ffi 1.11.3 → 1.12.0

data/ext/ffi_c/libffi/m4/ax_check_compile_flag.m4

@@ -29,33 +29,12 @@
 #   Copyright (c) 2008 Guido U. Draheim <guidod@gmx.de>
 #   Copyright (c) 2011 Maarten Bosmans <mkbosmans@gmail.com>
 #
-#   This program is free software: you can redistribute it and/or modify it
-#   under the terms of the GNU General Public License as published by the
-#   Free Software Foundation, either version 3 of the License, or (at your
-#   option) any later version.
-#
-#   This program is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
-#   Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License along
-#   with this program. If not, see <https://www.gnu.org/licenses/>.
-#
-#   As a special exception, the respective Autoconf Macro's copyright owner
-#   gives unlimited permission to copy, distribute and modify the configure
-#   scripts that are the output of Autoconf when processing the Macro. You
-#   need not follow the terms of the GNU General Public License when using
-#   or distributing such scripts, even though portions of the text of the
-#   Macro appear in them. The GNU General Public License (GPL) does govern
-#   all other use of the material that constitutes the Autoconf Macro.
-#
-#   This special exception to the GPL applies to versions of the Autoconf
-#   Macro released by the Autoconf Archive. When you make and distribute a
-#   modified version of the Autoconf Macro, you may extend this special
-#   exception to the GPL to apply to your modified version as well.
+#   Copying and distribution of this file, with or without modification, are
+#   permitted in any medium without royalty provided the copyright notice
+#   and this notice are preserved.  This file is offered as-is, without any
+#   warranty.
 
-#serial 5
+#serial 6
 
 AC_DEFUN([AX_CHECK_COMPILE_FLAG],
 [AC_PREREQ(2.64)dnl for _AC_LANG_PREFIX and AS_VAR_IF

data/ext/ffi_c/libffi/m4/ax_compiler_vendor.m4

@@ -44,7 +44,7 @@
 #   modified version of the Autoconf Macro, you may extend this special
 #   exception to the GPL to apply to your modified version as well.
 
-#serial 16
+#serial 17
 
 AC_DEFUN([AX_COMPILER_VENDOR],
 [AC_CACHE_CHECK([for _AC_LANG compiler vendor], ax_cv_[]_AC_LANG_ABBREV[]_compiler_vendor,
@@ -56,6 +56,7 @@ AC_DEFUN([AX_COMPILER_VENDOR],
            clang:     __clang__
            cray:      _CRAYC
            fujitsu:   __FUJITSU
+           sdcc:      SDCC, __SDCC
            gnu:       __GNUC__
            sun:       __SUNPRO_C,__SUNPRO_CC
            hp:        __HP_cc,__HP_aCC

data/ext/ffi_c/libffi/m4/ax_configure_args.m4

@@ -31,33 +31,12 @@
 #
 #   Copyright (c) 2008 Guido U. Draheim <guidod@gmx.de>
 #
-#   This program is free software; you can redistribute it and/or modify it
-#   under the terms of the GNU General Public License as published by the
-#   Free Software Foundation; either version 3 of the License, or (at your
-#   option) any later version.
-#
-#   This program is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
-#   Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License along
-#   with this program. If not, see <https://www.gnu.org/licenses/>.
-#
-#   As a special exception, the respective Autoconf Macro's copyright owner
-#   gives unlimited permission to copy, distribute and modify the configure
-#   scripts that are the output of Autoconf when processing the Macro. You
-#   need not follow the terms of the GNU General Public License when using
-#   or distributing such scripts, even though portions of the text of the
-#   Macro appear in them. The GNU General Public License (GPL) does govern
-#   all other use of the material that constitutes the Autoconf Macro.
-#
-#   This special exception to the GPL applies to versions of the Autoconf
-#   Macro released by the Autoconf Archive. When you make and distribute a
-#   modified version of the Autoconf Macro, you may extend this special
-#   exception to the GPL to apply to your modified version as well.
+#   Copying and distribution of this file, with or without modification, are
+#   permitted in any medium without royalty provided the copyright notice
+#   and this notice are preserved.  This file is offered as-is, without any
+#   warranty.
 
-#serial 13
+#serial 14
 
 AC_DEFUN([AX_CONFIGURE_ARGS],[
    # [$]@ is unusable in 2.60+ but earlier autoconf had no ac_configure_args

data/ext/ffi_c/libffi/m4/ax_gcc_archflag.m4

@@ -37,7 +37,7 @@
 #   Copyright (c) 2008 Steven G. Johnson <stevenj@alum.mit.edu>
 #   Copyright (c) 2008 Matteo Frigo
 #   Copyright (c) 2014 Tsukasa Oi
-#   Copyright (c) 2017 Alexey Kopytov
+#   Copyright (c) 2017-2018 Alexey Kopytov
 #
 #   This program is free software: you can redistribute it and/or modify it
 #   under the terms of the GNU General Public License as published by the
@@ -65,7 +65,7 @@
 #   modified version of the Autoconf Macro, you may extend this special
 #   exception to the GPL to apply to your modified version as well.
 
-#serial 20
+#serial 22
 
 AC_DEFUN([AX_GCC_ARCHFLAG],
 [AC_REQUIRE([AC_PROG_CC])
@@ -108,7 +108,7 @@ case $host_cpu in
 	    *2?6[[ad]]?:*:*:*) ax_gcc_arch="sandybridge corei7-avx corei7 core2 pentium-m pentium3 pentiumpro" ;;
 	    *3?6[[ae]]?:*:*:*) ax_gcc_arch="ivybridge core-avx-i corei7-avx corei7 core2 pentium-m pentium3 pentiumpro" ;;
 	    *3?6[[cf]]?:*:*:*|*4?6[[56]]?:*:*:*) ax_gcc_arch="haswell core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium3 pentiumpro" ;;
-	    *3?6d?:*:*:*) ax_gcc_arch="broadwell core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium3 pentiumpro" ;;
+	    *3?6d?:*:*:*|*4?6[[7f]]?:*:*:*|*5?66?:*:*:*) ax_gcc_arch="broadwell core-avx2 core-avx-i corei7-avx corei7 core2 pentium-m pentium3 pentiumpro" ;;
 	    *1?6c?:*:*:*|*2?6[[67]]?:*:*:*|*3?6[[56]]?:*:*:*) ax_gcc_arch="bonnell atom core2 pentium-m pentium3 pentiumpro" ;;
 	    *3?67?:*:*:*|*[[45]]?6[[ad]]?:*:*:*) ax_gcc_arch="silvermont atom core2 pentium-m pentium3 pentiumpro" ;;
 	    *000?f[[012]]?:*:*:*|?f[[012]]?:*:*:*|f[[012]]?:*:*:*) ax_gcc_arch="pentium4 pentiumpro" ;;
@@ -198,6 +198,10 @@ case $host_cpu in
        *POWER4*|*power4*|*gq*) ax_gcc_arch="power4 970";;
        *POWER5*|*power5*|*gr*|*gs*) ax_gcc_arch="power5 power4 970";;
        603ev|8240) ax_gcc_arch="$cputype 603e 603";;
+       *POWER7*) ax_gcc_arch="power7";;
+       *POWER8*) ax_gcc_arch="power8";;
+       *POWER9*) ax_gcc_arch="power9";;
+       *POWER10*) ax_gcc_arch="power10";;
        *) ax_gcc_arch=$cputype ;;
      esac
      ax_gcc_arch="$ax_gcc_arch powerpc"

data/ext/ffi_c/libffi/src/closures.c

@@ -1,5 +1,6 @@
 /* -----------------------------------------------------------------------
-   closures.c - Copyright (c) 2007, 2009, 2010  Red Hat, Inc.
+   closures.c - Copyright (c) 2019 Anthony Green
+                Copyright (c) 2007, 2009, 2010 Red Hat, Inc.
                 Copyright (C) 2007, 2009, 2010 Free Software Foundation, Inc
                 Copyright (c) 2011 Plausible Labs Cooperative, Inc.
 
@@ -936,7 +937,14 @@ void *
 ffi_data_to_code_pointer (void *data)
 {
   msegmentptr seg = segment_holding (gm, data);
-  return add_segment_exec_offset (data, seg);
+  /* We expect closures to be allocated with ffi_closure_alloc(), in
+     which case seg will be non-NULL.  However, some users take on the
+     burden of managing this memory themselves, in which case this
+     we'll just return data. */
+  if (seg)
+    return add_segment_exec_offset (data, seg);
+  else
+    return data;
 }
 
 /* Release a chunk of memory allocated with ffi_closure_alloc.  If

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/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;

data/ext/ffi_c/libffi/src/powerpc/ffi_linux64.c

@@ -38,7 +38,8 @@
 /* About the LINUX64 ABI.  */
 enum {
   NUM_GPR_ARG_REGISTERS64 = 8,
-  NUM_FPR_ARG_REGISTERS64 = 13
+  NUM_FPR_ARG_REGISTERS64 = 13,
+  NUM_VEC_ARG_REGISTERS64 = 12,
 };
 enum { ASM_NEEDS_REGISTERS64 = 4 };
 
@@ -81,11 +82,12 @@ discover_homogeneous_aggregate (ffi_abi abi,
          of FPRs, but according to the ABI must be considered
          distinct from doubles. They are also limited to a
          maximum of four members in a homogeneous aggregate. */
-      else
+      else if ((abi & FFI_LINUX_LONG_DOUBLE_IEEE128) == 0)
         {
           *elnum = 2;
           return FFI_TYPE_LONGDOUBLE;
         }
+      /* Fall through. */
 #endif
     case FFI_TYPE_FLOAT:
     case FFI_TYPE_DOUBLE:
@@ -130,13 +132,23 @@ ffi_prep_cif_linux64_core (ffi_cif *cif)
 {
   ffi_type **ptr;
   unsigned bytes;
-  unsigned i, fparg_count = 0, intarg_count = 0;
+  unsigned i, fparg_count = 0, intarg_count = 0, vecarg_count = 0;
   unsigned flags = cif->flags;
   unsigned elt, elnum, rtype;
 
 #if FFI_TYPE_LONGDOUBLE == FFI_TYPE_DOUBLE
-  /* If compiled without long double support..  */
-  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
+  /* If compiled without long double support... */
+  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0 ||
+      (cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+    return FFI_BAD_ABI;
+#elif !defined(__VEC__)
+  /* If compiled without vector register support (used by assembly)... */
+  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+    return FFI_BAD_ABI;
+#else
+  /* If the IEEE128 flag is set, but long double is only 64 bits wide... */
+  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) == 0 &&
+      (cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
     return FFI_BAD_ABI;
 #endif
 
@@ -166,6 +178,11 @@ homogeneous:
     {
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
     case FFI_TYPE_LONGDOUBLE:
+      if ((cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+        {
+          flags |= FLAG_RETURNS_VEC;
+          break;
+        }
       if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
 	flags |= FLAG_RETURNS_128BITS;
       /* Fall through.  */
@@ -221,6 +238,15 @@ homogeneous:
 	{
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 	case FFI_TYPE_LONGDOUBLE:
+          if ((cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+            {
+              vecarg_count++;
+              /* Align to 16 bytes, plus the 16-byte argument. */
+              intarg_count = (intarg_count + 3) & ~0x1;
+              if (vecarg_count > NUM_VEC_ARG_REGISTERS64)
+                flags |= FLAG_ARG_NEEDS_PSAVE;
+              break;
+            }
 	  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
 	    {
 	      fparg_count++;
@@ -248,6 +274,17 @@ homogeneous:
 	    }
 	  intarg_count += ((*ptr)->size + 7) / 8;
 	  elt = discover_homogeneous_aggregate (cif->abi, *ptr, &elnum);
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+          if (elt == FFI_TYPE_LONGDOUBLE &&
+              (cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+            {
+              vecarg_count += elnum;
+              if (vecarg_count > NUM_VEC_ARG_REGISTERS64)
+                flags |= FLAG_ARG_NEEDS_PSAVE;
+              break;
+            }
+	  else
+#endif
 	  if (elt)
 	    {
 	      fparg_count += elnum;
@@ -286,10 +323,17 @@ homogeneous:
     flags |= FLAG_FP_ARGUMENTS;
   if (intarg_count > 4)
     flags |= FLAG_4_GPR_ARGUMENTS;
+  if (vecarg_count != 0)
+    flags |= FLAG_VEC_ARGUMENTS;
 
   /* Space for the FPR registers, if needed.  */
   if (fparg_count != 0)
     bytes += NUM_FPR_ARG_REGISTERS64 * sizeof (double);
+  /* Space for the vector registers, if needed, aligned to 16 bytes. */
+  if (vecarg_count != 0) {
+    bytes = (bytes + 15) & ~0xF;
+    bytes += NUM_VEC_ARG_REGISTERS64 * sizeof (float128);
+  }
 
   /* Stack space.  */
 #if _CALL_ELF == 2
@@ -372,6 +416,8 @@ ffi_prep_cif_linux64_var (ffi_cif *cif,
    |--------------------------------------------| |
    |   FPR registers f1-f13 (optional)	13*8	| |
    |--------------------------------------------| |
+   |   VEC registers v2-v13 (optional)  12*16   | |
+   |--------------------------------------------| |
    |   Parameter save area		        | |
    |--------------------------------------------| |
    |   TOC save area			8	| |
@@ -401,6 +447,7 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
     unsigned long *ul;
     float *f;
     double *d;
+    float128 *f128;
     size_t p;
   } valp;
 
@@ -419,6 +466,11 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
   valp fpr_base;
   unsigned int fparg_count;
 
+  /* 'vec_base' points at the space for v2, and grows upwards as
+     we use vector registers.  */
+  valp vec_base;
+  unsigned int vecarg_count;
+
   unsigned int i, words, nargs, nfixedargs;
   ffi_type **ptr;
   double double_tmp;
@@ -435,6 +487,7 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
     unsigned long **ul;
     float **f;
     double **d;
+    float128 **f128;
   } p_argv;
   unsigned long gprvalue;
   unsigned long align;
@@ -449,11 +502,21 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
 #endif
   fpr_base.d = gpr_base.d - NUM_FPR_ARG_REGISTERS64;
   fparg_count = 0;
+  /* Place the vector args below the FPRs, if used, else the GPRs. */
+  if (ecif->cif->flags & FLAG_FP_ARGUMENTS)
+    vec_base.p = fpr_base.p & ~0xF;
+  else
+    vec_base.p = gpr_base.p;
+  vec_base.f128 -= NUM_VEC_ARG_REGISTERS64;
+  vecarg_count = 0;
   next_arg.ul = gpr_base.ul;
 
   /* Check that everything starts aligned properly.  */
   FFI_ASSERT (((unsigned long) (char *) stack & 0xF) == 0);
   FFI_ASSERT (((unsigned long) stacktop.c & 0xF) == 0);
+  FFI_ASSERT (((unsigned long) gpr_base.c & 0xF) == 0);
+  FFI_ASSERT (((unsigned long) gpr_end.c  & 0xF) == 0);
+  FFI_ASSERT (((unsigned long) vec_base.c & 0xF) == 0);
   FFI_ASSERT ((bytes & 0xF) == 0);
 
   /* Deal with return values that are actually pass-by-reference.  */
@@ -478,6 +541,22 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
 	{
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 	case FFI_TYPE_LONGDOUBLE:
+          if ((ecif->cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+            {
+              next_arg.p = FFI_ALIGN (next_arg.p, 16);
+              if (next_arg.ul == gpr_end.ul)
+                next_arg.ul = rest.ul;
+              if (vecarg_count < NUM_VEC_ARG_REGISTERS64 && i < nfixedargs)
+		memcpy (vec_base.f128++, *p_argv.f128, sizeof (float128));
+              else
+		memcpy (next_arg.f128, *p_argv.f128, sizeof (float128));
+              if (++next_arg.f128 == gpr_end.f128)
+                next_arg.f128 = rest.f128;
+              vecarg_count++;
+              FFI_ASSERT (__LDBL_MANT_DIG__ == 113);
+              FFI_ASSERT (flags & FLAG_VEC_ARGUMENTS);
+              break;
+            }
 	  if ((ecif->cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
 	    {
 	      double_tmp = (*p_argv.d)[0];
@@ -589,9 +668,29 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long *const stack)
 		void *v;
 		float *f;
 		double *d;
+		float128 *f128;
 	      } arg;
 
 	      arg.v = *p_argv.v;
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+              if (elt == FFI_TYPE_LONGDOUBLE &&
+                  (ecif->cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+                {
+                  do
+                    {
+                      if (vecarg_count < NUM_VEC_ARG_REGISTERS64
+                          && i < nfixedargs)
+		        memcpy (vec_base.f128++, arg.f128, sizeof (float128));
+                      else
+		        memcpy (next_arg.f128, arg.f128++, sizeof (float128));
+                      if (++next_arg.f128 == gpr_end.f128)
+                        next_arg.f128 = rest.f128;
+                      vecarg_count++;
+                    }
+                  while (--elnum != 0);
+                }
+              else
+#endif
 	      if (elt == FFI_TYPE_FLOAT)
 		{
 		  do
@@ -762,17 +861,20 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
 			    void *user_data,
 			    void *rvalue,
 			    unsigned long *pst,
-			    ffi_dblfl *pfr)
+                            ffi_dblfl *pfr,
+                            float128 *pvec)
 {
   /* rvalue is the pointer to space for return value in closure assembly */
   /* pst is the pointer to parameter save area
      (r3-r10 are stored into its first 8 slots by ffi_closure_LINUX64) */
   /* pfr is the pointer to where f1-f13 are stored in ffi_closure_LINUX64 */
+  /* pvec is the pointer to where v2-v13 are stored in ffi_closure_LINUX64 */
 
   void **avalue;
   ffi_type **arg_types;
   unsigned long i, avn, nfixedargs;
   ffi_dblfl *end_pfr = pfr + NUM_FPR_ARG_REGISTERS64;
+  float128 *end_pvec = pvec + NUM_VEC_ARG_REGISTERS64;
   unsigned long align;
 
   avalue = alloca (cif->nargs * sizeof (void *));
@@ -851,6 +953,7 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
 		unsigned long *ul;
 		float *f;
 		double *d;
+		float128 *f128;
 		size_t p;
 	      } to, from;
 
@@ -858,6 +961,17 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
 		 aggregate size is not greater than the space taken by
 		 the registers so store back to the register/parameter
 		 save arrays.  */
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+              if (elt == FFI_TYPE_LONGDOUBLE &&
+                  (cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+                {
+                  if (pvec + elnum <= end_pvec)
+                    to.v = pvec;
+                  else
+                    to.v = pst;
+                }
+              else
+#endif
 	      if (pfr + elnum <= end_pfr)
 		to.v = pfr;
 	      else
@@ -865,6 +979,23 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
 
 	      avalue[i] = to.v;
 	      from.ul = pst;
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+              if (elt == FFI_TYPE_LONGDOUBLE &&
+                  (cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+                {
+                  do
+                    {
+                      if (pvec < end_pvec && i < nfixedargs)
+		        memcpy (to.f128, pvec++, sizeof (float128));
+                      else
+		        memcpy (to.f128, from.f128, sizeof (float128));
+                      to.f128++;
+                      from.f128++;
+                    }
+                  while (--elnum != 0);
+                }
+              else
+#endif
 	      if (elt == FFI_TYPE_FLOAT)
 		{
 		  do
@@ -920,7 +1051,18 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
 
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 	case FFI_TYPE_LONGDOUBLE:
-	  if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
+          if ((cif->abi & FFI_LINUX_LONG_DOUBLE_IEEE128) != 0)
+            {
+              if (((unsigned long) pst & 0xF) != 0)
+                ++pst;
+              if (pvec < end_pvec && i < nfixedargs)
+                avalue[i] = pvec++;
+              else
+                avalue[i] = pst;
+              pst += 2;
+              break;
+            }
+          else if ((cif->abi & FFI_LINUX_LONG_DOUBLE_128) != 0)
 	    {
 	      if (pfr + 1 < end_pfr && i + 1 < nfixedargs)
 		{
@@ -995,13 +1137,17 @@ ffi_closure_helper_LINUX64 (ffi_cif *cif,
   /* Tell ffi_closure_LINUX64 how to perform return type promotions.  */
   if ((cif->flags & FLAG_RETURNS_SMST) != 0)
     {
-      if ((cif->flags & FLAG_RETURNS_FP) == 0)
+      if ((cif->flags & (FLAG_RETURNS_FP | FLAG_RETURNS_VEC)) == 0)
 	return FFI_V2_TYPE_SMALL_STRUCT + cif->rtype->size - 1;
+      else if ((cif->flags & FLAG_RETURNS_VEC) != 0)
+        return FFI_V2_TYPE_VECTOR_HOMOG;
       else if ((cif->flags & FLAG_RETURNS_64BITS) != 0)
 	return FFI_V2_TYPE_DOUBLE_HOMOG;
       else
 	return FFI_V2_TYPE_FLOAT_HOMOG;
     }
+  if ((cif->flags & FLAG_RETURNS_VEC) != 0)
+    return FFI_V2_TYPE_VECTOR;
   return cif->rtype->type;
 }
 #endif