/*

 *    Stack-less Just-In-Time compiler

 *

 *    Copyright Zoltan Herczeg (hzmester@freemail.hu). All rights reserved.

 *

 * Redistribution and use in source and binary forms, with or without modification, are

 * permitted provided that the following conditions are met:

 *

 *   1. Redistributions of source code must retain the above copyright notice, this list of

 *      conditions and the following disclaimer.

 *

 *   2. Redistributions in binary form must reproduce the above copyright notice, this list

 *      of conditions and the following disclaimer in the documentation and/or other materials

 *      provided with the distribution.

 *

 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) AND CONTRIBUTORS ``AS IS'' AND ANY

 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES

 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT

 * SHALL THE COPYRIGHT HOLDER(S) OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,

 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED

 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR

 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN

 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN

 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 */

#ifndef SLJIT_LIR_H_

#define SLJIT_LIR_H_

/*

   ------------------------------------------------------------------------

    Stack-Less JIT compiler for multiple architectures (x86, ARM, PowerPC)

   ------------------------------------------------------------------------

   Short description

    Advantages:

      - The execution can be continued from any LIR instruction. In other

        words, it is possible to jump to any label from anywhere, even from

        a code fragment, which is compiled later, as long as the compiling

        context is the same. See sljit_emit_enter for more details.

      - Supports self modifying code: target of any jump and call

        instructions and some constant values can be dynamically modified

        during runtime. See SLJIT_REWRITABLE_JUMP.

        - although it is not suggested to do it frequently

        - can be used for inline caching: save an important value once

          in the instruction stream

      - A fixed stack space can be allocated for local variables

      - The compiler is thread-safe

      - The compiler is highly configurable through preprocessor macros.

        You can disable unneeded features (multithreading in single

        threaded applications), and you can use your own system functions

        (including memory allocators). See sljitConfig.h.

    Disadvantages:

      - The compiler is more like a platform independent assembler, so

        there is no built-in variable management. Registers and stack must

        be managed manually (the name of the compiler refers to this).

    In practice:

      - This approach is very effective for interpreters

        - One of the saved registers typically points to a stack interface

        - It can jump to any exception handler anytime (even if it belongs

          to another function)

        - Hot paths can be modified during runtime reflecting the changes

          of the fastest execution path of the dynamic language

        - SLJIT supports complex memory addressing modes

        - mainly position and context independent code (except some cases)

    For valgrind users:

      - pass --smc-check=all argument to valgrind, since JIT is a "self-modifying code"

*/

#if (defined SLJIT_HAVE_CONFIG_PRE && SLJIT_HAVE_CONFIG_PRE)

#include "sljitConfigPre.h"

#endif /* SLJIT_HAVE_CONFIG_PRE */

#include "sljitConfigCPU.h"

#include "sljitConfig.h"

/* The following header file defines useful macros for fine tuning

SLJIT based code generators. They are listed in the beginning

of sljitConfigInternal.h */

#include "sljitConfigInternal.h"

#if (defined SLJIT_HAVE_CONFIG_POST && SLJIT_HAVE_CONFIG_POST)

#include "sljitConfigPost.h"

#endif /* SLJIT_HAVE_CONFIG_POST */

#ifdef __cplusplus

extern "C" {

#endif /* __cplusplus */

/* Version numbers. */

#define SLJIT_MAJOR_VERSION 0

#define SLJIT_MINOR_VERSION 95

/* --------------------------------------------------------------------- */

/*  Error codes                                                          */

/* --------------------------------------------------------------------- */

/* Indicates no error. */

#define SLJIT_SUCCESS     0

/* After the call of sljit_generate_code(), the error code of the compiler

   is set to this value to avoid further code generation.

   The complier should be freed after sljit_generate_code(). */

#define SLJIT_ERR_COMPILED    1

/* Cannot allocate non-executable memory. */

#define SLJIT_ERR_ALLOC_FAILED    2

/* Cannot allocate executable memory.

   Only sljit_generate_code() returns with this error code. */

#define SLJIT_ERR_EX_ALLOC_FAILED 3

/* Unsupported instruction form. */

#define SLJIT_ERR_UNSUPPORTED   4

/* An invalid argument is passed to any SLJIT function. */

#define SLJIT_ERR_BAD_ARGUMENT    5

/* --------------------------------------------------------------------- */

/*  Registers                                                            */

/* --------------------------------------------------------------------- */

/*

  Scratch (R) registers: registers which may not preserve their values

  across function calls.

  Saved (S) registers: registers which preserve their values across

  function calls.

  The scratch and saved register sets overlap. The last scratch register

  is the first saved register, the one before the last is the second saved

  register, and so on.

  For example, in an architecture with only five registers (A-E), if two

  are scratch and three saved registers, they will be defined as follows:

    A |   R0   |      |  R0 always represent scratch register A

    B |   R1   |      |  R1 always represent scratch register B

    C |  [R2]  |  S2  |  R2 and S2 represent the same physical register C

    D |  [R3]  |  S1  |  R3 and S1 represent the same physical register D

    E |  [R4]  |  S0  |  R4 and S0 represent the same physical register E

  Note: SLJIT_NUMBER_OF_SCRATCH_REGISTERS will be 2 and

        SLJIT_NUMBER_OF_SAVED_REGISTERS will be 3.

  Note: For all supported architectures SLJIT_NUMBER_OF_REGISTERS >= 12

        and SLJIT_NUMBER_OF_SAVED_REGISTERS >= 6. However, 6 registers

        are virtual on x86-32. See below.

  The purpose of this definition is convenience: saved registers can

  be used as extra scratch registers. For example, building in the

  previous example, four registers can be specified as scratch registers

  and the fifth one as saved register, allowing any user code which requires

  four scratch registers to run unmodified. The SLJIT compiler automatically

  saves the content of the two extra scratch register on the stack. Scratch

  registers can also be preserved by saving their value on the stack but

  that needs to be done manually.

  Note: To emphasize that registers assigned to R2-R4 are saved

        registers, they are enclosed by square brackets.

  Note: sljit_emit_enter and sljit_set_context define whether a register

        is S or R register. E.g: if in the previous example 3 scratches and

        1 saved are mapped by sljit_emit_enter, the allowed register set

        will be: R0-R2 and S0. Although S2 is mapped to the same register

        than R2, it is not available in that configuration. Furthermore

        the S1 register cannot be used at all.

*/

/* Scratch registers. */

#define SLJIT_R0  1

#define SLJIT_R1  2

#define SLJIT_R2  3

/* Note: on x86-32, R3 - R6 (same as S3 - S6) are emulated (they

   are allocated on the stack). These registers are called virtual

   and cannot be used for memory addressing (cannot be part of

   any SLJIT_MEM1, SLJIT_MEM2 construct). There is no such

   limitation on other CPUs. See sljit_get_register_index(). */

#define SLJIT_R3  4

#define SLJIT_R4  5

#define SLJIT_R5  6

#define SLJIT_R6  7

#define SLJIT_R7  8

#define SLJIT_R8  9

#define SLJIT_R9  10

/* All R registers provided by the architecture can be accessed by SLJIT_R(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_REGISTERS. */

#define SLJIT_R(i)  (1 + (i))

/* Saved registers. */

#define SLJIT_S0  (SLJIT_NUMBER_OF_REGISTERS)

#define SLJIT_S1  (SLJIT_NUMBER_OF_REGISTERS - 1)

#define SLJIT_S2  (SLJIT_NUMBER_OF_REGISTERS - 2)

/* Note: on x86-32, S3 - S6 (same as R3 - R6) are emulated (they

   are allocated on the stack). These registers are called virtual

   and cannot be used for memory addressing (cannot be part of

   any SLJIT_MEM1, SLJIT_MEM2 construct). There is no such

   limitation on other CPUs. See sljit_get_register_index(). */

#define SLJIT_S3  (SLJIT_NUMBER_OF_REGISTERS - 3)

#define SLJIT_S4  (SLJIT_NUMBER_OF_REGISTERS - 4)

#define SLJIT_S5  (SLJIT_NUMBER_OF_REGISTERS - 5)

#define SLJIT_S6  (SLJIT_NUMBER_OF_REGISTERS - 6)

#define SLJIT_S7  (SLJIT_NUMBER_OF_REGISTERS - 7)

#define SLJIT_S8  (SLJIT_NUMBER_OF_REGISTERS - 8)

#define SLJIT_S9  (SLJIT_NUMBER_OF_REGISTERS - 9)

/* All S registers provided by the architecture can be accessed by SLJIT_S(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_SAVED_REGISTERS. */

#define SLJIT_S(i)  (SLJIT_NUMBER_OF_REGISTERS - (i))

/* Registers >= SLJIT_FIRST_SAVED_REG are saved registers. */

#define SLJIT_FIRST_SAVED_REG (SLJIT_S0 - SLJIT_NUMBER_OF_SAVED_REGISTERS + 1)

/* The SLJIT_SP provides direct access to the linear stack space allocated by

   sljit_emit_enter. It can only be used in the following form: SLJIT_MEM1(SLJIT_SP).

   The immediate offset is extended by the relative stack offset automatically.

   sljit_get_local_base can be used to obtain the real address of a value. */

#define SLJIT_SP  (SLJIT_NUMBER_OF_REGISTERS + 1)

/* Return with machine word. */

#define SLJIT_RETURN_REG  SLJIT_R0

/* --------------------------------------------------------------------- */

/*  Floating point registers                                             */

/* --------------------------------------------------------------------- */

/* Each floating point register can store a 32 or a 64 bit precision

   value. The FR and FS register sets overlap in the same way as R

   and S register sets. See above. */

/* Floating point scratch registers. */

#define SLJIT_FR0 1

#define SLJIT_FR1 2

#define SLJIT_FR2 3

#define SLJIT_FR3 4

#define SLJIT_FR4 5

#define SLJIT_FR5 6

#define SLJIT_FR6 7

#define SLJIT_FR7 8

#define SLJIT_FR8 9

#define SLJIT_FR9 10

/* All FR registers provided by the architecture can be accessed by SLJIT_FR(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_FLOAT_REGISTERS. */

#define SLJIT_FR(i) (1 + (i))

/* Floating point saved registers. */

#define SLJIT_FS0 (SLJIT_NUMBER_OF_FLOAT_REGISTERS)

#define SLJIT_FS1 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 1)

#define SLJIT_FS2 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 2)

#define SLJIT_FS3 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 3)

#define SLJIT_FS4 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 4)

#define SLJIT_FS5 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 5)

#define SLJIT_FS6 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 6)

#define SLJIT_FS7 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 7)

#define SLJIT_FS8 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 8)

#define SLJIT_FS9 (SLJIT_NUMBER_OF_FLOAT_REGISTERS - 9)

/* All FS registers provided by the architecture can be accessed by SLJIT_FS(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_SAVED_FLOAT_REGISTERS. */

#define SLJIT_FS(i) (SLJIT_NUMBER_OF_FLOAT_REGISTERS - (i))

/* Float registers >= SLJIT_FIRST_SAVED_FLOAT_REG are saved registers. */

#define SLJIT_FIRST_SAVED_FLOAT_REG (SLJIT_FS0 - SLJIT_NUMBER_OF_SAVED_FLOAT_REGISTERS + 1)

/* Return with floating point arg. */

#define SLJIT_RETURN_FREG SLJIT_FR0

/* --------------------------------------------------------------------- */

/*  Vector registers                                                     */

/* --------------------------------------------------------------------- */

/* Vector registers are storage areas, which are used for Single Instruction

   Multiple Data (SIMD) computations. The VR and VS register sets overlap

   in the same way as R and S register sets. See above.

   The storage space of vector registers often overlap with floating point

   registers. In this case setting the value of SLJIT_VR(i) destroys the

   value of SLJIT_FR(i) and vice versa. See SLJIT_SEPARATE_VECTOR_REGISTERS

   macro. */

/* Vector scratch registers. */

#define SLJIT_VR0 1

#define SLJIT_VR1 2

#define SLJIT_VR2 3

#define SLJIT_VR3 4

#define SLJIT_VR4 5

#define SLJIT_VR5 6

#define SLJIT_VR6 7

#define SLJIT_VR7 8

#define SLJIT_VR8 9

#define SLJIT_VR9 10

/* All VR registers provided by the architecture can be accessed by SLJIT_VR(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_VECTOR_REGISTERS. */

#define SLJIT_VR(i) (1 + (i))

/* Vector saved registers. */

#define SLJIT_VS0 (SLJIT_NUMBER_OF_VECTOR_REGISTERS)

#define SLJIT_VS1 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 1)

#define SLJIT_VS2 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 2)

#define SLJIT_VS3 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 3)

#define SLJIT_VS4 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 4)

#define SLJIT_VS5 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 5)

#define SLJIT_VS6 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 6)

#define SLJIT_VS7 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 7)

#define SLJIT_VS8 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 8)

#define SLJIT_VS9 (SLJIT_NUMBER_OF_VECTOR_REGISTERS - 9)

/* All VS registers provided by the architecture can be accessed by SLJIT_VS(i)

   The i parameter must be >= 0 and < SLJIT_NUMBER_OF_SAVED_VECTOR_REGISTERS. */

#define SLJIT_VS(i) (SLJIT_NUMBER_OF_VECTOR_REGISTERS - (i))

/* Vector registers >= SLJIT_FIRST_SAVED_VECTOR_REG are saved registers. */

#define SLJIT_FIRST_SAVED_VECTOR_REG (SLJIT_VS0 - SLJIT_NUMBER_OF_SAVED_VECTOR_REGISTERS + 1)

/* --------------------------------------------------------------------- */

/*  Argument type definitions                                            */

/* --------------------------------------------------------------------- */

/* The following argument type definitions are used by sljit_emit_enter,

   sljit_set_context, sljit_emit_call, and sljit_emit_icall functions.

   For sljit_emit_call and sljit_emit_icall, the first integer argument

   must be placed into SLJIT_R0, the second one into SLJIT_R1, and so on.

   Similarly the first floating point argument must be placed into SLJIT_FR0,

   the second one into SLJIT_FR1, and so on.

   For sljit_emit_enter, the integer arguments can be stored in scratch

   or saved registers. Scratch registers are identified by a _R suffix.

   If only saved registers are used, then the allocation mirrors what is

   done for the "call" functions but using saved registers, meaning that

   the first integer argument goes to SLJIT_S0, the second one goes into

   SLJIT_S1, and so on.

   If scratch registers are used, then the way the integer registers are

   allocated changes so that SLJIT_S0, SLJIT_S1, etc; will be assigned

   only for the arguments not using scratch registers, while SLJIT_R<n>

   will be used for the ones using scratch registers.

   Furthermore, the index (shown as "n" above) that will be used for the

   scratch register depends on how many previous integer registers

   (scratch or saved) were used already, starting with SLJIT_R0.

   Eventhough some indexes will be likely skipped, they still need to be

   accounted for in the scratches parameter of sljit_emit_enter. See below

   for some examples.

   The floating point arguments always use scratch registers (but not the

   _R suffix like the integer arguments) and must use SLJIT_FR0, SLJIT_FR1,

   just like in the "call" functions.

   Note: the mapping for scratch registers is part of the compiler context

         and therefore a new context after sljit_emit_call/sljit_emit_icall

         could remove access to some scratch registers that were used as

         arguments.

   Example function definition:

     sljit_f32 SLJIT_FUNC example_c_callback(void *arg_a,

         sljit_f64 arg_b, sljit_u32 arg_c, sljit_f32 arg_d);

   Argument type definition:

     SLJIT_ARG_RETURN(SLJIT_ARG_TYPE_F32)

        | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_P, 1) | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_F64, 2)

        | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_32, 3) | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_F32, 4)

   Short form of argument type definition:

     SLJIT_ARGS4(F32, P, F64, 32, F32)

   Argument passing:

     arg_a must be placed in SLJIT_R0

     arg_b must be placed in SLJIT_FR0

     arg_c must be placed in SLJIT_R1

     arg_d must be placed in SLJIT_FR1

   Examples for argument processing by sljit_emit_enter:

     SLJIT_ARGS4V(P, 32_R, F32, W)

     Arguments are placed into: SLJIT_S0, SLJIT_R1, SLJIT_FR0, SLJIT_S1

     The type of the result is void.

     SLJIT_ARGS4(F32, W, W_R, W, W_R)

     Arguments are placed into: SLJIT_S0, SLJIT_R1, SLJIT_S1, SLJIT_R3

     The type of the result is sljit_f32.

     SLJIT_ARGS4(P, W, F32, P_R)

     Arguments are placed into: SLJIT_FR0, SLJIT_S0, SLJIT_FR1, SLJIT_R1

     The type of the result is pointer.

     Note: it is recommended to pass the scratch arguments first

     followed by the saved arguments:

       SLJIT_ARGS4(W, W_R, W_R, W, W)

       Arguments are placed into: SLJIT_R0, SLJIT_R1, SLJIT_S0, SLJIT_S1

       The type of the result is sljit_sw / sljit_uw.

*/

/* The following flag is only allowed for the integer arguments of

   sljit_emit_enter. When the flag is set, the integer argument is

   stored in a scratch register instead of a saved register. */

#define SLJIT_ARG_TYPE_SCRATCH_REG 0x8

/* No return value, only supported by SLJIT_ARG_RETURN. */

#define SLJIT_ARG_TYPE_RET_VOID   0

/* Machine word sized integer argument or result. */

#define SLJIT_ARG_TYPE_W    1

#define SLJIT_ARG_TYPE_W_R  (SLJIT_ARG_TYPE_W | SLJIT_ARG_TYPE_SCRATCH_REG)

/* 32 bit integer argument or result. */

#define SLJIT_ARG_TYPE_32   2

#define SLJIT_ARG_TYPE_32_R (SLJIT_ARG_TYPE_32 | SLJIT_ARG_TYPE_SCRATCH_REG)

/* Pointer sized integer argument or result. */

#define SLJIT_ARG_TYPE_P    3

#define SLJIT_ARG_TYPE_P_R  (SLJIT_ARG_TYPE_P | SLJIT_ARG_TYPE_SCRATCH_REG)

/* 64 bit floating point argument or result. */

#define SLJIT_ARG_TYPE_F64    4

/* 32 bit floating point argument or result. */

#define SLJIT_ARG_TYPE_F32    5

#define SLJIT_ARG_SHIFT 4

#define SLJIT_ARG_RETURN(type) (type)

#define SLJIT_ARG_VALUE(type, idx) ((type) << ((idx) * SLJIT_ARG_SHIFT))

/* Simplified argument list definitions.

   The following definition:

       SLJIT_ARG_RETURN(SLJIT_ARG_TYPE_W) | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_F32, 1)

   can be shortened to:

       SLJIT_ARGS1(W, F32)

   Another example where no value is returned:

       SLJIT_ARG_RETURN(SLJIT_ARG_TYPE_RET_VOID) | SLJIT_ARG_VALUE(SLJIT_ARG_TYPE_W_R, 1)

   can be shortened to:

       SLJIT_ARGS1V(W_R)

*/

#define SLJIT_ARG_TO_TYPE(type) SLJIT_ARG_TYPE_ ## type

#define SLJIT_ARGS0(ret) \

  SLJIT_ARG_RETURN(SLJIT_ARG_TO_TYPE(ret))

#define SLJIT_ARGS0V() \

  SLJIT_ARG_RETURN(SLJIT_ARG_TYPE_RET_VOID)

#define SLJIT_ARGS1(ret, arg1) \

  (SLJIT_ARGS0(ret) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg1), 1))

#define SLJIT_ARGS1V(arg1) \

  (SLJIT_ARGS0V() | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg1), 1))

#define SLJIT_ARGS2(ret, arg1, arg2) \

  (SLJIT_ARGS1(ret, arg1) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg2), 2))

#define SLJIT_ARGS2V(arg1, arg2) \

  (SLJIT_ARGS1V(arg1) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg2), 2))

#define SLJIT_ARGS3(ret, arg1, arg2, arg3) \

  (SLJIT_ARGS2(ret, arg1, arg2) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg3), 3))

#define SLJIT_ARGS3V(arg1, arg2, arg3) \

  (SLJIT_ARGS2V(arg1, arg2) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg3), 3))

#define SLJIT_ARGS4(ret, arg1, arg2, arg3, arg4) \

  (SLJIT_ARGS3(ret, arg1, arg2, arg3) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg4), 4))

#define SLJIT_ARGS4V(arg1, arg2, arg3, arg4) \

  (SLJIT_ARGS3V(arg1, arg2, arg3) | SLJIT_ARG_VALUE(SLJIT_ARG_TO_TYPE(arg4), 4))

/* --------------------------------------------------------------------- */

/*  Main structures and functions                                        */

/* --------------------------------------------------------------------- */

/*

  The following structures are private, and can be changed in the

  future. Keeping them here allows code inlining.

*/

struct sljit_memory_fragment {

  struct sljit_memory_fragment *next;

  sljit_uw used_size;

  /* Must be aligned to sljit_sw. */

  sljit_u8 memory[1];

};

struct sljit_label {

  struct sljit_label *next;

  union {

    sljit_uw index;

    sljit_uw addr;

  } u;

  /* The maximum size difference. */

  sljit_uw size;

};

struct sljit_jump {

  struct sljit_jump *next;

  sljit_uw addr;

  /* Architecture dependent flags. */

  sljit_uw flags;

  union {

    sljit_uw target;

    struct sljit_label *label;

  } u;

};

struct sljit_const {

  struct sljit_const *next;

  sljit_uw addr;

};

struct sljit_generate_code_buffer {

  void *buffer;

  sljit_uw size;

  sljit_sw executable_offset;

};

struct sljit_read_only_buffer {

  struct sljit_read_only_buffer *next;

  sljit_uw size;

  /* Label can be replaced by address after sljit_generate_code. */

  union {

    struct sljit_label *label;

    sljit_uw addr;

  } u;

};

struct sljit_compiler {

  sljit_s32 error;

  sljit_s32 options;

  struct sljit_label *labels;

  struct sljit_jump *jumps;

  struct sljit_const *consts;

  struct sljit_label *last_label;

  struct sljit_jump *last_jump;

  struct sljit_const *last_const;

  void *allocator_data;

  void *user_data;

  struct sljit_memory_fragment *buf;

  struct sljit_memory_fragment *abuf;

  /* Number of labels created by the compiler. */

  sljit_uw label_count;

  /* Available scratch registers. */

  sljit_s32 scratches;

  /* Available saved registers. */

  sljit_s32 saveds;

  /* Available float scratch registers. */

  sljit_s32 fscratches;

  /* Available float saved registers. */

  sljit_s32 fsaveds;

#if (defined SLJIT_SEPARATE_VECTOR_REGISTERS && SLJIT_SEPARATE_VECTOR_REGISTERS) \

    || (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) \

    || (defined SLJIT_DEBUG && SLJIT_DEBUG) \

    || (defined SLJIT_VERBOSE && SLJIT_VERBOSE)

  /* Available vector scratch registers. */

  sljit_s32 vscratches;

  /* Available vector saved registers. */

  sljit_s32 vsaveds;

#endif /* SLJIT_SEPARATE_VECTOR_REGISTERS || SLJIT_ARGUMENT_CHECKS || SLJIT_DEBUG || SLJIT_VERBOSE */

  /* Local stack size. */

  sljit_s32 local_size;

  /* Maximum code size. */

  sljit_uw size;

  /* Relative offset of the executable mapping from the writable mapping. */

  sljit_sw executable_offset;

  /* Executable size for statistical purposes. */

  sljit_uw executable_size;

#if (defined SLJIT_HAS_STATUS_FLAGS_STATE && SLJIT_HAS_STATUS_FLAGS_STATE)

  sljit_s32 status_flags_state;

#endif /* SLJIT_HAS_STATUS_FLAGS_STATE */

#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32)

  sljit_s32 args_size;

#endif /* SLJIT_CONFIG_X86_32 */

#if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64)

  /* Temporary fields. */

  sljit_s32 mode32;

#endif /* SLJIT_CONFIG_X86_64 */

#if (defined SLJIT_CONFIG_ARM_V6 && SLJIT_CONFIG_ARM_V6)

  /* Constant pool handling. */

  sljit_uw *cpool;

  sljit_u8 *cpool_unique;

  sljit_uw cpool_diff;

  sljit_uw cpool_fill;

  /* Other members. */

  /* Contains pointer, "ldr pc, [...]" pairs. */

  sljit_uw patches;

#endif /* SLJIT_CONFIG_ARM_V6 */

#if (defined SLJIT_CONFIG_ARM_V6 && SLJIT_CONFIG_ARM_V6) || (defined SLJIT_CONFIG_ARM_V7 && SLJIT_CONFIG_ARM_V7)

  /* Temporary fields. */

  sljit_uw shift_imm;

#endif /* SLJIT_CONFIG_ARM_V6 || SLJIT_CONFIG_ARM_V6 */

#if (defined SLJIT_CONFIG_ARM_32 && SLJIT_CONFIG_ARM_32) && (defined __SOFTFP__)

  sljit_uw args_size;

#endif /* SLJIT_CONFIG_ARM_32 && __SOFTFP__ */

#if (defined SLJIT_CONFIG_PPC && SLJIT_CONFIG_PPC)

  /* Temporary fields. */

  sljit_u32 imm;

#endif /* SLJIT_CONFIG_PPC */

#if (defined SLJIT_CONFIG_MIPS && SLJIT_CONFIG_MIPS)

  sljit_s32 delay_slot;

  /* Temporary fields. */

  sljit_s32 cache_arg;

  sljit_sw cache_argw;

#endif /* SLJIT_CONFIG_MIPS */

#if (defined SLJIT_CONFIG_MIPS_32 && SLJIT_CONFIG_MIPS_32)

  sljit_uw args_size;

#endif /* SLJIT_CONFIG_MIPS_32 */

#if (defined SLJIT_CONFIG_RISCV && SLJIT_CONFIG_RISCV)

  /* Temporary fields. */

  sljit_s32 cache_arg;

  sljit_sw cache_argw;

#endif /* SLJIT_CONFIG_RISCV */

#if (defined SLJIT_CONFIG_S390X && SLJIT_CONFIG_S390X)

  /* Need to allocate register save area to make calls. */

  /* Temporary fields. */

  sljit_s32 mode;

#endif /* SLJIT_CONFIG_S390X */

#if (defined SLJIT_CONFIG_LOONGARCH && SLJIT_CONFIG_LOONGARCH)

  /* Temporary fields. */

  sljit_s32 cache_arg;

  sljit_sw cache_argw;

#endif /* SLJIT_CONFIG_LOONGARCH */

#if (defined SLJIT_VERBOSE && SLJIT_VERBOSE)

  FILE* verbose;

#endif /* SLJIT_VERBOSE */

  /* Note: SLJIT_DEBUG enables SLJIT_ARGUMENT_CHECKS. */

#if (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) \

    || (defined SLJIT_DEBUG && SLJIT_DEBUG)

  /* Flags specified by the last arithmetic instruction.

     It contains the type of the variable flag. */

  sljit_s32 last_flags;

  /* Return value type set by entry functions. */

  sljit_s32 last_return;

  /* Local size passed to entry functions. */

  sljit_s32 logical_local_size;

#endif /* SLJIT_ARGUMENT_CHECKS || SLJIT_DEBUG */

#if (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) \

    || (defined SLJIT_DEBUG && SLJIT_DEBUG) \

    || (defined SLJIT_VERBOSE && SLJIT_VERBOSE)

#if !(defined SLJIT_SEPARATE_VECTOR_REGISTERS && SLJIT_SEPARATE_VECTOR_REGISTERS)

  /* Available float scratch registers. */

  sljit_s32 real_fscratches;

  /* Available float saved registers. */

  sljit_s32 real_fsaveds;

#endif /* !SLJIT_SEPARATE_VECTOR_REGISTERS */

  /* Trust arguments when an API function is called.

     Used internally for calling API functions. */

  sljit_s32 skip_checks;

#endif /* SLJIT_ARGUMENT_CHECKS || SLJIT_DEBUG || SLJIT_VERBOSE */

};

/* --------------------------------------------------------------------- */

/*  Main functions                                                       */

/* --------------------------------------------------------------------- */

/* Creates an SLJIT compiler. The allocator_data is required by some

   custom memory managers. This pointer is passed to SLJIT_MALLOC

   and SLJIT_FREE macros. Most allocators (including the default

   one) ignores this value, and it is recommended to pass NULL

   as a dummy value for allocator_data.

   Returns NULL if failed. */

SLJIT_API_FUNC_ATTRIBUTE struct sljit_compiler* sljit_create_compiler(void *allocator_data);

/* Frees everything except the compiled machine code. */

SLJIT_API_FUNC_ATTRIBUTE void sljit_free_compiler(struct sljit_compiler *compiler);

/* Returns the current error code. If an error occurres, future calls

   which uses the same compiler argument returns early with the same

   error code. Thus there is no need for checking the error after every

   call, it is enough to do it after the code is compiled. Removing

   these checks increases the performance of the compiling process. */

static SLJIT_INLINE sljit_s32 sljit_get_compiler_error(struct sljit_compiler *compiler) { return compiler->error; }

/* Sets the compiler error code to SLJIT_ERR_ALLOC_FAILED except

   if an error was detected before. After the error code is set

   the compiler behaves as if the allocation failure happened

   during an SLJIT function call. This can greatly simplify error

   checking, since it is enough to check the compiler status

   after the code is compiled. */

SLJIT_API_FUNC_ATTRIBUTE void sljit_set_compiler_memory_error(struct sljit_compiler *compiler);

/* Allocate a small amount of memory. The size must be <= 64 bytes on 32 bit,

   and <= 128 bytes on 64 bit architectures. The memory area is owned by the

   compiler, and freed by sljit_free_compiler. The returned pointer is

   sizeof(sljit_sw) aligned. Excellent for allocating small blocks during

   compiling, and no need to worry about freeing them. The size is enough

   to contain at most 16 pointers. If the size is outside of the range,

   the function will return with NULL. However, this return value does not

   indicate that there is no more memory (does not set the current error code

   of the compiler to out-of-memory status). */

SLJIT_API_FUNC_ATTRIBUTE void* sljit_alloc_memory(struct sljit_compiler *compiler, sljit_s32 size);

/* Returns the allocator data passed to sljit_create_compiler. */

static SLJIT_INLINE void* sljit_compiler_get_allocator_data(struct sljit_compiler *compiler) { return compiler->allocator_data; }

/* Sets/get the user data for a compiler. */

static SLJIT_INLINE void sljit_compiler_set_user_data(struct sljit_compiler *compiler, void *user_data) { compiler->user_data = user_data; }

static SLJIT_INLINE void* sljit_compiler_get_user_data(struct sljit_compiler *compiler) { return compiler->user_data; }

#if (defined SLJIT_VERBOSE && SLJIT_VERBOSE)

/* Passing NULL disables verbose. */

SLJIT_API_FUNC_ATTRIBUTE void sljit_compiler_verbose(struct sljit_compiler *compiler, FILE* verbose);

#endif /* SLJIT_VERBOSE */

/* Option bits for sljit_generate_code. */

/* The exec_allocator_data points to a pre-allocated

   buffer which type is sljit_generate_code_buffer. */

#define SLJIT_GENERATE_CODE_BUFFER    0x1

/* Create executable code from the instruction stream. This is the final step

   of the code generation, and no more instructions can be emitted after this call.

   options is the combination of SLJIT_GENERATE_CODE_* bits

   exec_allocator_data is passed to SLJIT_MALLOC_EXEC and

                       SLJIT_MALLOC_FREE functions */

SLJIT_API_FUNC_ATTRIBUTE void* sljit_generate_code(struct sljit_compiler *compiler, sljit_s32 options, void *exec_allocator_data);

/* Free executable code. */

SLJIT_API_FUNC_ATTRIBUTE void sljit_free_code(void* code, void *exec_allocator_data);

/* When the protected executable allocator is used the JIT code is mapped

   twice. The first mapping has read/write and the second mapping has read/exec

   permissions. This function returns with the relative offset of the executable

   mapping using the writable mapping as the base after the machine code is

   successfully generated. The returned value is always 0 for the normal executable

   allocator, since it uses only one mapping with read/write/exec permissions.

   Dynamic code modifications requires this value.

   Before a successful code generation, this function returns with 0. */

static SLJIT_INLINE sljit_sw sljit_get_executable_offset(struct sljit_compiler *compiler) { return compiler->executable_offset; }

/* The executable memory consumption of the generated code can be retrieved by

   this function. The returned value can be used for statistical purposes.

   Before a successful code generation, this function returns with 0. */

static SLJIT_INLINE sljit_uw sljit_get_generated_code_size(struct sljit_compiler *compiler) { return compiler->executable_size; }

/* Returns with non-zero if the feature or limitation type passed as its

   argument is present on the current CPU. The return value is one, if a

   feature is fully supported, and it is two, if partially supported.

   Some features (e.g. floating point operations) require hardware (CPU)

   support while others (e.g. move with update) are emulated if not available.

   However, even when a feature is emulated, specialized code paths may be

   faster than the emulation. Some limitations are emulated as well so their

   general case is supported but it has extra performance costs.

   Note: sljitConfigInternal.h also provides several feature detection macros. */

/* [Not emulated] Floating-point support is available. */

#define SLJIT_HAS_FPU     0

/* [Limitation] Some registers are virtual registers. */

#define SLJIT_HAS_VIRTUAL_REGISTERS 1

/* [Emulated] Has zero register (setting a memory location to zero is efficient). */

#define SLJIT_HAS_ZERO_REGISTER   2

/* [Emulated] Count leading zero is supported. */

#define SLJIT_HAS_CLZ     3

/* [Emulated] Count trailing zero is supported. */

#define SLJIT_HAS_CTZ     4

/* [Emulated] Reverse the order of bytes is supported. */

#define SLJIT_HAS_REV     5

/* [Emulated] Rotate left/right is supported. */

#define SLJIT_HAS_ROT     6

/* [Emulated] Conditional move is supported. */

#define SLJIT_HAS_CMOV      7

/* [Emulated] Prefetch instruction is available (emulated as a nop). */

#define SLJIT_HAS_PREFETCH    8

/* [Emulated] Copy from/to f32 operation is available (see sljit_emit_fcopy). */

#define SLJIT_HAS_COPY_F32    9

/* [Emulated] Copy from/to f64 operation is available (see sljit_emit_fcopy). */

#define SLJIT_HAS_COPY_F64    10

/* [Not emulated] The 64 bit floating point registers can be used as

   two separate 32 bit floating point registers (e.g. ARM32). The

   second 32 bit part can be accessed by SLJIT_F64_SECOND. */

#define SLJIT_HAS_F64_AS_F32_PAIR 11

/* [Not emulated] Some SIMD operations are supported by the compiler. */

#define SLJIT_HAS_SIMD      12

/* [Not emulated] SIMD registers are mapped to a pair of double precision

   floating point registers. E.g. passing either SLJIT_FR0 or SLJIT_FR1 to

   a simd operation represents the same 128 bit register, and both SLJIT_FR0

   and SLJIT_FR1 are overwritten. */

#define SLJIT_SIMD_REGS_ARE_PAIRS 13

/* [Not emulated] Atomic support is available. */

#define SLJIT_HAS_ATOMIC    14

/* [Not emulated] Memory barrier support is available. */

#define SLJIT_HAS_MEMORY_BARRIER    15

#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)

/* [Not emulated] AVX support is available on x86. */

#define SLJIT_HAS_AVX     100

/* [Not emulated] AVX2 support is available on x86. */

#define SLJIT_HAS_AVX2      101

#endif /* SLJIT_CONFIG_X86 */

#if (defined SLJIT_CONFIG_LOONGARCH)

/* [Not emulated] LASX support is available on LoongArch */

#define SLJIT_HAS_LASX        201

#endif /* SLJIT_CONFIG_LOONGARCH */

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type);

/* If type is between SLJIT_ORDERED_EQUAL and SLJIT_ORDERED_LESS_EQUAL,

   sljit_cmp_info returns with:

     zero - if the cpu supports the floating point comparison type

     one - if the comparison requires two machine instructions

     two - if the comparison requires more than two machine instructions

   When the result is non-zero, it is recommended to avoid

   using the specified comparison type if it is easy to do so.

   Otherwise it returns zero. */

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_cmp_info(sljit_s32 type);

/* The following functions generate machine code. If there is no

   error, they return with SLJIT_SUCCESS, otherwise they return

   with an error code. */

/*

   The executable code is a callable function from the viewpoint

   of the C language. Function calls must conform with the ABI

   (Application Binary Interface) of the target platform, which

   specify the purpose of machine registers and stack handling

   among other things. The sljit_emit_enter function emits the

   necessary instructions for setting up an entry point for the

   executable code. This is often called as function prologue.

   The "options" argument can be used to pass configuration options

   to the sljit compiler which affects the generated code, until

   another sljit_emit_enter or sljit_set_context is called. The

   available options are listed before sljit_emit_enter.

   The function argument list is specified by the SLJIT_ARGSx

   (SLJIT_ARGS0 .. SLJIT_ARGS4) macros. Currently maximum four

   arguments are supported. See the description of SLJIT_ARGSx

   macros about argument passing.

   The register set used by the function must be declared as well.

   The number of scratch and saved registers available to the

   function must be passed to sljit_emit_enter. Only R registers

   between R0 and "scratches" argument can be used later. E.g.

   if "scratches" is set to two, the scratch register set will

   be limited to SLJIT_R0 and SLJIT_R1. The S registers are

   declared in a similar manner, but their count is specified

   by "saveds" argument. The floating point scratch and saved

   registers can be set by using "scratches" and "saveds" argument

   as well, but their value must be passed to the SLJIT_ENTER_FLOAT

   macro, see below.

   The sljit_emit_enter is also capable of allocating a stack

   space for local data. The "local_size" argument contains the

   size in bytes of this local area, and it can be accessed using

   SLJIT_MEM1(SLJIT_SP). The memory area between SLJIT_SP (inclusive)

   and SLJIT_SP + local_size (exclusive) can be modified freely

   until the function returns. The alocated stack space is an

   uninitialized memory area.

   Floating point scratch and saved registers must be specified

   by the SLJIT_ENTER_FLOAT macro, which result value should be

   combined with scratches / saveds argument.

   Examples:

       To use three scratch and four floating point scratch

       registers, the "scratches" argument must be set to:

            3 | SLJIT_ENTER_FLOAT(4)

       To use six saved and five floating point saved

       registers, the "saveds" argument must be set to:

            6 | SLJIT_ENTER_FLOAT(5)

   Note: the following conditions must met:

         0 <= scratches <= SLJIT_NUMBER_OF_REGISTERS

         0 <= saveds <= SLJIT_NUMBER_OF_SAVED_REGISTERS

         scratches + saveds <= SLJIT_NUMBER_OF_REGISTERS

         0 <= float scratches <= SLJIT_NUMBER_OF_FLOAT_REGISTERS

         0 <= float saveds <= SLJIT_NUMBER_OF_SAVED_FLOAT_REGISTERS

         float scratches + float saveds <= SLJIT_NUMBER_OF_FLOAT_REGISTERS

   Note: the compiler can use saved registers as scratch registers,

         but the opposite is not supported

   Note: every call of sljit_emit_enter and sljit_set_context

         overwrites the previous context.

*/

/* The following options are available for sljit_emit_enter. */

/* Saved registers between SLJIT_S0 and SLJIT_S(n - 1) (inclusive)

   are not saved / restored on function enter / return. Instead,

   these registers can be used to pass / return data (such as

   global / local context pointers) across function calls. The

   value of n must be between 1 and 3. This option is only

   supported by SLJIT_ENTER_REG_ARG calling convention. */

#define SLJIT_ENTER_KEEP(n)   (n)

/* The compiled function uses an SLJIT specific register argument

   calling convention. This is a lightweight function call type where

   both the caller and the called functions must be compiled by

   SLJIT. The type argument of the call must be SLJIT_CALL_REG_ARG

   and all arguments must be stored in scratch registers. */

#define SLJIT_ENTER_REG_ARG   0x00000004

#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)

/* Use VEX prefix for all SIMD operations on x86. */

#define SLJIT_ENTER_USE_VEX   0x00010000

#endif /* !SLJIT_CONFIG_X86 */

/* Macros for other sljit_emit_enter arguments. */

/* Floating point scratch and saved registers can be

   specified by SLJIT_ENTER_FLOAT. */

#define SLJIT_ENTER_FLOAT(regs)   ((regs) << 8)

/* Vector scratch and saved registers can be specified

   by SLJIT_ENTER_VECTOR. */

#define SLJIT_ENTER_VECTOR(regs)  ((regs) << 16)

/* The local_size must be >= 0 and <= SLJIT_MAX_LOCAL_SIZE. */

#define SLJIT_MAX_LOCAL_SIZE    1048576

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compiler,

  sljit_s32 options, sljit_s32 arg_types,

  sljit_s32 scratches, sljit_s32 saveds, sljit_s32 local_size);

/* The SLJIT compiler has a current context (which contains the local

   stack space size, number of used registers, etc.) which is initialized

   by sljit_emit_enter. Several functions (such as sljit_emit_return)

   requires this context to be able to generate the appropriate code.

   However, some code fragments (compiled separately) may have no

   normal entry point so their context is unknown to the compiler.

   sljit_set_context and sljit_emit_enter have the same arguments,

   but sljit_set_context does not generate any machine code.

   Note: every call of sljit_emit_enter and sljit_set_context overwrites

         the previous context. */

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_set_context(struct sljit_compiler *compiler,

  sljit_s32 options, sljit_s32 arg_types,

  sljit_s32 scratches, sljit_s32 saveds, sljit_s32 local_size);

/* Return to the caller function. The sljit_emit_return_void function

   does not return with any value. The sljit_emit_return function returns

   with a single value loaded from its source operand. The load operation

   can be between SLJIT_MOV and SLJIT_MOV_P (see sljit_emit_op1) and

   SLJIT_MOV_F32/SLJIT_MOV_F64 (see sljit_emit_fop1) depending on the

   return value specified by sljit_emit_enter/sljit_set_context. */

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return_void(struct sljit_compiler *compiler);

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return(struct sljit_compiler *compiler, sljit_s32 op,

  sljit_s32 src, sljit_sw srcw);

/* Restores the saved registers and free the stack area, then the execution

   continues from the address specified by the source operand. This

   operation is similar to sljit_emit_return, but it ignores the return

   address. The code where the exection continues should use the same context

   as the caller function (see sljit_set_context). A word (pointer) value

   can be passed in the SLJIT_RETURN_REG register. This function can be used

   to jump to exception handlers. */

SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return_to(struct sljit_compiler *compiler,

  sljit_s32 src, sljit_sw srcw);

/*

   Source and destination operands for arithmetical instructions

    imm              - a simple immediate value (cannot be used as a destination)

    reg              - any of the available registers (immediate argument must be 0)

    [imm]            - absolute memory address

    [reg+imm]        - indirect memory address

    [reg+(reg<<imm)] - indirect indexed memory address (shift must be between 0 and 3)

                       useful for accessing arrays (fully supported by both x86 and

                       ARM architectures, and cheap operation on others)

*/

/*

   IMPORTANT NOTE: memory accesses MUST be naturally aligned unless

                   SLJIT_UNALIGNED macro is defined and its value is 1.

     length | alignment

   ---------+-----------

     byte   | 1 byte (any physical_address is accepted)

     half   | 2 byte (physical_address & 0x1 == 0)

     int    | 4 byte (physical_address & 0x3 == 0)

     word   | 4 byte if SLJIT_32BIT_ARCHITECTURE is defined and its value is 1

            | 8 byte if SLJIT_64BIT_ARCHITECTURE is defined and its value is 1

    pointer | size of sljit_up type (4 byte on 32 bit machines, 4 or 8 byte

            | on 64 bit machines)

   Note:   Different architectures have different addressing limitations.

           A single instruction is enough for the following addressing

           modes. Other addressing modes are emulated by instruction

           sequences. This information could help to improve those code

           generators which focuses only a few architectures.

   x86:    [reg+imm], -2^32+1 <= imm <= 2^32-1 (full address space on x86-32)

           [reg+(reg<<imm)] is supported

           [imm], -2^32+1 <= imm <= 2^32-1 is supported

           Write-back is not supported

   arm:    [reg+imm], -4095 <= imm <= 4095 or -255 <= imm <= 255 for signed

                bytes, any halfs or floating point values)

           [reg+(reg<<imm)] is supported

           Write-back is supported

   arm-t2: [reg+imm], -255 <= imm <= 4095

           [reg+(reg<<imm)] is supported

           Write back is supported only for [reg+imm], where -255 <= imm <= 255

   arm64:  [reg+imm], -256 <= imm <= 255, 0 <= aligned imm <= 4095 * alignment

           [reg+(reg<<imm)] is supported

           Write back is supported only for [reg+imm], where -256 <= imm <= 255

   ppc:    [reg+imm], -65536 <= imm <= 65535. 64 bit loads/stores and 32 bit

                signed load on 64 bit requires immediates divisible by 4.

                [reg+imm] is not supported for signed 8 bit values.

           [reg+reg] is supported

           Write-back is supported except for one instruction: 32 bit signed

                load with [reg+imm] addressing mode on 64 bit.

   mips:   [reg+imm], -65536 <= imm <= 65535

           Write-back is not supported

   riscv:  [reg+imm], -2048 <= imm <= 2047

           Write-back is not supported

   s390x:  [reg+imm], -2^19 <= imm < 2^19

           [reg+reg] is supported

           Write-back is not supported

   loongarch:  [reg+imm], -2048 <= imm <= 2047

           [reg+reg] is supported

           Write-back is not supported

*/

/* Macros for specifying operand types. */

#define SLJIT_MEM   0x80

#define SLJIT_MEM0()    (SLJIT_MEM)

#define SLJIT_MEM1(r1)    (SLJIT_MEM | (r1))

#define SLJIT_MEM2(r1, r2)  (SLJIT_MEM | (r1) | ((r2) << 8))

#define SLJIT_IMM   0x7f

#define SLJIT_REG_PAIR(r1, r2)  ((r1) | ((r2) << 8))

/* Macros for checking operand types (only for valid arguments). */

#define SLJIT_IS_REG(arg) ((arg) > 0 && (arg) < SLJIT_IMM)

#define SLJIT_IS_MEM(arg) ((arg) & SLJIT_MEM)

#define SLJIT_IS_MEM0(arg)  ((arg) == SLJIT_MEM)

#define SLJIT_IS_MEM1(arg)  ((arg) > SLJIT_MEM && (arg) < (SLJIT_MEM << 1))

#define SLJIT_IS_MEM2(arg)  (((arg) & SLJIT_MEM) && (arg) >= (SLJIT_MEM << 1))

#define SLJIT_IS_IMM(arg) ((arg) == SLJIT_IMM)

#define SLJIT_IS_REG_PAIR(arg)  (!((arg) & SLJIT_MEM) && (arg) >= (SLJIT_MEM << 1))

/* Macros for extracting registers from operands. */

/* Support operands which contains a single register or

   constructed using SLJIT_MEM1, SLJIT_MEM2, or SLJIT_REG_PAIR. */

#define SLJIT_EXTRACT_REG(arg)    ((arg) & 0x7f)

/* Support operands which constructed using SLJIT_MEM2, or SLJIT_REG_PAIR. */

#define SLJIT_EXTRACT_SECOND_REG(arg) ((arg) >> 8)

/* Sets 32 bit operation mode on 64 bit CPUs. This option is ignored on

   32 bit CPUs. When this option is set for an arithmetic operation, only

   the lower 32 bits of the input registers are used, and the CPU status

   flags are set according to the 32 bit result. Although the higher 32 bit

   of the input and the result registers are not defined by SLJIT, it might

   be defined by the CPU architecture (e.g. MIPS). To satisfy these CPU

   requirements all source registers must be the result of those operations

   where this option was also set. Memory loads read 32 bit values rather

   than 64 bit ones. In other words 32 bit and 64 bit operations cannot be

   mixed. The only exception is SLJIT_MOV32 which source register can hold

   any 32 or 64 bit value, and it is converted to a 32 bit compatible format

   first. When the source and destination registers are the same, this

   conversion is free (no instructions are emitted) on most CPUs. A 32 bit

   value can also be converted to a 64 bit value by SLJIT_MOV_S32

   (sign extension) or SLJIT_MOV_U32 (zero extension).

   As for floating-point operations, this option sets 32 bit single

   precision mode. Similar to the integer operations, all register arguments

   must be the result of those operations where this option was also set.

   Note: memory addressing always uses 64 bit values on 64 bit systems so

         the result of a 32 bit operation must not be used with SLJIT_MEMx

         macros.

   This option is part of the instruction name, so there is no need to

   manually set it. E.g:

     SLJIT_ADD32 == (SLJIT_ADD | SLJIT_32) */

#define SLJIT_32    0x100

/* Many CPUs (x86, ARM, PPC) have status flag bits which can be set according

   to the result of an operation. Other CPUs (MIPS) do not have status

   flag bits, and results must be stored in registers. To cover both

   architecture types efficiently only two flags are defined by SLJIT:

    * Zero (equal) flag: it is set if the result is zero

    * Variable flag: its value is defined by the arithmetic operation

   SLJIT instructions can set any or both of these flags. The value of

   these flags is undefined if the instruction does not specify their

   value. The description of each instruction contains the list of

   allowed flag types.

   Note: the logical or operation can be used to set flags.

   Example: SLJIT_ADD can set the Z, OVERFLOW, CARRY flags hence

     sljit_op2(..., SLJIT_ADD, ...)

       Both the zero and variable flags are undefined so they can

       have any value after the operation is completed.

     sljit_op2(..., SLJIT_ADD | SLJIT_SET_Z, ...)

       Sets the zero flag if the result is zero, clears it otherwise.

       The variable flag is undefined.

     sljit_op2(..., SLJIT_ADD | SLJIT_SET_OVERFLOW, ...)

       Sets the variable flag if an integer overflow occurs, clears

       it otherwise. The zero flag is undefined.

     sljit_op2(..., SLJIT_ADD | SLJIT_SET_Z | SLJIT_SET_CARRY, ...)

       Sets the zero flag if the result is zero, clears it otherwise.

       Sets the variable flag if unsigned overflow (carry) occurs,

       clears it otherwise.

   Certain instructions (e.g. SLJIT_MOV) does not modify flags, so

   status flags are unchanged.