ffi 0.5.4 → 0.6.0

data/ext/ffi_c/libffi/src/powerpc/aix_closure.S

@@ -1,5 +1,5 @@
 /* -----------------------------------------------------------------------
-   aix_closure.S - Copyright (c) 2002 2003 Free Software Foundation, Inc.
+   aix_closure.S - Copyright (c) 2002, 2003, 2009 Free Software Foundation, Inc.
    based on darwin_closure.S
 
    PowerPC Assembly glue.
@@ -94,65 +94,238 @@ LC..60:
 	.globl ffi_closure_ASM
 	.globl .ffi_closure_ASM
 .csect ffi_closure_ASM[DS]
-
 ffi_closure_ASM:
-	.long .ffi_closure_ASM, TOC[tc0], 0
+#ifdef __64BIT__
+	.llong .ffi_closure_ASM, TOC[tc0], 0
 	.csect .text[PR]
 .ffi_closure_ASM:
+/* we want to build up an area for the parameters passed */
+/* in registers (both floating point and integer) */
 
-	mflr r0			/* extract return address */
-	stw r0, 8(r1)		/* save the return address */
-
-	/* 24 Bytes (Linkage Area) */
-	/* 32 Bytes (params) */
+	/* we store gpr 3 to gpr 10 (aligned to 4)
+	in the parents outgoing area  */
+	std   r3, 48+(0*8)(r1)
+	std   r4, 48+(1*8)(r1)
+	std   r5, 48+(2*8)(r1)
+	std   r6, 48+(3*8)(r1)
+	mflr  r0
+
+	std   r7, 48+(4*8)(r1)
+	std   r8, 48+(5*8)(r1)
+	std   r9, 48+(6*8)(r1)
+	std   r10, 48+(7*8)(r1)
+	std   r0, 16(r1)	/* save the return address */
+
+
+	/* 48  Bytes (Linkage Area) */
+	/* 64  Bytes (params) */
+	/* 16  Bytes (result) */
 	/* 104 Bytes (13*8 from FPR) */
-	/* 8 Bytes (result) */
-	/* 168 Bytes */
+	/* 8   Bytes (alignment) */
+	/* 240 Bytes */
+
+	stdu  r1, -240(r1)	/* skip over caller save area
+				   keep stack aligned to 16  */
+
+	/* next save fpr 1 to fpr 13 (aligned to 8) */
+	stfd  f1, 128+(0*8)(r1)
+	stfd  f2, 128+(1*8)(r1)
+	stfd  f3, 128+(2*8)(r1)
+	stfd  f4, 128+(3*8)(r1)
+	stfd  f5, 128+(4*8)(r1)
+	stfd  f6, 128+(5*8)(r1)
+	stfd  f7, 128+(6*8)(r1)
+	stfd  f8, 128+(7*8)(r1)
+	stfd  f9, 128+(8*8)(r1)
+	stfd  f10, 128+(9*8)(r1)
+	stfd  f11, 128+(10*8)(r1)
+	stfd  f12, 128+(11*8)(r1)
+	stfd  f13, 128+(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, 112
+
+	/* now load up the pointer to the saved gpr registers */
+	addi r5, r1, 288
+
+	/* now load up the pointer to the saved fpr registers */
+	addi r6, r1, 128
+
+	/* 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 */
+	ld	r4, LC..60(2)	/* get address of jump table */
+	sldi	r3, r3, 4	/* now multiply return type by 16 */
+	ld	r0, 240+16(r1)	/* load return address */
+	add	r3, r3, r4	/* add contents of table to table address */
+	mtctr	r3
+	bctr			/* jump to it */
+
+/* Each fragment must be exactly 16 bytes long (4 instructions).
+   Align to 16 byte boundary for cache and dispatch efficiency.  */
+	.align 4
+
+L..60:
+/* case FFI_TYPE_VOID */
+	mtlr r0
+	addi r1, r1, 240
+	blr
+	nop
+
+/* case FFI_TYPE_INT */
+	lwa r3, 112+4(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_FLOAT */
+	lfs f1, 112+0(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
 
-	stwu r1,-176(r1)	/* skip over caller save area
-				keep stack aligned to 16  */
+/* case FFI_TYPE_DOUBLE */
+	lfd f1, 112+0(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_LONGDOUBLE */
+	lfd f1, 112+0(r1)
+	mtlr r0
+	lfd f2, 112+8(r1)
+	b L..finish
+
+/* case FFI_TYPE_UINT8 */
+	lbz r3, 112+7(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_SINT8 */
+	lbz r3, 112+7(r1)
+	mtlr r0
+	extsb r3, r3
+	b L..finish
+
+/* case FFI_TYPE_UINT16 */
+	lhz r3, 112+6(r1)
+	mtlr r0
+L..finish:
+	addi r1, r1, 240
+	blr
 
+/* case FFI_TYPE_SINT16 */
+	lha r3, 112+6(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_UINT32 */
+	lwz r3, 112+4(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_SINT32 */
+	lwa r3, 112+4(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_UINT64 */
+	ld r3, 112+0(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_SINT64 */
+	ld r3, 112+0(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+/* case FFI_TYPE_STRUCT */
+	mtlr r0
+	addi r1, r1, 240
+	blr
+	nop
+
+/* case FFI_TYPE_POINTER */
+	ld r3, 112+0(r1)
+	mtlr r0
+	addi r1, r1, 240
+	blr
+
+#else /* ! __64BIT__ */
+	
+	.long .ffi_closure_ASM, TOC[tc0], 0
+	.csect .text[PR]
+.ffi_closure_ASM:
 /* 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  */
-	stw   r3, 200(r1)
-	stw   r4, 204(r1)
-	stw   r5, 208(r1)
-	stw   r6, 212(r1)
-	stw   r7, 216(r1)
-	stw   r8, 220(r1)
-	stw   r9, 224(r1)
-	stw   r10, 228(r1)
+	stw   r3, 24+(0*4)(r1)
+	stw   r4, 24+(1*4)(r1)
+	stw   r5, 24+(2*4)(r1)
+	stw   r6, 24+(3*4)(r1)
+	mflr  r0
+
+	stw   r7, 24+(4*4)(r1)
+	stw   r8, 24+(5*4)(r1)
+	stw   r9, 24+(6*4)(r1)
+	stw   r10, 24+(7*4)(r1)
+	stw   r0, 8(r1)
+
+	/* 24 Bytes (Linkage Area) */
+	/* 32 Bytes (params) */
+	/* 16  Bytes (result) */
+	/* 104 Bytes (13*8 from FPR) */
+	/* 176 Bytes */
+
+	stwu  r1, -176(r1)	/* skip over caller save area
+				   keep stack aligned to 16  */
 
 	/* next save fpr 1 to fpr 13 (aligned to 8) */
-	stfd  f1, 56(r1)
-	stfd  f2, 64(r1)
-	stfd  f3, 72(r1)
-	stfd  f4, 80(r1)
-	stfd  f5, 88(r1)
-	stfd  f6, 96(r1)
-	stfd  f7, 104(r1)
-	stfd  f8, 112(r1)
-	stfd  f9, 120(r1)
-	stfd  f10, 128(r1)
-	stfd  f11, 136(r1)
-	stfd  f12, 144(r1)
-	stfd  f13, 152(r1)
+	stfd  f1, 72+(0*8)(r1)
+	stfd  f2, 72+(1*8)(r1)
+	stfd  f3, 72+(2*8)(r1)
+	stfd  f4, 72+(3*8)(r1)
+	stfd  f5, 72+(4*8)(r1)
+	stfd  f6, 72+(5*8)(r1)
+	stfd  f7, 72+(6*8)(r1)
+	stfd  f8, 72+(7*8)(r1)
+	stfd  f9, 72+(8*8)(r1)
+	stfd  f10, 72+(9*8)(r1)
+	stfd  f11, 72+(10*8)(r1)
+	stfd  f12, 72+(11*8)(r1)
+	stfd  f13, 72+(12*8)(r1)
 
 	/* set up registers for the routine that actually does the work */
 	/* get the context pointer from the trampoline */
-	mr r3,r11
+	mr r3, r11
 
 	/* now load up the pointer to the result storage */
-	addi r4,r1,160
+	addi r4, r1, 56
 
 	/* now load up the pointer to the saved gpr registers */
-	addi r5,r1,200
+	addi r5, r1, 200
 
 	/* now load up the pointer to the saved fpr registers */
-	addi r6,r1,56
+	addi r6, r1, 72
 
 	/* make the call */
 	bl .ffi_closure_helper_DARWIN
@@ -164,84 +337,107 @@ ffi_closure_ASM:
 
 	/* look up the proper starting point in table  */
 	/* by using return type as offset */
-	addi r5,r1,160		/* get pointer to results area */
-	lwz r4,LC..60(2)	/* get address of jump table */
-	slwi 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
+	lwz	r4, LC..60(2)	/* get address of jump table */
+	slwi	r3, r3, 4	/* now multiply return type by 4 */
+	lwz	r0, 176+8(r1)	/* load return address */
+	add	r3, r3, r4	/* add contents of table to table address */
+	mtctr	r3
 	bctr			/* jump to it */
 
+/* Each fragment must be exactly 16 bytes long (4 instructions).
+   Align to 16 byte boundary for cache and dispatch efficiency.  */
+	.align 4
+
 L..60:
-	.long L..44-L..60    /* FFI_TYPE_VOID */
-	.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..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..50-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..50-L..60    /* FFI_TYPE_POINTER */
-
-
-/* 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 */
-L..48:
-	lwz r3,0(r5)
-	lwz r4,4(r5)
-	b L..44
-
-/* case default / int32 / pointer */
-L..50:
-	lwz r3,0(r5)
-	b L..44
-
-/* case signed int8	 */
-L..55:
-	addi r5,r5,3
-	lbz r3,0(r5)
-	slwi r3,r3,24
-	srawi r3,r3,24
-	b L..44
-
-/* case unsigned int8	 */
-L..56:
-	addi r5,r5,3
-	lbz r3,0(r5)
-	b L..44
-
-/* case signed int16 */
-L..57:
-	addi r5,r5,2
-	lhz r3,0(r5)
-	extsh r3,r3
-	b L..44
-
-/* case unsigned int16 */
-L..58:
-	addi r5,r5,2
-	lhz r3,0(r5)
-
-/* case void / done	 */
-L..44:
-	addi r1,r1,176		/* restore stack pointer */
-	lwz r0,8(r1)		/* get return address */
-	mtlr r0			/* reset link register */
+/* case FFI_TYPE_VOID */
+	mtlr r0
+	addi r1, r1, 176
 	blr
+	nop
+
+/* case FFI_TYPE_INT */
+	lwz r3, 56+0(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_FLOAT */
+	lfs f1, 56+0(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_DOUBLE */
+	lfd f1, 56+0(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_LONGDOUBLE */
+	lfd f1, 56+0(r1)
+	mtlr r0
+	lfd f2, 56+8(r1)
+	b L..finish
+
+/* case FFI_TYPE_UINT8 */
+	lbz r3, 56+3(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_SINT8 */
+	lbz r3, 56+3(r1)
+	mtlr r0
+	extsb r3, r3
+	b L..finish
 
+/* case FFI_TYPE_UINT16 */
+	lhz r3, 56+2(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_SINT16 */
+	lha r3, 56+2(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_UINT32 */
+	lwz r3, 56+0(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_SINT32 */
+	lwz r3, 56+0(r1)
+	mtlr r0
+	addi r1, r1, 176
+	blr
+
+/* case FFI_TYPE_UINT64 */
+	lwz r3, 56+0(r1)
+	mtlr r0
+	lwz r4, 56+4(r1)
+	b L..finish
+
+/* case FFI_TYPE_SINT64 */
+	lwz r3, 56+0(r1)
+	mtlr r0
+	lwz r4, 56+4(r1)
+	b L..finish
+
+/* case FFI_TYPE_STRUCT */
+	mtlr r0
+	addi r1, r1, 176
+	blr
+	nop
+
+/* case FFI_TYPE_POINTER */
+	lwz r3, 56+0(r1)
+	mtlr r0
+L..finish:
+	addi r1, r1, 176
+	blr
+#endif
 /* END(ffi_closure_ASM) */