re PR libffi/35484 (libffi doesn't support AIX 64bit)

2009-11-30  David Edelsohn  <edelsohn@gnu.org>

        PR target/35484
        * src/powerpc/ffitarget.h (POWERPC64): Define for PPC64 Linux and
        AIX64.
        * src/powerpc/aix.S: Implement AIX64 version.
        * src/powerpc/aix_closure.S: Implement AIX64 version.
        (ffi_closure_ASM): Use extsb, lha and displament addresses.
        * src/powerpc/ffi_darwin.c (ffi_prep_args): Implement AIX64
        support.
        (ffi_prep_cif_machdep): Same.
        (ffi_call): Same.
        (ffi_closure_helper_DARWIN): Same.

From-SVN: r154855

5 files changed, 468 insertions, 77 deletions

diff --git a/libffi/ChangeLog b/libffi/ChangeLog
index 651a01e0b6e..c93d0b73574 100644
--- a/libffi/ChangeLog
+++ b/libffi/ChangeLog
@@ -1,3 +1,17 @@
+2009-11-30  David Edelsohn  <edelsohn@gnu.org>
+
+	PR target/35484
+	* src/powerpc/ffitarget.h (POWERPC64): Define for PPC64 Linux and
+	AIX64.
+	* src/powerpc/aix.S: Implement AIX64 version.
+	* src/powerpc/aix_closure.S: Implement AIX64 version.
+	(ffi_closure_ASM): Use extsb, lha and displament addresses.
+	* src/powerpc/ffi_darwin.c (ffi_prep_args): Implement AIX64
+	support.
+	(ffi_prep_cif_machdep): Same.
+	(ffi_call): Same.
+	(ffi_closure_helper_DARWIN): Same.
+
 2009-11-02  Andreas Tobler  <a.tobler@schweiz.org>
 
 	PR libffi/41908
diff --git a/libffi/src/powerpc/aix.S b/libffi/src/powerpc/aix.S
index 45502f7963e..7b73999f2a7 100644
--- a/libffi/src/powerpc/aix.S
+++ b/libffi/src/powerpc/aix.S
@@ -1,5 +1,5 @@
 /* -----------------------------------------------------------------------
-   aix.S - Copyright (c) 2002 Free Software Foundation, Inc.
+   aix.S - Copyright (c) 2002,2009 Free Software Foundation, Inc.
    based on darwin.S by John Hornkvist
 
    PowerPC Assembly glue.
@@ -86,9 +86,13 @@
 #define L(x) x
 	.file "aix.S"
 	.toc
-	.csect .text[PR]
-	.align 2
-.globl ffi_prep_args
+
+	/* void ffi_call_AIX(extended_cif *ecif, unsigned long bytes,
+	 *		     unsigned int flags, unsigned int *rvalue,
+	 *		     void (*fn)(),
+	 *		     void (*prep_args)(extended_cif*, unsigned *const));
+	 * r3=ecif, r4=bytes, r5=flags, r6=rvalue, r7=fn, r8=prep_args
+	 */
 
 .csect .text[PR]
 	.align 2
@@ -96,10 +100,124 @@
 	.globl .ffi_call_AIX
 .csect ffi_call_AIX[DS]
 ffi_call_AIX:
+#ifdef __64BIT__
+	.llong .ffi_call_AIX, TOC[tc0], 0
+	.csect .text[PR]
+.ffi_call_AIX:
+	mr	r12,r8 // We only need r12 until the call, so it doesn't have to be saved...
+	/* Save the old stack pointer as AP.  */
+	mr	r8,r1
+
+	/* Allocate the stack space we need.  */
+	stdux	r1,r1,r4
+
+	/* Save registers we use.  */
+	mflr	r9
+
+	std	r28,-32(r8)
+	std	r29,-24(r8)
+	std	r30,-16(r8)
+	std	r31, -8(r8)
+
+	std	r9, 16(r8)
+	std	r2, 40(r1)
+
+	/* Save arguments over call...  */
+	mr	r31,r5	/* flags, */
+	mr	r30,r6	/* rvalue, */
+	mr	r29,r7	/* function address, */
+	mr	r28,r8	/* our AP. */
+
+	/* Call ffi_prep_args.  */
+	mr	r4,r1
+	li	r9,0
+
+	ld	r2,8(r12)
+	ld	r12,0(r12)
+	mtctr	r12 // r12 holds address of _ffi_prep_args
+	bctrl
+	ld	r2,40(r1)
+
+	/* Now do the call.  */
+	ld	r12,0(r29)
+	/* Set up cr1 with bits 4-7 of the flags.  */
+	mtcrf	0x40,r31
+	std	r2,40(r1)
+	mtctr	r12
+	ld	r2,8(r29)
+	/* Load all those argument registers.  */
+	// We have set up a nice stack frame, just load it into registers.
+	ld	r3, 40+(1*8)(r1)
+	ld	r4, 40+(2*8)(r1)
+	ld	r5, 40+(3*8)(r1)
+	ld	r6, 40+(4*8)(r1)
+	nop
+	ld	r7, 40+(5*8)(r1)
+	ld	r8, 40+(6*8)(r1)
+	ld	r9, 40+(7*8)(r1)
+	ld	r10,40+(8*8)(r1)
+
+L1:
+	/* Load all the FP registers.  */
+	bf	6,L2 // 2f + 0x18
+	lfd	f1,-32-(13*8)(r28)
+	lfd	f2,-32-(12*8)(r28)
+	lfd	f3,-32-(11*8)(r28)
+	lfd	f4,-32-(10*8)(r28)
+	nop
+	lfd	f5,-32-(9*8)(r28)
+	lfd	f6,-32-(8*8)(r28)
+	lfd	f7,-32-(7*8)(r28)
+	lfd	f8,-32-(6*8)(r28)
+	nop
+	lfd	f9,-32-(5*8)(r28)
+	lfd	f10,-32-(4*8)(r28)
+	lfd	f11,-32-(3*8)(r28)
+	lfd	f12,-32-(2*8)(r28)
+	nop
+	lfd	f13,-32-(1*8)(r28)
+
+L2:
+	/* Make the call.  */
+	bctrl
+	ld	r2,40(r1)
+
+	/* Now, deal with the return value.  */
+	mtcrf	0x01,r31
+
+	bt	30,L(done_return_value)
+	bt	29,L(fp_return_value)
+	std	r3,0(r30)
+	bf	28,L(done_return_value)
+	std	r4,4(r30)
+
+	/* Fall through...  */
+
+L(done_return_value):
+	/* Restore the registers we used and return.  */
+	ld	r9,16(r28)
+	ld	r31,-8(r28)
+	mtlr	r9
+	ld	r30,-16(r28)
+	ld	r29,-24(r28)
+	ld	r28,-32(r28)
+	ld	r1,0(r1)
+	blr
+
+L(fp_return_value):
+	bf	28,L(float_return_value)
+	stfd	f1,0(r30)
+	b	L(done_return_value)
+L(float_return_value):
+	stfs	f1,0(r30)
+	b	L(done_return_value)
+
+#else /* ! __64BIT__ */
+	
 	.long .ffi_call_AIX, TOC[tc0], 0
 	.csect .text[PR]
 .ffi_call_AIX:
-	mr      r12,r8 // We only need r12 until the call, so it doesn't have to be saved...
+	mr	r12,r8 // We only need r12 until the call, so it doesn't have to be saved...
 	/* Save the old stack pointer as AP.  */
 	mr	r8,r1
 
@@ -142,15 +260,15 @@ ffi_call_AIX:
 	lwz	r2,4(r29)
 	/* Load all those argument registers.  */
 	// We have set up a nice stack frame, just load it into registers.
-	lwz     r3, 20+(1*4)(r1)
-	lwz     r4, 20+(2*4)(r1)
-	lwz     r5, 20+(3*4)(r1)
-	lwz     r6, 20+(4*4)(r1)
+	lwz	r3, 20+(1*4)(r1)
+	lwz	r4, 20+(2*4)(r1)
+	lwz	r5, 20+(3*4)(r1)
+	lwz	r6, 20+(4*4)(r1)
 	nop
-	lwz     r7, 20+(5*4)(r1)
-	lwz     r8, 20+(6*4)(r1)
-	lwz     r9, 20+(7*4)(r1)
-	lwz     r10,20+(8*4)(r1)
+	lwz	r7, 20+(5*4)(r1)
+	lwz	r8, 20+(6*4)(r1)
+	lwz	r9, 20+(7*4)(r1)
+	lwz	r10,20+(8*4)(r1)
 
 L1:
 	/* Load all the FP registers.  */
@@ -165,17 +283,17 @@ L1:
 	lfd	f7,-16-(7*8)(r28)
 	lfd	f8,-16-(6*8)(r28)
 	nop
-	lfd     f9,-16-(5*8)(r28)
-	lfd     f10,-16-(4*8)(r28)
-	lfd     f11,-16-(3*8)(r28)
-	lfd     f12,-16-(2*8)(r28)
+	lfd	f9,-16-(5*8)(r28)
+	lfd	f10,-16-(4*8)(r28)
+	lfd	f11,-16-(3*8)(r28)
+	lfd	f12,-16-(2*8)(r28)
 	nop
-	lfd     f13,-16-(1*8)(r28)
+	lfd	f13,-16-(1*8)(r28)
 
 L2:
 	/* Make the call.  */
 	bctrl
-	lwz r2,20(r1)
+	lwz	r2,20(r1)
 
 	/* Now, deal with the return value.  */
 	mtcrf	0x01,r31
@@ -190,8 +308,8 @@ L2:
 
 L(done_return_value):
 	/* Restore the registers we used and return.  */
-	lwz	r9,   8(r28)
-	lwz	r31,  -4(r28)
+	lwz	r9,8(r28)
+	lwz	r31,-4(r28)
 	mtlr	r9
 	lwz	r30, -8(r28)
 	lwz	r29,-12(r28)
@@ -206,6 +324,7 @@ L(fp_return_value):
 L(float_return_value):
 	stfs	f1,0(r30)
 	b	L(done_return_value)
+#endif
 	.long 0
 	.byte 0,0,0,1,128,4,0,0
 //END(ffi_call_AIX)
@@ -216,7 +335,11 @@ L(float_return_value):
 	.globl .ffi_call_DARWIN
 .csect ffi_call_DARWIN[DS]
 ffi_call_DARWIN:
+#ifdef __64BIT__
+	.llong .ffi_call_DARWIN, TOC[tc0], 0
+#else
 	.long .ffi_call_DARWIN, TOC[tc0], 0
+#endif
 	.csect .text[PR]
 .ffi_call_DARWIN:
 	blr
diff --git a/libffi/src/powerpc/aix_closure.S b/libffi/src/powerpc/aix_closure.S
index 7bf5c656063..e1a60b46832 100644
--- a/libffi/src/powerpc/aix_closure.S
+++ b/libffi/src/powerpc/aix_closure.S
@@ -1,5 +1,5 @@
 /* -----------------------------------------------------------------------
-   aix_closure.S - Copyright (c) 2002 2003 Free Software Foundation, Inc.
+   aix_closure.S - Copyright (c) 2002, 2003, 2009 Free Software Foundation, Inc.
    based on darwin_closure.S
 
    PowerPC Assembly glue.
@@ -94,8 +94,162 @@ LC..60:
 	.globl ffi_closure_ASM
 	.globl .ffi_closure_ASM
 .csect ffi_closure_ASM[DS]
-
 ffi_closure_ASM:
+#ifdef __64BIT__
+	.llong .ffi_closure_ASM, TOC[tc0], 0
+	.csect .text[PR]
+.ffi_closure_ASM:
+
+	mflr r0			/* extract return address */
+	std r0,16(r1)		/* save the return address */
+
+	/* 48  Bytes (Linkage Area) */
+	/* 64  Bytes (params) */
+	/* 104 Bytes (13*8 from FPR) */
+	/* 32  Bytes (result) */
+	/* 248 Bytes */
+
+	stdu r1,-248(r1)	/* skip over caller save area
+				   keep stack aligned to 16  */
+
+/* we want to build up an area for the parameters passed */
+/* in registers (both floating point and integer) */
+
+	/* we store gpr 3 to gpr 10 (aligned to 4)
+	in the parents outgoing area  */
+	std   r3, (304+0*8)(r1)
+	std   r4, (304+1*8)(r1)
+	std   r5, (304+2*8)(r1)
+	std   r6, (304+3*8)(r1)
+	std   r7, (304+4*8)(r1)
+	std   r8, (304+5*8)(r1)
+	std   r9, (304+6*8)(r1)
+	std   r10, (304+7*8)(r1)
+
+	/* next save fpr 1 to fpr 13 (aligned to 8) */
+	stfd  f1, (112+0*8)(r1)
+	stfd  f2, (112+1*8)(r1)
+	stfd  f3, (112+2*8)(r1)
+	stfd  f4, (112+3*8)(r1)
+	stfd  f5, (112+4*8)(r1)
+	stfd  f6, (112+5*8)(r1)
+	stfd  f7, (112+6*8)(r1)
+	stfd  f8, (112+7*8)(r1)
+	stfd  f9, (112+8*8)(r1)
+	stfd  f10, (112+9*8)(r1)
+	stfd  f11, (112+10*8)(r1)
+	stfd  f12, (112+11*8)(r1)
+	stfd  f13, (112+12*8)(r1)
+
+	/* set up registers for the routine that actually does the work */
+	/* get the context pointer from the trampoline */
+	mr r3,r11
+
+	/* now load up the pointer to the result storage */
+	addi r4,r1,216
+
+	/* now load up the pointer to the saved gpr registers */
+	addi r5,r1,304
+
+	/* now load up the pointer to the saved fpr registers */
+	addi r6,r1,112
+
+	/* make the call */
+	bl .ffi_closure_helper_DARWIN
+	nop
+
+	/* now r3 contains the return type */
+	/* so use it to look up in a table */
+	/* so we know how to deal with each type */
+
+	/* look up the proper starting point in table  */
+	/* by using return type as offset */
+	addi r5,r1,216		/* get pointer to results area */
+	ld r4,LC..60(2)		/* get address of jump table */
+	sldi r3,r3,2		/* now multiply return type by 4 */
+	lwzx r3,r4,r3		/* get the contents of that table value */
+	add r3,r3,r4		/* add contents of table to table address */
+	mtctr r3
+	bctr			/* jump to it */
+
+L..60:
+	.long L..44-L..60    /* FFI_TYPE_VOID */
+	.long L..51-L..60    /* FFI_TYPE_INT */
+	.long L..47-L..60    /* FFI_TYPE_FLOAT */
+	.long L..46-L..60    /* FFI_TYPE_DOUBLE */
+	.long L..45-L..60    /* FFI_TYPE_LONGDOUBLE */
+	.long L..56-L..60    /* FFI_TYPE_UINT8 */
+	.long L..55-L..60    /* FFI_TYPE_SINT8 */
+	.long L..58-L..60    /* FFI_TYPE_UINT16 */
+	.long L..57-L..60    /* FFI_TYPE_SINT16 */
+	.long L..50-L..60    /* FFI_TYPE_UINT32 */
+	.long L..51-L..60    /* FFI_TYPE_SINT32 */
+	.long L..48-L..60    /* FFI_TYPE_UINT64 */
+	.long L..48-L..60    /* FFI_TYPE_SINT64 */
+	.long L..44-L..60    /* FFI_TYPE_STRUCT */
+	.long L..48-L..60    /* FFI_TYPE_POINTER */
+
+
+/* case long double */
+L..45:
+	lfd f1,0(r5)
+	lfd f2,8(r5)
+	b L..44
+
+/* case double */
+L..46:
+	lfd f1,0(r5)
+	b L..44
+
+/* case float */
+L..47:
+	lfs f1,0(r5)
+	b L..44
+
+/* case long long / pointer */
+L..48:
+	ld r3,0(r5)
+	b L..44
+
+/* case uint32 */
+L..50:
+	lwz r3,4(r5)
+	b L..44
+
+/* case int / sint32 */
+L..51:
+	lwa r3,4(r5)
+	b L..44
+
+/* case signed int8 */
+L..55:
+	lbz r3,7(r5)
+	extsb r3,r3
+	b L..44
+
+/* case unsigned int8 */
+L..56:
+	lbz r3,7(r5)
+	b L..44
+
+/* case signed int16 */
+L..57:
+	lha r3,6(r5)
+	b L..44
+
+/* case unsigned int16 */
+L..58:
+	lhz r3,6(r5)
+
+/* case void / done	 */
+L..44:
+	addi r1,r1,248		/* restore stack pointer */
+	ld r0,16(r1)		/* get return address */
+	mtlr r0			/* reset link register */
+	blr
+
+#else /* ! __64BIT__ */
+	
 	.long .ffi_closure_ASM, TOC[tc0], 0
 	.csect .text[PR]
 .ffi_closure_ASM:
@@ -106,8 +260,8 @@ ffi_closure_ASM:
 	/* 24 Bytes (Linkage Area) */
 	/* 32 Bytes (params) */
 	/* 104 Bytes (13*8 from FPR) */
-	/* 8 Bytes (result) */
-	/* 168 Bytes */
+	/* 16  Bytes (result) */
+	/* 176 Bytes */
 
 	stwu r1,-176(r1)	/* skip over caller save area
 				keep stack aligned to 16  */
@@ -177,7 +331,7 @@ L..60:
 	.long L..50-L..60    /* FFI_TYPE_INT */
 	.long L..47-L..60    /* FFI_TYPE_FLOAT */
 	.long L..46-L..60    /* FFI_TYPE_DOUBLE */
-	.long L..46-L..60    /* FFI_TYPE_LONGDOUBLE */
+	.long L..45-L..60    /* FFI_TYPE_LONGDOUBLE */
 	.long L..56-L..60    /* FFI_TYPE_UINT8 */
 	.long L..55-L..60    /* FFI_TYPE_SINT8 */
 	.long L..58-L..60    /* FFI_TYPE_UINT16 */
@@ -190,6 +344,12 @@ L..60:
 	.long L..50-L..60    /* FFI_TYPE_POINTER */
 
 
+/* case long double */
+L..45:
+	lfd f1,0(r5)
+	lfd f2,8(r5)
+	b L..44
+
 /* case double */
 L..46:
 	lfd f1,0(r5)
@@ -211,31 +371,25 @@ L..50:
 	lwz r3,0(r5)
 	b L..44
 
-/* case signed int8	 */
+/* case signed int8 */
 L..55:
-	addi r5,r5,3
-	lbz r3,0(r5)
-	slwi r3,r3,24
-	srawi r3,r3,24
+	lbz r3,3(r5)
+	extsb r3,r3
 	b L..44
 
-/* case unsigned int8	 */
+/* case unsigned int8 */
 L..56:
-	addi r5,r5,3
-	lbz r3,0(r5)
+	lbz r3,3(r5)
 	b L..44
 
 /* case signed int16 */
 L..57:
-	addi r5,r5,2
-	lhz r3,0(r5)
-	extsh r3,r3
+	lha r3,2(r5)
 	b L..44
 
 /* case unsigned int16 */
 L..58:
-	addi r5,r5,2
-	lhz r3,0(r5)
+	lhz r3,2(r5)
 
 /* case void / done	 */
 L..44:
@@ -243,5 +397,5 @@ L..44:
 	lwz r0,8(r1)		/* get return address */
 	mtlr r0			/* reset link register */
 	blr
-
+#endif
 /* END(ffi_closure_ASM) */
diff --git a/libffi/src/powerpc/ffi_darwin.c b/libffi/src/powerpc/ffi_darwin.c
index 501035d75d0..fd2a3710434 100644
--- a/libffi/src/powerpc/ffi_darwin.c
+++ b/libffi/src/powerpc/ffi_darwin.c
@@ -3,7 +3,7 @@
 
    Copyright (C) 1998 Geoffrey Keating
    Copyright (C) 2001 John Hornkvist
-   Copyright (C) 2002, 2006, 2007 Free Software Foundation, Inc.
+   Copyright (C) 2002, 2006, 2007, 2009 Free Software Foundation, Inc.
 
    FFI support for Darwin and AIX.
    
@@ -80,27 +80,28 @@ enum { ASM_NEEDS_REGISTERS = 4 };
 
    */
 
-void ffi_prep_args(extended_cif *ecif, unsigned *const stack)
+void ffi_prep_args(extended_cif *ecif, unsigned long *const stack)
 {
   const unsigned bytes = ecif->cif->bytes;
   const unsigned flags = ecif->cif->flags;
 
   /* 'stacktop' points at the previous backchain pointer.  */
-  unsigned *const stacktop = stack + (bytes / sizeof(unsigned));
+  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;
+  double *fpr_base = (double *) ((stacktop - ASM_NEEDS_REGISTERS)
+				 - NUM_FPR_ARG_REGISTERS);
   int fparg_count = 0;
 
 
   /* 'next_arg' grows up as we put parameters in it.  */
-  unsigned *next_arg = stack + 6; /* 6 reserved positions.  */
+  unsigned long *next_arg = (unsigned long *) stack + 6; /* 6 reserved positions.  */
 
-  int i = ecif->cif->nargs;
+  int i;
   double double_tmp;
   void **p_argv = ecif->avalue;
-  unsigned gprvalue;
+  unsigned long gprvalue;
   ffi_type** ptr = ecif->cif->arg_types;
   char *dest_cpy;
   unsigned size_al = 0;
@@ -115,12 +116,10 @@ void ffi_prep_args(extended_cif *ecif, unsigned *const stack)
      Return values are referenced by r3, so r4 is the first parameter.  */
 
   if (flags & FLAG_RETVAL_REFERENCE)
-    *next_arg++ = (unsigned)(char *)ecif->rvalue;
+    *next_arg++ = (unsigned long)(char *)ecif->rvalue;
 
   /* Now for the arguments.  */
-  for (;
-       i > 0;
-       i--, ptr++, p_argv++)
+  for (i = ecif->cif->nargs; i > 0; i--, ptr++, p_argv++)
     {
       switch ((*ptr)->type)
 	{
@@ -133,7 +132,11 @@ void ffi_prep_args(extended_cif *ecif, unsigned *const stack)
 	    *(double *)next_arg = double_tmp;
 	  else
 	    *fpr_base++ = double_tmp;
+#ifdef POWERPC64
 	  next_arg++;
+#else
+	  next_arg += 2;
+#endif
 	  fparg_count++;
 	  FFI_ASSERT(flags & FLAG_FP_ARGUMENTS);
 	  break;
@@ -152,42 +155,69 @@ void ffi_prep_args(extended_cif *ecif, unsigned *const stack)
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 
 	case FFI_TYPE_LONGDOUBLE:
-	  double_tmp = ((double *)*p_argv)[0];
-	  if (fparg_count >= NUM_FPR_ARG_REGISTERS)
-	    *(double *)next_arg = double_tmp;
+#ifdef POWERPC64
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *((long double *) fpr_base)++ = *(long double *) *p_argv;
 	  else
-	    *fpr_base++ = double_tmp;
+	    *(long double *) next_arg = *(long double *) *p_argv;
 	  next_arg += 2;
-	  fparg_count++;
-	  double_tmp = ((double *)*p_argv)[1];
-	  if (fparg_count >= NUM_FPR_ARG_REGISTERS)
-	    *(double *)next_arg = double_tmp;
+	  fparg_count += 2;
+#else
+	  double_tmp = *((double *) *p_argv);
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
+	    *fpr_base++ = double_tmp;
 	  else
+	    *(double *) next_arg = double_tmp;
+
+	  double_tmp = ((double *) *p_argv)[1];
+	  if (fparg_count < NUM_FPR_ARG_REGISTERS)
 	    *fpr_base++ = double_tmp;
+	  else
+	    *(double *) next_arg = double_tmp;
 	  next_arg += 2;
 	  fparg_count++;
+#endif
 	  FFI_ASSERT(flags & FLAG_FP_ARGUMENTS);
 	  break;
 #endif
 	case FFI_TYPE_UINT64:
 	case FFI_TYPE_SINT64:
-	  *(long long *)next_arg = *(long long *)*p_argv;
+#ifdef POWERPC64
+	  gprvalue = *(long long *) p_argv;
+	  goto putgpr;
+#else
+	  *(long long *) next_arg = *(long long *) *p_argv;
 	  next_arg+=2;
+#endif
 	  break;
+	case FFI_TYPE_POINTER:
+	  gprvalue = *(unsigned long *) *p_argv;
+	  goto putgpr;
 	case FFI_TYPE_UINT8:
-	  gprvalue = *(unsigned char *)*p_argv;
+	  gprvalue = *(unsigned char *) *p_argv;
 	  goto putgpr;
 	case FFI_TYPE_SINT8:
-	  gprvalue = *(signed char *)*p_argv;
+	  gprvalue = *(signed char *) *p_argv;
 	  goto putgpr;
 	case FFI_TYPE_UINT16:
-	  gprvalue = *(unsigned short *)*p_argv;
+	  gprvalue = *(unsigned short *) *p_argv;
 	  goto putgpr;
 	case FFI_TYPE_SINT16:
-	  gprvalue = *(signed short *)*p_argv;
+	  gprvalue = *(signed short *) *p_argv;
 	  goto putgpr;
 
 	case FFI_TYPE_STRUCT:
+#ifdef POWERPC64
+	  dest_cpy = (char *) next_arg;
+	  size_al = (*ptr)->size;
+	  if ((*ptr)->elements[0]->type == 3)
+	    size_al = ALIGN((*ptr)->size, 8);
+	  if (size_al < 3 && ecif->cif->abi == FFI_DARWIN)
+	    dest_cpy += 4 - size_al;
+
+	  memcpy ((char *) dest_cpy, (char *) *p_argv, size_al);
+	  next_arg += (size_al + 7) / 8;
+#else
 	  dest_cpy = (char *) next_arg;
 
 	  /* Structures that match the basic modes (QI 1 byte, HI 2 bytes,
@@ -204,12 +234,12 @@ void ffi_prep_args(extended_cif *ecif, unsigned *const stack)
 
 	  memcpy((char *)dest_cpy, (char *)*p_argv, size_al);
 	  next_arg += (size_al + 3) / 4;
+#endif
 	  break;
 
 	case FFI_TYPE_INT:
 	case FFI_TYPE_UINT32:
 	case FFI_TYPE_SINT32:
-	case FFI_TYPE_POINTER:
 	  gprvalue = *(unsigned *)*p_argv;
 	putgpr:
 	  *next_arg++ = gprvalue;
@@ -324,6 +354,9 @@ ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
 
     case FFI_TYPE_UINT64:
     case FFI_TYPE_SINT64:
+#ifdef POWERPC64
+    case FFI_TYPE_POINTER:
+#endif
       flags |= FLAG_RETURNS_64BITS;
       break;
 
@@ -387,11 +420,14 @@ ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
 	case FFI_TYPE_STRUCT:
 	  size_al = (*ptr)->size;
 	  /* If the first member of the struct is a double, then align
-	     the struct to double-word.
-	     Type 3 is defined in include/ffi.h. #define FFI_TYPE_DOUBLE 3.  */
-	  if ((*ptr)->elements[0]->type == 3)
+	     the struct to double-word.  */
+	  if ((*ptr)->elements[0]->type == FFI_TYPE_DOUBLE)
 	    size_al = ALIGN((*ptr)->size, 8);
+#ifdef POWERPC64
+	  intarg_count += (size_al + 7) / 8;
+#else
 	  intarg_count += (size_al + 3) / 4;
+#endif
 	  break;
 
 	default:
@@ -410,13 +446,22 @@ ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
     bytes += NUM_FPR_ARG_REGISTERS * sizeof(double);
 
   /* 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.  */
+#ifdef POWERPC64
+  bytes = (bytes + 31) & -0x1F;
+#else
   bytes = (bytes + 15) & ~0xF;
+#endif
 
   cif->flags = flags;
   cif->bytes = bytes;
@@ -424,9 +469,9 @@ ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
   return FFI_OK;
 }
 
-extern void ffi_call_AIX(extended_cif *, unsigned, unsigned, unsigned *,
+extern void ffi_call_AIX(extended_cif *, long, unsigned, unsigned *,
 			 void (*fn)(void), void (*fn2)(void));
-extern void ffi_call_DARWIN(extended_cif *, unsigned, unsigned, unsigned *,
+extern void ffi_call_DARWIN(extended_cif *, long, unsigned, unsigned *,
 			    void (*fn)(void), void (*fn2)(void));
 
 void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
@@ -450,11 +495,11 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
   switch (cif->abi)
     {
     case FFI_AIX:
-      ffi_call_AIX(&ecif, -cif->bytes, cif->flags, ecif.rvalue, fn,
+      ffi_call_AIX(&ecif, -(long)cif->bytes, cif->flags, ecif.rvalue, fn,
 		   ffi_prep_args);
       break;
     case FFI_DARWIN:
-      ffi_call_DARWIN(&ecif, -cif->bytes, cif->flags, ecif.rvalue, fn,
+      ffi_call_DARWIN(&ecif, -(long)cif->bytes, cif->flags, ecif.rvalue, fn,
 		      ffi_prep_args);
       break;
     default:
@@ -650,7 +695,6 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
   ffi_cif *        cif;
   double           temp;
   unsigned         size_al;
-  union ldu        temp_ld;
 
   cif = closure->cif;
   avalue = alloca(cif->nargs * sizeof(void *));
@@ -678,34 +722,56 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
 	{
 	case FFI_TYPE_SINT8:
 	case FFI_TYPE_UINT8:
+#ifdef POWERPC64
+	  avalue[i] = (char *) pgr + 7;
+#else
 	  avalue[i] = (char *) pgr + 3;
+#endif
 	  ng++;
 	  pgr++;
 	  break;
 
 	case FFI_TYPE_SINT16:
 	case FFI_TYPE_UINT16:
+#ifdef POWERPC64
+	  avalue[i] = (char *) pgr + 6;
+#else
 	  avalue[i] = (char *) pgr + 2;
+#endif
 	  ng++;
 	  pgr++;
 	  break;
 
 	case FFI_TYPE_SINT32:
 	case FFI_TYPE_UINT32:
+#ifdef POWERPC64
+	  avalue[i] = (char *) pgr + 4;
+#else
 	case FFI_TYPE_POINTER:
 	  avalue[i] = pgr;
+#endif
 	  ng++;
 	  pgr++;
 	  break;
 
 	case FFI_TYPE_STRUCT:
+#ifdef POWERPC64
+	  size_al = arg_types[i]->size;
+	  if (arg_types[i]->elements[0]->type == FFI_TYPE_DOUBLE)
+	    size_al = ALIGN (arg_types[i]->size, 8);
+	  if (size_al < 3 && cif->abi == FFI_DARWIN)
+	    avalue[i] = (void *) pgr + 8 - size_al;
+	  else
+	    avalue[i] = (void *) pgr;
+	  ng += (size_al + 7) / 8;
+	  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.  */
 	  size_al = arg_types[i]->size;
 	  /* If the first member of the struct is a double, then align
-	     the struct to double-word.
-	     Type 3 is defined in include/ffi.h. #define FFI_TYPE_DOUBLE 3.  */
-	  if (arg_types[i]->elements[0]->type == 3)
+	     the struct to double-word.  */
+	  if (arg_types[i]->elements[0]->type == FFI_TYPE_DOUBLE)
 	    size_al = ALIGN(arg_types[i]->size, 8);
 	  if (size_al < 3 && cif->abi == FFI_DARWIN)
 	    avalue[i] = (void*) pgr + 4 - size_al;
@@ -713,15 +779,24 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
 	    avalue[i] = (void*) pgr;
 	  ng += (size_al + 3) / 4;
 	  pgr += (size_al + 3) / 4;
+#endif
 	  break;
 
 	case FFI_TYPE_SINT64:
 	case FFI_TYPE_UINT64:
+#ifdef POWERPC64
+	case FFI_TYPE_POINTER:
+	  avalue[i] = pgr;
+	  ng++;
+	  pgr++;
+	  break;
+#else
 	  /* Long long ints are passed in two gpr's.  */
 	  avalue[i] = pgr;
 	  ng += 2;
 	  pgr += 2;
 	  break;
+#endif
 
 	case FFI_TYPE_FLOAT:
 	  /* A float value consumes a GPR.
@@ -755,13 +830,32 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
 	      avalue[i] = pgr;
 	    }
 	  nf++;
+#ifdef POWERPC64
+	  ng++;
+	  pgr++;
+#else
 	  ng += 2;
 	  pgr += 2;
+#endif
 	  break;
 
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 
 	case FFI_TYPE_LONGDOUBLE:
+#ifdef POWERPC64
+	  if (nf < NUM_FPR_ARG_REGISTERS)
+	    {
+	      avalue[i] = pfr;
+	      pfr += 2;
+	    }
+	  else
+	    {
+	      avalue[i] = pgr;
+	    }
+	  nf += 2;
+	  ng += 2;
+	  pgr += 2;
+#else  /* POWERPC64 */
 	  /* A long double value consumes four GPRs and two FPRs.
 	     There are 13 64bit floating point registers.  */
 	  if (nf < NUM_FPR_ARG_REGISTERS - 1)
@@ -774,6 +868,7 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
 	     We use a union to pass the long double to avalue[i].  */
 	  else if (nf == NUM_FPR_ARG_REGISTERS - 1)
 	    {
+	      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;
@@ -785,6 +880,7 @@ int ffi_closure_helper_DARWIN (ffi_closure* closure, void * rvalue,
 	  nf += 2;
 	  ng += 4;
 	  pgr += 4;
+#endif  /* POWERPC64 */
 	  break;
 #endif
 	default:
diff --git a/libffi/src/powerpc/ffitarget.h b/libffi/src/powerpc/ffitarget.h
index 269f573987c..b4f992e6df4 100644
--- a/libffi/src/powerpc/ffitarget.h
+++ b/libffi/src/powerpc/ffitarget.h
@@ -30,7 +30,11 @@
 
 /* ---- System specific configurations ----------------------------------- */
 
-#if defined (POWERPC) && defined (__powerpc64__)
+#if defined (POWERPC) && defined (__powerpc64__)	/* linux64 */
+#define POWERPC64
+#elif defined (POWERPC_DARWIN) && defined (__ppc64__)	/* Darwin */
+#define POWERPC64
+#elif defined (POWERPC_AIX) && defined (__64BIT__)	/* AIX64 */
 #define POWERPC64
 #endif