chapter3-m5-ziavras

In the loop twice ld muld from the 1st iteration have

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 6 Execute F10 6 Yes ADD.D F6,F8,F2 Write result lt S. Ziavras DestiValue nation F6 Mem[34+Regs [R2]] F2 Mem[45+Regs [R3]] F0 #2xRegs[F4] F8 #1- #2 F10 F6 #4 + #2 FP Register Status Field Reorder # Busy F0 F1 F2 F3 F4 3 Yes F5 F6 F7 6 No No No No S. Ziavras No Yes F8 F10 4 … 5 Yes Yes Example Consider the code example used earlier for Tomasulo’s algorithm in execution Loop: L.D MUL.D S.D DADDIU BNE F0,0(R1) F4,F0,F2 F4,0(R1) R1,R1,#-8 R1,R2,Loop ;branches if R1≠R2 if R1 Assume: we have issued all the instrs. in the loop twice, & L.D & MUL.D from the 1st iteration have committed & all other instrs. have completed execution Normally, the store would wait in the ROB for both the EA operand (R1 here) & the value (F4 here). Since we are only considering the FP pipeline, assume the EA for the store is computed by Sthe time the instr. is issued . Ziavras Reorder Buffer En. Busy Instruction State 1 No L.D F0,0(R1) Commit 2 3 4 5 6 7 8 9 10 No MUL.D Yes S.D F4,F0,F2 Commit F4,0(R1) Write res. (WR) Yes DADDIU R1,R1,#-8 WR Yes BNE R1,R2,Loop WR R1 Yes L.D F0,0(R1) WR Yes MUL.D F4,F0,F2 WR F4 Yes S.D F4,0(R1) WR Yes DADDIU R1,R1,#-8 WR R1,R1,# Yes BNE R1,R2,Loop WR S. Ziavras Dest. F0 Value Mem[0+Regs [R1]] F4 0+Regs [R1] R1 #1xRegs[F2] #2 Regs[R1]-8 F0 F4 0+#4 R1 Mem[#4] #6xRegs[F2] #7 #4-8 FP Register Status Field Reorder # Busy F0 F1 F2 F3 F6 F7 F8 Yes No No No Yes No No … No 6 F4 F5 7 • Only the L.D & MUL.D instrs. have committed, although all the others have completed execution No RSs are busy & none are shown • The remaining instrs. will be committed as fast as possible • The first two reorder buffers are empty, but are shown fi for completeness S. Ziavras...
View Full Document

Ask a homework question - tutors are online