ffi 1.10.0 → 1.15.5

data/ext/ffi_c/libffi/src/powerpc/linux64.S

@@ -36,8 +36,10 @@
 	.cfi_startproc
 # if _CALL_ELF == 2
 ffi_call_LINUX64:
+#  ifndef __PCREL__
 	addis	%r2, %r12, .TOC.-ffi_call_LINUX64@ha
 	addi	%r2, %r2, .TOC.-ffi_call_LINUX64@l
+#  endif
 	.localentry ffi_call_LINUX64, . - ffi_call_LINUX64
 # else
 	.section	".opd","aw"
@@ -89,9 +91,15 @@ ffi_call_LINUX64:
 	/* Call ffi_prep_args64.  */
 	mr	%r4, %r1
 # if defined _CALL_LINUX || _CALL_ELF == 2
+#  ifdef __PCREL__
+	bl	ffi_prep_args64@notoc
+#  else
 	bl	ffi_prep_args64
+	nop
+#  endif
 # else
 	bl	.ffi_prep_args64
+	nop
 # endif
 
 # if _CALL_ELF == 2
@@ -101,40 +109,70 @@ ffi_call_LINUX64:
 	ld	%r2, 8(%r29)
 # endif
 	/* Now do the call.  */
-	/* Set up cr1 with bits 4-7 of the flags.  */
-	mtcrf	0x40, %r31
+	/* Set up cr1 with bits 3-7 of the flags.  */
+	mtcrf	0xc0, %r31
 
 	/* Get the address to call into CTR.  */
 	mtctr	%r12
 	/* Load all those argument registers.  */
-	ld	%r3, -32-(8*8)(%r28)
-	ld	%r4, -32-(7*8)(%r28)
-	ld	%r5, -32-(6*8)(%r28)
-	ld	%r6, -32-(5*8)(%r28)
+	addi	%r29, %r28, -32-(8*8)
+	ld	%r3,  (0*8)(%r29)
+	ld	%r4,  (1*8)(%r29)
+	ld	%r5,  (2*8)(%r29)
+	ld	%r6,  (3*8)(%r29)
 	bf-	5, 1f
-	ld	%r7, -32-(4*8)(%r28)
-	ld	%r8, -32-(3*8)(%r28)
-	ld	%r9, -32-(2*8)(%r28)
-	ld	%r10, -32-(1*8)(%r28)
+	ld	%r7,  (4*8)(%r29)
+	ld	%r8,  (5*8)(%r29)
+	ld	%r9,  (6*8)(%r29)
+	ld	%r10, (7*8)(%r29)
 1:
 
 	/* Load all the FP registers.  */
 	bf-	6, 2f
-	lfd	%f1, -32-(21*8)(%r28)
-	lfd	%f2, -32-(20*8)(%r28)
-	lfd	%f3, -32-(19*8)(%r28)
-	lfd	%f4, -32-(18*8)(%r28)
-	lfd	%f5, -32-(17*8)(%r28)
-	lfd	%f6, -32-(16*8)(%r28)
-	lfd	%f7, -32-(15*8)(%r28)
-	lfd	%f8, -32-(14*8)(%r28)
-	lfd	%f9, -32-(13*8)(%r28)
-	lfd	%f10, -32-(12*8)(%r28)
-	lfd	%f11, -32-(11*8)(%r28)
-	lfd	%f12, -32-(10*8)(%r28)
-	lfd	%f13, -32-(9*8)(%r28)
+	addi	%r29, %r29, -(14*8)
+	lfd	%f1,  ( 1*8)(%r29)
+	lfd	%f2,  ( 2*8)(%r29)
+	lfd	%f3,  ( 3*8)(%r29)
+	lfd	%f4,  ( 4*8)(%r29)
+	lfd	%f5,  ( 5*8)(%r29)
+	lfd	%f6,  ( 6*8)(%r29)
+	lfd	%f7,  ( 7*8)(%r29)
+	lfd	%f8,  ( 8*8)(%r29)
+	lfd	%f9,  ( 9*8)(%r29)
+	lfd	%f10, (10*8)(%r29)
+	lfd	%f11, (11*8)(%r29)
+	lfd	%f12, (12*8)(%r29)
+	lfd	%f13, (13*8)(%r29)
 2:
 
+	/* Load all the vector registers.  */
+	bf-	3, 3f
+	addi	%r29, %r29, -16
+	lvx	%v13, 0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v12, 0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v11, 0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v10, 0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v9,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v8,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v7,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v6,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v5,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v4,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v3,  0, %r29
+	addi	%r29, %r29, -16
+	lvx	%v2,  0, %r29
+3:
+
 	/* Make the call.  */
 	ld	%r11, 8(%r28)
 	bctrl
@@ -152,6 +190,7 @@ ffi_call_LINUX64:
 	bt	31, .Lstruct_return_value
 	bt	30, .Ldone_return_value
 	bt	29, .Lfp_return_value
+	bt	28, .Lvec_return_value
 	std	%r3, 0(%r30)
 	/* Fall through...  */
 
@@ -167,12 +206,16 @@ ffi_call_LINUX64:
 	ld	%r31, -8(%r1)
 	blr
 
+.Lvec_return_value:
+	stvx	%v2, 0, %r30
+	b	.Ldone_return_value
+
 .Lfp_return_value:
 	.cfi_def_cfa_register 28
-	bf	28, .Lfloat_return_value
-	stfd	%f1, 0(%r30)
 	mtcrf	0x02, %r31 /* cr6  */
-	bf	27, .Ldone_return_value
+	bf	27, .Lfloat_return_value
+	stfd	%f1, 0(%r30)
+	bf	26, .Ldone_return_value
 	stfd	%f2, 8(%r30)
 	b	.Ldone_return_value
 .Lfloat_return_value:
@@ -180,8 +223,9 @@ ffi_call_LINUX64:
 	b	.Ldone_return_value
 
 .Lstruct_return_value:
-	bf	29, .Lsmall_struct
-	bf	28, .Lfloat_homog_return_value
+	bf	29, .Lvec_homog_or_small_struct
+	mtcrf	0x02, %r31 /* cr6  */
+	bf	27, .Lfloat_homog_return_value
 	stfd	%f1, 0(%r30)
 	stfd	%f2, 8(%r30)
 	stfd	%f3, 16(%r30)
@@ -203,6 +247,25 @@ ffi_call_LINUX64:
 	stfs	%f8, 28(%r30)
 	b	.Ldone_return_value
 
+.Lvec_homog_or_small_struct:
+	bf	28, .Lsmall_struct
+	stvx	%v2, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v3, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v4, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v5, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v6, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v7, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v8, 0, %r30
+	addi	%r30, %r30, 16
+	stvx	%v9, 0, %r30
+	b	.Ldone_return_value
+
 .Lsmall_struct:
 	std	%r3, 0(%r30)
 	std	%r4, 8(%r30)

data/ext/ffi_c/libffi/src/powerpc/linux64_closure.S

@@ -37,8 +37,10 @@
 	.cfi_startproc
 # if _CALL_ELF == 2
 ffi_closure_LINUX64:
+#  ifndef __PCREL__
 	addis	%r2, %r12, .TOC.-ffi_closure_LINUX64@ha
 	addi	%r2, %r2, .TOC.-ffi_closure_LINUX64@l
+#  endif
 	.localentry ffi_closure_LINUX64, . - ffi_closure_LINUX64
 # else
 	.section        ".opd","aw"
@@ -61,9 +63,15 @@ ffi_closure_LINUX64:
 # endif
 
 # if _CALL_ELF == 2
-#  32 byte special reg save area + 64 byte parm save area
-#  + 64 byte retval area + 13*8 fpr save area + round to 16
-#  define STACKFRAME 272
+#  ifdef __VEC__
+#   32 byte special reg save area + 64 byte parm save area
+#   + 128 byte retval area + 13*8 fpr save area + 12*16 vec save area + round to 16
+#   define STACKFRAME 528
+#  else
+#   32 byte special reg save area + 64 byte parm save area
+#   + 64 byte retval area + 13*8 fpr save area + round to 16
+#   define STACKFRAME 272
+#  endif
 #  define PARMSAVE 32
 #  define RETVAL PARMSAVE+64
 # else
@@ -143,9 +151,38 @@ ffi_closure_LINUX64:
 	stfd	%f12, -104+(11*8)(%r1)
 	stfd	%f13, -104+(12*8)(%r1)
 
-	# load up the pointer to the saved fpr registers */
+	# load up the pointer to the saved fpr registers
 	addi	%r8, %r1, -104
 
+# ifdef __VEC__
+	# load up the pointer to the saved vector registers
+	# 8 bytes padding for 16-byte alignment at -112(%r1)
+	addi	%r9, %r8, -24
+	stvx	%v13, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v12, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v11, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v10, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v9, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v8, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v7, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v6, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v5, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v4, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v3, 0, %r9
+	addi	%r9, %r9, -16
+	stvx	%v2, 0, %r9
+# endif
+
 	# load up the pointer to the result storage
 	addi	%r6, %r1, -STACKFRAME+RETVAL
 
@@ -155,11 +192,19 @@ ffi_closure_LINUX64:
 
 	# make the call
 # if defined _CALL_LINUX || _CALL_ELF == 2
+#  ifdef __PCREL__
+	bl ffi_closure_helper_LINUX64@notoc
+.Lret:
+#  else
 	bl ffi_closure_helper_LINUX64
+.Lret:
+	nop
+#  endif
 # else
 	bl .ffi_closure_helper_LINUX64
-# endif
 .Lret:
+	nop
+# endif
 
 	# now r3 contains the return type
 	# so use it to look up in a table
@@ -313,6 +358,16 @@ ffi_closure_LINUX64:
 	.cfi_def_cfa_offset 0
 	blr
 	.cfi_def_cfa_offset STACKFRAME
+# case FFI_V2_TYPE_VECTOR
+	addi %r3, %r1, RETVAL
+	lvx %v2, 0, %r3
+	mtlr %r0
+	b .Lfinish
+# case FFI_V2_TYPE_VECTOR_HOMOG
+	addi %r3, %r1, RETVAL
+	lvx %v2, 0, %r3
+	addi %r3, %r3, 16
+	b .Lmorevector
 # case FFI_V2_TYPE_FLOAT_HOMOG
 	lfs %f1, RETVAL+0(%r1)
 	lfs %f2, RETVAL+4(%r1)
@@ -332,6 +387,25 @@ ffi_closure_LINUX64:
 	.cfi_def_cfa_offset 0
 	blr
 	.cfi_def_cfa_offset STACKFRAME
+.Lmorevector:
+	lvx %v3, 0, %r3
+	addi %r3, %r3, 16
+	lvx %v4, 0, %r3
+	addi %r3, %r3, 16
+	lvx %v5, 0, %r3
+	mtlr %r0
+	addi %r3, %r3, 16
+	lvx %v6, 0, %r3
+	addi %r3, %r3, 16
+	lvx %v7, 0, %r3
+	addi %r3, %r3, 16
+	lvx %v8, 0, %r3
+	addi %r3, %r3, 16
+	lvx %v9, 0, %r3
+	addi %r1, %r1, STACKFRAME
+	.cfi_def_cfa_offset 0
+	blr
+	.cfi_def_cfa_offset STACKFRAME
 .Lmorefloat:
 	lfs %f4, RETVAL+12(%r1)
 	mtlr %r0
@@ -396,8 +470,10 @@ ffi_closure_LINUX64:
 	.cfi_startproc
 # if _CALL_ELF == 2
 ffi_go_closure_linux64:
+#  ifndef __PCREL__
 	addis	%r2, %r12, .TOC.-ffi_go_closure_linux64@ha
 	addi	%r2, %r2, .TOC.-ffi_go_closure_linux64@l
+#  endif
 	.localentry ffi_go_closure_linux64, . - ffi_go_closure_linux64
 # else
 	.section        ".opd","aw"

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

@@ -104,17 +104,16 @@ ENTRY(ffi_call_SYSV)
 	bctrl
 
 	/* Now, deal with the return value.  */
-	mtcrf	0x01,%r31 /* cr7  */
+	mtcrf	0x03,%r31 /* cr6-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)
+	bf+	27,L(done_return_value)
 	stw	%r4,4(%r30)
-	mtcrf	0x02,%r31 /* cr6  */
-	bf	27,L(done_return_value)
+	bf	26,L(done_return_value)
 	stw     %r5,8(%r30)
 	stw	%r6,12(%r30)
 	/* Fall through...  */
@@ -145,10 +144,9 @@ L(done_return_value):
 #ifndef __NO_FPRS__
 L(fp_return_value):
 	.cfi_restore_state
-	bf	28,L(float_return_value)
+	bf	27,L(float_return_value)
 	stfd	%f1,0(%r30)
-	mtcrf   0x02,%r31 /* cr6  */
-	bf	27,L(done_return_value)
+	bf	26,L(done_return_value)
 	stfd	%f2,8(%r30)
 	b	L(done_return_value)
 L(float_return_value):

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

@@ -129,7 +129,9 @@ ffi_status FFI_HIDDEN ffi_prep_cif_core(ffi_cif *cif, ffi_abi abi,
   cif->rtype = rtype;
 
   cif->flags = 0;
-
+#if (defined(_M_ARM64) || defined(__aarch64__)) && defined(_WIN32)
+  cif->is_variadic = isvariadic;
+#endif
 #if HAVE_LONG_DOUBLE_VARIANT
   ffi_prep_types (abi);
 #endif
@@ -199,7 +201,7 @@ ffi_status FFI_HIDDEN ffi_prep_cif_core(ffi_cif *cif, ffi_abi abi,
 	    bytes = 6*4;
 #endif
 
-	  bytes += STACK_ARG_SIZE((*ptr)->size);
+	  bytes += (unsigned int)STACK_ARG_SIZE((*ptr)->size);
 	}
 #endif
     }

data/ext/ffi_c/libffi/src/riscv/ffi.c

@@ -324,9 +324,12 @@ ffi_status ffi_prep_cif_machdep_var(ffi_cif *cif, unsigned int nfixedargs, unsig
 }
 
 /* Low level routine for calling functions */
-extern void ffi_call_asm(void *stack, struct call_context *regs, void (*fn)(void)) FFI_HIDDEN;
+extern void ffi_call_asm (void *stack, struct call_context *regs,
+			  void (*fn) (void), void *closure) FFI_HIDDEN;
 
-void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
+static void
+ffi_call_int (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue,
+	      void *closure)
 {
     /* this is a conservative estimate, assuming a complex return value and
        that all remaining arguments are long long / __int128 */
@@ -366,13 +369,26 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
     for (i = 0; i < cif->nargs; i++)
         marshal(&cb, cif->arg_types[i], i >= cif->riscv_nfixedargs, avalue[i]);
 
-    ffi_call_asm((void*)alloc_base, cb.aregs, fn);
+    ffi_call_asm ((void *) alloc_base, cb.aregs, fn, closure);
 
     cb.used_float = cb.used_integer = 0;
     if (!return_by_ref && rvalue)
         unmarshal(&cb, cif->rtype, 0, rvalue);
 }
 
+void
+ffi_call (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue)
+{
+  ffi_call_int(cif, fn, rvalue, avalue, NULL);
+}
+
+void
+ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue,
+	     void **avalue, void *closure)
+{
+  ffi_call_int(cif, fn, rvalue, avalue, closure);
+}
+
 extern void ffi_closure_asm(void) FFI_HIDDEN;
 
 ffi_status ffi_prep_closure_loc(ffi_closure *closure, ffi_cif *cif, void (*fun)(ffi_cif*,void*,void**,void*), void *user_data, void *codeloc)
@@ -406,11 +422,31 @@ ffi_status ffi_prep_closure_loc(ffi_closure *closure, ffi_cif *cif, void (*fun)(
     return FFI_OK;
 }
 
+extern void ffi_go_closure_asm (void) FFI_HIDDEN;
+
+ffi_status
+ffi_prep_go_closure (ffi_go_closure *closure, ffi_cif *cif,
+		     void (*fun) (ffi_cif *, void *, void **, void *))
+{
+  if (cif->abi <= FFI_FIRST_ABI || cif->abi >= FFI_LAST_ABI)
+    return FFI_BAD_ABI;
+
+  closure->tramp = (void *) ffi_go_closure_asm;
+  closure->cif = cif;
+  closure->fun = fun;
+
+  return FFI_OK;
+}
+
 /* Called by the assembly code with aregs pointing to saved argument registers
    and stack pointing to the stacked arguments.  Return values passed in
    registers will be reloaded from aregs. */
-void FFI_HIDDEN ffi_closure_inner(size_t *stack, call_context *aregs, ffi_closure *closure) {
-    ffi_cif *cif = closure->cif;
+void FFI_HIDDEN
+ffi_closure_inner (ffi_cif *cif,
+		   void (*fun) (ffi_cif *, void *, void **, void *),
+		   void *user_data,
+		   size_t *stack, call_context *aregs)
+{
     void **avalue = alloca(cif->nargs * sizeof(void*));
     /* storage for arguments which will be copied by unmarshal().  We could
        theoretically avoid the copies in many cases and use at most 128 bytes
@@ -436,7 +472,7 @@ void FFI_HIDDEN ffi_closure_inner(size_t *stack, call_context *aregs, ffi_closur
         avalue[i] = unmarshal(&cb, cif->arg_types[i],
             i >= cif->riscv_nfixedargs, astorage + i*MAXCOPYARG);
 
-    (closure->fun)(cif, rvalue, avalue, closure->user_data);
+    fun (cif, rvalue, avalue, user_data);
 
     if (!return_by_ref && cif->rtype->type != FFI_TYPE_VOID) {
         cb.used_integer = cb.used_float = 0;

data/ext/ffi_c/libffi/src/riscv/ffitarget.h

@@ -59,6 +59,7 @@ typedef enum ffi_abi {
 /* ---- Definitions for closures ----------------------------------------- */
 
 #define FFI_CLOSURES 1
+#define FFI_GO_CLOSURES 1
 #define FFI_TRAMPOLINE_SIZE 24
 #define FFI_NATIVE_RAW_API 0
 #define FFI_EXTRA_CIF_FIELDS unsigned riscv_nfixedargs; unsigned riscv_unused;

data/ext/ffi_c/libffi/src/riscv/sysv.S

@@ -67,8 +67,8 @@
       intreg pad[rv32 ? 2 : 0];
       intreg save_fp, save_ra;
   }
-  void ffi_call_asm(size_t *stackargs, struct call_context *regargs,
-      void (*fn)(void));
+  void ffi_call_asm (size_t *stackargs, struct call_context *regargs,
+                     void (*fn) (void), void *closure);
 */
 
 #define FRAME_LEN (8 * FLTS + 8 * PTRS + 16)
@@ -98,6 +98,7 @@ ffi_call_asm:
 
     # Load arguments
     mv      t1, a2
+    mv      t2, a3
 
 #if FLTS
     FLARG   fa0, -FRAME_LEN+0*FLTS(fp)
@@ -145,8 +146,10 @@ ffi_call_asm:
 
 /*
   ffi_closure_asm. Expects address of the passed-in ffi_closure in t1.
-  void ffi_closure_inner(size_t *stackargs, struct call_context *regargs,
-      ffi_closure *closure);
+  void ffi_closure_inner (ffi_cif *cif,
+		          void (*fun) (ffi_cif *, void *, void **, void *),
+		          void *user_data,
+		          size_t *stackargs, struct call_context *regargs)
 */
 
     .globl ffi_closure_asm
@@ -187,9 +190,11 @@ ffi_closure_asm:
     SARG    a7, 8*FLTS+7*PTRS(sp)
 
     /* enter C */
-    addi    a0, sp, FRAME_LEN
-    mv      a1, sp
-    mv      a2, t1
+    LARG    a0, FFI_TRAMPOLINE_SIZE+0*PTRS(t1)
+    LARG    a1, FFI_TRAMPOLINE_SIZE+1*PTRS(t1)
+    LARG    a2, FFI_TRAMPOLINE_SIZE+2*PTRS(t1)
+    addi    a3, sp, FRAME_LEN
+    mv      a4, sp
 
     call    ffi_closure_inner
 
@@ -212,3 +217,77 @@ ffi_closure_asm:
     ret
     .cfi_endproc
     .size ffi_closure_asm, .-ffi_closure_asm
+
+/*
+  ffi_go_closure_asm.  Expects address of the passed-in ffi_go_closure in t2.
+  void ffi_closure_inner (ffi_cif *cif,
+		          void (*fun) (ffi_cif *, void *, void **, void *),
+		          void *user_data,
+		          size_t *stackargs, struct call_context *regargs)
+*/
+
+    .globl ffi_go_closure_asm
+    .hidden ffi_go_closure_asm
+    .type ffi_go_closure_asm, @function
+ffi_go_closure_asm:
+    .cfi_startproc
+
+    addi    sp,  sp, -FRAME_LEN
+    .cfi_def_cfa_offset FRAME_LEN
+
+    /* make a frame */
+    SARG    fp, FRAME_LEN - 2*PTRS(sp)
+    .cfi_offset 8, -2*PTRS
+    SARG    ra, FRAME_LEN - 1*PTRS(sp)
+    .cfi_offset 1, -1*PTRS
+    addi    fp, sp, FRAME_LEN
+
+    /* save arguments */
+#if FLTS
+    FSARG   fa0, 0*FLTS(sp)
+    FSARG   fa1, 1*FLTS(sp)
+    FSARG   fa2, 2*FLTS(sp)
+    FSARG   fa3, 3*FLTS(sp)
+    FSARG   fa4, 4*FLTS(sp)
+    FSARG   fa5, 5*FLTS(sp)
+    FSARG   fa6, 6*FLTS(sp)
+    FSARG   fa7, 7*FLTS(sp)
+#endif
+
+    SARG    a0, 8*FLTS+0*PTRS(sp)
+    SARG    a1, 8*FLTS+1*PTRS(sp)
+    SARG    a2, 8*FLTS+2*PTRS(sp)
+    SARG    a3, 8*FLTS+3*PTRS(sp)
+    SARG    a4, 8*FLTS+4*PTRS(sp)
+    SARG    a5, 8*FLTS+5*PTRS(sp)
+    SARG    a6, 8*FLTS+6*PTRS(sp)
+    SARG    a7, 8*FLTS+7*PTRS(sp)
+
+    /* enter C */
+    LARG    a0, 1*PTRS(t2)
+    LARG    a1, 2*PTRS(t2)
+    mv      a2, t2
+    addi    a3, sp, FRAME_LEN
+    mv      a4, sp
+
+    call    ffi_closure_inner
+
+    /* return values */
+#if FLTS
+    FLARG   fa0, 0*FLTS(sp)
+    FLARG   fa1, 1*FLTS(sp)
+#endif
+
+    LARG    a0, 8*FLTS+0*PTRS(sp)
+    LARG    a1, 8*FLTS+1*PTRS(sp)
+
+    /* restore and return */
+    LARG    ra, FRAME_LEN-1*PTRS(sp)
+    .cfi_restore 1
+    LARG    fp, FRAME_LEN-2*PTRS(sp)
+    .cfi_restore 8
+    addi    sp, sp, FRAME_LEN
+    .cfi_def_cfa_offset 0
+    ret
+    .cfi_endproc
+    .size ffi_go_closure_asm, .-ffi_go_closure_asm