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

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

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

From-SVN: r154855

libffi/ChangeLog

+2009-11-30  David Edelsohn  <edelsohn@gnu.org>
+
+	PR target/35484
+	* src/powerpc/ffitarget.h (POWERPC64): Define for PPC64 Linux and
+	AIX64.
+	* src/powerpc/aix.S: Implement AIX64 version.
+	* src/powerpc/aix_closure.S: Implement AIX64 version.
+	(ffi_closure_ASM): Use extsb, lha and displament addresses.
+	* src/powerpc/ffi_darwin.c (ffi_prep_args): Implement AIX64
+	support.
+	(ffi_prep_cif_machdep): Same.
+	(ffi_call): Same.
+	(ffi_closure_helper_DARWIN): Same.
+
 2009-11-02  Andreas Tobler  <a.tobler@schweiz.org>
 
 	PR libffi/41908

libffi/src/powerpc/aix.S

 /* -----------------------------------------------------------------------
-   aix.S - Copyright (c) 2002 Free Software Foundation, Inc.
+   aix.S - Copyright (c) 2002,2009 Free Software Foundation, Inc.
    based on darwin.S by John Hornkvist
 
    PowerPC Assembly glue.
@@ -86,9 +86,13 @@
 #define L(x) x
 	.file "aix.S"
 	.toc
-	.csect .text[PR]
-	.align 2
-.globl ffi_prep_args
+
+	/* void ffi_call_AIX(extended_cif *ecif, unsigned long bytes,
+	 *		     unsigned int flags, unsigned int *rvalue,
+	 *		     void (*fn)(),
+	 *		     void (*prep_args)(extended_cif*, unsigned *const));
+	 * r3=ecif, r4=bytes, r5=flags, r6=rvalue, r7=fn, r8=prep_args
+	 */
 
 .csect .text[PR]
 	.align 2
@@ -96,6 +100,120 @@
 	.globl .ffi_call_AIX
 .csect ffi_call_AIX[DS]
 ffi_call_AIX:
+#ifdef __64BIT__
+	.llong .ffi_call_AIX, TOC[tc0], 0
+	.csect .text[PR]
+.ffi_call_AIX:
+	mr	r12,r8 // We only need r12 until the call, so it doesn't have to be saved...
+	/* Save the old stack pointer as AP.  */
+	mr	r8,r1
+
+	/* Allocate the stack space we need.  */
+	stdux	r1,r1,r4
+
+	/* Save registers we use.  */
+	mflr	r9
+
+	std	r28,-32(r8)
+	std	r29,-24(r8)
+	std	r30,-16(r8)
+	std	r31, -8(r8)
+
+	std	r9, 16(r8)
+	std	r2, 40(r1)
+
+	/* Save arguments over call...  */
+	mr	r31,r5	/* flags, */
+	mr	r30,r6	/* rvalue, */
+	mr	r29,r7	/* function address, */
+	mr	r28,r8	/* our AP. */
+
+	/* Call ffi_prep_args.  */
+	mr	r4,r1
+	li	r9,0
+
+	ld	r2,8(r12)
+	ld	r12,0(r12)
+	mtctr	r12 // r12 holds address of _ffi_prep_args
+	bctrl
+	ld	r2,40(r1)
+
+	/* Now do the call.  */
+	ld	r12,0(r29)
+	/* Set up cr1 with bits 4-7 of the flags.  */
+	mtcrf	0x40,r31
+	std	r2,40(r1)
+	mtctr	r12
+	ld	r2,8(r29)
+	/* Load all those argument registers.  */
+	// We have set up a nice stack frame, just load it into registers.
+	ld	r3, 40+(1*8)(r1)
+	ld	r4, 40+(2*8)(r1)
+	ld	r5, 40+(3*8)(r1)
+	ld	r6, 40+(4*8)(r1)
+	nop
+	ld	r7, 40+(5*8)(r1)
+	ld	r8, 40+(6*8)(r1)
+	ld	r9, 40+(7*8)(r1)
+	ld	r10,40+(8*8)(r1)
+
+L1:
+	/* Load all the FP registers.  */
+	bf	6,L2 // 2f + 0x18
+	lfd	f1,-32-(13*8)(r28)
+	lfd	f2,-32-(12*8)(r28)
+	lfd	f3,-32-(11*8)(r28)
+	lfd	f4,-32-(10*8)(r28)
+	nop
+	lfd	f5,-32-(9*8)(r28)
+	lfd	f6,-32-(8*8)(r28)
+	lfd	f7,-32-(7*8)(r28)
+	lfd	f8,-32-(6*8)(r28)
+	nop
+	lfd	f9,-32-(5*8)(r28)
+	lfd	f10,-32-(4*8)(r28)
+	lfd	f11,-32-(3*8)(r28)
+	lfd	f12,-32-(2*8)(r28)
+	nop
+	lfd	f13,-32-(1*8)(r28)
+
+L2:
+	/* Make the call.  */
+	bctrl
+	ld	r2,40(r1)
+
+	/* Now, deal with the return value.  */
+	mtcrf	0x01,r31
+
+	bt	30,L(done_return_value)
+	bt	29,L(fp_return_value)
+	std	r3,0(r30)
+	bf	28,L(done_return_value)
+	std	r4,4(r30)
+
+	/* Fall through...  */
+
+L(done_return_value):
+	/* Restore the registers we used and return.  */
+	ld	r9,16(r28)
+	ld	r31,-8(r28)
+	mtlr	r9
+	ld	r30,-16(r28)
+	ld	r29,-24(r28)
+	ld	r28,-32(r28)
+	ld	r1,0(r1)
+	blr
+
+L(fp_return_value):
+	bf	28,L(float_return_value)
+	stfd	f1,0(r30)
+	b	L(done_return_value)
+L(float_return_value):
+	stfs	f1,0(r30)
+	b	L(done_return_value)
+
+#else /* ! __64BIT__ */
+	
 	.long .ffi_call_AIX, TOC[tc0], 0
 	.csect .text[PR]
 .ffi_call_AIX:
@@ -190,8 +308,8 @@ L2:
 
 L(done_return_value):
 	/* Restore the registers we used and return.  */
-	lwz	r9,   8(r28)
-	lwz	r31,  -4(r28)
+	lwz	r9,8(r28)
+	lwz	r31,-4(r28)
 	mtlr	r9
 	lwz	r30, -8(r28)
 	lwz	r29,-12(r28)
@@ -206,6 +324,7 @@ L(fp_return_value):
 L(float_return_value):
 	stfs	f1,0(r30)
 	b	L(done_return_value)
+#endif
 	.long 0
 	.byte 0,0,0,1,128,4,0,0
 //END(ffi_call_AIX)
@@ -216,7 +335,11 @@ L(float_return_value):
 	.globl .ffi_call_DARWIN
 .csect ffi_call_DARWIN[DS]
 ffi_call_DARWIN:
+#ifdef __64BIT__
+	.llong .ffi_call_DARWIN, TOC[tc0], 0
+#else
 	.long .ffi_call_DARWIN, TOC[tc0], 0
+#endif
 	.csect .text[PR]
 .ffi_call_DARWIN:
 	blr

libffi/src/powerpc/aix_closure.S

 /* -----------------------------------------------------------------------
-   aix_closure.S - Copyright (c) 2002 2003 Free Software Foundation, Inc.
+   aix_closure.S - Copyright (c) 2002, 2003, 2009 Free Software Foundation, Inc.
    based on darwin_closure.S
 
    PowerPC Assembly glue.
@@ -94,8 +94,162 @@ LC..60:
 	.globl ffi_closure_ASM
 	.globl .ffi_closure_ASM
 .csect ffi_closure_ASM[DS]
-
 ffi_closure_ASM:
+#ifdef __64BIT__
+	.llong .ffi_closure_ASM, TOC[tc0], 0
+	.csect .text[PR]
+.ffi_closure_ASM:
+
+	mflr r0			/* extract return address */
+	std r0,16(r1)		/* save the return address */
+
+	/* 48  Bytes (Linkage Area) */
+	/* 64  Bytes (params) */
+	/* 104 Bytes (13*8 from FPR) */
+	/* 32  Bytes (result) */
+	/* 248 Bytes */
+
+	stdu r1,-248(r1)	/* skip over caller save area
+				   keep stack aligned to 16  */
+
+/* we want to build up an area for the parameters passed */
+/* in registers (both floating point and integer) */
+
+	/* we store gpr 3 to gpr 10 (aligned to 4)
+	in the parents outgoing area  */
+	std   r3, (304+0*8)(r1)
+	std   r4, (304+1*8)(r1)
+	std   r5, (304+2*8)(r1)
+	std   r6, (304+3*8)(r1)
+	std   r7, (304+4*8)(r1)
+	std   r8, (304+5*8)(r1)
+	std   r9, (304+6*8)(r1)
+	std   r10, (304+7*8)(r1)
+
+	/* next save fpr 1 to fpr 13 (aligned to 8) */
+	stfd  f1, (112+0*8)(r1)
+	stfd  f2, (112+1*8)(r1)
+	stfd  f3, (112+2*8)(r1)
+	stfd  f4, (112+3*8)(r1)
+	stfd  f5, (112+4*8)(r1)
+	stfd  f6, (112+5*8)(r1)
+	stfd  f7, (112+6*8)(r1)
+	stfd  f8, (112+7*8)(r1)
+	stfd  f9, (112+8*8)(r1)
+	stfd  f10, (112+9*8)(r1)
+	stfd  f11, (112+10*8)(r1)
+	stfd  f12, (112+11*8)(r1)
+	stfd  f13, (112+12*8)(r1)
+
+	/* set up registers for the routine that actually does the work */
+	/* get the context pointer from the trampoline */
+	mr r3,r11
+
+	/* now load up the pointer to the result storage */
+	addi r4,r1,216
+
+	/* now load up the pointer to the saved gpr registers */
+	addi r5,r1,304
+
+	/* now load up the pointer to the saved fpr registers */
+	addi r6,r1,112
+
+	/* make the call */
+	bl .ffi_closure_helper_DARWIN
+	nop
+
+	/* now r3 contains the return type */
+	/* so use it to look up in a table */
+	/* so we know how to deal with each type */
+
+	/* look up the proper starting point in table  */
+	/* by using return type as offset */
+	addi r5,r1,216		/* get pointer to results area */
+	ld r4,LC..60(2)		/* get address of jump table */
+	sldi r3,r3,2		/* now multiply return type by 4 */
+	lwzx r3,r4,r3		/* get the contents of that table value */
+	add r3,r3,r4		/* add contents of table to table address */
+	mtctr r3
+	bctr			/* jump to it */
+
+L..60:
+	.long L..44-L..60    /* FFI_TYPE_VOID */
+	.long L..51-L..60    /* FFI_TYPE_INT */
+	.long L..47-L..60    /* FFI_TYPE_FLOAT */
+	.long L..46-L..60    /* FFI_TYPE_DOUBLE */
+	.long L..45-L..60    /* FFI_TYPE_LONGDOUBLE */
+	.long L..56-L..60    /* FFI_TYPE_UINT8 */
+	.long L..55-L..60    /* FFI_TYPE_SINT8 */
+	.long L..58-L..60    /* FFI_TYPE_UINT16 */
+	.long L..57-L..60    /* FFI_TYPE_SINT16 */
+	.long L..50-L..60    /* FFI_TYPE_UINT32 */
+	.long L..51-L..60    /* FFI_TYPE_SINT32 */
+	.long L..48-L..60    /* FFI_TYPE_UINT64 */
+	.long L..48-L..60    /* FFI_TYPE_SINT64 */
+	.long L..44-L..60    /* FFI_TYPE_STRUCT */
+	.long L..48-L..60    /* FFI_TYPE_POINTER */
+
+
+/* case long double */
+L..45:
+	lfd f1,0(r5)
+	lfd f2,8(r5)
+	b L..44
+
+/* case double */
+L..46:
+	lfd f1,0(r5)
+	b L..44
+
+/* case float */
+L..47:
+	lfs f1,0(r5)
+	b L..44
+
+/* case long long / pointer */
+L..48:
+	ld r3,0(r5)
+	b L..44
+
+/* case uint32 */
+L..50:
+	lwz r3,4(r5)
+	b L..44
+
+/* case int / sint32 */
+L..51:
+	lwa r3,4(r5)
+	b L..44
+
+/* case signed int8 */
+L..55:
+	lbz r3,7(r5)
+	extsb r3,r3
+	b L..44
+
+/* case unsigned int8 */
+L..56:
+	lbz r3,7(r5)
+	b L..44
+
+/* case signed int16 */
+L..57:
+	lha r3,6(r5)
+	b L..44
+
+/* case unsigned int16 */
+L..58:
+	lhz r3,6(r5)
+
+/* case void / done	 */
+L..44:
+	addi r1,r1,248		/* restore stack pointer */
+	ld r0,16(r1)		/* get return address */
+	mtlr r0			/* reset link register */
+	blr
+
+#else /* ! __64BIT__ */
+	
 	.long .ffi_closure_ASM, TOC[tc0], 0
 	.csect .text[PR]
 .ffi_closure_ASM:
@@ -106,8 +260,8 @@ ffi_closure_ASM:
 	/* 24 Bytes (Linkage Area) */
 	/* 32 Bytes (params) */
 	/* 104 Bytes (13*8 from FPR) */
-	/* 8 Bytes (result) */
-	/* 168 Bytes */
+	/* 16  Bytes (result) */
+	/* 176 Bytes */
 
 	stwu r1,-176(r1)	/* skip over caller save area
 				keep stack aligned to 16  */
@@ -177,7 +331,7 @@ L..60:
 	.long L..50-L..60    /* FFI_TYPE_INT */
 	.long L..47-L..60    /* FFI_TYPE_FLOAT */
 	.long L..46-L..60    /* FFI_TYPE_DOUBLE */
-	.long L..46-L..60    /* FFI_TYPE_LONGDOUBLE */
+	.long L..45-L..60    /* FFI_TYPE_LONGDOUBLE */
 	.long L..56-L..60    /* FFI_TYPE_UINT8 */
 	.long L..55-L..60    /* FFI_TYPE_SINT8 */
 	.long L..58-L..60    /* FFI_TYPE_UINT16 */
@@ -190,6 +344,12 @@ L..60:
 	.long L..50-L..60    /* FFI_TYPE_POINTER */
 
 
+/* case long double */
+L..45:
+	lfd f1,0(r5)
+	lfd f2,8(r5)
+	b L..44
+
 /* case double */
 L..46:
 	lfd f1,0(r5)
@@ -213,29 +373,23 @@ L..50:
 
 /* case signed int8 */
 L..55:
-	addi r5,r5,3
-	lbz r3,0(r5)
-	slwi r3,r3,24
-	srawi r3,r3,24
+	lbz r3,3(r5)
+	extsb r3,r3
 	b L..44
 
 /* case unsigned int8 */
 L..56:
-	addi r5,r5,3
-	lbz r3,0(r5)
+	lbz r3,3(r5)
 	b L..44
 
 /* case signed int16 */
 L..57:
-	addi r5,r5,2
-	lhz r3,0(r5)
-	extsh r3,r3
+	lha r3,2(r5)
 	b L..44
 
 /* case unsigned int16 */
 L..58:
-	addi r5,r5,2
-	lhz r3,0(r5)
+	lhz r3,2(r5)
 
 /* case void / done	 */
 L..44:
@@ -243,5 +397,5 @@ L..44:
 	lwz r0,8(r1)		/* get return address */
 	mtlr r0			/* reset link register */
 	blr
-
+#endif
 /* END(ffi_closure_ASM) */

libffi/src/powerpc/ffitarget.h

@@ -30,7 +30,11 @@
 
 /* ---- System specific configurations ----------------------------------- */
 
-#if defined (POWERPC) && defined (__powerpc64__)
+#if defined (POWERPC) && defined (__powerpc64__)	/* linux64 */
+#define POWERPC64
+#elif defined (POWERPC_DARWIN) && defined (__ppc64__)	/* Darwin */
+#define POWERPC64
+#elif defined (POWERPC_AIX) && defined (__64BIT__)	/* AIX64 */
 #define POWERPC64
 #endif