ffi 1.10.0 → 1.15.5

data/ext/ffi_c/libffi/src/kvx/ffitarget.h

@@ -0,0 +1,75 @@
+/* -----------------------------------------------------------------------
+   ffitarget.h - Copyright (c) 2020 Kalray
+
+   KVX Target configuration macros
+
+   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 AUTHORS OR COPYRIGHT
+   HOLDERS 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.
+   ----------------------------------------------------------------------- */
+
+#ifndef LIBFFI_TARGET_H
+#define LIBFFI_TARGET_H
+
+#ifndef LIBFFI_H
+#error "Please do not include ffitarget.h directly into your source. Use ffi.h instead."
+#endif
+
+/* ---- System specific configurations ----------------------------------- */
+
+#ifndef LIBFFI_ASM
+typedef unsigned long          ffi_arg;
+typedef signed long            ffi_sarg;
+
+typedef enum ffi_abi {
+  FFI_FIRST_ABI = 0,
+  FFI_SYSV,
+  FFI_LAST_ABI,
+  FFI_DEFAULT_ABI = FFI_SYSV
+} ffi_abi;
+
+/* Those values are set depending on return type
+ * they are used in the assembly code in sysv.S
+ */
+typedef enum kvx_intext_method {
+  KVX_RET_NONE = 0,
+  KVX_RET_SXBD = 1,
+  KVX_RET_SXHD = 2,
+  KVX_RET_SXWD = 3,
+  KVX_RET_ZXBD = 4,
+  KVX_RET_ZXHD = 5,
+  KVX_RET_ZXWD = 6
+} kvx_intext_method;
+
+#endif
+
+/* ---- Definitions for closures ----------------------------------------- */
+
+/* This is only to allow Python to compile
+ * but closures are not supported yet
+ */
+#define FFI_CLOSURES 1
+#define FFI_TRAMPOLINE_SIZE 0
+
+#define FFI_NATIVE_RAW_API 0
+#define FFI_TARGET_SPECIFIC_VARIADIC 1
+#define FFI_TARGET_HAS_COMPLEX_TYPE
+
+#endif
+

data/ext/ffi_c/libffi/src/kvx/sysv.S

@@ -0,0 +1,127 @@
+/* Copyright (c) 2020 Kalray
+
+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 AUTHORS OR COPYRIGHT HOLDERS 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.  */
+
+#if defined(__kvx__)
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+#include <ffi_cfi.h>
+#include <kvx/asm.h>
+
+.text
+.global ffi_call_SYSV
+.type ffi_call_SYSV, @function
+.type ffi_prep_args, @function
+.align 8
+
+/* ffi_call_SYSV
+
+  r0: total size to allocate on stack
+  r1: size of arg slots
+  r2: extended cif structure, DO NOT REMOVE: it is used by ffi_prep_args()
+  r3: return value address
+  r4: function to call
+  r5: integer sign extension method to be used
+*/
+ffi_call_SYSV:
+	addd $r12 = $r12, -64
+	so (-32)[$r12] = $r20r21r22r23
+	;;
+	sd (0)[$r12] = $r24
+	;;
+	get $r23 = $ra
+	copyd $r20 = $r12
+	sbfd $r12 = $r0, $r12
+	;;
+	copyd $r0 = $r12
+	copyd $r21 = $r3
+	copyd $r22 = $r4
+	copyd $r24 = $r5
+	call ffi_prep_args
+	;;
+	lo $r8r9r10r11 = (64)[$r12]
+	;;
+	lo $r4r5r6r7 = (32)[$r12]
+	;;
+	lo $r0r1r2r3 = (0)[$r12]
+	copyd $r12 = $r0
+	/* $r15 is the register used by the ABI to return big (>32 bytes)
+	 * structs by value.
+	 * It is also referred to as the "struct register" in the ABI.
+	 */
+	copyd $r15 = $r21
+	icall $r22
+	;;
+	pcrel $r4 = @pcrel(.Ltable)
+	cb.deqz $r24 ? .Lend
+	;;
+	addx8d $r24 = $r24, $r4
+	;;
+	igoto $r24
+	;;
+.Ltable:
+0: /* we should never arrive here */
+	goto .Lerror
+	nop
+	;;
+1: /* Sign extend byte to double */
+	sxbd $r0 = $r0
+	goto .Lend
+	;;
+2: /* Sign extend half to double */
+	sxhd $r0 = $r0
+	goto .Lend
+	;;
+3: /* Sign extend word to double */
+	sxwd $r0 = $r0
+	goto .Lend
+	;;
+4: /* Zero extend byte to double */
+	zxbd $r0 = $r0
+	goto .Lend
+	;;
+5: /* Zero extend half to double */
+	zxhd $r0 = $r0
+	goto .Lend
+	;;
+6: /* Zero extend word to double */
+	zxwd $r0 = $r0
+	/* Fallthrough to .Lend */
+	;;
+.Lend:
+	ld $r24 = (0)[$r12]
+	;;
+	set $ra = $r23
+	lo $r20r21r22r23 = (32)[$r20]
+	addd $r12 = $r20, 64
+	;;
+	ret
+	;;
+.Lerror:
+	errop
+	;;
+
+#endif /* __kvx__ */
+
+#if defined __ELF__ && defined __linux__
+	.section .note.GNU-stack,"",%progbits
+#endif
+

data/ext/ffi_c/libffi/src/metag/ffi.c

@@ -61,7 +61,7 @@ unsigned int ffi_prep_args(char *stack, extended_cif *ecif)
 		argp -= z;
 
 		/* Align if necessary */
-		argp = (char *) ALIGN_DOWN(ALIGN_DOWN(argp, (*p_arg)->alignment), 4);
+		argp = (char *) FFI_ALIGN_DOWN(FFI_ALIGN_DOWN(argp, (*p_arg)->alignment), 4);
 
 		if (z < sizeof(int)) {
 			z = sizeof(int);

data/ext/ffi_c/libffi/src/mips/ffi.c

@@ -39,7 +39,9 @@
 #endif
 
 #ifndef USE__BUILTIN___CLEAR_CACHE
-#  if defined(__OpenBSD__)
+#  if defined(__FreeBSD__)
+#    include <machine/sysarch.h>
+#  elif defined(__OpenBSD__)
 #    include <mips64/sysarch.h>
 #  else
 #    include <sys/cachectl.h>
@@ -777,11 +779,13 @@ ffi_prep_closure_loc (ffi_closure *closure,
   closure->fun = fun;
   closure->user_data = user_data;
 
+#if !defined(__FreeBSD__)
 #ifdef USE__BUILTIN___CLEAR_CACHE
   __builtin___clear_cache(clear_location, clear_location + FFI_TRAMPOLINE_SIZE);
 #else
   cacheflush (clear_location, FFI_TRAMPOLINE_SIZE, ICACHE);
 #endif
+#endif /* ! __FreeBSD__ */
   return FFI_OK;
 }
 

data/ext/ffi_c/libffi/src/mips/ffitarget.h

@@ -41,7 +41,7 @@
 #define _MIPS_SIM_ABI32		1
 #define _MIPS_SIM_NABI32	2
 #define _MIPS_SIM_ABI64		3
-#elif !defined(__OpenBSD__)
+#elif !defined(__OpenBSD__) && !defined(__FreeBSD__)
 # include <sgidefs.h>
 #endif
 

data/ext/ffi_c/libffi/src/mips/o32.S

@@ -282,9 +282,11 @@ $LCFI12:
 	li	$13, 1		# FFI_O32
 	bne	$16, $13, 1f	# Skip fp save if FFI_O32_SOFT_FLOAT
 	
+#ifndef __mips_soft_float
 	# Store all possible float/double registers.
 	s.d	$f12, FA_0_0_OFF2($fp)
 	s.d	$f14, FA_1_0_OFF2($fp)
+#endif
 1:
 	# prepare arguments for ffi_closure_mips_inner_O32
 	REG_L	a0, 4($15)	 # cif 

data/ext/ffi_c/libffi/src/moxie/ffi.c

@@ -100,7 +100,7 @@ void *ffi_prep_args(char *stack, extended_cif *ecif)
       count += z;
     }
 
-  return (stack + ((count > 24) ? 24 : ALIGN_DOWN(count, 8)));
+  return (stack + ((count > 24) ? 24 : FFI_ALIGN_DOWN(count, 8)));
 }
 
 /* Perform machine dependent cif processing */

data/ext/ffi_c/libffi/src/pa/ffi.c

@@ -421,12 +421,15 @@ ffi_status ffi_closure_inner_pa32(ffi_closure *closure, UINT32 *stack)
   ffi_cif *cif;
   void **avalue;
   void *rvalue;
-  UINT32 ret[2]; /* function can return up to 64-bits in registers */
+  /* Functions can return up to 64-bits in registers.  Return address
+     must be double word aligned.  */
+  union { double rd; UINT32 ret[2]; } u;
   ffi_type **p_arg;
   char *tmp;
   int i, avn;
   unsigned int slot = FIRST_ARG_SLOT;
   register UINT32 r28 asm("r28");
+  ffi_closure *c = (ffi_closure *)FFI_RESTORE_PTR (closure);
 
   cif = closure->cif;
 
@@ -434,7 +437,7 @@ ffi_status ffi_closure_inner_pa32(ffi_closure *closure, UINT32 *stack)
   if (cif->flags == FFI_TYPE_STRUCT)
     rvalue = (void *)r28;
   else
-    rvalue = &ret[0];
+    rvalue = &u;
 
   avalue = (void **)alloca(cif->nargs * FFI_SIZEOF_ARG);
   avn = cif->nargs;
@@ -529,35 +532,35 @@ ffi_status ffi_closure_inner_pa32(ffi_closure *closure, UINT32 *stack)
     }
 
   /* Invoke the closure.  */
-  (closure->fun) (cif, rvalue, avalue, closure->user_data);
+  (c->fun) (cif, rvalue, avalue, c->user_data);
 
-  debug(3, "after calling function, ret[0] = %08x, ret[1] = %08x\n", ret[0],
-	ret[1]);
+  debug(3, "after calling function, ret[0] = %08x, ret[1] = %08x\n", u.ret[0],
+	u.ret[1]);
 
   /* Store the result using the lower 2 bytes of the flags.  */
   switch (cif->flags)
     {
     case FFI_TYPE_UINT8:
-      *(stack - FIRST_ARG_SLOT) = (UINT8)(ret[0] >> 24);
+      *(stack - FIRST_ARG_SLOT) = (UINT8)(u.ret[0] >> 24);
       break;
     case FFI_TYPE_SINT8:
-      *(stack - FIRST_ARG_SLOT) = (SINT8)(ret[0] >> 24);
+      *(stack - FIRST_ARG_SLOT) = (SINT8)(u.ret[0] >> 24);
       break;
     case FFI_TYPE_UINT16:
-      *(stack - FIRST_ARG_SLOT) = (UINT16)(ret[0] >> 16);
+      *(stack - FIRST_ARG_SLOT) = (UINT16)(u.ret[0] >> 16);
       break;
     case FFI_TYPE_SINT16:
-      *(stack - FIRST_ARG_SLOT) = (SINT16)(ret[0] >> 16);
+      *(stack - FIRST_ARG_SLOT) = (SINT16)(u.ret[0] >> 16);
       break;
     case FFI_TYPE_INT:
     case FFI_TYPE_SINT32:
     case FFI_TYPE_UINT32:
-      *(stack - FIRST_ARG_SLOT) = ret[0];
+      *(stack - FIRST_ARG_SLOT) = u.ret[0];
       break;
     case FFI_TYPE_SINT64:
     case FFI_TYPE_UINT64:
-      *(stack - FIRST_ARG_SLOT) = ret[0];
-      *(stack - FIRST_ARG_SLOT - 1) = ret[1];
+      *(stack - FIRST_ARG_SLOT) = u.ret[0];
+      *(stack - FIRST_ARG_SLOT - 1) = u.ret[1];
       break;
 
     case FFI_TYPE_DOUBLE:
@@ -577,7 +580,7 @@ ffi_status ffi_closure_inner_pa32(ffi_closure *closure, UINT32 *stack)
     case FFI_TYPE_SMALL_STRUCT4:
       tmp = (void*)(stack -  FIRST_ARG_SLOT);
       tmp += 4 - cif->rtype->size;
-      memcpy((void*)tmp, &ret[0], cif->rtype->size);
+      memcpy((void*)tmp, &u, cif->rtype->size);
       break;
 
     case FFI_TYPE_SMALL_STRUCT5:
@@ -598,7 +601,7 @@ ffi_status ffi_closure_inner_pa32(ffi_closure *closure, UINT32 *stack)
 	  }
 
 	memset (ret2, 0, sizeof (ret2));
-	memcpy ((char *)ret2 + off, ret, 8 - off);
+	memcpy ((char *)ret2 + off, &u, 8 - off);
 
 	*(stack - FIRST_ARG_SLOT) = ret2[0];
 	*(stack - FIRST_ARG_SLOT - 1) = ret2[1];
@@ -630,89 +633,41 @@ ffi_prep_closure_loc (ffi_closure* closure,
 		      void *user_data,
 		      void *codeloc)
 {
-  UINT32 *tramp = (UINT32 *)(closure->tramp);
-#ifdef PA_HPUX
-  UINT32 *tmp;
-#endif
+  ffi_closure *c = (ffi_closure *)FFI_RESTORE_PTR (closure);
+
+  /* The layout of a function descriptor.  A function pointer with the PLABEL
+     bit set points to a function descriptor.  */
+  struct pa32_fd
+  {
+    UINT32 code_pointer;
+    UINT32 gp;
+  };
+
+  struct ffi_pa32_trampoline_struct
+  {
+     UINT32 code_pointer;        /* Pointer to ffi_closure_unix.  */
+     UINT32 fake_gp;             /* Pointer to closure, installed as gp.  */
+     UINT32 real_gp;             /* Real gp value.  */
+  };
+
+  struct ffi_pa32_trampoline_struct *tramp;
+  struct pa32_fd *fd;
 
   if (cif->abi != FFI_PA32)
     return FFI_BAD_ABI;
 
-  /* Make a small trampoline that will branch to our
-     handler function. Use PC-relative addressing.  */
-
-#ifdef PA_LINUX
-  tramp[0] = 0xeaa00000; /* b,l .+8,%r21        ; %r21 <- pc+8 */
-  tramp[1] = 0xd6a01c1e; /* depi 0,31,2,%r21    ; mask priv bits */
-  tramp[2] = 0x4aa10028; /* ldw 20(%r21),%r1    ; load plabel */
-  tramp[3] = 0x36b53ff1; /* ldo -8(%r21),%r21   ; get closure addr */
-  tramp[4] = 0x0c201096; /* ldw 0(%r1),%r22     ; address of handler */
-  tramp[5] = 0xeac0c000; /* bv%r0(%r22)         ; branch to handler */
-  tramp[6] = 0x0c281093; /* ldw 4(%r1),%r19     ; GP of handler */
-  tramp[7] = ((UINT32)(ffi_closure_pa32) & ~2);
-
-  /* Flush d/icache -- have to flush up 2 two lines because of
-     alignment.  */
-  __asm__ volatile(
-		   "fdc 0(%0)\n\t"
-		   "fdc %1(%0)\n\t"
-		   "fic 0(%%sr4, %0)\n\t"
-		   "fic %1(%%sr4, %0)\n\t"
-		   "sync\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n"
-		   :
-		   : "r"((unsigned long)tramp & ~31),
-		     "r"(32 /* stride */)
-		   : "memory");
-#endif
+  /* Get function descriptor address for ffi_closure_pa32.  */
+  fd = (struct pa32_fd *)((UINT32)ffi_closure_pa32 & ~3);
 
-#ifdef PA_HPUX
-  tramp[0] = 0xeaa00000; /* b,l .+8,%r21        ; %r21 <- pc+8  */
-  tramp[1] = 0xd6a01c1e; /* depi 0,31,2,%r21    ; mask priv bits  */
-  tramp[2] = 0x4aa10038; /* ldw 28(%r21),%r1    ; load plabel  */
-  tramp[3] = 0x36b53ff1; /* ldo -8(%r21),%r21   ; get closure addr  */
-  tramp[4] = 0x0c201096; /* ldw 0(%r1),%r22     ; address of handler  */
-  tramp[5] = 0x02c010b4; /* ldsid (%r22),%r20   ; load space id  */
-  tramp[6] = 0x00141820; /* mtsp %r20,%sr0      ; into %sr0  */
-  tramp[7] = 0xe2c00000; /* be 0(%sr0,%r22)     ; branch to handler  */
-  tramp[8] = 0x0c281093; /* ldw 4(%r1),%r19     ; GP of handler  */
-  tramp[9] = ((UINT32)(ffi_closure_pa32) & ~2);
-
-  /* Flush d/icache -- have to flush three lines because of alignment.  */
-  __asm__ volatile(
-		   "copy %1,%0\n\t"
-		   "fdc,m %2(%0)\n\t"
-		   "fdc,m %2(%0)\n\t"
-		   "fdc,m %2(%0)\n\t"
-		   "ldsid (%1),%0\n\t"
-		   "mtsp %0,%%sr0\n\t"
-		   "copy %1,%0\n\t"
-		   "fic,m %2(%%sr0,%0)\n\t"
-		   "fic,m %2(%%sr0,%0)\n\t"
-		   "fic,m %2(%%sr0,%0)\n\t"
-		   "sync\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n\t"
-		   "nop\n"
-		   : "=&r" ((unsigned long)tmp)
-		   : "r" ((unsigned long)tramp & ~31),
-		     "r" (32/* stride */)
-		   : "memory");
-#endif
+  /* Setup trampoline.  */
+  tramp = (struct ffi_pa32_trampoline_struct *)c->tramp;
+  tramp->code_pointer = fd->code_pointer;
+  tramp->fake_gp = (UINT32)codeloc & ~3;
+  tramp->real_gp = fd->gp;
 
-  closure->cif  = cif;
-  closure->user_data = user_data;
-  closure->fun  = fun;
+  c->cif  = cif;
+  c->user_data = user_data;
+  c->fun  = fun;
 
   return FFI_OK;
 }

data/ext/ffi_c/libffi/src/pa/ffitarget.h

@@ -68,12 +68,7 @@ typedef enum ffi_abi {
 
 #define FFI_CLOSURES 1
 #define FFI_NATIVE_RAW_API 0
-
-#ifdef PA_LINUX
-#define FFI_TRAMPOLINE_SIZE 32
-#else
-#define FFI_TRAMPOLINE_SIZE 40
-#endif
+#define FFI_TRAMPOLINE_SIZE 12
 
 #define FFI_TYPE_SMALL_STRUCT2 -1
 #define FFI_TYPE_SMALL_STRUCT3 -2

data/ext/ffi_c/libffi/src/pa/hpux32.S

@@ -259,7 +259,7 @@ L$done
 L$FE1
 
 	/* void ffi_closure_pa32(void);
-	   Called with closure argument in %r21 */
+	   Called with closure argument in %r19 */
 
 	.SPACE $TEXT$
 	.SUBSPA $CODE$
@@ -285,7 +285,9 @@ L$CFI22
 	stw	%arg2, -44(%r3)
 	stw	%arg3, -48(%r3)
 
-	copy	%r21, %arg0
+	/* Retrieve closure pointer and real gp.  */
+	copy	%r19, %arg0
+	ldw	8(%r19), %r19
 	bl	ffi_closure_inner_pa32, %r2
 	copy    %r3, %arg1
 	ldwm	-64(%sp), %r3

data/ext/ffi_c/libffi/src/pa/linux.S

@@ -252,7 +252,7 @@ ffi_call_pa32:
 .LFE1:
 
 	/* void ffi_closure_pa32(void);
-	   Called with closure argument in %r21 */
+	   Called with closure argument in %r19 */
 	.export ffi_closure_pa32,code
 	.import ffi_closure_inner_pa32,code
 
@@ -277,7 +277,9 @@ ffi_closure_pa32:
 	stw %arg2, -44(%r3)
 	stw %arg3, -48(%r3)
 
-	copy %r21, %arg0
+	/* Retrieve closure pointer and real gp.  */
+	copy    %r19, %arg0
+	ldw     8(%r19), %r19
 	bl ffi_closure_inner_pa32, %r2
 	copy %r3, %arg1
 
@@ -297,10 +299,18 @@ ffi_closure_pa32:
 .LSCIE1:
 	.word   0x0     ;# CIE Identifier Tag
 	.byte   0x1     ;# CIE Version
+#ifdef __PIC__
+	.ascii  "zR\0"  ;# CIE Augmentation: 'z' - data, 'R' - DW_EH_PE_... data
+#else
 	.ascii "\0"     ;# CIE Augmentation
+#endif
 	.uleb128 0x1    ;# CIE Code Alignment Factor
 	.sleb128 4      ;# CIE Data Alignment Factor
 	.byte   0x2     ;# CIE RA Column
+#ifdef __PIC__
+	.uleb128 0x1    ;# Augmentation size
+	.byte	0x1b    ;# FDE Encoding (DW_EH_PE_pcrel|DW_EH_PE_sdata4)
+#endif
 	.byte   0xc     ;# DW_CFA_def_cfa
 	.uleb128 0x1e
 	.uleb128 0x0
@@ -310,9 +320,15 @@ ffi_closure_pa32:
 	.word   .LEFDE1-.LASFDE1        ;# FDE Length
 .LASFDE1:
 	.word   .LASFDE1-.Lframe1       ;# FDE CIE offset
-	.word   .LFB1   ;# FDE initial location
+#ifdef __PIC__
+	.word	.LFB1-. ;# FDE initial location
+#else
+	.word	.LFB1   ;# FDE initial location
+#endif
 	.word   .LFE1-.LFB1     ;# FDE address range
-
+#ifdef __PIC__
+	.uleb128 0x0	;# Augmentation size: no data
+#endif
 	.byte   0x4     ;# DW_CFA_advance_loc4
 	.word   .LCFI11-.LFB1
 	.byte	0x83	;# DW_CFA_offset, column 0x3
@@ -338,8 +354,15 @@ ffi_closure_pa32:
 	.word   .LEFDE2-.LASFDE2        ;# FDE Length
 .LASFDE2:
 	.word   .LASFDE2-.Lframe1       ;# FDE CIE offset
+#ifdef __PIC__
+	.word   .LFB2-. ;# FDE initial location
+#else
 	.word   .LFB2   ;# FDE initial location
+#endif
 	.word   .LFE2-.LFB2     ;# FDE address range
+#ifdef __PIC__
+	.uleb128 0x0	;# Augmentation size: no data
+#endif
 	.byte   0x4     ;# DW_CFA_advance_loc4
 	.word   .LCFI21-.LFB2
 	.byte   0x83    ;# DW_CFA_offset, column 0x3

data/ext/ffi_c/libffi/src/powerpc/ffi.c

@@ -85,8 +85,9 @@ ffi_call_int (ffi_cif *cif,
      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.  */
-  unsigned long smst_buffer[8];
+     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;
@@ -121,8 +122,9 @@ ffi_call_int (ffi_cif *cif,
 # 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.  */
-	if (rsize <= 8)
+	   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

data/ext/ffi_c/libffi/src/powerpc/ffi_darwin.c

@@ -33,7 +33,10 @@
 #include <stdlib.h>
 
 extern void ffi_closure_ASM (void);
+
+#if defined (FFI_GO_CLOSURES)
 extern void ffi_go_closure_ASM (void);
+#endif
 
 enum {
   /* The assembly depends on these exact flags.  
@@ -909,8 +912,10 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 extern void ffi_call_AIX(extended_cif *, long, unsigned, unsigned *,
 			 void (*fn)(void), void (*fn2)(void));
 
+#if defined (FFI_GO_CLOSURES)
 extern void ffi_call_go_AIX(extended_cif *, long, unsigned, unsigned *,
 			    void (*fn)(void), void (*fn2)(void), void *closure);
+#endif
 
 extern void ffi_call_DARWIN(extended_cif *, long, unsigned, unsigned *,
 			    void (*fn)(void), void (*fn2)(void), ffi_type*);
@@ -950,6 +955,7 @@ ffi_call (ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
     }
 }
 
+#if defined (FFI_GO_CLOSURES)
 void
 ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue,
 	     void *closure)
@@ -981,6 +987,7 @@ ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue,
       break;
     }
 }
+#endif
 
 static void flush_icache(char *);
 static void flush_range(char *, int);
@@ -1110,6 +1117,7 @@ ffi_prep_closure_loc (ffi_closure* closure,
   return FFI_OK;
 }
 
+#if defined (FFI_GO_CLOSURES)
 ffi_status
 ffi_prep_go_closure (ffi_go_closure* closure,
 		     ffi_cif* cif,
@@ -1133,6 +1141,7 @@ ffi_prep_go_closure (ffi_go_closure* closure,
     }
   return FFI_OK;
 }
+#endif
 
 static void
 flush_icache(char *addr)
@@ -1168,9 +1177,11 @@ ffi_type *
 ffi_closure_helper_DARWIN (ffi_closure *, void *,
 			   unsigned long *, ffi_dblfl *);
 
+#if defined (FFI_GO_CLOSURES)
 ffi_type *
 ffi_go_closure_helper_DARWIN (ffi_go_closure*, void *,
 			      unsigned long *, ffi_dblfl *);
+#endif
 
 /* Basically the trampoline invokes ffi_closure_ASM, and on
    entry, r11 holds the address of the closure.
@@ -1430,6 +1441,7 @@ ffi_closure_helper_DARWIN (ffi_closure *closure, void *rvalue,
 				    closure->user_data, rvalue, pgr, pfr);
 }
 
+#if defined (FFI_GO_CLOSURES)
 ffi_type *
 ffi_go_closure_helper_DARWIN (ffi_go_closure *closure, void *rvalue,
 			      unsigned long *pgr, ffi_dblfl *pfr)
@@ -1437,4 +1449,4 @@ ffi_go_closure_helper_DARWIN (ffi_go_closure *closure, void *rvalue,
   return ffi_closure_helper_common (closure->cif, closure->fun,
 				    closure, rvalue, pgr, pfr);
 }
-
+#endif