Instruction Pipelining

The Instruction Cycle INSTRUCTION PIPELINING (I) Fetch instruction FI Decode DI 1. The Instruction Cycle 2. Instruction Pipelining 3. Pipeline Hazards Fetch operand 4. Structural Hazards - Calculate operand address (CO) - Fetch operand (FO) 5. Data Hazards Execute instruction 6. Control Hazards Instruction Pipelining • Acceleration by Pipelining Instruction execution is extremely complex and involves several operations which are executed successively (see slide 2). This implies a large amount of hardware, but only one part of this hardware works at a given moment. Two stage pipeline: • • Pipelining is an implementation technique whereby multiple instructions are overlapped in execution. This is solved without additional hardware but only by letting different parts of the hardware work for different instructions at the same time. The pipeline organization of a CPU is similar to an assembly line: the work to be done in an instruction is broken into smaller steps (pieces), each of which takes a fraction of the time needed to complete the entire instruction. Each of these steps is a pipe stage (or a pipe segment). Pipe stages are connected to form a pipe: Stage 1 • Stage 2 Stage n The time required for moving an instruction from one stage to the next: a machine cycle (often this is one clock cycle). The execution of one instruction takes several machine cycles as it passes through the pipeline. FI: fetch instruction EI: execute instruction Clock cycle → 1 2 3 4 5 6 7 8 Instr. i FI EI FI EI Instr. i+1 • - Execute instruction (EI) - Write back operand (WO) Instr. i+2 Instr. i+3 Instr. i+4 Instr. i+5 Instr. i+6 FI EI FI EI FI EI FI EI FI EI We consider that each instruction takes execution time Tex. Execution time for the 7 instructions, with pipelining: (Tex/2)*8= 4*Tex Acceleration by Pipelining (cont’d) Acceleration by Pipelining (cont’d) Six stage pipeline (see also slide 2): FI: fetch instruction FO: fetch operand DI: decode instruction EI: execute instruction CO: calculate operand address WO:write operand Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 Instr. i FI DI COFO EI WO Instr. i+1 Instr. i+2 Instr. i+3 • FI DI COFO EI WO FI DI COFO EI WO Apparently a greater number of stages always provides better performance. However: - a greater number of stages increases the overhead in moving information between stages and synchronization between stages. - with the number of stages the complexity of the CPU grows. - it is difficult to keep a large pipeline at maximum rate because of pipeline hazards. FI DI COFO EI WO Instr. i+4 FI DI COFO EI WO Instr. i+5 80486 and Pentium: FI DI COFO EI WO Instr. i+6 FI DI COFO EI WO PowerPC: five-stage pipeline for integer instr. eight-stage pipeline for FP instr. four-stage pipeline for integer instr. six-stage pipeline for FP instr. Execution time for the 7 instructions, with pipelining: (Tex/6)*12= 2*Tex • After a certain time (N-1 cycles) all the N stages of the pipeline are working: the pipeline is filled. Now, theoretically, the pipeline works providing maximal parallelism (N instructions are active simultaneously). Pipeline Hazards Structural Hazards • • • Pipeline hazards are situations that prevent the next instruction in the instruction stream from executing during its designated clock cycle. The instruction is said to be stalled. When an instruction is stalled, all instructions later in the pipeline than the stalled instruction are also stalled. Instructions earlier than the stalled one can continue. No new instructions are fetched during the stall. Types of hazards: 1. Structural hazards 2. Data hazards 3. Control hazards Structural hazards occur when a certain resource (memory, functional unit) is requested by more than one instruction at the same time. Instruction ADD R4,X fetches in the FO stage operand X from memory. The memory doesn’t accept another access during that cycle. Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 ADD R4,X FI DI COFO EI WO Instr. i+1 Instr. i+2 Instr. i+3 Instr. i+4 FI DI COFO EI WO FI DI COFO EI WO stall FI DI COFO EI WO FI DI COFO EI WO Penalty: 1 cycle • Certain resources are duplicated in order to avoid structural hazards. Functional units (ALU, FP unit) can be pipelined themselves in order to support several instructions at a time. A classical way to avoid hazards at memory access is by providing separate data and instruction caches. Data Hazards Data Hazards (cont’d) • • We have two instructions, I1 and I2. In a pipeline the execution of I2 can start before I1 has terminated. If in a certain stage of the pipeline, I2 needs the result produced by I1, but this result has not yet been generated, we have a data hazard. I1: I2: MUL ADD MUX R2 ← R2 * R3 R1 ← R1 + R2 R2,R3 R1,R2 MUX bypass path bypass path ALU Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 MUL R2,R3 FI DI COFO EI WO ADD R1,R2 Some of the penalty produced by data hazards can be avoided using a technique called forwarding (bypassing). from reg- from register or ister or memory memory to register or memory FI DI CO stall stall FO EI WO Instr. i+2 FI DI COFO EI WO Before executing its FO stage, the ADD instruction is stalled until the MUL instruction has written the result into R2. Penalty: 2 cycles • The ALU result is always fed back to the ALU input. If the hardware detects that the value needed for the current operation is the one produced by the previous operation (but which has not yet been written back) it selects the forwarded result as the ALU input, instead of the value read from register or memory. Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 MUL R2,R3 FI DI COFO EI WO ADD R1,R2 FI DI CO stall FO EI WO After the EI stage of the MUL instruction the result is available by forwarding. The penalty is reduced to one cycle. Control Hazards Control Hazards (cont’d) Conditional branch • Control hazards are produced by branch instructions. Unconditional branch -------------BR TARGET -------------TARGET -------------After the FO stage of the branch instruction the address of the target is known and it can be fetched 1 2 3 4 5 6 7 8 9 10 11 12 BR TARGET FI DI COFO EI WO FI target+1 stall stall Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 ADD R1,R2 FI DI COFO EI WO target FI DI COFO EI WO FI stall stall FI DI COFO EI WO Penalty: 3 cycles FI DI COFO EI WO FI DI COFO EI WO The instruction following the branch is fetched; before the DI is finished it is not known that a branch is executed. Later the fetched instruction is discarded Penalty: 3 cycles R1 ← R1 + R2 branch if zero At this moment, both the condition (set by ADD) and the target address are known. BEZ TARGET Clock cycle → target TARGET Branch is taken ADD R1,R2 BEZ TARGET instruction i+1 ------------------------- Branch not taken At this moment the condition is known and instr+1 can go on. Clock cycle → 1 2 3 4 5 6 7 8 9 10 11 12 ADD R1,R2 FI DI COFO EI WO BEZ TARGET instr i+1 Penalty: 2 cycles FI DI COFO EI WO FI stall stall DI COFO EI WO Summary Control Hazards (cont’d) • With conditional branch we have a penalty even if the branch has not been taken. This is because we have to wait until the branch condition is available. • Branch instructions represent a major problem in assuring an optimal flow through the pipeline. Several approaches have been taken for reducing branch penalties (see slides of the following lecture). • Instructions are executed by the CPU as a sequence of steps. Instruction execution can be substantially accelerated by instruction pipelining. • A pipeline is organized as a succession of N stages. At a certain moment N instructions can be active inside the pipeline. • Keeping a pipeline at its maximal rate is prevented by pipeline hazards. Structural hazards are due to resource conflicts. Data hazards are produced by data dependencies between instructions. Control hazards are produced as consequence of branch instructions