hschumann2/TempleOS-Source-Code

1 2/*The moral of this story is simple3inst level optimizations4don't matter much on a modern Intel CPU5because they convert complex insts6to a stream of RISC insts.7 8I learned this the hard way when I thought9I was greatly improving my compiler by10cutting code by a third.  No significant11speed-up.  Depressing.12*/13 14#define SAMPLES (8*10000000+1)15 16asm {17 18LIMIT:: DU64    SAMPLES;        //Memory reference should be bad, right?19 20_BADLY_UNOPTIMIZED::21        MOV     RAX,022        MOV     RCX,123@@05:   MOV     RDX,RCX24        INC     RCX             //if no dependencies, Free!25        ADD     RAX,RDX26        MOV     RDX,LIMIT-16    //added 16 displacement to make it worse27        CMP     RCX,U64 16[RDX]28        JB      @@0529        RET30 31_WELL_OPTIMIZED1::32        XOR     RAX,RAX33        MOV     RCX,SAMPLES-134@@05:   ADD     RAX,RCX35        DEC     RCX36        JNZ     @@0537        RET38 39_WELL_OPTIMIZED2:: //Unrolled40        XOR     RAX,RAX41        MOV     RCX,SAMPLES-142@@05:   ADD     RAX,RCX43        DEC     RCX44        ADD     RAX,RCX45        DEC     RCX46        ADD     RAX,RCX47        DEC     RCX48        ADD     RAX,RCX49        DEC     RCX50        ADD     RAX,RCX51        DEC     RCX52        ADD     RAX,RCX53        DEC     RCX54        ADD     RAX,RCX55        DEC     RCX56        ADD     RAX,RCX57        DEC     RCX58        JNZ     @@0559        RET60 61_WELL_OPTIMIZED3::62        XOR     RAX,RAX63        MOV     RCX,SAMPLES-164@@05:   ADD     RAX,RCX65        LOOP    @@05    //Inst has slow speed, but saves code size.66        RET67}68 69_extern _BADLY_UNOPTIMIZED I64 Loop1();70_extern _WELL_OPTIMIZED1   I64 Loop2();71_extern _WELL_OPTIMIZED2   I64 Loop3();72_extern _WELL_OPTIMIZED3   I64 Loop4();73 74I64 i;75F64 t0;76 77CPURep;78 79"Bad Code\n";80t0=tS;81i=Loop1;82"Res:%d Time:%9.6f\n",i,tS-t0;83 84"Good Code #1\n";85t0=tS;86i=Loop2;87"Res:%d Time:%9.6f\n",i,tS-t0;88 89"Good Code #2\n";90t0=tS;91i=Loop3;92"Res:%d Time:%9.6f\n",i,tS-t0;93 94"Good Code #3\n";95t0=tS;96i=Loop4;97"Res:%d Time:%9.6f\n",i,tS-t0;98 99/*  Program Output1008 Cores 2.660GHz101Bad Code102Res:3200000040000000 Time: 0.069966103Good Code #1104Res:3200000040000000 Time: 0.062567105Good Code #2106Res:3200000040000000 Time: 0.062907107Good Code #3108Res:3200000040000000 Time: 0.156359109*/110