xbyak profile with vtune

mytest.cpp

/*
Description:
Code2 is consistantly faster than Code because SSE2 instructions (like movdqu)
is not cooperating well with AVX registers. use AVX (vmovdqu) as much as possible.

------------------------------------------------------------------------------
Install vtune & make a symbolic link
    /opt/intel/vtune_amplifier -> /home/xxx/intel/oneapi/vtune/2021.5.0/

How to compile: add following rules to Makefile:
    mytest: mytest.cpp ../xbyak/xbyak_util.h
	    $(CXX) $(CFLAGS) mytest.cpp -o $@ -DXBYAK_USE_VTUNE -I /opt/intel/vtune_amplifier/include/ -L /opt/intel/vtune_amplifier/lib64 -ljitprofiling -ldl

How to profile JIT-code with perf or VTune
    $ source /opt/intel/vtune_amplifier/env/vars.sh
    $ vtune-gui

    application           :  mytest
    application parameters: 2

    Hotspots
*/

#include <chrono>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <xbyak/xbyak_util.h>


struct Code : public Xbyak::CodeGenerator {
  Code() {
    sub(rsp, 32 * 3);

    mov(rax, 200123 * 200);
    L(".sss");

    movdqu(ptr[rsp + 0], xmm5);
    movdqu(ptr[rsp + 16], xmm6);
    movdqu(ptr[rsp + 16 * 2], xmm7);

    vpxor(ymm0, ymm0, ymm0);
    vpxor(ymm1, ymm1, ymm1);
    vpxor(ymm2, ymm2, ymm2);

    sub(eax, 1);
    jnz(".sss");

    add(rsp, 32 * 3);
    mov(eax, 1);
    ret();
  }
};

struct Code2 : public Xbyak::CodeGenerator {
  Code2() {
    sub(rsp, 32 * 3);

    mov(rax, 200123 * 200);
    L(".sss");

    vmovdqu(ptr[rsp + 0], xmm5);
    vmovdqu(ptr[rsp + 32], xmm6);
    vmovdqu(ptr[rsp + 32 * 2], xmm7);

    vpxor(ymm5, ymm5, ymm5);
    vpxor(ymm6, ymm6, ymm6);
    vpxor(ymm7, ymm7, ymm7);
    /*
        vpxor(ymm0, ymm0, ymm0);
        vpxor(ymm1, ymm1, ymm1);
        vpxor(ymm2, ymm2, ymm2);
    */
    sub(eax, 1);
    jnz(".sss");

    add(rsp, 32 * 3);
    mov(eax, 1);
    ret();
  }
};

template <typename F> float dura(const char *name, F func) {
  auto t0 = std::chrono::high_resolution_clock::now();
  func();
  auto t1 = std::chrono::high_resolution_clock::now();
  std::chrono::duration<float> fsec = t1 - t0;
  float ret = fsec.count() * 1000;
  if (name)
    std::cout << name << " took " << ret << "ms\n";
  return ret;
}

int main(int argc, char *argv[]) {
  int mode = argc == 1 ? 0 : atoi(argv[1]);
  Code c;
  Code2 c2;
  int (*f)() = (int (*)())c.getCode();
  int (*g)() = (int (*)())c2.getCode();

  printf("f:%p, %d\n", (const void *)f, (int)c.getSize());
  printf("g:%p, %d\n", (const void *)g, (int)c2.getSize());

  Xbyak::util::Profiler prof;
  printf("mode=%d\n", mode);
  prof.init(mode);
  prof.set("f", (const void *)f, c.getSize());
  prof.set("g", (const void *)g, c2.getSize());

  auto ft = dura("f", f);
  auto gt = dura("g", g);

  printf("g is faster than f by %.1f%%\n", 100*((1./gt) - (1./ft))/(1./ft) );
  puts("end");
}