Add powerpc soft float support

.pc/applied-patches

@@ -34,3 +34,4 @@ kfreebsd-mmap
 minix
 interix-patch
 remove-debug-code
+powerpc-ffi-softfloat

.pc/powerpc-ffi-softfloat/src/powerpc/ffitarget.h

@@ -0,0 +1,139 @@
+/* -----------------------------------------------------------------*-C-*-
+   ffitarget.h - Copyright (c) 1996-2003  Red Hat, Inc.
+   Copyright (C) 2007, 2008, 2010 Free Software Foundation, Inc
+   Target configuration macros for PowerPC.
+
+   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
+
+/* ---- System specific configurations ----------------------------------- */
+
+#if defined (POWERPC) && defined (__powerpc64__)	/* linux64 */
+#ifndef POWERPC64
+#define POWERPC64
+#endif
+#elif defined (POWERPC_DARWIN) && defined (__ppc64__)	/* Darwin64 */
+#ifndef POWERPC64
+#define POWERPC64
+#endif
+#ifndef POWERPC_DARWIN64
+#define POWERPC_DARWIN64
+#endif
+#elif defined (POWERPC_AIX) && defined (__64BIT__)	/* AIX64 */
+#ifndef POWERPC64
+#define POWERPC64
+#endif
+#endif
+
+#ifndef LIBFFI_ASM
+typedef unsigned long          ffi_arg;
+typedef signed long            ffi_sarg;
+
+typedef enum ffi_abi {
+  FFI_FIRST_ABI = 0,
+
+#ifdef POWERPC
+  FFI_SYSV,
+  FFI_GCC_SYSV,
+  FFI_LINUX64,
+  FFI_LINUX,
+  FFI_LINUX_SOFT_FLOAT,
+# ifdef POWERPC64
+  FFI_DEFAULT_ABI = FFI_LINUX64,
+# else
+#  if (!defined(__NO_FPRS__) && (__LDBL_MANT_DIG__ == 106))
+  FFI_DEFAULT_ABI = FFI_LINUX,
+#  else
+#   ifdef __NO_FPRS__
+  FFI_DEFAULT_ABI = FFI_LINUX_SOFT_FLOAT,
+#   else
+  FFI_DEFAULT_ABI = FFI_GCC_SYSV,
+#   endif
+#  endif
+# endif
+#endif
+
+#ifdef POWERPC_AIX
+  FFI_AIX,
+  FFI_DARWIN,
+  FFI_DEFAULT_ABI = FFI_AIX,
+#endif
+
+#ifdef POWERPC_DARWIN
+  FFI_AIX,
+  FFI_DARWIN,
+  FFI_DEFAULT_ABI = FFI_DARWIN,
+#endif
+
+#ifdef POWERPC_FREEBSD
+  FFI_SYSV,
+  FFI_GCC_SYSV,
+  FFI_LINUX64,
+  FFI_LINUX,
+  FFI_LINUX_SOFT_FLOAT,
+  FFI_DEFAULT_ABI = FFI_SYSV,
+#endif
+
+  FFI_LAST_ABI
+} ffi_abi;
+#endif
+
+/* ---- Definitions for closures ----------------------------------------- */
+
+#define FFI_CLOSURES 1
+#define FFI_NATIVE_RAW_API 0
+
+/* For additional types like the below, take care about the order in
+   ppc_closures.S. They must follow after the FFI_TYPE_LAST.  */
+
+/* Needed for soft-float long-double-128 support.  */
+#define FFI_TYPE_UINT128 (FFI_TYPE_LAST + 1)
+
+/* Needed for FFI_SYSV small structure returns.
+   We use two flag bits, (FLAG_SYSV_SMST_R3, FLAG_SYSV_SMST_R4) which are
+   defined in ffi.c, to determine the exact return type and its size.  */
+#define FFI_SYSV_TYPE_SMALL_STRUCT (FFI_TYPE_LAST + 2)
+
+#if defined(POWERPC64) || defined(POWERPC_AIX)
+#  if defined(POWERPC_DARWIN64)
+#    define FFI_TRAMPOLINE_SIZE 48
+#  else
+#    define FFI_TRAMPOLINE_SIZE 24
+#  endif
+#else /* POWERPC || POWERPC_AIX */
+#  define FFI_TRAMPOLINE_SIZE 40
+#endif
+
+#ifndef LIBFFI_ASM
+#if defined(POWERPC_DARWIN) || defined(POWERPC_AIX)
+struct ffi_aix_trampoline_struct {
+    void * code_pointer;	/* Pointer to ffi_closure_ASM */
+    void * toc;			/* TOC */
+    void * static_chain;	/* Pointer to closure */
+};
+#endif
+#endif
+
+#endif

.pc/powerpc-ffi-softfloat/src/powerpc/ppc_closure.S

@@ -0,0 +1,327 @@
+/* -----------------------------------------------------------------------
+   sysv.h - Copyright (c) 2003 Jakub Jelinek <jakub@redhat.com>
+	    Copyright (c) 2008 Red Hat, Inc.
+
+   PowerPC Assembly glue.
+
+   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.
+   ----------------------------------------------------------------------- */
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+#include <powerpc/asm.h>
+
+	.file   "ppc_closure.S"
+
+#ifndef __powerpc64__
+
+ENTRY(ffi_closure_SYSV)
+.LFB1:
+	stwu %r1,-144(%r1)
+.LCFI0:
+	mflr %r0
+.LCFI1:
+	stw %r0,148(%r1)
+
+# we want to build up an areas for the parameters passed
+# in registers (both floating point and integer)
+
+	# so first save gpr 3 to gpr 10 (aligned to 4)
+	stw   %r3, 16(%r1)
+	stw   %r4, 20(%r1)
+	stw   %r5, 24(%r1)
+	stw   %r6, 28(%r1)
+	stw   %r7, 32(%r1)
+	stw   %r8, 36(%r1)
+	stw   %r9, 40(%r1)
+	stw   %r10,44(%r1)
+
+#ifndef __NO_FPRS__
+	# next save fpr 1 to fpr 8 (aligned to 8)
+	stfd  %f1, 48(%r1)
+	stfd  %f2, 56(%r1)
+	stfd  %f3, 64(%r1)
+	stfd  %f4, 72(%r1)
+	stfd  %f5, 80(%r1)
+	stfd  %f6, 88(%r1)
+	stfd  %f7, 96(%r1)
+	stfd  %f8, 104(%r1)
+#endif
+
+	# 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,%r1,112
+
+	# now load up the pointer to the saved gpr registers
+	addi %r5,%r1,16
+
+	# now load up the pointer to the saved fpr registers */
+	addi %r6,%r1,48
+
+	# now load up the pointer to the outgoing parameter
+	# stack in the previous frame
+	# i.e. the previous frame pointer + 8
+	addi %r7,%r1,152
+
+	# make the call
+	bl ffi_closure_helper_SYSV@local
+.Lret:
+	# 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
+
+	mflr %r4		# move address of .Lret to r4
+	slwi %r3,%r3,4		# now multiply return type by 16
+	addi %r4, %r4, .Lret_type0 - .Lret
+	lwz %r0,148(%r1)
+	add %r3,%r3,%r4		# add contents of table to table address
+	mtctr %r3
+	bctr			# jump to it
+.LFE1:
+
+# Each of the ret_typeX code fragments has to be exactly 16 bytes long
+# (4 instructions). For cache effectiveness we align to a 16 byte boundary
+# first.
+	.align 4
+# case FFI_TYPE_VOID
+.Lret_type0:
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+	nop
+
+# case FFI_TYPE_INT
+	lwz %r3,112+0(%r1)
+	mtlr %r0
+.Lfinish:
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_FLOAT
+	lfs %f1,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_DOUBLE
+	lfd %f1,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_LONGDOUBLE
+	lfd %f1,112+0(%r1)
+	lfd %f2,112+8(%r1)
+	mtlr %r0
+	b .Lfinish
+
+# case FFI_TYPE_UINT8
+	lbz %r3,112+3(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_SINT8
+	lbz %r3,112+3(%r1)
+	extsb %r3,%r3
+	mtlr %r0
+	b .Lfinish
+
+# case FFI_TYPE_UINT16
+	lhz %r3,112+2(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_SINT16
+	lha %r3,112+2(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_UINT32
+	lwz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_SINT32
+	lwz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_UINT64
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	mtlr %r0
+	b .Lfinish
+
+# case FFI_TYPE_SINT64
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	mtlr %r0
+	b .Lfinish
+
+# case FFI_TYPE_STRUCT
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+	nop
+
+# case FFI_TYPE_POINTER
+	lwz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_TYPE_UINT128
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	lwz %r5,112+8(%r1)
+	bl .Luint128
+
+# The return types below are only used when the ABI type is FFI_SYSV.
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 1. One byte struct.
+	lbz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 2. Two byte struct.
+	lhz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 3. Three byte struct.
+	lwz %r3,112+0(%r1)
+	srwi %r3,%r3,8
+	mtlr %r0
+	b .Lfinish
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 4. Four byte struct.
+	lwz %r3,112+0(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 5. Five byte struct.
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	li %r5,24
+	b .Lstruct567
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 6. Six byte struct.
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	li %r5,16
+	b .Lstruct567
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 7. Seven byte struct.
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	li %r5,8
+	b .Lstruct567
+
+# case FFI_SYSV_TYPE_SMALL_STRUCT + 8. Eight byte struct.
+	lwz %r3,112+0(%r1)
+	lwz %r4,112+4(%r1)
+	mtlr %r0
+	b .Lfinish
+
+.Lstruct567:
+	subfic %r6,%r5,32
+	srw %r4,%r4,%r5
+	slw %r6,%r3,%r6
+	srw %r3,%r3,%r5
+	or %r4,%r6,%r4
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+.Luint128:
+	lwz %r6,112+12(%r1)
+	mtlr %r0
+	addi %r1,%r1,144
+	blr
+
+END(ffi_closure_SYSV)
+
+	.section	".eh_frame",EH_FRAME_FLAGS,@progbits
+.Lframe1:
+	.4byte	.LECIE1-.LSCIE1	 # Length of Common Information Entry
+.LSCIE1:
+	.4byte	0x0	 # CIE Identifier Tag
+	.byte	0x1	 # CIE Version
+#if defined _RELOCATABLE || defined __PIC__
+	.ascii "zR\0"	 # CIE Augmentation
+#else
+	.ascii "\0"	 # CIE Augmentation
+#endif
+	.uleb128 0x1	 # CIE Code Alignment Factor
+	.sleb128 -4	 # CIE Data Alignment Factor
+	.byte	0x41	 # CIE RA Column
+#if defined _RELOCATABLE || defined __PIC__
+	.uleb128 0x1	 # Augmentation size
+	.byte	0x1b	 # FDE Encoding (pcrel sdata4)
+#endif
+	.byte	0xc	 # DW_CFA_def_cfa
+	.uleb128 0x1
+	.uleb128 0x0
+	.align 2
+.LECIE1:
+.LSFDE1:
+	.4byte	.LEFDE1-.LASFDE1	 # FDE Length
+.LASFDE1:
+	.4byte	.LASFDE1-.Lframe1	 # FDE CIE offset
+#if defined _RELOCATABLE || defined __PIC__
+	.4byte	.LFB1-.	 # FDE initial location
+#else
+	.4byte	.LFB1	 # FDE initial location
+#endif
+	.4byte	.LFE1-.LFB1	 # FDE address range
+#if defined _RELOCATABLE || defined __PIC__
+	.uleb128 0x0	 # Augmentation size
+#endif
+	.byte	0x4	 # DW_CFA_advance_loc4
+	.4byte	.LCFI0-.LFB1
+	.byte	0xe	 # DW_CFA_def_cfa_offset
+	.uleb128 144
+	.byte	0x4	 # DW_CFA_advance_loc4
+	.4byte	.LCFI1-.LCFI0
+	.byte	0x11	 # DW_CFA_offset_extended_sf
+	.uleb128 0x41
+	.sleb128 -1
+	.align 2
+.LEFDE1:
+
+#endif
+
+#if defined __ELF__ && defined __linux__
+	.section	.note.GNU-stack,"",@progbits
+#endif

.pc/powerpc-ffi-softfloat/src/powerpc/sysv.S

@@ -0,0 +1,214 @@
+/* -----------------------------------------------------------------------
+   sysv.S - Copyright (c) 1998 Geoffrey Keating
+   Copyright (C) 2007 Free Software Foundation, Inc
+
+   PowerPC Assembly glue.
+
+   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.
+   ----------------------------------------------------------------------- */
+
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+#include <powerpc/asm.h>
+
+#ifndef __powerpc64__
+	.globl ffi_prep_args_SYSV
+ENTRY(ffi_call_SYSV)
+.LFB1:
+	/* Save the old stack pointer as AP.  */
+	mr	%r8,%r1
+
+.LCFI0:
+	/* Allocate the stack space we need.  */
+	stwux	%r1,%r1,%r4
+	/* Save registers we use.  */
+	mflr	%r9
+	stw	%r28,-16(%r8)
+.LCFI1:
+	stw	%r29,-12(%r8)
+.LCFI2:
+	stw	%r30, -8(%r8)
+.LCFI3:
+	stw	%r31, -4(%r8)
+.LCFI4:
+	stw	%r9,   4(%r8)
+.LCFI5:
+
+	/* Save arguments over call...  */
+	mr	%r31,%r5	/* flags, */
+	mr	%r30,%r6	/* rvalue, */
+	mr	%r29,%r7	/* function address, */
+	mr	%r28,%r8	/* our AP. */
+.LCFI6:
+
+	/* Call ffi_prep_args_SYSV.  */
+	mr	%r4,%r1
+	bl	ffi_prep_args_SYSV@local
+
+	/* Now do the call.  */
+	/* Set up cr1 with bits 4-7 of the flags.  */
+	mtcrf	0x40,%r31
+	/* Get the address to call into CTR.  */
+	mtctr	%r29
+	/* Load all those argument registers.  */
+	lwz	%r3,-16-(8*4)(%r28)
+	lwz	%r4,-16-(7*4)(%r28)
+	lwz	%r5,-16-(6*4)(%r28)
+	lwz	%r6,-16-(5*4)(%r28)
+	bf-	5,1f
+	nop
+	lwz	%r7,-16-(4*4)(%r28)
+	lwz	%r8,-16-(3*4)(%r28)
+	lwz	%r9,-16-(2*4)(%r28)
+	lwz	%r10,-16-(1*4)(%r28)
+	nop
+1:
+
+	/* Load all the FP registers.  */
+	bf-	6,2f
+	lfd	%f1,-16-(8*4)-(8*8)(%r28)
+	lfd	%f2,-16-(8*4)-(7*8)(%r28)
+	lfd	%f3,-16-(8*4)-(6*8)(%r28)
+	lfd	%f4,-16-(8*4)-(5*8)(%r28)
+	nop
+	lfd	%f5,-16-(8*4)-(4*8)(%r28)
+	lfd	%f6,-16-(8*4)-(3*8)(%r28)
+	lfd	%f7,-16-(8*4)-(2*8)(%r28)
+	lfd	%f8,-16-(8*4)-(1*8)(%r28)
+2:
+
+	/* Make the call.  */
+	bctrl
+
+	/* Now, deal with the return value.  */
+	mtcrf	0x01,%r31 /* cr7  */
+	bt-	31,L(small_struct_return_value)
+	bt-	30,L(done_return_value)
+	bt-	29,L(fp_return_value)
+	stw	%r3,0(%r30)
+	bf+	28,L(done_return_value)
+	stw	%r4,4(%r30)
+	mtcrf	0x02,%r31 /* cr6  */
+	bf	27,L(done_return_value)
+	stw     %r5,8(%r30)
+	stw	%r6,12(%r30)
+	/* Fall through...  */
+
+L(done_return_value):
+	/* Restore the registers we used and return.  */
+	lwz	%r9,   4(%r28)
+	lwz	%r31, -4(%r28)
+	mtlr	%r9
+	lwz	%r30, -8(%r28)
+	lwz	%r29,-12(%r28)
+	lwz	%r28,-16(%r28)
+	lwz	%r1,0(%r1)
+	blr
+
+L(fp_return_value):
+	bf	28,L(float_return_value)
+	stfd	%f1,0(%r30)
+	mtcrf   0x02,%r31 /* cr6  */
+	bf	27,L(done_return_value)
+	stfd	%f2,8(%r30)
+	b	L(done_return_value)
+L(float_return_value):
+	stfs	%f1,0(%r30)
+	b	L(done_return_value)
+
+L(small_struct_return_value):
+	/*
+	 * The C code always allocates a properly-aligned 8-byte bounce
+	 * buffer to make this assembly code very simple.  Just write out
+	 * r3 and r4 to the buffer to allow the C code to handle the rest.
+	 */
+	stw %r3, 0(%r30)
+	stw %r4, 4(%r30)
+	b L(done_return_value)
+
+.LFE1:
+END(ffi_call_SYSV)
+
+      .section	".eh_frame",EH_FRAME_FLAGS,@progbits
+.Lframe1:
+      .4byte    .LECIE1-.LSCIE1  /*  Length of Common Information Entry */
+.LSCIE1:
+      .4byte    0x0      /*  CIE Identifier Tag */
+      .byte     0x1      /*  CIE Version */
+#if defined _RELOCATABLE || defined __PIC__
+      .ascii	"zR\0"   /*  CIE Augmentation */
+#else
+      .ascii	"\0"	 /*  CIE Augmentation */
+#endif
+      .uleb128  0x1      /*  CIE Code Alignment Factor */
+      .sleb128  -4	 /*  CIE Data Alignment Factor */
+      .byte     0x41     /*  CIE RA Column */
+#if defined _RELOCATABLE || defined __PIC__
+      .uleb128  0x1      /*  Augmentation size */
+      .byte	0x1b	 /*  FDE Encoding (pcrel sdata4) */
+#endif
+      .byte     0xc      /*  DW_CFA_def_cfa */
+      .uleb128  0x1
+      .uleb128  0x0
+      .align 2
+.LECIE1:
+.LSFDE1:
+      .4byte    .LEFDE1-.LASFDE1         /*  FDE Length */
+.LASFDE1:
+      .4byte    .LASFDE1-.Lframe1         /*  FDE CIE offset */
+#if defined _RELOCATABLE || defined __PIC__
+      .4byte    .LFB1-.  /*  FDE initial location */
+#else
+      .4byte    .LFB1    /*  FDE initial location */
+#endif
+      .4byte    .LFE1-.LFB1      /*  FDE address range */
+#if defined _RELOCATABLE || defined __PIC__
+      .uleb128  0x0	 /*  Augmentation size */
+#endif
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI0-.LFB1
+      .byte     0xd      /*  DW_CFA_def_cfa_register */
+      .uleb128  0x08
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI5-.LCFI0
+      .byte     0x11     /*  DW_CFA_offset_extended_sf */
+      .uleb128  0x41
+      .sleb128  -1
+      .byte     0x9f     /*  DW_CFA_offset, column 0x1f */
+      .uleb128  0x1
+      .byte     0x9e     /*  DW_CFA_offset, column 0x1e */
+      .uleb128  0x2
+      .byte     0x9d     /*  DW_CFA_offset, column 0x1d */
+      .uleb128  0x3
+      .byte     0x9c     /*  DW_CFA_offset, column 0x1c */
+      .uleb128  0x4
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI6-.LCFI5
+      .byte     0xd      /*  DW_CFA_def_cfa_register */
+      .uleb128  0x1c
+      .align 2
+.LEFDE1:
+#endif
+
+#if defined __ELF__ && defined __linux__
+	.section	.note.GNU-stack,"",@progbits
+#endif

ChangeLog

@@ -1,3 +1,9 @@
+2011-11-12  Kyle Moffett <Kyle.D.Moffett@boeing.com>
+
+	* src/powerpc/ffi.c, src/powerpc/ffitarget.h,
+	src/powerpc/ppc_closure.S, src/powerpc/sysv.S: Many changes for
+	softfloat powerpc variants.
+
 2011-11-12  Petr Salinger <Petr.Salinger@seznam.cz>
 
 	* configure.ac (FFI_EXEC_TRAMPOLINE_TABLE): Fix kfreebsd support.
@@ -17,7 +23,7 @@
 2011-11-12  Kimura Wataru  <kimuraw@i.nifty.jp>
 
 	* m4/ax_enable_builddir: Change from string comparison to numeric
-	comparison for wc output. 
+	comparison for wc output.
 	* configure.ac: Enable FFI_MMAP_EXEC_WRIT for darwin11 aka Mac OS
 	X 10.7.
 	* configure: Rebuilt.

patches/powerpc-ffi-softfloat

@@ -0,0 +1,970 @@
+Index: libffi/ChangeLog
+===================================================================
+--- libffi.orig/ChangeLog
++++ libffi/ChangeLog
+@@ -1,3 +1,9 @@
++2011-11-12  Kyle Moffett <Kyle.D.Moffett@boeing.com>
++
++	* src/powerpc/ffi.c, src/powerpc/ffitarget.h,
++	src/powerpc/ppc_closure.S, src/powerpc/sysv.S: Many changes for
++	softfloat powerpc variants.
++
+ 2011-11-12  Petr Salinger <Petr.Salinger@seznam.cz>
+ 
+ 	* configure.ac (FFI_EXEC_TRAMPOLINE_TABLE): Fix kfreebsd support.
+@@ -17,7 +23,7 @@
+ 2011-11-12  Kimura Wataru  <kimuraw@i.nifty.jp>
+ 
+ 	* m4/ax_enable_builddir: Change from string comparison to numeric
+-	comparison for wc output. 
++	comparison for wc output.
+ 	* configure.ac: Enable FFI_MMAP_EXEC_WRIT for darwin11 aka Mac OS
+ 	X 10.7.
+ 	* configure: Rebuilt.
+Index: libffi/src/powerpc/ffi.c
+===================================================================
+--- libffi.orig/src/powerpc/ffi.c
++++ libffi/src/powerpc/ffi.c
+@@ -41,27 +41,28 @@ enum {
+   /* The assembly depends on these exact flags.  */
+   FLAG_RETURNS_SMST	= 1 << (31-31), /* Used for FFI_SYSV small structs.  */
+   FLAG_RETURNS_NOTHING  = 1 << (31-30), /* These go in cr7 */
++#ifndef __NO_FPRS__
+   FLAG_RETURNS_FP       = 1 << (31-29),
++#endif
+   FLAG_RETURNS_64BITS   = 1 << (31-28),
+ 
+   FLAG_RETURNS_128BITS  = 1 << (31-27), /* cr6  */
+ 
+   FLAG_ARG_NEEDS_COPY   = 1 << (31- 7),
++#ifndef __NO_FPRS__
+   FLAG_FP_ARGUMENTS     = 1 << (31- 6), /* cr1.eq; specified by ABI */
++#endif
+   FLAG_4_GPR_ARGUMENTS  = 1 << (31- 5),
+   FLAG_RETVAL_REFERENCE = 1 << (31- 4)
+ };
+ 
+ /* About the SYSV ABI.  */
+-unsigned int NUM_GPR_ARG_REGISTERS = 8;
++#define ASM_NEEDS_REGISTERS 4
++#define NUM_GPR_ARG_REGISTERS 8
+ #ifndef __NO_FPRS__
+-unsigned int NUM_FPR_ARG_REGISTERS = 8;
+-#else
+-unsigned int NUM_FPR_ARG_REGISTERS = 0;
++# define NUM_FPR_ARG_REGISTERS 8
+ #endif
+ 
+-enum { ASM_NEEDS_REGISTERS = 4 };
+-
+ /* ffi_prep_args_SYSV is called by the assembly routine once stack space
+    has been allocated for the function's arguments.
+ 
+@@ -110,10 +111,12 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+   valp gpr_base;
+   int intarg_count;
+ 
++#ifndef __NO_FPRS__
+   /* 'fpr_base' points at the space for fpr1, and grows upwards as
+      we use FPR registers.  */
+   valp fpr_base;
+   int fparg_count;
++#endif
+ 
+   /* 'copy_space' grows down as we put structures in it.  It should
+      stay 16-byte aligned.  */
+@@ -122,9 +125,8 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+   /* 'next_arg' grows up as we put parameters in it.  */
+   valp next_arg;
+ 
+-  int i, ii MAYBE_UNUSED;
++  int i;
+   ffi_type **ptr;
+-  double double_tmp;
+   union {
+     void **v;
+     char **c;
+@@ -140,15 +142,16 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+   size_t struct_copy_size;
+   unsigned gprvalue;
+ 
+-  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
+-    NUM_FPR_ARG_REGISTERS = 0;
+-
+   stacktop.c = (char *) stack + bytes;
+   gpr_base.u = stacktop.u - ASM_NEEDS_REGISTERS - NUM_GPR_ARG_REGISTERS;
+   intarg_count = 0;
++#ifndef __NO_FPRS__
+   fpr_base.d = gpr_base.d - NUM_FPR_ARG_REGISTERS;
+   fparg_count = 0;
+   copy_space.c = ((flags & FLAG_FP_ARGUMENTS) ? fpr_base.c : gpr_base.c);
++#else
++  copy_space.c = gpr_base.c;
++#endif
+   next_arg.u = stack + 2;
+ 
+   /* Check that everything starts aligned properly.  */
+@@ -171,12 +174,28 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+        i > 0;
+        i--, ptr++, p_argv.v++)
+     {
+-      switch ((*ptr)->type)
+-	{
++      unsigned short typenum = (*ptr)->type;
++
++      /* We may need to handle some values depending on ABI */
++      if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT) {
++		if (typenum == FFI_TYPE_FLOAT)
++			typenum = FFI_TYPE_UINT32;
++		if (typenum == FFI_TYPE_DOUBLE)
++			typenum = FFI_TYPE_UINT64;
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_UINT128;
++      } else if (ecif->cif->abi != FFI_LINUX) {
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_STRUCT;
++#endif
++      }
++
++      /* Now test the translated value */
++      switch (typenum) {
++#ifndef __NO_FPRS__
+ 	case FFI_TYPE_FLOAT:
+ 	  /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
+-	  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	    goto soft_float_prep;
+ 	  double_tmp = **p_argv.f;
+ 	  if (fparg_count >= NUM_FPR_ARG_REGISTERS)
+ 	    {
+@@ -215,43 +234,6 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+ 
+ #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+ 	case FFI_TYPE_LONGDOUBLE:
+-	  if ((ecif->cif->abi != FFI_LINUX)
+-		&& (ecif->cif->abi != FFI_LINUX_SOFT_FLOAT))
+-	    goto do_struct;
+-	  /* The soft float ABI for long doubles works like this,
+-	     a long double is passed in four consecutive gprs if available.
+-	     A maximum of 2 long doubles can be passed in gprs.
+-	     If we do not have 4 gprs left, the long double is passed on the
+-	     stack, 4-byte aligned.  */
+-	  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	    {
+-	      unsigned int int_tmp = (*p_argv.ui)[0];
+-	      if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3)
+-		{
+-		  if (intarg_count < NUM_GPR_ARG_REGISTERS)
+-		    intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
+-		  *next_arg.u = int_tmp;
+-		  next_arg.u++;
+-		  for (ii = 1; ii < 4; ii++)
+-		    {
+-		      int_tmp = (*p_argv.ui)[ii];
+-		      *next_arg.u = int_tmp;
+-		      next_arg.u++;
+-		    }
+-		}
+-	      else
+-		{
+-		  *gpr_base.u++ = int_tmp;
+-		  for (ii = 1; ii < 4; ii++)
+-		    {
+-		      int_tmp = (*p_argv.ui)[ii];
+-		      *gpr_base.u++ = int_tmp;
+-		    }
+-		}
+-	      intarg_count +=4;
+-	    }
+-	  else
+-	    {
+ 	      double_tmp = (*p_argv.d)[0];
+ 
+ 	      if (fparg_count >= NUM_FPR_ARG_REGISTERS - 1)
+@@ -277,13 +259,40 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+ 
+ 	      fparg_count += 2;
+ 	      FFI_ASSERT (flags & FLAG_FP_ARGUMENTS);
+-	    }
+ 	  break;
+ #endif
++#endif /* have FPRs */
++
++	/*
++	 * The soft float ABI for long doubles works like this, a long double
++	 * is passed in four consecutive GPRs if available.  A maximum of 2
++	 * long doubles can be passed in gprs.  If we do not have 4 GPRs
++	 * left, the long double is passed on the stack, 4-byte aligned.
++	 */
++	case FFI_TYPE_UINT128: {
++		unsigned int int_tmp = (*p_argv.ui)[0];
++		unsigned int ii;
++		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3) {
++			if (intarg_count < NUM_GPR_ARG_REGISTERS)
++				intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
++			*(next_arg.u++) = int_tmp;
++			for (ii = 1; ii < 4; ii++) {
++				int_tmp = (*p_argv.ui)[ii];
++				*(next_arg.u++) = int_tmp;
++			}
++		} else {
++			*(gpr_base.u++) = int_tmp;
++			for (ii = 1; ii < 4; ii++) {
++				int_tmp = (*p_argv.ui)[ii];
++				*(gpr_base.u++) = int_tmp;
++			}
++		}
++		intarg_count += 4;
++		break;
++	}
+ 
+ 	case FFI_TYPE_UINT64:
+ 	case FFI_TYPE_SINT64:
+-	soft_double_prep:
+ 	  if (intarg_count == NUM_GPR_ARG_REGISTERS-1)
+ 	    intarg_count++;
+ 	  if (intarg_count >= NUM_GPR_ARG_REGISTERS)
+@@ -316,9 +325,6 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+ 	  break;
+ 
+ 	case FFI_TYPE_STRUCT:
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-	do_struct:
+-#endif
+ 	  struct_copy_size = ((*ptr)->size + 15) & ~0xF;
+ 	  copy_space.c -= struct_copy_size;
+ 	  memcpy (copy_space.c, *p_argv.c, (*ptr)->size);
+@@ -346,7 +352,6 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+ 	case FFI_TYPE_UINT32:
+ 	case FFI_TYPE_SINT32:
+ 	case FFI_TYPE_POINTER:
+-	soft_float_prep:
+ 
+ 	  gprvalue = **p_argv.ui;
+ 
+@@ -363,8 +368,10 @@ ffi_prep_args_SYSV (extended_cif *ecif,
+   /* Check that we didn't overrun the stack...  */
+   FFI_ASSERT (copy_space.c >= next_arg.c);
+   FFI_ASSERT (gpr_base.u <= stacktop.u - ASM_NEEDS_REGISTERS);
++#ifndef __NO_FPRS__
+   FFI_ASSERT (fpr_base.u
+ 	      <= stacktop.u - ASM_NEEDS_REGISTERS - NUM_GPR_ARG_REGISTERS);
++#endif
+   FFI_ASSERT (flags & FLAG_4_GPR_ARGUMENTS || intarg_count <= 4);
+ }
+ 
+@@ -601,9 +608,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+   unsigned type = cif->rtype->type;
+   unsigned size = cif->rtype->size;
+ 
+-  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-    NUM_FPR_ARG_REGISTERS = 0;
+-
+   if (cif->abi != FFI_LINUX64)
+     {
+       /* All the machine-independent calculation of cif->bytes will be wrong.
+@@ -643,25 +647,38 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+      - Single/double FP values in fpr1, long double in fpr1,fpr2.
+      - soft-float float/doubles are treated as UINT32/UINT64 respectivley.
+      - soft-float long doubles are returned in gpr3-gpr6.  */
++  /* First translate for softfloat/nonlinux */
++  if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
++	if (type == FFI_TYPE_FLOAT)
++		type = FFI_TYPE_UINT32;
++	if (type == FFI_TYPE_DOUBLE)
++		type = FFI_TYPE_UINT64;
++	if (type == FFI_TYPE_LONGDOUBLE)
++		type = FFI_TYPE_UINT128;
++  } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++	if (type == FFI_TYPE_LONGDOUBLE)
++		type = FFI_TYPE_STRUCT;
++#endif
++  }
++
+   switch (type)
+     {
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++#ifndef __NO_FPRS__
+     case FFI_TYPE_LONGDOUBLE:
+-      if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64
+-	&& cif->abi != FFI_LINUX_SOFT_FLOAT)
+-	goto byref;
+       flags |= FLAG_RETURNS_128BITS;
+       /* Fall through.  */
+-#endif
+     case FFI_TYPE_DOUBLE:
+       flags |= FLAG_RETURNS_64BITS;
+       /* Fall through.  */
+     case FFI_TYPE_FLOAT:
+-      /* With FFI_LINUX_SOFT_FLOAT no fp registers are used.  */
+-      if (cif->abi != FFI_LINUX_SOFT_FLOAT)
+-	flags |= FLAG_RETURNS_FP;
++      flags |= FLAG_RETURNS_FP;
+       break;
++#endif
+ 
++    case FFI_TYPE_UINT128:
++      flags |= FLAG_RETURNS_128BITS;
++      /* Fall through.  */
+     case FFI_TYPE_UINT64:
+     case FFI_TYPE_SINT64:
+       flags |= FLAG_RETURNS_64BITS;
+@@ -680,10 +697,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+        */
+       if (cif->abi == FFI_SYSV && size <= 8)
+ 	flags |= FLAG_RETURNS_SMST;
+-      
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-    byref:
+-#endif
+       intarg_count++;
+       flags |= FLAG_RETVAL_REFERENCE;
+       /* Fall through.  */
+@@ -704,39 +717,36 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+        Stuff on the stack needs to keep proper alignment.  */
+     for (ptr = cif->arg_types, i = cif->nargs; i > 0; i--, ptr++)
+       {
+-	switch ((*ptr)->type)
+-	  {
++	unsigned short typenum = (*ptr)->type;
++
++	/* We may need to handle some values depending on ABI */
++	if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
++		if (typenum == FFI_TYPE_FLOAT)
++			typenum = FFI_TYPE_UINT32;
++		if (typenum == FFI_TYPE_DOUBLE)
++			typenum = FFI_TYPE_UINT64;
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_UINT128;
++	} else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_STRUCT;
++#endif
++	}
++
++	switch (typenum) {
++#ifndef __NO_FPRS__
+ 	  case FFI_TYPE_FLOAT:
+-	    /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
+-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	      goto soft_float_cif;
+ 	    fparg_count++;
+ 	    /* floating singles are not 8-aligned on stack */
+ 	    break;
+ 
+ #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+ 	  case FFI_TYPE_LONGDOUBLE:
+-	    if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
+-	      goto do_struct;
+-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	      {
+-		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3
+-		  || intarg_count < NUM_GPR_ARG_REGISTERS)
+-		  /* A long double in FFI_LINUX_SOFT_FLOAT can use only
+-		     a set of four consecutive gprs. If we have not enough,
+-		     we have to adjust the intarg_count value.  */
+-		  intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
+-		intarg_count += 4;
+-		break;
+-	      }
+-	    else
+-	      fparg_count++;
++	    fparg_count++;
+ 	    /* Fall thru */
+ #endif
+ 	  case FFI_TYPE_DOUBLE:
+-	    /* With FFI_LINUX_SOFT_FLOAT doubles are handled like UINT64.  */
+-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	      goto soft_double_cif;
+ 	    fparg_count++;
+ 	    /* If this FP arg is going on the stack, it must be
+ 	       8-byte-aligned.  */
+@@ -745,10 +755,21 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+ 		&& intarg_count % 2 != 0)
+ 	      intarg_count++;
+ 	    break;
++#endif
++	  case FFI_TYPE_UINT128:
++		/*
++		 * A long double in FFI_LINUX_SOFT_FLOAT can use only a set
++		 * of four consecutive gprs. If we do not have enough, we
++		 * have to adjust the intarg_count value.
++		 */
++		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3
++				&& intarg_count < NUM_GPR_ARG_REGISTERS)
++			intarg_count = NUM_GPR_ARG_REGISTERS;
++		intarg_count += 4;
++		break;
+ 
+ 	  case FFI_TYPE_UINT64:
+ 	  case FFI_TYPE_SINT64:
+-	  soft_double_cif:
+ 	    /* '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
+@@ -765,9 +786,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+ 	    break;
+ 
+ 	  case FFI_TYPE_STRUCT:
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-	  do_struct:
+-#endif
+ 	    /* We must allocate space for a copy of these to enforce
+ 	       pass-by-value.  Pad the space up to a multiple of 16
+ 	       bytes (the maximum alignment required for anything under
+@@ -775,12 +793,20 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+ 	    struct_copy_size += ((*ptr)->size + 15) & ~0xF;
+ 	    /* Fall through (allocate space for the pointer).  */
+ 
+-	  default:
+-	  soft_float_cif:
++	  case FFI_TYPE_POINTER:
++	  case FFI_TYPE_INT:
++	  case FFI_TYPE_UINT32:
++	  case FFI_TYPE_SINT32:
++	  case FFI_TYPE_UINT16:
++	  case FFI_TYPE_SINT16:
++	  case FFI_TYPE_UINT8:
++	  case FFI_TYPE_SINT8:
+ 	    /* Everything else is passed as a 4-byte word in a GPR, either
+ 	       the object itself or a pointer to it.  */
+ 	    intarg_count++;
+ 	    break;
++	  default:
++		FFI_ASSERT (0);
+ 	  }
+       }
+   else
+@@ -809,16 +835,29 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+ 	    intarg_count += ((*ptr)->size + 7) / 8;
+ 	    break;
+ 
+-	  default:
++	  case FFI_TYPE_POINTER:
++	  case FFI_TYPE_UINT64:
++	  case FFI_TYPE_SINT64:
++	  case FFI_TYPE_INT:
++	  case FFI_TYPE_UINT32:
++	  case FFI_TYPE_SINT32:
++	  case FFI_TYPE_UINT16:
++	  case FFI_TYPE_SINT16:
++	  case FFI_TYPE_UINT8:
++	  case FFI_TYPE_SINT8:
+ 	    /* Everything else is passed as a 8-byte word in a GPR, either
+ 	       the object itself or a pointer to it.  */
+ 	    intarg_count++;
+ 	    break;
++	  default:
++		FFI_ASSERT (0);
+ 	  }
+       }
+ 
++#ifndef __NO_FPRS__
+   if (fparg_count != 0)
+     flags |= FLAG_FP_ARGUMENTS;
++#endif
+   if (intarg_count > 4)
+     flags |= FLAG_4_GPR_ARGUMENTS;
+   if (struct_copy_size != 0)
+@@ -826,21 +865,27 @@ ffi_prep_cif_machdep (ffi_cif *cif)
+ 
+   if (cif->abi != FFI_LINUX64)
+     {
++#ifndef __NO_FPRS__
+       /* Space for the FPR registers, if needed.  */
+       if (fparg_count != 0)
+ 	bytes += NUM_FPR_ARG_REGISTERS * sizeof (double);
++#endif
+ 
+       /* Stack space.  */
+       if (intarg_count > NUM_GPR_ARG_REGISTERS)
+ 	bytes += (intarg_count - NUM_GPR_ARG_REGISTERS) * sizeof (int);
++#ifndef __NO_FPRS__
+       if (fparg_count > NUM_FPR_ARG_REGISTERS)
+ 	bytes += (fparg_count - NUM_FPR_ARG_REGISTERS) * sizeof (double);
++#endif
+     }
+   else
+     {
++#ifndef __NO_FPRS__
+       /* Space for the FPR registers, if needed.  */
+       if (fparg_count != 0)
+ 	bytes += NUM_FPR_ARG_REGISTERS64 * sizeof (double);
++#endif
+ 
+       /* Stack space.  */
+       if (intarg_count > NUM_GPR_ARG_REGISTERS64)
+@@ -898,9 +943,11 @@ ffi_call(ffi_cif *cif, void (*fn)(void),
+   switch (cif->abi)
+     {
+ #ifndef POWERPC64
++# ifndef __NO_FPRS__
+     case FFI_SYSV:
+     case FFI_GCC_SYSV:
+     case FFI_LINUX:
++# endif
+     case FFI_LINUX_SOFT_FLOAT:
+       ffi_call_SYSV (&ecif, -cif->bytes, cif->flags, ecif.rvalue, fn);
+       break;
+@@ -1013,32 +1060,38 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+   void **          avalue;
+   ffi_type **      arg_types;
+   long             i, avn;
+-  long             nf;   /* number of floating registers already used */
+-  long             ng;   /* number of general registers already used */
+-  ffi_cif *        cif;
+-  double           temp;
+-  unsigned         size;
++#ifndef __NO_FPRS__
++  long             nf = 0;   /* number of floating registers already used */
++#endif
++  long             ng = 0;   /* number of general registers already used */
++
++  ffi_cif *cif = closure->cif;
++  unsigned       size     = cif->rtype->size;
++  unsigned short rtypenum = cif->rtype->type;
+ 
+-  cif = closure->cif;
+   avalue = alloca (cif->nargs * sizeof (void *));
+-  size = cif->rtype->size;
+ 
+-  nf = 0;
+-  ng = 0;
++  /* First translate for softfloat/nonlinux */
++  if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
++	if (rtypenum == FFI_TYPE_FLOAT)
++		rtypenum = FFI_TYPE_UINT32;
++	if (rtypenum == FFI_TYPE_DOUBLE)
++		rtypenum = FFI_TYPE_UINT64;
++	if (rtypenum == FFI_TYPE_LONGDOUBLE)
++		rtypenum = FFI_TYPE_UINT128;
++  } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++	if (rtypenum == FFI_TYPE_LONGDOUBLE)
++		rtypenum = FFI_TYPE_STRUCT;
++#endif
++  }
++
+ 
+   /* Copy the caller's structure return value address so that the closure
+      returns the data directly to the caller.
+      For FFI_SYSV the result is passed in r3/r4 if the struct size is less
+      or equal 8 bytes.  */
+-
+-  if ((cif->rtype->type == FFI_TYPE_STRUCT
+-       && !((cif->abi == FFI_SYSV) && (size <= 8)))
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-      || (cif->rtype->type == FFI_TYPE_LONGDOUBLE
+-	  && cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
+-#endif
+-      )
+-    {
++  if (rtypenum == FFI_TYPE_STRUCT && ((cif->abi != FFI_SYSV) || (size > 8))) {
+       rvalue = (void *) *pgr;
+       ng++;
+       pgr++;
+@@ -1049,10 +1102,109 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+   arg_types = cif->arg_types;
+ 
+   /* Grab the addresses of the arguments from the stack frame.  */
+-  while (i < avn)
+-    {
+-      switch (arg_types[i]->type)
+-	{
++  while (i < avn) {
++      unsigned short typenum = arg_types[i]->type;
++
++      /* We may need to handle some values depending on ABI */
++      if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
++		if (typenum == FFI_TYPE_FLOAT)
++			typenum = FFI_TYPE_UINT32;
++		if (typenum == FFI_TYPE_DOUBLE)
++			typenum = FFI_TYPE_UINT64;
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_UINT128;
++      } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++		if (typenum == FFI_TYPE_LONGDOUBLE)
++			typenum = FFI_TYPE_STRUCT;
++#endif
++      }
++
++      switch (typenum) {
++#ifndef __NO_FPRS__
++	case FFI_TYPE_FLOAT:
++	  /* unfortunately float values are stored as doubles
++	   * in the ffi_closure_SYSV code (since we don't check
++	   * the type in that routine).
++	   */
++
++	  /* there are 8 64bit floating point registers */
++
++	  if (nf < 8)
++	    {
++	      temp = pfr->d;
++	      pfr->f = (float) temp;
++	      avalue[i] = pfr;
++	      nf++;
++	      pfr++;
++	    }
++	  else
++	    {
++	      /* FIXME? here we are really changing the values
++	       * stored in the original calling routines outgoing
++	       * parameter stack.  This is probably a really
++	       * naughty thing to do but...
++	       */
++	      avalue[i] = pst;
++	      pst += 1;
++	    }
++	  break;
++
++	case FFI_TYPE_DOUBLE:
++	  /* On the outgoing stack all values are aligned to 8 */
++	  /* there are 8 64bit floating point registers */
++
++	  if (nf < 8)
++	    {
++	      avalue[i] = pfr;
++	      nf++;
++	      pfr++;
++	    }
++	  else
++	    {
++	      if (((long) pst) & 4)
++		pst++;
++	      avalue[i] = pst;
++	      pst += 2;
++	    }
++	  break;
++
++#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
++	case FFI_TYPE_LONGDOUBLE:
++	  if (nf < 7)
++	    {
++	      avalue[i] = pfr;
++	      pfr += 2;
++	      nf += 2;
++	    }
++	  else
++	    {
++	      if (((long) pst) & 4)
++		pst++;
++	      avalue[i] = pst;
++	      pst += 4;
++	      nf = 8;
++	    }
++	  break;
++#endif
++#endif /* have FPRS */
++
++	case FFI_TYPE_UINT128:
++		/*
++		 * Test if for the whole long double, 4 gprs are available.
++		 * otherwise the stuff ends up on the stack.
++		 */
++		if (ng < 5) {
++			avalue[i] = pgr;
++			pgr += 4;
++			ng += 4;
++		} else {
++			avalue[i] = pst;
++			pst += 4;
++			ng = 8+4;
++		}
++		break;
++
+ 	case FFI_TYPE_SINT8:
+ 	case FFI_TYPE_UINT8:
+ 	  /* there are 8 gpr registers used to pass values */
+@@ -1088,7 +1240,6 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+ 	case FFI_TYPE_SINT32:
+ 	case FFI_TYPE_UINT32:
+ 	case FFI_TYPE_POINTER:
+-	soft_float_closure:
+ 	  /* there are 8 gpr registers used to pass values */
+ 	  if (ng < 8)
+ 	    {
+@@ -1104,9 +1255,6 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+ 	  break;
+ 
+ 	case FFI_TYPE_STRUCT:
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-	do_struct:
+-#endif
+ 	  /* Structs are passed by reference. The address will appear in a
+ 	     gpr if it is one of the first 8 arguments.  */
+ 	  if (ng < 8)
+@@ -1124,7 +1272,6 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+ 
+ 	case FFI_TYPE_SINT64:
+ 	case FFI_TYPE_UINT64:
+-	soft_double_closure:
+ 	  /* passing long long ints are complex, they must
+ 	   * be passed in suitable register pairs such as
+ 	   * (r3,r4) or (r5,r6) or (r6,r7), or (r7,r8) or (r9,r10)
+@@ -1156,99 +1303,8 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+ 	    }
+ 	  break;
+ 
+-	case FFI_TYPE_FLOAT:
+-	  /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
+-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	    goto soft_float_closure;
+-	  /* unfortunately float values are stored as doubles
+-	   * in the ffi_closure_SYSV code (since we don't check
+-	   * the type in that routine).
+-	   */
+-
+-	  /* there are 8 64bit floating point registers */
+-
+-	  if (nf < 8)
+-	    {
+-	      temp = pfr->d;
+-	      pfr->f = (float) temp;
+-	      avalue[i] = pfr;
+-	      nf++;
+-	      pfr++;
+-	    }
+-	  else
+-	    {
+-	      /* FIXME? here we are really changing the values
+-	       * stored in the original calling routines outgoing
+-	       * parameter stack.  This is probably a really
+-	       * naughty thing to do but...
+-	       */
+-	      avalue[i] = pst;
+-	      pst += 1;
+-	    }
+-	  break;
+-
+-	case FFI_TYPE_DOUBLE:
+-	  /* With FFI_LINUX_SOFT_FLOAT doubles are handled like UINT64.  */
+-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	    goto soft_double_closure;
+-	  /* On the outgoing stack all values are aligned to 8 */
+-	  /* there are 8 64bit floating point registers */
+-
+-	  if (nf < 8)
+-	    {
+-	      avalue[i] = pfr;
+-	      nf++;
+-	      pfr++;
+-	    }
+-	  else
+-	    {
+-	      if (((long) pst) & 4)
+-		pst++;
+-	      avalue[i] = pst;
+-	      pst += 2;
+-	    }
+-	  break;
+-
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-	case FFI_TYPE_LONGDOUBLE:
+-	  if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
+-	    goto do_struct;
+-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-	    { /* Test if for the whole long double, 4 gprs are available.
+-		 otherwise the stuff ends up on the stack.  */
+-	      if (ng < 5)
+-		{
+-		  avalue[i] = pgr;
+-		  pgr += 4;
+-		  ng += 4;
+-		}
+-	      else
+-		{
+-		  avalue[i] = pst;
+-		  pst += 4;
+-		  ng = 8;
+-		}
+-	      break;
+-	    }
+-	  if (nf < 7)
+-	    {
+-	      avalue[i] = pfr;
+-	      pfr += 2;
+-	      nf += 2;
+-	    }
+-	  else
+-	    {
+-	      if (((long) pst) & 4)
+-		pst++;
+-	      avalue[i] = pst;
+-	      pst += 4;
+-	      nf = 8;
+-	    }
+-	  break;
+-#endif
+-
+ 	default:
+-	  FFI_ASSERT (0);
++		FFI_ASSERT (0);
+ 	}
+ 
+       i++;
+@@ -1265,39 +1321,9 @@ ffi_closure_helper_SYSV (ffi_closure *cl
+      already used and we never have a struct with size zero. That is the reason
+      for the subtraction of 1. See the comment in ffitarget.h about ordering.
+   */
+-  if (cif->abi == FFI_SYSV && cif->rtype->type == FFI_TYPE_STRUCT
+-      && size <= 8)
++  if (cif->abi == FFI_SYSV && rtypenum == FFI_TYPE_STRUCT && size <= 8)
+     return (FFI_SYSV_TYPE_SMALL_STRUCT - 1) + size;
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-  else if (cif->rtype->type == FFI_TYPE_LONGDOUBLE
+-	   && cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
+-    return FFI_TYPE_STRUCT;
+-#endif
+-  /* With FFI_LINUX_SOFT_FLOAT floats and doubles are handled like UINT32
+-     respectivley UINT64.  */
+-  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
+-    {
+-      switch (cif->rtype->type)
+-	{
+-	case FFI_TYPE_FLOAT:
+-	  return FFI_TYPE_UINT32;
+-	  break;
+-	case FFI_TYPE_DOUBLE:
+-	  return FFI_TYPE_UINT64;
+-	  break;
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-	case FFI_TYPE_LONGDOUBLE:
+-	  return FFI_TYPE_UINT128;
+-	  break;
+-#endif
+-	default:
+-	  return cif->rtype->type;
+-	}
+-    }
+-  else
+-    {
+-      return cif->rtype->type;
+-    }
++  return rtypenum;
+ }
+ 
+ int FFI_HIDDEN ffi_closure_helper_LINUX64 (ffi_closure *, void *,
+Index: libffi/src/powerpc/ffitarget.h
+===================================================================
+--- libffi.orig/src/powerpc/ffitarget.h
++++ libffi/src/powerpc/ffitarget.h
+@@ -60,18 +60,14 @@ typedef enum ffi_abi {
+   FFI_LINUX64,
+   FFI_LINUX,
+   FFI_LINUX_SOFT_FLOAT,
+-# ifdef POWERPC64
++# if defined(POWERPC64)
+   FFI_DEFAULT_ABI = FFI_LINUX64,
+-# else
+-#  if (!defined(__NO_FPRS__) && (__LDBL_MANT_DIG__ == 106))
+-  FFI_DEFAULT_ABI = FFI_LINUX,
+-#  else
+-#   ifdef __NO_FPRS__
++# elif defined(__NO_FPRS__)
+   FFI_DEFAULT_ABI = FFI_LINUX_SOFT_FLOAT,
+-#   else
++# elif (__LDBL_MANT_DIG__ == 106)
++  FFI_DEFAULT_ABI = FFI_LINUX,
++# else
+   FFI_DEFAULT_ABI = FFI_GCC_SYSV,
+-#   endif
+-#  endif
+ # endif
+ #endif
+ 
+Index: libffi/src/powerpc/ppc_closure.S
+===================================================================
+--- libffi.orig/src/powerpc/ppc_closure.S
++++ libffi/src/powerpc/ppc_closure.S
+@@ -122,22 +122,41 @@ ENTRY(ffi_closure_SYSV)
+ 	blr
+ 
+ # case FFI_TYPE_FLOAT
++#ifndef __NO_FPRS__
+ 	lfs %f1,112+0(%r1)
+ 	mtlr %r0
+ 	addi %r1,%r1,144
++#else
++	nop
++	nop
++	nop
++#endif
+ 	blr
+ 
+ # case FFI_TYPE_DOUBLE
++#ifndef __NO_FPRS__
+ 	lfd %f1,112+0(%r1)
+ 	mtlr %r0
+ 	addi %r1,%r1,144
++#else
++	nop
++	nop
++	nop
++#endif
+ 	blr
+ 
+ # case FFI_TYPE_LONGDOUBLE
++#ifndef __NO_FPRS__
+ 	lfd %f1,112+0(%r1)
+ 	lfd %f2,112+8(%r1)
+ 	mtlr %r0
+ 	b .Lfinish
++#else
++	nop
++	nop
++	nop
++	blr
++#endif
+ 
+ # case FFI_TYPE_UINT8
+ 	lbz %r3,112+3(%r1)
+Index: libffi/src/powerpc/sysv.S
+===================================================================
+--- libffi.orig/src/powerpc/sysv.S
++++ libffi/src/powerpc/sysv.S
+@@ -83,6 +83,7 @@ ENTRY(ffi_call_SYSV)
+ 	nop
+ 1:
+ 
++#ifndef __NO_FPRS__
+ 	/* Load all the FP registers.  */
+ 	bf-	6,2f
+ 	lfd	%f1,-16-(8*4)-(8*8)(%r28)
+@@ -94,6 +95,7 @@ ENTRY(ffi_call_SYSV)
+ 	lfd	%f6,-16-(8*4)-(3*8)(%r28)
+ 	lfd	%f7,-16-(8*4)-(2*8)(%r28)
+ 	lfd	%f8,-16-(8*4)-(1*8)(%r28)
++#endif
+ 2:
+ 
+ 	/* Make the call.  */
+@@ -103,7 +105,9 @@ ENTRY(ffi_call_SYSV)
+ 	mtcrf	0x01,%r31 /* cr7  */
+ 	bt-	31,L(small_struct_return_value)
+ 	bt-	30,L(done_return_value)
++#ifndef __NO_FPRS__
+ 	bt-	29,L(fp_return_value)
++#endif
+ 	stw	%r3,0(%r30)
+ 	bf+	28,L(done_return_value)
+ 	stw	%r4,4(%r30)
+@@ -124,6 +128,7 @@ L(done_return_value):
+ 	lwz	%r1,0(%r1)
+ 	blr
+ 
++#ifndef __NO_FPRS__
+ L(fp_return_value):
+ 	bf	28,L(float_return_value)
+ 	stfd	%f1,0(%r30)
+@@ -134,6 +139,7 @@ L(fp_return_value):
+ L(float_return_value):
+ 	stfs	%f1,0(%r30)
+ 	b	L(done_return_value)
++#endif
+ 
+ L(small_struct_return_value):
+ 	/*

patches/series

@@ -35,3 +35,4 @@ minix
 
 interix-patch
 remove-debug-code
+powerpc-ffi-softfloat

src/powerpc/ffi.c

@@ -41,27 +41,28 @@ enum {
   /* The assembly depends on these exact flags.  */
   FLAG_RETURNS_SMST	= 1 << (31-31), /* Used for FFI_SYSV small structs.  */
   FLAG_RETURNS_NOTHING  = 1 << (31-30), /* These go in cr7 */
+#ifndef __NO_FPRS__
   FLAG_RETURNS_FP       = 1 << (31-29),
+#endif
   FLAG_RETURNS_64BITS   = 1 << (31-28),
 
   FLAG_RETURNS_128BITS  = 1 << (31-27), /* cr6  */
 
   FLAG_ARG_NEEDS_COPY   = 1 << (31- 7),
+#ifndef __NO_FPRS__
   FLAG_FP_ARGUMENTS     = 1 << (31- 6), /* cr1.eq; specified by ABI */
+#endif
   FLAG_4_GPR_ARGUMENTS  = 1 << (31- 5),
   FLAG_RETVAL_REFERENCE = 1 << (31- 4)
 };
 
 /* About the SYSV ABI.  */
-unsigned int NUM_GPR_ARG_REGISTERS = 8;
+#define ASM_NEEDS_REGISTERS 4
+#define NUM_GPR_ARG_REGISTERS 8
 #ifndef __NO_FPRS__
-unsigned int NUM_FPR_ARG_REGISTERS = 8;
+# define NUM_FPR_ARG_REGISTERS 8
-#else
-unsigned int NUM_FPR_ARG_REGISTERS = 0;
 #endif
 
-enum { ASM_NEEDS_REGISTERS = 4 };
-
 /* ffi_prep_args_SYSV is called by the assembly routine once stack space
    has been allocated for the function's arguments.
 
@@ -110,10 +111,12 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
   valp gpr_base;
   int intarg_count;
 
+#ifndef __NO_FPRS__
   /* 'fpr_base' points at the space for fpr1, and grows upwards as
      we use FPR registers.  */
   valp fpr_base;
   int fparg_count;
+#endif
 
   /* 'copy_space' grows down as we put structures in it.  It should
      stay 16-byte aligned.  */
@@ -122,9 +125,8 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
   /* 'next_arg' grows up as we put parameters in it.  */
   valp next_arg;
 
-  int i, ii MAYBE_UNUSED;
+  int i;
   ffi_type **ptr;
-  double double_tmp;
   union {
     void **v;
     char **c;
@@ -140,15 +142,16 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
   size_t struct_copy_size;
   unsigned gprvalue;
 
-  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
-    NUM_FPR_ARG_REGISTERS = 0;
-
   stacktop.c = (char *) stack + bytes;
   gpr_base.u = stacktop.u - ASM_NEEDS_REGISTERS - NUM_GPR_ARG_REGISTERS;
   intarg_count = 0;
+#ifndef __NO_FPRS__
   fpr_base.d = gpr_base.d - NUM_FPR_ARG_REGISTERS;
   fparg_count = 0;
   copy_space.c = ((flags & FLAG_FP_ARGUMENTS) ? fpr_base.c : gpr_base.c);
+#else
+  copy_space.c = gpr_base.c;
+#endif
   next_arg.u = stack + 2;
 
   /* Check that everything starts aligned properly.  */
@@ -171,12 +174,28 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
        i > 0;
        i--, ptr++, p_argv.v++)
     {
-      switch ((*ptr)->type)
+      unsigned short typenum = (*ptr)->type;
-	{
+
+      /* We may need to handle some values depending on ABI */
+      if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT) {
+		if (typenum == FFI_TYPE_FLOAT)
+			typenum = FFI_TYPE_UINT32;
+		if (typenum == FFI_TYPE_DOUBLE)
+			typenum = FFI_TYPE_UINT64;
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_UINT128;
+      } else if (ecif->cif->abi != FFI_LINUX) {
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_STRUCT;
+#endif
+      }
+
+      /* Now test the translated value */
+      switch (typenum) {
+#ifndef __NO_FPRS__
 	case FFI_TYPE_FLOAT:
 	  /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
-	  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
-	    goto soft_float_prep;
 	  double_tmp = **p_argv.f;
 	  if (fparg_count >= NUM_FPR_ARG_REGISTERS)
 	    {
@@ -215,43 +234,6 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
 
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 	case FFI_TYPE_LONGDOUBLE:
-	  if ((ecif->cif->abi != FFI_LINUX)
-		&& (ecif->cif->abi != FFI_LINUX_SOFT_FLOAT))
-	    goto do_struct;
-	  /* The soft float ABI for long doubles works like this,
-	     a long double is passed in four consecutive gprs if available.
-	     A maximum of 2 long doubles can be passed in gprs.
-	     If we do not have 4 gprs left, the long double is passed on the
-	     stack, 4-byte aligned.  */
-	  if (ecif->cif->abi == FFI_LINUX_SOFT_FLOAT)
-	    {
-	      unsigned int int_tmp = (*p_argv.ui)[0];
-	      if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3)
-		{
-		  if (intarg_count < NUM_GPR_ARG_REGISTERS)
-		    intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
-		  *next_arg.u = int_tmp;
-		  next_arg.u++;
-		  for (ii = 1; ii < 4; ii++)
-		    {
-		      int_tmp = (*p_argv.ui)[ii];
-		      *next_arg.u = int_tmp;
-		      next_arg.u++;
-		    }
-		}
-	      else
-		{
-		  *gpr_base.u++ = int_tmp;
-		  for (ii = 1; ii < 4; ii++)
-		    {
-		      int_tmp = (*p_argv.ui)[ii];
-		      *gpr_base.u++ = int_tmp;
-		    }
-		}
-	      intarg_count +=4;
-	    }
-	  else
-	    {
 	      double_tmp = (*p_argv.d)[0];
 
 	      if (fparg_count >= NUM_FPR_ARG_REGISTERS - 1)
@@ -277,13 +259,40 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
 
 	      fparg_count += 2;
 	      FFI_ASSERT (flags & FLAG_FP_ARGUMENTS);
-	    }
 	  break;
 #endif
+#endif /* have FPRs */
+
+	/*
+	 * The soft float ABI for long doubles works like this, a long double
+	 * is passed in four consecutive GPRs if available.  A maximum of 2
+	 * long doubles can be passed in gprs.  If we do not have 4 GPRs
+	 * left, the long double is passed on the stack, 4-byte aligned.
+	 */
+	case FFI_TYPE_UINT128: {
+		unsigned int int_tmp = (*p_argv.ui)[0];
+		unsigned int ii;
+		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3) {
+			if (intarg_count < NUM_GPR_ARG_REGISTERS)
+				intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
+			*(next_arg.u++) = int_tmp;
+			for (ii = 1; ii < 4; ii++) {
+				int_tmp = (*p_argv.ui)[ii];
+				*(next_arg.u++) = int_tmp;
+			}
+		} else {
+			*(gpr_base.u++) = int_tmp;
+			for (ii = 1; ii < 4; ii++) {
+				int_tmp = (*p_argv.ui)[ii];
+				*(gpr_base.u++) = int_tmp;
+			}
+		}
+		intarg_count += 4;
+		break;
+	}
 
 	case FFI_TYPE_UINT64:
 	case FFI_TYPE_SINT64:
-	soft_double_prep:
 	  if (intarg_count == NUM_GPR_ARG_REGISTERS-1)
 	    intarg_count++;
 	  if (intarg_count >= NUM_GPR_ARG_REGISTERS)
@@ -316,9 +325,6 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
 	  break;
 
 	case FFI_TYPE_STRUCT:
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-	do_struct:
-#endif
 	  struct_copy_size = ((*ptr)->size + 15) & ~0xF;
 	  copy_space.c -= struct_copy_size;
 	  memcpy (copy_space.c, *p_argv.c, (*ptr)->size);
@@ -346,7 +352,6 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
 	case FFI_TYPE_UINT32:
 	case FFI_TYPE_SINT32:
 	case FFI_TYPE_POINTER:
-	soft_float_prep:
 
 	  gprvalue = **p_argv.ui;
 
@@ -363,8 +368,10 @@ ffi_prep_args_SYSV (extended_cif *ecif, unsigned *const stack)
   /* Check that we didn't overrun the stack...  */
   FFI_ASSERT (copy_space.c >= next_arg.c);
   FFI_ASSERT (gpr_base.u <= stacktop.u - ASM_NEEDS_REGISTERS);
+#ifndef __NO_FPRS__
   FFI_ASSERT (fpr_base.u
 	      <= stacktop.u - ASM_NEEDS_REGISTERS - NUM_GPR_ARG_REGISTERS);
+#endif
   FFI_ASSERT (flags & FLAG_4_GPR_ARGUMENTS || intarg_count <= 4);
 }
 
@@ -601,9 +608,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
   unsigned type = cif->rtype->type;
   unsigned size = cif->rtype->size;
 
-  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-    NUM_FPR_ARG_REGISTERS = 0;
-
   if (cif->abi != FFI_LINUX64)
     {
       /* All the machine-independent calculation of cif->bytes will be wrong.
@@ -643,25 +647,38 @@ ffi_prep_cif_machdep (ffi_cif *cif)
      - Single/double FP values in fpr1, long double in fpr1,fpr2.
      - soft-float float/doubles are treated as UINT32/UINT64 respectivley.
      - soft-float long doubles are returned in gpr3-gpr6.  */
+  /* First translate for softfloat/nonlinux */
+  if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
+	if (type == FFI_TYPE_FLOAT)
+		type = FFI_TYPE_UINT32;
+	if (type == FFI_TYPE_DOUBLE)
+		type = FFI_TYPE_UINT64;
+	if (type == FFI_TYPE_LONGDOUBLE)
+		type = FFI_TYPE_UINT128;
+  } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+	if (type == FFI_TYPE_LONGDOUBLE)
+		type = FFI_TYPE_STRUCT;
+#endif
+  }
+
   switch (type)
     {
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+#ifndef __NO_FPRS__
     case FFI_TYPE_LONGDOUBLE:
-      if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64
-	&& cif->abi != FFI_LINUX_SOFT_FLOAT)
-	goto byref;
       flags |= FLAG_RETURNS_128BITS;
       /* Fall through.  */
-#endif
     case FFI_TYPE_DOUBLE:
       flags |= FLAG_RETURNS_64BITS;
       /* Fall through.  */
     case FFI_TYPE_FLOAT:
-      /* With FFI_LINUX_SOFT_FLOAT no fp registers are used.  */
+      flags |= FLAG_RETURNS_FP;
-      if (cif->abi != FFI_LINUX_SOFT_FLOAT)
-	flags |= FLAG_RETURNS_FP;
       break;
+#endif
 
+    case FFI_TYPE_UINT128:
+      flags |= FLAG_RETURNS_128BITS;
+      /* Fall through.  */
     case FFI_TYPE_UINT64:
     case FFI_TYPE_SINT64:
       flags |= FLAG_RETURNS_64BITS;
@@ -680,10 +697,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
        */
       if (cif->abi == FFI_SYSV && size <= 8)
 	flags |= FLAG_RETURNS_SMST;
-      
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-    byref:
-#endif
       intarg_count++;
       flags |= FLAG_RETVAL_REFERENCE;
       /* Fall through.  */
@@ -704,39 +717,36 @@ ffi_prep_cif_machdep (ffi_cif *cif)
        Stuff on the stack needs to keep proper alignment.  */
     for (ptr = cif->arg_types, i = cif->nargs; i > 0; i--, ptr++)
       {
-	switch ((*ptr)->type)
+	unsigned short typenum = (*ptr)->type;
-	  {
+
+	/* We may need to handle some values depending on ABI */
+	if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
+		if (typenum == FFI_TYPE_FLOAT)
+			typenum = FFI_TYPE_UINT32;
+		if (typenum == FFI_TYPE_DOUBLE)
+			typenum = FFI_TYPE_UINT64;
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_UINT128;
+	} else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_STRUCT;
+#endif
+	}
+
+	switch (typenum) {
+#ifndef __NO_FPRS__
 	  case FFI_TYPE_FLOAT:
-	    /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	      goto soft_float_cif;
 	    fparg_count++;
 	    /* floating singles are not 8-aligned on stack */
 	    break;
 
 #if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
 	  case FFI_TYPE_LONGDOUBLE:
-	    if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
+	    fparg_count++;
-	      goto do_struct;
-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	      {
-		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3
-		  || intarg_count < NUM_GPR_ARG_REGISTERS)
-		  /* A long double in FFI_LINUX_SOFT_FLOAT can use only
-		     a set of four consecutive gprs. If we have not enough,
-		     we have to adjust the intarg_count value.  */
-		  intarg_count += NUM_GPR_ARG_REGISTERS - intarg_count;
-		intarg_count += 4;
-		break;
-	      }
-	    else
-	      fparg_count++;
 	    /* Fall thru */
 #endif
 	  case FFI_TYPE_DOUBLE:
-	    /* With FFI_LINUX_SOFT_FLOAT doubles are handled like UINT64.  */
-	    if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	      goto soft_double_cif;
 	    fparg_count++;
 	    /* If this FP arg is going on the stack, it must be
 	       8-byte-aligned.  */
@@ -745,10 +755,21 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 		&& intarg_count % 2 != 0)
 	      intarg_count++;
 	    break;
+#endif
+	  case FFI_TYPE_UINT128:
+		/*
+		 * A long double in FFI_LINUX_SOFT_FLOAT can use only a set
+		 * of four consecutive gprs. If we do not have enough, we
+		 * have to adjust the intarg_count value.
+		 */
+		if (intarg_count >= NUM_GPR_ARG_REGISTERS - 3
+				&& intarg_count < NUM_GPR_ARG_REGISTERS)
+			intarg_count = NUM_GPR_ARG_REGISTERS;
+		intarg_count += 4;
+		break;
 
 	  case FFI_TYPE_UINT64:
 	  case FFI_TYPE_SINT64:
-	  soft_double_cif:
 	    /* '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
@@ -765,9 +786,6 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 	    break;
 
 	  case FFI_TYPE_STRUCT:
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-	  do_struct:
-#endif
 	    /* We must allocate space for a copy of these to enforce
 	       pass-by-value.  Pad the space up to a multiple of 16
 	       bytes (the maximum alignment required for anything under
@@ -775,12 +793,20 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 	    struct_copy_size += ((*ptr)->size + 15) & ~0xF;
 	    /* Fall through (allocate space for the pointer).  */
 
-	  default:
+	  case FFI_TYPE_POINTER:
-	  soft_float_cif:
+	  case FFI_TYPE_INT:
+	  case FFI_TYPE_UINT32:
+	  case FFI_TYPE_SINT32:
+	  case FFI_TYPE_UINT16:
+	  case FFI_TYPE_SINT16:
+	  case FFI_TYPE_UINT8:
+	  case FFI_TYPE_SINT8:
 	    /* Everything else is passed as a 4-byte word in a GPR, either
 	       the object itself or a pointer to it.  */
 	    intarg_count++;
 	    break;
+	  default:
+		FFI_ASSERT (0);
 	  }
       }
   else
@@ -809,16 +835,29 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 	    intarg_count += ((*ptr)->size + 7) / 8;
 	    break;
 
-	  default:
+	  case FFI_TYPE_POINTER:
+	  case FFI_TYPE_UINT64:
+	  case FFI_TYPE_SINT64:
+	  case FFI_TYPE_INT:
+	  case FFI_TYPE_UINT32:
+	  case FFI_TYPE_SINT32:
+	  case FFI_TYPE_UINT16:
+	  case FFI_TYPE_SINT16:
+	  case FFI_TYPE_UINT8:
+	  case FFI_TYPE_SINT8:
 	    /* Everything else is passed as a 8-byte word in a GPR, either
 	       the object itself or a pointer to it.  */
 	    intarg_count++;
 	    break;
+	  default:
+		FFI_ASSERT (0);
 	  }
       }
 
+#ifndef __NO_FPRS__
   if (fparg_count != 0)
     flags |= FLAG_FP_ARGUMENTS;
+#endif
   if (intarg_count > 4)
     flags |= FLAG_4_GPR_ARGUMENTS;
   if (struct_copy_size != 0)
@@ -826,21 +865,27 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 
   if (cif->abi != FFI_LINUX64)
     {
+#ifndef __NO_FPRS__
       /* Space for the FPR registers, if needed.  */
       if (fparg_count != 0)
 	bytes += NUM_FPR_ARG_REGISTERS * sizeof (double);
+#endif
 
       /* Stack space.  */
       if (intarg_count > NUM_GPR_ARG_REGISTERS)
 	bytes += (intarg_count - NUM_GPR_ARG_REGISTERS) * sizeof (int);
+#ifndef __NO_FPRS__
       if (fparg_count > NUM_FPR_ARG_REGISTERS)
 	bytes += (fparg_count - NUM_FPR_ARG_REGISTERS) * sizeof (double);
+#endif
     }
   else
     {
+#ifndef __NO_FPRS__
       /* Space for the FPR registers, if needed.  */
       if (fparg_count != 0)
 	bytes += NUM_FPR_ARG_REGISTERS64 * sizeof (double);
+#endif
 
       /* Stack space.  */
       if (intarg_count > NUM_GPR_ARG_REGISTERS64)
@@ -898,9 +943,11 @@ ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
   switch (cif->abi)
     {
 #ifndef POWERPC64
+# ifndef __NO_FPRS__
     case FFI_SYSV:
     case FFI_GCC_SYSV:
     case FFI_LINUX:
+# endif
     case FFI_LINUX_SOFT_FLOAT:
       ffi_call_SYSV (&ecif, -cif->bytes, cif->flags, ecif.rvalue, fn);
       break;
@@ -1013,32 +1060,38 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
   void **          avalue;
   ffi_type **      arg_types;
   long             i, avn;
-  long             nf;   /* number of floating registers already used */
+#ifndef __NO_FPRS__
-  long             ng;   /* number of general registers already used */
+  long             nf = 0;   /* number of floating registers already used */
-  ffi_cif *        cif;
+#endif
-  double           temp;
+  long             ng = 0;   /* number of general registers already used */
-  unsigned         size;
+
+  ffi_cif *cif = closure->cif;
+  unsigned       size     = cif->rtype->size;
+  unsigned short rtypenum = cif->rtype->type;
 
-  cif = closure->cif;
   avalue = alloca (cif->nargs * sizeof (void *));
-  size = cif->rtype->size;
 
-  nf = 0;
+  /* First translate for softfloat/nonlinux */
-  ng = 0;
+  if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
+	if (rtypenum == FFI_TYPE_FLOAT)
+		rtypenum = FFI_TYPE_UINT32;
+	if (rtypenum == FFI_TYPE_DOUBLE)
+		rtypenum = FFI_TYPE_UINT64;
+	if (rtypenum == FFI_TYPE_LONGDOUBLE)
+		rtypenum = FFI_TYPE_UINT128;
+  } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+	if (rtypenum == FFI_TYPE_LONGDOUBLE)
+		rtypenum = FFI_TYPE_STRUCT;
+#endif
+  }
+
 
   /* Copy the caller's structure return value address so that the closure
      returns the data directly to the caller.
      For FFI_SYSV the result is passed in r3/r4 if the struct size is less
      or equal 8 bytes.  */
-
+  if (rtypenum == FFI_TYPE_STRUCT && ((cif->abi != FFI_SYSV) || (size > 8))) {
-  if ((cif->rtype->type == FFI_TYPE_STRUCT
-       && !((cif->abi == FFI_SYSV) && (size <= 8)))
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-      || (cif->rtype->type == FFI_TYPE_LONGDOUBLE
-	  && cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
-#endif
-      )
-    {
       rvalue = (void *) *pgr;
       ng++;
       pgr++;
@@ -1049,10 +1102,109 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
   arg_types = cif->arg_types;
 
   /* Grab the addresses of the arguments from the stack frame.  */
-  while (i < avn)
+  while (i < avn) {
-    {
+      unsigned short typenum = arg_types[i]->type;
-      switch (arg_types[i]->type)
+
-	{
+      /* We may need to handle some values depending on ABI */
+      if (cif->abi == FFI_LINUX_SOFT_FLOAT) {
+		if (typenum == FFI_TYPE_FLOAT)
+			typenum = FFI_TYPE_UINT32;
+		if (typenum == FFI_TYPE_DOUBLE)
+			typenum = FFI_TYPE_UINT64;
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_UINT128;
+      } else if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX64) {
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+		if (typenum == FFI_TYPE_LONGDOUBLE)
+			typenum = FFI_TYPE_STRUCT;
+#endif
+      }
+
+      switch (typenum) {
+#ifndef __NO_FPRS__
+	case FFI_TYPE_FLOAT:
+	  /* unfortunately float values are stored as doubles
+	   * in the ffi_closure_SYSV code (since we don't check
+	   * the type in that routine).
+	   */
+
+	  /* there are 8 64bit floating point registers */
+
+	  if (nf < 8)
+	    {
+	      temp = pfr->d;
+	      pfr->f = (float) temp;
+	      avalue[i] = pfr;
+	      nf++;
+	      pfr++;
+	    }
+	  else
+	    {
+	      /* FIXME? here we are really changing the values
+	       * stored in the original calling routines outgoing
+	       * parameter stack.  This is probably a really
+	       * naughty thing to do but...
+	       */
+	      avalue[i] = pst;
+	      pst += 1;
+	    }
+	  break;
+
+	case FFI_TYPE_DOUBLE:
+	  /* On the outgoing stack all values are aligned to 8 */
+	  /* there are 8 64bit floating point registers */
+
+	  if (nf < 8)
+	    {
+	      avalue[i] = pfr;
+	      nf++;
+	      pfr++;
+	    }
+	  else
+	    {
+	      if (((long) pst) & 4)
+		pst++;
+	      avalue[i] = pst;
+	      pst += 2;
+	    }
+	  break;
+
+#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+	case FFI_TYPE_LONGDOUBLE:
+	  if (nf < 7)
+	    {
+	      avalue[i] = pfr;
+	      pfr += 2;
+	      nf += 2;
+	    }
+	  else
+	    {
+	      if (((long) pst) & 4)
+		pst++;
+	      avalue[i] = pst;
+	      pst += 4;
+	      nf = 8;
+	    }
+	  break;
+#endif
+#endif /* have FPRS */
+
+	case FFI_TYPE_UINT128:
+		/*
+		 * Test if for the whole long double, 4 gprs are available.
+		 * otherwise the stuff ends up on the stack.
+		 */
+		if (ng < 5) {
+			avalue[i] = pgr;
+			pgr += 4;
+			ng += 4;
+		} else {
+			avalue[i] = pst;
+			pst += 4;
+			ng = 8+4;
+		}
+		break;
+
 	case FFI_TYPE_SINT8:
 	case FFI_TYPE_UINT8:
 	  /* there are 8 gpr registers used to pass values */
@@ -1088,7 +1240,6 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
 	case FFI_TYPE_SINT32:
 	case FFI_TYPE_UINT32:
 	case FFI_TYPE_POINTER:
-	soft_float_closure:
 	  /* there are 8 gpr registers used to pass values */
 	  if (ng < 8)
 	    {
@@ -1104,9 +1255,6 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
 	  break;
 
 	case FFI_TYPE_STRUCT:
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-	do_struct:
-#endif
 	  /* Structs are passed by reference. The address will appear in a
 	     gpr if it is one of the first 8 arguments.  */
 	  if (ng < 8)
@@ -1124,7 +1272,6 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
 
 	case FFI_TYPE_SINT64:
 	case FFI_TYPE_UINT64:
-	soft_double_closure:
 	  /* passing long long ints are complex, they must
 	   * be passed in suitable register pairs such as
 	   * (r3,r4) or (r5,r6) or (r6,r7), or (r7,r8) or (r9,r10)
@@ -1156,99 +1303,8 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
 	    }
 	  break;
 
-	case FFI_TYPE_FLOAT:
-	  /* With FFI_LINUX_SOFT_FLOAT floats are handled like UINT32.  */
-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	    goto soft_float_closure;
-	  /* unfortunately float values are stored as doubles
-	   * in the ffi_closure_SYSV code (since we don't check
-	   * the type in that routine).
-	   */
-
-	  /* there are 8 64bit floating point registers */
-
-	  if (nf < 8)
-	    {
-	      temp = pfr->d;
-	      pfr->f = (float) temp;
-	      avalue[i] = pfr;
-	      nf++;
-	      pfr++;
-	    }
-	  else
-	    {
-	      /* FIXME? here we are really changing the values
-	       * stored in the original calling routines outgoing
-	       * parameter stack.  This is probably a really
-	       * naughty thing to do but...
-	       */
-	      avalue[i] = pst;
-	      pst += 1;
-	    }
-	  break;
-
-	case FFI_TYPE_DOUBLE:
-	  /* With FFI_LINUX_SOFT_FLOAT doubles are handled like UINT64.  */
-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	    goto soft_double_closure;
-	  /* On the outgoing stack all values are aligned to 8 */
-	  /* there are 8 64bit floating point registers */
-
-	  if (nf < 8)
-	    {
-	      avalue[i] = pfr;
-	      nf++;
-	      pfr++;
-	    }
-	  else
-	    {
-	      if (((long) pst) & 4)
-		pst++;
-	      avalue[i] = pst;
-	      pst += 2;
-	    }
-	  break;
-
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-	case FFI_TYPE_LONGDOUBLE:
-	  if (cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
-	    goto do_struct;
-	  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-	    { /* Test if for the whole long double, 4 gprs are available.
-		 otherwise the stuff ends up on the stack.  */
-	      if (ng < 5)
-		{
-		  avalue[i] = pgr;
-		  pgr += 4;
-		  ng += 4;
-		}
-	      else
-		{
-		  avalue[i] = pst;
-		  pst += 4;
-		  ng = 8;
-		}
-	      break;
-	    }
-	  if (nf < 7)
-	    {
-	      avalue[i] = pfr;
-	      pfr += 2;
-	      nf += 2;
-	    }
-	  else
-	    {
-	      if (((long) pst) & 4)
-		pst++;
-	      avalue[i] = pst;
-	      pst += 4;
-	      nf = 8;
-	    }
-	  break;
-#endif
-
 	default:
-	  FFI_ASSERT (0);
+		FFI_ASSERT (0);
 	}
 
       i++;
@@ -1265,39 +1321,9 @@ ffi_closure_helper_SYSV (ffi_closure *closure, void *rvalue,
      already used and we never have a struct with size zero. That is the reason
      for the subtraction of 1. See the comment in ffitarget.h about ordering.
   */
-  if (cif->abi == FFI_SYSV && cif->rtype->type == FFI_TYPE_STRUCT
+  if (cif->abi == FFI_SYSV && rtypenum == FFI_TYPE_STRUCT && size <= 8)
-      && size <= 8)
     return (FFI_SYSV_TYPE_SMALL_STRUCT - 1) + size;
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+  return rtypenum;
-  else if (cif->rtype->type == FFI_TYPE_LONGDOUBLE
-	   && cif->abi != FFI_LINUX && cif->abi != FFI_LINUX_SOFT_FLOAT)
-    return FFI_TYPE_STRUCT;
-#endif
-  /* With FFI_LINUX_SOFT_FLOAT floats and doubles are handled like UINT32
-     respectivley UINT64.  */
-  if (cif->abi == FFI_LINUX_SOFT_FLOAT)
-    {
-      switch (cif->rtype->type)
-	{
-	case FFI_TYPE_FLOAT:
-	  return FFI_TYPE_UINT32;
-	  break;
-	case FFI_TYPE_DOUBLE:
-	  return FFI_TYPE_UINT64;
-	  break;
-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
-	case FFI_TYPE_LONGDOUBLE:
-	  return FFI_TYPE_UINT128;
-	  break;
-#endif
-	default:
-	  return cif->rtype->type;
-	}
-    }
-  else
-    {
-      return cif->rtype->type;
-    }
 }
 
 int FFI_HIDDEN ffi_closure_helper_LINUX64 (ffi_closure *, void *,

src/powerpc/ffi.c.rej

@@ -0,0 +1,12 @@
+--- src/powerpc/ffi.c
++++ src/powerpc/ffi.c
+@@ -717,9 +734,6 @@
+ 		}
+ 	    }
+ 	}
+-#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
+-    byref:
+-#endif
+       intarg_count++;
+       flags |= FLAG_RETVAL_REFERENCE;
+       /* Fall through.  */

src/powerpc/ffitarget.h

@@ -60,18 +60,14 @@ typedef enum ffi_abi {
   FFI_LINUX64,
   FFI_LINUX,
   FFI_LINUX_SOFT_FLOAT,
-# ifdef POWERPC64
+# if defined(POWERPC64)
   FFI_DEFAULT_ABI = FFI_LINUX64,
-# else
+# elif defined(__NO_FPRS__)
-#  if (!defined(__NO_FPRS__) && (__LDBL_MANT_DIG__ == 106))
-  FFI_DEFAULT_ABI = FFI_LINUX,
-#  else
-#   ifdef __NO_FPRS__
   FFI_DEFAULT_ABI = FFI_LINUX_SOFT_FLOAT,
-#   else
+# elif (__LDBL_MANT_DIG__ == 106)
+  FFI_DEFAULT_ABI = FFI_LINUX,
+# else
   FFI_DEFAULT_ABI = FFI_GCC_SYSV,
-#   endif
-#  endif
 # endif
 #endif
 

src/powerpc/ppc_closure.S

@@ -122,22 +122,41 @@ ENTRY(ffi_closure_SYSV)
 	blr
 
 # case FFI_TYPE_FLOAT
+#ifndef __NO_FPRS__
 	lfs %f1,112+0(%r1)
 	mtlr %r0
 	addi %r1,%r1,144
+#else
+	nop
+	nop
+	nop
+#endif
 	blr
 
 # case FFI_TYPE_DOUBLE
+#ifndef __NO_FPRS__
 	lfd %f1,112+0(%r1)
 	mtlr %r0
 	addi %r1,%r1,144
+#else
+	nop
+	nop
+	nop
+#endif
 	blr
 
 # case FFI_TYPE_LONGDOUBLE
+#ifndef __NO_FPRS__
 	lfd %f1,112+0(%r1)
 	lfd %f2,112+8(%r1)
 	mtlr %r0
 	b .Lfinish
+#else
+	nop
+	nop
+	nop
+	blr
+#endif
 
 # case FFI_TYPE_UINT8
 	lbz %r3,112+3(%r1)

src/powerpc/sysv.S

@@ -83,6 +83,7 @@ ENTRY(ffi_call_SYSV)
 	nop
 1:
 
+#ifndef __NO_FPRS__
 	/* Load all the FP registers.  */
 	bf-	6,2f
 	lfd	%f1,-16-(8*4)-(8*8)(%r28)
@@ -94,6 +95,7 @@ ENTRY(ffi_call_SYSV)
 	lfd	%f6,-16-(8*4)-(3*8)(%r28)
 	lfd	%f7,-16-(8*4)-(2*8)(%r28)
 	lfd	%f8,-16-(8*4)-(1*8)(%r28)
+#endif
 2:
 
 	/* Make the call.  */
@@ -103,7 +105,9 @@ ENTRY(ffi_call_SYSV)
 	mtcrf	0x01,%r31 /* cr7  */
 	bt-	31,L(small_struct_return_value)
 	bt-	30,L(done_return_value)
+#ifndef __NO_FPRS__
 	bt-	29,L(fp_return_value)
+#endif
 	stw	%r3,0(%r30)
 	bf+	28,L(done_return_value)
 	stw	%r4,4(%r30)
@@ -124,6 +128,7 @@ L(done_return_value):
 	lwz	%r1,0(%r1)
 	blr
 
+#ifndef __NO_FPRS__
 L(fp_return_value):
 	bf	28,L(float_return_value)
 	stfd	%f1,0(%r30)
@@ -134,6 +139,7 @@ L(fp_return_value):
 L(float_return_value):
 	stfs	%f1,0(%r30)
 	b	L(done_return_value)
+#endif
 
 L(small_struct_return_value):
 	/*

src/powerpc/sysv.S.orig

@@ -0,0 +1,214 @@
+/* -----------------------------------------------------------------------
+   sysv.S - Copyright (c) 1998 Geoffrey Keating
+   Copyright (C) 2007 Free Software Foundation, Inc
+
+   PowerPC Assembly glue.
+
+   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.
+   ----------------------------------------------------------------------- */
+
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+#include <powerpc/asm.h>
+
+#ifndef __powerpc64__
+	.globl ffi_prep_args_SYSV
+ENTRY(ffi_call_SYSV)
+.LFB1:
+	/* Save the old stack pointer as AP.  */
+	mr	%r8,%r1
+
+.LCFI0:
+	/* Allocate the stack space we need.  */
+	stwux	%r1,%r1,%r4
+	/* Save registers we use.  */
+	mflr	%r9
+	stw	%r28,-16(%r8)
+.LCFI1:
+	stw	%r29,-12(%r8)
+.LCFI2:
+	stw	%r30, -8(%r8)
+.LCFI3:
+	stw	%r31, -4(%r8)
+.LCFI4:
+	stw	%r9,   4(%r8)
+.LCFI5:
+
+	/* Save arguments over call...  */
+	mr	%r31,%r5	/* flags, */
+	mr	%r30,%r6	/* rvalue, */
+	mr	%r29,%r7	/* function address, */
+	mr	%r28,%r8	/* our AP. */
+.LCFI6:
+
+	/* Call ffi_prep_args_SYSV.  */
+	mr	%r4,%r1
+	bl	ffi_prep_args_SYSV@local
+
+	/* Now do the call.  */
+	/* Set up cr1 with bits 4-7 of the flags.  */
+	mtcrf	0x40,%r31
+	/* Get the address to call into CTR.  */
+	mtctr	%r29
+	/* Load all those argument registers.  */
+	lwz	%r3,-16-(8*4)(%r28)
+	lwz	%r4,-16-(7*4)(%r28)
+	lwz	%r5,-16-(6*4)(%r28)
+	lwz	%r6,-16-(5*4)(%r28)
+	bf-	5,1f
+	nop
+	lwz	%r7,-16-(4*4)(%r28)
+	lwz	%r8,-16-(3*4)(%r28)
+	lwz	%r9,-16-(2*4)(%r28)
+	lwz	%r10,-16-(1*4)(%r28)
+	nop
+1:
+
+	/* Load all the FP registers.  */
+	bf-	6,2f
+	lfd	%f1,-16-(8*4)-(8*8)(%r28)
+	lfd	%f2,-16-(8*4)-(7*8)(%r28)
+	lfd	%f3,-16-(8*4)-(6*8)(%r28)
+	lfd	%f4,-16-(8*4)-(5*8)(%r28)
+	nop
+	lfd	%f5,-16-(8*4)-(4*8)(%r28)
+	lfd	%f6,-16-(8*4)-(3*8)(%r28)
+	lfd	%f7,-16-(8*4)-(2*8)(%r28)
+	lfd	%f8,-16-(8*4)-(1*8)(%r28)
+2:
+
+	/* Make the call.  */
+	bctrl
+
+	/* Now, deal with the return value.  */
+	mtcrf	0x01,%r31 /* cr7  */
+	bt-	31,L(small_struct_return_value)
+	bt-	30,L(done_return_value)
+	bt-	29,L(fp_return_value)
+	stw	%r3,0(%r30)
+	bf+	28,L(done_return_value)
+	stw	%r4,4(%r30)
+	mtcrf	0x02,%r31 /* cr6  */
+	bf	27,L(done_return_value)
+	stw     %r5,8(%r30)
+	stw	%r6,12(%r30)
+	/* Fall through...  */
+
+L(done_return_value):
+	/* Restore the registers we used and return.  */
+	lwz	%r9,   4(%r28)
+	lwz	%r31, -4(%r28)
+	mtlr	%r9
+	lwz	%r30, -8(%r28)
+	lwz	%r29,-12(%r28)
+	lwz	%r28,-16(%r28)
+	lwz	%r1,0(%r1)
+	blr
+
+L(fp_return_value):
+	bf	28,L(float_return_value)
+	stfd	%f1,0(%r30)
+	mtcrf   0x02,%r31 /* cr6  */
+	bf	27,L(done_return_value)
+	stfd	%f2,8(%r30)
+	b	L(done_return_value)
+L(float_return_value):
+	stfs	%f1,0(%r30)
+	b	L(done_return_value)
+
+L(small_struct_return_value):
+	/*
+	 * The C code always allocates a properly-aligned 8-byte bounce
+	 * buffer to make this assembly code very simple.  Just write out
+	 * r3 and r4 to the buffer to allow the C code to handle the rest.
+	 */
+	stw %r3, 0(%r30)
+	stw %r4, 4(%r30)
+	b L(done_return_value)
+
+.LFE1:
+END(ffi_call_SYSV)
+
+      .section	".eh_frame",EH_FRAME_FLAGS,@progbits
+.Lframe1:
+      .4byte    .LECIE1-.LSCIE1  /*  Length of Common Information Entry */
+.LSCIE1:
+      .4byte    0x0      /*  CIE Identifier Tag */
+      .byte     0x1      /*  CIE Version */
+#if defined _RELOCATABLE || defined __PIC__
+      .ascii	"zR\0"   /*  CIE Augmentation */
+#else
+      .ascii	"\0"	 /*  CIE Augmentation */
+#endif
+      .uleb128  0x1      /*  CIE Code Alignment Factor */
+      .sleb128  -4	 /*  CIE Data Alignment Factor */
+      .byte     0x41     /*  CIE RA Column */
+#if defined _RELOCATABLE || defined __PIC__
+      .uleb128  0x1      /*  Augmentation size */
+      .byte	0x1b	 /*  FDE Encoding (pcrel sdata4) */
+#endif
+      .byte     0xc      /*  DW_CFA_def_cfa */
+      .uleb128  0x1
+      .uleb128  0x0
+      .align 2
+.LECIE1:
+.LSFDE1:
+      .4byte    .LEFDE1-.LASFDE1         /*  FDE Length */
+.LASFDE1:
+      .4byte    .LASFDE1-.Lframe1         /*  FDE CIE offset */
+#if defined _RELOCATABLE || defined __PIC__
+      .4byte    .LFB1-.  /*  FDE initial location */
+#else
+      .4byte    .LFB1    /*  FDE initial location */
+#endif
+      .4byte    .LFE1-.LFB1      /*  FDE address range */
+#if defined _RELOCATABLE || defined __PIC__
+      .uleb128  0x0	 /*  Augmentation size */
+#endif
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI0-.LFB1
+      .byte     0xd      /*  DW_CFA_def_cfa_register */
+      .uleb128  0x08
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI5-.LCFI0
+      .byte     0x11     /*  DW_CFA_offset_extended_sf */
+      .uleb128  0x41
+      .sleb128  -1
+      .byte     0x9f     /*  DW_CFA_offset, column 0x1f */
+      .uleb128  0x1
+      .byte     0x9e     /*  DW_CFA_offset, column 0x1e */
+      .uleb128  0x2
+      .byte     0x9d     /*  DW_CFA_offset, column 0x1d */
+      .uleb128  0x3
+      .byte     0x9c     /*  DW_CFA_offset, column 0x1c */
+      .uleb128  0x4
+      .byte     0x4      /*  DW_CFA_advance_loc4 */
+      .4byte    .LCFI6-.LCFI5
+      .byte     0xd      /*  DW_CFA_def_cfa_register */
+      .uleb128  0x1c
+      .align 2
+.LEFDE1:
+#endif
+
+#if defined __ELF__ && defined __linux__
+	.section	.note.GNU-stack,"",@progbits
+#endif