![Control Hazards (4)
Reducing the stall from branch hazards by moving the zero test and branch calculation into ID
phase of pipeline. It uses a separate adder to compute the branch target address during ID.
Because the branch target addition happens during ID, it will happen for all instructions. The
branch condition (Regs[IF/ID.IR6…10] op 0) will also be done for all instructions. The selection
of the sequential PC or the branch target PC will still occur during IF, but now it uses values
from ID phase, rather than from EX/MEM register. In this case, the branch instruction is done by
the end of ID phase, so EX, MEM and WB stages are not used for branch instructions anymore.](https://image.slidesharecdn.com/ct213processordesignpipelinehazard-130317101303-phpapp01/85/Ct213-processor-design_pipelinehazard-24-320.jpg)
![Modified Pipelined Instruction Fetch
• Instruction Fetch
– IF/ID.IR mem[PC]
– IF/ID.NPC, PC if (Regs[IF/ID. IR6…10] op 0)
{IF/ID.NPC +(IF/ID.IR16)16##IF/ID.IR16…31}else{PC+4}
• Operation:
– send out the PC and fetch the instruction from memory
– Increment the PC by 4 to address the next instruction or
save the address generated by a taken branch of a
previous instruction in decode stage](https://image.slidesharecdn.com/ct213processordesignpipelinehazard-130317101303-phpapp01/85/Ct213-processor-design_pipelinehazard-25-320.jpg)
![Modified Pipelined Instruction Decode
• Instruction Decode Cycle/Register Fetch
– ID/EX.A Regs[IR6…10]; ID/EX.B Regs[IR11…15]
– ID/EX.IR IF/EX.IR
– ID/EX.Imm (IF/ID.IR16)16##IF/ID.IR16…31
– Compute the condition: Regs[IF/ID.IR6..10] op 0
– Compute the branch address: IF/ID.NPC + (IF/ID.IR16)16##IF/ID.IR16…31
• Operation
– Decode the instruction and access the register files to access the registers; the
output of the general purpose registers are read into two temporary register (A
and B, part of the pipeline registers ID/EX stage) for use in latter clock cycles
– The lower 16 bits of IR, stored in pipeline registers from IF/ID stage are also
sign extended and stored into temporary register Imm (part of ID/EX pipeline
registers), for latter use
– Value IR is passed to the next stage of pipeline registers (from IF/ID to ID/EX)
– Compute the values for the cond and branch target and use them to set the PC if
necessary (if taken branch)](https://image.slidesharecdn.com/ct213processordesignpipelinehazard-130317101303-phpapp01/85/Ct213-processor-design_pipelinehazard-26-320.jpg)
Ct213 processor design_pipelinehazard
- 1. Content • Introduction to pipeline hazard • Structural Hazard • Data Hazard • Control Hazard
- 2. Pipeline Hazards (1) • Pipeline Hazards are situations that prevent the next instruction in the instruction stream from executing in its designated clock cycle • Hazards reduce the performance from the ideal speedup gained by pipelining • Three types of hazards – Structural hazards • Arise from resource conflicts when the hardware can’t support all possible combinations of overlapping instructions – Data hazards • Arise when an instruction depends on the results of a previous instruction in a way that is exposed by overlapping of instruction in pipeline – Control hazards • Arise from the pipelining of branches and other instructions that change the PC (Program Counter)
- 3. Pipeline Hazards (2) • Hazards in pipeline can make the pipeline to stall • Eliminating a hazard often requires that some instructions in the pipeline to be allowed to proceed while others are delayed – When an instruction is stalled, instructions issued latter than the stalled instruction are stopped, while the ones issued earlier must continue • No new instructions are fetched during the stall
- 4. Structural Hazards (1) • If certain combination of instructions can’t be accommodated because of resource conflicts, the machine is said to have a structural hazard • It can be generated by: – Some functional unit is not fully pipelined – Some resources has not been duplicated enough to allow all the combinations in the pipeline to execute – For example: a machine may have only one register file write port, but under certain conditions, the pipeline might want to perform two writes in one clock cycle – this will generate structural hazard • When a sequence of instructions encounter this hazard, the pipeline will stall one of the instructions until the required unit is available • Such stalls will increase the Clock cycle Per Instruction from its ideal 1 for pipelined machines
- 5. Structural Hazards (2) • Consider a Von Neumann architecture (same memory for instructions and data)
- 6. Structural Hazards (3) • Stall cycle added (commonly called pipeline bubble)
- 7. Structural Hazards (4) Instruction Clock number Number 1 2 3 4 5 6 7 8 9 10 load IF ID EX MEM WB Instruction i+1 IF ID EX MEM WB Instruction i+2 IF ID EX MEM WB Instruction i+3 stall IF ID EX MEM WB Instruction i+4 IF ID EX MEM WB Instruction i+5 IF ID EX MEM • Another way to represent the stall – no instruction is initiated in clock cycle 4
- 8. Structural Hazards (5) • A machine with structural hazard will have lower CPI • Why a designer allows structural hazard? – To reduce cost • Pipelining all the functional units or duplicating them may be too costly – To reduce latency • Introducing too many pipeline stages may cause latency issues
- 9. Data Hazards (1) • Data hazards occur when the pipeline changes the order of read/write accesses to operands so that the order differs from the order seen by sequentially executing instructions on an un-pipelined machine • Consider the execution of following instructions, on our pipelined example processor: – ADD R1, R2, R3 – SUB R4, R1, R5 – AND R6, R1, R7 – OR R8, R1, R9 – XOR R10, R1, R11
- 10. Data Hazards (2) • The use of results from ADD instruction causes hazard since the register is not written until after those instructions read it.
- 11. Data Hazards (3) • Eliminate the stalls for the hazard involving SUB and AND instructions using a technique called forwarding
- 12. Data Hazards (4) • Store requires an operand during MEM and forwarding is shown here. – The result of the load is forwarded from the output in MEM/WB to the memory input to be stored – In addition the ALUOutput is forwarded to ALU input for address calculation for both Load and Store
- 13. Data Hazards Classification • Depending on the order of read and write access in the instructions, data hazards could be classified as three types. • Consider two instructions i and j, with i occurring before j. Possible data hazards: – RAW (Read After Write) • j tries to read a source before i writes to it , so j incorrectly gets the old value; • most common type of hazard, that is what we tried to explain so far. – WAW (Write After Write) • j tries to write an operand before is written by i. The write ends up being performed in wrong order, having i overwrite the operand written by j, the destination containing the operand written by i rather than the one written by j • Present in pipelines that write in more than one pipe stage – WAR (Write After Read) • j tries to write a destination before it is read by i, so the instruction i incorrectly gets the new value • This doesn’t happen in our example, since all reads are early and writes late
- 14. Data Hazards Requiring Stalls (1) • Unfortunately not all data hazards can be handled by forwarding. Consider the following sequence: – LW R1, 0(R2) – SUB R4, R1, R5 – AND R6, R1, R7 – OR R8, R1, R9 • The problem with this sequence is that the Load operation will not have data until the end of MEM stage.
- 15. Data Hazards Requiring Stalls (2) • The load instruction can forward the results to AND and OR instruction, but not to the SUB instruction since that would mean forwarding results in “negative” time
- 16. Data Hazards Requiring Stalls (3) • The load interlock causes a stall to be inserted at clock cycle 4, delaying the SUB instruction and those that follow by one cycle. – This delay allows the value to be successfully forwarded onto the next clock cycle
- 17. Data Hazards Requiring Stalls (4) LW R1, 0(R2) IF ID EX MEM WB SUB R4, R1, R5 IF ID EX MEM WB AND R6, R1, R7 IF ID EX MEM WB OR R8, R1, R9 IF ID EX MEM WB • Before stall insertion LW R1, 0(R2) IF ID EX MEM WB SUB R4, R1, R5 IF ID stall EX MEM WB AND R6, R1, R7 IF stall ID EX MEM WB OR R8, R1, R9 stall IF ID EX MEM WB • After stall insertion
- 18. Compiler Scheduling for Data Hazards (1) • Consider a typical code, such as A = B+C LW R1, B IF ID EX MEM WB LW R2, C IF ID EX MEM WB ADD R3, R1, R2 IF ID stall EX MEM WB SW A, R3 IF stall ID EX MEM WB • The ADD instruction must be stalled to allow the load of C to complete • The SW needs not be delayed because the forwarding hardware passes the result from MEM/WB directly to the data memory input for storing
- 19. Compiler Scheduling for Data Hazards (2) • Rather than just allow the pipeline to stall, the compiler could try to schedule the pipeline to avoid the stalls, by rearranging the code – The compiler could try to avoid the generating the code with a load followed by an immediate use of the load destination register – This technique is called pipeline scheduling or instruction scheduling and it is a very used technique in modern compilers
- 20. Instruction scheduling example • Generate code for our example processor that avoids pipeline stalls from the following sequence: – A = B +C – D=E-F • Solution – LW Rb, B – LW Rc, C – LW Re, E ; swap instructions to avoid stall – ADD Ra, Rb, Rc – LW Rf, f – SW a, Ra ; swap instruction to avoid stall – SUB Rd, Re, Rf – SW d, Rd
- 21. Control Hazards (1) • Can cause a greater performance loss than the data hazards • When a branch is executed it may or it may not change the PC (to other value than its value + 4) – If a branch is changing the PC to its target address, than it is a taken branch – If a branch doesn’t change the PC to its target address, than it is a not taken branch • If instruction i is a taken branch, than the value of PC will not change until the end MEM stage of the instruction execution in the pipeline – A simple method to deal with branches is to stall the pipe as soon as we detect a branch until we know the result of the branch
- 22. Control Hazards (2) Branch Instruction IF ID EX MEM WB Branch Successor IF stall stall IF ID EX MEM WB Branch Successor +1 IF ID EX MEM WB Branch Successor +2 IF ID EX MEM • A branch causes three cycle stall in our example processor pipeline – One cycle is a repeated IF – necessary if the branch would be taken. If the branch is not taken, this IF is redundant – Two idle cycles
- 23. Control Hazards (3) • The three clock cycles lost for every branch is a significant loss – With a 30% branch frequency, the machine with branch stalls achieves only about half of the speedup from pipelining – Reducing the branch penalty becomes critical • The number of clock cycles in a branch stall can be reduced by two steps: – Find out if the branch is taken or not in early stage in the pipeline – Compute the taken PC (address of the branch target) earlier
- 24. Control Hazards (4) Reducing the stall from branch hazards by moving the zero test and branch calculation into ID phase of pipeline. It uses a separate adder to compute the branch target address during ID. Because the branch target addition happens during ID, it will happen for all instructions. The branch condition (Regs[IF/ID.IR6…10] op 0) will also be done for all instructions. The selection of the sequential PC or the branch target PC will still occur during IF, but now it uses values from ID phase, rather than from EX/MEM register. In this case, the branch instruction is done by the end of ID phase, so EX, MEM and WB stages are not used for branch instructions anymore.
- 25. Modified Pipelined Instruction Fetch • Instruction Fetch – IF/ID.IR mem[PC] – IF/ID.NPC, PC if (Regs[IF/ID. IR6…10] op 0) {IF/ID.NPC +(IF/ID.IR16)16##IF/ID.IR16…31}else{PC+4} • Operation: – send out the PC and fetch the instruction from memory – Increment the PC by 4 to address the next instruction or save the address generated by a taken branch of a previous instruction in decode stage
- 26. Modified Pipelined Instruction Decode • Instruction Decode Cycle/Register Fetch – ID/EX.A Regs[IR6…10]; ID/EX.B Regs[IR11…15] – ID/EX.IR IF/EX.IR – ID/EX.Imm (IF/ID.IR16)16##IF/ID.IR16…31 – Compute the condition: Regs[IF/ID.IR6..10] op 0 – Compute the branch address: IF/ID.NPC + (IF/ID.IR16)16##IF/ID.IR16…31 • Operation – Decode the instruction and access the register files to access the registers; the output of the general purpose registers are read into two temporary register (A and B, part of the pipeline registers ID/EX stage) for use in latter clock cycles – The lower 16 bits of IR, stored in pipeline registers from IF/ID stage are also sign extended and stored into temporary register Imm (part of ID/EX pipeline registers), for latter use – Value IR is passed to the next stage of pipeline registers (from IF/ID to ID/EX) – Compute the values for the cond and branch target and use them to set the PC if necessary (if taken branch)
- 27. References • “Computer Architecture – A Quantitative Approach”, John L Hennessy & David A Patterson, ISBN 1-55860-329-8 • “Computer Architecture”, Nicholas Charter, ISBN – 0-07-136207
Editor's Notes
- #6: As a result, when an instruction will perform a data reference, will conflict with an instruction fetch. In this example, the load instruction wants to access the memory to load data at the same time when instruction 3 wants to fetch an instruction from memory.
- #7: To solve the problem, a stall cycle is added. The effect of the pipeline bubble is actually to occupy the resources for that instruction slot as it travels through the pipeline. Performance wise, instruction 3 will not complete during clock cycle 8, but during clock cycle 9.
- #8: Sometime those diagrams are drawn with a stall occupying a whole raw, with instruction 3 being moved to the next raw. In either case, the effect is the same. The instruction 3 is not beginning execution until cycle 5.
- #10: All the instructions after ADD use the result from ADD.
- #11: The ADD instruction writes the result in register R1 only at the WB stage, but SUB instruction reads the value during its ID stage. This is what is called a data hazard. Unless precautions are taken, the SUB instruction will read the wrong value and will use it… The AND instruction is also affected by this hazard. As we can see from the figure, the write of R1 doesn’t complete until the end of clock cycle 5. Thus, the AND instruction that reads the registers in clock cycle 4 will receive the wrong results. XOR instruction operates correctly, it reads its inputs (in clock cycle 6) after the ADD has written its result (in clock cycle 5). OR instruction can also be made to work without incurring an hazard, using a simple implementation technique. The technique is to perform the register file reads in the second half of the clock cycle and the writes in the first half.
- #12: The data hazard, in certain circumstances can be solved using an implementation technique called forwarding. The idea behind the forwarding is that the result produced by ADD is not really needed by the SUB instruction until it is actually produced. If the result can be moved from where the ADD instruction produces it , the EX/MEM register, to where the SUB needs it, the ALU input latches, then the need for a stall can be avoided. Forwarding works as follow: The ALU result from EX/MEM register is always fed back to the ALU input latches If the forwarding hardware detects that the previous ALU operation has written the register corresponding to a source for the current ALU operation, control logic selects the forwarded result as the ALU input, rather than the value read from the register file. We need to forward results not only from immediately previous instruction, but possible from instructions that started two or three cycles earlier.
- #24: To optimize the branch behavior, both of the steps should be taken.
- #25: It uses a separate adder to compute the branch target address during ID. Because the branch target addition happens during ID, it will happen for all instructions. The branch condition (Regs[IF/ID.IR6…10] op 0) will also be done for all instructions. The selection of the sequential PC or the branch target PC will still occur during IF, but now it uses values from ID phase, rather than from EX/MEM register. In this case, the branch instruction is done by the end of ID phase, so EX, MEM and WB stages are not used for branch instructions anymore.