qiita_org 0.1.24 → 0.1.25

data/gems/ruby/2.7.0/gems/ffi-1.13.1/ext/ffi_c/libffi/src/powerpc/ffi.c

@@ -0,0 +1,175 @@
+/* -----------------------------------------------------------------------
+   ffi.c - Copyright (C) 2013 IBM
+           Copyright (C) 2011 Anthony Green
+           Copyright (C) 2011 Kyle Moffett
+           Copyright (C) 2008 Red Hat, Inc
+           Copyright (C) 2007, 2008 Free Software Foundation, Inc
+	   Copyright (c) 1998 Geoffrey Keating
+
+   PowerPC Foreign Function Interface
+
+   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.
+   ----------------------------------------------------------------------- */
+
+#include "ffi.h"
+#include "ffi_common.h"
+#include "ffi_powerpc.h"
+
+#if HAVE_LONG_DOUBLE_VARIANT
+/* Adjust ffi_type_longdouble.  */
+void FFI_HIDDEN
+ffi_prep_types (ffi_abi abi)
+{
+# if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+#  ifdef POWERPC64
+  ffi_prep_types_linux64 (abi);
+#  else
+  ffi_prep_types_sysv (abi);
+#  endif
+# endif
+}
+#endif
+
+/* Perform machine dependent cif processing */
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep (ffi_cif *cif)
+{
+#ifdef POWERPC64
+  return ffi_prep_cif_linux64 (cif);
+#else
+  return ffi_prep_cif_sysv (cif);
+#endif
+}
+
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep_var (ffi_cif *cif,
+			  unsigned int nfixedargs MAYBE_UNUSED,
+			  unsigned int ntotalargs MAYBE_UNUSED)
+{
+#ifdef POWERPC64
+  return ffi_prep_cif_linux64_var (cif, nfixedargs, ntotalargs);
+#else
+  return ffi_prep_cif_sysv (cif);
+#endif
+}
+
+static void
+ffi_call_int (ffi_cif *cif,
+	      void (*fn) (void),
+	      void *rvalue,
+	      void **avalue,
+	      void *closure)
+{
+  /* The final SYSV ABI says that structures smaller or equal 8 bytes
+     are returned in r3/r4.  A draft ABI used by linux instead returns
+     them in memory.
+
+     We bounce-buffer SYSV small struct return values so that sysv.S
+     can write r3 and r4 to memory without worrying about struct size.
+   
+     For ELFv2 ABI, use a bounce buffer for homogeneous structs too,
+     for similar reasons. This bounce buffer must be aligned to 16
+     bytes for use with homogeneous structs of vectors (float128).  */
+  float128 smst_buffer[8];
+  extended_cif ecif;
+
+  ecif.cif = cif;
+  ecif.avalue = avalue;
+
+  ecif.rvalue = rvalue;
+  if ((cif->flags & FLAG_RETURNS_SMST) != 0)
+    ecif.rvalue = smst_buffer;
+  /* Ensure that we have a valid struct return value.
+     FIXME: Isn't this just papering over a user problem?  */
+  else if (!rvalue && cif->rtype->type == FFI_TYPE_STRUCT)
+    ecif.rvalue = alloca (cif->rtype->size);
+
+#ifdef POWERPC64
+  ffi_call_LINUX64 (&ecif, fn, ecif.rvalue, cif->flags, closure,
+		    -(long) cif->bytes);
+#else
+  ffi_call_SYSV (&ecif, fn, ecif.rvalue, cif->flags, closure, -cif->bytes);
+#endif
+
+  /* Check for a bounce-buffered return value */
+  if (rvalue && ecif.rvalue == smst_buffer)
+    {
+      unsigned int rsize = cif->rtype->size;
+#ifndef __LITTLE_ENDIAN__
+      /* The SYSV ABI returns a structure of up to 4 bytes in size
+	 left-padded in r3.  */
+# ifndef POWERPC64
+      if (rsize <= 4)
+	memcpy (rvalue, (char *) smst_buffer + 4 - rsize, rsize);
+      else
+# endif
+	/* The SYSV ABI returns a structure of up to 8 bytes in size
+	   left-padded in r3/r4, and the ELFv2 ABI similarly returns a
+	   structure of up to 8 bytes in size left-padded in r3. But
+	   note that a structure of a single float is not paddded.  */
+	if (rsize <= 8 && (cif->flags & FLAG_RETURNS_FP) == 0)
+	  memcpy (rvalue, (char *) smst_buffer + 8 - rsize, rsize);
+	else
+#endif
+	  memcpy (rvalue, smst_buffer, rsize);
+    }
+}
+
+void
+ffi_call (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue)
+{
+  ffi_call_int (cif, fn, rvalue, avalue, NULL);
+}
+
+void
+ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue,
+	     void *closure)
+{
+  ffi_call_int (cif, fn, rvalue, avalue, closure);
+}
+
+ffi_status
+ffi_prep_closure_loc (ffi_closure *closure,
+		      ffi_cif *cif,
+		      void (*fun) (ffi_cif *, void *, void **, void *),
+		      void *user_data,
+		      void *codeloc)
+{
+#ifdef POWERPC64
+  return ffi_prep_closure_loc_linux64 (closure, cif, fun, user_data, codeloc);
+#else
+  return ffi_prep_closure_loc_sysv (closure, cif, fun, user_data, codeloc);
+#endif
+}
+
+ffi_status
+ffi_prep_go_closure (ffi_go_closure *closure,
+		     ffi_cif *cif,
+		     void (*fun) (ffi_cif *, void *, void **, void *))
+{
+#ifdef POWERPC64
+  closure->tramp = ffi_go_closure_linux64;
+#else
+  closure->tramp = ffi_go_closure_sysv;
+#endif
+  closure->cif = cif;
+  closure->fun = fun;
+  return FFI_OK;
+}

data/gems/ruby/2.7.0/gems/ffi-1.13.1/ext/ffi_c/libffi/src/powerpc/ffi_darwin.c

@@ -0,0 +1,1440 @@
+/* -----------------------------------------------------------------------
+   ffi_darwin.c
+
+   Copyright (C) 1998 Geoffrey Keating
+   Copyright (C) 2001 John Hornkvist
+   Copyright (C) 2002, 2006, 2007, 2009, 2010 Free Software Foundation, Inc.
+
+   FFI support for Darwin and AIX.
+   
+   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.
+   ----------------------------------------------------------------------- */
+
+#include <ffi.h>
+#include <ffi_common.h>
+
+#include <stdlib.h>
+
+extern void ffi_closure_ASM (void);
+extern void ffi_go_closure_ASM (void);
+
+enum {
+  /* The assembly depends on these exact flags.  
+     For Darwin64 (when FLAG_RETURNS_STRUCT is set):
+       FLAG_RETURNS_FP indicates that the structure embeds FP data.
+       FLAG_RETURNS_128BITS signals a special struct size that is not
+       expanded for float content.  */
+  FLAG_RETURNS_128BITS	= 1 << (31-31), /* These go in cr7  */
+  FLAG_RETURNS_NOTHING	= 1 << (31-30),
+  FLAG_RETURNS_FP	= 1 << (31-29),
+  FLAG_RETURNS_64BITS	= 1 << (31-28),
+
+  FLAG_RETURNS_STRUCT	= 1 << (31-27), /* This goes in cr6  */
+
+  FLAG_ARG_NEEDS_COPY   = 1 << (31- 7),
+  FLAG_FP_ARGUMENTS     = 1 << (31- 6), /* cr1.eq; specified by ABI  */
+  FLAG_4_GPR_ARGUMENTS  = 1 << (31- 5),
+  FLAG_RETVAL_REFERENCE = 1 << (31- 4)
+};
+
+/* About the DARWIN ABI.  */
+enum {
+  NUM_GPR_ARG_REGISTERS = 8,
+  NUM_FPR_ARG_REGISTERS = 13,
+  LINKAGE_AREA_GPRS = 6
+};
+
+enum { ASM_NEEDS_REGISTERS = 4 }; /* r28-r31 */
+
+/* ffi_prep_args is called by the assembly routine once stack space
+   has been allocated for the function's arguments.
+   
+   m32/m64
+
+   The stack layout we want looks like this:
+
+   |   Return address from ffi_call_DARWIN      |	higher addresses
+   |--------------------------------------------|
+   |   Previous backchain pointer	4/8	|	stack pointer here
+   |--------------------------------------------|<+ <<<	on entry to
+   |   ASM_NEEDS_REGISTERS=r28-r31   4*(4/8)	| |	ffi_call_DARWIN
+   |--------------------------------------------| |
+   |   When we have any FP activity... the	| |
+   |   FPRs occupy NUM_FPR_ARG_REGISTERS slots	| |
+   |   here fp13 .. fp1 from high to low addr.	| |
+   ~						~ ~
+   |   Parameters      (at least 8*4/8=32/64)	| | NUM_GPR_ARG_REGISTERS
+   |--------------------------------------------| |
+   |   TOC=R2 (AIX) Reserved (Darwin)   4/8	| |
+   |--------------------------------------------| |	stack	|
+   |   Reserved                       2*4/8	| |	grows	|
+   |--------------------------------------------| |	down	V
+   |   Space for callee's LR		4/8	| |
+   |--------------------------------------------| |	lower addresses
+   |   Saved CR [low word for m64]      4/8	| |
+   |--------------------------------------------| |     stack pointer here
+   |   Current backchain pointer	4/8	|-/	during
+   |--------------------------------------------|   <<<	ffi_call_DARWIN
+
+   */
+
+#if defined(POWERPC_DARWIN64)
+static void
+darwin64_pass_struct_by_value 
+  (ffi_type *, char *, unsigned, unsigned *, double **, unsigned long **);
+#endif
+
+/* This depends on GPR_SIZE = sizeof (unsigned long) */
+
+void
+ffi_prep_args (extended_cif *ecif, unsigned long *const stack)
+{
+  const unsigned bytes = ecif->cif->bytes;
+  const unsigned flags = ecif->cif->flags;
+  const unsigned nargs = ecif->cif->nargs;
+#if !defined(POWERPC_DARWIN64) 
+  const ffi_abi abi = ecif->cif->abi;
+#endif
+
+  /* 'stacktop' points at the previous backchain pointer.  */
+  unsigned long *const stacktop = stack + (bytes / sizeof(unsigned long));
+
+  /* 'fpr_base' points at the space for fpr1, and grows upwards as
+     we use FPR registers.  */
+  double *fpr_base = (double *) (stacktop - ASM_NEEDS_REGISTERS) - NUM_FPR_ARG_REGISTERS;
+  int gp_count = 0, fparg_count = 0;
+
+  /* 'next_arg' grows up as we put parameters in it.  */
+  unsigned long *next_arg = stack + LINKAGE_AREA_GPRS; /* 6 reserved positions.  */
+
+  int i;
+  double double_tmp;
+  void **p_argv = ecif->avalue;
+  unsigned long gprvalue;
+  ffi_type** ptr = ecif->cif->arg_types;
+#if !defined(POWERPC_DARWIN64) 
+  char *dest_cpy;
+#endif
+  unsigned size_al = 0;
+
+  /* Check that everything starts aligned properly.  */
+  FFI_ASSERT(((unsigned) (char *) stack & 0xF) == 0);
+  FFI_ASSERT(((unsigned) (char *) stacktop & 0xF) == 0);
+  FFI_ASSERT((bytes & 0xF) == 0);
+
+  /* Deal with return values that are actually pass-by-reference.
+     Rule:
+     Return values are referenced by r3, so r4 is the first parameter.  */
+
+  if (flags & FLAG_RETVAL_REFERENCE)
+    *next_arg++ = (unsigned long) (char *) ecif->rvalue;
+
+  /* Now for the arguments.  */
+  for (i = nargs; i > 0; i--, ptr++, p_argv++)
+    {
+      switch ((*ptr)->type)
+	{
+	/* If a floating-point parameter appears before all of the general-
+	   purpose registers are filled, the corresponding GPRs that match
+	   the size of the floating-point parameter are skipped.  */
+	case FFI_TYPE_FLOAT:
+	  double_tmp = *(float *) *p_argv;
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *fpr_base++ = double_tmp;
+#if defined(POWERPC_DARWIN)
+	  *(float *)next_arg = *(float *) *p_argv;
+#else
+	  *(double *)next_arg = double_tmp;
+#endif
+	  next_arg++;
+	  gp_count++;
+	  fparg_count++;
+	  FFI_ASSERT(flags & FLAG_FP_ARGUMENTS);
+	  break;
+
+	case FFI_TYPE_DOUBLE:
+	  double_tmp = *(double *) *p_argv;
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *fpr_base++ = double_tmp;
+	  *(double *)next_arg = double_tmp;
+#ifdef POWERPC64
+	  next_arg++;
+	  gp_count++;
+#else
+	  next_arg += 2;
+	  gp_count += 2;
+#endif
+	  fparg_count++;
+	  FFI_ASSERT(flags & FLAG_FP_ARGUMENTS);
+	  break;
+
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+
+	case FFI_TYPE_LONGDOUBLE:
+#  if defined(POWERPC64) && !defined(POWERPC_DARWIN64)
+	  /* ??? This will exceed the regs count when the value starts at fp13
+	     and it will not put the extra bit on the stack.  */
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *(long double *) fpr_base++ = *(long double *) *p_argv;
+	  else
+	    *(long double *) next_arg = *(long double *) *p_argv;
+	  next_arg += 2;
+	  fparg_count += 2;
+#  else
+	  double_tmp = ((double *) *p_argv)[0];
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *fpr_base++ = double_tmp;
+	  *(double *) next_arg = double_tmp;
+#    if defined(POWERPC_DARWIN64)
+	  next_arg++;
+	  gp_count++;
+#    else
+	  next_arg += 2;
+	  gp_count += 2;
+#    endif
+	  fparg_count++;
+	  double_tmp = ((double *) *p_argv)[1];
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *fpr_base++ = double_tmp;
+	  *(double *) next_arg = double_tmp;
+#    if defined(POWERPC_DARWIN64)
+	  next_arg++;
+	  gp_count++;
+#    else
+	  next_arg += 2;
+	  gp_count += 2;
+#    endif
+	  fparg_count++;
+#  endif
+	  FFI_ASSERT(flags & FLAG_FP_ARGUMENTS);
+	  break;
+#endif
+	case FFI_TYPE_UINT64:
+	case FFI_TYPE_SINT64:
+#ifdef POWERPC64
+	  gprvalue = *(long long *) *p_argv;
+	  goto putgpr;
+#else
+	  *(long long *) next_arg = *(long long *) *p_argv;
+	  next_arg += 2;
+	  gp_count += 2;
+#endif
+	  break;
+	case FFI_TYPE_POINTER:
+	  gprvalue = *(unsigned long *) *p_argv;
+	  goto putgpr;
+	case FFI_TYPE_UINT8:
+	  gprvalue = *(unsigned char *) *p_argv;
+	  goto putgpr;
+	case FFI_TYPE_SINT8:
+	  gprvalue = *(signed char *) *p_argv;
+	  goto putgpr;
+	case FFI_TYPE_UINT16:
+	  gprvalue = *(unsigned short *) *p_argv;
+	  goto putgpr;
+	case FFI_TYPE_SINT16:
+	  gprvalue = *(signed short *) *p_argv;
+	  goto putgpr;
+
+	case FFI_TYPE_STRUCT:
+	  size_al = (*ptr)->size;
+#if defined(POWERPC_DARWIN64)
+	  next_arg = (unsigned long *)FFI_ALIGN((char *)next_arg, (*ptr)->alignment);
+	  darwin64_pass_struct_by_value (*ptr, (char *) *p_argv, 
+					 (unsigned) size_al,
+					 (unsigned int *) &fparg_count,
+					 &fpr_base, &next_arg);
+#else
+	  dest_cpy = (char *) next_arg;
+
+	  /* If the first member of the struct is a double, then include enough
+	     padding in the struct size to align it to double-word.  */
+	  if ((*ptr)->elements[0]->type == FFI_TYPE_DOUBLE)
+	    size_al = FFI_ALIGN((*ptr)->size, 8);
+
+#  if defined(POWERPC64) 
+	  FFI_ASSERT (abi != FFI_DARWIN);
+	  memcpy ((char *) dest_cpy, (char *) *p_argv, size_al);
+	  next_arg += (size_al + 7) / 8;
+#  else
+	  /* Structures that match the basic modes (QI 1 byte, HI 2 bytes,
+	     SI 4 bytes) are aligned as if they were those modes.
+	     Structures with 3 byte in size are padded upwards.  */
+	  if (size_al < 3 && abi == FFI_DARWIN)
+	    dest_cpy += 4 - size_al;
+
+	  memcpy((char *) dest_cpy, (char *) *p_argv, size_al);
+	  next_arg += (size_al + 3) / 4;
+#  endif
+#endif
+	  break;
+
+	case FFI_TYPE_INT:
+	case FFI_TYPE_SINT32:
+	  gprvalue = *(signed int *) *p_argv;
+	  goto putgpr;
+
+	case FFI_TYPE_UINT32:
+	  gprvalue = *(unsigned int *) *p_argv;
+	putgpr:
+	  *next_arg++ = gprvalue;
+	  gp_count++;
+	  break;
+	default:
+	  break;
+	}
+    }
+
+  /* Check that we didn't overrun the stack...  */
+  /* FFI_ASSERT(gpr_base <= stacktop - ASM_NEEDS_REGISTERS);
+     FFI_ASSERT((unsigned *)fpr_base
+     	     <= stacktop - ASM_NEEDS_REGISTERS - NUM_GPR_ARG_REGISTERS);
+     FFI_ASSERT(flags & FLAG_4_GPR_ARGUMENTS || intarg_count <= 4);  */
+}
+
+#if defined(POWERPC_DARWIN64)
+
+/* See if we can put some of the struct into fprs.
+   This should not be called for structures of size 16 bytes, since these are not
+   broken out this way.  */
+static void
+darwin64_scan_struct_for_floats (ffi_type *s, unsigned *nfpr)
+{
+  int i;
+
+  FFI_ASSERT (s->type == FFI_TYPE_STRUCT)
+
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      ffi_type *p = s->elements[i];
+      switch (p->type)
+	{
+	  case FFI_TYPE_STRUCT:
+	    darwin64_scan_struct_for_floats (p, nfpr);
+	    break;
+	  case FFI_TYPE_LONGDOUBLE:
+	    (*nfpr) += 2;
+	    break;
+	  case FFI_TYPE_DOUBLE:
+	  case FFI_TYPE_FLOAT:
+	    (*nfpr) += 1;
+	    break;
+	  default:
+	    break;    
+	}
+    }
+}
+
+static int
+darwin64_struct_size_exceeds_gprs_p (ffi_type *s, char *src, unsigned *nfpr)
+{
+  unsigned struct_offset=0, i;
+
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      char *item_base;
+      ffi_type *p = s->elements[i];
+      /* Find the start of this item (0 for the first one).  */
+      if (i > 0)
+        struct_offset = FFI_ALIGN(struct_offset, p->alignment);
+
+      item_base = src + struct_offset;
+
+      switch (p->type)
+	{
+	  case FFI_TYPE_STRUCT:
+	    if (darwin64_struct_size_exceeds_gprs_p (p, item_base, nfpr))
+	      return 1;
+	    break;
+	  case FFI_TYPE_LONGDOUBLE:
+	    if (*nfpr >= NUM_FPR_ARG_REGISTERS)
+	      return 1;
+	    (*nfpr) += 1;
+	    item_base += 8;
+	  /* FALL THROUGH */
+	  case FFI_TYPE_DOUBLE:
+	    if (*nfpr >= NUM_FPR_ARG_REGISTERS)
+	      return 1;
+	    (*nfpr) += 1;
+	    break;
+	  case FFI_TYPE_FLOAT:
+	    if (*nfpr >= NUM_FPR_ARG_REGISTERS)
+	      return 1;
+	    (*nfpr) += 1;
+	    break;
+	  default:
+	    /* If we try and place any item, that is non-float, once we've
+	       exceeded the 8 GPR mark, then we can't fit the struct.  */
+	    if ((unsigned long)item_base >= 8*8) 
+	      return 1;
+	    break;    
+	}
+      /* now count the size of what we just used.  */
+      struct_offset += p->size;
+    }
+  return 0;
+}
+
+/* Can this struct be returned by value?  */
+int 
+darwin64_struct_ret_by_value_p (ffi_type *s)
+{
+  unsigned nfp = 0;
+
+  FFI_ASSERT (s && s->type == FFI_TYPE_STRUCT);
+  
+  /* The largest structure we can return is 8long + 13 doubles.  */
+  if (s->size > 168)
+    return 0;
+  
+  /* We can't pass more than 13 floats.  */
+  darwin64_scan_struct_for_floats (s, &nfp);
+  if (nfp > 13)
+    return 0;
+  
+  /* If there are not too many floats, and the struct is
+     small enough to accommodate in the GPRs, then it must be OK.  */
+  if (s->size <= 64)
+    return 1;
+  
+  /* Well, we have to look harder.  */
+  nfp = 0;
+  if (darwin64_struct_size_exceeds_gprs_p (s, NULL, &nfp))
+    return 0;
+  
+  return 1;
+}
+
+void
+darwin64_pass_struct_floats (ffi_type *s, char *src, 
+			     unsigned *nfpr, double **fprs)
+{
+  int i;
+  double *fpr_base = *fprs;
+  unsigned struct_offset = 0;
+
+  /* We don't assume anything about the alignment of the source.  */
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      char *item_base;
+      ffi_type *p = s->elements[i];
+      /* Find the start of this item (0 for the first one).  */
+      if (i > 0)
+        struct_offset = FFI_ALIGN(struct_offset, p->alignment);
+      item_base = src + struct_offset;
+
+      switch (p->type)
+	{
+	  case FFI_TYPE_STRUCT:
+	    darwin64_pass_struct_floats (p, item_base, nfpr,
+					   &fpr_base);
+	    break;
+	  case FFI_TYPE_LONGDOUBLE:
+	    if (*nfpr < NUM_FPR_ARG_REGISTERS)
+	      *fpr_base++ = *(double *)item_base;
+	    (*nfpr) += 1;
+	    item_base += 8;
+	  /* FALL THROUGH */
+	  case FFI_TYPE_DOUBLE:
+	    if (*nfpr < NUM_FPR_ARG_REGISTERS)
+	      *fpr_base++ = *(double *)item_base;
+	    (*nfpr) += 1;
+	    break;
+	  case FFI_TYPE_FLOAT:
+	    if (*nfpr < NUM_FPR_ARG_REGISTERS)
+	      *fpr_base++ = (double) *(float *)item_base;
+	    (*nfpr) += 1;
+	    break;
+	  default:
+	    break;    
+	}
+      /* now count the size of what we just used.  */
+      struct_offset += p->size;
+    }
+  /* Update the scores.  */
+  *fprs = fpr_base;
+}
+
+/* Darwin64 special rules.
+   Break out a struct into params and float registers.  */
+static void
+darwin64_pass_struct_by_value (ffi_type *s, char *src, unsigned size,
+			       unsigned *nfpr, double **fprs, unsigned long **arg)
+{
+  unsigned long *next_arg = *arg;
+  char *dest_cpy = (char *)next_arg;
+
+  FFI_ASSERT (s->type == FFI_TYPE_STRUCT)
+
+  if (!size)
+    return;
+
+  /* First... special cases.  */
+  if (size < 3
+      || (size == 4 
+	  && s->elements[0] 
+	  && s->elements[0]->type != FFI_TYPE_FLOAT))
+    {
+      /* Must be at least one GPR, padding is unspecified in value, 
+	 let's make it zero.  */
+      *next_arg = 0UL; 
+      dest_cpy += 8 - size;
+      memcpy ((char *) dest_cpy, src, size);
+      next_arg++;
+    }
+  else if (size == 16)
+    {
+      memcpy ((char *) dest_cpy, src, size);
+      next_arg += 2;
+    }
+  else
+    {
+      /* now the general case, we consider embedded floats.  */
+      memcpy ((char *) dest_cpy, src, size);
+      darwin64_pass_struct_floats (s, src, nfpr, fprs);
+      next_arg += (size+7)/8;
+    }
+    
+  *arg = next_arg;
+}
+
+double *
+darwin64_struct_floats_to_mem (ffi_type *s, char *dest, double *fprs, unsigned *nf)
+{
+  int i;
+  unsigned struct_offset = 0;
+
+  /* We don't assume anything about the alignment of the source.  */
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      char *item_base;
+      ffi_type *p = s->elements[i];
+      /* Find the start of this item (0 for the first one).  */
+      if (i > 0)
+        struct_offset = FFI_ALIGN(struct_offset, p->alignment);
+      item_base = dest + struct_offset;
+
+      switch (p->type)
+	{
+	  case FFI_TYPE_STRUCT:
+	    fprs = darwin64_struct_floats_to_mem (p, item_base, fprs, nf);
+	    break;
+	  case FFI_TYPE_LONGDOUBLE:
+	    if (*nf < NUM_FPR_ARG_REGISTERS)
+	      {
+		*(double *)item_base = *fprs++ ;
+		(*nf) += 1;
+	      }
+	    item_base += 8;
+	  /* FALL THROUGH */
+	  case FFI_TYPE_DOUBLE:
+	    if (*nf < NUM_FPR_ARG_REGISTERS)
+	      {
+		*(double *)item_base = *fprs++ ;
+		(*nf) += 1;
+	      }
+	    break;
+	  case FFI_TYPE_FLOAT:
+	    if (*nf < NUM_FPR_ARG_REGISTERS)
+	      {
+		*(float *)item_base = (float) *fprs++ ;
+		(*nf) += 1;
+	      }
+	    break;
+	  default:
+	    break;    
+	}
+      /* now count the size of what we just used.  */
+      struct_offset += p->size;
+    }
+  return fprs;
+}
+
+#endif
+
+/* Adjust the size of S to be correct for Darwin.
+   On Darwin m32, the first field of a structure has natural alignment.  
+   On Darwin m64, all fields have natural alignment.  */
+
+static void
+darwin_adjust_aggregate_sizes (ffi_type *s)
+{
+  int i;
+
+  if (s->type != FFI_TYPE_STRUCT)
+    return;
+
+  s->size = 0;
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      ffi_type *p;
+      int align;
+      
+      p = s->elements[i];
+      if (p->type == FFI_TYPE_STRUCT)
+	darwin_adjust_aggregate_sizes (p);
+#if defined(POWERPC_DARWIN64)
+      /* Natural alignment for all items.  */
+      align = p->alignment;
+#else
+      /* Natural alignment for the first item... */
+      if (i == 0)
+	align = p->alignment;
+      else if (p->alignment == 16 || p->alignment < 4)
+	/* .. subsequent items with vector or align < 4 have natural align.  */
+	align = p->alignment;
+      else
+	/* .. or align is 4.  */
+	align = 4;
+#endif
+      /* Pad, if necessary, before adding the current item.  */
+      s->size = FFI_ALIGN(s->size, align) + p->size;
+    }
+  
+  s->size = FFI_ALIGN(s->size, s->alignment);
+  
+  /* This should not be necessary on m64, but harmless.  */
+  if (s->elements[0]->type == FFI_TYPE_UINT64
+      || s->elements[0]->type == FFI_TYPE_SINT64
+      || s->elements[0]->type == FFI_TYPE_DOUBLE
+      || s->elements[0]->alignment == 8)
+    s->alignment = s->alignment > 8 ? s->alignment : 8;
+  /* Do not add additional tail padding.  */
+}
+
+/* Adjust the size of S to be correct for AIX.
+   Word-align double unless it is the first member of a structure.  */
+
+static void
+aix_adjust_aggregate_sizes (ffi_type *s)
+{
+  int i;
+
+  if (s->type != FFI_TYPE_STRUCT)
+    return;
+
+  s->size = 0;
+  for (i = 0; s->elements[i] != NULL; i++)
+    {
+      ffi_type *p;
+      int align;
+      
+      p = s->elements[i];
+      aix_adjust_aggregate_sizes (p);
+      align = p->alignment;
+      if (i != 0 && p->type == FFI_TYPE_DOUBLE)
+	align = 4;
+      s->size = FFI_ALIGN(s->size, align) + p->size;
+    }
+  
+  s->size = FFI_ALIGN(s->size, s->alignment);
+  
+  if (s->elements[0]->type == FFI_TYPE_UINT64
+      || s->elements[0]->type == FFI_TYPE_SINT64
+      || s->elements[0]->type == FFI_TYPE_DOUBLE
+      || s->elements[0]->alignment == 8)
+    s->alignment = s->alignment > 8 ? s->alignment : 8;
+  /* Do not add additional tail padding.  */
+}
+
+/* Perform machine dependent cif processing.  */
+ffi_status
+ffi_prep_cif_machdep (ffi_cif *cif)
+{
+  /* All this is for the DARWIN ABI.  */
+  unsigned i;
+  ffi_type **ptr;
+  unsigned bytes;
+  unsigned fparg_count = 0, intarg_count = 0;
+  unsigned flags = 0;
+  unsigned size_al = 0;
+
+  /* All the machine-independent calculation of cif->bytes will be wrong.
+     All the calculation of structure sizes will also be wrong.
+     Redo the calculation for DARWIN.  */
+
+  if (cif->abi == FFI_DARWIN)
+    {
+      darwin_adjust_aggregate_sizes (cif->rtype);
+      for (i = 0; i < cif->nargs; i++)
+	darwin_adjust_aggregate_sizes (cif->arg_types[i]);
+    }
+
+  if (cif->abi == FFI_AIX)
+    {
+      aix_adjust_aggregate_sizes (cif->rtype);
+      for (i = 0; i < cif->nargs; i++)
+	aix_adjust_aggregate_sizes (cif->arg_types[i]);
+    }
+
+  /* Space for the frame pointer, callee's LR, CR, etc, and for
+     the asm's temp regs.  */
+
+  bytes = (LINKAGE_AREA_GPRS + ASM_NEEDS_REGISTERS) * sizeof(unsigned long);
+
+  /* Return value handling.  
+    The rules m32 are as follows:
+     - 32-bit (or less) integer values are returned in gpr3;
+     - structures of size <= 4 bytes also returned in gpr3;
+     - 64-bit integer values [??? and structures between 5 and 8 bytes] are
+       returned in gpr3 and gpr4;
+     - Single/double FP values are returned in fpr1;
+     - Long double FP (if not equivalent to double) values are returned in
+       fpr1 and fpr2;
+     m64:
+     - 64-bit or smaller integral values are returned in GPR3
+     - Single/double FP values are returned in fpr1;
+     - Long double FP values are returned in fpr1 and fpr2;
+     m64 Structures:
+     - If the structure could be accommodated in registers were it to be the
+       first argument to a routine, then it is returned in those registers.
+     m32/m64 structures otherwise:
+     - Larger structures values are allocated space and a pointer is passed
+       as the first argument.  */
+  switch (cif->rtype->type)
+    {
+
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+    case FFI_TYPE_LONGDOUBLE:
+      flags |= FLAG_RETURNS_128BITS;
+      flags |= FLAG_RETURNS_FP;
+      break;
+#endif
+
+    case FFI_TYPE_DOUBLE:
+      flags |= FLAG_RETURNS_64BITS;
+      /* Fall through.  */
+    case FFI_TYPE_FLOAT:
+      flags |= FLAG_RETURNS_FP;
+      break;
+
+    case FFI_TYPE_UINT64:
+    case FFI_TYPE_SINT64:
+#ifdef POWERPC64
+    case FFI_TYPE_POINTER:
+#endif
+      flags |= FLAG_RETURNS_64BITS;
+      break;
+
+    case FFI_TYPE_STRUCT:
+#if defined(POWERPC_DARWIN64)
+      {
+	/* Can we fit the struct into regs?  */
+	if (darwin64_struct_ret_by_value_p (cif->rtype))
+	  {
+	    unsigned nfpr = 0;
+	    flags |= FLAG_RETURNS_STRUCT;
+	    if (cif->rtype->size != 16)
+	      darwin64_scan_struct_for_floats (cif->rtype, &nfpr) ;
+	    else
+	      flags |= FLAG_RETURNS_128BITS;
+	    /* Will be 0 for 16byte struct.  */
+	    if (nfpr)
+	      flags |= FLAG_RETURNS_FP;
+	  }
+	else /* By ref. */
+	  {
+	    flags |= FLAG_RETVAL_REFERENCE;
+	    flags |= FLAG_RETURNS_NOTHING;
+	    intarg_count++;
+	  }
+      }
+#elif defined(DARWIN_PPC)
+      if (cif->rtype->size <= 4)
+	flags |= FLAG_RETURNS_STRUCT;
+      else /* else by reference.  */
+	{
+	  flags |= FLAG_RETVAL_REFERENCE;
+	  flags |= FLAG_RETURNS_NOTHING;
+	  intarg_count++;
+	}
+#else /* assume we pass by ref.  */
+      flags |= FLAG_RETVAL_REFERENCE;
+      flags |= FLAG_RETURNS_NOTHING;
+      intarg_count++;
+#endif
+      break;
+    case FFI_TYPE_VOID:
+      flags |= FLAG_RETURNS_NOTHING;
+      break;
+
+    default:
+      /* Returns 32-bit integer, or similar.  Nothing to do here.  */
+      break;
+    }
+
+  /* The first NUM_GPR_ARG_REGISTERS words of integer arguments, and the
+     first NUM_FPR_ARG_REGISTERS fp arguments, go in registers; the rest
+     goes on the stack.  
+     ??? Structures are passed as a pointer to a copy of the structure. 
+     Stuff on the stack needs to keep proper alignment.  
+     For m64 the count is effectively of half-GPRs.  */
+  for (ptr = cif->arg_types, i = cif->nargs; i > 0; i--, ptr++)
+    {
+      unsigned align_words;
+      switch ((*ptr)->type)
+	{
+	case FFI_TYPE_FLOAT:
+	case FFI_TYPE_DOUBLE:
+	  fparg_count++;
+#if !defined(POWERPC_DARWIN64)
+	  /* If this FP arg is going on the stack, it must be
+	     8-byte-aligned.  */
+	  if (fparg_count > NUM_FPR_ARG_REGISTERS
+	      && (intarg_count & 0x01) != 0)
+	    intarg_count++;
+#endif
+	  break;
+
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+	case FFI_TYPE_LONGDOUBLE:
+	  fparg_count += 2;
+	  /* If this FP arg is going on the stack, it must be
+	     16-byte-aligned.  */
+	  if (fparg_count >= NUM_FPR_ARG_REGISTERS)
+#if defined (POWERPC64)
+	    intarg_count = FFI_ALIGN(intarg_count, 2);
+#else
+	    intarg_count = FFI_ALIGN(intarg_count, 4);
+#endif
+	  break;
+#endif
+
+	case FFI_TYPE_UINT64:
+	case FFI_TYPE_SINT64:
+#if defined(POWERPC64)
+	  intarg_count++;
+#else
+	  /* 'long long' arguments are passed as two words, but
+	     either both words must fit in registers or both go
+	     on the stack.  If they go on the stack, they must
+	     be 8-byte-aligned.  */
+	  if (intarg_count == NUM_GPR_ARG_REGISTERS-1
+	      || (intarg_count >= NUM_GPR_ARG_REGISTERS 
+	          && (intarg_count & 0x01) != 0))
+	    intarg_count++;
+	  intarg_count += 2;
+#endif
+	  break;
+
+	case FFI_TYPE_STRUCT:
+	  size_al = (*ptr)->size;
+#if defined(POWERPC_DARWIN64)
+	  align_words = (*ptr)->alignment >> 3;
+	  if (align_words)
+	    intarg_count = FFI_ALIGN(intarg_count, align_words);
+	  /* Base size of the struct.  */
+	  intarg_count += (size_al + 7) / 8;
+	  /* If 16 bytes then don't worry about floats.  */
+	  if (size_al != 16)
+	    /* Scan through for floats to be placed in regs.  */
+	    darwin64_scan_struct_for_floats (*ptr, &fparg_count) ;
+#else
+	  align_words = (*ptr)->alignment >> 2;
+	  if (align_words)
+	    intarg_count = FFI_ALIGN(intarg_count, align_words);
+	  /* If the first member of the struct is a double, then align
+	     the struct to double-word. 
+	  if ((*ptr)->elements[0]->type == FFI_TYPE_DOUBLE)
+	    size_al = FFI_ALIGN((*ptr)->size, 8); */
+#  ifdef POWERPC64
+	  intarg_count += (size_al + 7) / 8;
+#  else
+	  intarg_count += (size_al + 3) / 4;
+#  endif
+#endif
+	  break;
+
+	default:
+	  /* Everything else is passed as a 4-byte word in a GPR, either
+	     the object itself or a pointer to it.  */
+	  intarg_count++;
+	  break;
+	}
+    }
+
+  if (fparg_count != 0)
+    flags |= FLAG_FP_ARGUMENTS;
+
+#if defined(POWERPC_DARWIN64)
+  /* Space to image the FPR registers, if needed - which includes when they might be
+     used in a struct return.  */
+  if (fparg_count != 0 
+      || ((flags & FLAG_RETURNS_STRUCT)
+	   && (flags & FLAG_RETURNS_FP)))
+    bytes += NUM_FPR_ARG_REGISTERS * sizeof(double);
+#else
+  /* Space for the FPR registers, if needed.  */
+  if (fparg_count != 0)
+    bytes += NUM_FPR_ARG_REGISTERS * sizeof(double);
+#endif
+
+  /* Stack space.  */
+#ifdef POWERPC64
+  if ((intarg_count + fparg_count) > NUM_GPR_ARG_REGISTERS)
+    bytes += (intarg_count + fparg_count) * sizeof(long);
+#else
+  if ((intarg_count + 2 * fparg_count) > NUM_GPR_ARG_REGISTERS)
+    bytes += (intarg_count + 2 * fparg_count) * sizeof(long);
+#endif
+  else
+    bytes += NUM_GPR_ARG_REGISTERS * sizeof(long);
+
+  /* The stack space allocated needs to be a multiple of 16 bytes.  */
+  bytes = FFI_ALIGN(bytes, 16) ;
+
+  cif->flags = flags;
+  cif->bytes = bytes;
+
+  return FFI_OK;
+}
+
+extern void ffi_call_AIX(extended_cif *, long, unsigned, unsigned *,
+			 void (*fn)(void), void (*fn2)(void));
+
+extern void ffi_call_go_AIX(extended_cif *, long, unsigned, unsigned *,
+			    void (*fn)(void), void (*fn2)(void), void *closure);
+
+extern void ffi_call_DARWIN(extended_cif *, long, unsigned, unsigned *,
+			    void (*fn)(void), void (*fn2)(void), ffi_type*);
+
+void
+ffi_call (ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
+{
+  extended_cif ecif;
+
+  ecif.cif = cif;
+  ecif.avalue = avalue;
+
+  /* If the return value is a struct and we don't have a return
+     value address then we need to make one.  */
+
+  if ((rvalue == NULL) &&
+      (cif->rtype->type == FFI_TYPE_STRUCT))
+    {
+      ecif.rvalue = alloca (cif->rtype->size);
+    }
+  else
+    ecif.rvalue = rvalue;
+
+  switch (cif->abi)
+    {
+    case FFI_AIX:
+      ffi_call_AIX(&ecif, -(long)cif->bytes, cif->flags, ecif.rvalue, fn,
+		   FFI_FN(ffi_prep_args));
+      break;
+    case FFI_DARWIN:
+      ffi_call_DARWIN(&ecif, -(long)cif->bytes, cif->flags, ecif.rvalue, fn,
+		      FFI_FN(ffi_prep_args), cif->rtype);
+      break;
+    default:
+      FFI_ASSERT(0);
+      break;
+    }
+}
+
+void
+ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue,
+	     void *closure)
+{
+  extended_cif ecif;
+
+  ecif.cif = cif;
+  ecif.avalue = avalue;
+
+  /* If the return value is a struct and we don't have a return
+     value address then we need to make one.  */
+
+  if ((rvalue == NULL) &&
+      (cif->rtype->type == FFI_TYPE_STRUCT))
+    {
+      ecif.rvalue = alloca (cif->rtype->size);
+    }
+  else
+    ecif.rvalue = rvalue;
+
+  switch (cif->abi)
+    {
+    case FFI_AIX:
+      ffi_call_go_AIX(&ecif, -(long)cif->bytes, cif->flags, ecif.rvalue, fn,
+		      FFI_FN(ffi_prep_args), closure);
+      break;
+    default:
+      FFI_ASSERT(0);
+      break;
+    }
+}
+
+static void flush_icache(char *);
+static void flush_range(char *, int);
+
+/* The layout of a function descriptor.  A C function pointer really
+   points to one of these.  */
+
+typedef struct aix_fd_struct {
+  void *code_pointer;
+  void *toc;
+} aix_fd;
+
+/* here I'd like to add the stack frame layout we use in darwin_closure.S
+   and aix_closure.S
+
+   m32/m64
+
+   The stack layout looks like this:
+
+   |   Additional params...			| |     Higher address
+   ~						~ ~
+   |   Parameters      (at least 8*4/8=32/64)	| | NUM_GPR_ARG_REGISTERS
+   |--------------------------------------------| |
+   |   TOC=R2 (AIX) Reserved (Darwin)   4/8	| |
+   |--------------------------------------------| |
+   |   Reserved                       2*4/8	| |
+   |--------------------------------------------| |
+   |   Space for callee's LR		4/8	| |
+   |--------------------------------------------| |
+   |   Saved CR [low word for m64]      4/8	| |
+   |--------------------------------------------| |
+   |   Current backchain pointer	4/8	|-/ Parent's frame.
+   |--------------------------------------------| <+ <<< on entry to ffi_closure_ASM
+   |   Result Bytes			16	| |
+   |--------------------------------------------| |
+   ~   padding to 16-byte alignment		~ ~
+   |--------------------------------------------| |
+   |   NUM_FPR_ARG_REGISTERS slots		| |
+   |   here fp13 .. fp1		       13*8	| |
+   |--------------------------------------------| |
+   |   R3..R10			  8*4/8=32/64	| | NUM_GPR_ARG_REGISTERS
+   |--------------------------------------------| |
+   |   TOC=R2 (AIX) Reserved (Darwin)   4/8	| |
+   |--------------------------------------------| |	stack	|
+   |   Reserved [compiler,binder]     2*4/8	| |	grows	|
+   |--------------------------------------------| |	down	V
+   |   Space for callee's LR		4/8	| |
+   |--------------------------------------------| |	lower addresses
+   |   Saved CR [low word for m64]      4/8	| |
+   |--------------------------------------------| |     stack pointer here
+   |   Current backchain pointer	4/8	|-/	during
+   |--------------------------------------------|   <<<	ffi_closure_ASM.
+
+*/
+
+ffi_status
+ffi_prep_closure_loc (ffi_closure* closure,
+		      ffi_cif* cif,
+		      void (*fun)(ffi_cif*, void*, void**, void*),
+		      void *user_data,
+		      void *codeloc)
+{
+  unsigned int *tramp;
+  struct ffi_aix_trampoline_struct *tramp_aix;
+  aix_fd *fd;
+
+  switch (cif->abi)
+    {
+      case FFI_DARWIN:
+
+	FFI_ASSERT (cif->abi == FFI_DARWIN);
+
+	tramp = (unsigned int *) &closure->tramp[0];
+#if defined(POWERPC_DARWIN64)
+	tramp[0] = 0x7c0802a6;  /*   mflr    r0  */
+	tramp[1] = 0x429f0015;  /*   bcl-    20,4*cr7+so,  +0x18 (L1)  */
+	/* We put the addresses here.  */
+	tramp[6] = 0x7d6802a6;  /*L1:   mflr    r11  */
+	tramp[7] = 0xe98b0000;  /*   ld     r12,0(r11) function address  */
+	tramp[8] = 0x7c0803a6;  /*   mtlr    r0   */
+	tramp[9] = 0x7d8903a6;  /*   mtctr   r12  */
+	tramp[10] = 0xe96b0008;  /*   lwz     r11,8(r11) static chain  */
+	tramp[11] = 0x4e800420;  /*   bctr  */
+
+	*((unsigned long *)&tramp[2]) = (unsigned long) ffi_closure_ASM; /* function  */
+	*((unsigned long *)&tramp[4]) = (unsigned long) codeloc; /* context  */
+#else
+	tramp[0] = 0x7c0802a6;  /*   mflr    r0  */
+	tramp[1] = 0x429f000d;  /*   bcl-    20,4*cr7+so,0x10  */
+	tramp[4] = 0x7d6802a6;  /*   mflr    r11  */
+	tramp[5] = 0x818b0000;  /*   lwz     r12,0(r11) function address  */
+	tramp[6] = 0x7c0803a6;  /*   mtlr    r0   */
+	tramp[7] = 0x7d8903a6;  /*   mtctr   r12  */
+	tramp[8] = 0x816b0004;  /*   lwz     r11,4(r11) static chain  */
+	tramp[9] = 0x4e800420;  /*   bctr  */
+	tramp[2] = (unsigned long) ffi_closure_ASM; /* function  */
+	tramp[3] = (unsigned long) codeloc; /* context  */
+#endif
+	closure->cif = cif;
+	closure->fun = fun;
+	closure->user_data = user_data;
+
+	/* Flush the icache. Only necessary on Darwin.  */
+	flush_range(codeloc, FFI_TRAMPOLINE_SIZE);
+
+	break;
+
+    case FFI_AIX:
+
+      tramp_aix = (struct ffi_aix_trampoline_struct *) (closure->tramp);
+      fd = (aix_fd *)(void *)ffi_closure_ASM;
+
+      FFI_ASSERT (cif->abi == FFI_AIX);
+
+      tramp_aix->code_pointer = fd->code_pointer;
+      tramp_aix->toc = fd->toc;
+      tramp_aix->static_chain = codeloc;
+      closure->cif = cif;
+      closure->fun = fun;
+      closure->user_data = user_data;
+      break;
+
+    default:
+      return FFI_BAD_ABI;
+      break;
+    }
+  return FFI_OK;
+}
+
+ffi_status
+ffi_prep_go_closure (ffi_go_closure* closure,
+		     ffi_cif* cif,
+		     void (*fun)(ffi_cif*, void*, void**, void*))
+{
+  switch (cif->abi)
+    {
+      case FFI_AIX:
+
+        FFI_ASSERT (cif->abi == FFI_AIX);
+
+        closure->tramp = (void *)ffi_go_closure_ASM;
+        closure->cif = cif;
+        closure->fun = fun;
+        return FFI_OK;
+      
+      // For now, ffi_prep_go_closure is only implemented for AIX, not for Darwin
+      default:
+        return FFI_BAD_ABI;
+        break;
+    }
+  return FFI_OK;
+}
+
+static void
+flush_icache(char *addr)
+{
+#ifndef _AIX
+  __asm__ volatile (
+		"dcbf 0,%0\n"
+		"\tsync\n"
+		"\ticbi 0,%0\n"
+		"\tsync\n"
+		"\tisync"
+		: : "r"(addr) : "memory");
+#endif
+}
+
+static void
+flush_range(char * addr1, int size)
+{
+#define MIN_LINE_SIZE 32
+  int i;
+  for (i = 0; i < size; i += MIN_LINE_SIZE)
+    flush_icache(addr1+i);
+  flush_icache(addr1+size-1);
+}
+
+typedef union
+{
+  float f;
+  double d;
+} ffi_dblfl;
+
+ffi_type *
+ffi_closure_helper_DARWIN (ffi_closure *, void *,
+			   unsigned long *, ffi_dblfl *);
+
+ffi_type *
+ffi_go_closure_helper_DARWIN (ffi_go_closure*, void *,
+			      unsigned long *, ffi_dblfl *);
+
+/* Basically the trampoline invokes ffi_closure_ASM, and on
+   entry, r11 holds the address of the closure.
+   After storing the registers that could possibly contain
+   parameters to be passed into the stack frame and setting
+   up space for a return value, ffi_closure_ASM invokes the
+   following helper function to do most of the work.  */
+
+static ffi_type *
+ffi_closure_helper_common (ffi_cif* cif,
+			   void (*fun)(ffi_cif*, void*, void**, void*),
+			   void *user_data, void *rvalue,
+			   unsigned long *pgr, ffi_dblfl *pfr)
+{
+  /* rvalue is the pointer to space for return value in closure assembly
+     pgr is the pointer to where r3-r10 are stored in ffi_closure_ASM
+     pfr is the pointer to where f1-f13 are stored in ffi_closure_ASM.  */
+
+  typedef double ldbits[2];
+
+  union ldu
+  {
+    ldbits lb;
+    long double ld;
+  };
+
+  void **          avalue;
+  ffi_type **      arg_types;
+  long             i, avn;
+  ffi_dblfl *      end_pfr = pfr + NUM_FPR_ARG_REGISTERS;
+  unsigned         size_al;
+#if defined(POWERPC_DARWIN64)
+  unsigned 	   fpsused = 0;
+#endif
+
+  avalue = alloca (cif->nargs * sizeof(void *));
+
+  if (cif->rtype->type == FFI_TYPE_STRUCT)
+    {
+#if defined(POWERPC_DARWIN64)
+      if (!darwin64_struct_ret_by_value_p (cif->rtype))
+	{
+    	  /* Won't fit into the regs - return by ref.  */
+	  rvalue = (void *) *pgr;
+	  pgr++;
+	}
+#elif defined(DARWIN_PPC)
+      if (cif->rtype->size > 4)
+	{
+	  rvalue = (void *) *pgr;
+	  pgr++;
+	}
+#else /* assume we return by ref.  */
+      rvalue = (void *) *pgr;
+      pgr++;
+#endif
+    }
+
+  i = 0;
+  avn = cif->nargs;
+  arg_types = cif->arg_types;
+
+  /* Grab the addresses of the arguments from the stack frame.  */
+  while (i < avn)
+    {
+      switch (arg_types[i]->type)
+	{
+	case FFI_TYPE_SINT8:
+	case FFI_TYPE_UINT8:
+#if  defined(POWERPC64)
+	  avalue[i] = (char *) pgr + 7;
+#else
+	  avalue[i] = (char *) pgr + 3;
+#endif
+	  pgr++;
+	  break;
+
+	case FFI_TYPE_SINT16:
+	case FFI_TYPE_UINT16:
+#if  defined(POWERPC64)
+	  avalue[i] = (char *) pgr + 6;
+#else
+	  avalue[i] = (char *) pgr + 2;
+#endif
+	  pgr++;
+	  break;
+
+	case FFI_TYPE_SINT32:
+	case FFI_TYPE_UINT32:
+#if  defined(POWERPC64)
+	  avalue[i] = (char *) pgr + 4;
+#else
+	case FFI_TYPE_POINTER:
+	  avalue[i] = pgr;
+#endif
+	  pgr++;
+	  break;
+
+	case FFI_TYPE_STRUCT:
+	  size_al = arg_types[i]->size;
+#if defined(POWERPC_DARWIN64)
+	  pgr = (unsigned long *)FFI_ALIGN((char *)pgr, arg_types[i]->alignment);
+	  if (size_al < 3 || size_al == 4)
+	    {
+	      avalue[i] = ((char *)pgr)+8-size_al;
+	      if (arg_types[i]->elements[0]->type == FFI_TYPE_FLOAT
+		  && fpsused < NUM_FPR_ARG_REGISTERS)
+		{
+		  *(float *)pgr = (float) *(double *)pfr;
+		  pfr++;
+		  fpsused++;
+		}
+	    }
+	  else 
+	    {
+	      if (size_al != 16)
+		pfr = (ffi_dblfl *) 
+		    darwin64_struct_floats_to_mem (arg_types[i], (char *)pgr,
+						   (double *)pfr, &fpsused);
+	      avalue[i] = pgr;
+	    }
+	  pgr += (size_al + 7) / 8;
+#else
+	  /* If the first member of the struct is a double, then align
+	     the struct to double-word.  */
+	  if (arg_types[i]->elements[0]->type == FFI_TYPE_DOUBLE)
+	    size_al = FFI_ALIGN(arg_types[i]->size, 8);
+#  if defined(POWERPC64)
+	  FFI_ASSERT (cif->abi != FFI_DARWIN);
+	  avalue[i] = pgr;
+	  pgr += (size_al + 7) / 8;
+#  else
+	  /* Structures that match the basic modes (QI 1 byte, HI 2 bytes,
+	     SI 4 bytes) are aligned as if they were those modes.  */
+	  if (size_al < 3 && cif->abi == FFI_DARWIN)
+	    avalue[i] = (char*) pgr + 4 - size_al;
+	  else
+	    avalue[i] = pgr;
+	  pgr += (size_al + 3) / 4;
+#  endif
+#endif
+	  break;
+
+	case FFI_TYPE_SINT64:
+	case FFI_TYPE_UINT64:
+#if  defined(POWERPC64)
+	case FFI_TYPE_POINTER:
+	  avalue[i] = pgr;
+	  pgr++;
+	  break;
+#else
+	  /* Long long ints are passed in two gpr's.  */
+	  avalue[i] = pgr;
+	  pgr += 2;
+	  break;
+#endif
+
+	case FFI_TYPE_FLOAT:
+	  /* A float value consumes a GPR.
+	     There are 13 64bit floating point registers.  */
+	  if (pfr < end_pfr)
+	    {
+	      double temp = pfr->d;
+	      pfr->f = (float) temp;
+	      avalue[i] = pfr;
+	      pfr++;
+	    }
+	  else
+	    {
+	      avalue[i] = pgr;
+	    }
+	  pgr++;
+	  break;
+
+	case FFI_TYPE_DOUBLE:
+	  /* A double value consumes two GPRs.
+	     There are 13 64bit floating point registers.  */
+	  if (pfr < end_pfr)
+	    {
+	      avalue[i] = pfr;
+	      pfr++;
+	    }
+	  else
+	    {
+	      avalue[i] = pgr;
+	    }
+#ifdef POWERPC64
+	  pgr++;
+#else
+	  pgr += 2;
+#endif
+	  break;
+
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+
+	case FFI_TYPE_LONGDOUBLE:
+#ifdef POWERPC64
+	  if (pfr + 1 < end_pfr)
+	    {
+	      avalue[i] = pfr;
+	      pfr += 2;
+	    }
+	  else
+	    {
+	      if (pfr < end_pfr)
+		{
+		  *pgr = *(unsigned long *) pfr;
+		  pfr++;
+		}
+	      avalue[i] = pgr;
+	    }
+	  pgr += 2;
+#else  /* POWERPC64 */
+	  /* A long double value consumes four GPRs and two FPRs.
+	     There are 13 64bit floating point registers.  */
+	  if (pfr + 1 < end_pfr)
+	    {
+	      avalue[i] = pfr;
+	      pfr += 2;
+	    }
+	  /* Here we have the situation where one part of the long double
+	     is stored in fpr13 and the other part is already on the stack.
+	     We use a union to pass the long double to avalue[i].  */
+	  else if (pfr + 1 == end_pfr)
+	    {
+	      union ldu temp_ld;
+	      memcpy (&temp_ld.lb[0], pfr, sizeof(ldbits));
+	      memcpy (&temp_ld.lb[1], pgr + 2, sizeof(ldbits));
+	      avalue[i] = &temp_ld.ld;
+	      pfr++;
+	    }
+	  else
+	    {
+	      avalue[i] = pgr;
+	    }
+	  pgr += 4;
+#endif  /* POWERPC64 */
+	  break;
+#endif
+	default:
+	  FFI_ASSERT(0);
+	}
+      i++;
+    }
+
+  (fun) (cif, rvalue, avalue, user_data);
+
+  /* Tell ffi_closure_ASM to perform return type promotions.  */
+  return cif->rtype;
+}
+
+ffi_type *
+ffi_closure_helper_DARWIN (ffi_closure *closure, void *rvalue,
+			   unsigned long *pgr, ffi_dblfl *pfr)
+{
+  return ffi_closure_helper_common (closure->cif, closure->fun,
+				    closure->user_data, rvalue, pgr, pfr);
+}
+
+ffi_type *
+ffi_go_closure_helper_DARWIN (ffi_go_closure *closure, void *rvalue,
+			      unsigned long *pgr, ffi_dblfl *pfr)
+{
+  return ffi_closure_helper_common (closure->cif, closure->fun,
+				    closure, rvalue, pgr, pfr);
+}
+