Разбор перформансных задач с JBreak (часть 4)28.03.2018 14:18

habr.png

Разбор последней четвёртой задачи: 

    public double octaPow(double a) {
        return Math.pow(a, 8);
    }

    public double octaPow(double a) {
        return a * a * a * a * a * a * a * a;
    }

    public double octaPow(double a) {
        return Math.pow(Math.pow(Math.pow(a, 2), 2), 2);
    }

    public double octaPow(double a) {
        a *= a; a *= a; return a * a;
    }


Условие (упрощённо):

Определить, какие методы быстрые, а какие — медленные (JRE 1.8.0_161).


Под катом бенчмарки, куски ассемблера и разбор оптимизаций со стороны JVM.

Другие публикации серии: Часть 1, Часть 2 и Часть 3.

Комментарий к задаче


Как известно, операции над числами с плавающей точкой имеют дурную славу:

  1. Сложны и зависят от реализации.
  2. Не ассоциативны.
  3. Дают нелогичные результаты.
  4. Сравнение результатов на == в большинстве случаев не имеет смысла.


В связи с этим важно понимать, что предложенные методы могут будут давать разные результаты вычислений, но не в плане производительности, а в арифметическом смысле.

Пара примеров
    public static void main(String[] args) {
        double value = 1e15;
        double delta = 0.0001;
        System.out.println(value + delta == value); // true

        double a = 1.010101;
        double b = 101.0101;
        double c = 10101.01;

        System.out.println((a * b) * c != a * (b * c)); // true
    }


Очевидные неправильные ответы


В этой задаче было четыре варианта алгоритмов, поэтому и потенциальных ответов становится больше:

Все варианты одинаковые, т.к. в Java крутой JIT-компилятор! /* Славный ответ */
Второй или четвёртый вариант самые быстрые, т.к. это простое умножение.


Детальный разбор исследуемых методов


    public double mathOctaPow(double a) {
        return Math.pow(a, 8);
    }

    public double plainOctaPow(double a) {
        return a * a * a * a * a * a * a * a;
    }

    public double trickyMathOctaPow(double a) {
        return Math.pow(Math.pow(Math.pow(a, 2), 2), 2);
    }

    public double trickyPlainOctaPow(double a) {
        a *= a; a *= a; return a * a;
    }


Вывод дизассемблированного кода осуществлён с помощью следующего набора ключей: 
-XX:+UnlockDiagnosticVMOptions
-XX:CompileCommand=print,<класс>.<метод>
-XX:PrintAssemblyOptions=intel


plainOctaPow


Начнём с самого простого случая plainOctaPow. В действительности код
a * a * a * a * a * a * a * a
эквивалентен коду
((((((a * a) * a) * a) * a) * a) * a) * a
ввиду лево-ассоциативности операции умножения.

Содержательно данный код скомпилировался JIT-компилятором (c1) в следующий набор инструкций (в регистре xmm0 находится только значение параметра double a): 

  0x0000000002c96a3e: vmovapd xmm1, xmm0
  0x0000000002c96a42: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a46: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a4a: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a4e: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a52: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a56: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a5a: vmulsd  xmm1, xmm1, xmm0
  0x0000000002c96a5e: vmovapd xmm0, xmm1


Краткая памятка по инструкциям
vmovapd xmm1, xmm2 — поместить выровненные числа двойной точности с плавающей точкой (double-precision float-point далее везде будут называться double) из регистра xmm2 в регистр xmm1. Поскольку размер XMM регистров равен 128bit, то можно за один раз взять до двух double. Данная инструкция поддерживает регистры YMM и ZMM, размер которых 256bit и 512bit, соответственно.

vmulsd xmm1, xmm2, xmm3 — перемножить double-значения из регистра xmm2 и xmm3, а результат разместить в регистре xmm1. Аналогично предыдущей инструкции — можно перемножать одновременно до двух double. Если использовать YMM и ZMM регистры, то до четырёх и восьми double, соответственно.


Последовательность инструкций соответствует ровно тому, что написано у нас в коде — последовательное перемножение промежуточного результата на a. В данном случае нельзя нарушать лево-ассоциативность и каким-либо образом оптимизировать получившийся код.

trickyPlainOctaPow


Напомню, что у нас нет задачи в получении эквивалентного результата. Поэтому мы можем самостоятельно попробовать оптимизировать код путём сокращения количества операций, например, заменить последовательные умножения на 3 операции возведения в квадрат.

Код метода trickyPlainOctaPow() содержательно компилируется в следующий набор инструкций: 

0x0000000002b501be: vmovapd xmm1, xmm0
0x0000000002b501c2: vmulsd  xmm1, xmm1, xmm0
0x0000000002b501c6: vmovapd xmm0, xmm1
0x0000000002b501ca: vmulsd  xmm0, xmm0, xmm1
0x0000000002b501ce: vmovapd xmm1, xmm0
0x0000000002b501d2: vmulsd  xmm1, xmm1, xmm0
0x0000000002b501d6: vmovapd xmm0, xmm1


Как видим, общее количество операций сократилось: вместо 7 умножений мы получили 3 умножения и ещё 2 инструкции vmovapd для подготовки второго операнда в умножении. Получившийся код приблизительно вдвое быстрее если считать в условных latency каждой инструкции.

mathOctaPow


Заглянем внутрь реализации метода Math.pow(): 

    public static double pow(double a, double b) {
        return StrictMath.pow(a, b);
    }


Первое, что можно отметить, — значение степени, передаваемое вторым аргументом, имеет тип double. По этой причине реализация функции уже не может быть такой же простой, как в случае обычного умножения.

При этом StringMath.pow() является native-методом: 

    public static native double pow(double a, double b);


В практическом смысле это означает, что вызов Math.pow() сводится к вызову native-метода средствами JNI, что, как известно, дорого. С другой стороны, в JDK широко используются intrinsic-функции (см. полный перечень интринсиков в HotSpot). Среди которых есть _dpow — интринсик, заменяющий вызов Math.pow().

Последнее означает, что после прогрева, когда код будет скомпилирован JIT-компилятором, мы можем получить такой код вычисления степени в методе mathOctaPow():

Содержательная часть ассемблерного кода метода mathOctaPow ()
  0x0000000002aaacd0: vmovsd xmm1,QWORD PTR [rip+0xffffffffffffff68]        # 0x0000000002aaac40
                                                ;   {section_word}
  0x0000000002aaacd8: vmovsd QWORD PTR [rsp],xmm1
  0x0000000002aaacdd: fld    QWORD PTR [rsp]
  0x0000000002aaace0: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002aaace5: fld    QWORD PTR [rsp]
  0x0000000002aaace8: movabs rax,0x6c4ba7d0     ;   {external_word}
  0x0000000002aaacf2: fld    QWORD PTR [rax]
  0x0000000002aaacf4: fucomip st,st(2)
  0x0000000002aaacf6: jp     0x0000000002aaad0f
  0x0000000002aaacfc: jne    0x0000000002aaad0f
  0x0000000002aaad02: fxch   st(1)
  0x0000000002aaad04: ffree  st(0)
  0x0000000002aaad06: fincstp 
  0x0000000002aaad08: fmul   st,st(0)
  0x0000000002aaad0a: jmp    0x0000000002aab166
  0x0000000002aaad0f: fldz   
  0x0000000002aaad11: fucomip st,st(1)
  0x0000000002aaad13: ja     0x0000000002aaad96
  0x0000000002aaad19: fld    st(1)
  0x0000000002aaad1b: fld    st(1)
  0x0000000002aaad1d: sub    rsp,0x8
  0x0000000002aaad21: fstcw  WORD PTR [rsp]
  0x0000000002aaad25: mov    eax,DWORD PTR [rsp]
  0x0000000002aaad28: or     eax,0x300
  0x0000000002aaad2e: push   rax
  0x0000000002aaad2f: fldcw  WORD PTR [rsp]
  0x0000000002aaad32: pop    rax
  0x0000000002aaad33: fyl2x  
  0x0000000002aaad35: sub    rsp,0x8
  0x0000000002aaad39: fld    st(0)
  0x0000000002aaad3b: frndint 
  0x0000000002aaad3d: fsubr  st(1),st
  0x0000000002aaad3f: fistp  DWORD PTR [rsp]
  0x0000000002aaad42: f2xm1  
  0x0000000002aaad44: fld1   
  0x0000000002aaad46: faddp  st(1),st
  0x0000000002aaad48: mov    eax,DWORD PTR [rsp]
  0x0000000002aaad4b: mov    ecx,0xfffff800
  0x0000000002aaad50: add    eax,0x3ff
  0x0000000002aaad56: mov    edx,eax
  0x0000000002aaad58: shl    eax,0x14
  0x0000000002aaad5b: add    edx,0x1
  0x0000000002aaad5e: cmove  eax,ecx
  0x0000000002aaad61: cmp    edx,0x1
  0x0000000002aaad64: cmove  eax,ecx
  0x0000000002aaad67: test   ecx,edx
  0x0000000002aaad69: cmovne eax,ecx
  0x0000000002aaad6c: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002aaad70: mov    DWORD PTR [rsp],0x0
  0x0000000002aaad77: fmul   QWORD PTR [rsp]
  0x0000000002aaad7a: add    rsp,0x8
  0x0000000002aaad7e: fldcw  WORD PTR [rsp]
  0x0000000002aaad81: add    rsp,0x8
  0x0000000002aaad85: fucomi st,st(0)
  0x0000000002aaad87: jp     0x0000000002aaae36
  0x0000000002aaad8d: ffree  st(2)
  0x0000000002aaad8f: ffree  st(1)
  0x0000000002aaad91: jmp    0x0000000002aab166
  0x0000000002aaad96: fld    st(1)
  0x0000000002aaad98: frndint 
  0x0000000002aaad9a: fucomi st,st(2)
  0x0000000002aaad9c: jne    0x0000000002aaae36
  0x0000000002aaada2: sub    rsp,0x8
  0x0000000002aaada6: fistp  QWORD PTR [rsp]
  0x0000000002aaada9: fld    st(1)
  0x0000000002aaadab: fld    st(1)
  0x0000000002aaadad: fabs   
  0x0000000002aaadaf: sub    rsp,0x8
  0x0000000002aaadb3: fstcw  WORD PTR [rsp]
  0x0000000002aaadb7: mov    eax,DWORD PTR [rsp]
  0x0000000002aaadba: or     eax,0x300
  0x0000000002aaadc0: push   rax
  0x0000000002aaadc1: fldcw  WORD PTR [rsp]
  0x0000000002aaadc4: pop    rax
  0x0000000002aaadc5: fyl2x  
  0x0000000002aaadc7: sub    rsp,0x8
  0x0000000002aaadcb: fld    st(0)
  0x0000000002aaadcd: frndint 
  0x0000000002aaadcf: fsubr  st(1),st
  0x0000000002aaadd1: fistp  DWORD PTR [rsp]
  0x0000000002aaadd4: f2xm1  
  0x0000000002aaadd6: fld1   
  0x0000000002aaadd8: faddp  st(1),st
  0x0000000002aaadda: mov    eax,DWORD PTR [rsp]
  0x0000000002aaaddd: mov    ecx,0xfffff800
  0x0000000002aaade2: add    eax,0x3ff
  0x0000000002aaade8: mov    edx,eax
  0x0000000002aaadea: shl    eax,0x14
  0x0000000002aaaded: add    edx,0x1
  0x0000000002aaadf0: cmove  eax,ecx
  0x0000000002aaadf3: cmp    edx,0x1
  0x0000000002aaadf6: cmove  eax,ecx
  0x0000000002aaadf9: test   ecx,edx
  0x0000000002aaadfb: cmovne eax,ecx
  0x0000000002aaadfe: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002aaae02: mov    DWORD PTR [rsp],0x0
  0x0000000002aaae09: fmul   QWORD PTR [rsp]
  0x0000000002aaae0c: add    rsp,0x8
  0x0000000002aaae10: fldcw  WORD PTR [rsp]
  0x0000000002aaae13: add    rsp,0x8
  0x0000000002aaae17: fucomi st,st(0)
  0x0000000002aaae19: pop    rax
  0x0000000002aaae1a: jp     0x0000000002aaae36
  0x0000000002aaae20: ffree  st(2)
  0x0000000002aaae22: ffree  st(1)
  0x0000000002aaae24: test   eax,0x1
  0x0000000002aaae29: je     0x0000000002aab166
  0x0000000002aaae2f: fchs   
  0x0000000002aaae31: jmp    0x0000000002aab166
  0x0000000002aaae36: ffree  st(0)
  0x0000000002aaae38: fincstp 
  0x0000000002aaae3a: mov    QWORD PTR [rsp-0x28],rsp
  0x0000000002aaae3f: sub    rsp,0x80
  0x0000000002aaae46: mov    QWORD PTR [rsp+0x78],rax
  0x0000000002aaae4b: mov    QWORD PTR [rsp+0x70],rcx
  0x0000000002aaae50: mov    QWORD PTR [rsp+0x68],rdx
  0x0000000002aaae55: mov    QWORD PTR [rsp+0x60],rbx
  0x0000000002aaae5a: mov    QWORD PTR [rsp+0x50],rbp
  0x0000000002aaae5f: mov    QWORD PTR [rsp+0x48],rsi
  0x0000000002aaae64: mov    QWORD PTR [rsp+0x40],rdi
  0x0000000002aaae69: mov    QWORD PTR [rsp+0x38],r8
  0x0000000002aaae6e: mov    QWORD PTR [rsp+0x30],r9
  0x0000000002aaae73: mov    QWORD PTR [rsp+0x28],r10
  0x0000000002aaae78: mov    QWORD PTR [rsp+0x20],r11
  0x0000000002aaae7d: mov    QWORD PTR [rsp+0x18],r12
  0x0000000002aaae82: mov    QWORD PTR [rsp+0x10],r13
  0x0000000002aaae87: mov    QWORD PTR [rsp+0x8],r14
  0x0000000002aaae8c: mov    QWORD PTR [rsp],r15
  0x0000000002aaae90: sub    rsp,0x100
  0x0000000002aaae97: vextractf128 XMMWORD PTR [rsp],ymm0,0x1
  0x0000000002aaae9e: vextractf128 XMMWORD PTR [rsp+0x10],ymm1,0x1
  0x0000000002aaaea6: vextractf128 XMMWORD PTR [rsp+0x20],ymm2,0x1
  0x0000000002aaaeae: vextractf128 XMMWORD PTR [rsp+0x30],ymm3,0x1
  0x0000000002aaaeb6: vextractf128 XMMWORD PTR [rsp+0x40],ymm4,0x1
  0x0000000002aaaebe: vextractf128 XMMWORD PTR [rsp+0x50],ymm5,0x1
  0x0000000002aaaec6: vextractf128 XMMWORD PTR [rsp+0x60],ymm6,0x1
  0x0000000002aaaece: vextractf128 XMMWORD PTR [rsp+0x70],ymm7,0x1
  0x0000000002aaaed6: vextractf128 XMMWORD PTR [rsp+0x80],ymm8,0x1
  0x0000000002aaaee1: vextractf128 XMMWORD PTR [rsp+0x90],ymm9,0x1
  0x0000000002aaaeec: vextractf128 XMMWORD PTR [rsp+0xa0],ymm10,0x1
  0x0000000002aaaef7: vextractf128 XMMWORD PTR [rsp+0xb0],ymm11,0x1
  0x0000000002aaaf02: vextractf128 XMMWORD PTR [rsp+0xc0],ymm12,0x1
  0x0000000002aaaf0d: vextractf128 XMMWORD PTR [rsp+0xd0],ymm13,0x1
  0x0000000002aaaf18: vextractf128 XMMWORD PTR [rsp+0xe0],ymm14,0x1
  0x0000000002aaaf23: vextractf128 XMMWORD PTR [rsp+0xf0],ymm15,0x1
  0x0000000002aaaf2e: sub    rsp,0x100
  0x0000000002aaaf35: vmovdqu XMMWORD PTR [rsp],xmm0
  0x0000000002aaaf3a: vmovdqu XMMWORD PTR [rsp+0x10],xmm1
  0x0000000002aaaf40: vmovdqu XMMWORD PTR [rsp+0x20],xmm2
  0x0000000002aaaf46: vmovdqu XMMWORD PTR [rsp+0x30],xmm3
  0x0000000002aaaf4c: vmovdqu XMMWORD PTR [rsp+0x40],xmm4
  0x0000000002aaaf52: vmovdqu XMMWORD PTR [rsp+0x50],xmm5
  0x0000000002aaaf58: vmovdqu XMMWORD PTR [rsp+0x60],xmm6
  0x0000000002aaaf5e: vmovdqu XMMWORD PTR [rsp+0x70],xmm7
  0x0000000002aaaf64: vmovdqu XMMWORD PTR [rsp+0x80],xmm8
  0x0000000002aaaf6d: vmovdqu XMMWORD PTR [rsp+0x90],xmm9
  0x0000000002aaaf76: vmovdqu XMMWORD PTR [rsp+0xa0],xmm10
  0x0000000002aaaf7f: vmovdqu XMMWORD PTR [rsp+0xb0],xmm11
  0x0000000002aaaf88: vmovdqu XMMWORD PTR [rsp+0xc0],xmm12
  0x0000000002aaaf91: vmovdqu XMMWORD PTR [rsp+0xd0],xmm13
  0x0000000002aaaf9a: vmovdqu XMMWORD PTR [rsp+0xe0],xmm14
  0x0000000002aaafa3: vmovdqu XMMWORD PTR [rsp+0xf0],xmm15
  0x0000000002aaafac: sub    rsp,0x10
  0x0000000002aaafb0: fstp   QWORD PTR [rsp]
  0x0000000002aaafb3: fstp   QWORD PTR [rsp+0x8]
  0x0000000002aaafb7: vmovsd xmm0,QWORD PTR [rsp]
  0x0000000002aaafbc: vmovsd xmm1,QWORD PTR [rsp+0x8]
  0x0000000002aaafc2: sub    rsp,0x20
  0x0000000002aaafc6: test   esp,0xf
  0x0000000002aaafcc: je     0x0000000002aaafe4
  0x0000000002aaafd2: sub    rsp,0x8
  0x0000000002aaafd6: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002aaafdb: add    rsp,0x8
  0x0000000002aaafdf: jmp    0x0000000002aaafe9
  0x0000000002aaafe4: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002aaafe9: add    rsp,0x20
  0x0000000002aaafed: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002aaaff2: fld    QWORD PTR [rsp]
  0x0000000002aaaff5: add    rsp,0x10
  0x0000000002aaaff9: vmovdqu xmm0,XMMWORD PTR [rsp]
  0x0000000002aaaffe: vmovdqu xmm1,XMMWORD PTR [rsp+0x10]
  0x0000000002aab004: vmovdqu xmm2,XMMWORD PTR [rsp+0x20]
  0x0000000002aab00a: vmovdqu xmm3,XMMWORD PTR [rsp+0x30]
  0x0000000002aab010: vmovdqu xmm4,XMMWORD PTR [rsp+0x40]
  0x0000000002aab016: vmovdqu xmm5,XMMWORD PTR [rsp+0x50]
  0x0000000002aab01c: vmovdqu xmm6,XMMWORD PTR [rsp+0x60]
  0x0000000002aab022: vmovdqu xmm7,XMMWORD PTR [rsp+0x70]
  0x0000000002aab028: vmovdqu xmm8,XMMWORD PTR [rsp+0x80]
  0x0000000002aab031: vmovdqu xmm9,XMMWORD PTR [rsp+0x90]
  0x0000000002aab03a: vmovdqu xmm10,XMMWORD PTR [rsp+0xa0]
  0x0000000002aab043: vmovdqu xmm11,XMMWORD PTR [rsp+0xb0]
  0x0000000002aab04c: vmovdqu xmm12,XMMWORD PTR [rsp+0xc0]
  0x0000000002aab055: vmovdqu xmm13,XMMWORD PTR [rsp+0xd0]
  0x0000000002aab05e: vmovdqu xmm14,XMMWORD PTR [rsp+0xe0]
  0x0000000002aab067: vmovdqu xmm15,XMMWORD PTR [rsp+0xf0]
  0x0000000002aab070: add    rsp,0x100
  0x0000000002aab077: vinsertf128 ymm0,ymm0,XMMWORD PTR [rsp],0x1
  0x0000000002aab07e: vinsertf128 ymm1,ymm1,XMMWORD PTR [rsp+0x10],0x1
  0x0000000002aab086: vinsertf128 ymm2,ymm2,XMMWORD PTR [rsp+0x20],0x1
  0x0000000002aab08e: vinsertf128 ymm3,ymm3,XMMWORD PTR [rsp+0x30],0x1
  0x0000000002aab096: vinsertf128 ymm4,ymm4,XMMWORD PTR [rsp+0x40],0x1
  0x0000000002aab09e: vinsertf128 ymm5,ymm5,XMMWORD PTR [rsp+0x50],0x1
  0x0000000002aab0a6: vinsertf128 ymm6,ymm6,XMMWORD PTR [rsp+0x60],0x1
  0x0000000002aab0ae: vinsertf128 ymm7,ymm7,XMMWORD PTR [rsp+0x70],0x1
  0x0000000002aab0b6: vinsertf128 ymm8,ymm8,XMMWORD PTR [rsp+0x80],0x1
  0x0000000002aab0c1: vinsertf128 ymm9,ymm9,XMMWORD PTR [rsp+0x90],0x1
  0x0000000002aab0cc: vinsertf128 ymm10,ymm10,XMMWORD PTR [rsp+0xa0],0x1
  0x0000000002aab0d7: vinsertf128 ymm11,ymm11,XMMWORD PTR [rsp+0xb0],0x1
  0x0000000002aab0e2: vinsertf128 ymm12,ymm12,XMMWORD PTR [rsp+0xc0],0x1
  0x0000000002aab0ed: vinsertf128 ymm13,ymm13,XMMWORD PTR [rsp+0xd0],0x1
  0x0000000002aab0f8: vinsertf128 ymm14,ymm14,XMMWORD PTR [rsp+0xe0],0x1
  0x0000000002aab103: vinsertf128 ymm15,ymm15,XMMWORD PTR [rsp+0xf0],0x1
  0x0000000002aab10e: add    rsp,0x100
  0x0000000002aab115: mov    r15,QWORD PTR [rsp]
  0x0000000002aab119: mov    r14,QWORD PTR [rsp+0x8]
  0x0000000002aab11e: mov    r13,QWORD PTR [rsp+0x10]
  0x0000000002aab123: mov    r12,QWORD PTR [rsp+0x18]
  0x0000000002aab128: mov    r11,QWORD PTR [rsp+0x20]
  0x0000000002aab12d: mov    r10,QWORD PTR [rsp+0x28]
  0x0000000002aab132: mov    r9,QWORD PTR [rsp+0x30]
  0x0000000002aab137: mov    r8,QWORD PTR [rsp+0x38]
  0x0000000002aab13c: mov    rdi,QWORD PTR [rsp+0x40]
  0x0000000002aab141: mov    rsi,QWORD PTR [rsp+0x48]
  0x0000000002aab146: mov    rbp,QWORD PTR [rsp+0x50]
  0x0000000002aab14b: mov    rbx,QWORD PTR [rsp+0x60]
  0x0000000002aab150: mov    rdx,QWORD PTR [rsp+0x68]
  0x0000000002aab155: mov    rcx,QWORD PTR [rsp+0x70]
  0x0000000002aab15a: mov    rax,QWORD PTR [rsp+0x78]
  0x0000000002aab15f: add    rsp,0x80
  0x0000000002aab166: fstp   QWORD PTR [rsp]
  0x0000000002aab169: vmovsd xmm0,QWORD PTR [rsp]  ;*invokestatic pow
                                                ; - ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark::mathOctaPow@4 (line 55)


Здесь первая инструкция записывает константу 8.0 в регистр xmm1, а в регистре xmm0 уже находится значение a. Далее располагается тело intrinsic-функции.

trickyMathOctaPow


Замечательно, вместо одного вызова недешёвогоMath.pow() мы получили целых три. JIT-компилятор заменил тело метода trickyMathOctaPow() на три последовательных реализации _dpow.

Последовательный инлайнинг _dpow
  0x0000000002a70b14: vmovsd xmm1,QWORD PTR [rip+0xffffffffffffff44]        # 0x0000000002a70a60
                                                ;   {section_word}
  0x0000000002a70b1c: vmovsd QWORD PTR [rsp],xmm1
  0x0000000002a70b21: fld    QWORD PTR [rsp]
  0x0000000002a70b24: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a70b29: fld    QWORD PTR [rsp]
  0x0000000002a70b2c: movabs rax,0x6c4ba7d0     ;   {external_word}
  0x0000000002a70b36: fld    QWORD PTR [rax]
  0x0000000002a70b38: fucomip st,st(2)
  0x0000000002a70b3a: jp     0x0000000002a70b53
  0x0000000002a70b40: jne    0x0000000002a70b53
  0x0000000002a70b46: fxch   st(1)
  0x0000000002a70b48: ffree  st(0)
  0x0000000002a70b4a: fincstp 
  0x0000000002a70b4c: fmul   st,st(0)
  0x0000000002a70b4e: jmp    0x0000000002a70faa
  0x0000000002a70b53: fldz   
  0x0000000002a70b55: fucomip st,st(1)
  0x0000000002a70b57: ja     0x0000000002a70bda
  0x0000000002a70b5d: fld    st(1)
  0x0000000002a70b5f: fld    st(1)
  0x0000000002a70b61: sub    rsp,0x8
  0x0000000002a70b65: fstcw  WORD PTR [rsp]
  0x0000000002a70b69: mov    eax,DWORD PTR [rsp]
  0x0000000002a70b6c: or     eax,0x300
  0x0000000002a70b72: push   rax
  0x0000000002a70b73: fldcw  WORD PTR [rsp]
  0x0000000002a70b76: pop    rax
  0x0000000002a70b77: fyl2x  
  0x0000000002a70b79: sub    rsp,0x8
  0x0000000002a70b7d: fld    st(0)
  0x0000000002a70b7f: frndint 
  0x0000000002a70b81: fsubr  st(1),st
  0x0000000002a70b83: fistp  DWORD PTR [rsp]
  0x0000000002a70b86: f2xm1  
  0x0000000002a70b88: fld1   
  0x0000000002a70b8a: faddp  st(1),st
  0x0000000002a70b8c: mov    eax,DWORD PTR [rsp]
  0x0000000002a70b8f: mov    ecx,0xfffff800
  0x0000000002a70b94: add    eax,0x3ff
  0x0000000002a70b9a: mov    edx,eax
  0x0000000002a70b9c: shl    eax,0x14
  0x0000000002a70b9f: add    edx,0x1
  0x0000000002a70ba2: cmove  eax,ecx
  0x0000000002a70ba5: cmp    edx,0x1
  0x0000000002a70ba8: cmove  eax,ecx
  0x0000000002a70bab: test   ecx,edx
  0x0000000002a70bad: cmovne eax,ecx
  0x0000000002a70bb0: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a70bb4: mov    DWORD PTR [rsp],0x0
  0x0000000002a70bbb: fmul   QWORD PTR [rsp]
  0x0000000002a70bbe: add    rsp,0x8
  0x0000000002a70bc2: fldcw  WORD PTR [rsp]
  0x0000000002a70bc5: add    rsp,0x8
  0x0000000002a70bc9: fucomi st,st(0)
  0x0000000002a70bcb: jp     0x0000000002a70c7a
  0x0000000002a70bd1: ffree  st(2)
  0x0000000002a70bd3: ffree  st(1)
  0x0000000002a70bd5: jmp    0x0000000002a70faa
  0x0000000002a70bda: fld    st(1)
  0x0000000002a70bdc: frndint 
  0x0000000002a70bde: fucomi st,st(2)
  0x0000000002a70be0: jne    0x0000000002a70c7a
  0x0000000002a70be6: sub    rsp,0x8
  0x0000000002a70bea: fistp  QWORD PTR [rsp]
  0x0000000002a70bed: fld    st(1)
  0x0000000002a70bef: fld    st(1)
  0x0000000002a70bf1: fabs   
  0x0000000002a70bf3: sub    rsp,0x8
  0x0000000002a70bf7: fstcw  WORD PTR [rsp]
  0x0000000002a70bfb: mov    eax,DWORD PTR [rsp]
  0x0000000002a70bfe: or     eax,0x300
  0x0000000002a70c04: push   rax
  0x0000000002a70c05: fldcw  WORD PTR [rsp]
  0x0000000002a70c08: pop    rax
  0x0000000002a70c09: fyl2x  
  0x0000000002a70c0b: sub    rsp,0x8
  0x0000000002a70c0f: fld    st(0)
  0x0000000002a70c11: frndint 
  0x0000000002a70c13: fsubr  st(1),st
  0x0000000002a70c15: fistp  DWORD PTR [rsp]
  0x0000000002a70c18: f2xm1  
  0x0000000002a70c1a: fld1   
  0x0000000002a70c1c: faddp  st(1),st
  0x0000000002a70c1e: mov    eax,DWORD PTR [rsp]
  0x0000000002a70c21: mov    ecx,0xfffff800
  0x0000000002a70c26: add    eax,0x3ff
  0x0000000002a70c2c: mov    edx,eax
  0x0000000002a70c2e: shl    eax,0x14
  0x0000000002a70c31: add    edx,0x1
  0x0000000002a70c34: cmove  eax,ecx
  0x0000000002a70c37: cmp    edx,0x1
  0x0000000002a70c3a: cmove  eax,ecx
  0x0000000002a70c3d: test   ecx,edx
  0x0000000002a70c3f: cmovne eax,ecx
  0x0000000002a70c42: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a70c46: mov    DWORD PTR [rsp],0x0
  0x0000000002a70c4d: fmul   QWORD PTR [rsp]
  0x0000000002a70c50: add    rsp,0x8
  0x0000000002a70c54: fldcw  WORD PTR [rsp]
  0x0000000002a70c57: add    rsp,0x8
  0x0000000002a70c5b: fucomi st,st(0)
  0x0000000002a70c5d: pop    rax
  0x0000000002a70c5e: jp     0x0000000002a70c7a
  0x0000000002a70c64: ffree  st(2)
  0x0000000002a70c66: ffree  st(1)
  0x0000000002a70c68: test   eax,0x1
  0x0000000002a70c6d: je     0x0000000002a70faa
  0x0000000002a70c73: fchs   
  0x0000000002a70c75: jmp    0x0000000002a70faa
  0x0000000002a70c7a: ffree  st(0)
  0x0000000002a70c7c: fincstp 
  0x0000000002a70c7e: mov    QWORD PTR [rsp-0x28],rsp
  0x0000000002a70c83: sub    rsp,0x80
  0x0000000002a70c8a: mov    QWORD PTR [rsp+0x78],rax
  0x0000000002a70c8f: mov    QWORD PTR [rsp+0x70],rcx
  0x0000000002a70c94: mov    QWORD PTR [rsp+0x68],rdx
  0x0000000002a70c99: mov    QWORD PTR [rsp+0x60],rbx
  0x0000000002a70c9e: mov    QWORD PTR [rsp+0x50],rbp
  0x0000000002a70ca3: mov    QWORD PTR [rsp+0x48],rsi
  0x0000000002a70ca8: mov    QWORD PTR [rsp+0x40],rdi
  0x0000000002a70cad: mov    QWORD PTR [rsp+0x38],r8
  0x0000000002a70cb2: mov    QWORD PTR [rsp+0x30],r9
  0x0000000002a70cb7: mov    QWORD PTR [rsp+0x28],r10
  0x0000000002a70cbc: mov    QWORD PTR [rsp+0x20],r11
  0x0000000002a70cc1: mov    QWORD PTR [rsp+0x18],r12
  0x0000000002a70cc6: mov    QWORD PTR [rsp+0x10],r13
  0x0000000002a70ccb: mov    QWORD PTR [rsp+0x8],r14
  0x0000000002a70cd0: mov    QWORD PTR [rsp],r15
  0x0000000002a70cd4: sub    rsp,0x100
  0x0000000002a70cdb: vextractf128 XMMWORD PTR [rsp],ymm0,0x1
  0x0000000002a70ce2: vextractf128 XMMWORD PTR [rsp+0x10],ymm1,0x1
  0x0000000002a70cea: vextractf128 XMMWORD PTR [rsp+0x20],ymm2,0x1
  0x0000000002a70cf2: vextractf128 XMMWORD PTR [rsp+0x30],ymm3,0x1
  0x0000000002a70cfa: vextractf128 XMMWORD PTR [rsp+0x40],ymm4,0x1
  0x0000000002a70d02: vextractf128 XMMWORD PTR [rsp+0x50],ymm5,0x1
  0x0000000002a70d0a: vextractf128 XMMWORD PTR [rsp+0x60],ymm6,0x1
  0x0000000002a70d12: vextractf128 XMMWORD PTR [rsp+0x70],ymm7,0x1
  0x0000000002a70d1a: vextractf128 XMMWORD PTR [rsp+0x80],ymm8,0x1
  0x0000000002a70d25: vextractf128 XMMWORD PTR [rsp+0x90],ymm9,0x1
  0x0000000002a70d30: vextractf128 XMMWORD PTR [rsp+0xa0],ymm10,0x1
  0x0000000002a70d3b: vextractf128 XMMWORD PTR [rsp+0xb0],ymm11,0x1
  0x0000000002a70d46: vextractf128 XMMWORD PTR [rsp+0xc0],ymm12,0x1
  0x0000000002a70d51: vextractf128 XMMWORD PTR [rsp+0xd0],ymm13,0x1
  0x0000000002a70d5c: vextractf128 XMMWORD PTR [rsp+0xe0],ymm14,0x1
  0x0000000002a70d67: vextractf128 XMMWORD PTR [rsp+0xf0],ymm15,0x1
  0x0000000002a70d72: sub    rsp,0x100
  0x0000000002a70d79: vmovdqu XMMWORD PTR [rsp],xmm0
  0x0000000002a70d7e: vmovdqu XMMWORD PTR [rsp+0x10],xmm1
  0x0000000002a70d84: vmovdqu XMMWORD PTR [rsp+0x20],xmm2
  0x0000000002a70d8a: vmovdqu XMMWORD PTR [rsp+0x30],xmm3
  0x0000000002a70d90: vmovdqu XMMWORD PTR [rsp+0x40],xmm4
  0x0000000002a70d96: vmovdqu XMMWORD PTR [rsp+0x50],xmm5
  0x0000000002a70d9c: vmovdqu XMMWORD PTR [rsp+0x60],xmm6
  0x0000000002a70da2: vmovdqu XMMWORD PTR [rsp+0x70],xmm7
  0x0000000002a70da8: vmovdqu XMMWORD PTR [rsp+0x80],xmm8
  0x0000000002a70db1: vmovdqu XMMWORD PTR [rsp+0x90],xmm9
  0x0000000002a70dba: vmovdqu XMMWORD PTR [rsp+0xa0],xmm10
  0x0000000002a70dc3: vmovdqu XMMWORD PTR [rsp+0xb0],xmm11
  0x0000000002a70dcc: vmovdqu XMMWORD PTR [rsp+0xc0],xmm12
  0x0000000002a70dd5: vmovdqu XMMWORD PTR [rsp+0xd0],xmm13
  0x0000000002a70dde: vmovdqu XMMWORD PTR [rsp+0xe0],xmm14
  0x0000000002a70de7: vmovdqu XMMWORD PTR [rsp+0xf0],xmm15
  0x0000000002a70df0: sub    rsp,0x10
  0x0000000002a70df4: fstp   QWORD PTR [rsp]
  0x0000000002a70df7: fstp   QWORD PTR [rsp+0x8]
  0x0000000002a70dfb: vmovsd xmm0,QWORD PTR [rsp]
  0x0000000002a70e00: vmovsd xmm1,QWORD PTR [rsp+0x8]
  0x0000000002a70e06: sub    rsp,0x20
  0x0000000002a70e0a: test   esp,0xf
  0x0000000002a70e10: je     0x0000000002a70e28
  0x0000000002a70e16: sub    rsp,0x8
  0x0000000002a70e1a: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a70e1f: add    rsp,0x8
  0x0000000002a70e23: jmp    0x0000000002a70e2d
  0x0000000002a70e28: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a70e2d: add    rsp,0x20
  0x0000000002a70e31: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a70e36: fld    QWORD PTR [rsp]
  0x0000000002a70e39: add    rsp,0x10
  0x0000000002a70e3d: vmovdqu xmm0,XMMWORD PTR [rsp]
  0x0000000002a70e42: vmovdqu xmm1,XMMWORD PTR [rsp+0x10]
  0x0000000002a70e48: vmovdqu xmm2,XMMWORD PTR [rsp+0x20]
  0x0000000002a70e4e: vmovdqu xmm3,XMMWORD PTR [rsp+0x30]
  0x0000000002a70e54: vmovdqu xmm4,XMMWORD PTR [rsp+0x40]
  0x0000000002a70e5a: vmovdqu xmm5,XMMWORD PTR [rsp+0x50]
  0x0000000002a70e60: vmovdqu xmm6,XMMWORD PTR [rsp+0x60]
  0x0000000002a70e66: vmovdqu xmm7,XMMWORD PTR [rsp+0x70]
  0x0000000002a70e6c: vmovdqu xmm8,XMMWORD PTR [rsp+0x80]
  0x0000000002a70e75: vmovdqu xmm9,XMMWORD PTR [rsp+0x90]
  0x0000000002a70e7e: vmovdqu xmm10,XMMWORD PTR [rsp+0xa0]
  0x0000000002a70e87: vmovdqu xmm11,XMMWORD PTR [rsp+0xb0]
  0x0000000002a70e90: vmovdqu xmm12,XMMWORD PTR [rsp+0xc0]
  0x0000000002a70e99: vmovdqu xmm13,XMMWORD PTR [rsp+0xd0]
  0x0000000002a70ea2: vmovdqu xmm14,XMMWORD PTR [rsp+0xe0]
  0x0000000002a70eab: vmovdqu xmm15,XMMWORD PTR [rsp+0xf0]
  0x0000000002a70eb4: add    rsp,0x100
  0x0000000002a70ebb: vinsertf128 ymm0,ymm0,XMMWORD PTR [rsp],0x1
  0x0000000002a70ec2: vinsertf128 ymm1,ymm1,XMMWORD PTR [rsp+0x10],0x1
  0x0000000002a70eca: vinsertf128 ymm2,ymm2,XMMWORD PTR [rsp+0x20],0x1
  0x0000000002a70ed2: vinsertf128 ymm3,ymm3,XMMWORD PTR [rsp+0x30],0x1
  0x0000000002a70eda: vinsertf128 ymm4,ymm4,XMMWORD PTR [rsp+0x40],0x1
  0x0000000002a70ee2: vinsertf128 ymm5,ymm5,XMMWORD PTR [rsp+0x50],0x1
  0x0000000002a70eea: vinsertf128 ymm6,ymm6,XMMWORD PTR [rsp+0x60],0x1
  0x0000000002a70ef2: vinsertf128 ymm7,ymm7,XMMWORD PTR [rsp+0x70],0x1
  0x0000000002a70efa: vinsertf128 ymm8,ymm8,XMMWORD PTR [rsp+0x80],0x1
  0x0000000002a70f05: vinsertf128 ymm9,ymm9,XMMWORD PTR [rsp+0x90],0x1
  0x0000000002a70f10: vinsertf128 ymm10,ymm10,XMMWORD PTR [rsp+0xa0],0x1
  0x0000000002a70f1b: vinsertf128 ymm11,ymm11,XMMWORD PTR [rsp+0xb0],0x1
  0x0000000002a70f26: vinsertf128 ymm12,ymm12,XMMWORD PTR [rsp+0xc0],0x1
  0x0000000002a70f31: vinsertf128 ymm13,ymm13,XMMWORD PTR [rsp+0xd0],0x1
  0x0000000002a70f3c: vinsertf128 ymm14,ymm14,XMMWORD PTR [rsp+0xe0],0x1
  0x0000000002a70f47: vinsertf128 ymm15,ymm15,XMMWORD PTR [rsp+0xf0],0x1
  0x0000000002a70f52: add    rsp,0x100
  0x0000000002a70f59: mov    r15,QWORD PTR [rsp]
  0x0000000002a70f5d: mov    r14,QWORD PTR [rsp+0x8]
  0x0000000002a70f62: mov    r13,QWORD PTR [rsp+0x10]
  0x0000000002a70f67: mov    r12,QWORD PTR [rsp+0x18]
  0x0000000002a70f6c: mov    r11,QWORD PTR [rsp+0x20]
  0x0000000002a70f71: mov    r10,QWORD PTR [rsp+0x28]
  0x0000000002a70f76: mov    r9,QWORD PTR [rsp+0x30]
  0x0000000002a70f7b: mov    r8,QWORD PTR [rsp+0x38]
  0x0000000002a70f80: mov    rdi,QWORD PTR [rsp+0x40]
  0x0000000002a70f85: mov    rsi,QWORD PTR [rsp+0x48]
  0x0000000002a70f8a: mov    rbp,QWORD PTR [rsp+0x50]
  0x0000000002a70f8f: mov    rbx,QWORD PTR [rsp+0x60]
  0x0000000002a70f94: mov    rdx,QWORD PTR [rsp+0x68]
  0x0000000002a70f99: mov    rcx,QWORD PTR [rsp+0x70]
  0x0000000002a70f9e: mov    rax,QWORD PTR [rsp+0x78]
  0x0000000002a70fa3: add    rsp,0x80
  0x0000000002a70faa: fstp   QWORD PTR [rsp]
  0x0000000002a70fad: vmovsd xmm0,QWORD PTR [rsp]  ;*invokestatic pow
                                                ; - ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark::trickyMathOctaPow@4 (line 63)

  0x0000000002a70fb2: vmovsd xmm1,QWORD PTR [rip+0xfffffffffffffaae]        # 0x0000000002a70a68
                                                ;   {section_word}
  0x0000000002a70fba: vmovsd QWORD PTR [rsp],xmm1
  0x0000000002a70fbf: fld    QWORD PTR [rsp]
  0x0000000002a70fc2: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a70fc7: fld    QWORD PTR [rsp]
  0x0000000002a70fca: movabs rax,0x6c4ba7d0     ;   {external_word}
  0x0000000002a70fd4: fld    QWORD PTR [rax]
  0x0000000002a70fd6: fucomip st,st(2)
  0x0000000002a70fd8: jp     0x0000000002a70ff1
  0x0000000002a70fde: jne    0x0000000002a70ff1
  0x0000000002a70fe4: fxch   st(1)
  0x0000000002a70fe6: ffree  st(0)
  0x0000000002a70fe8: fincstp 
  0x0000000002a70fea: fmul   st,st(0)
  0x0000000002a70fec: jmp    0x0000000002a71448
  0x0000000002a70ff1: fldz   
  0x0000000002a70ff3: fucomip st,st(1)
  0x0000000002a70ff5: ja     0x0000000002a71078
  0x0000000002a70ffb: fld    st(1)
  0x0000000002a70ffd: fld    st(1)
  0x0000000002a70fff: sub    rsp,0x8
  0x0000000002a71003: fstcw  WORD PTR [rsp]
  0x0000000002a71007: mov    eax,DWORD PTR [rsp]
  0x0000000002a7100a: or     eax,0x300
  0x0000000002a71010: push   rax
  0x0000000002a71011: fldcw  WORD PTR [rsp]
  0x0000000002a71014: pop    rax
  0x0000000002a71015: fyl2x  
  0x0000000002a71017: sub    rsp,0x8
  0x0000000002a7101b: fld    st(0)
  0x0000000002a7101d: frndint 
  0x0000000002a7101f: fsubr  st(1),st
  0x0000000002a71021: fistp  DWORD PTR [rsp]
  0x0000000002a71024: f2xm1  
  0x0000000002a71026: fld1   
  0x0000000002a71028: faddp  st(1),st
  0x0000000002a7102a: mov    eax,DWORD PTR [rsp]
  0x0000000002a7102d: mov    ecx,0xfffff800
  0x0000000002a71032: add    eax,0x3ff
  0x0000000002a71038: mov    edx,eax
  0x0000000002a7103a: shl    eax,0x14
  0x0000000002a7103d: add    edx,0x1
  0x0000000002a71040: cmove  eax,ecx
  0x0000000002a71043: cmp    edx,0x1
  0x0000000002a71046: cmove  eax,ecx
  0x0000000002a71049: test   ecx,edx
  0x0000000002a7104b: cmovne eax,ecx
  0x0000000002a7104e: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a71052: mov    DWORD PTR [rsp],0x0
  0x0000000002a71059: fmul   QWORD PTR [rsp]
  0x0000000002a7105c: add    rsp,0x8
  0x0000000002a71060: fldcw  WORD PTR [rsp]
  0x0000000002a71063: add    rsp,0x8
  0x0000000002a71067: fucomi st,st(0)
  0x0000000002a71069: jp     0x0000000002a71118
  0x0000000002a7106f: ffree  st(2)
  0x0000000002a71071: ffree  st(1)
  0x0000000002a71073: jmp    0x0000000002a71448
  0x0000000002a71078: fld    st(1)
  0x0000000002a7107a: frndint 
  0x0000000002a7107c: fucomi st,st(2)
  0x0000000002a7107e: jne    0x0000000002a71118
  0x0000000002a71084: sub    rsp,0x8
  0x0000000002a71088: fistp  QWORD PTR [rsp]
  0x0000000002a7108b: fld    st(1)
  0x0000000002a7108d: fld    st(1)
  0x0000000002a7108f: fabs   
  0x0000000002a71091: sub    rsp,0x8
  0x0000000002a71095: fstcw  WORD PTR [rsp]
  0x0000000002a71099: mov    eax,DWORD PTR [rsp]
  0x0000000002a7109c: or     eax,0x300
  0x0000000002a710a2: push   rax
  0x0000000002a710a3: fldcw  WORD PTR [rsp]
  0x0000000002a710a6: pop    rax
  0x0000000002a710a7: fyl2x  
  0x0000000002a710a9: sub    rsp,0x8
  0x0000000002a710ad: fld    st(0)
  0x0000000002a710af: frndint 
  0x0000000002a710b1: fsubr  st(1),st
  0x0000000002a710b3: fistp  DWORD PTR [rsp]
  0x0000000002a710b6: f2xm1  
  0x0000000002a710b8: fld1   
  0x0000000002a710ba: faddp  st(1),st
  0x0000000002a710bc: mov    eax,DWORD PTR [rsp]
  0x0000000002a710bf: mov    ecx,0xfffff800
  0x0000000002a710c4: add    eax,0x3ff
  0x0000000002a710ca: mov    edx,eax
  0x0000000002a710cc: shl    eax,0x14
  0x0000000002a710cf: add    edx,0x1
  0x0000000002a710d2: cmove  eax,ecx
  0x0000000002a710d5: cmp    edx,0x1
  0x0000000002a710d8: cmove  eax,ecx
  0x0000000002a710db: test   ecx,edx
  0x0000000002a710dd: cmovne eax,ecx
  0x0000000002a710e0: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a710e4: mov    DWORD PTR [rsp],0x0
  0x0000000002a710eb: fmul   QWORD PTR [rsp]
  0x0000000002a710ee: add    rsp,0x8
  0x0000000002a710f2: fldcw  WORD PTR [rsp]
  0x0000000002a710f5: add    rsp,0x8
  0x0000000002a710f9: fucomi st,st(0)
  0x0000000002a710fb: pop    rax
  0x0000000002a710fc: jp     0x0000000002a71118
  0x0000000002a71102: ffree  st(2)
  0x0000000002a71104: ffree  st(1)
  0x0000000002a71106: test   eax,0x1
  0x0000000002a7110b: je     0x0000000002a71448
  0x0000000002a71111: fchs   
  0x0000000002a71113: jmp    0x0000000002a71448
  0x0000000002a71118: ffree  st(0)
  0x0000000002a7111a: fincstp 
  0x0000000002a7111c: mov    QWORD PTR [rsp-0x28],rsp
  0x0000000002a71121: sub    rsp,0x80
  0x0000000002a71128: mov    QWORD PTR [rsp+0x78],rax
  0x0000000002a7112d: mov    QWORD PTR [rsp+0x70],rcx
  0x0000000002a71132: mov    QWORD PTR [rsp+0x68],rdx
  0x0000000002a71137: mov    QWORD PTR [rsp+0x60],rbx
  0x0000000002a7113c: mov    QWORD PTR [rsp+0x50],rbp
  0x0000000002a71141: mov    QWORD PTR [rsp+0x48],rsi
  0x0000000002a71146: mov    QWORD PTR [rsp+0x40],rdi
  0x0000000002a7114b: mov    QWORD PTR [rsp+0x38],r8
  0x0000000002a71150: mov    QWORD PTR [rsp+0x30],r9
  0x0000000002a71155: mov    QWORD PTR [rsp+0x28],r10
  0x0000000002a7115a: mov    QWORD PTR [rsp+0x20],r11
  0x0000000002a7115f: mov    QWORD PTR [rsp+0x18],r12
  0x0000000002a71164: mov    QWORD PTR [rsp+0x10],r13
  0x0000000002a71169: mov    QWORD PTR [rsp+0x8],r14
  0x0000000002a7116e: mov    QWORD PTR [rsp],r15
  0x0000000002a71172: sub    rsp,0x100
  0x0000000002a71179: vextractf128 XMMWORD PTR [rsp],ymm0,0x1
  0x0000000002a71180: vextractf128 XMMWORD PTR [rsp+0x10],ymm1,0x1
  0x0000000002a71188: vextractf128 XMMWORD PTR [rsp+0x20],ymm2,0x1
  0x0000000002a71190: vextractf128 XMMWORD PTR [rsp+0x30],ymm3,0x1
  0x0000000002a71198: vextractf128 XMMWORD PTR [rsp+0x40],ymm4,0x1
  0x0000000002a711a0: vextractf128 XMMWORD PTR [rsp+0x50],ymm5,0x1
  0x0000000002a711a8: vextractf128 XMMWORD PTR [rsp+0x60],ymm6,0x1
  0x0000000002a711b0: vextractf128 XMMWORD PTR [rsp+0x70],ymm7,0x1
  0x0000000002a711b8: vextractf128 XMMWORD PTR [rsp+0x80],ymm8,0x1
  0x0000000002a711c3: vextractf128 XMMWORD PTR [rsp+0x90],ymm9,0x1
  0x0000000002a711ce: vextractf128 XMMWORD PTR [rsp+0xa0],ymm10,0x1
  0x0000000002a711d9: vextractf128 XMMWORD PTR [rsp+0xb0],ymm11,0x1
  0x0000000002a711e4: vextractf128 XMMWORD PTR [rsp+0xc0],ymm12,0x1
  0x0000000002a711ef: vextractf128 XMMWORD PTR [rsp+0xd0],ymm13,0x1
  0x0000000002a711fa: vextractf128 XMMWORD PTR [rsp+0xe0],ymm14,0x1
  0x0000000002a71205: vextractf128 XMMWORD PTR [rsp+0xf0],ymm15,0x1
  0x0000000002a71210: sub    rsp,0x100
  0x0000000002a71217: vmovdqu XMMWORD PTR [rsp],xmm0
  0x0000000002a7121c: vmovdqu XMMWORD PTR [rsp+0x10],xmm1
  0x0000000002a71222: vmovdqu XMMWORD PTR [rsp+0x20],xmm2
  0x0000000002a71228: vmovdqu XMMWORD PTR [rsp+0x30],xmm3
  0x0000000002a7122e: vmovdqu XMMWORD PTR [rsp+0x40],xmm4
  0x0000000002a71234: vmovdqu XMMWORD PTR [rsp+0x50],xmm5
  0x0000000002a7123a: vmovdqu XMMWORD PTR [rsp+0x60],xmm6
  0x0000000002a71240: vmovdqu XMMWORD PTR [rsp+0x70],xmm7
  0x0000000002a71246: vmovdqu XMMWORD PTR [rsp+0x80],xmm8
  0x0000000002a7124f: vmovdqu XMMWORD PTR [rsp+0x90],xmm9
  0x0000000002a71258: vmovdqu XMMWORD PTR [rsp+0xa0],xmm10
  0x0000000002a71261: vmovdqu XMMWORD PTR [rsp+0xb0],xmm11
  0x0000000002a7126a: vmovdqu XMMWORD PTR [rsp+0xc0],xmm12
  0x0000000002a71273: vmovdqu XMMWORD PTR [rsp+0xd0],xmm13
  0x0000000002a7127c: vmovdqu XMMWORD PTR [rsp+0xe0],xmm14
  0x0000000002a71285: vmovdqu XMMWORD PTR [rsp+0xf0],xmm15
  0x0000000002a7128e: sub    rsp,0x10
  0x0000000002a71292: fstp   QWORD PTR [rsp]
  0x0000000002a71295: fstp   QWORD PTR [rsp+0x8]
  0x0000000002a71299: vmovsd xmm0,QWORD PTR [rsp]
  0x0000000002a7129e: vmovsd xmm1,QWORD PTR [rsp+0x8]
  0x0000000002a712a4: sub    rsp,0x20
  0x0000000002a712a8: test   esp,0xf
  0x0000000002a712ae: je     0x0000000002a712c6
  0x0000000002a712b4: sub    rsp,0x8
  0x0000000002a712b8: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a712bd: add    rsp,0x8
  0x0000000002a712c1: jmp    0x0000000002a712cb
  0x0000000002a712c6: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a712cb: add    rsp,0x20
  0x0000000002a712cf: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a712d4: fld    QWORD PTR [rsp]
  0x0000000002a712d7: add    rsp,0x10
  0x0000000002a712db: vmovdqu xmm0,XMMWORD PTR [rsp]
  0x0000000002a712e0: vmovdqu xmm1,XMMWORD PTR [rsp+0x10]
  0x0000000002a712e6: vmovdqu xmm2,XMMWORD PTR [rsp+0x20]
  0x0000000002a712ec: vmovdqu xmm3,XMMWORD PTR [rsp+0x30]
  0x0000000002a712f2: vmovdqu xmm4,XMMWORD PTR [rsp+0x40]
  0x0000000002a712f8: vmovdqu xmm5,XMMWORD PTR [rsp+0x50]
  0x0000000002a712fe: vmovdqu xmm6,XMMWORD PTR [rsp+0x60]
  0x0000000002a71304: vmovdqu xmm7,XMMWORD PTR [rsp+0x70]
  0x0000000002a7130a: vmovdqu xmm8,XMMWORD PTR [rsp+0x80]
  0x0000000002a71313: vmovdqu xmm9,XMMWORD PTR [rsp+0x90]
  0x0000000002a7131c: vmovdqu xmm10,XMMWORD PTR [rsp+0xa0]
  0x0000000002a71325: vmovdqu xmm11,XMMWORD PTR [rsp+0xb0]
  0x0000000002a7132e: vmovdqu xmm12,XMMWORD PTR [rsp+0xc0]
  0x0000000002a71337: vmovdqu xmm13,XMMWORD PTR [rsp+0xd0]
  0x0000000002a71340: vmovdqu xmm14,XMMWORD PTR [rsp+0xe0]
  0x0000000002a71349: vmovdqu xmm15,XMMWORD PTR [rsp+0xf0]
  0x0000000002a71352: add    rsp,0x100
  0x0000000002a71359: vinsertf128 ymm0,ymm0,XMMWORD PTR [rsp],0x1
  0x0000000002a71360: vinsertf128 ymm1,ymm1,XMMWORD PTR [rsp+0x10],0x1
  0x0000000002a71368: vinsertf128 ymm2,ymm2,XMMWORD PTR [rsp+0x20],0x1
  0x0000000002a71370: vinsertf128 ymm3,ymm3,XMMWORD PTR [rsp+0x30],0x1
  0x0000000002a71378: vinsertf128 ymm4,ymm4,XMMWORD PTR [rsp+0x40],0x1
  0x0000000002a71380: vinsertf128 ymm5,ymm5,XMMWORD PTR [rsp+0x50],0x1
  0x0000000002a71388: vinsertf128 ymm6,ymm6,XMMWORD PTR [rsp+0x60],0x1
  0x0000000002a71390: vinsertf128 ymm7,ymm7,XMMWORD PTR [rsp+0x70],0x1
  0x0000000002a71398: vinsertf128 ymm8,ymm8,XMMWORD PTR [rsp+0x80],0x1
  0x0000000002a713a3: vinsertf128 ymm9,ymm9,XMMWORD PTR [rsp+0x90],0x1
  0x0000000002a713ae: vinsertf128 ymm10,ymm10,XMMWORD PTR [rsp+0xa0],0x1
  0x0000000002a713b9: vinsertf128 ymm11,ymm11,XMMWORD PTR [rsp+0xb0],0x1
  0x0000000002a713c4: vinsertf128 ymm12,ymm12,XMMWORD PTR [rsp+0xc0],0x1
  0x0000000002a713cf: vinsertf128 ymm13,ymm13,XMMWORD PTR [rsp+0xd0],0x1
  0x0000000002a713da: vinsertf128 ymm14,ymm14,XMMWORD PTR [rsp+0xe0],0x1
  0x0000000002a713e5: vinsertf128 ymm15,ymm15,XMMWORD PTR [rsp+0xf0],0x1
  0x0000000002a713f0: add    rsp,0x100
  0x0000000002a713f7: mov    r15,QWORD PTR [rsp]
  0x0000000002a713fb: mov    r14,QWORD PTR [rsp+0x8]
  0x0000000002a71400: mov    r13,QWORD PTR [rsp+0x10]
  0x0000000002a71405: mov    r12,QWORD PTR [rsp+0x18]
  0x0000000002a7140a: mov    r11,QWORD PTR [rsp+0x20]
  0x0000000002a7140f: mov    r10,QWORD PTR [rsp+0x28]
  0x0000000002a71414: mov    r9,QWORD PTR [rsp+0x30]
  0x0000000002a71419: mov    r8,QWORD PTR [rsp+0x38]
  0x0000000002a7141e: mov    rdi,QWORD PTR [rsp+0x40]
  0x0000000002a71423: mov    rsi,QWORD PTR [rsp+0x48]
  0x0000000002a71428: mov    rbp,QWORD PTR [rsp+0x50]
  0x0000000002a7142d: mov    rbx,QWORD PTR [rsp+0x60]
  0x0000000002a71432: mov    rdx,QWORD PTR [rsp+0x68]
  0x0000000002a71437: mov    rcx,QWORD PTR [rsp+0x70]
  0x0000000002a7143c: mov    rax,QWORD PTR [rsp+0x78]
  0x0000000002a71441: add    rsp,0x80
  0x0000000002a71448: fstp   QWORD PTR [rsp]
  0x0000000002a7144b: vmovsd xmm0,QWORD PTR [rsp]  ;*invokestatic pow
                                                ; - ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark::trickyMathOctaPow@10 (line 63)

  0x0000000002a71450: vmovsd xmm1,QWORD PTR [rip+0xfffffffffffff618]        # 0x0000000002a70a70
                                                ;   {section_word}
  0x0000000002a71458: vmovsd QWORD PTR [rsp],xmm1
  0x0000000002a7145d: fld    QWORD PTR [rsp]
  0x0000000002a71460: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a71465: fld    QWORD PTR [rsp]
  0x0000000002a71468: movabs rax,0x6c4ba7d0     ;   {external_word}
  0x0000000002a71472: fld    QWORD PTR [rax]
  0x0000000002a71474: fucomip st,st(2)
  0x0000000002a71476: jp     0x0000000002a7148f
  0x0000000002a7147c: jne    0x0000000002a7148f
  0x0000000002a71482: fxch   st(1)
  0x0000000002a71484: ffree  st(0)
  0x0000000002a71486: fincstp 
  0x0000000002a71488: fmul   st,st(0)
  0x0000000002a7148a: jmp    0x0000000002a718e6
  0x0000000002a7148f: fldz   
  0x0000000002a71491: fucomip st,st(1)
  0x0000000002a71493: ja     0x0000000002a71516
  0x0000000002a71499: fld    st(1)
  0x0000000002a7149b: fld    st(1)
  0x0000000002a7149d: sub    rsp,0x8
  0x0000000002a714a1: fstcw  WORD PTR [rsp]
  0x0000000002a714a5: mov    eax,DWORD PTR [rsp]
  0x0000000002a714a8: or     eax,0x300
  0x0000000002a714ae: push   rax
  0x0000000002a714af: fldcw  WORD PTR [rsp]
  0x0000000002a714b2: pop    rax
  0x0000000002a714b3: fyl2x  
  0x0000000002a714b5: sub    rsp,0x8
  0x0000000002a714b9: fld    st(0)
  0x0000000002a714bb: frndint 
  0x0000000002a714bd: fsubr  st(1),st
  0x0000000002a714bf: fistp  DWORD PTR [rsp]
  0x0000000002a714c2: f2xm1  
  0x0000000002a714c4: fld1   
  0x0000000002a714c6: faddp  st(1),st
  0x0000000002a714c8: mov    eax,DWORD PTR [rsp]
  0x0000000002a714cb: mov    ecx,0xfffff800
  0x0000000002a714d0: add    eax,0x3ff
  0x0000000002a714d6: mov    edx,eax
  0x0000000002a714d8: shl    eax,0x14
  0x0000000002a714db: add    edx,0x1
  0x0000000002a714de: cmove  eax,ecx
  0x0000000002a714e1: cmp    edx,0x1
  0x0000000002a714e4: cmove  eax,ecx
  0x0000000002a714e7: test   ecx,edx
  0x0000000002a714e9: cmovne eax,ecx
  0x0000000002a714ec: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a714f0: mov    DWORD PTR [rsp],0x0
  0x0000000002a714f7: fmul   QWORD PTR [rsp]
  0x0000000002a714fa: add    rsp,0x8
  0x0000000002a714fe: fldcw  WORD PTR [rsp]
  0x0000000002a71501: add    rsp,0x8
  0x0000000002a71505: fucomi st,st(0)
  0x0000000002a71507: jp     0x0000000002a715b6
  0x0000000002a7150d: ffree  st(2)
  0x0000000002a7150f: ffree  st(1)
  0x0000000002a71511: jmp    0x0000000002a718e6
  0x0000000002a71516: fld    st(1)
  0x0000000002a71518: frndint 
  0x0000000002a7151a: fucomi st,st(2)
  0x0000000002a7151c: jne    0x0000000002a715b6
  0x0000000002a71522: sub    rsp,0x8
  0x0000000002a71526: fistp  QWORD PTR [rsp]
  0x0000000002a71529: fld    st(1)
  0x0000000002a7152b: fld    st(1)
  0x0000000002a7152d: fabs   
  0x0000000002a7152f: sub    rsp,0x8
  0x0000000002a71533: fstcw  WORD PTR [rsp]
  0x0000000002a71537: mov    eax,DWORD PTR [rsp]
  0x0000000002a7153a: or     eax,0x300
  0x0000000002a71540: push   rax
  0x0000000002a71541: fldcw  WORD PTR [rsp]
  0x0000000002a71544: pop    rax
  0x0000000002a71545: fyl2x  
  0x0000000002a71547: sub    rsp,0x8
  0x0000000002a7154b: fld    st(0)
  0x0000000002a7154d: frndint 
  0x0000000002a7154f: fsubr  st(1),st
  0x0000000002a71551: fistp  DWORD PTR [rsp]
  0x0000000002a71554: f2xm1  
  0x0000000002a71556: fld1   
  0x0000000002a71558: faddp  st(1),st
  0x0000000002a7155a: mov    eax,DWORD PTR [rsp]
  0x0000000002a7155d: mov    ecx,0xfffff800
  0x0000000002a71562: add    eax,0x3ff
  0x0000000002a71568: mov    edx,eax
  0x0000000002a7156a: shl    eax,0x14
  0x0000000002a7156d: add    edx,0x1
  0x0000000002a71570: cmove  eax,ecx
  0x0000000002a71573: cmp    edx,0x1
  0x0000000002a71576: cmove  eax,ecx
  0x0000000002a71579: test   ecx,edx
  0x0000000002a7157b: cmovne eax,ecx
  0x0000000002a7157e: mov    DWORD PTR [rsp+0x4],eax
  0x0000000002a71582: mov    DWORD PTR [rsp],0x0
  0x0000000002a71589: fmul   QWORD PTR [rsp]
  0x0000000002a7158c: add    rsp,0x8
  0x0000000002a71590: fldcw  WORD PTR [rsp]
  0x0000000002a71593: add    rsp,0x8
  0x0000000002a71597: fucomi st,st(0)
  0x0000000002a71599: pop    rax
  0x0000000002a7159a: jp     0x0000000002a715b6
  0x0000000002a715a0: ffree  st(2)
  0x0000000002a715a2: ffree  st(1)
  0x0000000002a715a4: test   eax,0x1
  0x0000000002a715a9: je     0x0000000002a718e6
  0x0000000002a715af: fchs   
  0x0000000002a715b1: jmp    0x0000000002a718e6
  0x0000000002a715b6: ffree  st(0)
  0x0000000002a715b8: fincstp 
  0x0000000002a715ba: mov    QWORD PTR [rsp-0x28],rsp
  0x0000000002a715bf: sub    rsp,0x80
  0x0000000002a715c6: mov    QWORD PTR [rsp+0x78],rax
  0x0000000002a715cb: mov    QWORD PTR [rsp+0x70],rcx
  0x0000000002a715d0: mov    QWORD PTR [rsp+0x68],rdx
  0x0000000002a715d5: mov    QWORD PTR [rsp+0x60],rbx
  0x0000000002a715da: mov    QWORD PTR [rsp+0x50],rbp
  0x0000000002a715df: mov    QWORD PTR [rsp+0x48],rsi
  0x0000000002a715e4: mov    QWORD PTR [rsp+0x40],rdi
  0x0000000002a715e9: mov    QWORD PTR [rsp+0x38],r8
  0x0000000002a715ee: mov    QWORD PTR [rsp+0x30],r9
  0x0000000002a715f3: mov    QWORD PTR [rsp+0x28],r10
  0x0000000002a715f8: mov    QWORD PTR [rsp+0x20],r11
  0x0000000002a715fd: mov    QWORD PTR [rsp+0x18],r12
  0x0000000002a71602: mov    QWORD PTR [rsp+0x10],r13
  0x0000000002a71607: mov    QWORD PTR [rsp+0x8],r14
  0x0000000002a7160c: mov    QWORD PTR [rsp],r15
  0x0000000002a71610: sub    rsp,0x100
  0x0000000002a71617: vextractf128 XMMWORD PTR [rsp],ymm0,0x1
  0x0000000002a7161e: vextractf128 XMMWORD PTR [rsp+0x10],ymm1,0x1
  0x0000000002a71626: vextractf128 XMMWORD PTR [rsp+0x20],ymm2,0x1
  0x0000000002a7162e: vextractf128 XMMWORD PTR [rsp+0x30],ymm3,0x1
  0x0000000002a71636: vextractf128 XMMWORD PTR [rsp+0x40],ymm4,0x1
  0x0000000002a7163e: vextractf128 XMMWORD PTR [rsp+0x50],ymm5,0x1
  0x0000000002a71646: vextractf128 XMMWORD PTR [rsp+0x60],ymm6,0x1
  0x0000000002a7164e: vextractf128 XMMWORD PTR [rsp+0x70],ymm7,0x1
  0x0000000002a71656: vextractf128 XMMWORD PTR [rsp+0x80],ymm8,0x1
  0x0000000002a71661: vextractf128 XMMWORD PTR [rsp+0x90],ymm9,0x1
  0x0000000002a7166c: vextractf128 XMMWORD PTR [rsp+0xa0],ymm10,0x1
  0x0000000002a71677: vextractf128 XMMWORD PTR [rsp+0xb0],ymm11,0x1
  0x0000000002a71682: vextractf128 XMMWORD PTR [rsp+0xc0],ymm12,0x1
  0x0000000002a7168d: vextractf128 XMMWORD PTR [rsp+0xd0],ymm13,0x1
  0x0000000002a71698: vextractf128 XMMWORD PTR [rsp+0xe0],ymm14,0x1
  0x0000000002a716a3: vextractf128 XMMWORD PTR [rsp+0xf0],ymm15,0x1
  0x0000000002a716ae: sub    rsp,0x100
  0x0000000002a716b5: vmovdqu XMMWORD PTR [rsp],xmm0
  0x0000000002a716ba: vmovdqu XMMWORD PTR [rsp+0x10],xmm1
  0x0000000002a716c0: vmovdqu XMMWORD PTR [rsp+0x20],xmm2
  0x0000000002a716c6: vmovdqu XMMWORD PTR [rsp+0x30],xmm3
  0x0000000002a716cc: vmovdqu XMMWORD PTR [rsp+0x40],xmm4
  0x0000000002a716d2: vmovdqu XMMWORD PTR [rsp+0x50],xmm5
  0x0000000002a716d8: vmovdqu XMMWORD PTR [rsp+0x60],xmm6
  0x0000000002a716de: vmovdqu XMMWORD PTR [rsp+0x70],xmm7
  0x0000000002a716e4: vmovdqu XMMWORD PTR [rsp+0x80],xmm8
  0x0000000002a716ed: vmovdqu XMMWORD PTR [rsp+0x90],xmm9
  0x0000000002a716f6: vmovdqu XMMWORD PTR [rsp+0xa0],xmm10
  0x0000000002a716ff: vmovdqu XMMWORD PTR [rsp+0xb0],xmm11
  0x0000000002a71708: vmovdqu XMMWORD PTR [rsp+0xc0],xmm12
  0x0000000002a71711: vmovdqu XMMWORD PTR [rsp+0xd0],xmm13
  0x0000000002a7171a: vmovdqu XMMWORD PTR [rsp+0xe0],xmm14
  0x0000000002a71723: vmovdqu XMMWORD PTR [rsp+0xf0],xmm15
  0x0000000002a7172c: sub    rsp,0x10
  0x0000000002a71730: fstp   QWORD PTR [rsp]
  0x0000000002a71733: fstp   QWORD PTR [rsp+0x8]
  0x0000000002a71737: vmovsd xmm0,QWORD PTR [rsp]
  0x0000000002a7173c: vmovsd xmm1,QWORD PTR [rsp+0x8]
  0x0000000002a71742: sub    rsp,0x20
  0x0000000002a71746: test   esp,0xf
  0x0000000002a7174c: je     0x0000000002a71764
  0x0000000002a71752: sub    rsp,0x8
  0x0000000002a71756: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a7175b: add    rsp,0x8
  0x0000000002a7175f: jmp    0x0000000002a71769
  0x0000000002a71764: call   0x000000006bf240d0  ;   {runtime_call}
  0x0000000002a71769: add    rsp,0x20
  0x0000000002a7176d: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a71772: fld    QWORD PTR [rsp]
  0x0000000002a71775: add    rsp,0x10
  0x0000000002a71779: vmovdqu xmm0,XMMWORD PTR [rsp]
  0x0000000002a7177e: vmovdqu xmm1,XMMWORD PTR [rsp+0x10]
  0x0000000002a71784: vmovdqu xmm2,XMMWORD PTR [rsp+0x20]
  0x0000000002a7178a: vmovdqu xmm3,XMMWORD PTR [rsp+0x30]
  0x0000000002a71790: vmovdqu xmm4,XMMWORD PTR [rsp+0x40]
  0x0000000002a71796: vmovdqu xmm5,XMMWORD PTR [rsp+0x50]
  0x0000000002a7179c: vmovdqu xmm6,XMMWORD PTR [rsp+0x60]
  0x0000000002a717a2: vmovdqu xmm7,XMMWORD PTR [rsp+0x70]
  0x0000000002a717a8: vmovdqu xmm8,XMMWORD PTR [rsp+0x80]
  0x0000000002a717b1: vmovdqu xmm9,XMMWORD PTR [rsp+0x90]
  0x0000000002a717ba: vmovdqu xmm10,XMMWORD PTR [rsp+0xa0]
  0x0000000002a717c3: vmovdqu xmm11,XMMWORD PTR [rsp+0xb0]
  0x0000000002a717cc: vmovdqu xmm12,XMMWORD PTR [rsp+0xc0]
  0x0000000002a717d5: vmovdqu xmm13,XMMWORD PTR [rsp+0xd0]
  0x0000000002a717de: vmovdqu xmm14,XMMWORD PTR [rsp+0xe0]
  0x0000000002a717e7: vmovdqu xmm15,XMMWORD PTR [rsp+0xf0]
  0x0000000002a717f0: add    rsp,0x100
  0x0000000002a717f7: vinsertf128 ymm0,ymm0,XMMWORD PTR [rsp],0x1
  0x0000000002a717fe: vinsertf128 ymm1,ymm1,XMMWORD PTR [rsp+0x10],0x1
  0x0000000002a71806: vinsertf128 ymm2,ymm2,XMMWORD PTR [rsp+0x20],0x1
  0x0000000002a7180e: vinsertf128 ymm3,ymm3,XMMWORD PTR [rsp+0x30],0x1
  0x0000000002a71816: vinsertf128 ymm4,ymm4,XMMWORD PTR [rsp+0x40],0x1
  0x0000000002a7181e: vinsertf128 ymm5,ymm5,XMMWORD PTR [rsp+0x50],0x1
  0x0000000002a71826: vinsertf128 ymm6,ymm6,XMMWORD PTR [rsp+0x60],0x1
  0x0000000002a7182e: vinsertf128 ymm7,ymm7,XMMWORD PTR [rsp+0x70],0x1
  0x0000000002a71836: vinsertf128 ymm8,ymm8,XMMWORD PTR [rsp+0x80],0x1
  0x0000000002a71841: vinsertf128 ymm9,ymm9,XMMWORD PTR [rsp+0x90],0x1
  0x0000000002a7184c: vinsertf128 ymm10,ymm10,XMMWORD PTR [rsp+0xa0],0x1
  0x0000000002a71857: vinsertf128 ymm11,ymm11,XMMWORD PTR [rsp+0xb0],0x1
  0x0000000002a71862: vinsertf128 ymm12,ymm12,XMMWORD PTR [rsp+0xc0],0x1
  0x0000000002a7186d: vinsertf128 ymm13,ymm13,XMMWORD PTR [rsp+0xd0],0x1
  0x0000000002a71878: vinsertf128 ymm14,ymm14,XMMWORD PTR [rsp+0xe0],0x1
  0x0000000002a71883: vinsertf128 ymm15,ymm15,XMMWORD PTR [rsp+0xf0],0x1
  0x0000000002a7188e: add    rsp,0x100
  0x0000000002a71895: mov    r15,QWORD PTR [rsp]
  0x0000000002a71899: mov    r14,QWORD PTR [rsp+0x8]
  0x0000000002a7189e: mov    r13,QWORD PTR [rsp+0x10]
  0x0000000002a718a3: mov    r12,QWORD PTR [rsp+0x18]
  0x0000000002a718a8: mov    r11,QWORD PTR [rsp+0x20]
  0x0000000002a718ad: mov    r10,QWORD PTR [rsp+0x28]
  0x0000000002a718b2: mov    r9,QWORD PTR [rsp+0x30]
  0x0000000002a718b7: mov    r8,QWORD PTR [rsp+0x38]
  0x0000000002a718bc: mov    rdi,QWORD PTR [rsp+0x40]
  0x0000000002a718c1: mov    rsi,QWORD PTR [rsp+0x48]
  0x0000000002a718c6: mov    rbp,QWORD PTR [rsp+0x50]
  0x0000000002a718cb: mov    rbx,QWORD PTR [rsp+0x60]
  0x0000000002a718d0: mov    rdx,QWORD PTR [rsp+0x68]
  0x0000000002a718d5: mov    rcx,QWORD PTR [rsp+0x70]
  0x0000000002a718da: mov    rax,QWORD PTR [rsp+0x78]
  0x0000000002a718df: add    rsp,0x80
  0x0000000002a718e6: fstp   QWORD PTR [rsp]
  0x0000000002a718e9: vmovsd xmm0,QWORD PTR [rsp]  ;*invokestatic pow
                                                ; - ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark::trickyMathOctaPow@16 (line 63)


Однако, в реализации интринсика _dpow есть интересная особенность, а именно обработка «специального случая». Ниже фрагмент исходников library_call.cpp OpenJDK 8: 

//------------------------------inline_pow-------------------------------------
// Inline power instructions, if possible.
bool LibraryCallKit::inline_pow() {
  // Pseudocode for pow
  // if (y == 2) {
  //   return x * x;
  // } else {
  //   if (x <= 0.0) {
  //     long longy = (long)y;
  //     if ((double)longy == y) { // if y is long
  //       if (y + 1 == y) longy = 0; // huge number: even
  //       result = ((1&longy) == 0)?-DPow(abs(x), y):DPow(abs(x), y);
  //     } else {
  //       result = NaN;
  //     }
  //   } else {
  //     result = DPow(x,y);
  //   }
  //   if (result != result)?  {
  //     result = uncommon_trap() or runtime_call();
  //   }
  //   return result;
  // }

/* code omitted */

}


Разработчики HotSpot позаботились об одном частном случае — возведении числа в квадрат. Благодаря этому подставляемый JIT-компилятором код выполнит только x * x. Найдём эту проверку в дизассемблированном коде на примере первого вызова Math.pow(a, 2): 

  0x0000000002a70b14: vmovsd xmm1,QWORD PTR [rip+0xffffffffffffff44] ; в xmm1 поместили константу 2.0 
  0x0000000002a70b1c: vmovsd QWORD PTR [rsp],xmm1
  0x0000000002a70b21: fld    QWORD PTR [rsp] ; поместили значение 2.0 на FPU register stack
  0x0000000002a70b24: vmovsd QWORD PTR [rsp],xmm0
  0x0000000002a70b29: fld    QWORD PTR [rsp] ; поместили значение a на FPU register stack
  0x0000000002a70b2c: movabs rax,0x6c4ba7d0
  0x0000000002a70b36: fld    QWORD PTR [rax] ; поместили значение 2.0 на FPU register stack 
  0x0000000002a70b38: fucomip st,st(2) ; сравнили 2.0 и 2.0
  0x0000000002a70b3a: jp     0x0000000002a70b53
  0x0000000002a70b40: jne    0x0000000002a70b53
  0x0000000002a70b46: fxch   st(1) ; вытолкнули на вершину FPU стека значение a
  0x0000000002a70b48: ffree  st(0)
  0x0000000002a70b4a: fincstp 
  0x0000000002a70b4c: fmul   st,st(0) ; перемножили a на a
  0x0000000002a70b4e: jmp    0x0000000002a70faa
  ; code omitted
  0x0000000002a70faa: fstp   QWORD PTR [rsp]
  0x0000000002a70fad: vmovsd xmm0,QWORD PTR [rsp]  ; в xmm0 поместили значение a * a
  ; code omitted


Бенчмарки


Код бенчмарка: 

@Fork(value = 3, warmups = 0)
@Warmup(iterations = 5, time = 1_000, timeUnit = TimeUnit.MILLISECONDS)
@Measurement(iterations = 10, time = 1_000, timeUnit = TimeUnit.MILLISECONDS)
@OutputTimeUnit(value = TimeUnit.NANOSECONDS)
@BenchmarkMode(Mode.AverageTime)
@State(Scope.Benchmark)
public class MathBenchmark {
    public double a;

    @Setup
    public void setup() {
        a = 1234567.890;
    }

    @Benchmark
    public void mathOctaPowBenchmark(Blackhole bh) {
        bh.consume(mathOctaPow(a));
    }

    @Benchmark
    public void plainOctaPowBenchmark(Blackhole bh) {
        bh.consume(plainOctaPow(a));
    }

    @Benchmark
    public void trickyMathOctaPowBenchmark(Blackhole bh) {
        bh.consume(trickyMathOctaPow(a));
    }

    @Benchmark
    public void trickyPlainOctaPowBenchmark(Blackhole bh) {
        bh.consume(trickyPlainOctaPow(a));
    }

    public double mathOctaPow(double a) {
        return Math.pow(a, 8);
    }

    public double plainOctaPow(double a) {
        return a * a * a * a * a * a * a * a;
    }

    public double trickyMathOctaPow(double a) {
        return Math.pow(Math.pow(Math.pow(a, 2), 2), 2);
    }

    public double trickyPlainOctaPow(double a) {
        a *= a; a *= a; return a * a;
    }
}


Результаты: 

Benchmark                                  Mode  Cnt   Score   Error  Units
MathBenchmark.mathOctaPowBenchmark         avgt   30  76,041 ± 0,428  ns/op
MathBenchmark.plainOctaPowBenchmark        avgt   30   4,174 ± 0,027  ns/op
MathBenchmark.trickyMathOctaPowBenchmark   avgt   30   3,010 ± 0,014  ns/op
MathBenchmark.trickyPlainOctaPowBenchmark  avgt   30   3,011 ± 0,015  ns/op


Результаты бенчмарка целиком
# JMH version: 1.20
# VM version: JDK 1.8.0_161, VM 25.161-b12
# VM invoker: C:\Program Files\Java\jre1.8.0_161\bin\java.exe
# VM options: 
# Warmup: 5 iterations, 1000 ms each
# Measurement: 10 iterations, 1000 ms each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Average time, time/op
# Benchmark: ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark.mathOctaPowBenchmark

# Run progress: 0,00% complete, ETA 00:03:00
# Fork: 1 of 3
# Warmup Iteration   1: 77,026 ns/op
# Warmup Iteration   2: 76,561 ns/op
# Warmup Iteration   3: 77,623 ns/op
# Warmup Iteration   4: 76,192 ns/op
# Warmup Iteration   5: 76,012 ns/op
Iteration   1: 75,947 ns/op
Iteration   2: 75,739 ns/op
Iteration   3: 75,864 ns/op
Iteration   4: 76,179 ns/op
Iteration   5: 75,934 ns/op
Iteration   6: 75,783 ns/op
Iteration   7: 75,820 ns/op
Iteration   8: 75,898 ns/op
Iteration   9: 75,798 ns/op
Iteration  10: 76,053 ns/op

# Run progress: 8,33% complete, ETA 00:02:48
# Fork: 2 of 3
# Warmup Iteration   1: 75,975 ns/op
# Warmup Iteration   2: 76,008 ns/op
# Warmup Iteration   3: 75,867 ns/op
# Warmup Iteration   4: 76,061 ns/op
# Warmup Iteration   5: 75,710 ns/op
Iteration   1: 75,874 ns/op
Iteration   2: 75,862 ns/op
Iteration   3: 76,080 ns/op
Iteration   4: 75,948 ns/op
Iteration   5: 75,848 ns/op
Iteration   6: 75,883 ns/op
Iteration   7: 76,004 ns/op
Iteration   8: 75,790 ns/op
Iteration   9: 75,894 ns/op
Iteration  10: 75,847 ns/op

# Run progress: 16,67% complete, ETA 00:02:33
# Fork: 3 of 3
# Warmup Iteration   1: 75,778 ns/op
# Warmup Iteration   2: 75,850 ns/op
# Warmup Iteration   3: 75,878 ns/op
# Warmup Iteration   4: 76,025 ns/op
# Warmup Iteration   5: 76,450 ns/op
Iteration   1: 75,791 ns/op
Iteration   2: 75,941 ns/op
Iteration   3: 75,652 ns/op
Iteration   4: 75,795 ns/op
Iteration   5: 75,906 ns/op
Iteration   6: 78,971 ns/op
Iteration   7: 76,055 ns/op
Iteration   8: 75,736 ns/op
Iteration   9: 75,816 ns/op
Iteration  10: 77,537 ns/op


Result "ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark.mathOctaPowBenchmark":
  76,041 ±(99.9%) 0,428 ns/op [Average]
  (min, avg, max) = (75,652, 76,041, 78,971), stdev = 0,640
  CI (99.9%): [75,614, 76,469] (assumes normal distribution)


# JMH version: 1.20
# VM version: JDK 1.8.0_161, VM 25.161-b12
# VM invoker: C:\Program Files\Java\jre1.8.0_161\bin\java.exe
# VM options: 
# Warmup: 5 iterations, 1000 ms each
# Measurement: 10 iterations, 1000 ms each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Average time, time/op
# Benchmark: ru.gnkoshelev.jbreak2018.perf_tests.pow.MathBenchmark.plainOctaPowBenchmark

# Run progress: 25,00% complete, ETA 00:02:17
# Fork: 1 of 3
# Warmup Iteration   1: 4,622 ns/op
# Warmup Iteration   2: 4,406 ns/op
# Warmup Iteration   3: 4,169 ns/op
# Warmup Iteration   4: 4,163 ns/op
# Warmup Iteration   5: 4,153 ns/op
Iteration   1: 4,141 ns/op
Iteration   2: 4,144 ns/op
Iteration   3: 4,141 ns/op
Iteration   4: 4,141 ns/op
Iteration   5: 4,149 ns/op
Iteration   6: 4,136 ns/op
Iteration   7: 4,143 ns/op
Iteration   8: 4,136 ns/op
Iter
    
            
© Habrahabr.ru