koffi 2.15.3 → 2.15.5

package/CHANGELOG.md

@@ -7,6 +7,22 @@
 
 ### Koffi 2.15
 
+#### Koffi 2.15.5
+
+*Released on 2026-04-05*
+
+- Reduce repeated trampoline code size:
+  * Use single set of trampolines to forward general and floating-point values
+  * Minimize trampoline size by jumping early to common handling code
+- This reduces decompressed Koffi package size from 85.9 MB to 41.4 MB
+
+#### Koffi 2.15.4
+
+*Released on 2026-04-03*
+
+- Fix ABI issue with variadic parameters that are passed on the stack on Apple ARM64 platforms
+- Remove erroneous assertion from debug builds
+
 #### Koffi 2.15.3
 
 *Released on 2026-04-03*

package/index.js

@@ -4,9 +4,9 @@ var __commonJS = (cb, mod3) => function __require() {
   return mod3 || (0, cb[__getOwnPropNames(cb)[0]])((mod3 = { exports: {} }).exports, mod3), mod3.exports;
 };
 
-// bin/Koffi/package/src/cnoke/src/tools.js
+// ../../../bin/Koffi/package/src/cnoke/src/tools.js
 var require_tools = __commonJS({
-  "bin/Koffi/package/src/cnoke/src/tools.js"(exports2, module2) {
+  "../../../bin/Koffi/package/src/cnoke/src/tools.js"(exports2, module2) {
     "use strict";
     var crypto = require("crypto");
     var fs2 = require("fs");
@@ -393,12 +393,12 @@ var require_tools = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/package.json
+// ../../../bin/Koffi/package/src/koffi/package.json
 var require_package = __commonJS({
-  "bin/Koffi/package/src/koffi/package.json"(exports2, module2) {
+  "../../../bin/Koffi/package/src/koffi/package.json"(exports2, module2) {
     module2.exports = {
       name: "koffi",
-      version: "2.15.3",
+      version: "2.15.5",
       description: "Fast and simple C FFI (foreign function interface) for Node.js",
       keywords: [
         "foreign",
@@ -440,9 +440,9 @@ var require_package = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/src/init.js
+// ../../../bin/Koffi/package/src/koffi/src/init.js
 var require_init = __commonJS({
-  "bin/Koffi/package/src/koffi/src/init.js"(exports, module) {
+  "../../../bin/Koffi/package/src/koffi/src/init.js"(exports, module) {
     var fs = require("fs");
     var path = require("path");
     var util = require("util");
@@ -526,7 +526,7 @@ var require_init = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/index.js
+// ../../../bin/Koffi/package/src/koffi/index.js
 var { detect: detect2, init: init2 } = require_init();
 var triplet2 = detect2();
 var native2 = null;

package/indirect.js

@@ -4,9 +4,9 @@ var __commonJS = (cb, mod3) => function __require() {
   return mod3 || (0, cb[__getOwnPropNames(cb)[0]])((mod3 = { exports: {} }).exports, mod3), mod3.exports;
 };
 
-// bin/Koffi/package/src/cnoke/src/tools.js
+// ../../../bin/Koffi/package/src/cnoke/src/tools.js
 var require_tools = __commonJS({
-  "bin/Koffi/package/src/cnoke/src/tools.js"(exports2, module2) {
+  "../../../bin/Koffi/package/src/cnoke/src/tools.js"(exports2, module2) {
     "use strict";
     var crypto = require("crypto");
     var fs2 = require("fs");
@@ -393,12 +393,12 @@ var require_tools = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/package.json
+// ../../../bin/Koffi/package/src/koffi/package.json
 var require_package = __commonJS({
-  "bin/Koffi/package/src/koffi/package.json"(exports2, module2) {
+  "../../../bin/Koffi/package/src/koffi/package.json"(exports2, module2) {
     module2.exports = {
       name: "koffi",
-      version: "2.15.3",
+      version: "2.15.5",
       description: "Fast and simple C FFI (foreign function interface) for Node.js",
       keywords: [
         "foreign",
@@ -440,9 +440,9 @@ var require_package = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/src/init.js
+// ../../../bin/Koffi/package/src/koffi/src/init.js
 var require_init = __commonJS({
-  "bin/Koffi/package/src/koffi/src/init.js"(exports, module) {
+  "../../../bin/Koffi/package/src/koffi/src/init.js"(exports, module) {
     var fs = require("fs");
     var path = require("path");
     var util = require("util");
@@ -526,7 +526,7 @@ var require_init = __commonJS({
   }
 });
 
-// bin/Koffi/package/src/koffi/indirect.js
+// ../../../bin/Koffi/package/src/koffi/indirect.js
 var { detect: detect2, init: init2 } = require_init();
 var triplet2 = detect2();
 var mod2 = init2(triplet2, null);

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "koffi",
-    "version": "2.15.3",
+    "version": "2.15.5",
     "description": "Fast and simple C FFI (foreign function interface) for Node.js",
     "keywords": [
         "foreign",

package/src/koffi/src/abi_arm32.cc

@@ -38,8 +38,6 @@ extern "C" uint64_t ForwardCallXGG(const void *func, uint8_t *sp, uint8_t **out_
 extern "C" float ForwardCallXF(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 extern "C" HfaRet ForwardCallXDDDD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 
-#include "trampolines/prototypes.inc"
-
 static int IsHFA(const TypeInfo *type)
 {
 #if defined(__ARM_PCS_VFP)
@@ -1008,12 +1006,6 @@ void CallData::Relay(Size idx, uint8_t *sp)
     err_guard.Disable();
 }
 
-void *GetTrampoline(int16_t idx, const FunctionInfo *proto)
-{
-    bool vec = proto->forward_fp || IsFloat(proto->ret.type);
-    return Trampolines[idx][vec];
-}
-
 }
 
 #endif

package/src/koffi/src/abi_arm32_asm.S

@@ -94,49 +94,29 @@ ForwardCallXDDDD:
 .global SwitchAndRelay
 .global RelayDirect
 
-# First, make a copy of the GPR argument registers (r0 to r7).
+# First, make a copy of the argument registers.
 # Then call the C function RelayCallback with the following arguments:
 # static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
 # arguments of this call, and a pointer to a struct that will contain the result registers.
 # After the call, simply load these registers from the output struct.
 .macro trampoline id
     .cfi_startproc
-    push {fp, lr}
-    .cfi_def_cfa sp, 8
-    .cfi_offset 11, 4
-    .cfi_offset 14, 8
-    sub sp, sp, #120
-    .cfi_def_cfa sp, 128
-    add r12, sp, 64
-    stmia r12, {r0-r3}
-    mov r0, \id
-    mov r1, sp
-    bl RelayCallback
-    add sp, sp, #80
-    ldmia sp!, {r0-r1}
-    add sp, sp, #32
-    .cfi_def_cfa sp, 8
-    pop {fp, lr}
-    .cfi_def_cfa sp, 0
-    .cfi_restore 11
-    .cfi_restore 14
-    bx lr
+    mov r12, \id
+    b RelayTrampoline
     .cfi_endproc
 .endm
 
-# Same thing, but also forwards the floating-point argument registers and loads them at the end.
-.macro trampoline_vec id
+RelayTrampoline:
     .cfi_startproc
     push {fp, lr}
     .cfi_def_cfa sp, 8
     .cfi_offset 11, 4
     .cfi_offset 14, 8
-    sub sp, sp, #120
+    sub sp, sp, #40
+    stmdb sp!, {r0-r3}
+    vstmdb sp!, {d0-d7}
     .cfi_def_cfa sp, 128
-    mov r12, sp
-    vstmia r12!, {d0-d7}
-    stmia r12, {r0-r3}
-    mov r0, \id
+    mov r0, r12
     mov r1, sp
     bl RelayCallback
     add sp, sp, #80
@@ -149,7 +129,6 @@ ForwardCallXDDDD:
     .cfi_restore 14
     bx lr
     .cfi_endproc
-.endm
 
 # When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 # The problem is that we're still running on the separate Koffi stack, and V8 will

package/src/koffi/src/abi_arm64.cc

@@ -48,8 +48,6 @@ extern "C" X0X1Ret ForwardCallXGG(const void *func, uint8_t *sp, uint8_t **out_o
 extern "C" float ForwardCallXF(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 extern "C" HfaRet ForwardCallXDDDD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 
-#include "trampolines/prototypes.inc"
-
 static HfaInfo IsHFA(const TypeInfo *type)
 {
     bool float32 = false;
@@ -286,7 +284,7 @@ bool CallData::Prepare(const FunctionInfo *func, const Napi::CallbackInfo &info)
             if (param.gpr_count) [[likely]] { \
                 *(gpr_ptr++) = (uint64_t)v; \
             } else { \
-                args_ptr = AlignUp(args_ptr, param.type->align); \
+                args_ptr = AlignUp(args_ptr, param.variadic ? 8 : param.type->align); \
                 *args_ptr = (uint64_t)v; \
                 args_ptr = (uint64_t *)((uint8_t *)args_ptr + param.type->size); \
             } \
@@ -303,7 +301,7 @@ bool CallData::Prepare(const FunctionInfo *func, const Napi::CallbackInfo &info)
             if (param.gpr_count) [[likely]] { \
                 *(gpr_ptr++) = (uint64_t)ReverseBytes(v); \
             } else { \
-                args_ptr = AlignUp(args_ptr, param.type->align); \
+                args_ptr = AlignUp(args_ptr, param.variadic ? 8 : param.type->align); \
                 *args_ptr = (uint64_t)ReverseBytes(v); \
                 args_ptr = (uint64_t *)((uint8_t *)args_ptr + param.type->size); \
             } \
@@ -353,6 +351,7 @@ bool CallData::Prepare(const FunctionInfo *func, const Napi::CallbackInfo &info)
                 if (param.gpr_count) [[likely]] {
                     *(gpr_ptr++) = (uint64_t)b;
                 } else {
+                    args_ptr = AlignUp(args_ptr, param.variadic ? 8 : 1);
                     *(uint8_t *)args_ptr = b;
                     args_ptr = (uint64_t *)((uint8_t *)args_ptr + 1);
                 }
@@ -484,7 +483,7 @@ bool CallData::Prepare(const FunctionInfo *func, const Napi::CallbackInfo &info)
 #endif
                 } else {
 #if defined(__APPLE__)
-                    args_ptr = AlignUp(args_ptr, 4);
+                    args_ptr = AlignUp(args_ptr, param.variadic ? 8 : 4);
                     *(float *)args_ptr = f;
                     args_ptr = (uint64_t *)((uint8_t *)args_ptr + 4);
 #else
@@ -1302,12 +1301,6 @@ void CallData::Relay(Size idx, uint8_t *sp)
     err_guard.Disable();
 }
 
-void *GetTrampoline(int16_t idx, const FunctionInfo *proto)
-{
-    bool vec = proto->forward_fp || IsFloat(proto->ret.type);
-    return Trampolines[idx][vec];
-}
-
 }
 
 #endif

package/src/koffi/src/abi_arm64_asm.S

@@ -115,7 +115,7 @@ SYMBOL(ForwardCallXDDDD):
 .global SYMBOL(SwitchAndRelay)
 .global SYMBOL(RelayDirect)
 
-# First, make a copy of the GPR argument registers (x0 to x7).
+# First, make a copy of the argument registers.
 # Then call the C function RelayCallback with the following arguments:
 # static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
 # arguments of this call, and a pointer to a struct that will contain the result registers.
@@ -123,35 +123,13 @@ SYMBOL(ForwardCallXDDDD):
 .macro trampoline id
     .cfi_startproc
     hint #34
-    stp x29, x30, [sp, -16]!
-    .cfi_def_cfa sp, 16
-    .cfi_offset 29, 16
-    .cfi_offset 30, 8
-    sub sp, sp, #192
-    .cfi_def_cfa sp, 208
-    stp x0, x1, [sp, 0]
-    stp x2, x3, [sp, 16]
-    stp x4, x5, [sp, 32]
-    stp x6, x7, [sp, 48]
-    str x8, [sp, 64]
-    mov x0, \id
-    mov x1, sp
-    bl SYMBOL(RelayCallback)
-    ldp x0, x1, [sp, 136]
-    add sp, sp, #192
-    .cfi_def_cfa sp, 16
-    ldp x29, x30, [sp], 16
-    .cfi_def_cfa sp, 0
-    .cfi_restore 30
-    .cfi_restore 29
-    ret
+    mov x9, \id
+    b RelayTrampoline
     .cfi_endproc
 .endm
 
-# Same thing, but also forwards the floating-point argument registers and loads them at the end.
-.macro trampoline_vec id
+RelayTrampoline:
     .cfi_startproc
-    hint #34
     stp x29, x30, [sp, -16]!
     .cfi_def_cfa sp, 16
     .cfi_offset 29, 16
@@ -167,7 +145,7 @@ SYMBOL(ForwardCallXDDDD):
     stp d2, d3, [sp, 88]
     stp d4, d5, [sp, 104]
     stp d6, d7, [sp, 120]
-    mov x0, \id
+    mov x0, x9
     mov x1, sp
     bl SYMBOL(RelayCallback)
     ldp x0, x1, [sp, 136]
@@ -181,7 +159,6 @@ SYMBOL(ForwardCallXDDDD):
     .cfi_restore 29
     ret
     .cfi_endproc
-.endm
 
 # When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 # The problem is that we're still running on the separate Koffi stack, and V8 will

package/src/koffi/src/abi_arm64_asm.asm

@@ -110,7 +110,7 @@ ForwardCallXDDDD PROC
     EXPORT RelayDirect
     EXTERN RelayDirect
 
-    ; First, make a copy of the GPR argument registers (x0 to x7).
+    ; First, make a copy of the argument registers.
     ; Then call the C function RelayCallback with the following arguments:
     ; static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
     ; arguments of this call, and a pointer to a struct that will contain the result registers.
@@ -118,26 +118,11 @@ ForwardCallXDDDD PROC
     MACRO
     trampoline $ID
 
-    stp x29, x30, [sp, -16]!
-    sub sp, sp, #192
-    stp x0, x1, [sp, 0]
-    stp x2, x3, [sp, 16]
-    stp x4, x5, [sp, 32]
-    stp x6, x7, [sp, 48]
-    str x8, [sp, 64]
-    mov x0, $ID
-    mov x1, sp
-    bl RelayCallback
-    ldp x0, x1, [sp, 136]
-    add sp, sp, #192
-    ldp x29, x30, [sp], 16
-    ret
+    mov x9, $ID
+    b RelayTrampoline
     MEND
 
-    ; Same thing, but also forwards the floating-point argument registers and loads them at the end.
-    MACRO
-    trampoline_vec $ID
-
+RelayTrampoline PROC
     stp x29, x30, [sp, -16]!
     sub sp, sp, #192
     stp x0, x1, [sp, 0]
@@ -149,7 +134,7 @@ ForwardCallXDDDD PROC
     stp d2, d3, [sp, 88]
     stp d4, d5, [sp, 104]
     stp d6, d7, [sp, 120]
-    mov x0, $ID
+    mov x0, x9
     mov x1, sp
     bl RelayCallback
     ldp x0, x1, [sp, 136]
@@ -158,7 +143,7 @@ ForwardCallXDDDD PROC
     add sp, sp, #192
     ldp x29, x30, [sp], 16
     ret
-    MEND
+    ENDP
 
 ; When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 ; The problem is that we're still running on the separate Koffi stack, and V8 will

package/src/koffi/src/abi_loong64_asm.S

@@ -132,34 +132,17 @@ ForwardCallXDD:
 .global SwitchAndRelay
 .global RelayDirect
 
-# First, make a copy of the GPR argument registers (a0 to a7).
+# First, make a copy of the argument registers.
 # Then call the C function RelayCallback with the following arguments:
 # static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
 # arguments of this call, and a pointer to a struct that will contain the result registers.
 # After the call, simply load these registers from the output struct.
 .macro trampoline id
-    addi.d $sp, $sp, -176
-    st.d $ra, $sp, 0
-    st.d $a0, $sp, 8
-    st.d $a1, $sp, 16
-    st.d $a2, $sp, 24
-    st.d $a3, $sp, 32
-    st.d $a4, $sp, 40
-    st.d $a5, $sp, 48
-    st.d $a6, $sp, 56
-    st.d $a7, $sp, 64
-    li.d $a0, \id
-    addi.d $a1, $sp, 8
-    bl RelayCallback
-    ld.d $ra, $sp, 0
-    ld.d $a0, $sp, 136
-    ld.d $a1, $sp, 144
-    addi.d $sp, $sp, 176
-    jr $ra
+    li.d $t0, \id
+    b RelayTrampoline
 .endm
 
-# Same thing, but also forwards the floating-point argument registers and loads them at the end.
-.macro trampoline_vec id
+RelayTrampoline:
     addi.d $sp, $sp, -176
     st.d $ra, $sp, 0
     st.d $a0, $sp, 8
@@ -178,7 +161,7 @@ ForwardCallXDD:
     fst.d $fa5, $sp, 112
     fst.d $fa6, $sp, 120
     fst.d $fa7, $sp, 128
-    li.d $a0, \id
+    move $a0, $t0
     addi.d $a1, $sp, 8
     bl RelayCallback
     ld.d $ra, $sp, 0
@@ -188,7 +171,6 @@ ForwardCallXDD:
     fld.d $fa1, $sp, 160
     addi.d $sp, $sp, 176
     jr $ra
-.endm
 
 # When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 # The problem is that we're still running on the separate Koffi stack, and V8 will

package/src/koffi/src/abi_riscv64.cc

@@ -48,8 +48,6 @@ extern "C" Fa0A0Ret ForwardCallXDG(const void *func, uint8_t *sp, uint8_t **out_
 extern "C" A0Fa0Ret ForwardCallXGD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 extern "C" Fa0Fa1Ret ForwardCallXDD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 
-#include "trampolines/prototypes.inc"
-
 static inline void ExpandPair(const uint8_t raw[16], int size1, int size2, uint64_t out_regs[2])
 {
     memcpy(out_regs + 0, raw, size1);
@@ -905,12 +903,6 @@ void CallData::Relay(Size idx, uint8_t *sp)
     err_guard.Disable();
 }
 
-void *GetTrampoline(int16_t idx, const FunctionInfo *proto)
-{
-    bool fp = proto->forward_fp || proto->ret.vec_count;
-    return Trampolines[idx][fp];
-}
-
 }
 
 #endif

package/src/koffi/src/abi_riscv64_asm.S

@@ -132,34 +132,17 @@ ForwardCallXDD:
 .global SwitchAndRelay
 .global RelayDirect
 
-# First, make a copy of the GPR argument registers (a0 to a7).
+# First, make a copy of argument registers.
 # Then call the C function RelayCallback with the following arguments:
 # static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
 # arguments of this call, and a pointer to a struct that will contain the result registers.
 # After the call, simply load these registers from the output struct.
 .macro trampoline id
-    addi sp, sp, -176
-    sd ra, 0(sp)
-    sd a0, 8(sp)
-    sd a1, 16(sp)
-    sd a2, 24(sp)
-    sd a3, 32(sp)
-    sd a4, 40(sp)
-    sd a5, 48(sp)
-    sd a6, 56(sp)
-    sd a7, 64(sp)
-    li a0, \id
-    addi a1, sp, 8
-    call RelayCallback
-    ld ra, 0(sp)
-    ld a0, 136(sp)
-    ld a1, 144(sp)
-    addi sp, sp, 176
-    ret
+    li t0, \id
+    j RelayTrampoline
 .endm
 
-# Same thing, but also forwards the floating-point argument registers and loads them at the end.
-.macro trampoline_vec id
+RelayTrampoline:
     addi sp, sp, -176
     sd ra, 0(sp)
     sd a0, 8(sp)
@@ -178,7 +161,7 @@ ForwardCallXDD:
     fsd fa5, 112(sp)
     fsd fa6, 120(sp)
     fsd fa7, 128(sp)
-    li a0, \id
+    mv a0, t0
     addi a1, sp, 8
     call RelayCallback
     ld ra, 0(sp)
@@ -188,7 +171,6 @@ ForwardCallXDD:
     fld fa1, 160(sp)
     addi sp, sp, 176
     ret
-.endm
 
 # When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 # The problem is that we're still running on the separate Koffi stack, and V8 will

package/src/koffi/src/abi_x64_sysv.cc

@@ -55,8 +55,6 @@ extern "C" Xmm0RaxRet ForwardCallXDG(const void *func, uint8_t *sp, uint8_t **ou
 extern "C" RaxXmm0Ret ForwardCallXGD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 extern "C" Xmm0Xmm1Ret ForwardCallXDD(const void *func, uint8_t *sp, uint8_t **out_old_sp);
 
-#include "trampolines/prototypes.inc"
-
 static inline RegisterClass MergeClasses(RegisterClass cls1, RegisterClass cls2)
 {
     if (cls1 == cls2)
@@ -929,12 +927,6 @@ void CallData::Relay(Size idx, uint8_t *sp)
     err_guard.Disable();
 }
 
-void *GetTrampoline(int16_t idx, const FunctionInfo *proto)
-{
-    bool xmm = proto->forward_fp || IsFloat(proto->ret.type);
-    return Trampolines[idx][xmm];
-}
-
 }
 
 #endif

package/src/koffi/src/abi_x64_sysv_asm.S

@@ -149,7 +149,7 @@ SYMBOL(ForwardCallXDD):
 .global SYMBOL(SwitchAndRelay)
 .global SYMBOL(RelayDirect)
 
-# First, make a copy of the GPR argument registers (rdi, rsi, rdx, rcx, r8, r9).
+# First, make a copy of argument registers.
 # Then call the C function RelayCallback with the following arguments:
 # static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
 # arguments of this call, and a pointer to a struct that will contain the result registers.
@@ -158,34 +158,14 @@ SYMBOL(ForwardCallXDD):
     .cfi_startproc
     .cfi_def_cfa rsp, 8
     ENDBR64
-    subq $152, %rsp
-    .cfi_def_cfa rsp, 160
-    movq %rdi, 0(%rsp)
-    movq %rsi, 8(%rsp)
-    movq %rdx, 16(%rsp)
-    movq %rcx, 24(%rsp)
-    movq %r8, 32(%rsp)
-    movq %r9, 40(%rsp)
-    movq $\id, %rdi
-    movq %rsp, %rsi
-#if defined(__linux__)
-    call *RelayCallback@GOTPCREL(%rip)
-#else
-    call SYMBOL(RelayCallback)
-#endif
-    movq 112(%rsp), %rax
-    movq 120(%rsp), %rdx
-    addq $152, %rsp
-    .cfi_def_cfa rsp, 8
-    ret
+    movq $\id, %rax
+    jmp RelayTrampoline
     .cfi_endproc
 .endm
 
-# Same thing, but also forward the XMM argument registers and load the XMM result registers.
-.macro trampoline_vec id
+RelayTrampoline:
     .cfi_startproc
     .cfi_def_cfa rsp, 8
-    ENDBR64
     subq $152, %rsp
     .cfi_def_cfa rsp, 160
     movq %rdi, 0(%rsp)
@@ -202,7 +182,7 @@ SYMBOL(ForwardCallXDD):
     movsd %xmm5, 88(%rsp)
     movsd %xmm6, 96(%rsp)
     movsd %xmm7, 104(%rsp)
-    movq $\id, %rdi
+    movq %rax, %rdi
     movq %rsp, %rsi
 #if defined(__linux__)
     call *RelayCallback@GOTPCREL(%rip)
@@ -217,7 +197,6 @@ SYMBOL(ForwardCallXDD):
     .cfi_def_cfa rsp, 8
     ret
     .cfi_endproc
-.endm
 
 # When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
 # The problem is that we're still running on the separate Koffi stack, and V8 will