retrospective ;;Quote: except for spilling, graph coloring register allocation did well; avoided spilling by doubling hardware registers to 32

retrospective ;;Quote: optimize addressing and housekeeping computations through the intermediate language; necessary for efficiency

retrospective ;;Quote: PL.8 not standardized despite being a fully-checked language; enforcement of language rules cost less than 10%

39 ;;Quote: 32-bit, register-based intermediate language for PL.8; similar to target machines

39+;;Quote: PL.8 used intermediate language for optimization

40 ;;Quote: PL.8 intermediate language uses a status register for compares and conditional jumps; no source language control structures

40 ;;Quote: PL.8 intermediate language has string operations and symbolic registers; optimized

40 ;;Quote: generate intermediate code bottom up, independent of context

40 ;;Quote: replicate loop test at head and tail of loop; optimization eliminates the first test if constant operands

41 ;;Quote: all PL.8 data accessed from the stack frame; may be optimized to a register

41 ;;Quote: value numbering allows common subexpression elimination over assignments within an extended basic block

41 ;;Quote: dead store elimination eliminates stored values that are not used

41 ;;Quote: combine basic blocks by eliminating unnecessary branches; called straightening

41 ;;Quote: trap elimination moves bounds tests from loops by modifying the loop termination test

41 ;;Quote: consolidate constant values for a loop by rearranging expressions within loops; called reassociation

42 ;;Quote: PL.8 optimizer bugs produce catastrophic results since optimizer applies to all code; regression test of 150 self-checking programs

42 ;;Quote: instead of moving code out of a loop, insert a duplicate copy; common subexpression elimination to remove the original copy

42+;;Quote: simplify an optimization by depending on common subexpression elimination and other standard optimizations

42 ;;Quote: optimize complex operations, then expand the operation and reoptimize; e.g., max and min, boolean expressions, and function calls

43 ;;Quote: graph coloring allocator developed for PL.8; simplified the compiler

45 ;;Quote: PL.8 compiler was written in PL.8

45+;;Quote: 32 registers eliminated most spill code from PL.8; with sixteen registers about half of the routines used spill code

43 ;;Quote: instruction scheduling moves loads away from stores; conditions from branch; etc.

43+;;Quote: perform instruction scheduling before and after register allocation

43 ;;Quote: generate machine code after register allocation; table-driven, destructive operations, addressability, external references

44 ;;Quote: PL.8 generated type metadata for every external variable; used for link edit and load-time type checking

44 ;;Quote: condition codes difficult to optimize; use values instead and generate condition codes at final assembly; uses peephole optimization

44 ;;Quote: final assembly generates function prologs and epilogs; handles stack frame, preserved registers

44 ;;Quote: PL.8 guarantees that all array and offset references are within bounds; optimizes run-time checks; checking costs 5-10%

45 ;;Quote: self-test of compiler by fixed points; compile the compiler, then recompile; the two results should be identical; also recompile the runtime library for each machine

45+;;Quote: PL.8 was highly portable; most bugs occurred on all machines