1Thu Jun 26 14:43:04 CDT 2003 2 3Information about BinInterface 4------------------------------ 5 6Take in a set of instructions with some particular register 7allocation. It allows you to add, modify, or delete some instructions, 8in SSA form (kind of like LLVM's MachineInstrs.) Then re-allocate 9registers. It assumes that the transformations you are doing are safe. 10It does not update the mapping information or the LLVM representation 11for the modified trace (so it would not, for instance, support 12multiple optimization passes; passes have to be aware of and update 13manually the mapping information.) 14 15The way you use it is you take the original code and provide it to 16BinInterface; then you do optimizations to it, then you put it in the 17trace cache. 18 19The BinInterface tries to find live-outs for traces so that it can do 20register allocation on just the trace, and stitch the trace back into 21the original code. It has to preserve the live-ins and live-outs when 22it does its register allocation. (On exits from the trace we have 23epilogues that copy live-outs back into the right registers, but 24live-ins have to be in the right registers.) 25 26 27Limitations of BinInterface 28--------------------------- 29 30It does copy insertions for PHIs, which it infers from the machine 31code. The mapping info inserted by LLC is not sufficient to determine 32the PHIs. 33 34It does not handle integer or floating-point condition codes and it 35does not handle floating-point register allocation. 36 37It is not aggressively able to use lots of registers. 38 39There is a problem with alloca: we cannot find our spill space for 40spilling registers, normally allocated on the stack, if the trace 41follows an alloca(). What might be an acceptable solution would be to 42disable trace generation on functions that have variable-sized 43alloca()s. Variable-sized allocas in the trace would also probably 44screw things up. 45 46Because of the FP and alloca limitations, the BinInterface is 47completely disabled right now. 48 49 50Demo 51---- 52 53This is a demo of the Ball & Larus version that does NOT use 2-level 54profiling. 55 561. Compile program with llvm-gcc. 572. Run opt -lowerswitch -paths -emitfuncs on the bytecode. 58 -lowerswitch change switch statements to branches 59 -paths Ball & Larus path-profiling algorithm 60 -emitfuncs emit the table of functions 613. Run llc to generate SPARC assembly code for the result of step 2. 624. Use g++ to link the (instrumented) assembly code. 63 64We use a script to do all this: 65------------------------------------------------------------------------------ 66#!/bin/sh 67llvm-gcc $1.c -o $1 68opt -lowerswitch -paths -emitfuncs $1.bc > $1.run.bc 69llc -f $1.run.bc 70LIBS=$HOME/llvm_sparc/lib/Debug 71GXX=/usr/dcs/software/evaluation/bin/g++ 72$GXX -g -L $LIBS $1.run.s -o $1.run.llc \ 73$LIBS/tracecache.o \ 74$LIBS/mapinfo.o \ 75$LIBS/trigger.o \ 76$LIBS/profpaths.o \ 77$LIBS/bininterface.o \ 78$LIBS/support.o \ 79$LIBS/vmcore.o \ 80$LIBS/transformutils.o \ 81$LIBS/bcreader.o \ 82-lscalaropts -lscalaropts -lanalysis \ 83-lmalloc -lcpc -lm -ldl 84------------------------------------------------------------------------------ 85 865. Run the resulting binary. You will see output from BinInterface 87(described below) intermixed with the output from the program. 88 89 90Output from BinInterface 91------------------------ 92 93BinInterface's debugging code prints out the following stuff in order: 94 951. Initial code provided to BinInterface with original register 96allocation. 97 982. Section 0 is the trace prolog, consisting mainly of live-ins and 99register saves which will be restored in epilogs. 100 1013. Section 1 is the trace itself, in SSA form used by BinInterface, 102along with the PHIs that are inserted. 103PHIs are followed by the copies that implement them. 104Each branch (i.e., out of the trace) is annotated with the 105section number that represents the epilog it branches to. 106 1074. All the other sections starting with Section 2 are trace epilogs. 108Every branch from the trace has to go to some epilog. 109 1105. After the last section is the register allocation output. 111