root/trunk/blade/PerfTest.d

/**
 * Blade Performance tests
 */
module PerfTest;
import Blade.Performance.Perform;
import std.stdio;
//import std.cpuid;
import win32.core;
import std.string;


/* Values for agner

LDDQU xmm, mm128.  uops fused, latency, uops unfused (port unknown)
HADDPS xmm, xmm (same as HSUBPS)
HADDPD xmm, xmm. uops fused, uops unfused p01.

PMOVMSKB
*/

double X[1000];
double Y[2000];
double Z[5000];
double da = 3.5;
double SignBit[2] = [0,0];

void daxpyFunc(int REPS)
{
    asm {
    push ESI;
    push EDI;
    push EBX;
    mov EAX, REPS; //-128*8;
    neg EAX;

    lea ESI, X+8*128;
    lea EDI, Y+8*128+256;
//  lea ebx, Z+8*50000;
    movsd XMM2, da;
    xorpd XMM2, SignBit;
    shufpd XMM2, XMM2, 0;
    align 16;
L1:
//  movapd xmm3, [edi+eax];
//  cmp esi, 0; 
    movapd XMM1, [ESI+EAX];
    mulpd XMM1, XMM2;
//  addpd xmm3, [ebx+eax];
//  addpd xmm1, xmm3;
//  add edi, 0;
    add EAX, 16;
    addpd XMM1, [EDI+EAX-16];
    movapd [EDI+EAX-16], XMM1;
    js L1;
    pop EBX;
    pop EDI;
    pop ESI;
    }
}

void alignedmain()
{  
    void init() { X[] = 0; Y[]=0; Z[]=0; }
    void loop() {
        daxpyFunc(200*8);
    }
    void loop2() {
        daxpyFunc(100*8);
    }
    scope PerformanceTester perform = new PerformanceTester;
    perform.DoPerformanceTest(&loop, &init, &loop2);
    perform.describe();    
}


// Align the stack to a multiple of 64 bytes (this improves repeatability).
void main()
{
    asm {
        naked;
        mov EBP, ESP;
        and ESP, 0xFFFF_FFC0;        
        call alignedmain;
        mov ESP, EBP;
        ret;
    }
}