 instructions come from the trace cache, efficiently and continuously, while only a few instructions involve the microcode ROM. 1.5.2 The Second-level Cache A second-level cache miss initiates a transaction across the system bus interface to the memory sub-system. The system bus interface supports using a scalable bus clock and achieves an effective speed that quadruples the speed of the scalable bus clock. It takes on the order of 12 processor cycles to get to the bus and back within the processor, and 6-12 bus cycles to access memory if there is no bus congestion. Each bus cycle equals several processor cycles. The ratio of processor clock speed to the scalable bus clock speed is referred to as bus ratio. For example, one bus cycle for a 100 MHz bus is equal to 15 processor cycles on a 1.50 GHz processor.
