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

   darwin64.S - Copyright (c) 2006 Free Software Foundation, Inc.

   derived from unix64.S

   x86-64 Foreign Function Interface for Darwin.

   Permission is hereby granted, free of charge, to any person obtaining

   a copy of this software and associated documentation files (the

   ``Software''), to deal in the Software without restriction, including

   without limitation the rights to use, copy, modify, merge, publish,

   distribute, sublicense, and/or sell copies of the Software, and to

   permit persons to whom the Software is furnished to do so, subject to

   the following conditions:

   The above copyright notice and this permission notice shall be included

   in all copies or substantial portions of the Software.

   THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS

   OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF

   MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.

   IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY CLAIM, DAMAGES OR

   OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,

   ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR

   OTHER DEALINGS IN THE SOFTWARE.

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

#ifdef __x86_64__

#define LIBFFI_ASM

#include <fficonfig.h>

#include <ffi.h>

 .file "darwin64.S"

.text

/* ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,

       void *raddr, void (*fnaddr)());

   Bit o trickiness here -- ARGS+BYTES is the base of the stack frame

   for this function.  This has been allocated by ffi_call.  We also

   deallocate some of the stack that has been alloca'd.  */

 .align  3

 .globl  _ffi_call_unix64

_ffi_call_unix64:

LUW0:

 movq  (%rsp), %r10    /* Load return address.  */

 leaq  (%rdi, %rsi), %rax  /* Find local stack base.  */

 movq  %rdx, (%rax)    /* Save flags.  */

 movq  %rcx, 8(%rax)   /* Save raddr.  */

 movq  %rbp, 16(%rax)    /* Save old frame pointer.  */

 movq  %r10, 24(%rax)    /* Relocate return address.  */

 movq  %rax, %rbp    /* Finalize local stack frame.  */

LUW1:

 movq  %rdi, %r10    /* Save a copy of the register area. */

 movq  %r8, %r11   /* Save a copy of the target fn.  */

 movl  %r9d, %eax    /* Set number of SSE registers.  */

 /* Load up all argument registers.  */

 movq  (%r10), %rdi

 movq  8(%r10), %rsi

 movq  16(%r10), %rdx

 movq  24(%r10), %rcx

 movq  32(%r10), %r8

 movq  40(%r10), %r9

 testl %eax, %eax

 jnz Lload_sse

Lret_from_load_sse:

 /* Deallocate the reg arg area.  */

 leaq  176(%r10), %rsp

 /* Call the user function.  */

 call  *%r11

 /* Deallocate stack arg area; local stack frame in redzone.  */

 leaq  24(%rbp), %rsp

 movq  0(%rbp), %rcx   /* Reload flags.  */

 movq  8(%rbp), %rdi   /* Reload raddr.  */

 movq  16(%rbp), %rbp    /* Reload old frame pointer.  */

LUW2:

 /* The first byte of the flags contains the FFI_TYPE.  */

 movzbl  %cl, %r10d

 leaq  Lstore_table(%rip), %r11

 movslq  (%r11, %r10, 4), %r10

 addq  %r11, %r10

 jmp *%r10

Lstore_table:

 .long Lst_void-Lstore_table   /* FFI_TYPE_VOID */

 .long Lst_sint32-Lstore_table   /* FFI_TYPE_INT */

 .long Lst_float-Lstore_table    /* FFI_TYPE_FLOAT */

 .long Lst_double-Lstore_table   /* FFI_TYPE_DOUBLE */

 .long Lst_ldouble-Lstore_table  /* FFI_TYPE_LONGDOUBLE */

 .long Lst_uint8-Lstore_table    /* FFI_TYPE_UINT8 */

 .long Lst_sint8-Lstore_table    /* FFI_TYPE_SINT8 */

 .long Lst_uint16-Lstore_table   /* FFI_TYPE_UINT16 */

 .long Lst_sint16-Lstore_table   /* FFI_TYPE_SINT16 */

 .long Lst_uint32-Lstore_table   /* FFI_TYPE_UINT32 */

 .long Lst_sint32-Lstore_table   /* FFI_TYPE_SINT32 */

 .long Lst_int64-Lstore_table    /* FFI_TYPE_UINT64 */

 .long Lst_int64-Lstore_table    /* FFI_TYPE_SINT64 */

 .long Lst_struct-Lstore_table   /* FFI_TYPE_STRUCT */

 .long Lst_int64-Lstore_table    /* FFI_TYPE_POINTER */

 .text

 .align  3

Lst_void:

 ret

 .align  3

Lst_uint8:

 movzbq  %al, %rax

 movq  %rax, (%rdi)

 ret

 .align  3

Lst_sint8:

 movsbq  %al, %rax

 movq  %rax, (%rdi)

 ret

 .align  3

Lst_uint16:

 movzwq  %ax, %rax

 movq  %rax, (%rdi)

 .align  3

Lst_sint16:

 movswq  %ax, %rax

 movq  %rax, (%rdi)

 ret

 .align  3

Lst_uint32:

 movl  %eax, %eax

 movq  %rax, (%rdi)

 .align  3

Lst_sint32:

 cltq

 movq  %rax, (%rdi)

 ret

 .align  3

Lst_int64:

 movq  %rax, (%rdi)

 ret

 .align  3

Lst_float:

 movss %xmm0, (%rdi)

 ret

 .align  3

Lst_double:

 movsd %xmm0, (%rdi)

 ret

Lst_ldouble:

 fstpt (%rdi)

 ret

 .align  3

Lst_struct:

 leaq  -20(%rsp), %rsi   /* Scratch area in redzone.  */

 /* We have to locate the values now, and since we don't want to

    write too much data into the user's return value, we spill the

    value to a 16 byte scratch area first.  Bits 8, 9, and 10

    control where the values are located.  Only one of the three

    bits will be set; see ffi_prep_cif_machdep for the pattern.  */

 movd  %xmm0, %r10

 movd  %xmm1, %r11

 testl $0x100, %ecx

 cmovnz  %rax, %rdx

 cmovnz  %r10, %rax

 testl $0x200, %ecx

 cmovnz  %r10, %rdx

 testl $0x400, %ecx

 cmovnz  %r10, %rax

 cmovnz  %r11, %rdx

 movq  %rax, (%rsi)

 movq  %rdx, 8(%rsi)

 /* Bits 12-31 contain the true size of the structure.  Copy from

    the scratch area to the true destination.  */

 shrl  $12, %ecx

 rep movsb

 ret

 /* Many times we can avoid loading any SSE registers at all.

    It's not worth an indirect jump to load the exact set of

    SSE registers needed; zero or all is a good compromise.  */

 .align  3

LUW3:

Lload_sse:

 movdqa  48(%r10), %xmm0

 movdqa  64(%r10), %xmm1

 movdqa  80(%r10), %xmm2

 movdqa  96(%r10), %xmm3

 movdqa  112(%r10), %xmm4

 movdqa  128(%r10), %xmm5

 movdqa  144(%r10), %xmm6

 movdqa  160(%r10), %xmm7

 jmp Lret_from_load_sse

LUW4:

 .align  3

 .globl  _ffi_closure_unix64

_ffi_closure_unix64:

LUW5:

 /* The carry flag is set by the trampoline iff SSE registers

    are used.  Don't clobber it before the branch instruction.  */

 leaq    -200(%rsp), %rsp

LUW6:

 movq  %rdi, (%rsp)

 movq    %rsi, 8(%rsp)

 movq    %rdx, 16(%rsp)

 movq    %rcx, 24(%rsp)

 movq    %r8, 32(%rsp)

 movq    %r9, 40(%rsp)

 jc      Lsave_sse

Lret_from_save_sse:

 movq  %r10, %rdi

 leaq  176(%rsp), %rsi

 movq  %rsp, %rdx

 leaq  208(%rsp), %rcx

 call  _ffi_closure_unix64_inner

 /* Deallocate stack frame early; return value is now in redzone.  */

 addq  $200, %rsp

LUW7:

 /* The first byte of the return value contains the FFI_TYPE.  */

 movzbl  %al, %r10d

 leaq  Lload_table(%rip), %r11

 movslq  (%r11, %r10, 4), %r10

 addq  %r11, %r10

 jmp *%r10

Lload_table:

 .long Lld_void-Lload_table    /* FFI_TYPE_VOID */

 .long Lld_int32-Lload_table   /* FFI_TYPE_INT */

 .long Lld_float-Lload_table   /* FFI_TYPE_FLOAT */

 .long Lld_double-Lload_table    /* FFI_TYPE_DOUBLE */

 .long Lld_ldouble-Lload_table   /* FFI_TYPE_LONGDOUBLE */

 .long Lld_int8-Lload_table    /* FFI_TYPE_UINT8 */

 .long Lld_int8-Lload_table    /* FFI_TYPE_SINT8 */

 .long Lld_int16-Lload_table   /* FFI_TYPE_UINT16 */

 .long Lld_int16-Lload_table   /* FFI_TYPE_SINT16 */

 .long Lld_int32-Lload_table   /* FFI_TYPE_UINT32 */

 .long Lld_int32-Lload_table   /* FFI_TYPE_SINT32 */

 .long Lld_int64-Lload_table   /* FFI_TYPE_UINT64 */

 .long Lld_int64-Lload_table   /* FFI_TYPE_SINT64 */

 .long Lld_struct-Lload_table    /* FFI_TYPE_STRUCT */

 .long Lld_int64-Lload_table   /* FFI_TYPE_POINTER */

 .text

 .align  3

Lld_void:

 ret

 .align  3

Lld_int8:

 movzbl  -24(%rsp), %eax

 ret

 .align  3

Lld_int16:

 movzwl  -24(%rsp), %eax

 ret

 .align  3

Lld_int32:

 movl  -24(%rsp), %eax

 ret

 .align  3

Lld_int64:

 movq  -24(%rsp), %rax

 ret

 .align  3

Lld_float:

 movss -24(%rsp), %xmm0

 ret

 .align  3

Lld_double:

 movsd -24(%rsp), %xmm0

 ret

 .align  3

Lld_ldouble:

 fldt  -24(%rsp)

 ret

 .align  3

Lld_struct:

 /* There are four possibilities here, %rax/%rdx, %xmm0/%rax,

    %rax/%xmm0, %xmm0/%xmm1.  We collapse two by always loading

    both rdx and xmm1 with the second word.  For the remaining,

    bit 8 set means xmm0 gets the second word, and bit 9 means

    that rax gets the second word.  */

 movq  -24(%rsp), %rcx

 movq  -16(%rsp), %rdx

 movq  -16(%rsp), %xmm1

 testl $0x100, %eax

 cmovnz  %rdx, %rcx

 movd  %rcx, %xmm0

 testl $0x200, %eax

 movq  -24(%rsp), %rax

 cmovnz  %rdx, %rax

 ret

 /* See the comment above Lload_sse; the same logic applies here.  */

 .align  3

LUW8:

Lsave_sse:

 movdqa  %xmm0, 48(%rsp)

 movdqa  %xmm1, 64(%rsp)

 movdqa  %xmm2, 80(%rsp)

 movdqa  %xmm3, 96(%rsp)

 movdqa  %xmm4, 112(%rsp)

 movdqa  %xmm5, 128(%rsp)

 movdqa  %xmm6, 144(%rsp)

 movdqa  %xmm7, 160(%rsp)

 jmp Lret_from_save_sse

LUW9:

.section __TEXT,__eh_frame,coalesced,no_toc+strip_static_syms+live_support

EH_frame1:

 .set  L$set$0,LECIE1-LSCIE1   /* CIE Length */

 .long L$set$0

LSCIE1:

 .long 0x0   /* CIE Identifier Tag */

 .byte 0x1   /* CIE Version */

 .ascii  "zR\0"    /* CIE Augmentation */

 .byte 0x1   /* uleb128 0x1; CIE Code Alignment Factor */

 .byte 0x78    /* sleb128 -8; CIE Data Alignment Factor */

 .byte 0x10    /* CIE RA Column */

 .byte 0x1   /* uleb128 0x1; Augmentation size */

 .byte 0x10    /* FDE Encoding (pcrel sdata4) */

 .byte 0xc   /* DW_CFA_def_cfa, %rsp offset 8 */

 .byte 0x7   /* uleb128 0x7 */

 .byte 0x8   /* uleb128 0x8 */

 .byte 0x90    /* DW_CFA_offset, column 0x10 */

 .byte 0x1

 .align  3

LECIE1:

 .globl _ffi_call_unix64.eh

_ffi_call_unix64.eh:

LSFDE1:

 .set  L$set$1,LEFDE1-LASFDE1  /* FDE Length */

 .long L$set$1

LASFDE1:

 .long LASFDE1-EH_frame1 /* FDE CIE offset */

 .quad LUW0-.      /* FDE initial location */

 .set  L$set$2,LUW4-LUW0 /* FDE address range */

 .quad L$set$2

 .byte 0x0     /* Augmentation size */

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$3,LUW1-LUW0

 .long L$set$3

 /* New stack frame based off rbp.  This is a itty bit of unwind

    trickery in that the CFA *has* changed.  There is no easy way

    to describe it correctly on entry to the function.  Fortunately,

    it doesn't matter too much since at all points we can correctly

    unwind back to ffi_call.  Note that the location to which we

    moved the return address is (the new) CFA-8, so from the

    perspective of the unwind info, it hasn't moved.  */

 .byte 0xc     /* DW_CFA_def_cfa, %rbp offset 32 */

 .byte 0x6

 .byte 0x20

 .byte 0x80+6      /* DW_CFA_offset, %rbp offset 2*-8 */

 .byte 0x2

 .byte 0xa     /* DW_CFA_remember_state */

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$4,LUW2-LUW1

 .long L$set$4

 .byte 0xc     /* DW_CFA_def_cfa, %rsp offset 8 */

 .byte 0x7

 .byte 0x8

 .byte 0xc0+6      /* DW_CFA_restore, %rbp */

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$5,LUW3-LUW2

 .long L$set$5

 .byte 0xb     /* DW_CFA_restore_state */

 .align  3

LEFDE1:

 .globl _ffi_closure_unix64.eh

_ffi_closure_unix64.eh:

LSFDE3:

 .set  L$set$6,LEFDE3-LASFDE3  /* FDE Length */

 .long L$set$6

LASFDE3:

 .long LASFDE3-EH_frame1 /* FDE CIE offset */

 .quad LUW5-.      /* FDE initial location */

 .set  L$set$7,LUW9-LUW5 /* FDE address range */

 .quad L$set$7

 .byte 0x0     /* Augmentation size */

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$8,LUW6-LUW5

 .long L$set$8

 .byte 0xe     /* DW_CFA_def_cfa_offset */

 .byte 208,1     /* uleb128 208 */

 .byte 0xa     /* DW_CFA_remember_state */

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$9,LUW7-LUW6

 .long L$set$9

 .byte 0xe     /* DW_CFA_def_cfa_offset */

 .byte 0x8

 .byte 0x4     /* DW_CFA_advance_loc4 */

 .set  L$set$10,LUW8-LUW7

 .long L$set$10

 .byte 0xb     /* DW_CFA_restore_state */

 .align  3

LEFDE3:

 .subsections_via_symbols

#endif /* __x86_64__ */