diff --git a/src/Makefile b/src/Makefile
index b9101a74..fa5aed1c 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -479,7 +479,8 @@ TARGET_DEP= $(LIB_VMDEF) $(LUAJIT_SO)
 
 ifeq (Windows,$(TARGET_SYS))
   TARGET_DYNCC= $(STATIC_CC)
-  LJVM_MODE= coffasm
+  LJVM_MODE= peobj
+  LJVM_BOUT= $(LJVM_O)
   LUAJIT_T= luajit.exe
   ifeq (cygwin,$(HOST_MSYS))
     LUAJIT_SO= cyg$(TARGET_DLLNAME)
diff --git a/src/lj_ccall.c b/src/lj_ccall.c
index 34866dd2..a3ad1974 100644
--- a/src/lj_ccall.c
+++ b/src/lj_ccall.c
@@ -31,6 +31,26 @@
 
 #else
 
+#if LJ_TARGET_OSX
+
+#define CCALL_HANDLE_STRUCTRET \
+  /* Return structs of size 1, 2, 4 or 8 in registers. */ \
+  cc->retref = !(sz == 1 || sz == 2 || sz == 4 || sz == 8); \
+  if (cc->retref) { \
+    if (ngpr < maxgpr) \
+      cc->gpr[ngpr++] = (GPRArg)dp; \
+    else \
+      cc->stack[nsp++] = (GPRArg)dp; \
+  } else {  /* Struct with single FP field ends up in FPR. */ \
+    cc->resx87 = ccall_classify_struct(cts, ctr); \
+  }
+
+#define CCALL_HANDLE_STRUCTRET2 \
+  if (cc->resx87) sp = (uint8_t *)&cc->fpr[0]; \
+  memcpy(dp, sp, ctr->size);
+
+#else
+
 #define CCALL_HANDLE_STRUCTRET \
   cc->retref = 1;  /* Return all structs by reference (in reg or on stack). */ \
   if (ngpr < maxgpr) \
@@ -38,6 +58,8 @@
   else \
     cc->stack[nsp++] = (GPRArg)dp;
 
+#endif
+
 #define CCALL_HANDLE_COMPLEXRET \
   /* Return complex float in GPRs and complex double by reference. */ \
   cc->retref = (sz > 8); \
@@ -102,9 +124,9 @@
 /* Windows/x64 argument registers are strictly positional (use ngpr). */
 #define CCALL_HANDLE_REGARG \
   if (isfp) { \
-    if (ngpr < 4) { dp = &cc->fpr[ngpr++]; nfpr = ngpr; goto done; } \
+    if (ngpr < maxgpr) { dp = &cc->fpr[ngpr++]; nfpr = ngpr; goto done; } \
   } else { \
-    if (ngpr < 4) { dp = &cc->gpr[ngpr++]; goto done; } \
+    if (ngpr < maxgpr) { dp = &cc->gpr[ngpr++]; goto done; } \
   }
 
 #elif LJ_TARGET_X64
@@ -413,6 +435,42 @@
   memcpy(dp, sp, ctr->size);  /* Copy struct return value from GPRs. */
 #endif
 
+/* -- x86 OSX ABI struct classification ----------------------------------- */
+
+#if LJ_TARGET_X86 && LJ_TARGET_OSX
+
+/* Check for struct with single FP field. */
+static int ccall_classify_struct(CTState *cts, CType *ct)
+{
+  CTSize sz = ct->size;
+  if (!(sz == sizeof(float) || sz == sizeof(double))) return 0;
+  if ((ct->info & CTF_UNION)) return 0;
+  while (ct->sib) {
+    ct = ctype_get(cts, ct->sib);
+    if (ctype_isfield(ct->info)) {
+      CType *sct = ctype_rawchild(cts, ct);
+      if (ctype_isfp(sct->info)) {
+	if (sct->size == sz)
+	  return (sz >> 2);  /* Return 1 for float or 2 for double. */
+      } else if (ctype_isstruct(sct->info)) {
+	if (sct->size)
+	  return ccall_classify_struct(cts, sct);
+      } else {
+	break;
+      }
+    } else if (ctype_isbitfield(ct->info)) {
+      break;
+    } else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) {
+      CType *sct = ctype_rawchild(cts, ct);
+      if (sct->size)
+	return ccall_classify_struct(cts, sct);
+    }
+  }
+  return 0;
+}
+
+#endif
+
 /* -- x64 struct classification ------------------------------------------- */
 
 #if LJ_TARGET_X64 && !LJ_ABI_WIN
diff --git a/src/lj_ccallback.c b/src/lj_ccallback.c
index a3a0d798..00109923 100644
--- a/src/lj_ccallback.c
+++ b/src/lj_ccallback.c
@@ -286,9 +286,9 @@ void lj_ccallback_mcode_free(CTState *cts)
 /* Windows/x64 argument registers are strictly positional (use ngpr). */
 #define CALLBACK_HANDLE_REGARG \
   if (isfp) { \
-    if (ngpr < 4) { sp = &cts->cb.fpr[ngpr++]; nfpr = ngpr; goto done; } \
+    if (ngpr < maxgpr) { sp = &cts->cb.fpr[ngpr++]; UNUSED(nfpr); goto done; } \
   } else { \
-    if (ngpr < 4) { sp = &cts->cb.gpr[ngpr++]; goto done; } \
+    if (ngpr < maxgpr) { sp = &cts->cb.gpr[ngpr++]; goto done; } \
   }
 
 #elif LJ_TARGET_X64
diff --git a/src/lj_err.c b/src/lj_err.c
index 8ff0a455..7ae87a82 100644
--- a/src/lj_err.c
+++ b/src/lj_err.c
@@ -386,12 +386,17 @@ typedef struct UndocumentedDispatcherContext {
   ULONG Fill0;
 } UndocumentedDispatcherContext;
 
-#ifdef _MSC_VER
 /* Another wild guess. */
-extern __DestructExceptionObject(EXCEPTION_RECORD *rec, int nothrow);
+extern void __DestructExceptionObject(EXCEPTION_RECORD *rec, int nothrow);
+
+#ifdef MINGW_SDK_INIT
+/* Workaround for broken MinGW64 declaration. */
+VOID RtlUnwindEx_FIXED(PVOID,PVOID,PVOID,PVOID,PVOID,PVOID) asm("RtlUnwindEx");
+#define RtlUnwindEx RtlUnwindEx_FIXED
 #endif
 
 #define LJ_MSVC_EXCODE		((DWORD)0xe06d7363)
+#define LJ_GCC_EXCODE		((DWORD)0x20474343)
 
 #define LJ_EXCODE		((DWORD)0xe24c4a00)
 #define LJ_EXCODE_MAKE(c)	(LJ_EXCODE | (DWORD)(c))
@@ -411,10 +416,9 @@ LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
   } else {
     void *cf2 = err_unwind(L, cf, 0);
     if (cf2) {  /* We catch it, so start unwinding the upper frames. */
-      if (rec->ExceptionCode == LJ_MSVC_EXCODE) {
-#ifdef _MSC_VER
+      if (rec->ExceptionCode == LJ_MSVC_EXCODE ||
+	  rec->ExceptionCode == LJ_GCC_EXCODE) {
 	__DestructExceptionObject(rec, 1);
-#endif
 	setstrV(L, L->top++, lj_err_str(L, LJ_ERR_ERRCPP));
       } else if (!LJ_EXCODE_CHECK(rec->ExceptionCode)) {
 	/* Don't catch access violations etc. */
@@ -427,7 +431,7 @@ LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
       RtlUnwindEx(cf, (void *)((cframe_unwind_ff(cf2) && errcode != LUA_YIELD) ?
 			       lj_vm_unwind_ff_eh :
 			       lj_vm_unwind_c_eh),
-		  rec, (void *)errcode, ctx, dispatch->HistoryTable);
+		  rec, (void *)(uintptr_t)errcode, ctx, dispatch->HistoryTable);
       /* RtlUnwindEx should never return. */
     }
   }