Reduced instruction set computers

Reduced instruction set computers

• 1.
  Datorarkitektur Fö 5/6 - 1 Datorarkitektur Fö 5/6 - 2 What are RISCs and why do we need them? REDUCED INSTRUCTION SET COMPUTERS • RISC architectures represent an important (RISC) innovation in the area of computer organization. • The RISC architecture is an attempt to produce more CPU power by simplifying the instruction set of the CPU. • The opposed trend to RISC is that of complex instruction set computers (CISC). 1. Why do we need RISCs? 2. Some Typical Features of Current Programs Both RISC and CISC architectures have been developed as an attempt to cover the semantic gap. 3. Main Characteristics of RISC Architectures 4. Are RISCs Really Better than CISCs? 5. Examples Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH Datorarkitektur Fö 5/6 - 3 Datorarkitektur Fö 5/6 - 4 The Semantic Gap The Semantic Gap (cont’d) growing abstraction level High-level language (HLL) Problem: How should new HLL programs be compiled the semantic gap and executed efficiently on a processor architecture? Two possible answers: Machine language (machine architecture fully visible) 1. The CISC approach: design very complex architectures including a large number of instructions and addressing modes; include also instructions close to those present in HLL. • In order to improve the efficiency of software development, new and powerful programming 2. The RISC approach: simplify the instruction set and languages have been developed (Ada, C++, Java). adapt it to the real requirements of user programs. They provide: high level of abstraction, conciseness, power. • By this evolution the semantic gap grows. Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH
• 2.
  Datorarkitektur Fö 5/6 - 5 Datorarkitektur Fö 5/6 - 6 Evaluation of Program Execution Frequency of Instructions Executed or • Frequency distribution of executed machine What are Programs Doing Most of the Time? instructions: - moves: 33% - conditional branch: 20% • Several studies have been conducted to determine - Arithmetic/logic: 16% the execution characteristics of machine instruction - others: Between 0.1% and 10% sequences generated from HLL programs. • Addressing modes: the overwhelming majority of instructions uses simple addressing modes, in • Aspects of interest: which the address can be calculated in a single cycle (register, register indirect, displacement); 1. The frequency of operations performed. complex addressing modes (memory indirect, 2. The types of operands and their frequency of indexed+indirect, displacement+indexed, stack) are use. used only by ~18% of the instructions. 3. Execution sequencing (frequency of jumps, loops, subprogram calls). Operand Types • 74 to 80% of the operands are scalars (integers, reals, characters, etc.) which can be hold in registers; • the rest (20-26%) are arrays/structures; 90% of them are global variables; • 80% of the scalars are local variables. The majority of operands are local variables of scalar type, which can be stored in registers. Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH Datorarkitektur Fö 5/6 - 7 Datorarkitektur Fö 5/6 - 8 Procedure Calls (cont’d) Procedure Calls int x, y; float z; void proc3(int a, int b, int c) { - - - - - - - - - - - - - -- - - Investigations have been also performed about the } percentage of the total execution time spent executing a --------------------- certain HLL instruction. void proc2(int k) { int j, q; • It turned out that comparing HLL instructions, most ----------------- of the time is spent executing CALLs and proc3(j, 5, q); RETURNs. } • Even if only 15% of the executed HLL instructions --------------------- is a CALL or RETURN, they are executed most of void proc1() { the time, because of their complexity. int i; • A CALL or RETURN is compiled into a relatively ----------------- long sequence of machine instructions with a lot of proc2(i); memory references. } --------------------- main () { proc1(); } Some statistics concerning procedure calls: • Only 1.25% of called procedures have more than six parameters. • Only 6.7% of called procedures have more than six local variables. • Chains of nested procedure calls are usually short and only very seldom longer than 6. Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH
• 3.
  Datorarkitektur Fö 5/6 - 9 Datorarkitektur Fö 5/6 - 10 Main Characteristics of RISC Architectures Conclusions from Evaluation of Program Execution • The instruction set is limited and includes only simple instructions. • An overwhelming preponderance of simple (ALU and move) operations over complex operations. - The goal is to create an instruction set contain- • Preponderance of simple addressing modes. ing instructions that execute quickly; most of the • Large frequency of operand accesses; on average RISC instructions are executed in a single each instruction references 1.9 operands. machine cycle (after fetched and decoded). • Most of the referenced operands are scalars (so • Pipeline operation (without memory reference): they can be stored in a register) and are local variables or parameters. FI DI EI • Optimizing the procedure CALL/RETURN mechanism promises large benefits in speed. - RISC instructions, being simple, are hard-wired, while CISC architectures have to use micropro- gramming in order to implement complex These conclusions have been at the starting point to the instructions. Reduced Instruction Set Computer (RISC) approach. - Having only simple instructions results in re- duced complexity of the control unit and the data path; as a consequence, the processor can work at a high clock frequency. - The pipelines are used efficiently if instructions are simple and of similar execution time. - Complex operations on RISCs are executed as a sequence of simple RISC instructions. In the case of CISCs they are executed as one single or a few complex instruction. Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH Datorarkitektur Fö 5/6 - 11 Datorarkitektur Fö 5/6 - 12 Main Characteristics of RISC Architectures (cont’d) Main Characteristics of RISC Architectures (cont’d) Let’s see some small example: • Load-and-store architecture Assume: - Only LOAD and STORE instructions reference - we have a program with 80% of executed data in memory; all other instructions operate instructions being simple and 20% complex; only with registers (are register-to-register in- structions); thus, only the few instructions ac- - on a CISC machine simple instructions take 4 cessing memory need more than one cycle to cycles, complex instructions take 8 cycles; execute (after fetched and decoded). cycle time is 100 ns (10-7 s); • Pipeline operation with memory reference: - on a RISC machine simple instructions are executed in one cycle; complex operations are FI DI CA TR CA: compute address implemented as a sequence of instructions; we consider on average 14 instructions (14 cycles) TR: transfere for a complex operation; cycle time is 75 ns (0.75 * 10-7 s). • Instructions use only few addressing modes - Addressing modes are usually register, direct, How much time takes a program of 1 000 000 instructions? register indirect, displacement. CISC: (106*0.80*4 + 106*0.20*8)*10-7 = 0.48 s RISC: (106*0.80*1 + 106*0.20*14)*0.75*10-7 = 0.27 s • Instructions are of fixed length and uniform format • complex operations take more time on the RISC, - This makes the loading and decoding of but their number is small; instructions simple and fast; it is not needed to wait until the length of an instruction is known in • because of its simplicity, the RISC works at a order to start decoding the following one; smaller cycle time; with the CISC, simple instructions are slowed down because of the - Decoding is simplified because opcode and ad- increased data path length and the increased dress fields are located in the same position for control complexity. all instructions Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH
• 4.
  Datorarkitektur Fö 5/6 - 13 Datorarkitektur Fö 5/6 - 14 Main Characteristics of RISC Architectures (cont’d) Main Characteristics of RISC Architectures (cont’d) Is the above strategy realistic? • A large number of registers is available - The strategy is realistic, because the number of - Variables and intermediate results can be local variables in procedures is not large. The stored in registers and do not require repeated chains of nested procedure calls is only excep- loads and stores from/to memory. tionally larger than 6. - All local variables of procedures and the passed - If the chain of nested procedure calls becomes parameters can be stored in registers (see slide large, at a certain call there will be no registers 8 for comments on possible number of variables to be assigned to the called procedure; in this and parameters). case local variables and parameters have to be stored in memory. What happens when a new procedure is called? - Normally the registers have to be saved in Why is a large number of registers typical for RISC memory (they contain values of variables and architectures? parameters for the calling procedure); at return to the calling procedure, the values have to be again loaded from memory. This takes a lot of - Because of the reduced complexity of the proc- time. essor there is enough space on the chip to be allocated to a large number of registers. This, - If a large number of registers is available, a new usually, is not the case with CISCs. set of registers can be allocated to the called procedure and the register set assigned to the calling one remains untouched. Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH Datorarkitektur Fö 5/6 - 15 Datorarkitektur Fö 5/6 - 16 Main Characteristics of RISC Architectures (cont’d) The delayed load problem (cont’d) • With delayed load the processor always executes The delayed load problem the instruction following a LOAD, without a stall; It is the programmers (compilers) responsibility that this instruction does not need the loaded value. • LOAD instructions (similar to the STORE) require memory access and their execution cannot be completed in a single clock cycle. However, in the next cycle a new instruction is The following sequence is not correct with such a started by the processor. processor, and a compiler will not generate something like this: LOAD R1,X FI DI CA TR ADD R2,R1 FI DI stall EI LOAD R1,X loads from address X into R1 ADD R2,R1 R2 ← R2 + R1 ADD R4,R3 FI DI EI ADD R4,R3 R4 ← R4 + R3 SUB R2,R4 R2 ← R2 - R4 Two possible solutions: The value in R1 is 1. The hardware should delay the execution of not yet available! the instruction following the LOAD, if this in- LOAD R1,X FI DI CA TR struction needs the loaded value ADD R2,R1 FI DI EI 2. A more efficient, compiler based, solution, which has similarities with the delayed ADD R4,R3 FI DI EI branching, is the delayed load: SUB R2,R4 FI DI EI Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH
• 5.
  Datorarkitektur Fö 5/6 - 17 Datorarkitektur Fö 5/6 - 18 The delayed load problem (cont’d) Are RISCs Really Better than CISCs? The following one is the correct sequence: LOAD R1,X loads from address X into R1 ADD R4,R3 R4 ← R4 + R3 ADD R2,R1 R2 ← R2 + R1 • RISC architectures have several advantages and they were discussed throughout this lecture. SUB R2,R4 R2 ← R2 - R4 However, a definitive answer to the above question is difficult to give. LOAD R1,X FI DI CA TR • A lot of performance comparisons have shown that ADD R4,R3 FI DI EI benchmark programs are really running faster on RISC processors than on processors with CISC ADD R2,R1 FI DI EI characteristics. SUB R2,R4 FI DI EI • However, it is difficult to identify which feature of a processor produces the higher performance. Some "CISC fans" argue that the higher speed is not For the following sequence the compiler has generated a produced by the typical RISC features but because NOP after the LOAD because there is no instruction to of technology, better compilers, etc. fill the load-delay slot: • An argument in favour of the CISC: the simpler LOAD R1,X loads from address X into R1 instruction set of RISC processors results in a NOP larger memory requirement compared to the similar ADD R2,R1 R2 ← R2 + R1 program compiled for a CISC architecture. ADD R4,R2 R4 ← R4 + R2 STORE R4,X stores from R4 to address X Most of the current processors are not typical RISCs LOAD R1,X FI DI CA TR or CISCs but try to combine advantages of both approaches NOP ADD R2,R1 FI DI EI ADD R4,R2 FI DI EI STORE R4,X FI DI CA TR Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH Datorarkitektur Fö 5/6 - 19 Datorarkitektur Fö 5/6 - 20 Some Processor Examples Summary CISC Architectures: VAX 11/780 Nr. of instructions: 303 • Both RISCs and CISCs try to solve the same prob- Instruction size: 2 - 57 lem: to cover the semantic gap. They do it in differ- Instruction format: not fixed ent ways. CISCs are going the traditional way of implementing more and more complex instructions. Addressing modes: 22 RISCs try to simplify the instruction set. Number of general purpose registers: 16 • Innovations in RISC architectures are based on a Pentium close analysis of a large set of widely used Nr. of instructions: 235 programs. Instruction size: 1 - 11 • The main features of RISC architectures are: Instruction format: not fixed reduced number of simple instructions, few Addressing modes: 11 addressing modes, load-store architecture, Number of general purpose registers: 8 instructions are of fixed length and format, a large number of registers is available. • One of the main concerns of RISC designers was RISC Architectures: to maximise the efficiency of pipelining. Sun SPARC • Present architectures often include both RISC and Nr. of instructions: 52 CISC features. Instruction size: 4 Instruction format: fixed Addressing modes: 2 Number of general purpose registers: up to 520 PowerPC Nr. of instructions: 206 Instruction size: 4 Instruction format: not fixed (but small differences) Addressing modes: 2 Number of general purpose registers: 32 Petru Eles, IDA, LiTH Petru Eles, IDA, LiTH