github.com/prattmic/llgo-embedded@v0.0.0-20150820070356-41cfecea0e1e/third_party/gofrontend/libffi/src/x86/win64.S

#define LIBFFI_ASM
#include <fficonfig.h>
#include <ffi.h>
#include <ffi_cfi.h>

#if defined(HAVE_AS_CFI_PSEUDO_OP)
        .cfi_sections   .debug_frame
#endif

#define arg0	%rcx
#define arg1	%rdx
#define arg2	%r8
#define arg3	%r9

#ifdef SYMBOL_UNDERSCORE
#define SYMBOL_NAME(name) _##name
#else
#define SYMBOL_NAME(name) name
#endif

.macro E which
	.align	8
	.org	0b + \which * 8
.endm

	.text

/* ffi_call_win64 (void *stack, struct win64_call_frame *frame, void *r10)

   Bit o trickiness here -- FRAME 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	8
	.globl	ffi_call_win64

	.seh_proc ffi_call_win64
ffi_call_win64:
	cfi_startproc
	/* Set up the local stack frame and install it in rbp/rsp.  */
	movq	(%rsp), %rax
	movq	%rbp, (arg1)
	movq	%rax, 8(arg1)
	movq	arg1, %rbp
	cfi_def_cfa(%rbp, 16)
	cfi_rel_offset(%rbp, 0)
	.seh_pushreg %rbp
	.seh_setframe %rbp, 0
	.seh_endprologue
	movq	arg0, %rsp

	movq	arg2, %r10

	/* Load all slots into both general and xmm registers.  */
	movq	(%rsp), %rcx
	movsd	(%rsp), %xmm0
	movq	8(%rsp), %rdx
	movsd	8(%rsp), %xmm1
	movq	16(%rsp), %r8
	movsd	16(%rsp), %xmm2
	movq	24(%rsp), %r9
	movsd	24(%rsp), %xmm3

	call	*16(%rbp)

	movl	24(%rbp), %ecx
	movq	32(%rbp), %r8
	leaq	0f(%rip), %r10
	cmpl	$FFI_TYPE_SMALL_STRUCT_4B, %ecx
	leaq	(%r10, %rcx, 8), %r10
	ja	99f
	jmp	*%r10

/* Below, we're space constrained most of the time.  Thus we eschew the
   modern "mov, pop, ret" sequence (5 bytes) for "leave, ret" (2 bytes).  */
.macro epilogue
	leaveq
	cfi_remember_state
	cfi_def_cfa(%rsp, 8)
	cfi_restore(%rbp)
	ret
	cfi_restore_state
.endm

	.align	8
0:
E FFI_TYPE_VOID
	epilogue
E FFI_TYPE_INT
	movslq	%eax, %rax
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_FLOAT
	movss	%xmm0, (%r8)
	epilogue
E FFI_TYPE_DOUBLE
	movsd	%xmm0, (%r8)
	epilogue
E FFI_TYPE_LONGDOUBLE
	call	abort
E FFI_TYPE_UINT8
	movzbl	%al, %eax
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_SINT8
	movsbq	%al, %rax
	jmp	98f
E FFI_TYPE_UINT16
	movzwl	%ax, %eax
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_SINT16
	movswq	%ax, %rax
	jmp	98f
E FFI_TYPE_UINT32
	movl	%eax, %eax
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_SINT32
	movslq	%eax, %rax
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_UINT64
98:	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_SINT64
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_STRUCT
	epilogue
E FFI_TYPE_POINTER
	movq	%rax, (%r8)
	epilogue
E FFI_TYPE_COMPLEX
	call	abort
E FFI_TYPE_SMALL_STRUCT_1B
	movb	%al, (%r8)
	epilogue
E FFI_TYPE_SMALL_STRUCT_2B
	movw	%ax, (%r8)
	epilogue
E FFI_TYPE_SMALL_STRUCT_4B
	movl	%eax, (%r8)
	epilogue

	.align	8
99:	call	abort

.purgem epilogue

	cfi_endproc
	.seh_endproc


/* 32 bytes of outgoing register stack space, 8 bytes of alignment,
   16 bytes of result, 32 bytes of xmm registers.  */
#define ffi_clo_FS	(32+8+16+32)
#define ffi_clo_OFF_R	(32+8)
#define ffi_clo_OFF_X	(32+8+16)

	.align	8
	.globl	ffi_go_closure_win64

	.seh_proc ffi_go_closure_win64
ffi_go_closure_win64:
	cfi_startproc
	/* Save all integer arguments into the incoming reg stack space.  */
	movq	arg0, 8(%rsp)
	movq	arg1, 16(%rsp)
	movq	arg2, 24(%rsp)
	movq	arg3, 32(%rsp)

	movq	8(%r10), arg0			/* load cif */
	movq	16(%r10), arg1			/* load fun */
	movq	%r10, arg2			/* closure is user_data */
	jmp	0f
	cfi_endproc
	.seh_endproc

	.align	8
	.globl	ffi_closure_win64

	.seh_proc ffi_closure_win64
ffi_closure_win64:
	cfi_startproc
	/* Save all integer arguments into the incoming reg stack space.  */
	movq	arg0, 8(%rsp)
	movq	arg1, 16(%rsp)
	movq	arg2, 24(%rsp)
	movq	arg3, 32(%rsp)

	movq	FFI_TRAMPOLINE_SIZE(%r10), arg0		/* load cif */
	movq	FFI_TRAMPOLINE_SIZE+8(%r10), arg1	/* load fun */
	movq	FFI_TRAMPOLINE_SIZE+16(%r10), arg2	/* load user_data */
0:
	subq	$ffi_clo_FS, %rsp
	cfi_adjust_cfa_offset(ffi_clo_FS)
	.seh_stackalloc ffi_clo_FS
	.seh_endprologue

	/* Save all sse arguments into the stack frame.  */
	movsd	%xmm0, ffi_clo_OFF_X(%rsp)
	movsd	%xmm1, ffi_clo_OFF_X+8(%rsp)
	movsd	%xmm2, ffi_clo_OFF_X+16(%rsp)
	movsd	%xmm3, ffi_clo_OFF_X+24(%rsp)

	leaq	ffi_clo_OFF_R(%rsp), arg3
	call	ffi_closure_win64_inner

	/* Load the result into both possible result registers.  */
	movq    ffi_clo_OFF_R(%rsp), %rax
	movsd   ffi_clo_OFF_R(%rsp), %xmm0

	addq	$ffi_clo_FS, %rsp
	cfi_adjust_cfa_offset(-ffi_clo_FS)
	ret

	cfi_endproc
	.seh_endproc