Xbyak 5.751 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++

Abstract

This is a header file which enables dynamically to assemble x86(IA32), x64(AMD64, x86-64) mnemonic.

Feature

• header file only
• Intel/MASM like syntax
• fully support AVX-512

Note: Xbyak uses and(), or(), xor(), not() functions, so -fno-operator-names option is necessary for gcc/clang.

Or define XBYAK_NO_OP_NAMES before including xbyak.h and use and_(), or_(), xor_(), not_() instead of them.

and_(), or_(), xor_(), not_() are always available.

XBYAK_NO_OP_NAMES will be defined in the feature version.

Supported OS

• Windows Xp, Vista, Windows 7, Windows 10(32bit, 64bit)
• Linux(32bit, 64bit)
• Intel macOS

Supported Compilers

Almost C++03 or later compilers for x86/x64 such as Visual Studio, g++, clang++, Intel C++ compiler and g++ on mingw/cygwin.

Install

The following files are necessary. Please add the path to your compile directory.

• xbyak.h
• xbyak_mnemonic.h
• xbyak_util.h

Linux:

make install

These files are copied into /usr/local/include/xbyak.

How to use it

Inherit Xbyak::CodeGenerator class and make the class method.

#define XBYAK_NO_OP_NAMES
#include <xbyak/xbyak.h>

struct Code : Xbyak::CodeGenerator {
    Code(int x)
    {
        mov(eax, x);
        ret();
    }
};

Make an instance of the class and get the function pointer by calling getCode() and call it.

Code c(5);
int (*f)() = c.getCode<int (*)()>();
printf("ret=%d\n", f()); // ret = 5

Syntax

Similar to MASM/NASM syntax with parentheses.

NASM              Xbyak
mov eax, ebx  --> mov(eax, ebx);
inc ecx           inc(ecx);
ret           --> ret();

Addressing

Use qword, dword, word and byte if it is necessary to specify the size of memory, otherwise use ptr.

(ptr|qword|dword|word|byte) [base + index * (1|2|4|8) + displacement]
                            [rip + 32bit disp] ; x64 only

NASM                   Xbyak
mov eax, [ebx+ecx] --> mov(eax, ptr [ebx+ecx]);
mov al, [ebx+ecx]  --> mov(al, ptr [ebx + ecx]);
test byte [esp], 4 --> test(byte [esp], 4);
inc qword [rax]    --> inc(qword [rax]);

Note: qword, ... are member variables, then don't use dword as unsigned int type.

How to use Selector (Segment Register)

mov eax, [fs:eax] --> putSeg(fs);
                      mov(eax, ptr [eax]);
mov ax, cs        --> mov(ax, cs);

Note: Segment class is not derived from Operand.

AVX

vaddps(xmm1, xmm2, xmm3); // xmm1 <- xmm2 + xmm3
vaddps(xmm2, xmm3, ptr [rax]); // use ptr to access memory
vgatherdpd(xmm1, ptr [ebp + 256 + xmm2*4], xmm3);

Note: If XBYAK_ENABLE_OMITTED_OPERAND is defined, then you can use two operand version for backward compatibility. But the newer version will not support it.

vaddps(xmm2, xmm3); // xmm2 <- xmm2 + xmm3

AVX-512

vaddpd zmm2, zmm5, zmm30                --> vaddpd(zmm2, zmm5, zmm30);
vaddpd xmm30, xmm20, [rax]              --> vaddpd(xmm30, xmm20, ptr [rax]);
vaddps xmm30, xmm20, [rax]              --> vaddps(xmm30, xmm20, ptr [rax]);
vaddpd zmm2{k5}, zmm4, zmm2             --> vaddpd(zmm2 | k5, zmm4, zmm2);
vaddpd zmm2{k5}{z}, zmm4, zmm2          --> vaddpd(zmm2 | k5 | T_z, zmm4, zmm2);
vaddpd zmm2{k5}{z}, zmm4, zmm2,{rd-sae} --> vaddpd(zmm2 | k5 | T_z, zmm4, zmm2 | T_rd_sae);
                                            vaddpd(zmm2 | k5 | T_z | T_rd_sae, zmm4, zmm2); // the position of `|` is arbitrary.
vcmppd k4{k3}, zmm1, zmm2, {sae}, 5     --> vcmppd(k4 | k3, zmm1, zmm2 | T_sae, 5);

vaddpd xmm1, xmm2, [rax+256]            --> vaddpd(xmm1, xmm2, ptr [rax+256]);
vaddpd xmm1, xmm2, [rax+256]{1to2}      --> vaddpd(xmm1, xmm2, ptr_b [rax+256]);
vaddpd ymm1, ymm2, [rax+256]{1to4}      --> vaddpd(ymm1, ymm2, ptr_b [rax+256]);
vaddpd zmm1, zmm2, [rax+256]{1to8}      --> vaddpd(zmm1, zmm2, ptr_b [rax+256]);
vaddps zmm1, zmm2, [rax+rcx*8+8]{1to16} --> vaddps(zmm1, zmm2, ptr_b [rax+rcx*8+8]);
vmovsd [rax]{k1}, xmm4                  --> vmovsd(ptr [rax] | k1, xmm4);

vcvtpd2dq xmm16, oword [eax+33]         --> vcvtpd2dq(xmm16, xword [eax+33]); // use xword for m128 instead of oword
                                            vcvtpd2dq(xmm16, ptr [eax+33]); // default xword
vcvtpd2dq xmm21, [eax+32]{1to2}         --> vcvtpd2dq(xmm21, ptr_b [eax+32]);
vcvtpd2dq xmm0, yword [eax+33]          --> vcvtpd2dq(xmm0, yword [eax+33]); // use yword for m256
vcvtpd2dq xmm19, [eax+32]{1to4}         --> vcvtpd2dq(xmm19, yword_b [eax+32]); // use yword_b to broadcast

vfpclassps k5{k3}, zword [rax+64], 5    --> vfpclassps(k5|k3, zword [rax+64], 5); // specify m512
vfpclasspd k5{k3}, [rax+64]{1to2}, 5    --> vfpclasspd(k5|k3, xword_b [rax+64], 5); // broadcast 64-bit to 128-bit
vfpclassps k5{k3}, [rax+64]{1to4}, 5    --> vfpclassps(k5|k3, yword_b [rax+64], 5); // broadcast 64-bit to 256-bit

Remark

• k1, ..., k7 are opmask registers.
• use | T_z, | T_sae, | T_rn_sae, | T_rd_sae, | T_ru_sae, | T_rz_sae instead of ,{z}, ,{sae}, ,{rn-sae}, ,{rd-sae}, ,{ru-sae}, ,{rz-sae} respectively.
• k4 | k3 is different from k3 | k4.
• use ptr_b for broadcast {1toX}. X is automatically determined.
• specify xword/yword/zword(_b) for m128/m256/m512 if necessary.

Label

Two kinds of Label are supported. (String literal and Label class).

String literal

L("L1");
  jmp("L1");

  jmp("L2");
  ...
  a few mnemonics (8-bit displacement jmp)
  ...
L("L2");

  jmp("L3", T_NEAR);
  ...
  a lot of mnemonics (32-bit displacement jmp)
  ...
L("L3");

• Call hasUndefinedLabel() to verify your code has no undefined label.
• you can use a label for immediate value of mov like as mov(eax, "L2").

Support @@, @f, @b like MASM

L("@@"); // <A>
  jmp("@b"); // jmp to <A>
  jmp("@f"); // jmp to <B>
L("@@"); // <B>
  jmp("@b"); // jmp to <B>
  mov(eax, "@b");
  jmp(eax); // jmp to <B>

Local label

Label symbols beginning with a period between inLocalLabel() and outLocalLabel() are treated as a local label. inLocalLabel() and outLocalLabel() can be nested.

void func1()
{
    inLocalLabel();
  L(".lp"); // <A> ; local label
    ...
    jmp(".lp"); // jmp to <A>
  L("aaa"); // global label <C>
    outLocalLabel();

    inLocalLabel();
  L(".lp"); // <B> ; local label
    func1();
    jmp(".lp"); // jmp to <B>
    inLocalLabel();
    jmp("aaa"); // jmp to <C>
}

Label class

L() and jxx() support Label class.

  Xbyak::Label label1, label2;
L(label1);
  ...
  jmp(label1);
  ...
  jmp(label2);
  ...
L(label2);

Use putL for jmp table

    Label labelTbl, L0, L1, L2;
    mov(rax, labelTbl);
    // rdx is an index of jump table
    jmp(ptr [rax + rdx * sizeof(void*)]);
L(labelTbl);
    putL(L0);
    putL(L1);
    putL(L2);
L(L0);
    ....
L(L1);
    ....

assignL(dstLabel, srcLabel) binds dstLabel with srcLabel.

  Label label2;
  Label label1 = L(); // make label1 ; same to Label label1; L(label1);
  ...
  jmp(label2); // label2 is not determined here
  ...
  assignL(label2, label1); // label2 <- label1

The jmp in the above code jumps to label1 assigned by assignL.

Note:

• srcLabel must be used in L().
• dstLabel must not be used in L().

Label::getAddress() returns the address specified by the label instance and 0 if not specified.

// not AutoGrow mode
Label  label;
assert(label.getAddress() == 0);
L(label);
assert(label.getAddress() == getCurr());

Rip ; relative addressing

Label label;
mov(eax, ptr [rip + label]); // eax = 4
...

L(label);
dd(4);

int x;
...
  mov(eax, ptr[rip + &x]); // throw exception if the difference between &x and current position is larger than 2GiB

Code size

The default max code size is 4096 bytes. Specify the size in constructor of CodeGenerator() if necessary.

class Quantize : public Xbyak::CodeGenerator {
public:
  Quantize()
    : CodeGenerator(8192)
  {
  }
  ...
};

User allocated memory

You can make jit code on prepaired memory.

Call setProtectModeRE yourself to change memory mode if using the prepaired memory.

uint8_t alignas(4096) buf[8192]; // C++11 or later

struct Code : Xbyak::CodeGenerator {
    Code() : Xbyak::CodeGenerator(sizeof(buf), buf)
    {
        mov(rax, 123);
        ret();
    }
};

int main()
{
    Code c;
    c.setProtectModeRE(); // set memory to Read/Exec
    printf("%d\n", c.getCode<int(*)()>()());
}

Note: See sample/test0.cpp.

AutoGrow

The memory region for jit is automatically extended if necessary when AutoGrow is specified in a constructor of CodeGenerator.

Call ready() or readyRE() before calling getCode() to fix jump address.

struct Code : Xbyak::CodeGenerator {
  Code()
    : Xbyak::CodeGenerator(<default memory size>, Xbyak::AutoGrow)
  {
     ...
  }
};
Code c;
// generate code for jit
c.ready(); // mode = Read/Write/Exec

Note:

• Don't use the address returned by getCurr() before calling ready() because it may be invalid address.

Read/Exec mode

Xbyak set Read/Write/Exec mode to memory to run jit code. If you want to use Read/Exec mode for security, then specify DontSetProtectRWE for CodeGenerator and call setProtectModeRE() after generating jit code.

struct Code : Xbyak::CodeGenerator {
    Code()
        : Xbyak::CodeGenerator(4096, Xbyak::DontSetProtectRWE)
    {
        mov(eax, 123);
        ret();
    }
};

Code c;
c.setProtectModeRE();
...

Call readyRE() instead of ready() when using AutoGrow mode. See protect-re.cpp.

Macro

• XBYAK32 is defined on 32bit.
• XBYAK64 is defined on 64bit.
• XBYAK64_WIN is defined on 64bit Windows(VC)
• XBYAK64_GCC is defined on 64bit gcc, cygwin
• define XBYAK_NO_OP_NAMES on gcc without -fno-operator-names
• define XBYAK_ENABLE_OMITTED_OPERAND if you use omitted destination such as vaddps(xmm2, xmm3);(deprecated in the future)
• define XBYAK_UNDEF_JNL if Bessel function jnl is defined as macro

Sample

• test0.cpp ; tiny sample (x86, x64)
• quantize.cpp ; JIT optimized quantization by fast division (x86 only)
• calc.cpp ; assemble and estimate a given polynomial (x86, x64)
• bf.cpp ; JIT brainfuck (x86, x64)

License

modified new BSD License http://opensource.org/licenses/BSD-3-Clause

History

• 2018/Oct/31 ver 5.751 recover Xbyak::CastTo for compatibility
• 2018/Oct/29 ver 5.75 unlink LabelManager from Label when msg is destroyed
• 2018/Oct/21 ver 5.74 support RegRip +/- int. Xbyak::CastTo is removed
• 2018/Oct/15 util::AddressFrame uses push/pop instead of mov
• 2018/Sep/19 ver 5.73 fix evex encoding of vpslld, vpslldq, vpsllw, etc for (reg, mem, imm8)
• 2018/Sep/19 ver 5.72 fix the encoding of vinsertps for disp8N(Thanks to petercaday)
• 2018/Sep/04 ver 5.71 L() returns a new label instance
• 2018/Aug/27 ver 5.70 support setProtectMode() and DontUseProtect for read/exec setting
• 2018/Aug/24 ver 5.68 fix wrong VSIB encoding with vector index >= 16(thanks to petercaday)
• 2018/Aug/14 ver 5.67 remove mutable in Address ; fix setCacheHierarchy for cloud vm
• 2018/Jul/26 ver 5.661 support mingw64
• 2018/Jul/24 ver 5.66 add CodeArray::PROTECT_RE to mode of protect()
• 2018/Jun/26 ver 5.65 fix push(qword [mem])
• 2018/Mar/07 ver 5.64 fix zero division in Cpu() on some cpu
• 2018/Feb/14 ver 5.63 fix Cpu::setCacheHierarchy() and fix EvexModifierZero for clang<3.9(thanks to mgouicem)
• 2018/Feb/13 ver 5.62 Cpu::setCacheHierarchy() by mgouicem and rsdubtso
• 2018/Feb/07 ver 5.61 vmov* supports mem{k}{z}(I forgot it)
• 2018/Jan/24 ver 5.601 add xword, yword, etc. into Xbyak::util namespace
• 2018/Jan/05 ver 5.60 support AVX-512 for Ice lake(319433-030.pdf)
• 2017/Aug/22 ver 5.53 fix mpx encoding, add bnd() prefix
• 2017/Aug/18 ver 5.52 fix align (thanks to MerryMage)
• 2017/Aug/17 ver 5.51 add multi-byte nop and align() uses it(thanks to inolen)
• 2017/Aug/08 ver 5.50 add mpx(thanks to magurosan)
• 2017/Aug/08 ver 5.45 add sha(thanks to magurosan)
• 2017/Aug/08 ver 5.44 add prefetchw(thanks to rsdubtso)
• 2017/Jul/12 ver 5.432 reduce warnings of PVS studio
• 2017/Jul/09 ver 5.431 fix hasRex() (no affect) (thanks to drillsar)
• 2017/May/14 ver 5.43 fix CodeGenerator::resetSize() (thanks to gibbed)
• 2017/May/13 ver 5.42 add movs{b,w,d,q}
• 2017/Jan/26 ver 5.41 add prefetchwt1 and support for scale == 0(thanks to rsdubtso)
• 2016/Dec/14 ver 5.40 add Label::getAddress() method to get the pointer specified by the label
• 2016/Dec/09 ver 5.34 fix handling of negative offsets when encoding disp8N(thanks to rsdubtso)
• 2016/Dec/08 ver 5.33 fix encoding of vpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w} for disp8N
• 2016/Dec/01 ver 5.32 rename __xgetbv() to _xgetbv() to support clang for Visual Studio(thanks to freiro)
• 2016/Nov/27 ver 5.31 rename AVX512_4VNNI to AVX512_4VNNIW
• 2016/Nov/27 ver 5.30 add AVX512_4VNNI, AVX512_4FMAPS instructions(thanks to rsdubtso)
• 2016/Nov/26 ver 5.20 add detection of AVX512_4VNNI and AVX512_4FMAPS(thanks to rsdubtso)
• 2016/Nov/20 ver 5.11 lost vptest for ymm(thanks to gregory38)
• 2016/Nov/20 ver 5.10 add addressing [rip+&var]
• 2016/Sep/29 ver 5.03 fix detection ERR_INVALID_OPMASK_WITH_MEMORY(thanks to PVS-Studio)
• 2016/Aug/15 ver 5.02 xbyak does not include xbyak_bin2hex.h
• 2016/Aug/15 ver 5.011 fix detection of version of gcc 5.4
• 2016/Aug/03 ver 5.01 disable omitted operand
• 2016/Jun/24 ver 5.00 support avx-512 instruction set
• 2016/Jun/13 avx-512 add mask instructions
• 2016/May/05 ver 4.91 add detection of AVX-512 to Xbyak::util::Cpu
• 2016/Mar/14 ver 4.901 comment to ready() function(thanks to skmp)
• 2016/Feb/04 ver 4.90 add jcc(const void *addr);
• 2016/Jan/30 ver 4.89 vpblendvb supports ymm reg(thanks to John Funnell)
• 2016/Jan/24 ver 4.88 lea, cmov supports 16-bit register(thanks to whyisthisfieldhere)
• 2015/Oct/05 ver 4.87 support segment selectors
• 2015/Aug/18 ver 4.86 fix [rip + label] addressing with immediate value(thanks to whyisthisfieldhere)
• 2015/Aug/10 ver 4.85 Address::operator==() is not correct(thanks to inolen)
• 2015/Jun/22 ver 4.84 call() support variadic template if available(thanks to randomstuff)
• 2015/Jun/16 ver 4.83 support movbe(thanks to benvanik)
• 2015/May/24 ver 4.82 support detection of F16C
• 2015/Apr/25 ver 4.81 fix the condition to throw exception for setSize(thanks to whyisthisfieldhere)
• 2015/Apr/22 ver 4.80 rip supports label(thanks to whyisthisfieldhere)
• 2015/Jar/28 ver 4.71 support adcx, adox, cmpxchg, rdseed, stac
• 2014/Oct/14 ver 4.70 support MmapAllocator
• 2014/Jun/13 ver 4.62 disable warning of VC2014
• 2014/May/30 ver 4.61 support bt, bts, btr, btc
• 2014/May/28 ver 4.60 support vcvtph2ps, vcvtps2ph
• 2014/Apr/11 ver 4.52 add detection of rdrand
• 2014/Mar/25 ver 4.51 remove state information of unreferenced labels
• 2014/Mar/16 ver 4.50 support new Label
• 2014/Mar/05 ver 4.40 fix wrong detection of BMI/enhanced rep on VirtualBox
• 2013/Dec/03 ver 4.30 support Reg::cvt8(), cvt16(), cvt32(), cvt64()
• 2013/Oct/16 ver 4.21 label support std::string
• 2013/Jul/30 ver 4.20 [break backward compatibility] split Reg32e class into RegExp(base+index*scale+disp) and Reg32e(means Reg32 or Reg64)
• 2013/Jul/04 ver 4.10 [break backward compatibility] change the type of Xbyak::Error from enum to a class
• 2013/Jun/21 ver 4.02 add putL(LABEL) function to put the address of the label
• 2013/Jun/21 ver 4.01 vpsllw, vpslld, vpsllq, vpsraw, vpsrad, vpsrlw, vpsrld, vpsrlq support (ymm, ymm, xmm). support vpbroadcastb, vpbroadcastw, vpbroadcastd, vpbroadcastq(thanks to Gabest).
• 2013/May/30 ver 4.00 support AVX2, VEX-encoded GPR-instructions
• 2013/Mar/27 ver 3.80 support mov(reg, "label");
• 2013/Mar/13 ver 3.76 add cqo(), jcxz(), jecxz(), jrcxz()
• 2013/Jan/15 ver 3.75 add setSize() to modify generated code
• 2013/Jan/12 ver 3.74 add CodeGenerator::reset() ; add Allocator::useProtect()
• 2013/Jan/06 ver 3.73 use unordered_map if possible
• 2012/Dec/04 ver 3.72 eax, ebx, ... are member variables of CodeGenerator(revert), Xbyak::util::eax, ... are static const.
• 2012/Nov/17 ver 3.71 and_(), or_(), xor_(), not_() are available if XBYAK_NO_OP_NAMES is not defined.
• 2012/Nov/17 change eax, ebx, ptr and so on in CodeGenerator as static member and alias of them are defined in Xbyak::util.
• 2012/Nov/09 ver 3.70 XBYAK_NO_OP_NAMES macro is added to use and_() instead of and() (thanks to Mattias)
• 2012/Nov/01 ver 3.62 add fwait/fnwait/finit/fninit
• 2012/Nov/01 ver 3.61 add fldcw/fstcw
• 2012/May/03 ver 3.60 change interface of Allocator
• 2012/Mar/23 ver 3.51 fix userPtr mode
• 2012/Mar/19 ver 3.50 support AutoGrow mode
• 2011/Nov/09 ver 3.05 fix bit property of rip addresing / support movsxd
• 2011/Aug/15 ver 3.04 fix dealing with imm8 such as add(dword [ebp-8], 0xda); (thanks to lolcat)
• 2011/Jun/16 ver 3.03 fix __GNUC_PREREQ macro for Mac gcc(thanks to t_teruya)
• 2011/Apr/28 ver 3.02 do not use xgetbv on Mac gcc
• 2011/May/24 ver 3.01 fix typo of OSXSAVE
• 2011/May/23 ver 3.00 add vcmpeqps and so on
• 2011/Feb/16 ver 2.994 beta add vmovq for 32-bit mode(I forgot it)
• 2011/Feb/16 ver 2.993 beta remove cvtReg to avoid thread unsafe
• 2011/Feb/10 ver 2.992 beta support one argument syntax for fadd like nasm
• 2011/Feb/07 ver 2.991 beta fix pextrw reg, xmm, imm(Thanks to Gabest)
• 2011/Feb/04 ver 2.99 beta support AVX
• 2010/Dec/08 ver 2.31 fix ptr [rip + 32bit offset], support rdtscp
• 2010/Oct/19 ver 2.30 support pclmulqdq, aesdec, aesdeclast, aesenc, aesenclast, aesimc, aeskeygenassist
• 2010/Jun/07 ver 2.29 fix call()
• 2010/Jun/17 ver 2.28 move some member functions to public
• 2010/Jun/01 ver 2.27 support encoding of mov(reg64, imm) like yasm(not nasm)
• 2010/May/24 ver 2.26 fix sub(rsp, 1000)
• 2010/Apr/26 ver 2.25 add jc/jnc(I forgot to implement them...)
• 2010/Apr/16 ver 2.24 change the prototype of rewrite() method
• 2010/Apr/15 ver 2.23 fix align() and xbyak_util.h for Mac
• 2010/Feb/16 ver 2.22 fix inLocalLabel()/outLocalLabel()
• 2009/Dec/09 ver 2.21 support cygwin(gcc 4.3.2)
• 2009/Nov/28 support a part of FPU
• 2009/Jun/25 fix mov(qword[rax], imm); (thanks to Martin)
• 2009/Mar/10 fix redundant REX.W prefix on jmp/call reg64
• 2009/Feb/24 add movq reg64, mmx/xmm; movq mmx/xmm, reg64
• 2009/Feb/13 movd(xmm7, dword[eax]) drops 0x66 prefix (thanks to Gabest)
• 2008/Dec/30 fix call in short relative address(thanks to kato san)
• 2008/Sep/18 support @@, @f, @b and localization of label(thanks to nobu-q san)
• 2008/Sep/18 support (ptr[rip + 32bit offset]) (thanks to Dango-Chu san)
• 2008/Jun/03 fix align(). mov(ptr[eax],1) throws ERR_MEM_SIZE_IS_NOT_SPECIFIED.
• 2008/Jun/02 support memory interface allocated by user
• 2008/May/26 fix protect() to avoid invalid setting(thanks to shinichiro_h san)
• 2008/Apr/30 add cmpxchg16b, cdqe
• 2008/Apr/29 support x64
• 2008/Apr/14 code refactoring
• 2008/Mar/12 add bsr/bsf
• 2008/Feb/14 fix output of sub eax, 1234 (thanks to Robert)
• 2007/Nov/5 support lock, xadd, xchg
• 2007/Nov/2 support SSSE3/SSE4 (thanks to Dango-Chu san)
• 2007/Feb/4 fix the bug that exception doesn't occur under the condition which the offset of jmp mnemonic without T_NEAR is over 127.
• 2007/Jan/21 fix the bug to create address like [disp] select smaller representation for mov (eax|ax|al, [disp])
• 2007/Jan/4 first version

Author

MITSUNARI Shigeo([email protected])