ffi 1.1.2 → 1.17.0

data/ext/ffi_c/libffi/src/mips/n32.S

@@ -37,35 +37,43 @@
 #define flags	 a3
 #define raddr    a4
 #define fn       a5
+#define closure  a6
 
-#define SIZEOF_FRAME	( 8 * FFI_SIZEOF_ARG )
+/* Note: to keep stack 16 byte aligned we need even number slots 
+   used 9 slots here
+*/
+#define SIZEOF_FRAME	( 10 * FFI_SIZEOF_ARG )
 
 #ifdef __GNUC__
 	.abicalls
+#endif
+#if !defined(__mips_isa_rev) || (__mips_isa_rev<6)
+	.set mips4
 #endif
 	.text
 	.align	2
 	.globl	ffi_call_N32
 	.ent	ffi_call_N32
 ffi_call_N32:	
-.LFB3:
+.LFB0:
 	.frame	$fp, SIZEOF_FRAME, ra
 	.mask	0xc0000000,-FFI_SIZEOF_ARG
 	.fmask	0x00000000,0
 
 	# Prologue
 	SUBU	$sp, SIZEOF_FRAME			# Frame size
-.LCFI0:
+.LCFI00:
 	REG_S	$fp, SIZEOF_FRAME - 2*FFI_SIZEOF_ARG($sp)	# Save frame pointer
 	REG_S	ra, SIZEOF_FRAME - 1*FFI_SIZEOF_ARG($sp)	# Save return address
-.LCFI1:
+.LCFI01:
 	move	$fp, $sp
-.LCFI3:
+.LCFI02:
 	move	t9, callback	# callback function pointer
 	REG_S	bytes, 2*FFI_SIZEOF_ARG($fp) # bytes
 	REG_S	flags, 3*FFI_SIZEOF_ARG($fp) # flags
 	REG_S	raddr, 4*FFI_SIZEOF_ARG($fp) # raddr
 	REG_S	fn,    5*FFI_SIZEOF_ARG($fp) # fn
+	REG_S	closure, 6*FFI_SIZEOF_ARG($fp) # closure
 
 	# Allocate at least 4 words in the argstack
 	move	v0, bytes
@@ -106,11 +114,29 @@ loadregs:
 
 	REG_L	t6, 3*FFI_SIZEOF_ARG($fp)  # load the flags word into t6.
 
-	and	t4, t6, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg1_floatp
+	# when retval is _Complex long double, $f12/$a0, $f13/$a1 will be skipped
+	# no idea why, but gcc does it.
+	SRL	t4, t6, 8*FFI_FLAG_BITS
+	move	t8, t6
+	bne	t4, FFI_TYPE_COMPLEX_LDLD, loadregs1
+
+	SLL	t8, t6, 2*FFI_FLAG_BITS
+
+
+loadregs1:
+#ifdef __mips_soft_float
 	REG_L	a0, 0*FFI_SIZEOF_ARG(t9)
-	b	arg1_next
-arg1_floatp:	
+	REG_L	a1, 1*FFI_SIZEOF_ARG(t9)
+	REG_L	a2, 2*FFI_SIZEOF_ARG(t9)
+	REG_L	a3, 3*FFI_SIZEOF_ARG(t9)
+	REG_L	a4, 4*FFI_SIZEOF_ARG(t9)
+	REG_L	a5, 5*FFI_SIZEOF_ARG(t9)
+	REG_L	a6, 6*FFI_SIZEOF_ARG(t9)
+	REG_L	a7, 7*FFI_SIZEOF_ARG(t9)
+#else
+	and	t4, t8, ((1<<FFI_FLAG_BITS)-1)
+	REG_L	a0, 0*FFI_SIZEOF_ARG(t9)
+	beqz	t4, arg1_next
 	bne	t4, FFI_TYPE_FLOAT, arg1_doublep
 	l.s	$f12, 0*FFI_SIZEOF_ARG(t9)
 	b	arg1_next
@@ -118,12 +144,10 @@ arg1_doublep:
 	l.d	$f12, 0*FFI_SIZEOF_ARG(t9)
 arg1_next:	
 	
-	SRL	t4, t6, 1*FFI_FLAG_BITS
+	SRL	t4, t8, 1*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg2_floatp
 	REG_L	a1, 1*FFI_SIZEOF_ARG(t9)
-	b	arg2_next
-arg2_floatp:
+	beqz	t4, arg2_next
 	bne	t4, FFI_TYPE_FLOAT, arg2_doublep
 	l.s	$f13, 1*FFI_SIZEOF_ARG(t9)	
 	b	arg2_next
@@ -131,12 +155,10 @@ arg2_doublep:
 	l.d	$f13, 1*FFI_SIZEOF_ARG(t9)	
 arg2_next:	
 	
-	SRL	t4, t6, 2*FFI_FLAG_BITS
+	SRL	t4, t8, 2*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg3_floatp
 	REG_L	a2, 2*FFI_SIZEOF_ARG(t9)
-	b	arg3_next
-arg3_floatp:
+	beqz	t4, arg3_next
 	bne	t4, FFI_TYPE_FLOAT, arg3_doublep
 	l.s	$f14, 2*FFI_SIZEOF_ARG(t9)	
 	b	arg3_next
@@ -144,12 +166,10 @@ arg3_doublep:
 	l.d	$f14, 2*FFI_SIZEOF_ARG(t9)	
 arg3_next:	
 	
-	SRL	t4, t6, 3*FFI_FLAG_BITS
+	SRL	t4, t8, 3*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg4_floatp
 	REG_L	a3, 3*FFI_SIZEOF_ARG(t9)
-	b	arg4_next
-arg4_floatp:
+	beqz	t4, arg4_next
 	bne	t4, FFI_TYPE_FLOAT, arg4_doublep
 	l.s	$f15, 3*FFI_SIZEOF_ARG(t9)	
 	b	arg4_next
@@ -157,12 +177,10 @@ arg4_doublep:
 	l.d	$f15, 3*FFI_SIZEOF_ARG(t9)	
 arg4_next:	
 	
-	SRL	t4, t6, 4*FFI_FLAG_BITS
+	SRL	t4, t8, 4*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg5_floatp
 	REG_L	a4, 4*FFI_SIZEOF_ARG(t9)
-	b	arg5_next
-arg5_floatp:
+	beqz	t4, arg5_next
 	bne	t4, FFI_TYPE_FLOAT, arg5_doublep
 	l.s	$f16, 4*FFI_SIZEOF_ARG(t9)	
 	b	arg5_next
@@ -170,12 +188,10 @@ arg5_doublep:
 	l.d	$f16, 4*FFI_SIZEOF_ARG(t9)	
 arg5_next:	
 	
-	SRL	t4, t6, 5*FFI_FLAG_BITS
+	SRL	t4, t8, 5*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg6_floatp
 	REG_L	a5, 5*FFI_SIZEOF_ARG(t9)
-	b	arg6_next
-arg6_floatp:
+	beqz	t4, arg6_next
 	bne	t4, FFI_TYPE_FLOAT, arg6_doublep
 	l.s	$f17, 5*FFI_SIZEOF_ARG(t9)	
 	b	arg6_next
@@ -183,12 +199,10 @@ arg6_doublep:
 	l.d	$f17, 5*FFI_SIZEOF_ARG(t9)	
 arg6_next:	
 	
-	SRL	t4, t6, 6*FFI_FLAG_BITS
+	SRL	t4, t8, 6*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg7_floatp
 	REG_L	a6, 6*FFI_SIZEOF_ARG(t9)
-	b	arg7_next
-arg7_floatp:
+	beqz	t4, arg7_next
 	bne	t4, FFI_TYPE_FLOAT, arg7_doublep
 	l.s	$f18, 6*FFI_SIZEOF_ARG(t9)	
 	b	arg7_next
@@ -196,23 +210,25 @@ arg7_doublep:
 	l.d	$f18, 6*FFI_SIZEOF_ARG(t9)	
 arg7_next:	
 	
-	SRL	t4, t6, 7*FFI_FLAG_BITS
+	SRL	t4, t8, 7*FFI_FLAG_BITS
 	and	t4, ((1<<FFI_FLAG_BITS)-1)
-	bnez	t4, arg8_floatp
 	REG_L	a7, 7*FFI_SIZEOF_ARG(t9)
-	b	arg8_next
-arg8_floatp:
+	beqz	t4, arg8_next
 	bne	t4, FFI_TYPE_FLOAT, arg8_doublep
  	l.s	$f19, 7*FFI_SIZEOF_ARG(t9)	
 	b	arg8_next
 arg8_doublep:	
  	l.d	$f19, 7*FFI_SIZEOF_ARG(t9)	
 arg8_next:	
+#endif
 
-callit:		
+callit:
 	# Load the function pointer
 	REG_L	t9, 5*FFI_SIZEOF_ARG($fp)
 
+	# install the static chain(t7=$15)
+	REG_L	t7, 6*FFI_SIZEOF_ARG($fp)
+
 	# If the return value pointer is NULL, assume no return value.
 	REG_L	t5, 4*FFI_SIZEOF_ARG($fp)
 	beqz	t5, noretval
@@ -220,15 +236,58 @@ callit:
 	# Shift the return type flag over
 	SRL	t6, 8*FFI_FLAG_BITS
 
-	beq	t6, FFI_TYPE_SINT32, retint	
-	bne     t6, FFI_TYPE_INT, retfloat
-retint:
+	bne     t6, FFI_TYPE_UINT64, retsint32
+
+retuint64:
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
-	REG_S	v0, 0(t4)
+	sd	v0, 0(t4)
+	b	epilogue
+
+retsint32:
+	bne	t6, FFI_TYPE_SINT32, retuint16
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	sll	v0, v0, 0
+	sd	v0, 0(t4)
+	b	epilogue
+
+retuint16:
+	bne	t6, FFI_TYPE_UINT16, retsint16
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	andi	v0, v0, 0xffff
+	sd	v0, 0(t4)
+	b	epilogue
+
+retsint16:
+	bne	t6, FFI_TYPE_SINT16, retuint8
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	dsll	v0, v0, 48
+	dsra	v0, v0, 48
+	sd	v0, 0(t4)
+	b	epilogue
+
+retuint8:
+	bne	t6, FFI_TYPE_UINT8, retsint8
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	andi	v0, v0, 0xff
+	sd	v0, 0(t4)
+	b	epilogue
+
+retsint8:
+	bne	t6, FFI_TYPE_SINT8, retfloat
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	sd	v0, 0(t4)
+	dsll	v0, v0, 56
+	dsra	v0, v0, 56
 	b	epilogue
 
 retfloat:
+#ifndef __mips_soft_float
 	bne     t6, FFI_TYPE_FLOAT, retdouble
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
@@ -242,7 +301,7 @@ retdouble:
 	s.d	$f0, 0(t4)
 	b	epilogue
 
-retstruct_d:	
+retstruct_d:
 	bne	t6, FFI_TYPE_STRUCT_D, retstruct_f
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
@@ -256,16 +315,37 @@ retstruct_f:
 	s.s	$f0, 0(t4)
 	b	epilogue
 	
-retstruct_d_d:	
-	bne	t6, FFI_TYPE_STRUCT_DD, retstruct_f_f
+retstruct_d_d:
+	bne	t6, FFI_TYPE_STRUCT_DD, retcomplex_d_d
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
 	s.d	$f0, 0(t4)
 	s.d	$f2, 8(t4)
 	b	epilogue
+
+retcomplex_d_d:
+	bne	t6, FFI_TYPE_COMPLEX_DD, retcomplex_ld_ld
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	s.d	$f0, 0(t4)
+	s.d	$f2, 8(t4)
+	b	epilogue
+
+retcomplex_ld_ld:
+	bne	t6, FFI_TYPE_COMPLEX_LDLD, retstruct_f_f
+	jal	t9
+	b	epilogue
 	
-retstruct_f_f:	
-	bne	t6, FFI_TYPE_STRUCT_FF, retstruct_d_f
+retstruct_f_f:
+	bne	t6, FFI_TYPE_STRUCT_FF, retcomplex_f_f
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	s.s	$f0, 0(t4)
+	s.s	$f2, 4(t4)
+	b	epilogue
+
+retcomplex_f_f:
+	bne	t6, FFI_TYPE_COMPLEX_FF, retstruct_d_f
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
 	s.s	$f0, 0(t4)
@@ -287,6 +367,7 @@ retstruct_f_d:
 	s.s	$f0, 0(t4)
 	s.d	$f2, 8(t4)
 	b	epilogue
+#endif
 
 retstruct_d_soft:
 	bne	t6, FFI_TYPE_STRUCT_D_SOFT, retstruct_f_soft
@@ -307,7 +388,7 @@ retstruct_d_d_soft:
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
 	sd	v0, 0(t4)
-	sd	v1, 8(t4)
+	sd	a0, 8(t4) # not typo, it is a0, I have no idea, gcc does do it
 	b	epilogue
 	
 retstruct_f_f_soft:	
@@ -331,7 +412,7 @@ retstruct_f_d_soft:
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
 	sw	v0, 0(t4)
-	sd	v1, 8(t4)
+	sd	a0, 8(t4) # not typo, it is a0, I have no idea, gcc does do it
 	b	epilogue
 	
 retstruct_small:	
@@ -342,14 +423,37 @@ retstruct_small:
 	b	epilogue
 	
 retstruct_small2:	
-	bne	t6, FFI_TYPE_STRUCT_SMALL2, retstruct
+	bne	t6, FFI_TYPE_STRUCT_SMALL2, retlongdouble_soft
 	jal	t9
 	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
 	REG_S	v0, 0(t4)
 	REG_S	v1, 8(t4)
 	b	epilogue
 	
-retstruct:	
+retlongdouble_soft:
+	bne	t6, FFI_TYPE_LONGDOUBLE, retcomplex_small
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	REG_S	v0, 0(t4)
+	REG_S	a0, 8(t4) # not typo, it is a0, I have no idea, gcc does do it
+	b	epilogue
+
+retcomplex_small:
+	bne	t6, FFI_TYPE_COMPLEX_SMALL, retcomplex_small2
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	REG_S	v0, 0(t4)
+	b	epilogue
+
+retcomplex_small2:
+	bne	t6, FFI_TYPE_COMPLEX_SMALL2, retstruct
+	jal	t9
+	REG_L	t4, 4*FFI_SIZEOF_ARG($fp)
+	REG_S	v0, 0(t4)
+	REG_S	v1, 8(t4)
+	b	epilogue
+
+retstruct:
 noretval:	
 	jal	t9
 	
@@ -361,7 +465,7 @@ epilogue:
 	ADDU	$sp, SIZEOF_FRAME		      # Fix stack pointer
 	j	ra
 
-.LFE3:
+.LFE0:
 	.end	ffi_call_N32
 
 /* ffi_closure_N32. Expects address of the passed-in ffi_closure in t0
@@ -395,7 +499,7 @@ epilogue:
 	 */
 
 #define SIZEOF_FRAME2	(20 * FFI_SIZEOF_ARG)
-	
+
 #define A7_OFF2		(19 * FFI_SIZEOF_ARG)
 #define A6_OFF2		(18 * FFI_SIZEOF_ARG)
 #define A5_OFF2		(17 * FFI_SIZEOF_ARG)
@@ -420,6 +524,46 @@ epilogue:
 #define RA_OFF2		(1  * FFI_SIZEOF_ARG)
 #define GP_OFF2		(0  * FFI_SIZEOF_ARG)
 
+	.align	2
+	.globl	ffi_go_closure_N32
+	.ent	ffi_go_closure_N32
+ffi_go_closure_N32:
+.LFB1:
+	.frame	$sp, SIZEOF_FRAME2, ra
+	.mask	0x90000000,-(SIZEOF_FRAME2 - RA_OFF2)
+	.fmask	0x00000000,0
+	SUBU	$sp, SIZEOF_FRAME2
+.LCFI10:
+	.cpsetup t9, GP_OFF2, ffi_go_closure_N32
+	REG_S	ra, RA_OFF2($sp)	# Save return address
+.LCFI11:
+
+	REG_S	a0, A0_OFF2($sp)
+	REG_S	a1, A1_OFF2($sp)
+	REG_S	a2, A2_OFF2($sp)
+	REG_S	a3, A3_OFF2($sp)
+	REG_S	a4, A4_OFF2($sp)
+	REG_S	a5, A5_OFF2($sp)
+
+	# Call ffi_closure_mips_inner_N32 to do the real work.
+	LA	t9, ffi_closure_mips_inner_N32
+#if _MIPS_SIM==_ABIN32
+	lw	a0, 4($15)   # cif
+	lw	a1, 8($15) # fun
+#else
+	REG_L	a0, 8($15)   # cif
+	REG_L	a1, 16($15) # fun
+#endif
+	move	a2, t7                     # userdata=closure
+	ADDU	a3, $sp, V0_OFF2           # rvalue
+	ADDU	a4, $sp, A0_OFF2           # ar
+	ADDU	a5, $sp, F12_OFF2          # fpr
+
+	b	$do_closure
+
+.LFE1:	
+	.end	ffi_go_closure_N32
+
 	.align	2
 	.globl	ffi_closure_N32
 	.ent	ffi_closure_N32
@@ -429,21 +573,40 @@ ffi_closure_N32:
 	.mask	0x90000000,-(SIZEOF_FRAME2 - RA_OFF2)
 	.fmask	0x00000000,0
 	SUBU	$sp, SIZEOF_FRAME2
-.LCFI5:
+.LCFI20:
 	.cpsetup t9, GP_OFF2, ffi_closure_N32
 	REG_S	ra, RA_OFF2($sp)	# Save return address
-.LCFI6:
-	# Store all possible argument registers. If there are more than
-	# fit in registers, then they were stored on the stack.
+.LCFI21:
 	REG_S	a0, A0_OFF2($sp)
 	REG_S	a1, A1_OFF2($sp)
 	REG_S	a2, A2_OFF2($sp)
 	REG_S	a3, A3_OFF2($sp)
 	REG_S	a4, A4_OFF2($sp)
 	REG_S	a5, A5_OFF2($sp)
+
+	# Call ffi_closure_mips_inner_N32 to do the real work.
+	LA	t9, ffi_closure_mips_inner_N32
+#if _MIPS_SIM==_ABIN32
+	lw	a0, 20($12)   # cif
+	lw	a1, 24($12)   # fun
+	lw	a2, 28($12) # user_data
+#else
+	REG_L	a0, 56($12)   # cif
+	REG_L	a1, 64($12)   # fun
+	REG_L	a2, 72($12) # user_data
+#endif
+	ADDU	a3, $sp, V0_OFF2
+	# FIXME: a4 does work, while if ret is _Complex long double, it will overwrite Fn_OFF2
+	ADDU	a4, $sp, A0_OFF2
+	ADDU	a5, $sp, F12_OFF2
+
+$do_closure:
+	# Store all possible argument registers. If there are more than
+	# fit in registers, then they were stored on the stack.
 	REG_S	a6, A6_OFF2($sp)
 	REG_S	a7, A7_OFF2($sp)
 
+#ifndef __mips_soft_float
 	# Store all possible float/double registers.
 	s.d	$f12, F12_OFF2($sp)
 	s.d	$f13, F13_OFF2($sp)
@@ -453,26 +616,43 @@ ffi_closure_N32:
 	s.d	$f17, F17_OFF2($sp)
 	s.d	$f18, F18_OFF2($sp)
 	s.d	$f19, F19_OFF2($sp)
+#endif
 
-	# Call ffi_closure_mips_inner_N32 to do the real work.
-	LA	t9, ffi_closure_mips_inner_N32
-	move	a0, $12	 # Pointer to the ffi_closure
-	ADDU	a1, $sp, V0_OFF2
-	ADDU	a2, $sp, A0_OFF2
-	ADDU	a3, $sp, F12_OFF2
 	jalr	t9
 
+cls_retuint64:
 	# Return flags are in v0
-	bne     v0, FFI_TYPE_SINT32, cls_retint
+	bne     v0, FFI_TYPE_UINT64, cls_retsint32
+	ld	v0, V0_OFF2($sp)
+	b	cls_epilogue
+
+cls_retsint32:
+	bne     v0, FFI_TYPE_SINT32, cls_retsint16
 	lw	v0, V0_OFF2($sp)
 	b	cls_epilogue
 
-cls_retint:
-	bne     v0, FFI_TYPE_INT, cls_retfloat
-	REG_L	v0, V0_OFF2($sp)
+cls_retsint16:
+	bne     v0, FFI_TYPE_SINT16, cls_retuint16
+	lh	v0, V0_OFF2($sp)
+	b	cls_epilogue
+
+cls_retuint16:
+	bne     v0, FFI_TYPE_UINT16, cls_retsint8
+	lhu	v0, V0_OFF2($sp)
+	b	cls_epilogue
+
+cls_retsint8:
+	bne     v0, FFI_TYPE_SINT8, cls_retuint8
+	lb	v0, V0_OFF2($sp)
+	b	cls_epilogue
+
+cls_retuint8:
+	bne     v0, FFI_TYPE_UINT8, cls_retfloat
+	lbu	v0, V0_OFF2($sp)
 	b	cls_epilogue
 
 cls_retfloat:
+#ifndef __mips_soft_float
 	bne     v0, FFI_TYPE_FLOAT, cls_retdouble
 	l.s	$f0, V0_OFF2($sp)
 	b	cls_epilogue
@@ -493,17 +673,29 @@ cls_retstruct_f:
 	b	cls_epilogue
 	
 cls_retstruct_d_d:	
-	bne	v0, FFI_TYPE_STRUCT_DD, cls_retstruct_f_f
+	bne	v0, FFI_TYPE_STRUCT_DD, cls_retcomplex_d_d
+	l.d	$f0, V0_OFF2($sp)
+	l.d	$f2, V1_OFF2($sp)
+	b	cls_epilogue
+
+cls_retcomplex_d_d:
+	bne	v0, FFI_TYPE_COMPLEX_DD, cls_retcomplex_f_f
 	l.d	$f0, V0_OFF2($sp)
 	l.d	$f2, V1_OFF2($sp)
 	b	cls_epilogue
 	
-cls_retstruct_f_f:	
-	bne	v0, FFI_TYPE_STRUCT_FF, cls_retstruct_d_f
+cls_retstruct_f_f:
+	bne	v0, FFI_TYPE_STRUCT_FF, cls_retcomplex_f_f
 	l.s	$f0, V0_OFF2($sp)
 	l.s	$f2, V1_OFF2($sp)
 	b	cls_epilogue
 	
+cls_retcomplex_f_f:
+	bne	v0, FFI_TYPE_COMPLEX_FF, cls_retstruct_d_f
+	l.s	$f0, V0_OFF2($sp)
+	l.s	$f2, (V0_OFF2+4)($sp)
+	b	cls_epilogue
+
 cls_retstruct_d_f:	
 	bne	v0, FFI_TYPE_STRUCT_DF, cls_retstruct_f_d
 	l.d	$f0, V0_OFF2($sp)
@@ -511,11 +703,31 @@ cls_retstruct_d_f:
 	b	cls_epilogue
 	
 cls_retstruct_f_d:	
-	bne	v0, FFI_TYPE_STRUCT_FD, cls_retstruct_small2
+	bne	v0, FFI_TYPE_STRUCT_FD, cls_retcomplex_ld_ld
 	l.s	$f0, V0_OFF2($sp)
 	l.d	$f2, V1_OFF2($sp)
 	b	cls_epilogue
-	
+#else
+cls_longdouble_soft:
+	bne	v0, FFI_TYPE_LONGDOUBLE, cls_retcomplex_ld_ld
+	REG_L	v0, V0_OFF2($sp)
+	REG_L	a0, V1_OFF2($sp) # not typo, it is a0, I have no idea, gcc does do it
+	b	cls_epilogue
+#endif
+
+cls_retcomplex_ld_ld:
+	bne	v0, FFI_TYPE_COMPLEX_LDLD, cls_retstruct_small2
+	REG_L	t8, A0_OFF2($sp)
+	REG_L	t9, 16($sp)
+	REG_S	t9, 0(t8)
+	REG_L	t9, 24($sp)
+	REG_S	t9, 8(t8)
+	REG_L	t9, 32($sp)
+	REG_S	t9, 16(t8)
+	REG_L	t9, 40($sp)
+	REG_S	t9, 24(t8)
+	b	cls_epilogue
+
 cls_retstruct_small2:	
 	REG_L	v0, V0_OFF2($sp)
 	REG_L	v1, V1_OFF2($sp)
@@ -530,7 +742,7 @@ cls_epilogue:
 	.end	ffi_closure_N32
 
 #ifdef __GNUC__
-        .section        .eh_frame,"aw",@progbits
+        .section        .eh_frame,EH_FRAME_FLAGS,@progbits
 .Lframe1:
         .4byte  .LECIE1-.LSCIE1		# length
 .LSCIE1:
@@ -546,46 +758,66 @@ cls_epilogue:
         .align  EH_FRAME_ALIGN
 .LECIE1:
 
-.LSFDE1:
-        .4byte  .LEFDE1-.LASFDE1	# length.
-.LASFDE1:
-        .4byte  .LASFDE1-.Lframe1	# CIE_pointer.
-        FDE_ADDR_BYTES  .LFB3		# initial_location.
-        FDE_ADDR_BYTES  .LFE3-.LFB3	# address_range.
+.LSFDE0:
+        .4byte  .LEFDE0-.LASFDE0	# length.
+.LASFDE0:
+        .4byte  .LASFDE0-.Lframe1	# CIE_pointer.
+        FDE_ADDR_BYTES  .LFB0		# initial_location.
+        FDE_ADDR_BYTES  .LFE0-.LFB0	# address_range.
         .byte   0x4			# DW_CFA_advance_loc4
-        .4byte  .LCFI0-.LFB3		# to .LCFI0
+        .4byte  .LCFI00-.LFB0		# to .LCFI00
         .byte   0xe			# DW_CFA_def_cfa_offset
         .uleb128 SIZEOF_FRAME		# adjust stack.by SIZEOF_FRAME
         .byte   0x4			# DW_CFA_advance_loc4
-        .4byte  .LCFI1-.LCFI0		# to .LCFI1
+        .4byte  .LCFI01-.LCFI00		# to .LCFI01
         .byte   0x9e			# DW_CFA_offset of $fp
         .uleb128 2*FFI_SIZEOF_ARG/4	# 
         .byte   0x9f			# DW_CFA_offset of ra
         .uleb128 1*FFI_SIZEOF_ARG/4	# 
         .byte   0x4			# DW_CFA_advance_loc4
-        .4byte  .LCFI3-.LCFI1		# to .LCFI3
+        .4byte  .LCFI02-.LCFI01		# to .LCFI02
         .byte   0xd			# DW_CFA_def_cfa_register
         .uleb128 0x1e			# in $fp
         .align  EH_FRAME_ALIGN
+.LEFDE0:
+
+.LSFDE1:
+	.4byte	.LEFDE1-.LASFDE1	# length
+.LASFDE1:
+	.4byte	.LASFDE1-.Lframe1	# CIE_pointer.
+	FDE_ADDR_BYTES	.LFB1		# initial_location.
+	FDE_ADDR_BYTES	.LFE1-.LFB1	# address_range.
+	.byte	0x4			# DW_CFA_advance_loc4
+	.4byte	.LCFI10-.LFB1		# to .LCFI10
+	.byte	0xe			# DW_CFA_def_cfa_offset
+	.uleb128 SIZEOF_FRAME2		# adjust stack.by SIZEOF_FRAME
+	.byte	0x4			# DW_CFA_advance_loc4
+	.4byte	.LCFI11-.LCFI10		# to .LCFI11
+	.byte	0x9c			# DW_CFA_offset of $gp ($28)
+	.uleb128 (SIZEOF_FRAME2 - GP_OFF2)/4
+	.byte	0x9f			# DW_CFA_offset of ra ($31)
+	.uleb128 (SIZEOF_FRAME2 - RA_OFF2)/4
+	.align	EH_FRAME_ALIGN
 .LEFDE1:
-.LSFDE3:
-	.4byte	.LEFDE3-.LASFDE3	# length
-.LASFDE3:
-	.4byte	.LASFDE3-.Lframe1	# CIE_pointer.
+
+.LSFDE2:
+	.4byte	.LEFDE2-.LASFDE2	# length
+.LASFDE2:
+	.4byte	.LASFDE2-.Lframe1	# CIE_pointer.
 	FDE_ADDR_BYTES	.LFB2		# initial_location.
 	FDE_ADDR_BYTES	.LFE2-.LFB2	# address_range.
 	.byte	0x4			# DW_CFA_advance_loc4
-	.4byte	.LCFI5-.LFB2		# to .LCFI5
+	.4byte	.LCFI20-.LFB2		# to .LCFI20
 	.byte	0xe			# DW_CFA_def_cfa_offset
 	.uleb128 SIZEOF_FRAME2		# adjust stack.by SIZEOF_FRAME
 	.byte	0x4			# DW_CFA_advance_loc4
-	.4byte	.LCFI6-.LCFI5		# to .LCFI6
+	.4byte	.LCFI21-.LCFI20		# to .LCFI21
 	.byte	0x9c			# DW_CFA_offset of $gp ($28)
 	.uleb128 (SIZEOF_FRAME2 - GP_OFF2)/4
 	.byte	0x9f			# DW_CFA_offset of ra ($31)
 	.uleb128 (SIZEOF_FRAME2 - RA_OFF2)/4
 	.align	EH_FRAME_ALIGN
-.LEFDE3:
+.LEFDE2:
 #endif /* __GNUC__ */	
 	
 #endif