main/luajit: update ppc64le support patch to latest

This should fix lua-penlight testsuite on ppc64le

3 files changed, 3525 insertions, 4825 deletions

diff --git a/main/luajit/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch b/main/luajit/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
new file mode 100644
index 0000000000..a879f3fc9d
--- /dev/null
+++ b/main/luajit/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
@@ -0,0 +1,3522 @@
+From: "Rodrigo R. Galvao" <rosattig@br.ibm.com>
+Date: Wed, 11 Oct 2017 08:41:47 +0000
+Subject: New patch proposal for PPC64 support
+
+ Create a patch for PPC64 support based on 
+https://github.com/LuaJIT/LuaJIT/pull/140.
+ It replaces the old patch since this new one is more likely to be merged 
+with luajit upstream.
+
+
+Author: Rodrigo R. Galvao <rosattig@br.ibm.com>
+---
+ dynasm/dasm_ppc.lua    |    5 +
+ src/Makefile           |   11 +-
+ src/host/buildvm_asm.c |   16 +-
+ src/lj_arch.h          |   18 +-
+ src/lj_ccall.c         |  166 ++++++-
+ src/lj_ccall.h         |   13 +
+ src/lj_ccallback.c     |   68 ++-
+ src/lj_ctype.h         |    2 +-
+ src/lj_def.h           |    4 +
+ src/lj_frame.h         |    9 +
+ src/lj_target_ppc.h    |   14 +
+ src/vm_ppc.dasc        | 1290 ++++++++++++++++++++++++++++++++----------------
+ 12 files changed, 1162 insertions(+), 454 deletions(-)
+
+diff --git a/dynasm/dasm_ppc.lua b/dynasm/dasm_ppc.lua
+index f73974d..a4ad70b 100644
+--- a/dynasm/dasm_ppc.lua
++++ b/dynasm/dasm_ppc.lua
+@@ -257,9 +257,11 @@ map_op = {
+   addic_3 =	"30000000RRI",
+   ["addic._3"] = "34000000RRI",
+   addi_3 =	"38000000RR0I",
++  addil_3 =	"38000000RR0J",
+   li_2 =	"38000000RI",
+   la_2 =	"38000000RD",
+   addis_3 =	"3c000000RR0I",
++  addisl_3 =	"3c000000RR0J",
+   lis_2 =	"3c000000RI",
+   lus_2 =	"3c000000RU",
+   bc_3 =	"40000000AAK",
+@@ -842,6 +844,9 @@ map_op = {
+   srdi_3 =	op_alias("rldicl_4", function(p)
+     p[4] = p[3]; p[3] = "64-("..p[3]..")"
+   end),
++  ["srdi._3"] =	op_alias("rldicl._4", function(p)
++    p[4] = p[3]; p[3] = "64-("..p[3]..")"
++  end),
+   clrldi_3 =	op_alias("rldicl_4", function(p)
+     p[4] = p[3]; p[3] = "0"
+   end),
+diff --git a/src/Makefile b/src/Makefile
+index 6b73a89..cc50bae 100644
+--- a/src/Makefile
++++ b/src/Makefile
+@@ -453,7 +453,16 @@ ifeq (ppc,$(TARGET_LJARCH))
+     DASM_AFLAGS+= -D GPR64
+   endif
+   ifeq (PS3,$(TARGET_SYS))
+-    DASM_AFLAGS+= -D PPE -D TOC
++    DASM_AFLAGS+= -D PPE
++  endif
++  ifneq (,$(findstring LJ_ARCH_PPC_OPD 1,$(TARGET_TESTARCH)))
++    DASM_AFLAGS+= -D OPD
++  endif
++  ifneq (,$(findstring LJ_ARCH_PPC_OPDENV 1,$(TARGET_TESTARCH)))
++    DASM_AFLAGS+= -D OPDENV
++  endif
++  ifneq (,$(findstring LJ_ARCH_PPC_ELFV2 1,$(TARGET_TESTARCH)))
++    DASM_AFLAGS+= -D ELFV2
+   endif
+   ifneq (,$(findstring LJ_ARCH_PPC64 ,$(TARGET_TESTARCH)))
+     DASM_ARCH= ppc64
+diff --git a/src/host/buildvm_asm.c b/src/host/buildvm_asm.c
+index ffd1490..6bb995e 100644
+--- a/src/host/buildvm_asm.c
++++ b/src/host/buildvm_asm.c
+@@ -140,18 +140,14 @@ static void emit_asm_wordreloc(BuildCtx *ctx, uint8_t *p, int n,
+ #else
+ #define TOCPREFIX ""
+ #endif
+-  if ((ins >> 26) == 16) {
++  if ((ins >> 26) == 14) {
++    fprintf(ctx->fp, "\taddi %d,%d,%s\n", (ins >> 21) & 31, (ins >> 16) & 31, sym);
++  } else if ((ins >> 26) == 15) {
++    fprintf(ctx->fp, "\taddis %d,%d,%s\n", (ins >> 21) & 31, (ins >> 16) & 31, sym);
++  } else if ((ins >> 26) == 16) {
+     fprintf(ctx->fp, "\t%s %d, %d, " TOCPREFIX "%s\n",
+ 	    (ins & 1) ? "bcl" : "bc", (ins >> 21) & 31, (ins >> 16) & 31, sym);
+   } else if ((ins >> 26) == 18) {
+-#if LJ_ARCH_PPC64
+-    const char *suffix = strchr(sym, '@');
+-    if (suffix && suffix[1] == 'h') {
+-      fprintf(ctx->fp, "\taddis 11, 2, %s\n", sym);
+-    } else if (suffix && suffix[1] == 'l') {
+-      fprintf(ctx->fp, "\tld 12, %s\n", sym);
+-    } else
+-#endif
+     fprintf(ctx->fp, "\t%s " TOCPREFIX "%s\n", (ins & 1) ? "bl" : "b", sym);
+   } else {
+     fprintf(stderr,
+@@ -250,7 +246,7 @@ void emit_asm(BuildCtx *ctx)
+   int i, rel;
+ 
+   fprintf(ctx->fp, "\t.file \"buildvm_%s.dasc\"\n", ctx->dasm_arch);
+-#if LJ_ARCH_PPC64
++#if LJ_ARCH_PPC_ELFV2
+   fprintf(ctx->fp, "\t.abiversion 2\n");
+ #endif
+   fprintf(ctx->fp, "\t.text\n");
+diff --git a/src/lj_arch.h b/src/lj_arch.h
+index d609b37..53bc651 100644
+--- a/src/lj_arch.h
++++ b/src/lj_arch.h
+@@ -269,10 +269,18 @@
+ #if LJ_TARGET_CONSOLE
+ #define LJ_ARCH_PPC32ON64	1
+ #define LJ_ARCH_NOFFI		1
++#if LJ_TARGET_PS3
++#define LJ_ARCH_PPC_OPD		1
++#endif
+ #elif LJ_ARCH_BITS == 64
+-#define LJ_ARCH_PPC64		1
+-#define LJ_TARGET_GC64		1
++#define LJ_ARCH_PPC32ON64	1
+ #define LJ_ARCH_NOJIT		1	/* NYI */
++#if _CALL_ELF == 2
++#define LJ_ARCH_PPC_ELFV2	1
++#else
++#define LJ_ARCH_PPC_OPD		1
++#define LJ_ARCH_PPC_OPDENV	1
++#endif
+ #endif
+ 
+ #if _ARCH_PWR7
+@@ -423,12 +431,6 @@
+ #if defined(_SOFT_FLOAT) || defined(_SOFT_DOUBLE)
+ #error "No support for PowerPC CPUs without double-precision FPU"
+ #endif
+-#if !LJ_ARCH_PPC64 && LJ_ARCH_ENDIAN == LUAJIT_LE
+-#error "No support for little-endian PPC32"
+-#endif
+-#if LJ_ARCH_PPC64
+-#error "No support for PowerPC 64 bit mode (yet)"
+-#endif
+ #ifdef __NO_FPRS__
+ #error "No support for PPC/e500 anymore (use LuaJIT 2.0)"
+ #endif
+diff --git a/src/lj_ccall.c b/src/lj_ccall.c
+index 5c252e5..b891591 100644
+--- a/src/lj_ccall.c
++++ b/src/lj_ccall.c
+@@ -369,21 +369,97 @@
+ #elif LJ_TARGET_PPC
+ /* -- PPC calling conventions --------------------------------------------- */
+ 
++#if LJ_ARCH_BITS == 64
++
++#if LJ_ARCH_PPC_ELFV2
++
++#define CCALL_HANDLE_STRUCTRET \
++  if (sz > 16 && ccall_classify_fp(cts, ctr) <= 0) { \
++    cc->retref = 1;  /* Return by reference. */ \
++    cc->gpr[ngpr++] = (GPRArg)dp; \
++  }
++
++#define CCALL_HANDLE_STRUCTRET2 \
++  int isfp = ccall_classify_fp(cts, ctr); \
++  int i; \
++  if (isfp == FTYPE_FLOAT) { \
++    for (i = 0; i < ctr->size / 4; i++) \
++      ((float *)dp)[i] = cc->fpr[i]; \
++  } else if (isfp == FTYPE_DOUBLE) { \
++    for (i = 0; i < ctr->size / 8; i++) \
++      ((double *)dp)[i] = cc->fpr[i]; \
++  } else { \
++    if (ctr->size < 8 && LJ_BE) { \
++      sp += 8 - ctr->size; \
++    } \
++    memcpy(dp, sp, ctr->size); \
++  }
++
++#else
++
+ #define CCALL_HANDLE_STRUCTRET \
+   cc->retref = 1;  /* Return all structs by reference. */ \
+   cc->gpr[ngpr++] = (GPRArg)dp;
+ 
++#endif
++
+ #define CCALL_HANDLE_COMPLEXRET \
+   /* Complex values are returned in 2 or 4 GPRs. */ \
+   cc->retref = 0;
+ 
++#define CCALL_HANDLE_STRUCTARG
++
+ #define CCALL_HANDLE_COMPLEXRET2 \
+-  memcpy(dp, sp, ctr->size);  /* Copy complex from GPRs. */
++  if (ctr->size == 2*sizeof(float)) {  /* Copy complex float from FPRs. */ \
++    ((float *)dp)[0] = cc->fpr[0]; \
++    ((float *)dp)[1] = cc->fpr[1]; \
++  } else {  /* Copy complex double from FPRs. */ \
++    ((double *)dp)[0] = cc->fpr[0]; \
++    ((double *)dp)[1] = cc->fpr[1]; \
++  }
++
++#define CCALL_HANDLE_COMPLEXARG \
++  isfp = 1; \
++  if (d->size == sizeof(float) * 2) { \
++    d = ctype_get(cts, CTID_COMPLEX_DOUBLE); \
++    isf32 = 1; \
++  }
++
++#define CCALL_HANDLE_REGARG \
++  if (isfp && d->size == sizeof(float)) { \
++    d = ctype_get(cts, CTID_DOUBLE); \
++    isf32 = 1; \
++  } \
++  if (ngpr < maxgpr) { \
++   dp = &cc->gpr[ngpr]; \
++   ngpr += n; \
++   if (ngpr > maxgpr) { \
++     nsp += ngpr - 8; \
++     ngpr = 8; \
++     if (nsp > CCALL_MAXSTACK) { \
++       goto err_nyi; \
++     } \
++   } \
++   goto done; \
++  }
++
++#else
++
++#define CCALL_HANDLE_STRUCTRET \
++  cc->retref = 1;  /* Return all structs by reference. */ \
++  cc->gpr[ngpr++] = (GPRArg)dp;
++
++#define CCALL_HANDLE_COMPLEXRET \
++  /* Complex values are returned in 2 or 4 GPRs. */ \
++  cc->retref = 0;
+ 
+ #define CCALL_HANDLE_STRUCTARG \
+   rp = cdataptr(lj_cdata_new(cts, did, sz)); \
+   sz = CTSIZE_PTR;  /* Pass all structs by reference. */
+ 
++#define CCALL_HANDLE_COMPLEXRET2 \
++  memcpy(dp, sp, ctr->size);  /* Copy complex from GPRs. */
++
+ #define CCALL_HANDLE_COMPLEXARG \
+   /* Pass complex by value in 2 or 4 GPRs. */
+ 
+@@ -410,6 +486,8 @@
+     } \
+   }
+ 
++#endif
++
+ #define CCALL_HANDLE_RET \
+   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+     ctr = ctype_get(cts, CTID_DOUBLE);  /* FPRs always hold doubles. */
+@@ -801,6 +879,50 @@ noth:  /* Not a homogeneous float/double aggregate. */
+ 
+ #endif
+ 
++/* -- PowerPC64 ELFv2 ABI struct classification ------------------- */
++
++#if LJ_ARCH_PPC_ELFV2
++
++#define FTYPE_FLOAT	1
++#define FTYPE_DOUBLE	2
++
++static unsigned int ccall_classify_fp(CTState *cts, CType *ct) {
++  if (ctype_isfp(ct->info)) {
++    if (ct->size == sizeof(float))
++      return FTYPE_FLOAT;
++    else
++      return FTYPE_DOUBLE;
++  } else if (ctype_iscomplex(ct->info)) {
++    if (ct->size == sizeof(float) * 2)
++      return FTYPE_FLOAT;
++    else
++      return FTYPE_DOUBLE;
++  } else if (ctype_isstruct(ct->info)) {
++    int res = -1;
++    int sz = ct->size;
++    while (ct->sib) {
++      ct = ctype_get(cts, ct->sib);
++      if (ctype_isfield(ct->info)) {
++        int sub = ccall_classify_fp(cts, ctype_rawchild(cts, ct));
++        if (res == -1)
++          res = sub;
++        if (sub != -1 && sub != res)
++          return 0;
++      } else if (ctype_isbitfield(ct->info) ||
++        ctype_isxattrib(ct->info, CTA_SUBTYPE)) {
++        return 0;
++      }
++    }
++    if (res > 0 && sz > res * 4 * 8)
++      return 0;
++    return res;
++  } else {
++    return 0;
++  }
++}
++
++#endif
++
+ /* -- MIPS64 ABI struct classification ---------------------------- */
+ 
+ #if LJ_TARGET_MIPS64
+@@ -974,6 +1096,9 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
+     CTSize sz;
+     MSize n, isfp = 0, isva = 0;
+     void *dp, *rp = NULL;
++#if LJ_TARGET_PPC && LJ_ARCH_BITS == 64
++    int isf32 = 0;
++#endif
+ 
+     if (fid) {  /* Get argument type from field. */
+       CType *ctf = ctype_get(cts, fid);
+@@ -1030,7 +1155,37 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
+       *(void **)dp = rp;
+       dp = rp;
+     }
++#if LJ_TARGET_PPC && LJ_ARCH_BITS == 64 && LJ_BE
++    if (ctype_isstruct(d->info) && sz < CTSIZE_PTR) {
++      dp = (char *)dp + (CTSIZE_PTR - sz);
++    }
++#endif
+     lj_cconv_ct_tv(cts, d, (uint8_t *)dp, o, CCF_ARG(narg));
++#if LJ_TARGET_PPC && LJ_ARCH_BITS == 64
++    if (isfp) {
++      int i;
++      for (i = 0; i < d->size / 8 && nfpr < CCALL_NARG_FPR; i++)
++        cc->fpr[nfpr++] = ((double *)dp)[i];
++    }
++    if (isf32) {
++      int i;
++      for (i = 0; i < d->size / 8; i++)
++        ((float *)dp)[i*2] = ((double *)dp)[i];
++    }
++#endif
++#if LJ_ARCH_PPC_ELFV2
++    if (ctype_isstruct(d->info)) {
++      isfp = ccall_classify_fp(cts, d);
++      int i;
++      if (isfp == FTYPE_FLOAT) {
++        for (i = 0; i < d->size / 4 && nfpr < CCALL_NARG_FPR; i++)
++          cc->fpr[nfpr++] = ((float *)dp)[i];
++      } else if (isfp == FTYPE_DOUBLE) {
++        for (i = 0; i < d->size / 8 && nfpr < CCALL_NARG_FPR; i++)
++          cc->fpr[nfpr++] = ((double *)dp)[i];
++      }
++    }
++#endif
+     /* Extend passed integers to 32 bits at least. */
+     if (ctype_isinteger_or_bool(d->info) && d->size < 4) {
+       if (d->info & CTF_UNSIGNED)
+@@ -1044,6 +1199,15 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
+     if (isfp && d->size == sizeof(float))
+       ((float *)dp)[1] = ((float *)dp)[0];  /* Floats occupy high slot. */
+ #endif
++#if LJ_TARGET_PPC && LJ_ARCH_BITS == 64
++    if ((ctype_isinteger_or_bool(d->info) || ctype_isenum(d->info))
++	&& d->size <= 4) {
++      if (d->info & CTF_UNSIGNED)
++	*(uint64_t *)dp = (uint64_t)*(uint32_t *)dp;
++      else
++        *(int64_t *)dp = (int64_t)*(int32_t *)dp;
++    }
++#endif
+ #if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+     if ((ctype_isinteger_or_bool(d->info) || ctype_isenum(d->info)
+ #if LJ_TARGET_MIPS64
+diff --git a/src/lj_ccall.h b/src/lj_ccall.h
+index 59f6648..bbf309f 100644
+--- a/src/lj_ccall.h
++++ b/src/lj_ccall.h
+@@ -86,10 +86,23 @@ typedef union FPRArg {
+ #elif LJ_TARGET_PPC
+ 
+ #define CCALL_NARG_GPR		8
++#if LJ_ARCH_BITS == 64
++#define CCALL_NARG_FPR		13
++#if LJ_ARCH_PPC_ELFV2
++#define CCALL_NRET_GPR		2
++#define CCALL_NRET_FPR		8
++#define CCALL_SPS_EXTRA		14
++#else
++#define CCALL_NRET_GPR		1
++#define CCALL_NRET_FPR		2
++#define CCALL_SPS_EXTRA		16
++#endif
++#else
+ #define CCALL_NARG_FPR		8
+ #define CCALL_NRET_GPR		4	/* For complex double. */
+ #define CCALL_NRET_FPR		1
+ #define CCALL_SPS_EXTRA		4
++#endif
+ #define CCALL_SPS_FREE		0
+ 
+ typedef intptr_t GPRArg;
+diff --git a/src/lj_ccallback.c b/src/lj_ccallback.c
+index 846827b..eb7f445 100644
+--- a/src/lj_ccallback.c
++++ b/src/lj_ccallback.c
+@@ -61,8 +61,24 @@ static MSize CALLBACK_OFS2SLOT(MSize ofs)
+ 
+ #elif LJ_TARGET_PPC
+ 
++#if LJ_ARCH_PPC_OPD
++
++#define CALLBACK_SLOT2OFS(slot)		(24*(slot))
++#define CALLBACK_OFS2SLOT(ofs)		((ofs)/24)
++#define CALLBACK_MAX_SLOT		(CALLBACK_OFS2SLOT(CALLBACK_MCODE_SIZE))
++
++#elif LJ_ARCH_PPC_ELFV2
++
++#define CALLBACK_SLOT2OFS(slot)		(4*(slot))
++#define CALLBACK_OFS2SLOT(ofs)		((ofs)/4)
++#define CALLBACK_MAX_SLOT		(CALLBACK_MCODE_SIZE/4 - 10)
++
++#else
++
+ #define CALLBACK_MCODE_HEAD		24
+ 
++#endif
++
+ #elif LJ_TARGET_MIPS32
+ 
+ #define CALLBACK_MCODE_HEAD		20
+@@ -188,24 +204,59 @@ static void callback_mcode_init(global_State *g, uint32_t *page)
+   lua_assert(p - page <= CALLBACK_MCODE_SIZE);
+ }
+ #elif LJ_TARGET_PPC
++#if LJ_ARCH_PPC_OPD
++register void *vm_toc __asm__("r2");
++static void callback_mcode_init(global_State *g, uint64_t *page)
++{
++  uint64_t *p = page;
++  void *target = (void *)lj_vm_ffi_callback;
++  MSize slot;
++  for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
++    *p++ = (uint64_t)target;
++    *p++ = (uint64_t)vm_toc;
++    *p++ = (uint64_t)g | ((uint64_t)slot << 47);
++  }
++  lua_assert(p - page <= CALLBACK_MCODE_SIZE / 8);
++}
++#else
+ static void callback_mcode_init(global_State *g, uint32_t *page)
+ {
+   uint32_t *p = page;
+   void *target = (void *)lj_vm_ffi_callback;
+   MSize slot;
++#if LJ_ARCH_PPC_ELFV2
++  // Needs to be in sync with lj_vm_ffi_callback.
++  lua_assert(CALLBACK_MCODE_SIZE == 4096);
++  for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
++    *p = PPCI_B | (((page+CALLBACK_MAX_SLOT-p) & 0x00ffffffu) << 2);
++    p++;
++  }
++  *p++ = PPCI_LI | PPCF_T(RID_SYS1) | ((((intptr_t)target) >> 32) & 0xffff);
++  *p++ = PPCI_LI | PPCF_T(RID_R11) | ((((intptr_t)g) >> 32) & 0xffff);
++  *p++ = PPCI_RLDICR | PPCF_T(RID_SYS1) | PPCF_A(RID_SYS1) | PPCF_SH(32) | PPCF_M6(63-32);  /* sldi */
++  *p++ = PPCI_RLDICR | PPCF_T(RID_R11) | PPCF_A(RID_R11) | PPCF_SH(32) | PPCF_M6(63-32);  /* sldi */
++  *p++ = PPCI_ORIS | PPCF_A(RID_SYS1) | PPCF_T(RID_SYS1) | ((((intptr_t)target) >> 16) & 0xffff);
++  *p++ = PPCI_ORIS | PPCF_A(RID_R11) | PPCF_T(RID_R11) | ((((intptr_t)g) >> 16) & 0xffff);
++  *p++ = PPCI_ORI | PPCF_A(RID_SYS1) | PPCF_T(RID_SYS1) | (((intptr_t)target) & 0xffff);
++  *p++ = PPCI_ORI | PPCF_A(RID_R11) | PPCF_T(RID_R11) | (((intptr_t)g) & 0xffff);
++  *p++ = PPCI_MTCTR | PPCF_T(RID_SYS1);
++  *p++ = PPCI_BCTR;
++#else
+   *p++ = PPCI_LIS | PPCF_T(RID_TMP) | (u32ptr(target) >> 16);
+-  *p++ = PPCI_LIS | PPCF_T(RID_R12) | (u32ptr(g) >> 16);
++  *p++ = PPCI_LIS | PPCF_T(RID_R11) | (u32ptr(g) >> 16);
+   *p++ = PPCI_ORI | PPCF_A(RID_TMP)|PPCF_T(RID_TMP) | (u32ptr(target) & 0xffff);
+-  *p++ = PPCI_ORI | PPCF_A(RID_R12)|PPCF_T(RID_R12) | (u32ptr(g) & 0xffff);
++  *p++ = PPCI_ORI | PPCF_A(RID_R11)|PPCF_T(RID_R11) | (u32ptr(g) & 0xffff);
+   *p++ = PPCI_MTCTR | PPCF_T(RID_TMP);
+   *p++ = PPCI_BCTR;
+   for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
+-    *p++ = PPCI_LI | PPCF_T(RID_R11) | slot;
++    *p++ = PPCI_LI | PPCF_T(RID_R12) | slot;
+     *p = PPCI_B | (((page-p) & 0x00ffffffu) << 2);
+     p++;
+   }
+-  lua_assert(p - page <= CALLBACK_MCODE_SIZE);
++#endif
++  lua_assert(p - page <= CALLBACK_MCODE_SIZE / 4);
+ }
++#endif
+ #elif LJ_TARGET_MIPS
+ static void callback_mcode_init(global_State *g, uint32_t *page)
+ {
+@@ -641,6 +692,15 @@ static void callback_conv_result(CTState *cts, lua_State *L, TValue *o)
+ 	*(int32_t *)dp = ctr->size == 1 ? (int32_t)*(int8_t *)dp :
+ 					  (int32_t)*(int16_t *)dp;
+     }
++#if LJ_TARGET_PPC && LJ_ARCH_BITS == 64
++    if (ctr->size <= 4 &&
++       (ctype_isinteger_or_bool(ctr->info) || ctype_isenum(ctr->info))) {
++      if (ctr->info & CTF_UNSIGNED)
++        *(uint64_t *)dp = (uint64_t)*(uint32_t *)dp;
++      else
++        *(int64_t *)dp = (int64_t)*(int32_t *)dp;
++    }
++#endif
+ #if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+     /* Always sign-extend results to 64 bits. Even a soft-fp 'float'. */
+     if (ctr->size <= 4 &&
+diff --git a/src/lj_ctype.h b/src/lj_ctype.h
+index 0c220a8..105865b 100644
+--- a/src/lj_ctype.h
++++ b/src/lj_ctype.h
+@@ -153,7 +153,7 @@ typedef struct CType {
+ 
+ /* Simplify target-specific configuration. Checked in lj_ccall.h. */
+ #define CCALL_MAX_GPR		8
+-#define CCALL_MAX_FPR		8
++#define CCALL_MAX_FPR		14
+ 
+ typedef LJ_ALIGN(8) union FPRCBArg { double d; float f[2]; } FPRCBArg;
+ 
+diff --git a/src/lj_def.h b/src/lj_def.h
+index 2d8fff6..381d6f5 100644
+--- a/src/lj_def.h
++++ b/src/lj_def.h
+@@ -71,7 +71,11 @@ typedef unsigned int uintptr_t;
+ #define LJ_MAX_IDXCHAIN	100		/* __index/__newindex chain limit. */
+ #define LJ_STACK_EXTRA	(5+2*LJ_FR2)	/* Extra stack space (metamethods). */
+ 
++#if defined(__powerpc64__) && _CALL_ELF != 2
++#define LJ_NUM_CBPAGE	4		/* Number of FFI callback pages. */
++#else
+ #define LJ_NUM_CBPAGE	1		/* Number of FFI callback pages. */
++#endif
+ 
+ /* Minimum table/buffer sizes. */
+ #define LJ_MIN_GLOBAL	6		/* Min. global table size (hbits). */
+diff --git a/src/lj_frame.h b/src/lj_frame.h
+index 19c49a4..c666418 100644
+--- a/src/lj_frame.h
++++ b/src/lj_frame.h
+@@ -210,6 +210,15 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
+ #define CFRAME_OFS_MULTRES	408
+ #define CFRAME_SIZE		384
+ #define CFRAME_SHIFT_MULTRES	3
++#elif LJ_ARCH_PPC_ELFV2
++#define CFRAME_OFS_ERRF		360
++#define CFRAME_OFS_NRES		356
++#define CFRAME_OFS_PREV		336
++#define CFRAME_OFS_L		352
++#define CFRAME_OFS_PC		348
++#define CFRAME_OFS_MULTRES	344
++#define CFRAME_SIZE		368
++#define CFRAME_SHIFT_MULTRES	3
+ #elif LJ_ARCH_PPC32ON64
+ #define CFRAME_OFS_ERRF		472
+ #define CFRAME_OFS_NRES		468
+diff --git a/src/lj_target_ppc.h b/src/lj_target_ppc.h
+index c5c991a..f0c8c94 100644
+--- a/src/lj_target_ppc.h
++++ b/src/lj_target_ppc.h
+@@ -30,8 +30,13 @@ enum {
+ 
+   /* Calling conventions. */
+   RID_RET = RID_R3,
++#if LJ_LE
++  RID_RETHI = RID_R4,
++  RID_RETLO = RID_R3,
++#else
+   RID_RETHI = RID_R3,
+   RID_RETLO = RID_R4,
++#endif
+   RID_FPRET = RID_F1,
+ 
+   /* These definitions must match with the *.dasc file(s): */
+@@ -131,6 +136,8 @@ static LJ_AINLINE uint32_t *exitstub_trace_addr_(uint32_t *p, uint32_t exitno)
+ #define PPCF_C(r)	((r) << 6)
+ #define PPCF_MB(n)	((n) << 6)
+ #define PPCF_ME(n)	((n) << 1)
++#define PPCF_SH(n)	((((n) & 31) << (11+1)) | (((n) & 32) >> (5-1)))
++#define PPCF_M6(n)	((((n) & 31) << (5+1)) | (((n) & 32) << (11-5)))
+ #define PPCF_Y		0x00200000
+ #define PPCF_DOT	0x00000001
+ 
+@@ -200,6 +207,13 @@ typedef enum PPCIns {
+   PPCI_RLWINM = 0x54000000,
+   PPCI_RLWIMI = 0x50000000,
+ 
++  PPCI_RLDICL = 0x78000000,
++  PPCI_RLDICR = 0x78000004,
++  PPCI_RLDIC = 0x78000008,
++  PPCI_RLDIMI = 0x7800000c,
++  PPCI_RLDCL = 0x78000010,
++  PPCI_RLDCR = 0x78000012,
++
+   PPCI_B = 0x48000000,
+   PPCI_BL = 0x48000001,
+   PPCI_BC = 0x40800000,
+diff --git a/src/vm_ppc.dasc b/src/vm_ppc.dasc
+index b4260eb..abb381e 100644
+--- a/src/vm_ppc.dasc
++++ b/src/vm_ppc.dasc
+@@ -22,35 +22,40 @@
+ |// GPR64   64 bit registers (but possibly 32 bit pointers, e.g. PS3).
+ |//         Affects reg saves, stack layout, carry/overflow/dot flags etc.
+ |// FRAME32 Use 32 bit frame layout, even with GPR64 (Xbox 360).
+-|// TOC     Need table of contents (64 bit or 32 bit variant, e.g. PS3).
++|// OPD     Need function descriptors (64 bit or 32 bit variant, e.g. PS3).
+ |//         Function pointers are really a struct: code, TOC, env (optional).
+-|// TOCENV  Function pointers have an environment pointer, too (not on PS3).
++|// OPDENV  Function pointers have an environment pointer, too (not on PS3).
++|// ELFV2   The 64-bit ELF V2 ABI is in use.
+ |// PPE     Power Processor Element of Cell (PS3) or Xenon (Xbox 360).
+ |//         Must avoid (slow) micro-coded instructions.
+ |
+ |.if P64
+-|.define TOC, 1
+-|.define TOCENV, 1
+ |.macro lpx, a, b, c; ldx a, b, c; .endmacro
+ |.macro lp, a, b; ld a, b; .endmacro
+ |.macro stp, a, b; std a, b; .endmacro
++|.macro stpx, a, b, c; stdx a, b, c; .endmacro
+ |.define decode_OPP, decode_OP8
+-|.if FFI
+-|// Missing: Calling conventions, 64 bit regs, TOC.
+-|.error lib_ffi not yet implemented for PPC64
+-|.endif
++|.define PSIZE, 8
+ |.else
+ |.macro lpx, a, b, c; lwzx a, b, c; .endmacro
+ |.macro lp, a, b; lwz a, b; .endmacro
+ |.macro stp, a, b; stw a, b; .endmacro
++|.macro stpx, a, b, c; stwx a, b, c; .endmacro
+ |.define decode_OPP, decode_OP4
++|.define PSIZE, 4
+ |.endif
+ |
+ |// Convenience macros for TOC handling.
+-|.if TOC
++|.if OPD or ELFV2
+ |// Linker needs a TOC patch area for every external call relocation.
+-|.macro blex, target; bl extern target@plt; nop; .endmacro
++|.macro blex, target; bl extern target; nop; .endmacro
+ |.macro .toc, a, b; a, b; .endmacro
++|.else
++|.macro blex, target; bl extern target@plt; .endmacro
++|.macro .toc, a, b; .endmacro
++|.endif
++|.if OPD
++|.macro .opd, a, b; a, b; .endmacro
+ |.if P64
+ |.define TOC_OFS,	 8
+ |.define ENV_OFS,	16
+@@ -58,13 +63,13 @@
+ |.define TOC_OFS,	4
+ |.define ENV_OFS,	8
+ |.endif
+-|.else  // No TOC.
+-|.macro blex, target; bl extern target@plt; .endmacro
+-|.macro .toc, a, b; .endmacro
++|.else  // No OPD.
++|.macro .opd, a, b; .endmacro
+ |.endif
+-|.macro .tocenv, a, b; .if TOCENV; a, b; .endif; .endmacro
++|.macro .opdenv, a, b; .if OPDENV; a, b; .endif; .endmacro
+ |
+ |.macro .gpr64, a, b; .if GPR64; a, b; .endif; .endmacro
++|.macro .elfv2, a, b; .if ELFV2; a, b; .endif; .endmacro
+ |
+ |.macro andix., y, a, i
+ |.if PPE
+@@ -75,29 +80,6 @@
+ |.endif
+ |.endmacro
+ |
+-|.macro clrso, reg
+-|.if PPE
+-|  li reg, 0
+-|  mtxer reg
+-|.else
+-|  mcrxr cr0
+-|.endif
+-|.endmacro
+-|
+-|.macro checkov, reg, noov
+-|.if PPE
+-|  mfxer reg
+-|  add reg, reg, reg
+-|  cmpwi reg, 0
+-|   li reg, 0
+-|   mtxer reg
+-|  bgey noov
+-|.else
+-|  mcrxr cr0
+-|  bley noov
+-|.endif
+-|.endmacro
+-|
+ |//-----------------------------------------------------------------------
+ |
+ |// Fixed register assignments for the interpreter.
+@@ -111,6 +93,7 @@
+ |.define LREG,		r18	// Register holding lua_State (also in SAVE_L).
+ |.define MULTRES,	r19	// Size of multi-result: (nresults+1)*8.
+ |.define JGL,		r31	// On-trace: global_State + 32768.
++|.define BASEP4,	r25	// Equal to BASE + 4
+ |
+ |// Constants for type-comparisons, stores and conversions. C callee-save.
+ |.define TISNUM,	r22
+@@ -143,12 +126,19 @@
+ |
+ |.define FARG1,		f1
+ |.define FARG2,		f2
++|.define FARG3,		f3
++|.define FARG4,		f4
++|.define FARG5,		f5
++|.define FARG6,		f6
++|.define FARG7,		f7
++|.define FARG8,		f8
+ |
+ |.define CRET1,		r3
+ |.define CRET2,		r4
+ |
+ |.define TOCREG,	r2	// TOC register (only used by C code).
+ |.define ENVREG,	r11	// Environment pointer (nested C functions).
++|.define FUNCREG,	r12	// ELFv2 function pointer (overlaps RD)
+ |
+ |// Stack layout while in interpreter. Must match with lj_frame.h.
+ |.if GPR64
+@@ -182,6 +172,49 @@
+ |.define TMPD,		TMPD_HI
+ |.define TONUM_D,	TONUM_HI
+ |
++|.elif ELFV2
++|
++|//			392(sp) // \ 32 bit C frame info.
++|.define SAVE_LR,	384(sp)
++|.define SAVE_CR,	376(sp) // 64 bit CR save.
++|.define CFRAME_SPACE,	368     // Delta for sp.
++|// Back chain for sp:	368(sp) <-- sp entering interpreter
++|.define SAVE_ERRF,	360(sp) // |
++|.define SAVE_NRES,	356(sp) // |
++|.define SAVE_L,	352(sp) //  > Parameter save area.
++|.define SAVE_PC,	348(sp) // |
++|.define SAVE_MULTRES,	344(sp) // |
++|.define SAVE_CFRAME,	336(sp) // / 64 bit C frame chain.
++|.define SAVE_FPR_,	192     // .. 192+18*8: 64 bit FPR saves.
++|.define SAVE_GPR_,	48      // .. 48+18*8: 64 bit GPR saves.
++|.if ENDIAN_LE
++|.define TMPD_HI,	44(sp)
++|.define TMPD_LO,	40(sp)
++|.define TONUM_HI,	36(sp)
++|.define TONUM_LO,	32(sp)
++|.else
++|.define TMPD_LO,	44(sp)
++|.define TMPD_HI,	40(sp)
++|.define TONUM_LO,	36(sp)
++|.define TONUM_HI,	32(sp)
++|.endif
++|.define SAVE_TOC,	24(sp)  // TOC save area.
++|// Next frame lr:	16(sp)
++|// Next frame cr:	8(sp)
++|// Back chain for sp:	0(sp)	<-- sp while in interpreter
++|
++|.if ENDIAN_LE
++|.define TMPD_BLO,	32(sp)
++|.define TMPD,		TMPD_LO
++|.define TONUM_D,	TONUM_LO
++|.else
++|.define TMPD_BLO,	39(sp)
++|.define TMPD,		TMPD_HI
++|.define TONUM_D,	TONUM_HI
++|.endif
++|
++|.define EXIT_OFFSET,	32
++|
+ |.else
+ |
+ |//			508(sp) // \ 32 bit C frame info.
+@@ -192,23 +225,39 @@
+ |.define SAVE_MULTRES,	456(sp) // |
+ |.define SAVE_CFRAME,	448(sp) // / 64 bit C frame chain.
+ |.define SAVE_LR,	416(sp)
++|.define SAVE_CR,	408(sp)  // 64 bit CR save.
+ |.define CFRAME_SPACE,	400     // Delta for sp.
+ |// Back chain for sp:	400(sp) <-- sp entering interpreter
+ |.define SAVE_FPR_,	256     // .. 256+18*8: 64 bit FPR saves.
+ |.define SAVE_GPR_,	112     // .. 112+18*8: 64 bit GPR saves.
+ |//			48(sp)  // Callee parameter save area (ABI mandated).
+ |.define SAVE_TOC,	40(sp)  // TOC save area.
++|.if ENDIAN_LE
++|.define TMPD_HI,	36(sp)  // \ Link editor temp (ABI mandated).
++|.define TMPD_LO,	32(sp)  // /
++|.define TONUM_HI,	28(sp)  // \ Compiler temp (ABI mandated).
++|.define TONUM_LO,	24(sp)  // /
++|.else
+ |.define TMPD_LO,	36(sp)  // \ Link editor temp (ABI mandated).
+ |.define TMPD_HI,	32(sp)  // /
+ |.define TONUM_LO,	28(sp)  // \ Compiler temp (ABI mandated).
+ |.define TONUM_HI,	24(sp)  // /
++|.endif
+ |// Next frame lr:	16(sp)
+-|.define SAVE_CR,	8(sp)  // 64 bit CR save.
++|// Next frame cr:	8(sp)
+ |// Back chain for sp:	0(sp)	<-- sp while in interpreter
+ |
++|.if ENDIAN_LE
++|.define TMPD_BLO,	32(sp)
++|.define TMPD,		TMPD_LO
++|.define TONUM_D,	TONUM_LO
++|.else
+ |.define TMPD_BLO,	39(sp)
+ |.define TMPD,		TMPD_HI
+ |.define TONUM_D,	TONUM_HI
++|.endif
++|
++|.define EXIT_OFFSET,	112
+ |
+ |.endif
+ |.else
+@@ -226,16 +275,31 @@
+ |.define SAVE_PC,	32(sp)
+ |.define SAVE_MULTRES,	28(sp)
+ |.define UNUSED1,	24(sp)
++|.if ENDIAN_LE
++|.define TMPD_HI,	20(sp)
++|.define TMPD_LO,	16(sp)
++|.define TONUM_HI,	12(sp)
++|.define TONUM_LO,	8(sp)
++|.else
+ |.define TMPD_LO,	20(sp)
+ |.define TMPD_HI,	16(sp)
+ |.define TONUM_LO,	12(sp)
+ |.define TONUM_HI,	8(sp)
++|.endif
+ |// Next frame lr:	4(sp)
+ |// Back chain for sp:	0(sp)	<-- sp while in interpreter
+ |
++|.if ENDIAN_LE
++|.define TMPD_BLO,	16(sp)
++|.define TMPD,		TMPD_LO
++|.define TONUM_D,	TONUM_LO
++|.else
+ |.define TMPD_BLO,	23(sp)
+ |.define TMPD,		TMPD_HI
+ |.define TONUM_D,	TONUM_HI
++|.endif
++|
++|.define EXIT_OFFSET,	16
+ |
+ |.endif
+ |
+@@ -350,8 +414,35 @@
+ |//-----------------------------------------------------------------------
+ |
+ |// Access to frame relative to BASE.
++|.if ENDIAN_LE
++|.define FRAME_PC,	-4
++|.define FRAME_FUNC,	-8
++|.define FRAME_CONTPC,	-12
++|.define FRAME_CONTRET,	-16
++|.define WORD_LO,	0
++|.define WORD_HI,	4
++|.define WORD_BLO,	0
++|.define BASE_LO,	BASE
++|.define BASE_HI,	BASEP4
++|.macro lwzux2, hi, lo, base, idx
++|  lwzux lo, base, idx
++|  lwz hi, 4(base)
++|.endmacro
++|.else
+ |.define FRAME_PC,	-8
+ |.define FRAME_FUNC,	-4
++|.define FRAME_CONTPC,	-16
++|.define FRAME_CONTRET,	-12
++|.define WORD_LO,	4
++|.define WORD_HI,	0
++|.define WORD_BLO,	7
++|.define BASE_LO,	BASEP4
++|.define BASE_HI,	BASE
++|.macro lwzux2, hi, lo, base, idx
++|  lwzux hi, base, idx
++|  lwz lo, 4(base)
++|.endmacro
++|.endif
+ |
+ |// Instruction decode.
+ |.macro decode_OP4, dst, ins; rlwinm dst, ins, 2, 22, 29; .endmacro
+@@ -412,6 +503,7 @@
+ |// Call decode and dispatch.
+ |.macro ins_callt
+ |  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC
++|  addi BASEP4, BASE, 4
+ |  lwz PC, LFUNC:RB->pc
+ |  lwz INS, 0(PC)
+ |   addi PC, PC, 4
+@@ -504,7 +596,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lwz PC, FRAME_PC(TMP2)		// Fetch PC of previous frame.
+   |  mr BASE, TMP2			// Restore caller base.
+   |  // Prepending may overwrite the pcall frame, so do it at the end.
+-  |   stwu TMP1, FRAME_PC(RA)		// Prepend true to results.
++  |  .if ENDIAN_LE
++  |    addi RA, RA, -8
++  |    stw TMP1, WORD_HI(RA)		// Prepend true to results.
++  |  .else
++  |    stwu TMP1, -8(RA)		// Prepend true to results.
++  |  .endif
+   |
+   |->vm_returnc:
+   |  addi RD, RD, 8			// RD = (nresults+1)*8.
+@@ -560,7 +657,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lwz TMP1, L->maxstack
+   |  cmplw BASE, TMP1
+   |  bge >8
+-  |  stw TISNIL, 0(BASE)
++  |  stw TISNIL, WORD_HI(BASE)
+   |  addi RD, RD, 8
+   |  addi BASE, BASE, 8
+   |  b <2
+@@ -611,7 +708,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |->vm_unwind_ff_eh:			// Landing pad for external unwinder.
+   |  lwz L, SAVE_L
+   |  .toc ld TOCREG, SAVE_TOC
++  |.if P64
++  |     lus TISNUM, LJ_TISNUM >> 16	// Setup type comparison constants.
++  |     ori TISNUM, TISNUM, LJ_TISNUM & 0xffff
++  |.else
+   |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
++  |.endif
+   |  lp BASE, L->base
+   |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
+   |   lwz DISPATCH, L->glref		// Setup pointer to dispatch table.
+@@ -626,7 +728,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la RA, -8(BASE)			// Results start at BASE-8.
+   |     stw TMP3, TMPD
+   |   addi DISPATCH, DISPATCH, GG_G2DISP
+-  |  stw TMP1, 0(RA)			// Prepend false to error message.
++  |  stw TMP1, WORD_HI(RA)		// Prepend false to error message.
+   |  li RD, 16				// 2 results: false + error message.
+   |    st_vmstate
+   |     lfs TONUM, TMPD
+@@ -687,7 +789,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |  stw L, DISPATCH_GL(cur_L)(DISPATCH)
+   |  mr RA, BASE
+   |   lp BASE, L->base
++  |.if P64
++  |     lus TISNUM, LJ_TISNUM >> 16	// Setup type comparison constants.
++  |     ori TISNUM, TISNUM, LJ_TISNUM & 0xffff
++  |.else
+   |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
++  |.endif
+   |   lp TMP1, L->top
+   |  lwz PC, FRAME_PC(BASE)
+   |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
+@@ -737,7 +844,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |3:  // Entry point for vm_cpcall/vm_resume (BASE = base, PC = ftype).
+   |  stw L, DISPATCH_GL(cur_L)(DISPATCH)
+   |  lp TMP2, L->base			// TMP2 = old base (used in vmeta_call).
++  |.if P64
++  |     lus TISNUM, LJ_TISNUM >> 16	// Setup type comparison constants.
++  |     ori TISNUM, TISNUM, LJ_TISNUM & 0xffff
++  |.else
+   |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
++  |.endif
+   |   lp TMP1, L->top
+   |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
+   |  add PC, PC, BASE
+@@ -757,8 +869,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |->vm_call_dispatch:
+   |  // TMP2 = old base, BASE = new base, RC = nargs*8, PC = caller PC
+-  |  lwz TMP0, FRAME_PC(BASE)
+-  |   lwz LFUNC:RB, FRAME_FUNC(BASE)
++  |  lwz TMP0, WORD_HI-8(BASE)
++  |   lwz LFUNC:RB, WORD_LO-8(BASE)
+   |  checkfunc TMP0; bne ->vmeta_call
+   |
+   |->vm_call_dispatch_f:
+@@ -777,7 +889,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |   sub TMP0, TMP0, TMP1		// Compute -savestack(L, L->top).
+   |    lp TMP1, L->cframe
+   |     addi DISPATCH, DISPATCH, GG_G2DISP
+-  |  .toc lp CARG4, 0(CARG4)
++  |  .opd lp TOCREG, TOC_OFS(CARG4)
++  |  .opdenv lp ENVREG, ENV_OFS(CARG4)
++  |  .opd lp CARG4, 0(CARG4)
+   |  li TMP2, 0
+   |   stw TMP0, SAVE_NRES		// Neg. delta means cframe w/o frame.
+   |  stw TMP2, SAVE_ERRF		// No error function.
+@@ -785,7 +899,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |    stp sp, L->cframe		// Add our C frame to cframe chain.
+   |     stw L, DISPATCH_GL(cur_L)(DISPATCH)
+   |  mtctr CARG4
++  |  .elfv2 mr FUNCREG, CARG4
+   |  bctrl			// (lua_State *L, lua_CFunction func, void *ud)
++  |  .toc lp TOCREG, SAVE_TOC
+   |.if PPE
+   |  mr BASE, CRET1
+   |  cmpwi CRET1, 0
+@@ -807,20 +923,27 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |->cont_dispatch:
+   |  // BASE = meta base, RA = resultptr, RD = (nresults+1)*8
+-  |  lwz TMP0, -12(BASE)		// Continuation.
++  |  lwz TMP0, FRAME_CONTRET(BASE)	// Continuation.
+   |   mr RB, BASE
+   |   mr BASE, TMP2			// Restore caller BASE.
+   |    lwz LFUNC:TMP1, FRAME_FUNC(TMP2)
+   |.if FFI
+   |  cmplwi TMP0, 1
+   |.endif
+-  |     lwz PC, -16(RB)			// Restore PC from [cont|PC].
+-  |   subi TMP2, RD, 8
++  |     lwz PC, FRAME_CONTPC(RB)	// Restore PC from [cont|PC].
++  |  addi BASEP4, BASE, 4
++  |   addi TMP2, RD, WORD_HI-8
+   |    lwz TMP1, LFUNC:TMP1->pc
+   |   stwx TISNIL, RA, TMP2		// Ensure one valid arg.
++  |.if P64
++  |   ld TMP3, 0(DISPATCH)
++  |.endif
+   |.if FFI
+   |  ble >1
+   |.endif
++  |.if P64
++  |  add TMP0, TMP0, TMP3
++  |.endif
+   |    lwz KBASE, PC2PROTO(k)(TMP1)
+   |  // BASE = base, RA = resultptr, RB = meta base
+   |  mtctr TMP0
+@@ -856,20 +979,20 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
+   |  li TMP0, LJ_TSTR
+   |   decode_RB8 RB, INS
+-  |  stw STR:RC, 4(CARG3)
++  |  stw STR:RC, WORD_LO(CARG3)
+   |   add CARG2, BASE, RB
+-  |  stw TMP0, 0(CARG3)
++  |  stw TMP0, WORD_HI(CARG3)
+   |  b >1
+   |
+   |->vmeta_tgets:
+   |  la CARG2, DISPATCH_GL(tmptv)(DISPATCH)
+   |  li TMP0, LJ_TTAB
+-  |  stw TAB:RB, 4(CARG2)
++  |  stw TAB:RB, WORD_LO(CARG2)
+   |   la CARG3, DISPATCH_GL(tmptv2)(DISPATCH)
+-  |  stw TMP0, 0(CARG2)
++  |  stw TMP0, WORD_HI(CARG2)
+   |   li TMP1, LJ_TSTR
+-  |   stw STR:RC, 4(CARG3)
+-  |   stw TMP1, 0(CARG3)
++  |   stw STR:RC, WORD_LO(CARG3)
++  |   stw TMP1, WORD_HI(CARG3)
+   |  b >1
+   |
+   |->vmeta_tgetb:			// TMP0 = index
+@@ -880,8 +1003,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
+   |   add CARG2, BASE, RB
+   |.if DUALNUM
+-  |  stw TISNUM, 0(CARG3)
+-  |  stw TMP0, 4(CARG3)
++  |  stw TISNUM, WORD_HI(CARG3)
++  |  stw TMP0, WORD_LO(CARG3)
+   |.else
+   |  stfd f0, 0(CARG3)
+   |.endif
+@@ -909,7 +1032,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  // BASE = base, L->top = new base, stack = cont/func/t/k
+   |  subfic TMP1, BASE, FRAME_CONT
+   |  lp BASE, L->top
+-  |  stw PC, -16(BASE)			// [cont|PC]
++  |  stw PC, FRAME_CONTPC(BASE)		// [cont|PC]
+   |   add PC, TMP1, BASE
+   |  lwz LFUNC:RB, FRAME_FUNC(BASE)	// Guaranteed to be a function here.
+   |   li NARGS8:RC, 16			// 2 args for func(t, k).
+@@ -923,7 +1046,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lfd f14, 0(CRET1)
+   |  b ->BC_TGETR_Z
+   |1:
+-  |  stwx TISNIL, BASE, RA
++  |  stwx TISNIL, BASE_HI, RA
+   |  b ->cont_nop
+   |
+   |//-----------------------------------------------------------------------
+@@ -932,20 +1055,20 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
+   |  li TMP0, LJ_TSTR
+   |   decode_RB8 RB, INS
+-  |  stw STR:RC, 4(CARG3)
++  |  stw STR:RC, WORD_LO(CARG3)
+   |   add CARG2, BASE, RB
+-  |  stw TMP0, 0(CARG3)
++  |  stw TMP0, WORD_HI(CARG3)
+   |  b >1
+   |
+   |->vmeta_tsets:
+   |  la CARG2, DISPATCH_GL(tmptv)(DISPATCH)
+   |  li TMP0, LJ_TTAB
+-  |  stw TAB:RB, 4(CARG2)
++  |  stw TAB:RB, WORD_LO(CARG2)
+   |   la CARG3, DISPATCH_GL(tmptv2)(DISPATCH)
+-  |  stw TMP0, 0(CARG2)
++  |  stw TMP0, WORD_HI(CARG2)
+   |   li TMP1, LJ_TSTR
+-  |   stw STR:RC, 4(CARG3)
+-  |   stw TMP1, 0(CARG3)
++  |   stw STR:RC, WORD_LO(CARG3)
++  |   stw TMP1, WORD_HI(CARG3)
+   |  b >1
+   |
+   |->vmeta_tsetb:			// TMP0 = index
+@@ -956,8 +1079,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
+   |   add CARG2, BASE, RB
+   |.if DUALNUM
+-  |  stw TISNUM, 0(CARG3)
+-  |  stw TMP0, 4(CARG3)
++  |  stw TISNUM, WORD_HI(CARG3)
++  |  stw TMP0, WORD_LO(CARG3)
+   |.else
+   |  stfd f0, 0(CARG3)
+   |.endif
+@@ -986,7 +1109,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  // BASE = base, L->top = new base, stack = cont/func/t/k/(v)
+   |  subfic TMP1, BASE, FRAME_CONT
+   |  lp BASE, L->top
+-  |  stw PC, -16(BASE)			// [cont|PC]
++  |  stw PC, FRAME_CONTPC(BASE)		// [cont|PC]
+   |   add PC, TMP1, BASE
+   |  lwz LFUNC:RB, FRAME_FUNC(BASE)	// Guaranteed to be a function here.
+   |   li NARGS8:RC, 24			// 3 args for func(t, k, v)
+@@ -1006,17 +1129,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |->vmeta_comp:
+   |  mr CARG1, L
+   |   subi PC, PC, 4
+-  |.if DUALNUM
+-  |  mr CARG2, RA
+-  |.else
+   |  add CARG2, BASE, RA
+-  |.endif
+   |   stw PC, SAVE_PC
+-  |.if DUALNUM
+-  |  mr CARG3, RD
+-  |.else
+   |  add CARG3, BASE, RD
+-  |.endif
+   |   stp BASE, L->base
+   |  decode_OP1 CARG4, INS
+   |  bl extern lj_meta_comp  // (lua_State *L, TValue *o1, *o2, int op)
+@@ -1043,7 +1158,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  b ->cont_nop
+   |
+   |->cont_condt:			// RA = resultptr
+-  |  lwz TMP0, 0(RA)
++  |  lwz TMP0, WORD_HI(RA)
+   |  .gpr64 extsw TMP0, TMP0
+   |  subfic TMP0, TMP0, LJ_TTRUE	// Branch if result is true.
+   |  subfe CRET1, CRET1, CRET1
+@@ -1051,7 +1166,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  b <4
+   |
+   |->cont_condf:			// RA = resultptr
+-  |  lwz TMP0, 0(RA)
++  |  lwz TMP0, WORD_HI(RA)
+   |  .gpr64 extsw TMP0, TMP0
+   |  subfic TMP0, TMP0, LJ_TTRUE	// Branch if result is false.
+   |  subfe CRET1, CRET1, CRET1
+@@ -1103,8 +1218,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |.endif
+   |
+   |->vmeta_unm:
+-  |  mr CARG3, RD
+-  |  mr CARG4, RD
++  |  add CARG3, BASE, RD
++  |  add CARG4, BASE, RD
+   |  b >1
+   |
+   |->vmeta_arith_vn:
+@@ -1139,7 +1254,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |->vmeta_binop:
+   |  // BASE = old base, CRET1 = new base, stack = cont/func/o1/o2
+   |  sub TMP1, CRET1, BASE
+-  |   stw PC, -16(CRET1)		// [cont|PC]
++  |   stw PC, FRAME_CONTPC(CRET1)	// [cont|PC]
+   |   mr TMP2, BASE
+   |  addi PC, TMP1, FRAME_CONT
+   |   mr BASE, CRET1
+@@ -1150,7 +1265,7 @@ static void build_subroutines(BuildCtx *ctx)
+ #if LJ_52
+   |  mr SAVE0, CARG1
+ #endif
+-  |  mr CARG2, RD
++  |  add CARG2, BASE, RD
+   |   stp BASE, L->base
+   |  mr CARG1, L
+   |   stw PC, SAVE_PC
+@@ -1227,25 +1342,25 @@ static void build_subroutines(BuildCtx *ctx)
+   |.macro .ffunc_1, name
+   |->ff_ .. name:
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
+-  |    lwz CARG1, 4(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
++  |    lwz CARG1, WORD_LO(BASE)
+   |  blt ->fff_fallback
+   |.endmacro
+   |
+   |.macro .ffunc_2, name
+   |->ff_ .. name:
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG3, 0(BASE)
+-  |    lwz CARG4, 8(BASE)
+-  |   lwz CARG1, 4(BASE)
+-  |    lwz CARG2, 12(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
++  |    lwz CARG4, WORD_HI+8(BASE)
++  |   lwz CARG1, WORD_LO(BASE)
++  |    lwz CARG2, WORD_LO+8(BASE)
+   |  blt ->fff_fallback
+   |.endmacro
+   |
+   |.macro .ffunc_n, name
+   |->ff_ .. name:
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
+   |    lfd FARG1, 0(BASE)
+   |  blt ->fff_fallback
+   |  checknum CARG3; bge ->fff_fallback
+@@ -1254,9 +1369,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |.macro .ffunc_nn, name
+   |->ff_ .. name:
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG3, 0(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
+   |    lfd FARG1, 0(BASE)
+-  |   lwz CARG4, 8(BASE)
++  |   lwz CARG4, WORD_HI+8(BASE)
+   |    lfd FARG2, 8(BASE)
+   |  blt ->fff_fallback
+   |  checknum CARG3; bge ->fff_fallback
+@@ -1279,9 +1394,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |  cmplw cr1, CARG3, TMP1
+   |    lwz PC, FRAME_PC(BASE)
+   |  bge cr1, ->fff_fallback
+-  |   stw CARG3, 0(RA)
++  |   stw CARG3, WORD_HI(RA)
+   |  addi RD, NARGS8:RC, 8		// Compute (nresults+1)*8.
+-  |   stw CARG1, 4(RA)
++  |   stw CARG1, WORD_LO(RA)
+   |  beq ->fff_res			// Done if exactly 1 argument.
+   |  li TMP1, 8
+   |  subi RC, RC, 8
+@@ -1295,17 +1410,36 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc type
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG1, 0(BASE)
++  |   lwz CARG1, WORD_HI(BASE)
+   |  blt ->fff_fallback
+   |  .gpr64 extsw CARG1, CARG1
++  |.if P64
++  |  li TMP0, LJ_TNUMX
++  |    srawi TMP3, CARG1, 15
++  |  subfc TMP1, TMP0, CARG1
++  |.else
+   |  subfc TMP0, TISNUM, CARG1
+-  |  subfe TMP2, CARG1, CARG1
++  |.endif
++  |    subfe TMP2, CARG1, CARG1
++  |.if P64
++  |  cmpwi TMP3, -2
++  |    orc TMP1, TMP2, TMP1
++  |    subf TMP1, TMP0, TMP1
++  |  beq >1
++  |.else
+   |  orc TMP1, TMP2, TMP0
+-  |  addi TMP1, TMP1, ~LJ_TISNUM+1
++  |  subf TMP1, TISNUM, TMP1
++  |.endif
+   |  slwi TMP1, TMP1, 3
++  |2:
+   |   la TMP2, CFUNC:RB->upvalue
+   |  lfdx FARG1, TMP2, TMP1
+   |  b ->fff_resn
++  |.if P64
++  |1:
++  |  li TMP1, ~LJ_TLIGHTUD<<3
++  |  b <2
++  |.endif
+   |
+   |//-- Base library: getters and setters ---------------------------------
+   |
+@@ -1328,10 +1462,10 @@ static void build_subroutines(BuildCtx *ctx)
+   |  sub TMP1, TMP0, TMP1
+   |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
+   |3:  // Rearranged logic, because we expect _not_ to find the key.
+-  |  lwz CARG4, NODE:TMP2->key
+-  |   lwz TMP0, 4+offsetof(Node, key)(NODE:TMP2)
+-  |    lwz CARG2, NODE:TMP2->val
+-  |     lwz TMP1, 4+offsetof(Node, val)(NODE:TMP2)
++  |  lwz CARG4, WORD_HI+offsetof(Node, key)(NODE:TMP2)
++  |   lwz TMP0, WORD_LO+offsetof(Node, key)(NODE:TMP2)
++  |    lwz CARG2, WORD_HI+offsetof(Node, val)(NODE:TMP2)
++  |     lwz TMP1, WORD_LO+offsetof(Node, val)(NODE:TMP2)
+   |  checkstr CARG4; bne >4
+   |   cmpw TMP0, STR:RC; beq >5
+   |4:
+@@ -1349,14 +1483,33 @@ static void build_subroutines(BuildCtx *ctx)
+   |6:
+   |  cmpwi CARG3, LJ_TUDATA; beq <1
+   |  .gpr64 extsw CARG3, CARG3
++  |.if P64
++  |  li TMP0, LJ_TNUMX
++  |    srawi TMP3, CARG3, 15
++  |  subfc TMP1, TMP0, CARG3
++  |.else
+   |  subfc TMP0, TISNUM, CARG3
++  |.endif
+   |  subfe TMP2, CARG3, CARG3
++  |.if P64
++  |  cmpwi TMP3, -2
++  |    orc TMP1, TMP2, TMP1
++  |    subf TMP1, TMP0, TMP1
++  |  beq >7
++  |.else
+   |  orc TMP1, TMP2, TMP0
+-  |  addi TMP1, TMP1, ~LJ_TISNUM+1
++  |  subf TMP1, TISNUM, TMP1
++  |.endif
+   |  slwi TMP1, TMP1, 2
++  |8:
+   |   la TMP2, DISPATCH_GL(gcroot[GCROOT_BASEMT])(DISPATCH)
+   |  lwzx TAB:CARG1, TMP2, TMP1
+   |  b <2
++  |.if P64
++  |7:
++  |  li TMP1, ~LJ_TLIGHTUD<<2
++  |  b <8
++  |.endif
+   |
+   |.ffunc_2 setmetatable
+   |  // Fast path: no mt for table yet and not clearing the mt.
+@@ -1374,8 +1527,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc rawget
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG4, 0(BASE)
+-  |    lwz TAB:CARG2, 4(BASE)
++  |   lwz CARG4, WORD_HI(BASE)
++  |    lwz TAB:CARG2, WORD_LO(BASE)
+   |  blt ->fff_fallback
+   |  checktab CARG4; bne ->fff_fallback
+   |   la CARG3, 8(BASE)
+@@ -1390,7 +1543,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |.ffunc tonumber
+   |  // Only handles the number case inline (without a base argument).
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG1, 0(BASE)
++  |   lwz CARG1, WORD_HI(BASE)
+   |    lfd FARG1, 0(BASE)
+   |  bne ->fff_fallback			// Exactly one argument.
+   |   checknum CARG1; bgt ->fff_fallback
+@@ -1425,10 +1578,15 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc next
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG1, 0(BASE)
+-  |    lwz TAB:CARG2, 4(BASE)
++  |   lwz CARG1, WORD_HI(BASE)
++  |    lwz TAB:CARG2, WORD_LO(BASE)
+   |  blt ->fff_fallback
++  |.if ENDIAN_LE
++  |   add TMP1, BASE, NARGS8:RC
++  |   stw TISNIL, WORD_HI(TMP1)		// Set missing 2nd arg to nil.
++  |.else
+   |   stwx TISNIL, BASE, NARGS8:RC	// Set missing 2nd arg to nil.
++  |.endif
+   |  checktab CARG1
+   |   lwz PC, FRAME_PC(BASE)
+   |  bne ->fff_fallback
+@@ -1464,18 +1622,18 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lfd f0, CFUNC:RB->upvalue[0]
+   |  la RA, -8(BASE)
+ #endif
+-  |   stw TISNIL, 8(BASE)
++  |   stw TISNIL, 8+WORD_HI(BASE)
+   |  li RD, (3+1)*8
+   |  stfd f0, 0(RA)
+   |  b ->fff_res
+   |
+   |.ffunc ipairs_aux
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG3, 0(BASE)
+-  |    lwz TAB:CARG1, 4(BASE)
+-  |   lwz CARG4, 8(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
++  |    lwz TAB:CARG1, WORD_LO(BASE)
++  |   lwz CARG4, 8+WORD_HI(BASE)
+   |.if DUALNUM
+-  |    lwz TMP2, 12(BASE)
++  |    lwz TMP2, 8+WORD_LO(BASE)
+   |.else
+   |    lfd FARG2, 8(BASE)
+   |.endif
+@@ -1504,16 +1662,16 @@ static void build_subroutines(BuildCtx *ctx)
+   |   la RA, -8(BASE)
+   |  cmplw TMP0, TMP2
+   |.if DUALNUM
+-  |  stw TISNUM, 0(RA)
++  |  stw TISNUM, WORD_HI(RA)
+   |   slwi TMP3, TMP2, 3
+-  |  stw TMP2, 4(RA)
++  |  stw TMP2, WORD_LO(RA)
+   |.else
+   |   slwi TMP3, TMP2, 3
+   |  stfd FARG2, 0(RA)
+   |.endif
+   |  ble >2				// Not in array part?
+-  |  lwzx TMP2, TMP1, TMP3
+-  |  lfdx f0, TMP1, TMP3
++  |  lfdux f0, TMP1, TMP3
++  |  lwz TMP2, WORD_HI(TMP1)
+   |1:
+   |  checknil TMP2
+   |   li RD, (0+1)*8
+@@ -1532,7 +1690,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  cmplwi CRET1, 0
+   |   li RD, (0+1)*8
+   |  beq ->fff_res
+-  |  lwz TMP2, 0(CRET1)
++  |  lwz TMP2, WORD_HI(CRET1)
+   |  lfd f0, 0(CRET1)
+   |  b <1
+   |
+@@ -1551,11 +1709,11 @@ static void build_subroutines(BuildCtx *ctx)
+   |  la RA, -8(BASE)
+ #endif
+   |.if DUALNUM
+-  |  stw TISNUM, 8(BASE)
++  |  stw TISNUM, 8+WORD_HI(BASE)
+   |.else
+-  |  stw ZERO, 8(BASE)
++  |  stw ZERO, 8+WORD_HI(BASE)
+   |.endif
+-  |   stw ZERO, 12(BASE)
++  |   stw ZERO, 8+WORD_LO(BASE)
+   |  li RD, (3+1)*8
+   |  stfd f0, 0(RA)
+   |  b ->fff_res
+@@ -1576,7 +1734,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc xpcall
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG4, 8(BASE)
++  |   lwz CARG4, 8+WORD_HI(BASE)
+   |    lfd FARG2, 8(BASE)
+   |    lfd FARG1, 0(BASE)
+   |  blt ->fff_fallback
+@@ -1673,7 +1831,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |.if resume
+   |  li TMP1, LJ_TTRUE
+   |   la RA, -8(BASE)
+-  |  stw TMP1, -8(BASE)			// Prepend true to results.
++  |  stw TMP1, WORD_HI-8(BASE)		// Prepend true to results.
+   |  addi RD, RD, 16
+   |.else
+   |  mr RA, BASE
+@@ -1693,7 +1851,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lfd f0, 0(TMP3)
+   |   stp TMP3, L:SAVE0->top		// Remove error from coroutine stack.
+   |    li RD, (2+1)*8
+-  |   stw TMP1, -8(BASE)		// Prepend false to results.
++  |   stw TMP1, WORD_HI-8(BASE)		// Prepend false to results.
+   |    la RA, -8(BASE)
+   |  stfd f0, 0(BASE)			// Copy error message.
+   |  b <7
+@@ -1746,8 +1904,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |->fff_resi:
+   |  lwz PC, FRAME_PC(BASE)
+   |  la RA, -8(BASE)
+-  |  stw TISNUM, -8(BASE)
+-  |  stw CRET1, -4(BASE)
++  |  stw TISNUM, WORD_HI-8(BASE)
++  |  stw CRET1, WORD_LO-8(BASE)
+   |  b ->fff_res1
+   |1:
+   |  lus CARG3, 0x41e0	// 2^31.
+@@ -1762,9 +1920,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |->fff_restv:
+   |  // CARG3/CARG1 = TValue result.
+   |  lwz PC, FRAME_PC(BASE)
+-  |   stw CARG3, -8(BASE)
++  |   stw CARG3, WORD_HI-8(BASE)
+   |  la RA, -8(BASE)
+-  |   stw CARG1, -4(BASE)
++  |   stw CARG1, WORD_LO-8(BASE)
+   |->fff_res1:
+   |  // RA = results, PC = return.
+   |  li RD, (1+1)*8
+@@ -1782,10 +1940,11 @@ static void build_subroutines(BuildCtx *ctx)
+   |  ins_next1
+   |  // Adjust BASE. KBASE is assumed to be set for the calling frame.
+   |   sub BASE, RA, TMP0
++  |   addi BASEP4, BASE, 4
+   |  ins_next2
+   |
+   |6:  // Fill up results with nil.
+-  |  subi TMP1, RD, 8
++  |  addi TMP1, RD, WORD_HI-8
+   |   addi RD, RD, 8
+   |  stwx TISNIL, RA, TMP1
+   |  b <5
+@@ -1898,7 +2057,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc math_log
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
+   |    lfd FARG1, 0(BASE)
+   |  bne ->fff_fallback			// Need exactly 1 argument.
+   |  checknum CARG3; bge ->fff_fallback
+@@ -1923,13 +2082,13 @@ static void build_subroutines(BuildCtx *ctx)
+   |.if DUALNUM
+   |.ffunc math_ldexp
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG3, 0(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
+   |    lfd FARG1, 0(BASE)
+-  |   lwz CARG4, 8(BASE)
++  |   lwz CARG4, WORD_HI+8(BASE)
+   |.if GPR64
+-  |    lwz CARG2, 12(BASE)
++  |    lwz CARG2, WORD_LO+8(BASE)
+   |.else
+-  |    lwz CARG1, 12(BASE)
++  |    lwz CARG1, WORD_LO+8(BASE)
+   |.endif
+   |  blt ->fff_fallback
+   |  checknum CARG3; bge ->fff_fallback
+@@ -1961,8 +2120,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |  stfd FARG1, 0(RA)
+   |  li RD, (2+1)*8
+   |.if DUALNUM
+-  |   stw TISNUM, 8(RA)
+-  |   stw TMP1, 12(RA)
++  |   stw TISNUM, WORD_HI+8(RA)
++  |   stw TMP1, WORD_LO+8(RA)
+   |.else
+   |   stfd FARG2, 8(RA)
+   |.endif
+@@ -1989,9 +2148,9 @@ static void build_subroutines(BuildCtx *ctx)
+   |   add TMP2, BASE, NARGS8:RC
+   |  bne >4
+   |1:  // Handle integers.
+-  |  lwz CARG4, 0(TMP1)
++  |  lwz CARG4, WORD_HI(TMP1)
+   |   cmplw cr1, TMP1, TMP2
+-  |  lwz CARG2, 4(TMP1)
++  |  lwz CARG2, WORD_LO(TMP1)
+   |   bge cr1, ->fff_resi
+   |  checknum CARG4
+   |   xoris TMP0, CARG1, 0x8000
+@@ -2020,7 +2179,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |   lfd FARG1, 0(BASE)
+   |  bge ->fff_fallback
+   |5:  // Handle numbers.
+-  |  lwz CARG4, 0(TMP1)
++  |  lwz CARG4, WORD_HI(TMP1)
+   |   cmplw cr1, TMP1, TMP2
+   |  lfd FARG2, 0(TMP1)
+   |   bge cr1, ->fff_resn
+@@ -2035,7 +2194,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |.endif
+   |  b <5
+   |7:  // Convert integer to number and continue above.
+-  |   lwz CARG2, 4(TMP1)
++  |   lwz CARG2, WORD_LO(TMP1)
+   |  bne ->fff_fallback
+   |  tonum_i FARG2, CARG2
+   |  b <6
+@@ -2043,7 +2202,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |  .ffunc_n name
+   |  li TMP1, 8
+   |1:
++  |.if ENDIAN_LE
++  |   add CARG2, BASE, TMP1
++  |   lwz CARG2, WORD_HI(CARG2)
++  |.else
+   |   lwzx CARG2, BASE, TMP1
++  |.endif
+   |   lfdx FARG2, BASE, TMP1
+   |  cmplw cr1, TMP1, NARGS8:RC
+   |   checknum CARG2
+@@ -2067,8 +2231,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |.ffunc string_byte			// Only handle the 1-arg case here.
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
+-  |    lwz STR:CARG1, 4(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
++  |    lwz STR:CARG1, WORD_LO(BASE)
+   |  bne ->fff_fallback			// Need exactly 1 argument.
+   |   checkstr CARG3
+   |   bne ->fff_fallback
+@@ -2099,12 +2263,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |.ffunc string_char			// Only handle the 1-arg case here.
+   |  ffgccheck
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
+   |.if DUALNUM
+-  |    lwz TMP0, 4(BASE)
++  |    lwz TMP0, WORD_LO(BASE)
+   |  bne ->fff_fallback			// Exactly 1 argument.
+   |  checknum CARG3; bne ->fff_fallback
+-  |   la CARG2, 7(BASE)
++  |   la CARG2, WORD_BLO(BASE)
+   |.else
+   |    lfd FARG1, 0(BASE)
+   |  bne ->fff_fallback			// Exactly 1 argument.
+@@ -2128,16 +2292,16 @@ static void build_subroutines(BuildCtx *ctx)
+   |.ffunc string_sub
+   |  ffgccheck
+   |  cmplwi NARGS8:RC, 16
+-  |   lwz CARG3, 16(BASE)
++  |   lwz CARG3, WORD_HI+16(BASE)
+   |.if not DUALNUM
+   |    lfd f0, 16(BASE)
+   |.endif
+-  |   lwz TMP0, 0(BASE)
+-  |    lwz STR:CARG1, 4(BASE)
++  |   lwz TMP0, WORD_HI(BASE)
++  |    lwz STR:CARG1, WORD_LO(BASE)
+   |  blt ->fff_fallback
+-  |   lwz CARG2, 8(BASE)
++  |   lwz CARG2, WORD_HI+8(BASE)
+   |.if DUALNUM
+-  |    lwz TMP1, 12(BASE)
++  |    lwz TMP1, WORD_LO+8(BASE)
+   |.else
+   |    lfd f1, 8(BASE)
+   |.endif
+@@ -2145,7 +2309,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  beq >1
+   |.if DUALNUM
+   |  checknum CARG3
+-  |   lwz TMP2, 20(BASE)
++  |   lwz TMP2, WORD_LO+16(BASE)
+   |  bne ->fff_fallback
+   |1:
+   |  checknum CARG2; bne ->fff_fallback
+@@ -2201,8 +2365,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |  .ffunc string_ .. name
+   |  ffgccheck
+   |  cmplwi NARGS8:RC, 8
+-  |   lwz CARG3, 0(BASE)
+-  |    lwz STR:CARG2, 4(BASE)
++  |   lwz CARG3, WORD_HI(BASE)
++  |    lwz STR:CARG2, WORD_LO(BASE)
+   |  blt ->fff_fallback
+   |  checkstr CARG3
+   |   la SBUF:CARG1, DISPATCH_GL(tmpbuf)(DISPATCH)
+@@ -2240,10 +2404,10 @@ static void build_subroutines(BuildCtx *ctx)
+   |  addi TMP1, BASE, 8
+   |  add TMP2, BASE, NARGS8:RC
+   |1:
+-  |  lwz CARG4, 0(TMP1)
++  |  lwz CARG4, WORD_HI(TMP1)
+   |   cmplw cr1, TMP1, TMP2
+   |.if DUALNUM
+-  |  lwz CARG2, 4(TMP1)
++  |  lwz CARG2, WORD_LO(TMP1)
+   |.else
+   |  lfd FARG1, 0(TMP1)
+   |.endif
+@@ -2344,20 +2508,23 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |->fff_fallback:			// Call fast function fallback handler.
+   |  // BASE = new base, RB = CFUNC, RC = nargs*8
+-  |  lp TMP3, CFUNC:RB->f
++  |  lp FUNCREG, CFUNC:RB->f
+   |    add TMP1, BASE, NARGS8:RC
+   |   lwz PC, FRAME_PC(BASE)		// Fallback may overwrite PC.
+   |    addi TMP0, TMP1, 8*LUA_MINSTACK
+   |     lwz TMP2, L->maxstack
+   |   stw PC, SAVE_PC			// Redundant (but a defined value).
+-  |  .toc lp TMP3, 0(TMP3)
++  |  .opd lp TOCREG, TOC_OFS(FUNCREG)
++  |  .opdenv lp ENVREG, ENV_OFS(FUNCREG)
++  |  .opd lp FUNCREG, 0(FUNCREG)
+   |  cmplw TMP0, TMP2
+   |     stp BASE, L->base
+   |    stp TMP1, L->top
+   |   mr CARG1, L
+   |  bgt >5				// Need to grow stack.
+-  |  mtctr TMP3
++  |  mtctr FUNCREG
+   |  bctrl				// (lua_State *L)
++  |  .toc lp TOCREG, SAVE_TOC
+   |  // Either throws an error, or recovers and returns -1, 0 or nresults+1.
+   |  lp BASE, L->base
+   |  cmpwi CRET1, 0
+@@ -2459,6 +2626,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |3:
+   |  lp BASE, L->base
+   |4:  // Re-dispatch to static ins.
++  |  addi BASEP4, BASE, 4
+   |  lwz INS, -4(PC)
+   |  decode_OPP TMP1, INS
+   |   decode_RB8 RB, INS
+@@ -2472,7 +2640,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |
+   |->cont_hook:				// Continue from hook yield.
+   |  addi PC, PC, 4
+-  |  lwz MULTRES, -20(RB)		// Restore MULTRES for *M ins.
++  |  lwz MULTRES, WORD_LO-24(RB)		// Restore MULTRES for *M ins.
+   |  b <4
+   |
+   |->vm_hotloop:			// Hot loop counter underflow.
+@@ -2514,6 +2682,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lp BASE, L->base
+   |   lp TMP0, L->top
+   |   stw ZERO, SAVE_PC			// Invalidate for subsequent line hook.
++  |  addi BASEP4, BASE, 4
+   |  sub NARGS8:RC, TMP0, BASE
+   |  add RA, BASE, RA
+   |  lwz LFUNC:RB, FRAME_FUNC(BASE)
+@@ -2525,7 +2694,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |.if JIT
+   |  // RA = resultptr, RB = meta base
+   |  lwz INS, -4(PC)
+-  |    lwz TRACE:TMP2, -20(RB)		// Save previous trace.
++  |    lwz TRACE:TMP2, WORD_LO-24(RB)	// Save previous trace.
+   |   addic. TMP1, MULTRES, -8
+   |  decode_RA8 RC, INS			// Call base.
+   |   beq >2
+@@ -2560,10 +2729,16 @@ static void build_subroutines(BuildCtx *ctx)
+   |  mr CARG2, PC
+   |  bl extern lj_dispatch_stitch	// (jit_State *J, const BCIns *pc)
+   |  lp BASE, L->base
++  |  addi BASEP4, BASE, 4
+   |  b ->cont_nop
+   |
+   |9:
++  |.if ENDIAN_LE
++  |  addi BASEP4, BASE, 4
++  |  stwx TISNIL, BASEP4, RC
++  |.else
+   |  stwx TISNIL, BASE, RC
++  |.endif
+   |  addi RC, RC, 8
+   |  b <3
+   |.endif
+@@ -2578,6 +2753,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  // HOOK_PROFILE is off again, so re-dispatch to dynamic instruction.
+   |  lp BASE, L->base
+   |  subi PC, PC, 4
++  |  addi BASEP4, BASE, 4
+   |  b ->cont_nop
+ #endif
+   |
+@@ -2586,39 +2762,72 @@ static void build_subroutines(BuildCtx *ctx)
+   |//-----------------------------------------------------------------------
+   |
+   |.macro savex_, a, b, c, d
+-  |  stfd f..a, 16+a*8(sp)
+-  |  stfd f..b, 16+b*8(sp)
+-  |  stfd f..c, 16+c*8(sp)
+-  |  stfd f..d, 16+d*8(sp)
++  |  stfd f..a, EXIT_OFFSET+a*8(sp)
++  |  stfd f..b, EXIT_OFFSET+b*8(sp)
++  |  stfd f..c, EXIT_OFFSET+c*8(sp)
++  |  stfd f..d, EXIT_OFFSET+d*8(sp)
++  |.endmacro
++  |
++  |.macro saver, a
++  |  stp r..a, EXIT_OFFSET+32*8+a*PSIZE(sp)
+   |.endmacro
+   |
+   |->vm_exit_handler:
+   |.if JIT
+-  |  addi sp, sp, -(16+32*8+32*4)
+-  |  stmw r2, 16+32*8+2*4(sp)
++  |  addi sp, TMP0, sp, -(EXIT_OFFSET+32*8+32*PSIZE)
++  |  saver 3 // CARG1
++  |  saver 4 // CARG2
++  |  saver 5 // CARG3
++  |  saver 17 // DISPATCH
+   |    addi DISPATCH, JGL, -GG_DISP2G-32768
+   |    li CARG2, ~LJ_VMST_EXIT
+-  |   lwz CARG1, 16+32*8+32*4(sp)	// Get stack chain.
++  |   lp CARG1, EXIT_OFFSET+32*8+32*PSIZE(sp)	// Get stack chain.
+   |    stw CARG2, DISPATCH_GL(vmstate)(DISPATCH)
++  |  saver 2
++  |  saver 6
++  |  saver 7
++  |  saver 8
++  |  saver 9
++  |  saver 10
++  |  saver 11
++  |  saver 12
++  |  saver 13
+   |  savex_ 0,1,2,3
+-  |   stw CARG1, 0(sp)			// Store extended stack chain.
+-  |   clrso TMP1
++  |   stp CARG1, 0(sp)			// Store extended stack chain.
++
+   |  savex_ 4,5,6,7
+-  |   addi CARG2, sp, 16+32*8+32*4	// Recompute original value of sp.
++  |  saver 14
++  |  saver 15
++  |  saver 16
++  |  saver 18
++  |   addi CARG2, sp, EXIT_OFFSET+32*8+32*PSIZE	// Recompute original value of sp.
+   |  savex_ 8,9,10,11
+-  |   stw CARG2, 16+32*8+1*4(sp)	// Store sp in RID_SP.
++  |   stp CARG2, EXIT_OFFSET+32*8+1*PSIZE(sp)	// Store sp in RID_SP.
+   |  savex_ 12,13,14,15
+   |   mflr CARG3
+   |   li TMP1, 0
+   |  savex_ 16,17,18,19
+-  |   stw TMP1, 16+32*8+0*4(sp)		// Clear RID_TMP.
++  |   stw TMP1, EXIT_OFFSET+32*8+0*PSIZE(sp)		// Clear RID_TMP.
+   |  savex_ 20,21,22,23
+   |   lhz CARG4, 2(CARG3)		// Load trace number.
+   |  savex_ 24,25,26,27
+   |  lwz L, DISPATCH_GL(cur_L)(DISPATCH)
+   |  savex_ 28,29,30,31
++  |  saver 19
++  |  saver 20
++  |  saver 21
++  |  saver 22
++  |  saver 23
++  |  saver 24
++  |  saver 25
++  |  saver 26
++  |  saver 27
++  |  saver 28
++  |  saver 29
++  |  saver 30
++  |  saver 31
+   |   sub CARG3, TMP0, CARG3		// Compute exit number.
+-  |  lp BASE, DISPATCH_GL(jit_base)(DISPATCH)
++  |  lwz BASE, DISPATCH_GL(jit_base)(DISPATCH)
+   |   srwi CARG3, CARG3, 2
+   |  stp L, DISPATCH_J(L)(DISPATCH)
+   |   subi CARG3, CARG3, 2
+@@ -2627,11 +2836,11 @@ static void build_subroutines(BuildCtx *ctx)
+   |  stw TMP1, DISPATCH_GL(jit_base)(DISPATCH)
+   |  addi CARG1, DISPATCH, GG_DISP2J
+   |   stw CARG3, DISPATCH_J(exitno)(DISPATCH)
+-  |  addi CARG2, sp, 16
++  |  addi CARG2, sp, EXIT_OFFSET
+   |  bl extern lj_trace_exit		// (jit_State *J, ExitState *ex)
+   |  // Returns MULTRES (unscaled) or negated error code.
+   |  lp TMP1, L->cframe
+-  |  lwz TMP2, 0(sp)
++  |  lp TMP2, 0(sp)
+   |   lp BASE, L->base
+   |.if GPR64
+   |  rldicr sp, TMP1, 0, 61
+@@ -2639,7 +2848,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  rlwinm sp, TMP1, 0, 0, 29
+   |.endif
+   |   lwz PC, SAVE_PC			// Get SAVE_PC.
+-  |  stw TMP2, 0(sp)
++  |  stp TMP2, 0(sp)
+   |  stw L, SAVE_L			// Set SAVE_L (on-trace resume/yield).
+   |  b >1
+   |.endif
+@@ -2660,7 +2869,12 @@ static void build_subroutines(BuildCtx *ctx)
+   |    stw TMP2, DISPATCH_GL(jit_base)(DISPATCH)
+   |  lwz KBASE, PC2PROTO(k)(TMP1)
+   |  // Setup type comparison constants.
++  |.if P64
++  |  lus TISNUM, LJ_TISNUM >> 16
++  |  ori TISNUM, TISNUM, LJ_TISNUM & 0xffff
++  |.else
+   |  li TISNUM, LJ_TISNUM
++  |.endif
+   |  lus TMP3, 0x59c0			// TOBIT = 2^52 + 2^51 (float).
+   |  stw TMP3, TMPD
+   |  li ZERO, 0
+@@ -2680,14 +2894,14 @@ static void build_subroutines(BuildCtx *ctx)
+   |   decode_RA8 RA, INS
+   |  lpx TMP0, DISPATCH, TMP1
+   |  mtctr TMP0
+-  |  cmplwi TMP1, BC_FUNCF*4		// Function header?
++  |  cmplwi TMP1, BC_FUNCF*PSIZE	// Function header?
+   |  bge >2
+   |   decode_RB8 RB, INS
+   |   decode_RD8 RD, INS
+   |   decode_RC8 RC, INS
+   |  bctr
+   |2:
+-  |  cmplwi TMP1, (BC_FUNCC+2)*4	// Fast function?
++  |  cmplwi TMP1, (BC_FUNCC+2)*PSIZE	// Fast function?
+   |  blt >3
+   |  // Check frame below fast function.
+   |  lwz TMP1, FRAME_PC(BASE)
+@@ -2697,7 +2911,7 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lwz TMP2, -4(TMP1)
+   |  decode_RA8 TMP0, TMP2
+   |  sub TMP1, BASE, TMP0
+-  |  lwz LFUNC:TMP2, -12(TMP1)
++  |  lwz LFUNC:TMP2, WORD_LO-16(TMP1)
+   |  lwz TMP1, LFUNC:TMP2->pc
+   |  lwz KBASE, PC2PROTO(k)(TMP1)
+   |3:
+@@ -2718,6 +2932,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |// NYI: Use internal implementations of floor, ceil, trunc.
+   |
+   |->vm_modi:
++  |  li TMP1, 0
++  |  mtxer TMP1
+   |  divwo. TMP0, CARG1, CARG2
+   |  bso >1
+   |.if GPR64
+@@ -2736,7 +2952,8 @@ static void build_subroutines(BuildCtx *ctx)
+   |  cmpwi CARG2, 0
+   |   li CARG1, 0
+   |  beqlr
+-  |  clrso TMP0			// Clear SO for -2147483648 % -1 and return 0.
++  |  // Clear SO for -2147483648 % -1 and return 0.
++  |  crxor 4*cr0+so, 4*cr0+so, 4*cr0+so
+   |  blr
+   |
+   |//-----------------------------------------------------------------------
+@@ -2749,10 +2966,18 @@ static void build_subroutines(BuildCtx *ctx)
+   |->vm_cachesync:
+   |.if JIT or FFI
+   |  // Compute start of first cache line and number of cache lines.
++  |  .if GPR64
++  |  rldicr CARG1, CARG1, 0, 58
++  |  .else
+   |  rlwinm CARG1, CARG1, 0, 0, 26
++  |  .endif
+   |  sub CARG2, CARG2, CARG1
+   |  addi CARG2, CARG2, 31
++  |  .if GPR64
++  |  srdi. CARG2, CARG2, 5
++  |  .else
+   |  rlwinm. CARG2, CARG2, 27, 5, 31
++  |  .endif
+   |  beqlr
+   |  mtctr CARG2
+   |  mr CARG3, CARG1
+@@ -2774,39 +2999,70 @@ static void build_subroutines(BuildCtx *ctx)
+   |//-- FFI helper functions -----------------------------------------------
+   |//-----------------------------------------------------------------------
+   |
+-  |// Handler for callback functions. Callback slot number in r11, g in r12.
++  |// Handler for callback functions.
++  |// 32-bit: Callback slot number in r12, g in r11.
++  |// 64-bit v1: Callback slot number in bits 47+ of r11, g in 0-46, TOC in r2.
++  |// 64-bit v2: Callback slot number in bits 2-11 of r12, g in r11,
++  |// vm_ffi_callback in r2.
+   |->vm_ffi_callback:
+   |.if FFI
+   |.type CTSTATE, CTState, PC
++  |  .if OPD
++  |   rldicl r12, r11, 17, 47
++  |   rldicl r11, r11, 0, 17
++  |  .endif
++  |  .if ELFV2
++  |   rlwinm r12, r12, 30, 22, 31
++  |   addisl TOCREG, TOCREG, extern .TOC.-lj_vm_ffi_callback@ha
++  |   addil TOCREG, TOCREG, extern .TOC.-lj_vm_ffi_callback@l
++  |  .endif
+   |  saveregs
+-  |  lwz CTSTATE, GL:r12->ctype_state
+-  |   addi DISPATCH, r12, GG_G2DISP
+-  |  stw r11, CTSTATE->cb.slot
+-  |  stw r3, CTSTATE->cb.gpr[0]
++  |  lwz CTSTATE, GL:r11->ctype_state
++  |   addi DISPATCH, r11, GG_G2DISP
++  |  stw r12, CTSTATE->cb.slot
++  |  stp r3, CTSTATE->cb.gpr[0]
+   |   stfd f1, CTSTATE->cb.fpr[0]
+-  |  stw r4, CTSTATE->cb.gpr[1]
++  |  stp r4, CTSTATE->cb.gpr[1]
+   |   stfd f2, CTSTATE->cb.fpr[1]
+-  |  stw r5, CTSTATE->cb.gpr[2]
++  |  stp r5, CTSTATE->cb.gpr[2]
+   |   stfd f3, CTSTATE->cb.fpr[2]
+-  |  stw r6, CTSTATE->cb.gpr[3]
++  |  stp r6, CTSTATE->cb.gpr[3]
+   |   stfd f4, CTSTATE->cb.fpr[3]
+-  |  stw r7, CTSTATE->cb.gpr[4]
++  |  stp r7, CTSTATE->cb.gpr[4]
+   |   stfd f5, CTSTATE->cb.fpr[4]
+-  |  stw r8, CTSTATE->cb.gpr[5]
++  |  stp r8, CTSTATE->cb.gpr[5]
+   |   stfd f6, CTSTATE->cb.fpr[5]
+-  |  stw r9, CTSTATE->cb.gpr[6]
++  |  stp r9, CTSTATE->cb.gpr[6]
+   |   stfd f7, CTSTATE->cb.fpr[6]
+-  |  stw r10, CTSTATE->cb.gpr[7]
++  |  stp r10, CTSTATE->cb.gpr[7]
+   |   stfd f8, CTSTATE->cb.fpr[7]
++  |  .if GPR64
++  |   stfd f9, CTSTATE->cb.fpr[8]
++  |   stfd f10, CTSTATE->cb.fpr[9]
++  |   stfd f11, CTSTATE->cb.fpr[10]
++  |   stfd f12, CTSTATE->cb.fpr[11]
++  |   stfd f13, CTSTATE->cb.fpr[12]
++  |  .endif
++  |  .if ELFV2
++  |  addi TMP0, sp, CFRAME_SPACE+96
++  |  .elif GPR64
++  |  addi TMP0, sp, CFRAME_SPACE+112
++  |  .else
+   |  addi TMP0, sp, CFRAME_SPACE+8
+-  |  stw TMP0, CTSTATE->cb.stack
++  |  .endif
++  |  stp TMP0, CTSTATE->cb.stack
+   |   mr CARG1, CTSTATE
+   |  stw CTSTATE, SAVE_PC		// Any value outside of bytecode is ok.
+   |   mr CARG2, sp
+   |  bl extern lj_ccallback_enter	// (CTState *cts, void *cf)
+   |  // Returns lua_State *.
+   |  lp BASE, L:CRET1->base
++  |.if P64
++  |     lus TISNUM, LJ_TISNUM >> 16	// Setup type comparison constants.
++  |     ori TISNUM, TISNUM, LJ_TISNUM & 0xffff
++  |.else
+   |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
++  |.endif
+   |  lp RC, L:CRET1->top
+   |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
+   |     li ZERO, 0
+@@ -2835,9 +3091,21 @@ static void build_subroutines(BuildCtx *ctx)
+   |  mr CARG1, CTSTATE
+   |  mr CARG2, RA
+   |  bl extern lj_ccallback_leave	// (CTState *cts, TValue *o)
+-  |  lwz CRET1, CTSTATE->cb.gpr[0]
++  |  lp CRET1, CTSTATE->cb.gpr[0]
+   |  lfd FARG1, CTSTATE->cb.fpr[0]
+-  |  lwz CRET2, CTSTATE->cb.gpr[1]
++  |  lp CRET2, CTSTATE->cb.gpr[1]
++  |  .if GPR64
++  |    lfd FARG2, CTSTATE->cb.fpr[1]
++  |  .else
++  |    lp CARG3, CTSTATE->cb.gpr[2]
++  |    lp CARG4, CTSTATE->cb.gpr[3]
++  |  .endif
++  |  .elfv2 lfd f3, CTSTATE->cb.fpr[2]
++  |  .elfv2 lfd f4, CTSTATE->cb.fpr[3]
++  |  .elfv2 lfd f5, CTSTATE->cb.fpr[4]
++  |  .elfv2 lfd f6, CTSTATE->cb.fpr[5]
++  |  .elfv2 lfd f7, CTSTATE->cb.fpr[6]
++  |  .elfv2 lfd f8, CTSTATE->cb.fpr[7]
+   |  b ->vm_leave_unw
+   |.endif
+   |
+@@ -2850,23 +3118,46 @@ static void build_subroutines(BuildCtx *ctx)
+   |   lbz CARG2, CCSTATE->nsp
+   |   lbz CARG3, CCSTATE->nfpr
+   |  neg TMP1, TMP1
++  |  .if GPR64
++  |    std TMP0, 16(sp)
++  |  .else
+   |    stw TMP0, 4(sp)
++  |  .endif
+   |   cmpwi cr1, CARG3, 0
+   |  mr TMP2, sp
+   |   addic. CARG2, CARG2, -1
++  |  .if GPR64
++  |  stdux sp, sp, TMP1
++  |  .else
+   |  stwux sp, sp, TMP1
++  |  .endif
+   |   crnot 4*cr1+eq, 4*cr1+eq		// For vararg calls.
+-  |  stw r14, -4(TMP2)
+-  |  stw CCSTATE, -8(TMP2)
++  |  .if GPR64
++  |    std r14, -8(TMP2)
++  |    std CCSTATE, -16(TMP2)
++  |  .else
++  |    stw r14, -4(TMP2)
++  |    stw CCSTATE, -8(TMP2)
++  |  .endif
+   |  mr r14, TMP2
+   |  la TMP1, CCSTATE->stack
++  |  .if GPR64
++  |   sldi CARG2, CARG2, 3
++  |  .else
+   |   slwi CARG2, CARG2, 2
++  |  .endif
+   |   blty >2
+-  |  la TMP2, 8(sp)
++  |  .if ELFV2
++  |    la TMP2, 96(sp)
++  |  .elif GPR64
++  |    la TMP2, 112(sp)
++  |  .else
++  |    la TMP2, 8(sp)
++  |  .endif
+   |1:
+-  |  lwzx TMP0, TMP1, CARG2
+-  |  stwx TMP0, TMP2, CARG2
+-  |   addic. CARG2, CARG2, -4
++  |  lpx TMP0, TMP1, CARG2
++  |  stpx TMP0, TMP2, CARG2
++  |   addic. CARG2, CARG2, -PSIZE
+   |  bge <1
+   |2:
+   |  bney cr1, >3
+@@ -2878,28 +3169,55 @@ static void build_subroutines(BuildCtx *ctx)
+   |  lfd f6, CCSTATE->fpr[5]
+   |  lfd f7, CCSTATE->fpr[6]
+   |  lfd f8, CCSTATE->fpr[7]
++  |  .if GPR64
++  |  lfd f9, CCSTATE->fpr[8]
++  |  lfd f10, CCSTATE->fpr[9]
++  |  lfd f11, CCSTATE->fpr[10]
++  |  lfd f12, CCSTATE->fpr[11]
++  |  lfd f13, CCSTATE->fpr[12]
++  |  .endif
+   |3:
+-  |   lp TMP0, CCSTATE->func
+-  |  lwz CARG2, CCSTATE->gpr[1]
+-  |  lwz CARG3, CCSTATE->gpr[2]
+-  |  lwz CARG4, CCSTATE->gpr[3]
+-  |  lwz CARG5, CCSTATE->gpr[4]
+-  |   mtctr TMP0
+-  |  lwz r8, CCSTATE->gpr[5]
+-  |  lwz r9, CCSTATE->gpr[6]
+-  |  lwz r10, CCSTATE->gpr[7]
+-  |  lwz CARG1, CCSTATE->gpr[0]		// Do this last, since CCSTATE is CARG1.
++  |  .toc std TOCREG, SAVE_TOC
++  |   lp FUNCREG, CCSTATE->func
++  |  lp CARG2, CCSTATE->gpr[1]
++  |  lp CARG3, CCSTATE->gpr[2]
++  |  .opd lp TOCREG, TOC_OFS(FUNCREG)
++  |  .opdenv lp ENVREG, ENV_OFS(FUNCREG)
++  |  .opd lp FUNCREG, 0(FUNCREG)
++  |  lp CARG4, CCSTATE->gpr[3]
++  |  lp CARG5, CCSTATE->gpr[4]
++  |   mtctr FUNCREG
++  |  lp r8, CCSTATE->gpr[5]
++  |  lp r9, CCSTATE->gpr[6]
++  |  lp r10, CCSTATE->gpr[7]
++  |  lp CARG1, CCSTATE->gpr[0]		// Do this last, since CCSTATE is CARG1.
+   |   bctrl
+-  |  lwz CCSTATE:TMP1, -8(r14)
+-  |  lwz TMP2, -4(r14)
++  |   .toc lp TOCREG, SAVE_TOC
++  |  .if GPR64
++  |   ld CCSTATE:TMP1, -16(r14)
++  |   ld TMP2, -8(r14)
++  |   ld TMP0, 16(r14)
++  |  .else
++  |   lwz CCSTATE:TMP1, -8(r14)
++  |   lwz TMP2, -4(r14)
+   |   lwz TMP0, 4(r14)
+-  |  stw CARG1, CCSTATE:TMP1->gpr[0]
++  |  .endif
++  |  stp CARG1, CCSTATE:TMP1->gpr[0]
+   |  stfd FARG1, CCSTATE:TMP1->fpr[0]
+-  |  stw CARG2, CCSTATE:TMP1->gpr[1]
++  |  stp CARG2, CCSTATE:TMP1->gpr[1]
++  |  .if GPR64
++  |   stfd FARG2, CCSTATE:TMP1->fpr[1]
++  |  .endif
++  |  .elfv2 stfd FARG3, CCSTATE:TMP1->fpr[2]
++  |  .elfv2 stfd FARG4, CCSTATE:TMP1->fpr[3]
++  |  .elfv2 stfd FARG5, CCSTATE:TMP1->fpr[4]
++  |  .elfv2 stfd FARG6, CCSTATE:TMP1->fpr[5]
++  |  .elfv2 stfd FARG7, CCSTATE:TMP1->fpr[6]
++  |  .elfv2 stfd FARG8, CCSTATE:TMP1->fpr[7]
+   |   mtlr TMP0
+-  |  stw CARG3, CCSTATE:TMP1->gpr[2]
++  |  stp CARG3, CCSTATE:TMP1->gpr[2]
+   |   mr sp, r14
+-  |  stw CARG4, CCSTATE:TMP1->gpr[3]
++  |  stp CARG4, CCSTATE:TMP1->gpr[3]
+   |   mr r14, TMP2
+   |  blr
+   |.endif
+@@ -2923,13 +3241,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_ISLT: case BC_ISGE: case BC_ISLE: case BC_ISGT:
+     |  // RA = src1*8, RD = src2*8, JMP with RD = target
+     |.if DUALNUM
+-    |  lwzux TMP0, RA, BASE
++    |  lwzx TMP0, BASE_HI, RA
+     |    addi PC, PC, 4
+-    |   lwz CARG2, 4(RA)
+-    |  lwzux TMP1, RD, BASE
++    |   lwzx CARG2, BASE_LO, RA
++    |  lwzx TMP1, BASE_HI, RD
+     |    lwz TMP2, -4(PC)
+     |  checknum cr0, TMP0
+-    |   lwz CARG3, 4(RD)
++    |   lwzx CARG3, BASE_LO, RD
+     |    decode_RD4 TMP2, TMP2
+     |  checknum cr1, TMP1
+     |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
+@@ -2953,7 +3271,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |7:  // RA is not an integer.
+     |  bgt cr0, ->vmeta_comp
+     |  // RA is a number.
+-    |   lfd f0, 0(RA)
++    |   lfdx f0, BASE, RA
+     |  bgt cr1, ->vmeta_comp
+     |  blt cr1, >4
+     |  // RA is a number, RD is an integer.
+@@ -2965,7 +3283,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // RA is an integer, RD is a number.
+     |  tonum_i f0, CARG2
+     |4:
+-    |  lfd f1, 0(RD)
++    |  lfdx f1, BASE, RD
+     |5:
+     |  fcmpu cr0, f0, f1
+     if (op == BC_ISLT) {
+@@ -2981,10 +3299,10 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     }
+     |  b <1
+     |.else
+-    |  lwzx TMP0, BASE, RA
++    |  lwzx TMP0, BASE_HI, RA
+     |    addi PC, PC, 4
+     |   lfdx f0, BASE, RA
+-    |  lwzx TMP1, BASE, RD
++    |  lwzx TMP1, BASE_HI, RD
+     |  checknum cr0, TMP0
+     |    lwz TMP2, -4(PC)
+     |   lfdx f1, BASE, RD
+@@ -3015,15 +3333,23 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     vk = op == BC_ISEQV;
+     |  // RA = src1*8, RD = src2*8, JMP with RD = target
+     |.if DUALNUM
+-    |  lwzux TMP0, RA, BASE
++    |  lwzx TMP0, BASE_HI, RA
+     |    addi PC, PC, 4
+-    |   lwz CARG2, 4(RA)
+-    |  lwzux TMP1, RD, BASE
++    |   lwzx CARG2, BASE_LO, RA
++    |  .if ENDIAN_LE
++    |    lwzx TMP1, BASE_HI, RD
++    |  .else
++    |    lwzux TMP1, RD, BASE_HI
++    |  .endif
+     |  checknum cr0, TMP0
+     |    lwz TMP2, -4(PC)
+     |  checknum cr1, TMP1
+     |    decode_RD4 TMP2, TMP2
+-    |   lwz CARG3, 4(RD)
++    |  .if ENDIAN_LE
++    |   lwzux CARG3, RD, BASE_LO
++    |  .else
++    |   lwz CARG3, WORD_LO(RD)
++    |  .endif
+     |  cror 4*cr7+gt, 4*cr0+gt, 4*cr1+gt
+     |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
+     if (vk) {
+@@ -3032,14 +3358,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+       |  ble cr7, ->BC_ISNEN_Z
+     }
+     |.else
+-    |  lwzux TMP0, RA, BASE
++    |  lwzx TMP0, BASE_HI, RA
+     |   lwz TMP2, 0(PC)
+-    |    lfd f0, 0(RA)
++    |    lfdx f0, BASE, RA
+     |   addi PC, PC, 4
+-    |  lwzux TMP1, RD, BASE
++    |  lwzx TMP1, BASE_HI, RD
+     |  checknum cr0, TMP0
+     |   decode_RD4 TMP2, TMP2
+-    |    lfd f1, 0(RD)
++    |    lfdx f1, BASE, RD
+     |  checknum cr1, TMP1
+     |   addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
+     |  bge cr0, >5
+@@ -3057,8 +3383,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |.endif
+     |5:  // Either or both types are not numbers.
+     |.if not DUALNUM
+-    |    lwz CARG2, 4(RA)
+-    |    lwz CARG3, 4(RD)
++    |    lwzx CARG2, BASE_LO, RA
++    |    lwzx CARG3, BASE_LO, RD
+     |.endif
+     |.if FFI
+     |  cmpwi cr7, TMP0, LJ_TCDATA
+@@ -3074,10 +3400,16 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |.if FFI
+     |  beq cr7, ->vmeta_equal_cd
+     |.endif
++    |.if P64
++    |   cmplwi cr7, TMP3, ~LJ_TUDATA		// Avoid 64 bit lightuserdata.
++    |.endif
+     |    cmplw cr5, CARG2, CARG3
+     |  crandc 4*cr0+gt, 4*cr0+eq, 4*cr1+gt	// 2: Same type and primitive.
+     |  crorc 4*cr0+lt, 4*cr5+eq, 4*cr0+eq	// 1: Same tv or different type.
+     |  crand 4*cr0+eq, 4*cr0+eq, 4*cr5+eq	// 0: Same type and same tv.
++    |.if P64
++    |   cror 4*cr6+lt, 4*cr6+lt, 4*cr7+gt
++    |.endif
+     |   mr SAVE0, PC
+     |  cror 4*cr0+eq, 4*cr0+eq, 4*cr0+gt	// 0 or 2.
+     |  cror 4*cr0+lt, 4*cr0+lt, 4*cr0+gt	// 1 or 2.
+@@ -3116,9 +3448,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_ISEQS: case BC_ISNES:
+     vk = op == BC_ISEQS;
+     |  // RA = src*8, RD = str_const*8 (~), JMP with RD = target
+-    |  lwzux TMP0, RA, BASE
++    |  lwzx TMP0, BASE_HI, RA
+     |   srwi RD, RD, 1
+-    |  lwz STR:TMP3, 4(RA)
++    |  lwzx STR:TMP3, BASE_LO, RA
+     |    lwz TMP2, 0(PC)
+     |   subfic RD, RD, -4
+     |    addi PC, PC, 4
+@@ -3150,15 +3482,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     vk = op == BC_ISEQN;
+     |  // RA = src*8, RD = num_const*8, JMP with RD = target
+     |.if DUALNUM
+-    |  lwzux TMP0, RA, BASE
++    |  lwzx TMP0, BASE_HI, RA
+     |    addi PC, PC, 4
+-    |   lwz CARG2, 4(RA)
+-    |  lwzux TMP1, RD, KBASE
++    |   lwzx CARG2, BASE_LO, RA
++    |  lwzux2 TMP1, CARG3, RD, KBASE
+     |  checknum cr0, TMP0
+     |    lwz TMP2, -4(PC)
+     |  checknum cr1, TMP1
+     |    decode_RD4 TMP2, TMP2
+-    |   lwz CARG3, 4(RD)
+     |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
+     if (vk) {
+       |->BC_ISEQN_Z:
+@@ -3175,7 +3506,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     } else {
+       |->BC_ISNEN_Z:  // Dummy label.
+     }
+-    |  lwzx TMP0, BASE, RA
++    |  lwzx TMP0, BASE_HI, RA
+     |    addi PC, PC, 4
+     |   lfdx f0, BASE, RA
+     |    lwz TMP2, -4(PC)
+@@ -3213,7 +3544,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |7:  // RA is not an integer.
+     |  bge cr0, <3
+     |  // RA is a number.
+-    |   lfd f0, 0(RA)
++    |   lfdx f0, BASE, RA
+     |  blt cr1, >1
+     |  // RA is a number, RD is an integer.
+     |  tonum_i f1, CARG3
+@@ -3232,7 +3563,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_ISEQP: case BC_ISNEP:
+     vk = op == BC_ISEQP;
+     |  // RA = src*8, RD = primitive_type*8 (~), JMP with RD = target
+-    |  lwzx TMP0, BASE, RA
++    |  lwzx TMP0, BASE_HI, RA
+     |   srwi TMP1, RD, 3
+     |    lwz TMP2, 0(PC)
+     |   not TMP1, TMP1
+@@ -3262,7 +3593,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+ 
+   case BC_ISTC: case BC_ISFC: case BC_IST: case BC_ISF:
+     |  // RA = dst*8 or unused, RD = src*8, JMP with RD = target
+-    |  lwzx TMP0, BASE, RD
++    |  lwzx TMP0, BASE_HI, RD
+     |   lwz INS, 0(PC)
+     |   addi PC, PC, 4
+     if (op == BC_IST || op == BC_ISF) {
+@@ -3297,7 +3628,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+ 
+   case BC_ISTYPE:
+     |  // RA = src*8, RD = -type*8
+-    |  lwzx TMP0, BASE, RA
++    |  lwzx TMP0, BASE_HI, RA
+     |  srwi TMP1, RD, 3
+     |  ins_next1
+     |.if not PPE and not GPR64
+@@ -3311,7 +3642,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_ISNUM:
+     |  // RA = src*8, RD = -(TISNUM-1)*8
+-    |  lwzx TMP0, BASE, RA
++    |  lwzx TMP0, BASE_HI, RA
+     |  ins_next1
+     |  checknum TMP0
+     |  bge ->vmeta_istype
+@@ -3330,17 +3661,20 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_NOT:
+     |  // RA = dst*8, RD = src*8
+     |  ins_next1
+-    |  lwzx TMP0, BASE, RD
++    |  lwzx TMP0, BASE_HI, RD
+     |  .gpr64 extsw TMP0, TMP0
+     |  subfic TMP1, TMP0, LJ_TTRUE
+     |  adde TMP0, TMP0, TMP1
+-    |  stwx TMP0, BASE, RA
++    |  stwx TMP0, BASE_HI, RA
+     |  ins_next2
+     break;
+   case BC_UNM:
+     |  // RA = dst*8, RD = src*8
+-    |  lwzux TMP1, RD, BASE
+-    |   lwz TMP0, 4(RD)
++    |  lwzx TMP1, BASE_HI, RD
++    |   lwzx TMP0, BASE_LO, RD
++    |.if DUALNUM and not GPR64
++    |  mtxer ZERO
++    |.endif
+     |  checknum TMP1
+     |.if DUALNUM
+     |  bne >5
+@@ -3352,18 +3686,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |.else
+     |  nego. TMP0, TMP0
+     |  bso >4
+-    |1:
+     |.endif
+     |  ins_next1
+-    |  stwux TISNUM, RA, BASE
+-    |   stw TMP0, 4(RA)
++    |  stwx TISNUM, BASE_HI, RA
++    |   stwx TMP0, BASE_LO, RA
+     |3:
+     |  ins_next2
+     |4:
+-    |.if not GPR64
+-    |  // Potential overflow.
+-    |  checkov TMP1, <1			// Ignore unrelated overflow.
+-    |.endif
+     |  lus TMP1, 0x41e0			// 2^31.
+     |  li TMP0, 0
+     |  b >7
+@@ -3373,8 +3702,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  xoris TMP1, TMP1, 0x8000
+     |7:
+     |  ins_next1
+-    |  stwux TMP1, RA, BASE
+-    |   stw TMP0, 4(RA)
++    |  stwx TMP1, BASE_HI, RA
++    |   stwx TMP0, BASE_LO, RA
+     |.if DUALNUM
+     |  b <3
+     |.else
+@@ -3383,15 +3712,15 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_LEN:
+     |  // RA = dst*8, RD = src*8
+-    |  lwzux TMP0, RD, BASE
+-    |   lwz CARG1, 4(RD)
++    |  lwzx TMP0, BASE_HI, RD
++    |   lwzx CARG1, BASE_LO, RD
+     |  checkstr TMP0; bne >2
+     |  lwz CRET1, STR:CARG1->len
+     |1:
+     |.if DUALNUM
+     |  ins_next1
+-    |  stwux TISNUM, RA, BASE
+-    |   stw CRET1, 4(RA)
++    |  stwx TISNUM, BASE_HI, RA
++    |   stwx CRET1, BASE_LO, RA
+     |.else
+     |  tonum_u f0, CRET1		// Result is a non-negative integer.
+     |  ins_next1
+@@ -3426,9 +3755,16 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN);
+     ||switch (vk) {
+     ||case 0:
+-    |   lwzx TMP1, BASE, RB
++    |   .if ENDIAN_LE and DUALNUM
++    |     addi TMP2, RC, 4
++    |   .endif
++    |   lwzx TMP1, BASE_HI, RB
+     |   .if DUALNUM
+-    |     lwzx TMP2, KBASE, RC
++    |     .if ENDIAN_LE
++    |       lwzx TMP2, KBASE, TMP2
++    |     .else
++    |       lwzx TMP2, KBASE, RC
++    |     .endif
+     |   .endif
+     |    lfdx f14, BASE, RB
+     |    lfdx f15, KBASE, RC
+@@ -3442,9 +3778,16 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   .endif
+     ||  break;
+     ||case 1:
+-    |   lwzx TMP1, BASE, RB
++    |   .if ENDIAN_LE and DUALNUM
++    |     addi TMP2, RC, 4
++    |   .endif
++    |   lwzx TMP1, BASE_HI, RB
+     |   .if DUALNUM
+-    |     lwzx TMP2, KBASE, RC
++    |     .if ENDIAN_LE
++    |       lwzx TMP2, KBASE, TMP2
++    |     .else
++    |       lwzx TMP2, KBASE, RC
++    |     .endif
+     |   .endif
+     |    lfdx f15, BASE, RB
+     |    lfdx f14, KBASE, RC
+@@ -3458,8 +3801,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   .endif
+     ||  break;
+     ||default:
+-    |   lwzx TMP1, BASE, RB
+-    |   lwzx TMP2, BASE, RC
++    |   lwzx TMP1, BASE_HI, RB
++    |   lwzx TMP2, BASE_HI, RC
+     |    lfdx f14, BASE, RB
+     |    lfdx f15, BASE, RC
+     |   checknum cr0, TMP1
+@@ -3514,41 +3857,62 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN);
+     ||switch (vk) {
+     ||case 0:
+-    |   lwzux TMP1, RB, BASE
+-    |   lwzux TMP2, RC, KBASE
+-    |    lwz CARG1, 4(RB)
+-    |   checknum cr0, TMP1
+-    |    lwz CARG2, 4(RC)
++    |   .if ENDIAN_LE
++    |     lwzux CARG2, RC, KBASE
++    |     lwzx TMP1, RB, BASE_HI
++    |      lwz TMP2, 4(RC)
++    |     checknum cr0, TMP1
++    |      lwzux CARG1, RB, BASE
++    |   .else
++    |     lwzux TMP1, RB, BASE
++    |     lwzux TMP2, RC, KBASE
++    |      lwz CARG1, 4(RB)
++    |     checknum cr0, TMP1
++    |      lwz CARG2, 4(RC)
++    |   .endif
+     ||  break;
+     ||case 1:
+-    |   lwzux TMP1, RB, BASE
+-    |   lwzux TMP2, RC, KBASE
+-    |    lwz CARG2, 4(RB)
+-    |   checknum cr0, TMP1
+-    |    lwz CARG1, 4(RC)
++    |   .if ENDIAN_LE
++    |     lwzux CARG1, RC, KBASE
++    |     lwzx TMP1, RB, BASE_HI
++    |      lwz TMP2, 4(RC)
++    |     checknum cr0, TMP1
++    |      lwzux CARG2, RB, BASE
++    |   .else
++    |     lwzux TMP1, RB, BASE
++    |     lwzux TMP2, RC, KBASE
++    |      lwz CARG2, 4(RB)
++    |     checknum cr0, TMP1
++    |      lwz CARG1, 4(RC)
++    |   .endif
+     ||  break;
+     ||default:
+-    |   lwzux TMP1, RB, BASE
+-    |   lwzux TMP2, RC, BASE
+-    |    lwz CARG1, 4(RB)
+-    |   checknum cr0, TMP1
+-    |    lwz CARG2, 4(RC)
++    |   .if ENDIAN_LE
++    |     lwzx TMP1, RB, BASE_HI
++    |     lwzx TMP2, RC, BASE_HI
++    |      lwzux CARG1, RB, BASE
++    |     checknum cr0, TMP1
++    |      lwzux CARG2, RC, BASE
++    |   .else
++    |     lwzux TMP1, RB, BASE
++    |     lwzux TMP2, RC, BASE
++    |      lwz CARG1, 4(RB)
++    |     checknum cr0, TMP1
++    |      lwz CARG2, 4(RC)
++    |   .endif
+     ||  break;
+     ||}
++    |  mtxer ZERO
+     |  checknum cr1, TMP2
+     |  bne >5
+     |  bne cr1, >5
+     |  intins CARG1, CARG1, CARG2
+-    |  bso >4
+-    |1:
++    |  ins_arithfallback bso
+     |  ins_next1
+-    |  stwux TISNUM, RA, BASE
+-    |  stw CARG1, 4(RA)
++    |  stwx TISNUM, BASE_HI, RA
++    |  stwx CARG1, BASE_LO, RA
+     |2:
+     |  ins_next2
+-    |4:  // Overflow.
+-    |  checkov TMP0, <1			// Ignore unrelated overflow.
+-    |  ins_arithfallback b
+     |5:  // FP variant.
+     ||if (vk == 1) {
+     |  lfd f15, 0(RB)
+@@ -3620,9 +3984,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_POW:
+     |  // NYI: (partial) integer arithmetic.
+-    |  lwzx TMP1, BASE, RB
++    |  lwzx TMP1, BASE_HI, RB
+     |   lfdx FARG1, BASE, RB
+-    |  lwzx TMP2, BASE, RC
++    |  lwzx TMP2, BASE_HI, RC
+     |   lfdx FARG2, BASE, RC
+     |  checknum cr0, TMP1
+     |  checknum cr1, TMP2
+@@ -3648,6 +4012,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // Returns NULL (finished) or TValue * (metamethod).
+     |  cmplwi CRET1, 0
+     |   lp BASE, L->base
++    |   addi BASEP4, BASE, 4
+     |  bne ->vmeta_binop
+     |  ins_next1
+     |  lfdx f0, BASE, SAVE0		// Copy result from RB to RA.
+@@ -3664,8 +4029,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  ins_next1
+     |  lwzx TMP0, KBASE, TMP1		// KBASE-4-str_const*4
+     |  li TMP2, LJ_TSTR
+-    |  stwux TMP2, RA, BASE
+-    |  stw TMP0, 4(RA)
++    |  stwx TMP2, BASE_HI, RA
++    |  stwx TMP0, BASE_LO, RA
+     |  ins_next2
+     break;
+   case BC_KCDATA:
+@@ -3676,8 +4041,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  ins_next1
+     |  lwzx TMP0, KBASE, TMP1		// KBASE-4-cdata_const*4
+     |  li TMP2, LJ_TCDATA
+-    |  stwux TMP2, RA, BASE
+-    |  stw TMP0, 4(RA)
++    |  stwx TMP2, BASE_HI, RA
++    |  stwx TMP0, BASE_LO, RA
+     |  ins_next2
+     |.endif
+     break;
+@@ -3687,14 +4052,15 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  slwi RD, RD, 13
+     |  srawi RD, RD, 16
+     |  ins_next1
+-    |   stwux TISNUM, RA, BASE
+-    |   stw RD, 4(RA)
++    |   stwx TISNUM, BASE_HI, RA
++    |   stwx RD, BASE_LO, RA
+     |  ins_next2
+     |.else
+     |  // The soft-float approach is faster.
+     |  slwi RD, RD, 13
+     |  srawi TMP1, RD, 31
+     |  xor TMP2, TMP1, RD
++    |  .gpr64 extsw RD, RD
+     |  sub TMP2, TMP2, TMP1		// TMP2 = abs(x)
+     |  cntlzw TMP3, TMP2
+     |  subfic TMP1, TMP3, 0x40d		// TMP1 = exponent-1
+@@ -3706,8 +4072,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   add RD, RD, TMP1		// hi = hi + exponent-1
+     |    and RD, RD, TMP0		// hi = x == 0 ? 0 : hi
+     |  ins_next1
+-    |    stwux RD, RA, BASE
+-    |    stw ZERO, 4(RA)
++    |    stwx RD, BASE_HI, RA
++    |    stwx ZERO, BASE_LO, RA
+     |  ins_next2
+     |.endif
+     break;
+@@ -3723,15 +4089,15 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  srwi TMP1, RD, 3
+     |  not TMP0, TMP1
+     |  ins_next1
+-    |  stwx TMP0, BASE, RA
++    |  stwx TMP0, BASE_HI, RA
+     |  ins_next2
+     break;
+   case BC_KNIL:
+     |  // RA = base*8, RD = end*8
+-    |  stwx TISNIL, BASE, RA
++    |  stwx TISNIL, BASE_HI, RA
+     |   addi RA, RA, 8
+     |1:
+-    |  stwx TISNIL, BASE, RA
++    |  stwx TISNIL, BASE_HI, RA
+     |  cmpw RA, RD
+     |   addi RA, RA, 8
+     |  blt <1
+@@ -3763,10 +4129,10 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   lwz CARG2, UPVAL:RB->v
+     |  andix. TMP3, TMP3, LJ_GC_BLACK	// isblack(uv)
+     |    lbz TMP0, UPVAL:RB->closed
+-    |   lwz TMP2, 0(RD)
++    |   lwz TMP2, WORD_HI(RD)
+     |   stfd f0, 0(CARG2)
+     |    cmplwi cr1, TMP0, 0
+-    |   lwz TMP1, 4(RD)
++    |   lwz TMP1, WORD_LO(RD)
+     |  cror 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
+     |   subi TMP2, TMP2, (LJ_TNUMX+1)
+     |  bne >2				// Upvalue is closed and black?
+@@ -3799,8 +4165,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   lbz TMP3, STR:TMP1->marked
+     |   lbz TMP2, UPVAL:RB->closed
+     |   li TMP0, LJ_TSTR
+-    |   stw STR:TMP1, 4(CARG2)
+-    |   stw TMP0, 0(CARG2)
++    |   stw STR:TMP1, WORD_LO(CARG2)
++    |   stw TMP0, WORD_HI(CARG2)
+     |  bne >2
+     |1:
+     |  ins_next
+@@ -3837,7 +4203,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  lwzx UPVAL:RB, LFUNC:RB, RA
+     |  ins_next1
+     |  lwz TMP1, UPVAL:RB->v
+-    |  stw TMP0, 0(TMP1)
++    |  stw TMP0, WORD_HI(TMP1)
+     |  ins_next2
+     break;
+ 
+@@ -3852,6 +4218,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   add CARG2, BASE, RA
+     |  bl extern lj_func_closeuv	// (lua_State *L, TValue *level)
+     |  lp BASE, L->base
++    |  addi BASEP4, BASE, 4
+     |1:
+     |  ins_next
+     break;
+@@ -3870,8 +4237,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // Returns GCfuncL *.
+     |  lp BASE, L->base
+     |   li TMP0, LJ_TFUNC
+-    |  stwux TMP0, RA, BASE
+-    |  stw LFUNC:CRET1, 4(RA)
++    |  addi BASEP4, BASE, 4
++    |  stwx TMP0, BASE_HI, RA
++    |  stwx LFUNC:CRET1, BASE_LO, RA
+     |  ins_next
+     break;
+ 
+@@ -3904,8 +4272,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     }
+     |  lp BASE, L->base
+     |   li TMP0, LJ_TTAB
+-    |  stwux TMP0, RA, BASE
+-    |  stw TAB:CRET1, 4(RA)
++    |  addi BASEP4, BASE, 4
++    |  stwx TMP0, BASE_HI, RA
++    |  stwx TAB:CRET1, BASE_LO, RA
+     |  ins_next
+     if (op == BC_TNEW) {
+       |3:
+@@ -3938,13 +4307,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+ 
+   case BC_TGETV:
+     |  // RA = dst*8, RB = table*8, RC = key*8
+-    |  lwzux CARG1, RB, BASE
+-    |  lwzux CARG2, RC, BASE
+-    |   lwz TAB:RB, 4(RB)
++    |  lwzx CARG1, BASE_HI, RB
++    |  lwzx CARG2, BASE_HI, RC
++    |   lwzx TAB:RB, BASE_LO, RB
+     |.if DUALNUM
+-    |   lwz RC, 4(RC)
++    |   lwzx RC, BASE_LO, RC
+     |.else
+-    |   lfd f0, 0(RC)
++    |   lfdx f0, BASE, RC
+     |.endif
+     |  checktab CARG1
+     |   checknum cr1, CARG2
+@@ -3971,8 +4340,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   slwi TMP2, TMP2, 3
+     |.endif
+     |  ble ->vmeta_tgetv		// Integer key and in array part?
+-    |  lwzx TMP0, TMP1, TMP2
+-    |   lfdx f14, TMP1, TMP2
++    |  .if ENDIAN_LE
++    |    lfdux f14, TMP1, TMP2
++    |    lwz TMP0, WORD_HI(TMP1)
++    |  .else
++    |    lwzx TMP0, TMP1, TMP2
++    |    lfdx f14, TMP1, TMP2
++    |  .endif
+     |  checknil TMP0; beq >2
+     |1:
+     |  ins_next1
+@@ -3991,15 +4365,15 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |5:
+     |  checkstr CARG2; bne ->vmeta_tgetv
+     |.if not DUALNUM
+-    |  lwz STR:RC, 4(RC)
++    |  lwzx STR:RC, BASE_LO, RC
+     |.endif
+     |  b ->BC_TGETS_Z			// String key?
+     break;
+   case BC_TGETS:
+     |  // RA = dst*8, RB = table*8, RC = str_const*8 (~)
+-    |  lwzux CARG1, RB, BASE
++    |  lwzx CARG1, BASE_HI, RB
+     |   srwi TMP1, RC, 1
+-    |    lwz TAB:RB, 4(RB)
++    |    lwzx TAB:RB, BASE_LO, RB
+     |   subfic TMP1, TMP1, -4
+     |  checktab CARG1
+     |   lwzx STR:RC, KBASE, TMP1	// KBASE-4-str_const*4
+@@ -4015,16 +4389,16 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  sub TMP1, TMP0, TMP1
+     |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
+     |1:
+-    |  lwz CARG1, NODE:TMP2->key
+-    |   lwz TMP0, 4+offsetof(Node, key)(NODE:TMP2)
+-    |    lwz CARG2, NODE:TMP2->val
+-    |     lwz TMP1, 4+offsetof(Node, val)(NODE:TMP2)
++    |  lwz CARG1, WORD_HI+offsetof(Node, key)(NODE:TMP2)
++    |   lwz TMP0, WORD_LO+offsetof(Node, key)(NODE:TMP2)
++    |    lwz CARG2, WORD_HI+offsetof(Node, val)(NODE:TMP2)
++    |     lwz TMP1, WORD_LO+offsetof(Node, val)(NODE:TMP2)
+     |  checkstr CARG1; bne >4
+     |   cmpw TMP0, STR:RC; bne >4
+     |    checknil CARG2; beq >5		// Key found, but nil value?
+     |3:
+-    |    stwux CARG2, RA, BASE
+-    |     stw TMP1, 4(RA)
++    |    stwx CARG2, BASE_HI, RA
++    |     stwx TMP1, BASE_LO, RA
+     |  ins_next
+     |
+     |4:  // Follow hash chain.
+@@ -4045,15 +4419,20 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_TGETB:
+     |  // RA = dst*8, RB = table*8, RC = index*8
+-    |  lwzux CARG1, RB, BASE
++    |  lwzx CARG1, BASE_HI, RB
+     |   srwi TMP0, RC, 3
+-    |   lwz TAB:RB, 4(RB)
++    |   lwzx TAB:RB, BASE_LO, RB
+     |  checktab CARG1; bne ->vmeta_tgetb
+     |  lwz TMP1, TAB:RB->asize
+     |   lwz TMP2, TAB:RB->array
+     |  cmplw TMP0, TMP1; bge ->vmeta_tgetb
+-    |  lwzx TMP1, TMP2, RC
+-    |   lfdx f0, TMP2, RC
++    |  .if ENDIAN_LE
++    |    lfdux f0, TMP2, RC
++    |    lwz TMP1, WORD_HI(TMP2)
++    |  .else
++    |    lwzx TMP1, TMP2, RC
++    |    lfdx f0, TMP2, RC
++    |  .endif
+     |  checknil TMP1; beq >5
+     |1:
+     |  ins_next1
+@@ -4071,12 +4450,10 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_TGETR:
+     |  // RA = dst*8, RB = table*8, RC = key*8
+-    |  add RB, BASE, RB
+-    |  lwz TAB:CARG1, 4(RB)
++    |  lwzx TAB:CARG1, BASE_LO, RB
+     |.if DUALNUM
+-    |  add RC, BASE, RC
+     |  lwz TMP0, TAB:CARG1->asize
+-    |  lwz CARG2, 4(RC)
++    |  lwzx CARG2, BASE_LO, RC
+     |   lwz TMP1, TAB:CARG1->array
+     |.else
+     |  lfdx f0, BASE, RC
+@@ -4096,13 +4473,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+ 
+   case BC_TSETV:
+     |  // RA = src*8, RB = table*8, RC = key*8
+-    |  lwzux CARG1, RB, BASE
+-    |  lwzux CARG2, RC, BASE
+-    |   lwz TAB:RB, 4(RB)
++    |  lwzx CARG1, BASE_HI, RB
++    |  lwzx CARG2, BASE_HI, RC
++    |   lwzx TAB:RB, BASE_LO, RB
+     |.if DUALNUM
+-    |   lwz RC, 4(RC)
++    |   lwzx RC, BASE_LO, RC
+     |.else
+-    |   lfd f0, 0(RC)
++    |   lfdx f0, BASE, RC
+     |.endif
+     |  checktab CARG1
+     |   checknum cr1, CARG2
+@@ -4129,7 +4506,12 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   slwi TMP0, TMP2, 3
+     |.endif
+     |  ble ->vmeta_tsetv		// Integer key and in array part?
++    |  .if ENDIAN_LE
++    |   addi TMP2, TMP1, 4
++    |   lwzx TMP2, TMP2, TMP0
++    |  .else
+     |   lwzx TMP2, TMP1, TMP0
++    |  .endif
+     |  lbz TMP3, TAB:RB->marked
+     |    lfdx f14, BASE, RA
+     |   checknil TMP2; beq >3
+@@ -4152,7 +4534,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |5:
+     |  checkstr CARG2; bne ->vmeta_tsetv
+     |.if not DUALNUM
+-    |  lwz STR:RC, 4(RC)
++    |  lwzx STR:RC, BASE_LO, RC
+     |.endif
+     |  b ->BC_TSETS_Z			// String key?
+     |
+@@ -4162,9 +4544,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_TSETS:
+     |  // RA = src*8, RB = table*8, RC = str_const*8 (~)
+-    |  lwzux CARG1, RB, BASE
++    |  lwzx CARG1, BASE_HI, RB
+     |   srwi TMP1, RC, 1
+-    |    lwz TAB:RB, 4(RB)
++    |    lwzx TAB:RB, BASE_LO, RB
+     |   subfic TMP1, TMP1, -4
+     |  checktab CARG1
+     |   lwzx STR:RC, KBASE, TMP1	// KBASE-4-str_const*4
+@@ -4183,9 +4565,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |    lbz TMP3, TAB:RB->marked
+     |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
+     |1:
+-    |  lwz CARG1, NODE:TMP2->key
+-    |   lwz TMP0, 4+offsetof(Node, key)(NODE:TMP2)
+-    |    lwz CARG2, NODE:TMP2->val
++    |  lwz CARG1, WORD_HI+offsetof(Node, key)(NODE:TMP2)
++    |   lwz TMP0, WORD_LO+offsetof(Node, key)(NODE:TMP2)
++    |    lwz CARG2, WORD_HI+offsetof(Node, val)(NODE:TMP2)
+     |     lwz NODE:TMP1, NODE:TMP2->next
+     |  checkstr CARG1; bne >5
+     |   cmpw TMP0, STR:RC; bne >5
+@@ -4225,13 +4607,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  beq ->vmeta_tsets		// 'no __newindex' flag NOT set: check.
+     |6:
+     |  li TMP0, LJ_TSTR
+-    |   stw STR:RC, 4(CARG3)
++    |   stw STR:RC, WORD_LO(CARG3)
+     |   mr CARG2, TAB:RB
+-    |  stw TMP0, 0(CARG3)
++    |  stw TMP0, WORD_HI(CARG3)
+     |  bl extern lj_tab_newkey		// (lua_State *L, GCtab *t, TValue *k)
+     |  // Returns TValue *.
+     |  lp BASE, L->base
+     |  stfd f14, 0(CRET1)
++    |   addi BASEP4, BASE, 4
+     |  b <3				// No 2nd write barrier needed.
+     |
+     |7:  // Possible table write barrier for the value. Skip valiswhite check.
+@@ -4240,9 +4623,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_TSETB:
+     |  // RA = src*8, RB = table*8, RC = index*8
+-    |  lwzux CARG1, RB, BASE
++    |  lwzx CARG1, BASE_HI, RB
+     |   srwi TMP0, RC, 3
+-    |   lwz TAB:RB, 4(RB)
++    |   lwzx TAB:RB, BASE_LO, RB
+     |  checktab CARG1; bne ->vmeta_tsetb
+     |  lwz TMP1, TAB:RB->asize
+     |   lwz TMP2, TAB:RB->array
+@@ -4250,7 +4633,12 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  cmplw TMP0, TMP1
+     |   lfdx f14, BASE, RA
+     |  bge ->vmeta_tsetb
+-    |  lwzx TMP1, TMP2, RC
++    |  .if ENDIAN_LE
++    |    addi TMP1, TMP2, 4
++    |    lwzx TMP1, TMP1, RC
++    |  .else
++    |    lwzx TMP1, TMP2, RC
++    |  .endif
+     |  checknil TMP1; beq >5
+     |1:
+     |  andix. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
+@@ -4274,13 +4662,11 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_TSETR:
+     |  // RA = dst*8, RB = table*8, RC = key*8
+-    |  add RB, BASE, RB
+-    |  lwz TAB:CARG2, 4(RB)
++    |  lwzx TAB:CARG2, BASE_LO, RB
+     |.if DUALNUM
+-    |  add RC, BASE, RC
+     |    lbz TMP3, TAB:CARG2->marked
+     |  lwz TMP0, TAB:CARG2->asize
+-    |  lwz CARG3, 4(RC)
++    |  lwzx CARG3, BASE_LO, RC
+     |   lwz TMP1, TAB:CARG2->array
+     |.else
+     |  lfdx f0, BASE, RC
+@@ -4311,9 +4697,9 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  add RA, BASE, RA
+     |1:
+     |   add TMP3, KBASE, RD
+-    |  lwz TAB:CARG2, -4(RA)		// Guaranteed to be a table.
++    |  lwz TAB:CARG2, WORD_LO-8(RA)	// Guaranteed to be a table.
+     |    addic. TMP0, MULTRES, -8
+-    |   lwz TMP3, 4(TMP3)		// Integer constant is in lo-word.
++    |   lwz TMP3, WORD_LO(TMP3)		// Integer constant is in lo-word.
+     |    srwi CARG3, TMP0, 3
+     |    beq >4				// Nothing to copy?
+     |  add CARG3, CARG3, TMP3
+@@ -4362,8 +4748,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_CALL:
+     |  // RA = base*8, (RB = (nresults+1)*8,) RC = (nargs+1)*8
+     |  mr TMP2, BASE
+-    |  lwzux TMP0, BASE, RA
+-    |   lwz LFUNC:RB, 4(BASE)
++    |  lwzux2 TMP0, LFUNC:RB, BASE, RA
+     |    subi NARGS8:RC, NARGS8:RC, 8
+     |   addi BASE, BASE, 8
+     |  checkfunc TMP0; bne ->vmeta_call
+@@ -4377,8 +4762,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     break;
+   case BC_CALLT:
+     |  // RA = base*8, (RB = 0,) RC = (nargs+1)*8
+-    |  lwzux TMP0, RA, BASE
+-    |   lwz LFUNC:RB, 4(RA)
++    |  lwzux2 TMP0, LFUNC:RB, RA, BASE
+     |    subi NARGS8:RC, NARGS8:RC, 8
+     |    lwz TMP1, FRAME_PC(BASE)
+     |  checkfunc TMP0
+@@ -4430,12 +4814,12 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // RA = base*8, (RB = (nresults+1)*8, RC = (nargs+1)*8 ((2+1)*8))
+     |  mr TMP2, BASE
+     |  add BASE, BASE, RA
+-    |  lwz TMP1, -24(BASE)
+-    |   lwz LFUNC:RB, -20(BASE)
++    |  lwz TMP1, WORD_HI-24(BASE)
++    |   lwz LFUNC:RB, WORD_LO-24(BASE)
+     |    lfd f1, -8(BASE)
+     |    lfd f0, -16(BASE)
+-    |  stw TMP1, 0(BASE)		// Copy callable.
+-    |   stw LFUNC:RB, 4(BASE)
++    |  stw TMP1, WORD_HI(BASE)		// Copy callable.
++    |   stw LFUNC:RB, WORD_LO(BASE)
+     |  checkfunc TMP1
+     |    stfd f1, 16(BASE)		// Copy control var.
+     |     li NARGS8:RC, 16		// Iterators get 2 arguments.
+@@ -4450,8 +4834,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // NYI: add hotloop, record BC_ITERN.
+     |.endif
+     |  add RA, BASE, RA
+-    |  lwz TAB:RB, -12(RA)
+-    |  lwz RC, -4(RA)			// Get index from control var.
++    |  lwz TAB:RB, WORD_LO-16(RA)
++    |  lwz RC, WORD_LO-8(RA)		// Get index from control var.
+     |  lwz TMP0, TAB:RB->asize
+     |  lwz TMP1, TAB:RB->array
+     |   addi PC, PC, 4
+@@ -4459,14 +4843,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  cmplw RC, TMP0
+     |   slwi TMP3, RC, 3
+     |  bge >5				// Index points after array part?
+-    |  lwzx TMP2, TMP1, TMP3
+-    |   lfdx f0, TMP1, TMP3
++    |  lfdux f0, TMP3, TMP1
++    |   lwz TMP2, WORD_HI(TMP3)
+     |  checknil TMP2
+     |     lwz INS, -4(PC)
+     |  beq >4
+     |.if DUALNUM
+-    |   stw RC, 4(RA)
+-    |   stw TISNUM, 0(RA)
++    |   stw RC, WORD_LO(RA)
++    |   stw TISNUM, WORD_HI(RA)
+     |.else
+     |   tonum_u f1, RC
+     |.endif
+@@ -4474,7 +4858,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |     addis TMP3, PC, -(BCBIAS_J*4 >> 16)
+     |  stfd f0, 8(RA)
+     |     decode_RD4 TMP1, INS
+-    |    stw RC, -4(RA)			// Update control var.
++    |    stw RC, WORD_LO-8(RA)		// Update control var.
+     |     add PC, TMP1, TMP3
+     |.if not DUALNUM
+     |   stfd f1, 0(RA)
+@@ -4496,9 +4880,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  bgty <3
+     |   slwi RB, RC, 3
+     |   sub TMP3, TMP3, RB
+-    |  lwzx RB, TMP2, TMP3
+-    |  lfdx f0, TMP2, TMP3
+-    |   add NODE:TMP3, TMP2, TMP3
++    |  lfdux f0, TMP3, TMP2
++    |  lwz RB, WORD_HI(TMP3)
+     |  checknil RB
+     |     lwz INS, -4(PC)
+     |  beq >7
+@@ -4510,7 +4893,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |   stfd f1, 0(RA)
+     |    addi RC, RC, 1
+     |     add PC, TMP1, TMP2
+-    |    stw RC, -4(RA)			// Update control var.
++    |    stw RC, WORD_LO-8(RA)		// Update control var.
+     |  b <3
+     |
+     |7:  // Skip holes in hash part.
+@@ -4521,10 +4904,10 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_ISNEXT:
+     |  // RA = base*8, RD = target (points to ITERN)
+     |  add RA, BASE, RA
+-    |  lwz TMP0, -24(RA)
+-    |  lwz CFUNC:TMP1, -20(RA)
+-    |   lwz TMP2, -16(RA)
+-    |    lwz TMP3, -8(RA)
++    |  lwz TMP0, WORD_HI-24(RA)
++    |  lwz CFUNC:TMP1, WORD_LO-24(RA)
++    |   lwz TMP2, WORD_HI-16(RA)
++    |    lwz TMP3, WORD_HI-8(RA)
+     |   cmpwi cr0, TMP2, LJ_TTAB
+     |  cmpwi cr1, TMP0, LJ_TFUNC
+     |    cmpwi cr6, TMP3, LJ_TNIL
+@@ -4538,17 +4921,25 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  bne cr0, >5
+     |  lus TMP1, 0xfffe
+     |  ori TMP1, TMP1, 0x7fff
+-    |  stw ZERO, -4(RA)			// Initialize control var.
+-    |  stw TMP1, -8(RA)
++    |  stw ZERO, WORD_LO-8(RA)		// Initialize control var.
++    |  stw TMP1, WORD_HI-8(RA)
+     |    addis PC, TMP3, -(BCBIAS_J*4 >> 16)
+     |1:
+     |  ins_next
+     |5:  // Despecialize bytecode if any of the checks fail.
+     |  li TMP0, BC_JMP
+     |   li TMP1, BC_ITERC
++    |  .if ENDIAN_LE
++    |  stb TMP0, -4(PC)
++    |  .else
+     |  stb TMP0, -1(PC)
++    |  .endif
+     |    addis PC, TMP3, -(BCBIAS_J*4 >> 16)
++    |  .if ENDIAN_LE
++    |   stb TMP1, 0(PC)
++    |  .else
+     |   stb TMP1, 3(PC)
++    |  .endif
+     |  b <1
+     break;
+ 
+@@ -4582,7 +4973,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |    addi RA, RA, 8
+     |   blt cr1, <1			// More vararg slots?
+     |2:  // Fill up remainder with nil.
+-    |  stw TISNIL, 0(RA)
++    |  stw TISNIL, WORD_HI(RA)
+     |  cmplw RA, TMP2
+     |   addi RA, RA, 8
+     |  blt <2
+@@ -4619,6 +5010,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  add RA, BASE, RA
+     |  add RC, BASE, SAVE0
+     |  subi TMP3, BASE, 8
++    |  addi BASEP4, BASE, 4
+     |  b <6
+     break;
+ 
+@@ -4667,13 +5059,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  bgt >6
+     |   sub BASE, TMP2, RA
+     |  lwz LFUNC:TMP1, FRAME_FUNC(BASE)
++    |  addi BASEP4, BASE, 4
+     |  ins_next1
+     |  lwz TMP1, LFUNC:TMP1->pc
+     |  lwz KBASE, PC2PROTO(k)(TMP1)
+     |  ins_next2
+     |
+     |6:  // Fill up results with nil.
+-    |  subi TMP1, RD, 8
++    |  addi TMP1, RD, WORD_HI-8
+     |   addi RD, RD, 8
+     |  stwx TISNIL, TMP2, TMP1
+     |  b <5
+@@ -4709,13 +5102,14 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  bgt >6
+     |   sub BASE, TMP2, RA
+     |  lwz LFUNC:TMP1, FRAME_FUNC(BASE)
++    |  addi BASEP4, BASE, 4
+     |  ins_next1
+     |  lwz TMP1, LFUNC:TMP1->pc
+     |  lwz KBASE, PC2PROTO(k)(TMP1)
+     |  ins_next2
+     |
+     |6:  // Fill up results with nil.
+-    |  subi TMP1, RD, 8
++    |  addi TMP1, RD, WORD_HI-8
+     |   addi RD, RD, 8
+     |  stwx TISNIL, TMP2, TMP1
+     |  b <5
+@@ -4741,11 +5135,13 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     vk = (op == BC_IFORL || op == BC_JFORL);
+     |.if DUALNUM
+     |  // Integer loop.
+-    |  lwzux TMP1, RA, BASE
+-    |   lwz CARG1, FORL_IDX*8+4(RA)
++    |  lwzux2 TMP1, CARG1, RA, BASE
++    if (vk) {
++      |  mtxer ZERO
++    }
+     |  cmplw cr0, TMP1, TISNUM
+     if (vk) {
+-      |   lwz CARG3, FORL_STEP*8+4(RA)
++      |   lwz CARG3, FORL_STEP*8+WORD_LO(RA)
+       |  bne >9
+       |.if GPR64
+       |  // Need to check overflow for (a<<32) + (b<<32).
+@@ -4757,15 +5153,15 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+       |  addo. CARG1, CARG1, CARG3
+       |.endif
+       |    cmpwi cr6, CARG3, 0
+-      |   lwz CARG2, FORL_STOP*8+4(RA)
+-      |  bso >6
++      |   lwz CARG2, FORL_STOP*8+WORD_LO(RA)
++      |  bso >2
+       |4:
+-      |  stw CARG1, FORL_IDX*8+4(RA)
++      |  stw CARG1, FORL_IDX*8+WORD_LO(RA)
+     } else {
+-      |  lwz TMP3, FORL_STEP*8(RA)
+-      |   lwz CARG3, FORL_STEP*8+4(RA)
+-      |  lwz TMP2, FORL_STOP*8(RA)
+-      |   lwz CARG2, FORL_STOP*8+4(RA)
++      |  lwz TMP3, FORL_STEP*8+WORD_HI(RA)
++      |   lwz CARG3, FORL_STEP*8+WORD_LO(RA)
++      |  lwz TMP2, FORL_STOP*8+WORD_HI(RA)
++      |   lwz CARG2, FORL_STOP*8+WORD_LO(RA)
+       |  cmplw cr7, TMP3, TISNUM
+       |  cmplw cr1, TMP2, TISNUM
+       |  crand 4*cr0+eq, 4*cr0+eq, 4*cr7+eq
+@@ -4776,11 +5172,11 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |    blt cr6, >5
+     |  cmpw CARG1, CARG2
+     |1:
+-    |   stw TISNUM, FORL_EXT*8(RA)
++    |   stw TISNUM, FORL_EXT*8+WORD_HI(RA)
+     if (op != BC_JFORL) {
+       |  srwi RD, RD, 1
+     }
+-    |   stw CARG1, FORL_EXT*8+4(RA)
++    |   stw CARG1, FORL_EXT*8+WORD_LO(RA)
+     if (op != BC_JFORL) {
+       |  add RD, PC, RD
+     }
+@@ -4800,11 +5196,6 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |5:  // Invert check for negative step.
+     |  cmpw CARG2, CARG1
+     |  b <1
+-    if (vk) {
+-      |6:  // Potential overflow.
+-      |  checkov TMP0, <4		// Ignore unrelated overflow.
+-      |  b <2
+-    }
+     |.endif
+     if (vk) {
+       |.if DUALNUM
+@@ -4815,14 +5206,19 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+       |.endif
+       |  lfd f3, FORL_STEP*8(RA)
+       |  lfd f2, FORL_STOP*8(RA)
+-      |   lwz TMP3, FORL_STEP*8(RA)
++      |   lwz TMP3, FORL_STEP*8+WORD_HI(RA)
+       |  fadd f1, f1, f3
+       |  stfd f1, FORL_IDX*8(RA)
+     } else {
+       |.if DUALNUM
+       |9:  // FP loop.
+       |.else
++      |.if ENDIAN_LE
++      |  lwzx TMP1, RA, BASE_LO
++      |  add RA, RA, BASE
++      |.else
+       |  lwzux TMP1, RA, BASE
++      |.endif
+       |  lwz TMP3, FORL_STEP*8(RA)
+       |  lwz TMP2, FORL_STOP*8(RA)
+       |  cmplw cr0, TMP1, TISNUM
+@@ -4903,17 +5299,16 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+ #endif
+   case BC_IITERL:
+     |  // RA = base*8, RD = target
+-    |  lwzux TMP1, RA, BASE
+-    |   lwz TMP2, 4(RA)
++    |  lwzux2 TMP1, TMP2, RA, BASE
+     |  checknil TMP1; beq >1		// Stop if iterator returned nil.
+     if (op == BC_JITERL) {
+-      |  stw TMP1, -8(RA)
+-      |   stw TMP2, -4(RA)
++      |  stw TMP1, WORD_HI-8(RA)
++      |   stw TMP2, WORD_LO-8(RA)
+       |  b =>BC_JLOOP
+     } else {
+       |  branch_RD			// Otherwise save control var + branch.
+-      |  stw TMP1, -8(RA)
+-      |   stw TMP2, -4(RA)
++      |  stw TMP1, WORD_HI-8(RA)
++      |   stw TMP2, WORD_LO-8(RA)
+     }
+     |1:
+     |  ins_next
+@@ -4942,7 +5337,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  // Traces on PPC don't store the trace number, so use 0.
+     |   stw ZERO, DISPATCH_GL(vmstate)(DISPATCH)
+     |  lwzx TRACE:TMP2, TMP1, RD
+-    |  clrso TMP1
++    |  mtxer ZERO
+     |  lp TMP2, TRACE:TMP2->mcode
+     |   stw BASE, DISPATCH_GL(jit_base)(DISPATCH)
+     |  mtctr TMP2
+@@ -4994,7 +5389,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     }
+     |
+     |3:  // Clear missing parameters.
+-    |  stwx TISNIL, BASE, NARGS8:RC
++    |  stwx TISNIL, BASE_HI, NARGS8:RC
+     |  addi NARGS8:RC, NARGS8:RC, 8
+     |  b <2
+     break;
+@@ -5011,11 +5406,11 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  lwz TMP2, L->maxstack
+     |   add TMP1, BASE, RC
+     |  add TMP0, RA, RC
+-    |   stw LFUNC:RB, 4(TMP1)		// Store copy of LFUNC.
++    |   stw LFUNC:RB, WORD_LO(TMP1)	// Store copy of LFUNC.
+     |   addi TMP3, RC, 8+FRAME_VARG
+     |    lwz KBASE, -4+PC2PROTO(k)(PC)
+     |  cmplw TMP0, TMP2
+-    |   stw TMP3, 0(TMP1)		// Store delta + FRAME_VARG.
++    |   stw TMP3, WORD_HI(TMP1)		// Store delta + FRAME_VARG.
+     |  bge ->vm_growstack_l
+     |  lbz TMP2, -4+PC2PROTO(numparams)(PC)
+     |   mr RA, BASE
+@@ -5026,18 +5421,19 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+     |  beq >3
+     |1:
+     |  cmplw RA, RC			// Less args than parameters?
+-    |   lwz TMP0, 0(RA)
+-    |   lwz TMP3, 4(RA)
++    |   lwz TMP0, WORD_HI(RA)
++    |   lwz TMP3, WORD_LO(RA)
+     |  bge >4
+-    |    stw TISNIL, 0(RA)		// Clear old fixarg slot (help the GC).
++    |    stw TISNIL, WORD_HI(RA)	// Clear old fixarg slot (help the GC).
+     |    addi RA, RA, 8
+     |2:
+     |  addic. TMP2, TMP2, -1
+-    |   stw TMP0, 8(TMP1)
+-    |   stw TMP3, 12(TMP1)
++    |   stw TMP0, WORD_HI+8(TMP1)
++    |   stw TMP3, WORD_LO+8(TMP1)
+     |    addi TMP1, TMP1, 8
+     |  bne <1
+     |3:
++    |  addi BASEP4, BASE, 4
+     |  ins_next2
+     |
+     |4:  // Clear missing parameters.
+@@ -5049,35 +5445,35 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
+   case BC_FUNCCW:
+     |  // BASE = new base, RA = BASE+framesize*8, RB = CFUNC, RC = nargs*8
+     if (op == BC_FUNCC) {
+-      |  lp RD, CFUNC:RB->f
++      |  lp FUNCREG, CFUNC:RB->f
+     } else {
+-      |  lp RD, DISPATCH_GL(wrapf)(DISPATCH)
++      |  lp FUNCREG, DISPATCH_GL(wrapf)(DISPATCH)
+     }
+     |   add TMP1, RA, NARGS8:RC
+     |   lwz TMP2, L->maxstack
+-    |  .toc lp TMP3, 0(RD)
++    |  .opd lp TMP3, 0(FUNCREG)
+     |    add RC, BASE, NARGS8:RC
+     |   stp BASE, L->base
+     |   cmplw TMP1, TMP2
+     |    stp RC, L->top
+     |     li_vmstate C
+-    |.if TOC
++    |.if OPD
+     |  mtctr TMP3
+     |.else
+-    |  mtctr RD
++    |  mtctr FUNCREG
+     |.endif
+     if (op == BC_FUNCCW) {
+       |  lp CARG2, CFUNC:RB->f
+     }
+     |  mr CARG1, L
+     |   bgt ->vm_growstack_c		// Need to grow stack.
+-    |  .toc lp TOCREG, TOC_OFS(RD)
+-    |  .tocenv lp ENVREG, ENV_OFS(RD)
++    |  .opd lp TOCREG, TOC_OFS(FUNCREG)
++    |  .opdenv lp ENVREG, ENV_OFS(FUNCREG)
+     |     st_vmstate
+     |  bctrl				// (lua_State *L [, lua_CFunction f])
++    |  .toc lp TOCREG, SAVE_TOC
+     |  // Returns nresults.
+     |  lp BASE, L->base
+-    |  .toc ld TOCREG, SAVE_TOC
+     |   slwi RD, CRET1, 3
+     |  lp TMP1, L->top
+     |    li_vmstate INTERP
+@@ -5128,7 +5524,11 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.byte 0x1\n"
+ 	"\t.string \"\"\n"
+ 	"\t.uleb128 0x1\n"
++#if LJ_ARCH_PPC32ON64
++	"\t.sleb128 -8\n"
++#else
+ 	"\t.sleb128 -4\n"
++#endif
+ 	"\t.byte 65\n"
+ 	"\t.byte 0xc\n\t.uleb128 1\n\t.uleb128 0\n"
+ 	"\t.align 2\n"
+@@ -5141,14 +5541,24 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.long .Lbegin\n"
+ 	"\t.long %d\n"
+ 	"\t.byte 0xe\n\t.uleb128 %d\n"
++#if LJ_ARCH_PPC32ON64
++	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -2\n"
++	"\t.byte 0x11\n\t.uleb128 70\n\t.sleb128 -1\n",
++#else
+ 	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -1\n"
+ 	"\t.byte 0x5\n\t.uleb128 70\n\t.uleb128 55\n",
++#endif
+ 	fcofs, CFRAME_SIZE);
+     for (i = 14; i <= 31; i++)
+       fprintf(ctx->fp,
+ 	"\t.byte %d\n\t.uleb128 %d\n"
+ 	"\t.byte %d\n\t.uleb128 %d\n",
+-	0x80+i, 37+(31-i), 0x80+32+i, 2+2*(31-i));
++#if LJ_ARCH_PPC32ON64
++	0x80+i, 19+(31-i), 0x80+32+i, 1+(31-i)
++#else
++	0x80+i, 37+(31-i), 0x80+32+i, 2+2*(31-i)
++#endif
++      );
+     fprintf(ctx->fp,
+ 	"\t.align 2\n"
+ 	".LEFDE0:\n\n");
+@@ -5164,8 +5574,12 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.long lj_vm_ffi_call\n"
+ #endif
+ 	"\t.long %d\n"
++#if LJ_ARCH_PPC32ON64
++	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -2\n"
++#else
+ 	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -1\n"
+-	"\t.byte 0x8e\n\t.uleb128 2\n"
++#endif
++	"\t.byte 0x8e\n\t.uleb128 1\n"
+ 	"\t.byte 0xd\n\t.uleb128 0xe\n"
+ 	"\t.align 2\n"
+ 	".LEFDE1:\n\n", (int)ctx->codesz - fcofs);
+@@ -5180,7 +5594,11 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.byte 0x1\n"
+ 	"\t.string \"zPR\"\n"
+ 	"\t.uleb128 0x1\n"
++#if LJ_ARCH_PPC32ON64
++	"\t.sleb128 -8\n"
++#else
+ 	"\t.sleb128 -4\n"
++#endif
+ 	"\t.byte 65\n"
+ 	"\t.uleb128 6\n"			/* augmentation length */
+ 	"\t.byte 0x1b\n"			/* pcrel|sdata4 */
+@@ -5198,14 +5616,24 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.long %d\n"
+ 	"\t.uleb128 0\n"			/* augmentation length */
+ 	"\t.byte 0xe\n\t.uleb128 %d\n"
++#if LJ_ARCH_PPC32ON64
++	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -2\n"
++	"\t.byte 0x11\n\t.uleb128 70\n\t.sleb128 -1\n",
++#else
+ 	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -1\n"
+ 	"\t.byte 0x5\n\t.uleb128 70\n\t.uleb128 55\n",
++#endif
+ 	fcofs, CFRAME_SIZE);
+     for (i = 14; i <= 31; i++)
+       fprintf(ctx->fp,
+ 	"\t.byte %d\n\t.uleb128 %d\n"
+ 	"\t.byte %d\n\t.uleb128 %d\n",
+-	0x80+i, 37+(31-i), 0x80+32+i, 2+2*(31-i));
++#if LJ_ARCH_PPC32ON64
++	0x80+i, 19+(31-i), 0x80+32+i, 1+(31-i)
++#else
++	0x80+i, 37+(31-i), 0x80+32+i, 2+2*(31-i)
++#endif
++      );
+     fprintf(ctx->fp,
+ 	"\t.align 2\n"
+ 	".LEFDE2:\n\n");
+@@ -5233,8 +5661,12 @@ static void emit_asm_debug(BuildCtx *ctx)
+ 	"\t.long lj_vm_ffi_call-.\n"
+ 	"\t.long %d\n"
+ 	"\t.uleb128 0\n"			/* augmentation length */
++#if LJ_ARCH_PPC32ON64
++	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -2\n"
++#else
+ 	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 -1\n"
+-	"\t.byte 0x8e\n\t.uleb128 2\n"
++#endif
++	"\t.byte 0x8e\n\t.uleb128 1\n"
+ 	"\t.byte 0xd\n\t.uleb128 0xe\n"
+ 	"\t.align 2\n"
+ 	".LEFDE3:\n\n", (int)ctx->codesz - fcofs);
+
diff --git a/main/luajit/APKBUILD b/main/luajit/APKBUILD
index 1dee244d20..a13ccc6e5f 100644
--- a/main/luajit/APKBUILD
+++ b/main/luajit/APKBUILD
@@ -4,7 +4,7 @@ pkgname=luajit
 _pkgname=LuaJIT
 pkgver=2.1.0_beta3
 _pkgver=${pkgver/_/-}
-pkgrel=4
+pkgrel=5
 pkgdesc='Just-in-time compiler and replacement for Lua 5.1'
 url='http://luajit.org'
 arch="all !s390x"
@@ -13,7 +13,7 @@ makedepends="$depends_dev paxmark"
 provides="lua"
 subpackages="$pkgname-dev $pkgname-doc"
 source="http://luajit.org/download/$_pkgname-$_pkgver.tar.gz
-	enable-support-for-ppc64le.patch
+	0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
 	module-paths.patch"
 builddir="$srcdir/$_pkgname-$_pkgver"
 
@@ -46,5 +46,5 @@ package() {
 }
 
 sha512sums="c44e967a0f671ed32b55aee810bc8b3b63737a2d7363b8984ae1949b24f98dbb3f9be7c1e10239fdeb96a3e3c836f606342cbd61838cf9bcadb077443eb5bc12  LuaJIT-2.1.0-beta3.tar.gz
-c2b1a39fe19dfe1817811015ef1fea24b0966ff791fcbde1090527331c2b9ceab9b923960c01cfb8aa81bab40aa437eb0f285385378e7eaa83016abf36beae27  enable-support-for-ppc64le.patch
+2e4a4416ad60f6463b2f4ce2c5b8f0519dfb7bfd85e377fed8423477088c5ac9143d0c30a772e65f492317da63bcd06ab01dabe85933c2bd32d6f061167e5515  0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
 9c6c410e142838c5128c70570e66beb53cc5422df5ecfb8ab0b5f362637b21ab5978c00f19bc1759165df162fd6438a2ed43e25020400011b7bb14014d87c62e  module-paths.patch"
diff --git a/main/luajit/enable-support-for-ppc64le.patch b/main/luajit/enable-support-for-ppc64le.patch
deleted file mode 100644
index f6a46c37c9..0000000000
--- a/main/luajit/enable-support-for-ppc64le.patch
+++ /dev/null
@@ -1,4822 +0,0 @@
-From d2400df1c753683571c5df9ea29afb37b6583d36 Mon Sep 17 00:00:00 2001
-From: Gustavo Serra Scalet <gsscalet@gmail.com>
-Date: Tue, 2 Jun 2015 14:36:20 -0300
-Subject: [PATCH] PPC64: Enable support for ppc64 little endian
-
----
- dynasm/dasm_ppc.lua    |   11 +-
- src/host/buildvm_asm.c |   19 +-
- src/lj_arch.h          |    4 +-
- src/lj_ccall.c         |   37 +
- src/lj_ccall.h         |    9 +
- src/lj_ccallback.c     |   15 +
- src/lj_ctype.h         |    2 +-
- src/lj_frame.h         |    9 +
- src/lj_target_ppc.h    |   11 +-
- src/vm_ppc64.dasc      | 4515 ++++++++++++++++++++++++++++++++++++++++++++++++
- 10 files changed, 4621 insertions(+), 11 deletions(-)
- create mode 100644 src/vm_ppc64.dasc
-
-diff --git a/dynasm/dasm_ppc.lua b/dynasm/dasm_ppc.lua
-index 278f095..7a36454 100644
---- a/dynasm/dasm_ppc.lua
-+++ b/dynasm/dasm_ppc.lua
-@@ -257,9 +257,11 @@ map_op = {
-   addic_3 =	"30000000RRI",
-   ["addic._3"] = "34000000RRI",
-   addi_3 =	"38000000RR0I",
-+  addil_3 =	"38000000RR0J",
-   li_2 =	"38000000RI",
-   la_2 =	"38000000RD",
-   addis_3 =	"3c000000RR0I",
-+  addisl_3 =	"3c000000RR0J",
-   lis_2 =	"3c000000RI",
-   lus_2 =	"3c000000RU",
-   bc_3 =	"40000000AAK",
-@@ -764,7 +766,7 @@ map_op = {
-   lfddx_3 =	"7c000646FRR",
-   stvepx_3 =	"7c00064eVRR",
-   srawi_3 =	"7c000670RR~A.",
--  sradi_3 =	"7c000674RR~H.",
-+  sradi_3 =	"7c000674RR~f.",
-   eieio_0 =	"7c0006ac",
-   lfiwax_3 =	"7c0006aeFR0R",
-   divdeuo_3 =	"7c000712RRR.",
-@@ -1718,7 +1720,12 @@ op_template = function(params, template, nparams)
-     elseif p == "G" then
-       op = op + parse_imm(params[n], 8, 12, 0, false); n = n + 1
-     elseif p == "H" then
--      op = op + parse_shiftmask(params[n], true); n = n + 1
-+      v = parse_imm(params[n], 6, 0, 0, false);
-+      op = op + shl(band(v,31), 11)+shl(shr(v,5), 1);
-+      n = n + 1;
-+    elseif p == "f" then
-+      v = tonumber(params[n]);
-+      op = op + shl(band(v,31), 11)+shl(shr(v,5), 1);
-     elseif p == "M" then
-       op = op + parse_shiftmask(params[n], false); n = n + 1
-     elseif p == "J" or p == "K" then
-diff --git a/src/host/buildvm_asm.c b/src/host/buildvm_asm.c
-index 9b7ae53..14ae701 100644
---- a/src/host/buildvm_asm.c
-+++ b/src/host/buildvm_asm.c
-@@ -139,13 +139,21 @@ static void emit_asm_wordreloc(BuildCtx *ctx, uint8_t *p, int n,
-   if ((ins >> 26) == 16) {
-     fprintf(ctx->fp, "\t%s %d, %d, " TOCPREFIX "%s\n",
- 	    (ins & 1) ? "bcl" : "bc", (ins >> 21) & 31, (ins >> 16) & 31, sym);
-+#if LJ_ARCH_PPC64
-+  } else if ((ins >> 26) == 14) {
-+    if (strcmp(sym, "TOC") < 0) {
-+      fprintf(ctx->fp, "\taddi 2,2,%s\n", sym);
-+    }
-+  } else if ((ins >> 26) == 15) {
-+    if (strcmp(sym, "TOC") < 0) {
-+      fprintf(ctx->fp, "\taddis 2,12,%s\n", sym);
-+    }
-+#endif
-   } else if ((ins >> 26) == 18) {
- #if LJ_ARCH_PPC64
--    const char *suffix = strchr(sym, '@');
--    if (suffix && suffix[1] == 'h') {
--      fprintf(ctx->fp, "\taddis 11, 2, %s\n", sym);
--    } else if (suffix && suffix[1] == 'l') {
--      fprintf(ctx->fp, "\tld 12, %s\n", sym);
-+    char *suffix = strchr(sym, '@');
-+    if (suffix) {
-+      fprintf(ctx->fp, "\tld 12, %s(2)\n", sym);
-     } else
- #endif
-     fprintf(ctx->fp, "\t%s " TOCPREFIX "%s\n", (ins & 1) ? "bl" : "b", sym);
-@@ -247,6 +255,7 @@ void emit_asm(BuildCtx *ctx)
-   fprintf(ctx->fp, "\t.file \"buildvm_%s.dasc\"\n", ctx->dasm_arch);
- #if LJ_ARCH_PPC64
-   fprintf(ctx->fp, "\t.abiversion 2\n");
-+  fprintf(ctx->fp, "\t.section\t\t\".toc\",\"aw\"\n");
- #endif
-   fprintf(ctx->fp, "\t.text\n");
-   emit_asm_align(ctx, 4);
-diff --git a/src/lj_arch.h b/src/lj_arch.h
-index f1e7d7f..e2cdda7 100644
---- a/src/lj_arch.h
-+++ b/src/lj_arch.h
-@@ -375,8 +375,8 @@
- #if !LJ_ARCH_PPC64 && LJ_ARCH_ENDIAN == LUAJIT_LE
- #error "No support for little-endian PPC32"
- #endif
--#if LJ_ARCH_PPC64
--#error "No support for PowerPC 64 bit mode (yet)"
-+#if LJ_ARCH_PPC64 && LJ_ARCH_ENDIAN == LUAJIT_BE
-+#error "No support for big-endian PPC64"
- #endif
- #ifdef __NO_FPRS__
- #error "No support for PPC/e500 anymore (use LuaJIT 2.0)"
-diff --git a/src/lj_ccall.c b/src/lj_ccall.c
-index 5ab5b60..9ecad9d 100644
---- a/src/lj_ccall.c
-+++ b/src/lj_ccall.c
-@@ -380,6 +380,42 @@
- #define CCALL_HANDLE_COMPLEXARG \
-   /* Pass complex by value in 2 or 4 GPRs. */
- 
-+#if LJ_ARCH_PPC64
-+#define CCALL_HANDLE_REGARG \
-+  if (isva) {  /* only GPRs will be used on C ellipsis operator */ \
-+    goto gpr; \
-+  } \
-+  else { \
-+    if (isfp) {  /* Try to pass argument in FPRs. */ \
-+      if (nfpr + 1 <= CCALL_NARG_FPR) { \
-+	dp = &cc->fpr[nfpr]; \
-+	nfpr += 1; \
-+	d = ctype_get(cts, CTID_DOUBLE);  /* FPRs always hold doubles. */ \
-+	if (ngpr + 1 <= maxgpr) \
-+	  ngpr += 1;  /* align GPRs */ \
-+	else if (nsp + 1 <= CCALL_MAXSTACK) \
-+	  nsp += 1; /* align save area slots */ \
-+	else \
-+	  goto err_nyi; /* Too many args */ \
-+	goto done; \
-+      } \
-+    } else {  /* Try to pass argument in GPRs. */ \
-+  gpr: \
-+      if (n > 1) { \
-+	lua_assert(n == 2 || n == 4);  /* int64_t or complex (float). */ \
-+	if (ctype_isinteger(d->info)) \
-+	  ngpr = (ngpr + 1u) & ~1u;  /* Align int64_t to regpair. */ \
-+	else if (ngpr + n > maxgpr) \
-+	  ngpr = maxgpr;  /* Prevent reordering. */ \
-+      } \
-+      if (ngpr + n <= maxgpr) { \
-+	dp = &cc->gpr[ngpr]; \
-+	ngpr += n; \
-+	goto done; \
-+      } \
-+    } \
-+  }
-+#else	/* 32 bits */
- #define CCALL_HANDLE_REGARG \
-   if (isfp) {  /* Try to pass argument in FPRs. */ \
-     if (nfpr + 1 <= CCALL_NARG_FPR) { \
-@@ -402,6 +438,7 @@
-       goto done; \
-     } \
-   }
-+#endif
- 
- #define CCALL_HANDLE_RET \
-   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
-diff --git a/src/lj_ccall.h b/src/lj_ccall.h
-index 91983fe..0994d84 100644
---- a/src/lj_ccall.h
-+++ b/src/lj_ccall.h
-@@ -85,12 +85,21 @@ typedef union FPRArg {
- 
- #elif LJ_TARGET_PPC
- 
-+#if LJ_ARCH_PPC64
-+#define CCALL_NARG_GPR		8
-+#define CCALL_NARG_FPR		13
-+#define CCALL_NRET_GPR		4	/* For complex double. */
-+#define CCALL_NRET_FPR		1
-+#define CCALL_SPS_EXTRA		14
-+#define CCALL_SPS_FREE		0
-+#else
- #define CCALL_NARG_GPR		8
- #define CCALL_NARG_FPR		8
- #define CCALL_NRET_GPR		4	/* For complex double. */
- #define CCALL_NRET_FPR		1
- #define CCALL_SPS_EXTRA		4
- #define CCALL_SPS_FREE		0
-+#endif
- 
- typedef intptr_t GPRArg;
- typedef double FPRArg;
-diff --git a/src/lj_ccallback.c b/src/lj_ccallback.c
-index 065c329..0ef527b 100644
---- a/src/lj_ccallback.c
-+++ b/src/lj_ccallback.c
-@@ -61,7 +61,11 @@ static MSize CALLBACK_OFS2SLOT(MSize ofs)
- 
- #elif LJ_TARGET_PPC
- 
-+#if LJ_ARCH_PPC64
-+#define CALLBACK_MCODE_HEAD		40
-+#else  /* PPC 32bits */
- #define CALLBACK_MCODE_HEAD		24
-+#endif
- 
- #elif LJ_TARGET_MIPS
- 
-@@ -189,10 +193,21 @@ static void callback_mcode_init(global_State *g, uint32_t *page)
-   uint32_t *p = page;
-   void *target = (void *)lj_vm_ffi_callback;
-   MSize slot;
-+#if LJ_ARCH_PPC64
-+  *p++ = PPCI_LI | PPCF_T(RID_TMP) | ((((intptr_t)target) >> 32) & 0xffff);
-+  *p++ = PPCI_LI | PPCF_T(RID_R12) | ((((intptr_t)g) >> 32) & 0xffff);
-+  *p++ = PPCI_RLDICR | PPCF_T(RID_TMP) | PPCF_A(RID_TMP) | PPCF_SH(32) | PPCF_M6(63-32);  /* sldi */
-+  *p++ = PPCI_RLDICR | PPCF_T(RID_R12) | PPCF_A(RID_R12) | PPCF_SH(32) | PPCF_M6(63-32);  /* sldi */
-+  *p++ = PPCI_ORIS | PPCF_A(RID_TMP) | PPCF_T(RID_TMP) | ((((intptr_t)target) >> 16) & 0xffff);
-+  *p++ = PPCI_ORIS | PPCF_A(RID_R12) | PPCF_T(RID_R12) | ((((intptr_t)g) >> 16) & 0xffff);
-+  *p++ = PPCI_ORI | PPCF_A(RID_TMP) | PPCF_T(RID_TMP) | (((intptr_t)target) & 0xffff);
-+  *p++ = PPCI_ORI | PPCF_A(RID_R12) | PPCF_T(RID_R12) | (((intptr_t)g) & 0xffff);
-+#else  /* PPC 32bits */
-   *p++ = PPCI_LIS | PPCF_T(RID_TMP) | (u32ptr(target) >> 16);
-   *p++ = PPCI_LIS | PPCF_T(RID_R12) | (u32ptr(g) >> 16);
-   *p++ = PPCI_ORI | PPCF_A(RID_TMP)|PPCF_T(RID_TMP) | (u32ptr(target) & 0xffff);
-   *p++ = PPCI_ORI | PPCF_A(RID_R12)|PPCF_T(RID_R12) | (u32ptr(g) & 0xffff);
-+#endif
-   *p++ = PPCI_MTCTR | PPCF_T(RID_TMP);
-   *p++ = PPCI_BCTR;
-   for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
-diff --git a/src/lj_ctype.h b/src/lj_ctype.h
-index 6639547..2f09d7d 100644
---- a/src/lj_ctype.h
-+++ b/src/lj_ctype.h
-@@ -153,7 +153,7 @@ typedef struct CType {
- 
- /* Simplify target-specific configuration. Checked in lj_ccall.h. */
- #define CCALL_MAX_GPR		8
--#define CCALL_MAX_FPR		8
-+#define CCALL_MAX_FPR		13
- 
- typedef LJ_ALIGN(8) union FPRCBArg { double d; float f[2]; } FPRCBArg;
- 
-diff --git a/src/lj_frame.h b/src/lj_frame.h
-index a86c36b..7ec6478 100644
---- a/src/lj_frame.h
-+++ b/src/lj_frame.h
-@@ -207,6 +207,15 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
- #define CFRAME_OFS_MULTRES	456
- #define CFRAME_SIZE		400
- #define CFRAME_SHIFT_MULTRES	3
-+#elif LJ_ARCH_PPC64
-+#define CFRAME_OFS_ERRF         88
-+#define CFRAME_OFS_NRES         80
-+#define CFRAME_OFS_L            72
-+#define CFRAME_OFS_PC           64
-+#define CFRAME_OFS_MULTRES      56
-+#define CFRAME_OFS_PREV         48
-+#define CFRAME_SIZE             400
-+#define CFRAME_SHIFT_MULTRES    3
- #else
- #define CFRAME_OFS_ERRF		48
- #define CFRAME_OFS_NRES		44
-diff --git a/src/lj_target_ppc.h b/src/lj_target_ppc.h
-index 9986768..7a68e2a 100644
---- a/src/lj_target_ppc.h
-+++ b/src/lj_target_ppc.h
-@@ -1,5 +1,5 @@
- /*
--** Definitions for PPC CPUs.
-+** Definitions for PPC/PPC64 CPUs.
- ** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
- */
- 
-@@ -131,6 +131,8 @@ static LJ_AINLINE uint32_t *exitstub_trace_addr_(uint32_t *p, uint32_t exitno)
- #define PPCF_C(r)	((r) << 6)
- #define PPCF_MB(n)	((n) << 6)
- #define PPCF_ME(n)	((n) << 1)
-+#define PPCF_SH(n)	((((n) & 31) << (11+1)) | (((n) & 32) >> (5-1)))
-+#define PPCF_M6(n)	((((n) & 31) << (5+1)) | (((n) & 32) << (11-5)))
- #define PPCF_Y		0x00200000
- #define PPCF_DOT	0x00000001
- 
-@@ -200,6 +202,13 @@ typedef enum PPCIns {
-   PPCI_RLWINM = 0x54000000,
-   PPCI_RLWIMI = 0x50000000,
- 
-+  PPCI_RLDICL = 0x78000000,
-+  PPCI_RLDICR = 0x78000004,
-+  PPCI_RLDIC = 0x78000008,
-+  PPCI_RLDIMI = 0x7800000c,
-+  PPCI_RLDCL = 0x78000010,
-+  PPCI_RLDCR = 0x78000012,
-+
-   PPCI_B = 0x48000000,
-   PPCI_BL = 0x48000001,
-   PPCI_BC = 0x40800000,
-diff --git a/src/vm_ppc64.dasc b/src/vm_ppc64.dasc
-new file mode 100644
-index 0000000..d7e6bb6
---- /dev/null
-+++ b/src/vm_ppc64.dasc
-@@ -0,0 +1,4515 @@
-+|// Low-level VM code for PPC64 CPUs.
-+|// Bytecode interpreter, fast functions and helper functions.
-+|// Copyright (C) 2005-2015 Mike Pall. See Copyright Notice in luajit.h
-+|
-+|.arch ppc
-+|.section code_op, code_sub
-+|
-+|.actionlist build_actionlist
-+|.globals GLOB_
-+|.globalnames globnames
-+|.externnames extnames
-+|
-+|// Note: The ragged indentation of the instructions is intentional.
-+|//       The starting columns indicate data dependencies.
-+|
-+|// Convenience macros for TOC handling.
-+|.macro blex, target
-+|  bl extern target@got		// ld 12, target@got(2)
-+|  mtctr r12
-+|  bctrl
-+|//It is require to restore the TOC register after a function call
-+|//once the linker will not imply a function from mtctr; bctrl; pair
-+|//it is required to perform such step manually in this macro
-+|  ld TOCREG, SAVE_TOC
-+|.endmacro
-+|
-+|.macro pic_code_setup, target_name
-+|  addisl r2, r12, extern .TOC.-lj_..target_name@ha
-+|  addil  r2, r2,  extern .TOC.-lj_..target_name@l
-+|.endmacro
-+|
-+|.macro checkov, noov
-+|  mcrxr cr0
-+|  bley noov
-+|.endmacro
-+|
-+|//-----------------------------------------------------------------------
-+|
-+|// Fixed register assignments for the interpreter.
-+|// Don't use: r1 = sp, r2 and r13 = reserved (TOC, TLS or SDATA)
-+|
-+|// The following must be C callee-save (but BASE is often refetched).
-+|.define BASE,		r14	// Base of current Lua stack frame.
-+|.define KBASE,		r15	// Constants of current Lua function.
-+|.define PC,		r16	// Next PC.
-+|.define DISPATCH,	r17	// Opcode dispatch table.
-+|.define LREG,		r18	// Register holding lua_State (also in SAVE_L).
-+|.define MULTRES,	r19	// Size of multi-result: (nresults+1)*8.
-+|.define JGL,		r31	// On-trace: global_State + 32768.
-+|
-+|// Constants for type-comparisons, stores and conversions. C callee-save.
-+|.define TISNUM,	r22	// Constant LJ_TISNUM << 47.
-+|.define TISNIL,	r23
-+|.define ZERO,		r24
-+|.define TOBIT,		f30	// 2^52 + 2^51.
-+|.define TONUM,		f31	// 2^52 + 2^51 + 2^31.
-+|
-+|// The following temporaries are not saved across C calls, except for RA.
-+|.define RA,		r20	// Callee-save.
-+|.define RB,		r10
-+|.define RC,		r11
-+|.define RD,		r12	// Also used as function linkage register
-+|.define INS,		r7	// Overlaps CARG5.
-+|
-+|.define TMP0,		r0
-+|.define TMP1,		r8
-+|.define TMP2,		r9
-+|.define TMP3,		r6	// Overlaps CARG4.
-+|.define TMP4,		r25	// reserved for check.* macros
-+|.define TMP5,		r26	// reserved for check.* macros
-+|
-+|// Saved temporaries.
-+|.define SAVE0,		r21
-+|
-+|// Calling conventions.
-+|.define CARG1,		r3
-+|.define CARG2,		r4
-+|.define CARG3,		r5
-+|.define CARG4,		r6	// Overlaps TMP3.
-+|.define CARG5,		r7	// Overlaps INS.
-+|
-+|.define FARG1,		f1
-+|.define FARG2,		f2
-+|
-+|.define CRET1,		r3
-+|.define CRET2,		r4
-+|
-+|.define TOCREG,	r2	// TOC register (only used by C code).
-+|.define ENVREG,	r11	// Environment pointer (nested C functions).
-+|
-+|// Stack layout while in interpreter. Must match with lj_frame.h.
-+|
-+|.define SAVE_LR,	416(sp)
-+|.define CFRAME_SPACE,	400	// Delta for sp.
-+|// Back chain for sp:	400(sp)	<-- sp entering interpreter
-+|.define SAVE_FPR_,	256	// .. 256+18*8: 64 bit FPR saves.
-+|.define SAVE_GPR_,	112	// .. 112+18*8: 64 bit GPR saves.
-+|//			92(sp)	// \ 32 bit C frame info.
-+|.define SAVE_ERRF,	88(sp)	// |
-+|.define SAVE_NRES,	80(sp)	// |
-+|.define SAVE_L,	72(sp)	//  > Parameter save area.
-+|.define SAVE_PC,	64(sp)	// |
-+|.define SAVE_MULTRES,	56(sp)	// |
-+|.define SAVE_CFRAME,	48(sp)	// / 64 bit C frame chain.
-+|.define TMPD_HI,	44(sp)	// \ Link editor temp (ABI mandated).
-+|.define TMPD_LO,	40(sp)	// /
-+|.define TONUM_HI,	36(sp)	// \ Compiler temp (ABI mandated).
-+|.define TONUM_LO,	32(sp)	// /
-+|//			32(sp)	// Callee parameter save area (ABI mandated).
-+|.define SAVE_TOC,	24(sp)	// TOC save area.
-+|// Next frame lr:	16(sp)
-+|.define SAVE_CR,	8(sp)	// 64 bit CR save.
-+|// Back chain for sp:	0(sp)   <-- sp while in interpreter
-+|
-+|.define TMPD_BLO,	40(sp)	// LSB
-+|.define TMPD,		TMPD_LO	// base address of TMPD doubleword
-+|.define TONUM_D,	TONUM_LO// base address of TONUM doubleword
-+|
-+|.macro save_, reg
-+|  std r..reg, SAVE_GPR_+(reg-14)*8(sp)
-+|  stfd f..reg, SAVE_FPR_+(reg-14)*8(sp)
-+|.endmacro
-+|.macro rest_, reg
-+|  ld r..reg, SAVE_GPR_+(reg-14)*8(sp)
-+|  lfd f..reg, SAVE_FPR_+(reg-14)*8(sp)
-+|.endmacro
-+|
-+|.macro saveregs
-+|  stdu sp, -CFRAME_SPACE(sp)
-+|  save_ 14; save_ 15; save_ 16
-+|  mflr r0
-+|  save_ 17; save_ 18; save_ 19; save_ 20; save_ 21; save_ 22
-+|  std r0, SAVE_LR
-+|  save_ 23; save_ 24; save_ 25
-+|  mfcr r0
-+|  save_ 26; save_ 27; save_ 28; save_ 29; save_ 30; save_ 31
-+|  std r0, SAVE_CR
-+|  std TOCREG, SAVE_TOC
-+|.endmacro
-+|
-+|.macro restoreregs
-+|  ld r0, SAVE_LR
-+|  ld r12, SAVE_CR
-+|  rest_ 14; rest_ 15; rest_ 16; rest_ 17; rest_ 18; rest_ 19
-+|  mtlr r0
-+|  mtcrf 0xff, r12
-+|  rest_ 20; rest_ 21; rest_ 22; rest_ 23; rest_ 24; rest_ 25
-+|  rest_ 26; rest_ 27; rest_ 28; rest_ 29; rest_ 30; rest_ 31
-+|  addi sp, sp, CFRAME_SPACE
-+|.endmacro
-+|
-+|// Type definitions. Some of these are only used for documentation.
-+|.type L,		lua_State,	LREG
-+|.type GL,		global_State
-+|.type TVALUE,		TValue
-+|.type GCOBJ,		GCobj
-+|.type STR,		GCstr
-+|.type TAB,		GCtab
-+|.type LFUNC,		GCfuncL
-+|.type CFUNC,		GCfuncC
-+|.type PROTO,		GCproto
-+|.type UPVAL,		GCupval
-+|.type NODE,		Node
-+|.type NARGS8,		int
-+|.type TRACE,		GCtrace
-+|.type SBUF,		SBuf
-+|
-+|//-----------------------------------------------------------------------
-+|
-+|// Trap for not-yet-implemented parts.
-+|.macro NYI; tw 4, sp, sp; .endmacro
-+|
-+|// int/FP conversions.
-+|.macro tonum_i, freg, reg
-+|  xoris reg, reg, 0x8000
-+|  stw reg, TONUM_LO
-+|  lfd freg, TONUM_D
-+|  fsub freg, freg, TONUM
-+|.endmacro
-+|
-+|.macro toint, reg, freg, tmpfreg
-+|  fctiwz tmpfreg, freg
-+|  stfd tmpfreg, TMPD
-+|  lwz reg, TMPD_LO
-+|.endmacro
-+|
-+|.macro toint, reg, freg
-+|  toint reg, freg, freg
-+|.endmacro
-+|
-+|//-----------------------------------------------------------------------
-+|
-+|// Access to frame relative to BASE.
-+|.define FRAME_PC,	-8
-+|.define FRAME_FUNC,	-16
-+|
-+|// Instruction decode.
-+|.macro decode_OP8, dst, ins; rlwinm dst, ins,  3, 21, 28; .endmacro
-+|.macro decode_RA8, dst, ins; rlwinm dst, ins, 27, 21, 28; .endmacro
-+|.macro decode_RB8, dst, ins; rlwinm dst, ins, 11, 21, 28; .endmacro
-+|.macro decode_RC8, dst, ins; rlwinm dst, ins, 19, 21, 28; .endmacro
-+|.macro decode_RD8, dst, ins; rlwinm dst, ins, 19, 13, 28; .endmacro
-+|
-+|.macro decode_OP1, dst, ins; rlwinm dst, ins,  0, 24, 31; .endmacro
-+|.macro decode_RD4, dst, ins; rlwinm dst, ins, 18, 14, 29; .endmacro
-+|
-+|// Instruction fetch.
-+|.macro ins_NEXT1
-+|  lwz INS, 0(PC)
-+|   addi PC, PC, 4
-+|.endmacro
-+|// Instruction decode+dispatch. Note: optimized for e300!
-+|.macro ins_NEXT2
-+|  decode_OP8 TMP1, INS
-+|  ldx TMP0, DISPATCH, TMP1
-+|  mtctr TMP0
-+|   decode_RB8 RB, INS
-+|   decode_RD8 RD, INS
-+|   decode_RA8 RA, INS
-+|   decode_RC8 RC, INS
-+|  bctr
-+|.endmacro
-+|.macro ins_NEXT
-+|  ins_NEXT1
-+|  ins_NEXT2
-+|.endmacro
-+|
-+|// Instruction footer.
-+|.if 1
-+|  // Replicated dispatch. Less unpredictable branches, but higher I-Cache use.
-+|  .define ins_next, ins_NEXT
-+|  .define ins_next_, ins_NEXT
-+|  .define ins_next1, ins_NEXT1
-+|  .define ins_next2, ins_NEXT2
-+|.else
-+|  // Common dispatch. Lower I-Cache use, only one (very) unpredictable branch.
-+|  // Affects only certain kinds of benchmarks (and only with -j off).
-+|  .macro ins_next
-+|    b ->ins_next
-+|  .endmacro
-+|  .macro ins_next1
-+|  .endmacro
-+|  .macro ins_next2
-+|    b ->ins_next
-+|  .endmacro
-+|  .macro ins_next_
-+|  ->ins_next:
-+|    ins_NEXT
-+|  .endmacro
-+|.endif
-+|
-+|// Call decode and dispatch.
-+|.macro ins_callt
-+|  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC
-+|  ld PC, LFUNC:RB->pc
-+|  lwz INS, 0(PC)
-+|   addi PC, PC, 4
-+|  decode_OP8 TMP1, INS
-+|   decode_RA8 RA, INS
-+|  ldx TMP0, DISPATCH, TMP1
-+|   add RA, RA, BASE
-+|  mtctr TMP0
-+|  bctr
-+|.endmacro
-+|
-+|.macro ins_call
-+|  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, PC = caller PC
-+|  std PC, FRAME_PC(BASE)
-+|  ins_callt
-+|.endmacro
-+|
-+|//-----------------------------------------------------------------------
-+|
-+|// Macros to test operand types.
-+|.macro get_oper_type, dst, reg
-+|  sradi dst, reg, 47
-+|.endmacro
-+|
-+|.macro set_oper_type, dst, reg
-+|  sldi dst, reg, 47
-+|.endmacro
-+|
-+|.macro set_bool, reg, type; li TMP4, type; rotldi reg, TMP4, 47; .endmacro
-+|
-+|.macro add_oper_type, value, type
-+|  clear_field value
-+|  add value, value, type
-+|.endmacro
-+|
-+|.macro clear_field, reg
-+|  clrldi reg, reg, 17
-+|.endmacro
-+|
-+|// clear bits 32-63 of 32-bit numbers and extend sign
-+|.macro get_value, reg
-+|  srawi reg, reg, 0
-+|.endmacro
-+|
-+|.macro checktp, cr, reg, clear, cmp_type, cmp_arg
-+|  get_oper_type TMP5, reg
-+|  cmp_type cr, TMP5, cmp_arg
-+|  clear reg
-+|.endmacro
-+|
-+|.macro nop_arg, p; .endmacro
-+|
-+|.macro checknum_sig, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr0, reg, get_value, cmpd, TMP4
-+|.endmacro
-+|.macro checknum_sig, cr, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr, reg, get_value, cmpd, TMP4
-+|.endmacro
-+|.macro checknum, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr0, reg, get_value, cmpld, TMP4
-+|.endmacro
-+|.macro checknum, cr, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr, reg, get_value, cmpld, TMP4
-+|.endmacro
-+|.macro checkstr, reg
-+|  li TMP4, LJ_TSTR
-+|  checktp cr0, reg, clear_field, cmpd, TMP4
-+|.endmacro
-+|.macro checktab, reg
-+|  li TMP4, LJ_TTAB
-+|  checktp cr0, reg, clear_field, cmpd, TMP4
-+|.endmacro
-+|.macro checkfunc, reg
-+|  li TMP4, LJ_TFUNC
-+|  checktp cr0, reg, clear_field, cmpd, TMP4
-+|.endmacro
-+|.macro checkthread, reg
-+|  li TMP4, LJ_TTHREAD
-+|  checktp cr0, reg, clear_field, cmpd, TMP4
-+|.endmacro
-+|
-+|.macro checknil_noclear, reg
-+|  li TMP4, LJ_TNIL
-+|  checktp cr0, reg, nop_arg, cmpd, TMP4
-+|.endmacro
-+|.macro checknum_noclear, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr0, reg, nop_arg, cmpld, TMP4
-+|.endmacro
-+|.macro checknum_noclear, cr, reg
-+|  li TMP4, LJ_TISNUM
-+|  checktp cr, reg, nop_arg, cmpld, TMP4
-+|.endmacro
-+|.macro checkfunc_noclear, reg
-+|  li TMP4, LJ_TFUNC
-+|  checktp cr0, reg, nop_arg, cmpd, TMP4
-+|.endmacro
-+|
-+|.macro branch_RD
-+|  srdi TMP0, RD, 1
-+|  addis PC, PC, -(BCBIAS_J*4 >> 16)
-+|  add PC, PC, TMP0
-+|.endmacro
-+|
-+|// Assumes DISPATCH is relative to GL.
-+#define DISPATCH_GL(field)	(GG_DISP2G + (int)offsetof(global_State, field))
-+#define DISPATCH_J(field)	(GG_DISP2J + (int)offsetof(jit_State, field))
-+|
-+#define PC2PROTO(field)  ((int)offsetof(GCproto, field)-(int)sizeof(GCproto))
-+|
-+|.macro hotcheck, delta, target
-+| NYI
-+|.endmacro
-+|
-+|.macro hotloop
-+|  hotcheck HOTCOUNT_LOOP, ->vm_hotloop
-+|.endmacro
-+|
-+|.macro hotcall
-+|  hotcheck HOTCOUNT_CALL, ->vm_hotcall
-+|.endmacro
-+|
-+|// Set current VM state. Uses TMP0.
-+|.macro li_vmstate, st; li TMP0, ~LJ_VMST_..st; .endmacro
-+|.macro st_vmstate; std TMP0, DISPATCH_GL(vmstate)(DISPATCH); .endmacro
-+|
-+|// Move table write barrier back. Overwrites mark and tmp.
-+|.macro barrierback, tab, mark, tmp
-+|  ld tmp, DISPATCH_GL(gc.grayagain)(DISPATCH)
-+|  // Assumes LJ_GC_BLACK is 0x04.
-+|   rlwinm mark, mark, 0, 30, 28	// black2gray(tab)
-+|  std tab, DISPATCH_GL(gc.grayagain)(DISPATCH)
-+|   stb mark, tab->marked
-+|  std tmp, tab->gclist
-+|.endmacro
-+|
-+|//-----------------------------------------------------------------------
-+
-+/* Generate subroutines used by opcodes and other parts of the VM. */
-+/* The .code_sub section should be last to help static branch prediction. */
-+static void build_subroutines(BuildCtx *ctx)
-+{
-+  |.code_sub
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Return handling ----------------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vm_returnp:
-+  |  // See vm_return. Also: TMP2 = previous base.
-+  |  andi. TMP0, PC, FRAME_P
-+  |   set_bool TMP1, LJ_TTRUE
-+  |  beq ->cont_dispatch
-+  |
-+  |  // Return from pcall or xpcall fast func.
-+  |  ld PC, FRAME_PC(TMP2)		// Fetch PC of previous frame.
-+  |  mr BASE, TMP2			// Restore caller base.
-+  |  // Prepending may overwrite the pcall frame, so do it at the end.
-+  |   stdu TMP1, FRAME_PC(RA)		// Prepend true to results.
-+  |
-+  |->vm_returnc:
-+  |  addi RD, RD, 8			// RD = (nresults+1)*8.
-+  |   andi. TMP0, PC, FRAME_TYPE
-+  |  cmpdi cr1, RD, 0
-+  |  li CRET1, LUA_YIELD
-+  |  beq cr1, ->vm_unwind_c_eh
-+  |  mr MULTRES, RD
-+  |   beq ->BC_RET_Z			// Handle regular return to Lua.
-+  |
-+  |->vm_return:
-+  |  // BASE = base, RA = resultptr, RD/MULTRES = (nresults+1)*8, PC = return
-+  |  // TMP0 = PC & FRAME_TYPE
-+  |  cmpdi TMP0, FRAME_C
-+  |   li TMP4, FRAME_TYPEP
-+  |   not TMP4, TMP4
-+  |    li_vmstate C
-+  |   and TMP2, PC, TMP4
-+  |   sub TMP2, BASE, TMP2		// TMP2 = previous base.
-+  |  bney ->vm_returnp
-+  |
-+  |  addic. TMP1, RD, -8
-+  |   std TMP2, L->base
-+  |   lwz TMP2, SAVE_NRES
-+  |   extsw TMP2, TMP2
-+  |    subi BASE, BASE, 16
-+  |    st_vmstate
-+  |   sldi TMP2, TMP2, 3
-+  |  beq >2
-+  |1:
-+  |  addic. TMP1, TMP1, -8
-+  |   lfd f0, 0(RA)
-+  |    addi RA, RA, 8
-+  |   stfd f0, 0(BASE)
-+  |    addi BASE, BASE, 8
-+  |  bney <1
-+  |
-+  |2:
-+  |  cmpd TMP2, RD			// More/less results wanted?
-+  |  bne >6
-+  |3:
-+  |  std BASE, L->top			// Store new top.
-+  |
-+  |->vm_leave_cp:
-+  |  ld TMP0, SAVE_CFRAME		// Restore previous C frame.
-+  |   li CRET1, 0			// Ok return status for vm_pcall.
-+  |  std TMP0, L->cframe
-+  |
-+  |->vm_leave_unw:
-+  |  restoreregs
-+  |  blr
-+  |
-+  |6:
-+  |  ble >7				// Less results wanted?
-+  |  // More results wanted. Check stack size and fill up results with nil.
-+  |  ld TMP1, L->maxstack
-+  |  cmpld BASE, TMP1
-+  |  bge >8
-+  |  std TISNIL, 0(BASE)
-+  |  addi RD, RD, 8
-+  |  addi BASE, BASE, 8
-+  |  b <2
-+  |
-+  |7:  // Less results wanted.
-+  |  subfic TMP3, TMP2, 0		// LUA_MULTRET+1 case?
-+  |   sub TMP0, RD, TMP2
-+  |  subfe TMP1, TMP1, TMP1		// TMP1 = TMP2 == 0 ? 0 : -1
-+  |   and TMP0, TMP0, TMP1
-+  |  sub BASE, BASE, TMP0		// Either keep top or shrink it.
-+  |  b <3
-+  |
-+  |8:  // Corner case: need to grow stack for filling up results.
-+  |  // This can happen if:
-+  |  // - A C function grows the stack (a lot).
-+  |  // - The GC shrinks the stack in between.
-+  |  // - A return back from a lua_call() with (high) nresults adjustment.
-+  |  std BASE, L->top			// Save current top held in BASE (yes).
-+  |   mr SAVE0, RD
-+  |  srdi CARG2, TMP2, 3
-+  |  mr CARG1, L
-+  |  bl extern lj_state_growstack	// (lua_State *L, int n)
-+  |    ld TMP2, SAVE_NRES
-+  |   mr RD, SAVE0
-+  |    sldi TMP2, TMP2, 3
-+  |  ld BASE, L->top			// Need the (realloced) L->top in BASE.
-+  |  b <2
-+  |
-+  |->vm_unwind_c:			// Unwind C stack, return from vm_pcall.
-+  |  // (void *cframe, int errcode)
-+  |  mr sp, CARG1
-+  |  mr CRET1, CARG2
-+  |->vm_unwind_c_eh:			// Landing pad for external unwinder.
-+  |  ld L, SAVE_L
-+  |  ld TOCREG, SAVE_TOC
-+  |   li TMP0, ~LJ_VMST_C
-+  |  ld GL:TMP1, L->glref
-+  |   std TMP0, GL:TMP1->vmstate
-+  |  b ->vm_leave_unw
-+  |
-+  |->vm_unwind_ff:			// Unwind C stack, return from ff pcall.
-+  |  // (void *cframe)
-+  |  rldicr sp, CARG1, 0, 61
-+  |->vm_unwind_ff_eh:			// Landing pad for external unwinder.
-+  |  ld L, SAVE_L
-+  |  ld TOCREG, SAVE_TOC
-+  |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
-+  |     set_oper_type TISNUM, TISNUM
-+  |  ld BASE, L->base
-+  |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
-+  |   ld DISPATCH, L->glref		// Setup pointer to dispatch table.
-+  |     li ZERO, 0
-+  |     std TMP3, TMPD
-+  |  set_bool TMP1, LJ_TFALSE
-+  |     ori TMP3, TMP3, 0x0004		// TONUM = 2^52 + 2^51 + 2^31 (float).
-+  |     li TISNIL, LJ_TNIL
-+  |    li_vmstate INTERP
-+  |     lfs TOBIT, TMPD
-+  |  ld PC, FRAME_PC(BASE)		// Fetch PC of previous frame.
-+  |  la RA, -8(BASE)			// Results start at BASE-8.
-+  |     std TMP3, TMPD
-+  |   addi DISPATCH, DISPATCH, GG_G2DISP
-+  |  std TMP1, 0(RA)			// Prepend false to error message.
-+  |  li RD, 16				// 2 results: false + error message.
-+  |    st_vmstate
-+  |     lfs TONUM, TMPD
-+  |  b ->vm_returnc
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Grow stack for calls -----------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vm_growstack_c:			// Grow stack for C function.
-+  |  li CARG2, LUA_MINSTACK
-+  |  b >2
-+  |
-+  |->vm_growstack_l:			// Grow stack for Lua function.
-+  |  // BASE = new base, RA = BASE+framesize*8, RC = nargs*8, PC = first PC
-+  |  add RC, BASE, RC
-+  |   sub RA, RA, BASE
-+  |  std BASE, L->base
-+  |   addi PC, PC, 4			// Must point after first instruction.
-+  |  std RC, L->top
-+  |   srdi CARG2, RA, 3
-+  |2:
-+  |  // L->base = new base, L->top = top
-+  |   std PC, SAVE_PC
-+  |  mr CARG1, L
-+  |  bl extern lj_state_growstack	// (lua_State *L, int n)
-+  |  ld BASE, L->base
-+  |  ld RC, L->top
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+  |  sub RC, RC, BASE
-+  |  clear_field RB
-+  |  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC
-+  |  ins_callt				// Just retry the call.
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Entry points into the assembler VM ---------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vm_resume:				// Setup C frame and resume thread.
-+  |  // (lua_State *L, TValue *base, int nres1 = 0, ptrdiff_t ef = 0)
-+  |  saveregs
-+  |  mr L, CARG1
-+  |    ld DISPATCH, L->glref		// Setup pointer to dispatch table.
-+  |  mr BASE, CARG2
-+  |    lbz TMP1, L->status
-+  |   std L, SAVE_L
-+  |  li PC, FRAME_CP
-+  |  addi TMP0, sp, CFRAME_RESUME
-+  |    addi DISPATCH, DISPATCH, GG_G2DISP
-+  |   std CARG3, SAVE_NRES
-+  |    cmpldi TMP1, 0
-+  |   std CARG3, SAVE_ERRF
-+  |   std CARG3, SAVE_CFRAME
-+  |   std CARG1, SAVE_PC		// Any value outside of bytecode is ok.
-+  |  std TMP0, L->cframe
-+  |    beq >3
-+  |
-+  |  // Resume after yield (like a return).
-+  |  std L, DISPATCH_GL(cur_L)(DISPATCH)
-+  |  mr RA, BASE
-+  |   ld BASE, L->base
-+  |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
-+  |     set_oper_type TISNUM, TISNUM
-+  |   ld TMP1, L->top
-+  |  ld PC, FRAME_PC(BASE)
-+  |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
-+  |    stb CARG3, L->status
-+  |     std TMP3, TMPD
-+  |     ori TMP3, TMP3, 0x0004		// TONUM = 2^52 + 2^51 + 2^31 (float).
-+  |     lfs TOBIT, TMPD
-+  |   sub RD, TMP1, BASE
-+  |     std TMP3, TMPD
-+  |     lus TMP0, 0x4338		// Hiword of 2^52 + 2^51 (double)
-+  |   addi RD, RD, 8
-+  |     stw TMP0, TONUM_HI
-+  |    li_vmstate INTERP
-+  |     li ZERO, 0
-+  |    st_vmstate
-+  |  andi. TMP0, PC, FRAME_TYPE
-+  |   mr MULTRES, RD
-+  |     lfs TONUM, TMPD
-+  |     li TISNIL, LJ_TNIL
-+  |  beq ->BC_RET_Z
-+  |  b ->vm_return
-+  |
-+  |->vm_pcall:				// Setup protected C frame and enter VM.
-+  |  // (lua_State *L, TValue *base, int nres1, ptrdiff_t ef)
-+  |  saveregs
-+  |  li PC, FRAME_CP
-+  |  std CARG4, SAVE_ERRF
-+  |  b >1
-+  |
-+  |->vm_call:				// Setup C frame and enter VM.
-+  |  // (lua_State *L, TValue *base, int nres1)
-+  |  saveregs
-+  |  li PC, FRAME_C
-+  |
-+  |1:  // Entry point for vm_pcall above (PC = ftype).
-+  |  ld TMP1, L:CARG1->cframe
-+  |    mr L, CARG1
-+  |   std CARG3, SAVE_NRES
-+  |    ld DISPATCH, L->glref		// Setup pointer to dispatch table.
-+  |   std CARG1, SAVE_L
-+  |     mr BASE, CARG2
-+  |    addi DISPATCH, DISPATCH, GG_G2DISP
-+  |   std CARG1, SAVE_PC		// Any value outside of bytecode is ok.
-+  |  std TMP1, SAVE_CFRAME
-+  |  std sp, L->cframe			// Add our C frame to cframe chain.
-+  |
-+  |3:  // Entry point for vm_cpcall/vm_resume (BASE = base, PC = ftype).
-+  |  std L, DISPATCH_GL(cur_L)(DISPATCH)
-+  |  ld TMP2, L->base			// TMP2 = old base (used in vmeta_call).
-+  |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
-+  |     set_oper_type TISNUM, TISNUM
-+  |   ld TMP1, L->top
-+  |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
-+  |  add PC, PC, BASE
-+  |     stw TMP3, TMPD
-+  |     li ZERO, 0
-+  |     ori TMP3, TMP3, 0x0004		// TONUM = 2^52 + 2^51 + 2^31 (float).
-+  |     lfs TOBIT, TMPD
-+  |  sub PC, PC, TMP2			// PC = frame delta + frame type
-+  |     stw TMP3, TMPD
-+  |     lus TMP0, 0x4338		// Hiword of 2^52 + 2^51 (double)
-+  |   sub NARGS8:RC, TMP1, BASE
-+  |     stw TMP0, TONUM_HI
-+  |    li_vmstate INTERP
-+  |     lfs TONUM, TMPD
-+  |     li TISNIL, LJ_TNIL
-+  |    st_vmstate
-+  |
-+  |->vm_call_dispatch:
-+  |  // TMP2 = old base, BASE = new base, RC = nargs*8, PC = caller PC
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+  |  checkfunc RB
-+  |  bne ->vmeta_call
-+  |
-+  |->vm_call_dispatch_f:
-+  |  ins_call
-+  |  // BASE = new base, RB = func, RC = nargs*8, PC = caller PC
-+  |
-+  |->vm_cpcall:				// Setup protected C frame, call C.
-+  |  // (lua_State *L, lua_CFunction func, void *ud, lua_CPFunction cp)
-+  |  saveregs
-+  |  mr L, CARG1
-+  |   ld TMP0, L:CARG1->stack
-+  |  std CARG1, SAVE_L
-+  |   ld TMP1, L->top
-+  |     ld DISPATCH, L->glref		// Setup pointer to dispatch table.
-+  |  std CARG1, SAVE_PC			// Any value outside of bytecode is ok.
-+  |   sub TMP0, TMP0, TMP1		// Compute -savestack(L, L->top).
-+  |    ld TMP1, L->cframe
-+  |     addi DISPATCH, DISPATCH, GG_G2DISP
-+  |  li TMP2, 0
-+  |   std TMP0, SAVE_NRES		// Neg. delta means cframe w/o frame.
-+  |  std TMP2, SAVE_ERRF		// No error function.
-+  |    std TMP1, SAVE_CFRAME
-+  |    std sp, L->cframe		// Add our C frame to cframe chain.
-+  |     std L, DISPATCH_GL(cur_L)(DISPATCH)
-+  |  mr r12, CARG4			// keep r12 for function linkage.
-+  |  mtctr r12
-+  |  bctrl			// (lua_State *L, lua_CFunction func, void *ud)
-+  |  mr. BASE, CRET1
-+  |   li PC, FRAME_CP
-+  |  bne <3				// Else continue with the call.
-+  |  b ->vm_leave_cp			// No base? Just remove C frame.
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Metamethod handling ------------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |// The lj_meta_* functions (except for lj_meta_cat) don't reallocate the
-+  |// stack, so BASE doesn't need to be reloaded across these calls.
-+  |
-+  |//-- Continuation dispatch ----------------------------------------------
-+  |
-+  |->cont_dispatch:
-+  |  // BASE = meta base, RA = resultptr, RD = (nresults+1)*8
-+  |  ld TMP0, -32(BASE)			// Continuation.
-+  |  clear_field TMP0
-+  |   mr RB, BASE
-+  |   mr BASE, TMP2			// Restore caller BASE.
-+  |    ld LFUNC:TMP1, FRAME_FUNC(TMP2)
-+  |    clear_field LFUNC:TMP1
-+  |.if FFI
-+  |  cmpldi TMP0, 1
-+  |.endif
-+  |     ld PC, -24(RB)			// Restore PC from [cont|PC].
-+  |   subi TMP2, RD, 8
-+  |    ld TMP1, LFUNC:TMP1->pc
-+  |   stdx TISNIL, RA, TMP2		// Ensure one valid arg.
-+  |.if FFI
-+  |  ble >1
-+  |.endif
-+  |    ld KBASE, PC2PROTO(k)(TMP1)
-+  |  // BASE = base, RA = resultptr, RB = meta base
-+  |  mtctr TMP0
-+  |  bctr				// Jump to continuation.
-+  |
-+  |.if FFI
-+  |1:
-+  |  beq ->cont_ffi_callback		// cont = 1: return from FFI callback.
-+  |  // cont = 0: tailcall from C function.
-+  |  subi TMP1, RB, 32
-+  |  sub RC, TMP1, BASE
-+  |  b ->vm_call_tail
-+  |.endif
-+  |
-+  |->cont_cat:				// RA = resultptr, RB = meta base
-+  |  lwz INS, -4(PC)
-+  |   subi CARG2, RB, 32
-+  |  decode_RB8 SAVE0, INS
-+  |   lfd f0, 0(RA)
-+  |  add TMP1, BASE, SAVE0
-+  |   std BASE, L->base
-+  |  cmpld TMP1, CARG2
-+  |   sub CARG3, CARG2, TMP1
-+  |  decode_RA8 RA, INS
-+  |   stfd f0, 0(CARG2)
-+  |  bney ->BC_CAT_Z
-+  |   stfdx f0, BASE, RA
-+  |  b ->cont_nop
-+  |
-+  |//-- Table indexing metamethods -----------------------------------------
-+  |
-+  |->vmeta_tgets1:
-+  |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
-+  |  li TMP0, LJ_TSTR
-+  |  set_oper_type TMP0, TMP0
-+  |   decode_RB8 RB, INS
-+  |  add STR:RC, RC, TMP0
-+  |   add CARG2, BASE, RB
-+  |  std STR:RC, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tgets:
-+  |  la CARG2, DISPATCH_GL(tmptv)(DISPATCH)
-+  |  li TMP0, LJ_TTAB
-+  |  set_oper_type TMP0, TMP0
-+  |  add TMP0, TMP0, TAB:RB
-+  |   la CARG3, DISPATCH_GL(tmptv2)(DISPATCH)
-+  |  std TMP0, 0(CARG2)
-+  |   li TMP1, LJ_TSTR
-+  |   set_oper_type TMP1, TMP1
-+  |   add TMP1, TMP1, TAB:RC
-+  |   std TMP1, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tgetb:			// TMP0 = index
-+  |   decode_RB8 RB, INS
-+  |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
-+  |   add CARG2, BASE, RB
-+  | li TISNUM, LJ_TISNUM
-+  | set_oper_type TISNUM, TISNUM
-+  | add TMP0, TMP0, TISNUM
-+  | std TMP0, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tgetv:
-+  |  decode_RB8 RB, INS
-+  |   decode_RC8 RC, INS
-+  |  add CARG2, BASE, RB
-+  |   add CARG3, BASE, RC
-+  |1:
-+  |  std BASE, L->base
-+  |  mr CARG1, L
-+  |  std PC, SAVE_PC
-+  |  bl extern lj_meta_tget		// (lua_State *L, TValue *o, TValue *k)
-+  |  // Returns TValue * (finished) or NULL (metamethod).
-+  |  cmpldi CRET1, 0
-+  |  beq >3
-+  |   lfd f0, 0(CRET1)
-+  |  ins_next1
-+  |   stfdx f0, BASE, RA
-+  |  ins_next2
-+  |
-+  |3:  // Call __index metamethod.
-+  |  // BASE = base, L->top = new base, stack = cont/func/t/k
-+  |  subfic TMP1, BASE, FRAME_CONT
-+  |  ld BASE, L->top
-+  |  std PC, -24(BASE)			// [cont|PC]
-+  |   add PC, TMP1, BASE
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)	// Guaranteed to be a function here.
-+  |   li NARGS8:RC, 16			// 2 args for func(t, k).
-+  |  clear_field RB
-+  |  b ->vm_call_dispatch_f
-+  |
-+  |->vmeta_tgetr:
-+  |  bl extern lj_tab_getinth		// (GCtab *t, int32_t key)
-+  |  // Returns cTValue * or NULL.
-+  |  cmpldi CRET1, 0
-+  |  beq >1
-+  |  lfd f14, 0(CRET1)
-+  |  b ->BC_TGETR_Z
-+  |1:
-+  |  stdx TISNIL, BASE, RA
-+  |  b ->cont_nop
-+  |
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vmeta_tsets1:
-+  |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
-+  |  li TMP0, LJ_TSTR
-+  |  set_oper_type TMP0, TMP0
-+  |   decode_RB8 RB, INS
-+  |  add TMP0, TMP0, STR:RC
-+  |   add CARG2, BASE, RB
-+  |  std TMP0, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tsets:
-+  |  la CARG2, DISPATCH_GL(tmptv)(DISPATCH)
-+  |  li TMP0, LJ_TTAB
-+  |  set_oper_type TMP0, TMP0
-+  |  add TAB:RB, TMP0, TAB:RB
-+  |   la CARG3, DISPATCH_GL(tmptv2)(DISPATCH)
-+  |  std TAB:RB, 0(CARG2)
-+  |   li TMP1, LJ_TSTR
-+  |   set_oper_type TMP1, TMP1
-+  |   add STR:RC, TMP1, STR:RC
-+  |   la CARG3, DISPATCH_GL(tmptv2)(DISPATCH)
-+  |   std STR:RC, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tsetb:			// TMP0 = index
-+  |   decode_RB8 RB, INS
-+  |  la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
-+  |   add CARG2, BASE, RB
-+  |  li TISNUM, LJ_TISNUM
-+  |  set_oper_type TISNUM, TISNUM
-+  |  add TMP0, TMP0, TISNUM		// assume TMP0 has no type
-+  |  std TMP0, 0(CARG3)
-+  |  b >1
-+  |
-+  |->vmeta_tsetv:
-+  |  decode_RB8 RB, INS
-+  |   decode_RC8 RC, INS
-+  |  add CARG2, BASE, RB
-+  |   add CARG3, BASE, RC
-+  |1:
-+  |  std BASE, L->base
-+  |  mr CARG1, L
-+  |  std PC, SAVE_PC
-+  |  bl extern lj_meta_tset		// (lua_State *L, TValue *o, TValue *k)
-+  |  // Returns TValue * (finished) or NULL (metamethod).
-+  |  cmpldi CRET1, 0
-+  |   lfdx f0, BASE, RA
-+  |  beq >3
-+  |  // NOBARRIER: lj_meta_tset ensures the table is not black.
-+  |  ins_next1
-+  |   stfd f0, 0(CRET1)
-+  |  ins_next2
-+  |
-+  |3:  // Call __newindex metamethod.
-+  |  // BASE = base, L->top = new base, stack = cont/func/t/k/(v)
-+  |  subfic TMP1, BASE, FRAME_CONT
-+  |  ld BASE, L->top
-+  |  std PC, -24(BASE)			// [cont|PC]
-+  |   add PC, TMP1, BASE
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)	// Guaranteed to be a function here.
-+  |   li NARGS8:RC, 24			// 3 args for func(t, k, v)
-+  |  stfd f0, 16(BASE)			// Copy value to third argument.
-+  |  clear_field RB
-+  |  b ->vm_call_dispatch_f
-+  |
-+  |->vmeta_tsetr:
-+  |  std BASE, L->base
-+  |  std PC, SAVE_PC
-+  |  bl extern lj_tab_setinth  // (lua_State *L, GCtab *t, int32_t key)
-+  |  // Returns TValue *.
-+  |  stfd f14, 0(CRET1)
-+  |  b ->cont_nop
-+  |
-+  |//-- Comparison metamethods ---------------------------------------------
-+  |
-+  |->vmeta_comp:
-+  |  mr CARG1, L
-+  |   subi PC, PC, 4
-+  |  mr CARG2, RA
-+  |   std PC, SAVE_PC
-+  |  mr CARG3, RD
-+  |   std BASE, L->base
-+  |  decode_OP1 CARG4, INS
-+  |  bl extern lj_meta_comp  // (lua_State *L, TValue *o1, *o2, int op)
-+  |  // Returns 0/1 or TValue * (metamethod).
-+  |3:
-+  |  cmpldi CRET1, 1
-+  |  bgt ->vmeta_binop
-+  |  subfic CRET1, CRET1, 0
-+  |4:
-+  |  lwz INS, 0(PC)
-+  |   addi PC, PC, 4
-+  |  decode_RD4 TMP2, INS
-+  |  addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+  |  and TMP2, TMP2, CRET1
-+  |  add PC, PC, TMP2
-+  |->cont_nop:
-+  |  ins_next
-+  |
-+  |->cont_ra:				// RA = resultptr
-+  |  lwz INS, -4(PC)
-+  |   lfd f0, 0(RA)
-+  |  decode_RA8 TMP1, INS
-+  |   stfdx f0, BASE, TMP1
-+  |  b ->cont_nop
-+  |
-+  |->cont_condt:			// RA = resultptr
-+  |  ld TMP0, 0(RA)
-+  |  get_oper_type TMP0, TMP0
-+  |  subfic TMP0, TMP0, LJ_TTRUE	// Branch if result is true.
-+  |  set_oper_type TMP0, TMP0
-+  |  subfe CRET1, CRET1, CRET1
-+  |  not CRET1, CRET1
-+  |  b <4
-+  |
-+  |->cont_condf:			// RA = resultptr
-+  |  ld TMP0, 0(RA)
-+  |  get_oper_type TMP0, TMP0
-+  |  subfic TMP0, TMP0, LJ_TTRUE	// Branch if result is false.
-+  |  set_oper_type TMP0, TMP0
-+  |  subfe CRET1, CRET1, CRET1
-+  |  b <4
-+  |
-+  |->vmeta_equal:
-+  |  // CARG2, CARG3, CARG4 are already set by BC_ISEQV/BC_ISNEV.
-+  |  clear_field CARG3
-+  |  subi PC, PC, 4
-+  |   std BASE, L->base
-+  |  mr CARG1, L
-+  |   std PC, SAVE_PC
-+  |  bl extern lj_meta_equal  // (lua_State *L, GCobj *o1, *o2, int ne)
-+  |  // Returns 0/1 or TValue * (metamethod).
-+  |  b <3
-+  |
-+  |->vmeta_equal_cd:
-+  |.if FFI
-+  |  mr CARG2, INS
-+  |  subi PC, PC, 4
-+  |   std BASE, L->base
-+  |  mr CARG1, L
-+  |   std PC, SAVE_PC
-+  |  bl extern lj_meta_equal_cd		// (lua_State *L, BCIns op)
-+  |  // Returns 0/1 or TValue * (metamethod).
-+  |  b <3
-+  |.endif
-+  |
-+  |->vmeta_istype:
-+  |  subi PC, PC, 4
-+  |   std BASE, L->base
-+  |   srdi CARG2, RA, 3
-+  |   mr CARG1, L
-+  |   srdi CARG3, RD, 3
-+  |  std PC, SAVE_PC
-+  |  bl extern lj_meta_istype  // (lua_State *L, BCReg ra, BCReg tp)
-+  |  b ->cont_nop
-+  |
-+  |//-- Arithmetic metamethods ---------------------------------------------
-+  |
-+  |->vmeta_arith_nv:
-+  |  add CARG3, KBASE, RC
-+  |  add CARG4, BASE, RB
-+  |  b >1
-+  |->vmeta_arith_nv2:
-+  |  mr CARG3, RC
-+  |  mr CARG4, RB
-+  |  b >1
-+  |
-+  |->vmeta_unm:
-+  |  mr CARG3, RD
-+  |  mr CARG4, RD
-+  |  b >1
-+  |
-+  |->vmeta_arith_vn:
-+  |  add CARG3, BASE, RB
-+  |  add CARG4, KBASE, RC
-+  |  b >1
-+  |
-+  |->vmeta_arith_vv:
-+  |  add CARG3, BASE, RB
-+  |  add CARG4, BASE, RC
-+  |  b >1
-+  |->vmeta_arith_vn2:
-+  |->vmeta_arith_vv2:
-+  |  mr CARG3, RB
-+  |  mr CARG4, RC
-+  |1:
-+  |  add CARG2, BASE, RA
-+  |   std BASE, L->base
-+  |  mr CARG1, L
-+  |   std PC, SAVE_PC
-+  |  decode_OP1 CARG5, INS		// Caveat: CARG5 overlaps INS.
-+  |  bl extern lj_meta_arith  // (lua_State *L, TValue *ra,*rb,*rc, BCReg op)
-+  |  // Returns NULL (finished) or TValue * (metamethod).
-+  |  cmpldi CRET1, 0
-+  |  beq ->cont_nop
-+  |
-+  |  // Call metamethod for binary op.
-+  |->vmeta_binop:
-+  |  // BASE = old base, CRET1 = new base, stack = cont/func/o1/o2
-+  |  sub TMP1, CRET1, BASE
-+  |   std PC, -24(CRET1)		// [cont|PC]
-+  |   mr TMP2, BASE
-+  |  addi PC, TMP1, FRAME_CONT
-+  |   mr BASE, CRET1
-+  |  li NARGS8:RC, 16			// 2 args for func(o1, o2).
-+  |  b ->vm_call_dispatch
-+  |
-+  |->vmeta_len:
-+#if LJ_52
-+  |  mr SAVE0, CARG1
-+#endif
-+  |  mr CARG2, RD
-+  |   std BASE, L->base
-+  |  mr CARG1, L
-+  |   std PC, SAVE_PC
-+  |  bl extern lj_meta_len		// (lua_State *L, TValue *o)
-+  |  // Returns NULL (retry) or TValue * (metamethod base).
-+#if LJ_52
-+  |  cmpldi CRET1, 0
-+  |  bne ->vmeta_binop			// Binop call for compatibility.
-+  |  mr CARG1, SAVE0
-+  |  b ->BC_LEN_Z
-+#else
-+  |  b ->vmeta_binop			// Binop call for compatibility.
-+#endif
-+  |
-+  |//-- Call metamethod ----------------------------------------------------
-+  |
-+  |->vmeta_call:			// Resolve and call __call metamethod.
-+  |  // TMP2 = old base, BASE = new base, RC = nargs*8
-+  |  mr CARG1, L
-+  |   std TMP2, L->base			// This is the callers base!
-+  |  subi CARG2, BASE, 16
-+  |   std PC, SAVE_PC
-+  |  add CARG3, BASE, RC
-+  |   mr SAVE0, NARGS8:RC
-+  |  bl extern lj_meta_call	// (lua_State *L, TValue *func, TValue *top)
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)	// Guaranteed to be a function here.
-+  |   addi NARGS8:RC, SAVE0, 8		// Got one more argument now.
-+  |  clear_field RB
-+  |  ins_call
-+  |
-+  |->vmeta_callt:			// Resolve __call for BC_CALLT.
-+  |  // BASE = old base, RA = new base, RC = nargs*8
-+  |  mr CARG1, L
-+  |   std BASE, L->base
-+  |  subi CARG2, RA, 16
-+  |   std PC, SAVE_PC
-+  |  add CARG3, RA, RC
-+  |   mr SAVE0, NARGS8:RC
-+  |  bl extern lj_meta_call	// (lua_State *L, TValue *func, TValue *top)
-+  |  ld TMP1, FRAME_PC(BASE)
-+  |   addi NARGS8:RC, SAVE0, 8		// Got one more argument now.
-+  |   ld LFUNC:RB, FRAME_FUNC(RA)	// Guaranteed to be a function here.
-+  |  clear_field RB
-+  |  b ->BC_CALLT_Z
-+  |
-+  |//-- Argument coercion for 'for' statement ------------------------------
-+  |
-+  |->vmeta_for:
-+  |  mr CARG1, L
-+  |   std BASE, L->base
-+  |  mr CARG2, RA
-+  |   std PC, SAVE_PC
-+  |  mr SAVE0, INS
-+  |  bl extern lj_meta_for	// (lua_State *L, TValue *base)
-+  |.if JIT
-+  |   decode_OP1 TMP0, SAVE0
-+  |.endif
-+  |  decode_RA8 RA, SAVE0
-+  |.if JIT
-+  |   cmpdi TMP0, BC_JFORI
-+  |.endif
-+  |  decode_RD8 RD, SAVE0
-+  |.if JIT
-+  |   beqy =>BC_JFORI
-+  |.endif
-+  |  b =>BC_FORI
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Fast functions -----------------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |.macro .ffunc, name
-+  |->ff_ .. name:
-+  |.endmacro
-+  |
-+  |.macro .ffunc_1, name
-+  |->ff_ .. name:
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |.endmacro
-+  |
-+  |.macro .ffunc_2, name
-+  |->ff_ .. name:
-+  | cmpldi NARGS8:RC, 16
-+  |  ld CARG1, 0(BASE)
-+  |    ld CARG2, 8(BASE)
-+  |  blt ->fff_fallback
-+  |.endmacro
-+  |
-+  |.macro .ffunc_n, name
-+  |->ff_ .. name:
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG3, 0(BASE)
-+  |    lfd FARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  checknum CARG3; bge ->fff_fallback
-+  |.endmacro
-+  |
-+  |.macro .ffunc_nn, name
-+  |->ff_ .. name:
-+  |  cmpldi NARGS8:RC, 16
-+  |   ld CARG3, 0(BASE)
-+  |    lfd FARG1, 0(BASE)
-+  |   ld CARG4, 8(BASE)
-+  |    lfd FARG2, 8(BASE)
-+  |  blt ->fff_fallback
-+  |  checknum CARG3; bge ->fff_fallback
-+  |  checknum CARG4; bge ->fff_fallback
-+  |.endmacro
-+  |
-+  |// Inlined GC threshold check. Caveat: uses TMP0 and TMP1.
-+  |.macro ffgccheck
-+  |  ld TMP0, DISPATCH_GL(gc.total)(DISPATCH)
-+  |  ld TMP1, DISPATCH_GL(gc.threshold)(DISPATCH)
-+  |  cmpld TMP0, TMP1
-+  |  bgel ->fff_gcstep
-+  |.endmacro
-+  |
-+  |//-- Base library: checks -----------------------------------------------
-+  |
-+  |.ffunc_1 assert
-+  |   set_bool TMP1, LJ_TFALSE
-+  |   la RA, -16(BASE)
-+  |  cmpld cr1, CARG1, TMP1
-+  |    ld PC, FRAME_PC(BASE)
-+  |  bge cr1, ->fff_fallback
-+  |   std CARG1, 0(RA)
-+  |  addi RD, NARGS8:RC, 8		// Compute (nresults+1)*8.
-+  |  beq ->fff_res			// Done if exactly 1 argument.
-+  |  li TMP1, 8
-+  |  subi RC, RC, 8
-+  |1:
-+  |  cmpld TMP1, RC
-+  |   ldx TMP0, BASE, TMP1
-+  |   stdx TMP0, RA, TMP1
-+  |    addi TMP1, TMP1, 8
-+  |  bney <1
-+  |  b ->fff_res
-+  |
-+  |.ffunc type
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG1, 0(BASE)
-+  |   get_oper_type CARG1, CARG1	// only type is needed
-+  |  blt ->fff_fallback
-+  |  get_oper_type TMP0, TISNUM		// comparing with type shifted
-+  |  subfc TMP0, TMP0, CARG1
-+  |  subfe TMP2, CARG1, CARG1
-+  |  orc TMP1, TMP2, TMP0
-+  |  addi TMP1, TMP1, ~LJ_TISNUM+1
-+  |  sldi TMP1, TMP1, 3
-+  |   la TMP2, CFUNC:RB->upvalue
-+  |  lfdx FARG1, TMP2, TMP1
-+  |  ldx CARG1, TMP2, TMP1
-+  |  b ->fff_restv
-+  |
-+  |//-- Base library: getters and setters ---------------------------------
-+  |
-+  |.ffunc_1 getmetatable
-+  | get_oper_type CARG3, CARG1		// saving for :6
-+  | checktab CARG1; bne >6
-+  |1:  // Field metatable must be at same offset for GCtab and GCudata!
-+  |  ld TAB:RB, TAB:CARG1->metatable
-+  |2:
-+  |  mr CARG1, TISNIL
-+  |   cmpldi TAB:RB, 0
-+  |  ld STR:RC, DISPATCH_GL(gcroot[GCROOT_MMNAME+MM_metatable])(DISPATCH)
-+  |   beq ->fff_restv
-+  |  lwz TMP0, TAB:RB->hmask
-+  |   li CARG1, LJ_TTAB			// Use metatable as default result.
-+  |  ld TMP1, STR:RC->hash
-+  |   set_oper_type CARG1, CARG1
-+  |  ld NODE:TMP2, TAB:RB->node
-+  |  add CARG1, CARG1, TAB:RB
-+  |  and TMP1, TMP1, TMP0		// idx = str->hash & tab->hmask
-+  |  sldi TMP0, TMP1, 5
-+  |  sldi TMP1, TMP1, 3
-+  |  sub TMP1, TMP0, TMP1
-+  |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
-+  |3:  // Rearranged logic, because we expect _not_ to find the key.
-+  |  ld CARG4, NODE:TMP2->key
-+  |    ld CARG2, NODE:TMP2->val
-+  |  checkstr CARG4; bne >4
-+  |   cmpd CARG4, STR:RC; beq >5
-+  |4:
-+  |  ld NODE:TMP2, NODE:TMP2->next
-+  |  cmpldi NODE:TMP2, 0
-+  |  beq ->fff_restv			// Not found, keep default result.
-+  |  b <3
-+  |5:
-+  |  checknil_noclear CARG2
-+  |  beq ->fff_restv			// Ditto for nil value.
-+  |  mr CARG1, CARG2
-+  |  b ->fff_restv
-+  |
-+  |6:
-+  |  get_oper_type TMP3, TISNUM
-+  |  cmpdi CARG3, LJ_TUDATA; beq <1
-+  |  subfc TMP0, TMP3, CARG3
-+  |  subfe TMP2, CARG3, CARG3
-+  |  orc TMP1, TMP2, TMP0
-+  |  addi TMP1, TMP1, ~LJ_TISNUM+1
-+  |  sldi TMP1, TMP1, 3
-+  |   la TMP2, DISPATCH_GL(gcroot[GCROOT_BASEMT])(DISPATCH)
-+  |  ldx TAB:RB, TMP2, TMP1
-+  |  b <2
-+  |
-+  |.ffunc_2 setmetatable
-+  |  // Fast path: no mt for table yet and not clearing the mt.
-+  |  mr TMP2, CARG1
-+  |  checktab TMP2; bne ->fff_fallback
-+  |  ld TAB:TMP1, TAB:TMP2->metatable
-+  |   checktab CARG2; bne ->fff_fallback
-+  |  cmpldi TAB:TMP1, 0
-+  |   lbz TMP3, TAB:TMP2->marked
-+  |  bne ->fff_fallback
-+  |   andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
-+  |    std TAB:CARG2, TAB:TMP2->metatable
-+  |   beq ->fff_restv
-+  |  barrierback TAB:TMP2, TMP3, TMP0
-+  |  b ->fff_restv
-+  |
-+  |.ffunc rawget
-+  |  cmpldi NARGS8:RC, 16
-+  |   ld CARG2, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  checktab CARG2; bne ->fff_fallback
-+  |   la CARG3, 8(BASE)
-+  |   mr CARG1, L
-+  |  bl extern lj_tab_get  // (lua_State *L, GCtab *t, cTValue *key)
-+  |  // Returns cTValue *.
-+  |  lfd FARG1, 0(CRET1)
-+  |  ld CARG1, 0(CRET1)
-+  |  b ->fff_restv
-+  |
-+  |//-- Base library: conversions ------------------------------------------
-+  |
-+  |.ffunc tonumber
-+  |  // Only handles the number case inline (without a base argument).
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG1, 0(BASE)
-+  |    lfd FARG1, 0(BASE)
-+  |  bne ->fff_fallback			// Exactly one argument.
-+  |   checknum_noclear CARG1; ble ->fff_restv
-+  |  b ->fff_fallback
-+  |
-+  |.ffunc_1 tostring
-+  |  // Only handles the string or number case inline.
-+  |  // save CARG1 type for checknum_sig and fff_restv
-+  |  mr CARG3, CARG1
-+  |  checkstr CARG3
-+  |  // A __tostring method in the string base metatable is ignored.
-+  |  beq ->fff_restv			// String key?
-+  |  // Handle numbers inline, unless a number base metatable is present.
-+  |  ld TMP0, DISPATCH_GL(gcroot[GCROOT_BASEMT_NUM])(DISPATCH)
-+  |   checknum_sig CARG1
-+  |  cmpldi cr1, TMP0, 0
-+  |   std BASE, L->base			// Add frame since C call can throw.
-+  |  crorc 4*cr0+eq, 4*cr0+gt, 4*cr1+eq
-+  |   std PC, SAVE_PC			// Redundant (but a defined value).
-+  |  beq ->fff_fallback
-+  |  ffgccheck
-+  |  mr CARG1, L
-+  |  mr CARG2, BASE
-+  |  bl extern lj_strfmt_number		// (lua_State *L, cTValue *o)
-+  |  // Returns GCstr *.
-+  |  li CARG3, LJ_TSTR
-+  |  set_oper_type CARG3, CARG3
-+  |  add CARG1, CARG1, CARG3
-+  |  b ->fff_restv
-+  |
-+  |//-- Base library: iterators -------------------------------------------
-+  |
-+  |.ffunc next
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |   stdx TISNIL, BASE, NARGS8:RC	// Set missing 2nd arg to nil.
-+  |  checktab CARG1
-+  |   ld PC, FRAME_PC(BASE)
-+  |  bne ->fff_fallback
-+  |  mr CARG2, CARG1
-+  |   std BASE, L->base			// Add frame since C call can throw.
-+  |  mr CARG1, L
-+  |   std BASE, L->top			// Dummy frame length is ok.
-+  |  la CARG3, 8(BASE)
-+  |   std PC, SAVE_PC
-+  |  bl extern lj_tab_next	// (lua_State *L, GCtab *t, TValue *key)
-+  |  // Returns 0 at end of traversal.
-+  |  cmpldi CRET1, 0
-+  |   li CARG3, LJ_TNIL
-+  |  std CARG3, -16(BASE)
-+  |  beq ->fff_res1			// End of traversal: return nil.
-+  |  ld CARG1, 8(BASE)			// Copy key and value to results.
-+  |   la RA, -16(BASE)
-+  |  ld CARG2, 16(BASE)
-+  |  std CARG1, 0(RA)
-+  |   li RD, (2+1)*8
-+  |  std CARG2, 8(RA)
-+  |  b ->fff_res
-+  |
-+  |.ffunc_1 pairs
-+  |  mr TMP1, CARG1
-+  |  checktab CARG1
-+  |   ld PC, FRAME_PC(BASE)
-+  |  bne ->fff_fallback
-+#if LJ_52
-+  |   ld TAB:TMP2, TAB:CARG1->metatable
-+  |  ld TMP0, CFUNC:RB->upvalue[0]
-+  |   cmpldi TAB:TMP2, 0
-+  |  la RA, -16(BASE)
-+  |   bne ->fff_fallback
-+#else
-+  |  ld TMP0, CFUNC:RB->upvalue[0]
-+  |  la RA, -16(BASE)
-+#endif
-+  |   std TMP1, 8(RA)
-+  |   std TISNIL, 16(RA)
-+  |  li RD, (3+1)*8
-+  |  std TMP0, 0(RA)
-+  |  b ->fff_res
-+  |
-+  |.ffunc ipairs_aux
-+  |  cmpldi NARGS8:RC, 16
-+  |   ld CARG4, 8(BASE)
-+  |   ld CARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  checktab CARG1
-+  |  checknum cr1, CARG4
-+  |   ld PC, FRAME_PC(BASE)
-+  |  bne ->fff_fallback
-+  |  bne cr1, ->fff_fallback
-+  |   lwz TMP0, TAB:CARG1->asize
-+  |   ld TMP1, TAB:CARG1->array
-+  |  addi TMP2, CARG4, 1
-+  |   la RA, -16(BASE)
-+  |  cmplw TMP0, TMP2
-+  |   sldi TMP3, TMP2, 3		// TMP2 is array index
-+  |  add TMP2, TMP2, TISNUM		// TMP2 is now lua number
-+  |  std TMP2, 0(RA)
-+  |  ble >2				// Not in array part?
-+  |  ldx TMP2, TMP1, TMP3
-+  |1:
-+  |  checknil_noclear TMP2
-+  |   li RD, (0+1)*8
-+  |  beq ->fff_res			// End of iteration, return 0 results.
-+  |   li RD, (2+1)*8
-+  |  std TMP2, 8(RA)
-+  |  b ->fff_res
-+  |2:  // Check for empty hash part first. Otherwise call C function.
-+  |  lwz TMP0, TAB:CARG1->hmask
-+  |  cmpldi TMP0, 0
-+  |   li RD, (0+1)*8
-+  |  beq ->fff_res
-+  |   mr CARG2, TMP2
-+  |  bl extern lj_tab_getinth		// (GCtab *t, int32_t key)
-+  |  // Returns cTValue * or NULL.
-+  |  cmpldi CRET1, 0
-+  |   li RD, (0+1)*8
-+  |  beq ->fff_res
-+  |  ld TMP2, 0(CRET1)
-+  |  b <1
-+  |
-+  |.ffunc_1 ipairs
-+  |  mr TMP1, CARG1
-+  |  checktab CARG1
-+  |   ld PC, FRAME_PC(BASE)
-+  |  bne ->fff_fallback
-+#if LJ_52
-+  |   ld TAB:TMP2, TAB:CARG1->metatable
-+  |  ld TMP0, CFUNC:RB->upvalue[0]
-+  |   cmpldi TAB:TMP2, 0
-+  |  la RA, -16(BASE)
-+  |   bne ->fff_fallback
-+#else
-+  |  ld TMP0, CFUNC:RB->upvalue[0]
-+  |  la RA, -16(BASE)
-+#endif
-+  |  std TMP1, -8(BASE)
-+  |  std TISNUM, 0(BASE)
-+  |  li RD, (3+1)*8
-+  |  std TMP0, -16(BASE)
-+  |  b ->fff_res
-+  |
-+  |//-- Base library: catch errors ----------------------------------------
-+  |
-+  |.ffunc pcall
-+  |  cmpldi NARGS8:RC, 8
-+  |   lbz TMP3, DISPATCH_GL(hookmask)(DISPATCH)
-+  |  blt ->fff_fallback
-+  |   mr TMP2, BASE
-+  |   la BASE, 16(BASE)
-+  |  // Remember active hook before pcall.
-+  |  rlwinm TMP3, TMP3, 32-HOOK_ACTIVE_SHIFT, 31, 31
-+  |   subi NARGS8:RC, NARGS8:RC, 8
-+  |  addi PC, TMP3, 16+FRAME_PCALL
-+  |   cmpdi NARGS8:RC, 0
-+  |  beq ->vm_call_dispatch
-+  |1:
-+  |   add TMP1, BASE, NARGS8:RC
-+  |2:
-+  |   ld TMP0, -16(TMP1)
-+  |   stdu TMP0, -8(TMP1)
-+  |   cmpld TMP1, BASE
-+  |   bne <2
-+  |  b ->vm_call_dispatch
-+  |
-+  |.ffunc xpcall
-+  |  subic. NARGS8:RC, NARGS8:RC, 16
-+  |    ld CARG2, 8(BASE)
-+  |    ld CARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  lbz TMP1, DISPATCH_GL(hookmask)(DISPATCH)
-+  |   mr TMP2, BASE
-+  |  // Traceback must be a function.
-+  |  checkfunc_noclear CARG2; bne ->fff_fallback
-+  |   la BASE, 24(BASE)
-+  |  // Remember active hook before pcall.
-+  |  rlwinm TMP1, TMP1, 32-HOOK_ACTIVE_SHIFT, 31, 31
-+  |    std CARG2, 0(TMP2)		// Swap function and traceback.
-+  |  cmpdi RC, 0
-+  |    std CARG1, 8(TMP2)
-+  |   addi PC, TMP1, 24+FRAME_PCALL
-+  |  beq ->vm_call_dispatch
-+  |  b <1
-+  |
-+  |//-- Coroutine library --------------------------------------------------
-+  |
-+  |.macro coroutine_resume_wrap, resume
-+  |.if resume
-+  |.ffunc_1 coroutine_resume
-+  |  checkthread CARG1; bne ->fff_fallback
-+  |.else
-+  |.ffunc coroutine_wrap_aux
-+  |  ld L:CARG1, CFUNC:RB->upvalue[0].gcr
-+  |  clear_field L:CARG1
-+  |.endif
-+  |  lbz TMP0, L:CARG1->status
-+  |   ld TMP1, L:CARG1->cframe
-+  |     la TMP3, L:CARG1->base
-+  |     ld TMP2, 0(TMP3)
-+  |     ld CARG2, 8(TMP3)
-+  |  cmpldi cr0, TMP0, LUA_YIELD
-+  |     add TMP3, CARG2, TMP0
-+  |   cmpldi cr1, TMP1, 0
-+  |     cmpd cr7, TMP3, TMP2
-+  |   ld TMP0, L:CARG1->maxstack
-+  |     beq cr7, ->fff_fallback
-+  |     cmpld cr7, CARG2, TMP2
-+  |   ld PC, FRAME_PC(BASE)
-+  |  bge cr0, >9
-+  |   addi CARG2, CARG2, 8
-+  |9:
-+  |  crorc 4*cr6+lt, 4*cr0+gt, 4*cr1+eq		// st>LUA_YIELD || cframe!=0
-+  |   add TMP2, CARG2, NARGS8:RC
-+  |  crandc 4*cr6+gt, 4*cr7+eq, 4*cr0+eq	// base==top && st!=LUA_YIELD
-+  |   cmpld cr1, TMP2, TMP0
-+  |  cror 4*cr6+lt, 4*cr6+lt, 4*cr6+gt
-+  |   std PC, SAVE_PC
-+  |  cror 4*cr6+lt, 4*cr6+lt, 4*cr1+gt		// cond1 || cond2 || stackov
-+  |   std BASE, L->base
-+  |  blt cr6, ->fff_fallback
-+  |1:
-+  |.if resume
-+  |  addi BASE, BASE, 8			// Keep resumed thread in stack for GC.
-+  |  subi NARGS8:RC, NARGS8:RC, 8
-+  |  subi TMP2, TMP2, 8
-+  |.endif
-+  |  std TMP2, L:CARG1->top
-+  |  li TMP1, 0
-+  |  std BASE, L->top
-+  |2:  // Move args to coroutine.
-+  |  cmpd TMP1, NARGS8:RC
-+  |   lfdx f0, BASE, TMP1
-+  |  beq >3
-+  |   stfdx f0, CARG2, TMP1
-+  |  addi TMP1, TMP1, 8
-+  |  b <2
-+  |3:
-+  |  li CARG3, 0
-+  |   mr L:SAVE0, L:CARG1
-+  |  li CARG4, 0
-+  |  bl ->vm_resume			// (lua_State *L, TValue *base, 0, 0)
-+  |  // Returns thread status.
-+  |4:
-+  |  ld TMP2, L:SAVE0->base
-+  |   cmpldi CRET1, LUA_YIELD
-+  |  ld TMP3, L:SAVE0->top
-+  |    li_vmstate INTERP
-+  |  ld BASE, L->base
-+  |    std L, DISPATCH_GL(cur_L)(DISPATCH)
-+  |    st_vmstate
-+  |   bgt >8
-+  |  sub RD, TMP3, TMP2
-+  |   ld TMP0, L->maxstack
-+  |  cmpldi RD, 0
-+  |   add TMP1, BASE, RD
-+  |  beq >6				// No results?
-+  |  cmpld TMP1, TMP0
-+  |   li TMP1, 0
-+  |  bgt >9				// Need to grow stack?
-+  |
-+  |  subi TMP3, RD, 8
-+  |   std TMP2, L:SAVE0->top		// Clear coroutine stack.
-+  |5:  // Move results from coroutine.
-+  |  cmpld TMP1, TMP3
-+  |   lfdx f0, TMP2, TMP1
-+  |   stfdx f0, BASE, TMP1
-+  |    addi TMP1, TMP1, 8
-+  |  bne <5
-+  |6:
-+  |  andi. TMP0, PC, FRAME_TYPE
-+  |.if resume
-+  |  set_bool TMP1, LJ_TTRUE
-+  |   la RA, -8(BASE)
-+  |  std TMP1, -8(BASE)			// Prepend true to results.
-+  |  addi RD, RD, 16
-+  |.else
-+  |  mr RA, BASE
-+  |  addi RD, RD, 8
-+  |.endif
-+  |7:
-+  |    std PC, SAVE_PC
-+  |   mr MULTRES, RD
-+  |  beq ->BC_RET_Z
-+  |  b ->vm_return
-+  |
-+  |8:  // Coroutine returned with error (at co->top-1).
-+  |.if resume
-+  |  andi. TMP0, PC, FRAME_TYPE
-+  |  la TMP3, -8(TMP3)
-+  |   set_bool TMP1, LJ_TFALSE
-+  |  lfd f0, 0(TMP3)
-+  |   std TMP3, L:SAVE0->top		// Remove error from coroutine stack.
-+  |    li RD, (2+1)*8
-+  |   std TMP1, -16(BASE)		// Prepend false to results.
-+  |    la RA, -16(BASE)
-+  |  stfd f0, -8(BASE)			// Copy error message.
-+  |  b <7
-+  |.else
-+  |  mr CARG1, L
-+  |  mr CARG2, L:SAVE0
-+  |  bl extern lj_ffh_coroutine_wrap_err  // (lua_State *L, lua_State *co)
-+  |.endif
-+  |
-+  |9:  // Handle stack expansion on return from yield.
-+  |  mr CARG1, L
-+  |  srdi CARG2, RD, 3
-+  |  bl extern lj_state_growstack	// (lua_State *L, int n)
-+  |  li CRET1, 0
-+  |  b <4
-+  |.endmacro
-+  |
-+  |  coroutine_resume_wrap 1		// coroutine.resume
-+  |  coroutine_resume_wrap 0		// coroutine.wrap
-+  |
-+  |.ffunc coroutine_yield
-+  |  ld TMP0, L->cframe
-+  |   add TMP1, BASE, NARGS8:RC
-+  |   std BASE, L->base
-+  |  andi. TMP0, TMP0, CFRAME_RESUME
-+  |   std TMP1, L->top
-+  |    li CRET1, LUA_YIELD
-+  |  beq ->fff_fallback
-+  |   std ZERO, L->cframe
-+  |    stb CRET1, L->status
-+  |  b ->vm_leave_unw
-+  |
-+  |//-- Math library -------------------------------------------------------
-+  |
-+  |.ffunc_1 math_abs
-+  |  checknum_noclear CARG1
-+  |  bne >2
-+  |  get_value CARG1
-+  |  sradi TMP1, CARG1, 31
-+  |  xor TMP2, TMP1, CARG1
-+  |  lus TMP0, 0x8000
-+  |  sub CARG1, TMP2, TMP1
-+  |  get_value CARG1
-+  |  cmpld CARG1, TMP0
-+  |  beq >1
-+  |->fff_resi:
-+  |  ld PC, FRAME_PC(BASE)
-+  |  la RA, -16(BASE)
-+  |  add_oper_type CRET1, TISNUM
-+  |  std CRET1, -16(BASE)
-+  |  b ->fff_res1
-+  |1:
-+  |  lus CARG1, 0x41e0	// 2^31.
-+  |  sldi CARG1, CARG1, 32
-+  |  b ->fff_restv
-+  |2:
-+  |  bge ->fff_fallback
-+  |  clrldi CARG1,CARG1,1
-+  |  // Fallthrough.
-+  |
-+  |->fff_restv:
-+  |  // CARG1 = TValue result.
-+  |  ld PC, FRAME_PC(BASE)
-+  |   std CARG1, -16(BASE)
-+  |  la RA, -16(BASE)
-+  |->fff_res1:
-+  |  // RA = results, PC = return.
-+  |  li RD, (1+1)*8
-+  |->fff_res:
-+  |  // RA = results, RD = (nresults+1)*8, PC = return.
-+  |  andi. TMP0, PC, FRAME_TYPE
-+  |   mr MULTRES, RD
-+  |  subi RA, BASE, 16
-+  |  bney ->vm_return
-+  |  lwz INS, -4(PC)
-+  |  decode_RB8 RB, INS
-+  |5:
-+  |  cmpld RB, RD			// More results expected?
-+  |   decode_RA8 TMP0, INS
-+  |  bgt >6
-+  |  ins_next1
-+  |  // Adjust BASE. KBASE is assumed to be set for the calling frame.
-+  |   sub BASE, RA, TMP0
-+  |  ins_next2
-+  |
-+  |6:  // Fill up results with nil.
-+  |  subi TMP1, RD, 8
-+  |   addi RD, RD, 8
-+  |  stdx TISNIL, RA, TMP1
-+  |  b <5
-+  |
-+  |.macro math_extern, func
-+  |  .ffunc_n math_ .. func
-+  |  blex func
-+  |  b ->fff_resn
-+  |.endmacro
-+  |
-+  |.macro math_extern2, func
-+  |  .ffunc_nn math_ .. func
-+  |  blex func
-+  |  b ->fff_resn
-+  |.endmacro
-+  |
-+  |.macro math_round, func, round
-+  |  .ffunc math_ .. func
-+  |  ld CARG1, 0(BASE)
-+  |  cmplwi NARGS8:RC, 8
-+  |  lfd FARG1, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  checknum CARG1; bgt ->fff_fallback
-+  |  srdi TMP0, CARG1, 32
-+  |  sldi TMP1, TISNUM, 15
-+  |  cmpld TMP0, TMP1
-+  |  beq ->fff_restv
-+  |  round FARG1, FARG1
-+  |  b ->fff_resn
-+  |.endmacro
-+  |
-+  |  math_round floor, frim
-+  |  math_round ceil, frip
-+  |
-+  |.if SQRT
-+  |.ffunc_n math_sqrt
-+  |  fsqrt FARG1, FARG1
-+  |  b ->fff_resn
-+  |.else
-+  |  math_extern sqrt
-+  |.endif
-+  |
-+  |.ffunc math_log
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG3, 0(BASE)
-+  |    lfd FARG1, 0(BASE)
-+  |  bne ->fff_fallback			// Need exactly 1 argument.
-+  |  checknum CARG3; bge ->fff_fallback
-+  |  blex log
-+  |  b ->fff_resn
-+  |
-+  |  math_extern log10
-+  |  math_extern exp
-+  |  math_extern sin
-+  |  math_extern cos
-+  |  math_extern tan
-+  |  math_extern asin
-+  |  math_extern acos
-+  |  math_extern atan
-+  |  math_extern sinh
-+  |  math_extern cosh
-+  |  math_extern tanh
-+  |  math_extern2 pow
-+  |  math_extern2 atan2
-+  |  math_extern2 fmod
-+  |
-+  |.ffunc math_ldexp
-+  |  cmpldi NARGS8:RC, 16
-+  |   ld CARG1, 0(BASE)
-+  |    ld CARG2, 8(BASE)
-+  |  blt ->fff_fallback
-+  |  checknum_noclear CARG1; bge ->fff_fallback
-+  |  checknum_noclear CARG2; bne ->fff_fallback
-+  |  std CARG1, 0(BASE)
-+  |  lfd FARG1, 0(BASE)
-+  |  std TOCREG, SAVE_TOC
-+  |  blex ldexp
-+  |  b ->fff_resn
-+  |
-+  |.ffunc_n math_frexp
-+  |  la CARG2, DISPATCH_GL(tmptv)(DISPATCH)
-+  |   ld PC, FRAME_PC(BASE)
-+  |  blex frexp
-+  |   lwz TMP1, DISPATCH_GL(tmptv)(DISPATCH)
-+  |   la RA, -16(BASE)
-+  |  stfd FARG1, 0(RA)
-+  |  li RD, (2+1)*8
-+  |   add TMP1, TMP1, TISNUM
-+  |   std TMP1, 8(RA)
-+  |  b ->fff_res
-+  |
-+  |.ffunc_n math_modf
-+  |  la CARG2, -16(BASE)
-+  |   ld PC, FRAME_PC(BASE)
-+  |  blex modf
-+  |   la RA, -16(BASE)
-+  |  stfd FARG1, -8(BASE)
-+  |  li RD, (2+1)*8
-+  |  b ->fff_res
-+  |
-+  |.macro math_minmax, name, ismax
-+  |  .ffunc_1 name
-+  |  checknum_noclear CARG1
-+  |   addi TMP1, BASE, 8
-+  |   add TMP2, BASE, NARGS8:RC
-+  |  bne >4
-+  |1:  // Handle integers.
-+  |  ld CARG2, 0(TMP1)
-+  |   cmpld cr1, TMP1, TMP2
-+  |   bge cr1, ->fff_restv
-+  |  checknum_noclear CARG2
-+  |   xoris TMP0, CARG1, 0x8000
-+  |   clrldi TMP0, TMP0, 32
-+  |   xoris TMP3, CARG2, 0x8000
-+  |   clrldi TMP3, TMP3, 32
-+  |  bne >3
-+  |  subfc TMP3, TMP3, TMP0
-+  |  subfe TMP0, TMP0, TMP0
-+  |.if ismax
-+  |  andc TMP3, TMP3, TMP0
-+  |.else
-+  |  and TMP3, TMP3, TMP0
-+  |.endif
-+  |  add CARG1, TMP3, CARG2
-+  |   add_oper_type CARG1, TISNUM
-+  |   addi TMP1, TMP1, 8
-+  |  b <1
-+  |3:
-+  |  bge ->fff_fallback
-+  |  // Convert intermediate result to number and continue below.
-+  |  tonum_i FARG1, CARG1
-+  |  lfd FARG2, 0(TMP1)
-+  |  b >6
-+  |4:
-+  |   lfd FARG1, 0(BASE)
-+  |  bge ->fff_fallback
-+  |5:  // Handle numbers.
-+  |  ld CARG2, 0(TMP1)
-+  |   cmpld cr1, TMP1, TMP2
-+  |  lfd FARG2, 0(TMP1)
-+  |   bge cr1, ->fff_resn
-+  |  checknum_noclear CARG2; bge >7
-+  |6:
-+  |  fsub f0, FARG1, FARG2
-+  |   addi TMP1, TMP1, 8
-+  |.if ismax
-+  |  fsel FARG1, f0, FARG1, FARG2
-+  |.else
-+  |  fsel FARG1, f0, FARG2, FARG1
-+  |.endif
-+  |  b <5
-+  |7:  // Convert integer to number and continue above.
-+  |   ld CARG2, 0(TMP1)
-+  |  bne ->fff_fallback
-+  |  tonum_i FARG2, CARG2
-+  |  b <6
-+  |.endmacro
-+  |
-+  |  math_minmax math_min, 0
-+  |  math_minmax math_max, 1
-+  |
-+  |//-- String library -----------------------------------------------------
-+  |
-+  |.ffunc string_byte			// Only handle the 1-arg case here.
-+  |  cmpldi NARGS8:RC, 8
-+  |    ld STR:CARG1, 0(BASE)
-+  |  bne ->fff_fallback			// Need exactly 1 argument.
-+  |   checkstr CARG1
-+  |   bne ->fff_fallback
-+  |  ld TMP0, STR:CARG1->len
-+  |   lbz CARG1, STR:CARG1[1]		// Access is always ok (NUL at end).
-+  |   li RD, (0+1)*8
-+  |   ld PC, FRAME_PC(BASE)
-+  |  cmpldi TMP0, 0
-+  |   la RA, -16(BASE)
-+  |  beqy ->fff_res
-+  |  b ->fff_resi
-+  |
-+  |.ffunc string_char			// Only handle the 1-arg case here.
-+  |  ffgccheck
-+  |  cmpldi NARGS8:RC, 8
-+  |   ld CARG3, 0(BASE)
-+  |  bne ->fff_fallback			// Exactly 1 argument.
-+  |  checknum CARG3; bne ->fff_fallback
-+  |   la CARG2, 0(BASE)			// Points to stack. Little-endian.
-+
-+  |    cmpldi CARG3, 255
-+  |   li CARG3, 1
-+  |    bgt ->fff_fallback
-+  |->fff_newstr:
-+  |  mr CARG1, L
-+  |  std BASE, L->base
-+  |  std PC, SAVE_PC
-+  |  bl extern lj_str_new		// (lua_State *L, char *str, size_t l)
-+  |->fff_resstr:
-+  |  // Returns GCstr *.
-+  |  ld BASE, L->base
-+  |  li TMP1, LJ_TSTR
-+  |  set_oper_type, TMP1, TMP1
-+  |  add CARG1, CARG1, TMP1
-+  |  b ->fff_restv
-+  |
-+  |.ffunc string_sub
-+  |  ffgccheck
-+  |  cmpldi NARGS8:RC, 16
-+  |   ld STR:CARG1, 0(BASE)
-+  |   blt ->fff_fallback
-+  |   ld TMP1, 8(BASE)
-+  |    get_oper_type TMP0,CARG1
-+  |    get_oper_type CARG2,TMP1
-+  |   li TMP2,-1
-+  |  beq >1
-+  |   ld TMP2, 16(BASE)
-+  |   get_oper_type CARG3,TMP2
-+  |   checknum TMP2
-+  |  bne ->fff_fallback
-+  |1:
-+  |  checknum TMP1; bne ->fff_fallback
-+  |  checkstr CARG1; bne ->fff_fallback
-+  |   lwz TMP0, STR:CARG1->len
-+  |  cmpld TMP0, TMP2			// len < end? (unsigned compare)
-+  |   addi TMP3, TMP2, 1
-+  |  blt >5
-+  |2:
-+  |  cmpdi TMP1, 0			// start <= 0?
-+  |   add TMP3, TMP1, TMP0
-+  |  ble >7
-+  |3:
-+  |  sub CARG3, TMP2, TMP1
-+  |    addi CARG2, STR:CARG1, #STR-1
-+  |  sradi TMP0, CARG3, 31
-+  |   addi CARG3, CARG3, 1
-+  |    add CARG2, CARG2, TMP1
-+  |  andc CARG3, CARG3, TMP0
-+  |  b ->fff_newstr
-+  |
-+  |5:  // Negative end or overflow.
-+  |  cmpd TMP0, TMP2			// len >= end? (signed compare)
-+  |   add TMP2, TMP0, TMP3		// Negative end: end = end+len+1.
-+  |  bge <2
-+  |   mr TMP2, TMP0			// Overflow: end = len.
-+  |  b <2
-+  |
-+  |7:  // Negative start or underflow.
-+  |  extsw TMP1, TMP1
-+  |  addic CARG3, TMP1, -1
-+  |  subfe CARG3, CARG3, CARG3
-+  |   sradi CARG2, TMP3, 31		// Note: modifies carry.
-+  |  andc TMP3, TMP3, CARG3
-+  |   andc TMP1, TMP3, CARG2
-+  |  addi TMP1, TMP1, 1			// start = 1 + (start ? start+len : 0)
-+  |  b <3
-+  |
-+  |.macro ffstring_op, name
-+  |  .ffunc string_ .. name
-+  |  ffgccheck
-+  |  cmpldi NARGS8:RC, 8
-+  |    ld STR:CARG2, 0(BASE)
-+  |  blt ->fff_fallback
-+  |  checkstr CARG2
-+  |   la SBUF:CARG1, DISPATCH_GL(tmpbuf)(DISPATCH)
-+  |  bne ->fff_fallback
-+  |   ld TMP0, SBUF:CARG1->b
-+  |  std L, SBUF:CARG1->L
-+  |  std BASE, L->base
-+  |  std PC, SAVE_PC
-+  |   std TMP0, SBUF:CARG1->p
-+  |  bl extern lj_buf_putstr_ .. name
-+  |  bl extern lj_buf_tostr
-+  |  b ->fff_resstr
-+  |.endmacro
-+  |
-+  |ffstring_op reverse
-+  |ffstring_op lower
-+  |ffstring_op upper
-+  |
-+  |//-- Bit library --------------------------------------------------------
-+  |
-+  |.macro .ffunc_bit, name
-+  |  .ffunc_1 bit_..name
-+  |  checknum CARG1; bnel ->fff_tobit_fb
-+  |.endmacro
-+  |
-+  |.macro .ffunc_bit_op, name, ins
-+  |  .ffunc_bit name
-+  |   mr TMP0, CARG1
-+  |   rldicl TMP0, TMP0, 0, 32
-+  |  addi TMP1, BASE, 8
-+  |  add TMP2, BASE, NARGS8:RC
-+  |1:
-+  |  ld CARG2, 0(TMP1)
-+  |   cmpld cr1, TMP1, TMP2
-+  |   bgey cr1, >9
-+  |  checknum CARG2
-+  |  bnel ->fff_bitop_fb
-+  |  rldicl CARG2, CARG2, 0, 32
-+  |  ins TMP0, TMP0, CARG2
-+  |   addi TMP1, TMP1, 8
-+  |  b <1
-+  |9:
-+  |  add CARG1, TMP0, TISNUM
-+  |  b ->fff_restv
-+  |
-+  |.endmacro
-+  |
-+  |.ffunc_bit_op band, and
-+  |.ffunc_bit_op bor, or
-+  |.ffunc_bit_op bxor, xor
-+  |
-+  |.ffunc_bit bswap
-+  |  rlwinm TMP0, CARG1, 8, 0, 31
-+  |  rlwimi TMP0, CARG1, 24, 0, 7
-+  |  rlwimi TMP0, CARG1, 24, 16, 23
-+  |  mr CRET1, TMP0
-+  |  b ->fff_resi
-+  |
-+  |.ffunc_bit bnot
-+  |  not CRET1, CARG1
-+  |  rldicl CRET1, CRET1, 0, 32
-+  |  add CRET1, CRET1, TISNUM
-+  |  b ->fff_resi
-+  |
-+  |.macro .ffunc_bit_sh, name, ins, shmod
-+  |  .ffunc_2 bit_..name
-+  |  checknum CARG1; bnel ->fff_tobit_fb
-+  |  // Note: no inline conversion from number for 2nd argument!
-+  |  checknum CARG2; bne ->fff_fallback
-+  |  rldicl TMP0, CARG1, 0, 32
-+  |  rldicl TMP1, CARG2, 0, 32
-+  |.if shmod == 1
-+  |  rlwinm TMP1, TMP1, 0, 27, 31
-+  |.elif shmod == 2
-+  |  neg TMP1, TMP1
-+  |.endif
-+  |  ins CARG1, TMP0, TMP1
-+  |  rldicl CARG1, CARG1, 0, 32
-+  |  add CARG1, CARG1, TISNUM
-+  |  b ->fff_restv
-+  |.endmacro
-+  |
-+  |.ffunc_bit_sh lshift, slw, 1
-+  |.ffunc_bit_sh rshift, srw, 1
-+  |.ffunc_bit_sh arshift, sraw, 1
-+  |.ffunc_bit_sh rol, rotlw, 0
-+  |.ffunc_bit_sh ror, rotlw, 2
-+  |
-+  |.ffunc_bit tobit
-+  |  rldicl TMP0, CARG1, 0, 32
-+  |  add CARG1, TMP0, TISNUM
-+  |  b ->fff_restv
-+  |
-+  |->fff_resn:
-+  |  ld PC, FRAME_PC(BASE)
-+  |  la RA, -16(BASE)
-+  |  stfd FARG1, -16(BASE)
-+  |  b ->fff_res1
-+  |
-+  |// Fallback FP number to bit conversion.
-+  |->fff_tobit_fb:
-+  |  lfd FARG1, 0(BASE)
-+  |  bgt ->fff_fallback
-+  |  fadd FARG1, FARG1, TOBIT
-+  |  stfd FARG1, TMPD
-+  |  ld CARG1, TMPD
-+  |  blr
-+  |->fff_bitop_fb:
-+  |  lfd FARG1, 0(TMP1)
-+  |  bgt ->fff_fallback
-+  |  fadd FARG1, FARG1, TOBIT
-+  |  stfd FARG1, TMPD
-+  |  ld CARG2, TMPD
-+  |  blr
-+  |
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->fff_fallback:			// Call fast function fallback handler.
-+  |  // BASE = new base, RB = CFUNC, RC = nargs*8
-+  |  ld TMP3, CFUNC:RB->f
-+  |    add TMP1, BASE, NARGS8:RC
-+  |   ld PC, FRAME_PC(BASE)		// Fallback may overwrite PC.
-+  |    addi TMP0, TMP1, 8*LUA_MINSTACK
-+  |     ld TMP2, L->maxstack
-+  |   std PC, SAVE_PC			// Redundant (but a defined value).
-+  |  cmpld TMP0, TMP2
-+  |     std BASE, L->base
-+  |    std TMP1, L->top
-+  |   mr CARG1, L
-+  |  bgt >5				// Need to grow stack.
-+  |  mr r12, TMP3			// keep r12 for function linkage.
-+  |  mtctr r12
-+  |  bctrl				// (lua_State *L)
-+  |  // Either throws an error, or recovers and returns -1, 0 or nresults+1.
-+  |  ld BASE, L->base
-+  |  cmpdi CRET1, 0
-+  |   sldi RD, CRET1, 3
-+  |   la RA, -16(BASE)
-+  |  bgt ->fff_res			// Returned nresults+1?
-+  |1:  // Returned 0 or -1: retry fast path.
-+  |  ld TMP0, L->top
-+  |   ld LFUNC:RB, FRAME_FUNC(BASE)
-+  |  sub NARGS8:RC, TMP0, BASE
-+  |  bne ->vm_call_tail			// Returned -1?
-+  |  clear_field RB
-+  |  ins_callt				// Returned 0: retry fast path.
-+  |
-+  |// Reconstruct previous base for vmeta_call during tailcall.
-+  |->vm_call_tail:
-+  |  andi. TMP0, PC, FRAME_TYPE
-+  |   rlwinm TMP1, PC, 0, 0, 28
-+  |  bne >3
-+  |  lwz INS, -4(PC)
-+  |  decode_RA8 TMP1, INS
-+  |  addi TMP1, TMP1, 8
-+  |3:
-+  |  sub TMP2, BASE, TMP1
-+  |  b ->vm_call_dispatch		// Resolve again for tailcall.
-+  |
-+  |5:  // Grow stack for fallback handler.
-+  |  li CARG2, LUA_MINSTACK
-+  |  bl extern lj_state_growstack	// (lua_State *L, int n)
-+  |  ld BASE, L->base
-+  |  cmpd TMP0, TMP0			// Set 4*cr0+eq to force retry.
-+  |  b <1
-+  |
-+  |->fff_gcstep:			// Call GC step function.
-+  |  // BASE = new base, RC = nargs*8
-+  |  mflr SAVE0
-+  |   std BASE, L->base
-+  |  add TMP0, BASE, NARGS8:RC
-+  |   std PC, SAVE_PC			// Redundant (but a defined value).
-+  |  std TMP0, L->top
-+  |  mr CARG1, L
-+  |  bl extern lj_gc_step		// (lua_State *L)
-+  |   ld BASE, L->base
-+  |  mtlr SAVE0
-+  |    ld TMP0, L->top
-+  |   sub NARGS8:RC, TMP0, BASE
-+  |   ld CFUNC:RB, FRAME_FUNC(BASE)
-+  |   clear_field CFUNC:RB
-+  |  blr
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Special dispatch targets -------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vm_record:				// Dispatch target for recording phase.
-+  |  NYI
-+  |
-+  |->vm_rethook:			// Dispatch target for return hooks.
-+  |  lbz TMP3, DISPATCH_GL(hookmask)(DISPATCH)
-+  |  andi. TMP0, TMP3, HOOK_ACTIVE	// Hook already active?
-+  |  beq >1
-+  |5:  // Re-dispatch to static ins.
-+  |  addi TMP1, TMP1, GG_DISP2STATIC	// Assumes decode_OP8 TMP1, INS.
-+  |  ldx TMP0, DISPATCH, TMP1
-+  |  mtctr TMP0
-+  |  bctr
-+  |
-+  |->vm_inshook:			// Dispatch target for instr/line hooks.
-+  |  lbz TMP3, DISPATCH_GL(hookmask)(DISPATCH)
-+  |  lwz TMP2, DISPATCH_GL(hookcount)(DISPATCH)
-+  |  andi. TMP0, TMP3, HOOK_ACTIVE	// Hook already active?
-+  |   rlwinm TMP0, TMP3, 31-LUA_HOOKLINE, 31, 0
-+  |  bne <5
-+  |
-+  |   cmpdi cr1, TMP0, 0
-+  |  addic. TMP2, TMP2, -1
-+  |   beq cr1, <5
-+  |  stw TMP2, DISPATCH_GL(hookcount)(DISPATCH)
-+  |  beq >1
-+  |   bge cr1, <5
-+  |1:
-+  |  mr CARG1, L
-+  |   std MULTRES, SAVE_MULTRES
-+  |  mr CARG2, PC
-+  |   std BASE, L->base
-+  |  // SAVE_PC must hold the _previous_ PC. The callee updates it with PC.
-+  |  bl extern lj_dispatch_ins		// (lua_State *L, const BCIns *pc)
-+  |3:
-+  |  ld BASE, L->base
-+  |4:  // Re-dispatch to static ins.
-+  |  lwz INS, -4(PC)
-+  |  decode_OP8 TMP1, INS
-+  |   decode_RB8 RB, INS
-+  |  addi TMP1, TMP1, GG_DISP2STATIC
-+  |   decode_RD8 RD, INS
-+  |  ldx TMP0, DISPATCH, TMP1
-+  |   decode_RA8 RA, INS
-+  |   decode_RC8 RC, INS
-+  |  mtctr TMP0
-+  |  bctr
-+  |
-+  |->cont_hook:				// Continue from hook yield.
-+  |  addi PC, PC, 4
-+  |  ld MULTRES, -40(RB)		// Restore MULTRES for *M ins.
-+  |  b <4
-+  |
-+  |->vm_hotloop:			// Hot loop counter underflow.
-+  |  NYI
-+  |
-+  |->vm_callhook:			// Dispatch target for call hooks.
-+  |  mr CARG2, PC
-+  |.if JIT
-+  |  b >1
-+  |.endif
-+  |
-+  |->vm_hotcall:			// Hot call counter underflow.
-+  |.if JIT
-+  |  ori CARG2, PC, 1
-+  |1:
-+  |.endif
-+  |  add TMP0, BASE, RC
-+  |   std PC, SAVE_PC
-+  |  mr CARG1, L
-+  |   std BASE, L->base
-+  |  sub RA, RA, BASE
-+  |   std TMP0, L->top
-+  |  bl extern lj_dispatch_call		// (lua_State *L, const BCIns *pc)
-+  |  // Returns ASMFunction.
-+  |  ld BASE, L->base
-+  |   ld TMP0, L->top
-+  |   std ZERO, SAVE_PC			// Invalidate for subsequent line hook.
-+  |  sub NARGS8:RC, TMP0, BASE
-+  |  add RA, BASE, RA
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+  |  clear_field RB
-+  |  lwz INS, -4(PC)
-+  |  mtctr CRET1
-+  |  bctr
-+  |
-+  |->cont_stitch:			// Trace stitching.
-+  |  NYI
-+  |
-+  |->vm_profhook:			// Dispatch target for profiler hook.
-+#if LJ_HASPROFILE
-+  |  mr CARG1, L
-+  |   std MULTRES, SAVE_MULTRES
-+  |  mr CARG2, PC
-+  |   std BASE, L->base
-+  |  bl extern lj_dispatch_profile	// (lua_State *L, const BCIns *pc)
-+  |  // HOOK_PROFILE is off again, so re-dispatch to dynamic instruction.
-+  |  ld BASE, L->base
-+  |  subi PC, PC, 4
-+  |  b ->cont_nop
-+#endif
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Trace exit handler -------------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |->vm_exit_handler:
-+  |  NYI
-+  |->vm_exit_interp:
-+  |  NYI
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Math helper functions ----------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |// NYI: Use internal implementations of floor, ceil, trunc.
-+  |
-+  |->vm_modi:
-+  |  divdo. TMP0, CARG1, CARG2
-+  |  bso >1
-+  |   xor CARG3, CARG1, CARG2
-+  |   cmpdi CARG3, 0
-+  |  mulld TMP0, TMP0, CARG2
-+  |  sub CARG1, CARG1, TMP0
-+  |   bgelr
-+  |  cmpdi CARG1, 0; beqlr
-+  |  add CARG1, CARG1, CARG2
-+  |  blr
-+  |1:
-+  |  cmpdi CARG2, 0
-+  |   li CARG1, 0
-+  |  beqlr
-+  |  mcrxr cr0			// Clear SO for -2147483648 % -1 and return 0.
-+  |  blr
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- Miscellaneous functions --------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |// void lj_vm_cachesync(void *start, void *end)
-+  |// Flush D-Cache and invalidate I-Cache. Assumes 32 byte cache line size.
-+  |// This is a good lower bound, except for very ancient PPC models.
-+  |->vm_cachesync:
-+  |.if JIT or FFI
-+  |  // Compute start of first cache line and number of cache lines.
-+  |  clrldi CARG1, CARG1, 17
-+  |  sub CARG2, CARG2, CARG1
-+  |  addi CARG2, CARG2, 31
-+  |  rlwinm. CARG2, CARG2, 27, 5, 31
-+  |  beqlr
-+  |  mtctr CARG2
-+  |  mr CARG3, CARG1
-+  |1:  // Flush D-Cache.
-+  |  dcbst r0, CARG1
-+  |  addi CARG1, CARG1, 32
-+  |  bdnz <1
-+  |  sync
-+  |  mtctr CARG2
-+  |1:  // Invalidate I-Cache.
-+  |  icbi r0, CARG3
-+  |  addi CARG3, CARG3, 32
-+  |  bdnz <1
-+  |  isync
-+  |  blr
-+  |.endif
-+  |
-+  |//-----------------------------------------------------------------------
-+  |//-- FFI helper functions -----------------------------------------------
-+  |//-----------------------------------------------------------------------
-+  |
-+  |// Handler for callback functions. Callback slot number in r11, g in r12.
-+  |->vm_ffi_callback:
-+  |.if FFI
-+  |.type CTSTATE, CTState, PC
-+  |  pic_code_setup vm_ffi_callback
-+  |  saveregs
-+  |  ld CTSTATE, GL:r12->ctype_state
-+  |   addi DISPATCH, r12, GG_G2DISP
-+  |  std r11, CTSTATE->cb.slot
-+  |  std r3, CTSTATE->cb.gpr[0]
-+  |   stfd f1, CTSTATE->cb.fpr[0]
-+  |  std r4, CTSTATE->cb.gpr[1]
-+  |   stfd f2, CTSTATE->cb.fpr[1]
-+  |  std r5, CTSTATE->cb.gpr[2]
-+  |   stfd f3, CTSTATE->cb.fpr[2]
-+  |  std r6, CTSTATE->cb.gpr[3]
-+  |   stfd f4, CTSTATE->cb.fpr[3]
-+  |  std r7, CTSTATE->cb.gpr[4]
-+  |   stfd f5, CTSTATE->cb.fpr[4]
-+  |  std r8, CTSTATE->cb.gpr[5]
-+  |   stfd f6, CTSTATE->cb.fpr[5]
-+  |  std r9, CTSTATE->cb.gpr[6]
-+  |   stfd f7, CTSTATE->cb.fpr[6]
-+  |  std r10, CTSTATE->cb.gpr[7]
-+  |   stfd f8, CTSTATE->cb.fpr[7]
-+  |  addi TMP0, sp, CFRAME_SPACE+8
-+  |  std TMP0, CTSTATE->cb.stack
-+  |   mr CARG1, CTSTATE
-+  |  std CTSTATE, SAVE_PC		// Any value outside of bytecode is ok.
-+  |   mr CARG2, sp
-+  |  bl extern lj_ccallback_enter	// (CTState *cts, void *cf)
-+  |  // Returns lua_State *.
-+  |  ld BASE, L:CRET1->base
-+  |     li TISNUM, LJ_TISNUM		// Setup type comparison constants.
-+  |     set_oper_type TISNUM, TISNUM
-+  |  ld RC, L:CRET1->top
-+  |     lus TMP3, 0x59c0		// TOBIT = 2^52 + 2^51 (float).
-+  |     li ZERO, 0
-+  |   mr L, CRET1
-+  |     std TMP3, TMPD
-+  |     lus TMP0, 0x4338		// Hiword of 2^52 + 2^51 (double)
-+  |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+  |     ori TMP3, TMP3, 0x0004		// TONUM = 2^52 + 2^51 + 2^31 (float).
-+  |     stw TMP0, TONUM_HI
-+  |     li TISNIL, LJ_TNIL
-+  |    li_vmstate INTERP
-+  |     lfs TOBIT, TMPD
-+  |     std TMP3, TMPD
-+  |  sub RC, RC, BASE
-+  |    st_vmstate
-+  |     lfs TONUM, TMPD
-+  |  clear_field RB
-+  |  ins_callt
-+  |.endif
-+  |
-+  |->cont_ffi_callback:			// Return from FFI callback.
-+  |.if FFI
-+  |  ld CTSTATE, DISPATCH_GL(ctype_state)(DISPATCH)
-+  |   std BASE, L->base
-+  |   std RB, L->top
-+  |  std L, CTSTATE->L
-+  |  mr CARG1, CTSTATE
-+  |  mr CARG2, RA
-+  |  bl extern lj_ccallback_leave	// (CTState *cts, TValue *o)
-+  |  ld CRET1, CTSTATE->cb.gpr[0]
-+  |  lfd FARG1, CTSTATE->cb.fpr[0]
-+  |  ld CRET2, CTSTATE->cb.gpr[1]
-+  |  b ->vm_leave_unw
-+  |.endif
-+  |
-+  |->vm_ffi_call:			// Call C function via FFI.
-+  |  // Caveat: needs special frame unwinding, see below.
-+  |.if FFI
-+  |  .type CCSTATE, CCallState, CARG1
-+  |  lwz TMP1, CCSTATE->spadj
-+  |    mflr TMP0
-+  |   lbz CARG2, CCSTATE->nsp
-+  |   lbz CARG3, CCSTATE->nfpr
-+  |  neg TMP1, TMP1
-+  |    std TMP0, 16(sp)
-+  |   cmpdi cr1, CARG3, 0
-+  |   std TOCREG, 24(sp)
-+  |  mr TMP2, sp
-+  |   addic. CARG2, CARG2, -1
-+  |  stdux sp, sp, TMP1
-+  |   crnot 4*cr1+eq, 4*cr1+eq		// For vararg calls.
-+  |  std r14, -8(TMP2)
-+  |  std CCSTATE, -16(TMP2)
-+  |  mr r14, TMP2
-+  |  la TMP1, CCSTATE->stack
-+  |   sldi CARG2, CARG2, 3
-+  |   blty >2
-+  |  la TMP2, 96(sp)			// stack params after regs
-+  |1:  // Copy stack slots
-+  |  ldx TMP0, TMP1, CARG2
-+  |  stdx TMP0, TMP2, CARG2
-+  |   addic. CARG2, CARG2, -8
-+  |  bge <1
-+  |2:
-+  |  bney cr1, >3
-+  |  lfd f1, CCSTATE->fpr[0]
-+  |  lfd f2, CCSTATE->fpr[1]
-+  |  lfd f3, CCSTATE->fpr[2]
-+  |  lfd f4, CCSTATE->fpr[3]
-+  |  lfd f5, CCSTATE->fpr[4]
-+  |  lfd f6, CCSTATE->fpr[5]
-+  |  lfd f7, CCSTATE->fpr[6]
-+  |  lfd f8, CCSTATE->fpr[7]
-+  |  lfd f9, CCSTATE->fpr[8]
-+  |  lfd f10, CCSTATE->fpr[9]
-+  |  lfd f11, CCSTATE->fpr[10]
-+  |  lfd f12, CCSTATE->fpr[11]
-+  |  lfd f13, CCSTATE->fpr[12]
-+  |3:
-+  |  ld r12, CCSTATE->func
-+  |  ld CARG2, CCSTATE->gpr[1]
-+  |  ld CARG3, CCSTATE->gpr[2]
-+  |  ld CARG4, CCSTATE->gpr[3]
-+  |  ld CARG5, CCSTATE->gpr[4]
-+  |   mtctr r12
-+  |  ld r8, CCSTATE->gpr[5]
-+  |  ld r9, CCSTATE->gpr[6]
-+  |  ld r10, CCSTATE->gpr[7]
-+  |  ld CARG1, CCSTATE->gpr[0]		// Do this last, since CCSTATE is CARG1.
-+  |   bctrl
-+  |  ld CCSTATE:TMP1, -16(r14)
-+  |  ld TMP2, -8(r14)
-+  |   ld TMP0, 16(r14)
-+  |   ld TOCREG, 24(r14)
-+  |  std CARG1, CCSTATE:TMP1->gpr[0]
-+  |  stfd FARG1, CCSTATE:TMP1->fpr[0]
-+  |  std CARG2, CCSTATE:TMP1->gpr[1]
-+  |   mtlr TMP0
-+  |  std CARG3, CCSTATE:TMP1->gpr[2]
-+  |   mr sp, r14
-+  |  std CARG4, CCSTATE:TMP1->gpr[3]
-+  |   mr r14, TMP2
-+  |  blr
-+  |.endif
-+  |// Note: vm_ffi_call must be the last function in this object file!
-+  |
-+  |//-----------------------------------------------------------------------
-+}
-+
-+/* Generate the code for a single instruction. */
-+static void build_ins(BuildCtx *ctx, BCOp op, int defop)
-+{
-+  int vk = 0;
-+  |=>defop:
-+
-+  switch (op) {
-+
-+  /* -- Comparison ops ---------------------------------------------------- */
-+
-+  /* Remember: all ops branch for a true comparison, fall through otherwise. */
-+
-+  case BC_ISLT: case BC_ISGE: case BC_ISLE: case BC_ISGT:
-+    |  // RA = src1*8, RD = src2*8, JMP with RD = target
-+    |  ldux CARG2, RA, BASE
-+    |    addi PC, PC, 4
-+    |     get_oper_type TMP0, CARG2
-+    |  ldux CARG3, RD, BASE
-+    |    lwz TMP2, -4(PC)
-+    |  checknum cr0, CARG2
-+    |    decode_RD4 TMP2, TMP2
-+    |    get_oper_type TMP1, CARG3
-+    |  checknum cr1, CARG3
-+    |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+    |  bne cr0, >7
-+    |  bne cr1, >8
-+    |   cmpd CARG2, CARG3
-+    if (op == BC_ISLT) {
-+      |  bge >2
-+    } else if (op == BC_ISGE) {
-+      |  blt >2
-+    } else if (op == BC_ISLE) {
-+      |  bgt >2
-+    } else {
-+      |  ble >2
-+    }
-+    |1:
-+    |  add PC, PC, TMP2
-+    |2:
-+    |  ins_next
-+    |
-+    |7:  // RA is not an integer.
-+    |  bgt cr0, ->vmeta_comp
-+    |  // RA is a number.
-+    |   lfd f0, 0(RA)
-+    |  bgt cr1, ->vmeta_comp
-+    |  blt cr1, >4
-+    |  // RA is a number, RD is an integer.
-+    |  tonum_i f1, CARG3
-+    |  b >5
-+    |
-+    |8: // RA is an integer, RD is not an integer.
-+    |  bgt cr1, ->vmeta_comp
-+    |  // RA is an integer, RD is a number.
-+    |  tonum_i f0, CARG2
-+    |4:
-+    |  lfd f1, 0(RD)
-+    |5:
-+    |  fcmpu cr0, f0, f1
-+    if (op == BC_ISLT) {
-+      |  bge <2
-+    } else if (op == BC_ISGE) {
-+      |  blt <2
-+    } else if (op == BC_ISLE) {
-+      |  cror 4*cr0+lt, 4*cr0+lt, 4*cr0+eq
-+      |  bge <2
-+    } else {
-+      |  cror 4*cr0+lt, 4*cr0+lt, 4*cr0+eq
-+      |  blt <2
-+    }
-+    |  b <1
-+    break;
-+
-+  case BC_ISEQV: case BC_ISNEV:
-+    vk = op == BC_ISEQV;
-+    |  // RA = src1*8, RD = src2*8, JMP with RD = target
-+    |  ldux CARG2, RA, BASE
-+    |    addi PC, PC, 4
-+    |  get_oper_type TMP0, CARG2
-+    |  ldux CARG3, RD, BASE
-+    |  li TMP3, LJ_TISNUM
-+    |  cmpld cr0, TMP0, TMP3
-+    |    ld TMP2, -4(PC)
-+    |  get_oper_type TMP1, CARG3
-+    |  cmpld cr1, TMP1, TMP3
-+    |    decode_RD4 TMP2, TMP2
-+    |  cror 4*cr7+gt, 4*cr0+gt, 4*cr1+gt
-+    |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+    if (vk) {
-+      |  ble cr7, ->BC_ISEQN_Z
-+    } else {
-+      |  ble cr7, ->BC_ISNEN_Z
-+    }
-+    |5:  // Either or both types are not numbers.
-+    |.if FFI
-+    |  cmpdi cr7, TMP0, LJ_TCDATA
-+    |  cmpdi cr5, TMP1, LJ_TCDATA
-+    |.endif
-+    |   not TMP3, TMP0
-+    |  cmpld TMP0, TMP1
-+    |   cmpldi cr1, TMP3, ~LJ_TISPRI		// Primitive?
-+    |.if FFI
-+    |  cror 4*cr7+eq, 4*cr7+eq, 4*cr5+eq
-+    |.endif
-+    |   cmpldi cr6, TMP3, ~LJ_TISTABUD		// Table or userdata?
-+    |.if FFI
-+    |  beq cr7, ->vmeta_equal_cd
-+    |.endif
-+    |    cmpld cr5, CARG2, CARG3
-+    |  crandc 4*cr0+gt, 4*cr0+eq, 4*cr1+gt	// 2: Same type and primitive.
-+    |  crorc 4*cr0+lt, 4*cr5+eq, 4*cr0+eq	// 1: Same tv or different type.
-+    |  crand 4*cr0+eq, 4*cr0+eq, 4*cr5+eq	// 0: Same type and same tv.
-+    |   mr SAVE0, PC
-+    |  cror 4*cr0+eq, 4*cr0+eq, 4*cr0+gt	// 0 or 2.
-+    |  cror 4*cr0+lt, 4*cr0+lt, 4*cr0+gt	// 1 or 2.
-+    if (vk) {
-+      |  bne cr0, >6
-+      |  add PC, PC, TMP2
-+      |6:
-+    } else {
-+      |  beq cr0, >6
-+      |  add PC, PC, TMP2
-+      |6:
-+    }
-+    |  bge cr0, >2			// Done if 1 or 2.
-+    |1:
-+    |  ins_next
-+    |2:
-+    |  blt cr6, <1			// Done if not tab/ud.
-+    |
-+    |  // Different tables or userdatas. Need to check __eq metamethod.
-+    |  // Field metatable must be at same offset for GCtab and GCudata!
-+    |  clear_field TAB:CARG2
-+    |  ld TAB:TMP2, TAB:CARG2->metatable
-+    |   li CARG4, 1-vk			// ne = 0 or 1.
-+    |  cmpldi TAB:TMP2, 0
-+    |  beq <1				// No metatable?
-+    |  lbz TMP2, TAB:TMP2->nomm
-+    |  andi. TMP2, TMP2, 1<<MM_eq
-+    |  bne <1				// Or 'no __eq' flag set?
-+    |  mr PC, SAVE0			// Restore old PC.
-+    |  b ->vmeta_equal			// Handle __eq metamethod.
-+    break;
-+
-+  case BC_ISEQS: case BC_ISNES:
-+    vk = op == BC_ISEQS;
-+    |  // RA = src*8, RD = str_const*8 (~), JMP with RD = target
-+    |  ldux TMP3, RA, BASE
-+    |    get_oper_type TMP0,TMP3
-+    |    clear_field TMP3
-+    |    lwz TMP2, 0(PC)
-+    |   subfic RD, RD, -8
-+    |    addi PC, PC, 4
-+    |.if FFI
-+    |  cmpdi TMP0, LJ_TCDATA
-+    |.endif
-+    |   ldx STR:TMP1, KBASE, RD		// KBASE-8-str_const*8
-+    |  subfic TMP0, TMP0, LJ_TSTR
-+    |.if FFI
-+    |  beq ->vmeta_equal_cd
-+    |.endif
-+    |  sub TMP1, STR:TMP1, STR:TMP3
-+    |  or TMP0, TMP0, TMP1
-+    |    decode_RD4 TMP2, TMP2
-+    |  subfic TMP0, TMP0, 0
-+    |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+    |  subfe TMP1, TMP1, TMP1
-+    if (vk) {
-+      |  andc TMP2, TMP2, TMP1
-+    } else {
-+      |  and TMP2, TMP2, TMP1
-+    }
-+    |  add PC, PC, TMP2
-+    |  ins_next
-+    break;
-+
-+  case BC_ISEQN: case BC_ISNEN:
-+    vk = op == BC_ISEQN;
-+    |  // RA = src*8, RD = num_const*8, JMP with RD = target
-+    |  ldux CARG2, RA, BASE
-+    |    addi PC, PC, 4
-+    |   get_oper_type TMP0, CARG2
-+    |  ldux CARG3, RD, KBASE
-+    |   li TMP4, LJ_TISNUM
-+    |  get_value CARG2
-+    |   cmpld cr0, TMP0, TMP4
-+    |    ld TMP2, -4(PC)
-+    |  checknum cr1, CARG3
-+    |    decode_RD4 TMP2, TMP2
-+    |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+    if (vk) {
-+      |->BC_ISEQN_Z:
-+    } else {
-+      |->BC_ISNEN_Z:
-+    }
-+    |  bne cr0, >7
-+    |  bne cr1, >8
-+    |   cmpw CARG2, CARG3
-+    |4:
-+    if (vk) {
-+      |  bne >1
-+      |  add PC, PC, TMP2
-+      |1:
-+      |.if not FFI
-+      |3:
-+      |.endif
-+    } else {
-+      |  beq >2
-+      |1:
-+      |.if not FFI
-+      |3:
-+      |.endif
-+      |  add PC, PC, TMP2
-+      |2:
-+    }
-+    |  ins_next
-+    |.if FFI
-+    |3:
-+    |  cmpdi TMP0, LJ_TCDATA
-+    |  beq ->vmeta_equal_cd
-+    |  b <1
-+    |.endif
-+    |7:  // RA is not an integer.
-+    |  bge cr0, <3
-+    |  // RA is a number.
-+    |   lfd f0, 0(RA)
-+    |  blt cr1, >1
-+    |  // RA is a number, RD is an integer.
-+    |  tonum_i f1, CARG3
-+    |  b >2
-+    |
-+    |8: // RA is an integer, RD is a number.
-+    |  tonum_i f0, CARG2
-+    |1:
-+    |  lfd f1, 0(RD)
-+    |2:
-+    |  fcmpu cr0, f0, f1
-+    |  b <4
-+    break;
-+
-+  case BC_ISEQP: case BC_ISNEP:
-+    vk = op == BC_ISEQP;
-+    |  // RA = src*8, RD = primitive_type*8 (~), JMP with RD = target
-+    |  ldx TMP0, BASE, RA
-+    |   srdi TMP1, RD, 3
-+    |    lwz TMP2, 0(PC)
-+    |  get_oper_type TMP0, TMP0
-+    |   not TMP1, TMP1
-+    |    addi PC, PC, 4
-+    |.if FFI
-+    |  cmpdi TMP0, LJ_TCDATA
-+    |.endif
-+    |  sub TMP0, TMP0, TMP1
-+    |.if FFI
-+    |  beq ->vmeta_equal_cd
-+    |.endif
-+    |    decode_RD4 TMP2, TMP2
-+    |   extsw TMP0, TMP0
-+    |  addic TMP0, TMP0, -1
-+    |    addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+    |  subfe TMP1, TMP1, TMP1
-+    if (vk) {
-+      |  and TMP2, TMP2, TMP1
-+    } else {
-+      |  andc TMP2, TMP2, TMP1
-+    }
-+    |  add PC, PC, TMP2
-+    |  ins_next
-+    break;
-+
-+  /* -- Unary test and copy ops ------------------------------------------- */
-+
-+  case BC_ISTC: case BC_ISFC: case BC_IST: case BC_ISF:
-+    |  // RA = dst*8 or unused, RD = src*8, JMP with RD = target
-+    |  ldx TMP0, BASE, RD
-+    |   lwz INS, 0(PC)
-+    |   addi PC, PC, 4
-+    if (op == BC_IST || op == BC_ISF) {
-+      |  get_oper_type TMP0, TMP0
-+      |  subfic TMP0, TMP0, LJ_TTRUE
-+      |   decode_RD4 TMP2, INS
-+      |  subfe TMP1, TMP1, TMP1
-+      |   addis TMP2, TMP2, -(BCBIAS_J*4 >> 16)
-+      if (op == BC_IST) {
-+	|  andc TMP2, TMP2, TMP1
-+      } else {
-+	|  and TMP2, TMP2, TMP1
-+      }
-+      |  add PC, PC, TMP2
-+    } else {
-+      |  set_bool TMP1, LJ_TFALSE
-+      |   lfdx f0, BASE, RD
-+      |  cmpld TMP0, TMP1		// compare directly due to get_oper_type
-+      if (op == BC_ISTC) {
-+	|  bge >1
-+      } else {
-+	|  blt >1
-+      }
-+      |  addis PC, PC, -(BCBIAS_J*4 >> 16)
-+      |  decode_RD4 TMP2, INS
-+      |   stfdx f0, BASE, RA
-+      |  add PC, PC, TMP2
-+      |1:
-+    }
-+    |  ins_next
-+    break;
-+
-+  case BC_ISTYPE:
-+    |  // RA = src*8, RD = -type*8
-+    |  ldx TMP0, BASE, RA
-+    |  srdi TMP1, RD, 3
-+    |  ins_next1
-+    |  neg TMP1, TMP1
-+    |  get_oper_type TMP0, TMP0		// TMP1 is not shifted
-+    |  cmpd TMP0, TMP1
-+    |  bne ->vmeta_istype
-+    |  ins_next2
-+    break;
-+  case BC_ISNUM:
-+    |  // RA = src*8, RD = -(TISNUM-1)*8
-+    |  ldx TMP0, BASE, RA
-+    |  ins_next1
-+    |  checknum TMP0
-+    |  bge ->vmeta_istype
-+    |  ins_next2
-+    break;
-+
-+  /* -- Unary ops --------------------------------------------------------- */
-+
-+  case BC_MOV:
-+    |  // RA = dst*8, RD = src*8
-+    |  ins_next1
-+    |  lfdx f0, BASE, RD
-+    |  stfdx f0, BASE, RA
-+    |  ins_next2
-+    break;
-+  case BC_NOT:
-+    |  // RA = dst*8, RD = src*8
-+    |  ins_next1
-+    |  ldx TMP0, BASE, RD
-+    |  rotldi TMP0, TMP0, 17
-+    |  subfic TMP1, TMP0, LJ_TTRUE
-+    |  adde TMP0, TMP0, TMP1
-+    |  rotldi TMP0, TMP0, 47
-+    |  stdx TMP0, BASE, RA
-+    |  ins_next2
-+    break;
-+  case BC_UNM:
-+    |  // RA = dst*8, RD = src*8
-+    |  ldux TMP0, RD, BASE		// RD is Used in vmeta_unm
-+    |  mr TMP1, TMP0
-+    |  lus TMP2, 0x8000
-+    |  checknum TMP1
-+    |  bne >5
-+    |  neg TMP1, TMP1
-+    |  cmplw TMP1, TMP2
-+    |  beq >4
-+    |1:
-+    |  ins_next1
-+    |  rldicl TMP0, TMP1, 0, 32		// clear the high order 32 bits
-+    |  add TMP0, TMP0, TISNUM
-+    |  stdx TMP0, RA, BASE
-+    |3:
-+    |  ins_next2
-+    |4:
-+    |  li TMP0, 0x41e0			// 2^31.
-+    |  sldi TMP0, TMP0, 48
-+    |  b >7
-+    |5:
-+    |  sldi TMP2, TMP2, 32
-+    |  bge ->vmeta_unm
-+    |  xor TMP0, TMP0, TMP2
-+    |7:
-+    |  ins_next1
-+    |  stdx TMP0, RA, BASE
-+    |  b <3
-+    break;
-+  case BC_LEN:
-+    |  // RA = dst*8, RD = src*8
-+    |  ldux CARG1, RD, BASE
-+    |  mr TMP0, CARG1
-+    |  checkstr CARG1; bne >2
-+    |  lwz CRET1, STR:CARG1->len
-+    |1:
-+    |  ins_next1
-+    |  add CRET1, CRET1, TISNUM
-+    |  stdux CRET1, RA, BASE
-+    |  ins_next2
-+    |2:
-+    |  checktab TMP0; bne ->vmeta_len
-+#if LJ_52
-+    |  ld TAB:TMP2, TAB:CARG1->metatable
-+    |  cmpldi TAB:TMP2, 0
-+    |  bne >9
-+    |3:
-+#endif
-+    |->BC_LEN_Z:
-+    |  bl extern lj_tab_len		// (GCtab *t)
-+    |  // Returns uint32_t (but less than 2^31).
-+    |  b <1
-+#if LJ_52
-+    |9:
-+    |  lbz TMP0, TAB:TMP2->nomm
-+    |  andi. TMP0, TMP0, 1<<MM_len
-+    |  bne <3				// 'no __len' flag set: done.
-+    |  b ->vmeta_len
-+#endif
-+    break;
-+
-+  /* -- Binary ops -------------------------------------------------------- */
-+
-+    |.macro ins_arithpre
-+    |  // RA = dst*8, RB = src1*8, RC = src2*8 | num_const*8
-+    ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN);
-+    ||switch (vk) {
-+    ||case 0:
-+    |   ldx TMP1, BASE, RB
-+    |     ldx TMP2, KBASE, RC
-+    |    lfdx f14, BASE, RB
-+    |    lfdx f15, KBASE, RC
-+    |     checknum cr0, TMP1
-+    |     checknum cr1, TMP2
-+    |     crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |     bge ->vmeta_arith_vn
-+    ||  break;
-+    ||case 1:
-+    |   ldx TMP1, BASE, RB
-+    |     ldx TMP2, KBASE, RC
-+    |    lfdx f15, BASE, RB
-+    |    lfdx f14, KBASE, RC
-+    |     checknum cr0, TMP1
-+    |     checknum cr1, TMP2
-+    |     crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |     bge ->vmeta_arith_nv
-+    ||  break;
-+    ||default:
-+    |   ldx TMP1, BASE, RB
-+    |   ldx TMP2, BASE, RC
-+    |    lfdx f14, BASE, RB
-+    |    lfdx f15, BASE, RC
-+    |   checknum cr0, TMP1
-+    |   checknum cr1, TMP2
-+    |   crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |   bge ->vmeta_arith_vv
-+    ||  break;
-+    ||}
-+    |.endmacro
-+    |
-+    |.macro ins_arithfallback, ins
-+    ||switch (vk) {
-+    ||case 0:
-+    |   ins ->vmeta_arith_vn2
-+    ||  break;
-+    ||case 1:
-+    |   ins ->vmeta_arith_nv2
-+    ||  break;
-+    ||default:
-+    |   ins ->vmeta_arith_vv2
-+    ||  break;
-+    ||}
-+    |.endmacro
-+    |
-+    |.macro intmod, a, b, c
-+    |  bl ->vm_modi
-+    |.endmacro
-+    |
-+    |.macro fpmod, a, b, c
-+    |->BC_MODVN_Z:
-+    |  fdiv FARG1, b, c
-+    |  // NYI: Use internal implementation of floor.
-+    |  blex floor			// floor(b/c)
-+    |  fmul a, FARG1, c
-+    |  fsub a, b, a			// b - floor(b/c)*c
-+    |.endmacro
-+    |
-+    |.macro ins_arithfp, fpins
-+    |  ins_arithpre
-+    |.if "fpins" == "fpmod_"
-+    |  b ->BC_MODVN_Z			// Avoid 3 copies. It's slow anyway.
-+    |.else
-+    |  fpins f0, f14, f15
-+    |  ins_next1
-+    |  stfdx f0, BASE, RA
-+    |  ins_next2
-+    |.endif
-+    |.endmacro
-+    |
-+    |.macro ins_arithdn, intins, fpins
-+    |  // RA = dst*8, RB = src1*8, RC = src2*8 | num_const*8
-+    ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN);
-+    ||switch (vk) {
-+    ||case 0:
-+    |   ldux CARG1, RB, BASE
-+    |   ldux CARG2, RC, KBASE
-+    |   checknum cr0, CARG1
-+    ||  break;
-+    ||case 1:
-+    |   ldux CARG1, RC, KBASE
-+    |   ldux CARG2, RB, BASE
-+    |   checknum cr0, CARG1
-+    ||  break;
-+    ||default:
-+    |   ldux CARG1, RB, BASE
-+    |   ldux CARG2, RC, BASE
-+    |   checknum cr0, CARG1
-+    ||  break;
-+    ||}
-+    |  checknum cr1, CARG2
-+    |  bne >5
-+    |  bne cr1, >5
-+    |  intins CARG1, CARG1, CARG2
-+    |  bso >4
-+    |1:
-+    |  ins_next1
-+    |  li TISNUM, LJ_TISNUM
-+    |  set_oper_type TISNUM, TISNUM
-+    |  clear_field CARG1
-+    |  add CARG1, CARG1, TISNUM
-+    |  stdux CARG1, RA, BASE
-+    |2:
-+    |  ins_next2
-+    |4:  // Overflow.
-+    |  checkov <1			// Ignore unrelated overflow.
-+    |  ins_arithfallback b
-+    |5:  // FP variant.
-+    ||if (vk == 1) {
-+    |  lfd f15, 0(RB)
-+    |   crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |  lfd f14, 0(RC)
-+    ||} else {
-+    |  lfd f14, 0(RB)
-+    |   crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |  lfd f15, 0(RC)
-+    ||}
-+    |   ins_arithfallback bge
-+    |.if "fpins" == "fpmod_"
-+    |  b ->BC_MODVN_Z			// Avoid 3 copies. It's slow anyway.
-+    |.else
-+    |  fpins f0, f14, f15
-+    |  ins_next1
-+    |  stfdx f0, BASE, RA
-+    |  b <2
-+    |.endif
-+    |.endmacro
-+    |
-+    |.macro ins_arith, intins, fpins
-+    |  ins_arithdn intins, fpins
-+    |.endmacro
-+
-+  case BC_ADDVN: case BC_ADDNV: case BC_ADDVV:
-+    |.macro addo32., y, a, b
-+    |  // Need to check overflow for (a<<32) + (b<<32).
-+    |  rldicr TMP0, a, 32, 31
-+    |  rldicr TMP3, b, 32, 31
-+    |  addo. TMP0, TMP0, TMP3
-+    |  add y, a, b
-+    |.endmacro
-+    |  ins_arith addo32., fadd
-+    break;
-+  case BC_SUBVN: case BC_SUBNV: case BC_SUBVV:
-+    |.macro subo32., y, a, b
-+    |  // Need to check overflow for (a<<32) - (b<<32).
-+    |  rldicr TMP0, a, 32, 31
-+    |  rldicr TMP3, b, 32, 31
-+    |  subo. TMP0, TMP0, TMP3
-+    |  sub y, a, b
-+    |.endmacro
-+    |  ins_arith subo32., fsub
-+    break;
-+  case BC_MULVN: case BC_MULNV: case BC_MULVV:
-+    |  ins_arith mullwo., fmul
-+    break;
-+  case BC_DIVVN: case BC_DIVNV: case BC_DIVVV:
-+    |  ins_arithfp fdiv
-+    break;
-+  case BC_MODVN:
-+    |  ins_arith intmod, fpmod
-+    break;
-+  case BC_MODNV: case BC_MODVV:
-+    |  ins_arith intmod, fpmod_
-+    break;
-+  case BC_POW:
-+    |  // NYI: (partial) integer arithmetic.
-+    |  ldx TMP1, BASE, RB
-+    |   lfdx FARG1, BASE, RB
-+    |  ldx TMP2, BASE, RC
-+    |   lfdx FARG2, BASE, RC
-+    |  checknum cr0, TMP1
-+    |  checknum cr1, TMP2
-+    |  crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+    |  bge ->vmeta_arith_vv
-+    |  blex pow
-+    |  ins_next1
-+    |  stfdx FARG1, BASE, RA
-+    |  ins_next2
-+    break;
-+
-+  case BC_CAT:
-+    |  // RA = dst*8, RB = src_start*8, RC = src_end*8
-+    |  sub CARG3, RC, RB
-+    |   std BASE, L->base
-+    |  add CARG2, BASE, RC
-+    |  mr SAVE0, RB
-+    |->BC_CAT_Z:
-+    |   std PC, SAVE_PC
-+    |  mr CARG1, L
-+    |  srdi CARG3, CARG3, 3
-+    |  bl extern lj_meta_cat		// (lua_State *L, TValue *top, int left)
-+    |  // Returns NULL (finished) or TValue * (metamethod).
-+    |  cmpldi CRET1, 0
-+    |   ld BASE, L->base
-+    |  bne ->vmeta_binop
-+    |  ins_next1
-+    |  lfdx f0, BASE, SAVE0		// Copy result from RB to RA.
-+    |  stfdx f0, BASE, RA
-+    |  ins_next2
-+    break;
-+  /* -- Constant ops ------------------------------------------------------ */
-+
-+  case BC_KSTR:
-+    |  // RA = dst*8, RD = str_const*8 (~)
-+    |  subfic TMP1, RD, -8
-+    |  ins_next1
-+    |  ldx TMP0, KBASE, TMP1		// KBASE-8-str_const*8
-+    |  li TMP2, LJ_TSTR
-+    |  set_oper_type TMP2, TMP2
-+    |  add TMP0, TMP2, TMP0
-+    |  stdx TMP0, RA, BASE
-+    |  ins_next2
-+    break;
-+  case BC_KCDATA:
-+    |.if FFI
-+    |  // RA = dst*8, RD = cdata_const*8 (~)
-+    |  subfic TMP1, RD, -8
-+    |  ins_next1
-+    |  ldx TMP0, KBASE, TMP1		// KBASE-8-cdata_const*8
-+    |  li TMP2, LJ_TCDATA
-+    |  set_oper_type TMP2, TMP2
-+    |  add TMP2, TMP2, TMP0
-+    |  stdx TMP2, RA, BASE
-+    |  ins_next2
-+    |.endif
-+    break;
-+  case BC_KSHORT:
-+    |  // RA = dst*8, RD = int16_literal*8
-+    |  srdi RD, RD, 3
-+    |  extsh RD, RD			// extend sign for negative numbers
-+    |  clear_field RD
-+    |  ins_next1
-+    |   add TMP0, RD, TISNUM
-+    |   stdx TMP0, RA, BASE
-+    |  ins_next2
-+    break;
-+  case BC_KNUM:
-+    |  // RA = dst*8, RD = num_const*8
-+    |  ins_next1
-+    |  lfdx f0, KBASE, RD
-+    |  stfdx f0, BASE, RA
-+    |  ins_next2
-+    break;
-+  case BC_KPRI:
-+    |  // RA = dst*8, RD = primitive_type*8 (~)
-+    |  srdi TMP1, RD, 3
-+    |  set_oper_type TMP1,TMP1
-+    |  not TMP0, TMP1
-+    |  ins_next1
-+    |  stdx TMP0, BASE, RA
-+    |  ins_next2
-+    break;
-+  case BC_KNIL:
-+    |  // RA = base*8, RD = end*8
-+    |  stdx TISNIL, BASE, RA
-+    |   addi RA, RA, 8
-+    |1:
-+    |  stdx TISNIL, BASE, RA
-+    |  cmpd RA, RD
-+    |   addi RA, RA, 8
-+    |  blt <1
-+    |  ins_next_
-+    break;
-+
-+  /* -- Upvalue and function ops ------------------------------------------ */
-+
-+  case BC_UGET:
-+    |  // RA = dst*8, RD = uvnum*8
-+    |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+    |   addi RD, RD, offsetof(GCfuncL, uvptr)
-+    |  clear_field RB
-+    |  ldx UPVAL:RB, LFUNC:RB, RD
-+    |  ins_next1
-+    |  ld TMP1, UPVAL:RB->v
-+    |  ld TMP1, 0(TMP1)
-+    |  stdx TMP1, BASE, RA
-+    |  ins_next2
-+    break;
-+  case BC_USETV:
-+    |  // RA = uvnum*8, RD = src*8
-+    |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+    |    addi RA, RA, offsetof(GCfuncL, uvptr)
-+    |   lfdux f0, RD, BASE
-+    |  clear_field RB
-+    |  ldx UPVAL:RB, LFUNC:RB, RA
-+    |  lbz TMP3, UPVAL:RB->marked
-+    |   ld CARG2, UPVAL:RB->v
-+    |  andi. TMP3, TMP3, LJ_GC_BLACK	// isblack(uv)
-+    |    lbz TMP0, UPVAL:RB->closed
-+    |   ld TMP1, 0(RD)
-+    |   get_oper_type TMP2, TMP1
-+    |   stfd f0, 0(CARG2)
-+    |    cmpldi cr1, TMP0, 0
-+    |  cror 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
-+    |   subi TMP2, TMP2, (LJ_TNUMX+1)
-+    |  bne >2				// Upvalue is closed and black?
-+    |1:
-+    |  ins_next
-+    |
-+    |2:  // Check if new value is collectable.
-+    |  cmpldi TMP2, LJ_TISGCV - (LJ_TNUMX+1)
-+    |  bge <1				// tvisgcv(v)
-+    |  clear_field GCOBJ:TMP1
-+    |  lbz TMP3, GCOBJ:TMP1->gch.marked
-+    |  andi. TMP3, TMP3, LJ_GC_WHITES	// iswhite(v)
-+    |   la CARG1, GG_DISP2G(DISPATCH)
-+    |  // Crossed a write barrier. Move the barrier forward.
-+    |  beq <1
-+    |  bl extern lj_gc_barrieruv	// (global_State *g, TValue *tv)
-+    |  b <1
-+    break;
-+  case BC_USETS:
-+    |  // RA = uvnum*8, RD = str_const*8 (~)
-+    |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+    |   subfic TMP1, RD, -8
-+    |    addi RA, RA, offsetof(GCfuncL, uvptr)
-+    |  clear_field RB
-+    |   ldx STR:TMP1, KBASE, TMP1	// KBASE-8-str_const*8
-+    |  ldx UPVAL:RB, LFUNC:RB, RA
-+    |  lbz TMP3, UPVAL:RB->marked
-+    |   ld CARG2, UPVAL:RB->v
-+    |  andi. TMP3, TMP3, LJ_GC_BLACK	// isblack(uv)
-+    |   lbz TMP3, STR:TMP1->marked
-+    |   lbz TMP2, UPVAL:RB->closed
-+    |   li TMP0, LJ_TSTR
-+    |   set_oper_type TMP0, TMP0
-+    |   add TMP0, TMP0, STR:TMP1
-+    |   std TMP0, 0(CARG2)
-+    |  bne >2
-+    |1:
-+    |  ins_next
-+    |
-+    |2:  // Check if string is white and ensure upvalue is closed.
-+    |  andi. TMP3, TMP3, LJ_GC_WHITES	// iswhite(str)
-+    |   cmpldi cr1, TMP2, 0
-+    |  cror 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
-+    |   la CARG1, GG_DISP2G(DISPATCH)
-+    |  // Crossed a write barrier. Move the barrier forward.
-+    |  beq <1
-+    |  bl extern lj_gc_barrieruv	// (global_State *g, TValue *tv)
-+    |  b <1
-+    break;
-+  case BC_USETN:
-+    |  // RA = uvnum*8, RD = num_const*8
-+    |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+    |   addi RA, RA, offsetof(GCfuncL, uvptr)
-+    |  clear_field RB
-+    |    lfdx f0, KBASE, RD
-+    |  ldx UPVAL:RB, LFUNC:RB, RA
-+    |  ins_next1
-+    |  ld TMP1, UPVAL:RB->v
-+    |  stfd f0, 0(TMP1)
-+    |  ins_next2
-+    break;
-+  case BC_USETP:
-+    |  // RA = uvnum*8, RD = primitive_type*8 (~)
-+    |  ld LFUNC:RB, FRAME_FUNC(BASE)
-+    |    srdi TMP0, RD, 3
-+    |   addi RA, RA, offsetof(GCfuncL, uvptr)
-+    |    set_oper_type TMP0, TMP0
-+    |  clear_field RB
-+    |    not TMP0, TMP0
-+    |  ldx UPVAL:RB, LFUNC:RB, RA
-+    |  ins_next1
-+    |  ld TMP1, UPVAL:RB->v
-+    |  std TMP0, 0(TMP1)
-+    |  ins_next2
-+    break;
-+
-+  case BC_UCLO:
-+    |  // RA = level*8, RD = target
-+    |  ld TMP1, L->openupval
-+    |  branch_RD			// Do this first since RD is not saved.
-+    |   std BASE, L->base
-+    |  cmpldi TMP1, 0
-+    |   mr CARG1, L
-+    |  beq >1
-+    |   add CARG2, BASE, RA
-+    |  bl extern lj_func_closeuv	// (lua_State *L, TValue *level)
-+    |  ld BASE, L->base
-+    |1:
-+    |  ins_next
-+    break;
-+
-+  case BC_FNEW:
-+    |  // RA = dst*8, RD = proto_const*8 (~) (holding function prototype)
-+    |   std BASE, L->base
-+    |  subfic TMP1, RD, -8
-+    |   std PC, SAVE_PC
-+    |  ldx CARG2, KBASE, TMP1		// KBASE-8-tab_const*8
-+    |   mr CARG1, L
-+    |  ld CARG3, FRAME_FUNC(BASE)
-+    |  clear_field CARG3
-+    |  // (lua_State *L, GCproto *pt, GCfuncL *parent)
-+    |  bl extern lj_func_newL_gc
-+    |  // Returns GCfuncL *.
-+    |  ld BASE, L->base
-+    |   li TMP0, LJ_TFUNC
-+    |   set_oper_type TMP0, TMP0
-+    |  add TMP0, TMP0, LFUNC:CRET1
-+    |  stdx TMP0, RA, BASE
-+    |  ins_next
-+    break;
-+
-+  /* -- Table ops --------------------------------------------------------- */
-+
-+  case BC_TNEW:
-+  case BC_TDUP:
-+    |  // RA = dst*8, RD = (hbits|asize)*8 | tab_const*8 (~)
-+    |  ld TMP0, DISPATCH_GL(gc.total)(DISPATCH)
-+    |   mr CARG1, L
-+    |  ld TMP1, DISPATCH_GL(gc.threshold)(DISPATCH)
-+    |   std BASE, L->base
-+    |  cmpld TMP0, TMP1
-+    |   std PC, SAVE_PC
-+    |  bge >5
-+    |1:
-+    if (op == BC_TNEW) {
-+      |  rlwinm CARG2, RD, 29, 21, 31
-+      |  rlwinm CARG3, RD, 18, 27, 31
-+      |  cmpdi CARG2, 0x7ff; beq >3
-+      |2:
-+      |  bl extern lj_tab_new	// (lua_State *L, int32_t asize, uint32_t hbits)
-+      |  // Returns Table *.
-+    } else {
-+      |  subfic TMP1, RD, -8
-+      |  ldx CARG2, KBASE, TMP1		// KBASE-8-tab_const*8
-+      |  bl extern lj_tab_dup	// (lua_State *L, Table *kt)
-+      |  // Returns Table *.
-+    }
-+    |  ld BASE, L->base
-+    |   li TMP0, LJ_TTAB
-+    |   set_oper_type TMP0, TMP0
-+    |   add TAB:CRET1, TMP0, TAB:CRET1
-+    |  stdx TAB:CRET1, RA, BASE
-+    |  ins_next
-+    if (op == BC_TNEW) {
-+      |3:
-+      |  li CARG2, 0x801
-+      |  b <2
-+    }
-+    |5:
-+    |  mr SAVE0, RD
-+    |  bl extern lj_gc_step_fixtop	// (lua_State *L)
-+    |  mr RD, SAVE0
-+    |  mr CARG1, L
-+    |  b <1
-+    break;
-+
-+  case BC_GGET:
-+    |  // RA = dst*8, RD = str_const*8 (~)
-+  case BC_GSET:
-+    |  // RA = src*8, RD = str_const*8 (~)
-+    |  ld LFUNC:TMP2, FRAME_FUNC(BASE)
-+    |  clear_field LFUNC:TMP2
-+    |  ld TAB:RB, LFUNC:TMP2->env
-+    |   subfic TMP1, RD, -8
-+    |   ldx STR:RC, KBASE, TMP1		// KBASE-8-str_const*8
-+    if (op == BC_GGET) {
-+      |  b ->BC_TGETS_Z
-+    } else {
-+      |  b ->BC_TSETS_Z
-+    }
-+    break;
-+
-+  case BC_TGETV:
-+    |  // RA = dst*8, RB = table*8, RC = key*8
-+    |  ldux CARG1, RB, BASE
-+    |  ldux CARG2, RC, BASE
-+    |  checktab CARG1
-+    |   checknum_noclear cr1, CARG2
-+    |  bne ->vmeta_tgetv
-+    |  lwz TMP0, TAB:CARG1->asize
-+    |   bne cr1, >5
-+    |   get_value CARG2			// Number is cleared through get_value
-+    |   ld TMP1, TAB:CARG1->array
-+    |  cmpld TMP0, CARG2
-+    |   sldi TMP2, CARG2, 3
-+    |  ble ->vmeta_tgetv		// Integer key and in array part?
-+    |  ldx TMP0, TMP1, TMP2
-+    |   lfdx f14, TMP1, TMP2
-+    |  checknil_noclear TMP0; beq >2
-+    |1:
-+    |  ins_next1
-+    |   stfdx f14, BASE, RA
-+    |  ins_next2
-+    |
-+    |2:  // Check for __index if table value is nil.
-+    |  ld TAB:TMP2, TAB:CARG1->metatable
-+    |  cmpldi TAB:TMP2, 0
-+    |  beq <1				// No metatable: done.
-+    |  lbz TMP0, TAB:TMP2->nomm
-+    |  andi. TMP0, TMP0, 1<<MM_index
-+    |  bne <1				// 'no __index' flag set: done.
-+    |  b ->vmeta_tgetv
-+    |
-+    |5:
-+    |  checkstr CARG2; bne ->vmeta_tgetv
-+    |  mr RC, CARG2			// BC_TGETS_Z needs plain RB and RC
-+    |  mr RB, CARG1			// pointers without its type
-+    |  b ->BC_TGETS_Z			// String key?
-+    break;
-+  case BC_TGETS:
-+    |  // RA = dst*8, RB = table*8, RC = str_const*8 (~)
-+    |  ldux CARG1, RB, BASE
-+    |    ld TAB:RB, 0(RB)
-+    |    clear_field TAB:RB
-+    |   subfic TMP1, RC, -8
-+    |  checktab CARG1
-+    |   ldx STR:RC, KBASE, TMP1		// KBASE-8-str_const*8
-+    |  bne ->vmeta_tgets1
-+    |->BC_TGETS_Z:
-+    |  // TAB:RB = GCtab *, STR:RC = GCstr *, RA = dst*8
-+    |  lwz TMP0, TAB:RB->hmask
-+    |  lwz TMP1, STR:RC->hash
-+    |  ld NODE:TMP2, TAB:RB->node
-+    |  and TMP1, TMP1, TMP0		// idx = str->hash & tab->hmask
-+    |  sldi TMP0, TMP1, 5
-+    |  sldi TMP1, TMP1, 3
-+    |  sub TMP1, TMP0, TMP1
-+    |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
-+    |1:
-+    |  ld CARG1, NODE:TMP2->key
-+    |    ld CARG2, NODE:TMP2->val
-+    |  checkstr CARG1; bne >4
-+    |   cmpd CARG1, STR:RC; bne >4
-+    |    checknil_noclear CARG2; beq >5	// Key found, but nil value?
-+    |3:
-+    |    stdx CARG2, RA, BASE
-+    |  ins_next
-+    |
-+    |4:  // Follow hash chain.
-+    |  ld NODE:TMP2, NODE:TMP2->next
-+    |  cmpldi NODE:TMP2, 0
-+    |  bne <1
-+    |  // End of hash chain: key not found, nil result.
-+    |   li CARG2, LJ_TNIL
-+    |
-+    |5:  // Check for __index if table value is nil.
-+    |  ld TAB:TMP2, TAB:RB->metatable
-+    |  cmpldi TAB:TMP2, 0
-+    |  beq <3				// No metatable: done.
-+    |  lbz TMP0, TAB:TMP2->nomm
-+    |  andi. TMP0, TMP0, 1<<MM_index
-+    |  bne <3				// 'no __index' flag set: done.
-+    |  b ->vmeta_tgets
-+    break;
-+  case BC_TGETB:
-+    |  // RA = dst*8, RB = table*8, RC = index*8
-+    |  ldux CARG1, RB, BASE
-+    |   srdi TMP0, RC, 3
-+    |   ld TAB:RB, 8(RB)
-+    |  checktab CARG1; bne ->vmeta_tgetb
-+    |  lwz TMP1, TAB:CARG1->asize
-+    |   ld TMP2, TAB:CARG1->array
-+    |  cmpld TMP0, TMP1; bge ->vmeta_tgetb
-+    |  ldx TMP1, TMP2, RC
-+    |   lfdx f0, TMP2, RC
-+    |  checknil_noclear TMP1; beq >5
-+    |1:
-+    |  ins_next1
-+    |   stfdx f0, BASE, RA
-+    |  ins_next2
-+    |
-+    |5:  // Check for __index if table value is nil.
-+    |  ld TAB:TMP2, TAB:CARG1->metatable
-+    |  cmpldi TAB:TMP2, 0
-+    |  beq <1				// No metatable: done.
-+    |  lbz TMP2, TAB:TMP2->nomm
-+    |  andi. TMP2, TMP2, 1<<MM_index
-+    |  bne <1				// 'no __index' flag set: done.
-+    |  b ->vmeta_tgetb			// Caveat: preserve TMP0!
-+    break;
-+  case BC_TGETR:
-+    |  // RA = dst*8, RB = table*8, RC = key*8
-+    |  ldx TAB:CARG1, BASE, RB
-+    |  clear_field TAB:CARG1
-+    |  lwz TMP0, TAB:CARG1->asize
-+    |   ldx CARG2, BASE, RC
-+    |   clear_field TAB:CARG2
-+    |   ld TMP1, TAB:CARG1->array
-+    |  cmpld TMP0, CARG2
-+    |   sldi TMP2, CARG2, 3
-+    |  ble ->vmeta_tgetr		// In array part?
-+    |   lfdx f14, TMP1, TMP2
-+    |->BC_TGETR_Z:
-+    |  ins_next1
-+    |   stfdx f14, BASE, RA
-+    |  ins_next2
-+    break;
-+
-+  case BC_TSETV:
-+    |  // RA = src*8, RB = table*8, RC = key*8
-+    |  ldux CARG1, RB, BASE
-+    |  ldux CARG2, RC, BASE
-+    |  checktab CARG1
-+    |  bne ->vmeta_tsetv
-+    |   checknum_noclear CARG2
-+    |   bne  >5
-+    |  lwz TMP0, TAB:CARG1->asize
-+    |  clear_field CARG2
-+    |   ld TMP1, TAB:CARG1->array
-+    |  cmpld TMP0, CARG2
-+    |   sldi TMP0, CARG2, 3
-+    |  ble ->vmeta_tsetv		// Integer key and in array part?
-+    |   ldx TMP2, TMP1, TMP0
-+    |  lbz TMP3, TAB:CARG1->marked
-+    |    lfdx f14, BASE, RA
-+    |   checknil_noclear TMP2; beq >3
-+    |1:
-+    |  andi. TMP2, TMP3, LJ_GC_BLACK	// isblack(table)
-+    |    stfdx f14, TMP1, TMP0
-+    |  bne >7
-+    |2:
-+    |  ins_next
-+    |
-+    |3:  // Check for __newindex if previous value is nil.
-+    |  ld TAB:TMP2, TAB:CARG1->metatable
-+    |  cmpldi TAB:TMP2, 0
-+    |  beq <1				// No metatable: done.
-+    |  lbz TMP2, TAB:TMP2->nomm
-+    |  andi. TMP2, TMP2, 1<<MM_newindex
-+    |  bne <1				// 'no __newindex' flag set: done.
-+    |  b ->vmeta_tsetv
-+    |
-+    |5:
-+    |  checkstr CARG2; bne ->vmeta_tsetv
-+    |  mr RC, CARG2
-+    |  ld TAB:RB, 0(RB)
-+    |    clear_field TAB:RB
-+    |  b ->BC_TSETS_Z			// String key?
-+    |
-+    |7:  // Possible table write barrier for the value. Skip valiswhite check.
-+    |  barrierback TAB:CARG1, TMP3, TMP0
-+    |  b <2
-+    break;
-+  case BC_TSETS:
-+    |  // RA = src*8, RB = table*8, RC = str_const*8 (~)
-+    |  ldux CARG1, RB, BASE
-+    |    ld TAB:RB, 0(RB)
-+    |    clear_field TAB:RB
-+    |   subfic TMP1, RC, -8
-+    |  checktab CARG1
-+    |   ldx STR:RC, KBASE, TMP1		// KBASE-8-str_const*8
-+    |  bne ->vmeta_tsets1
-+    |->BC_TSETS_Z:
-+    |  // TAB:RB = GCtab *, STR:RC = GCstr *, RA = src*8
-+    |  lwz TMP0, TAB:RB->hmask
-+    |  lwz TMP1, STR:RC->hash
-+    |  ld NODE:TMP2, TAB:RB->node
-+    |    stb ZERO, TAB:RB->nomm		// Clear metamethod cache.
-+    |  and TMP1, TMP1, TMP0		// idx = str->hash & tab->hmask
-+    |    lfdx f14, BASE, RA
-+    |  sldi TMP0, TMP1, 5
-+    |  sldi TMP1, TMP1, 3
-+    |  sub TMP1, TMP0, TMP1
-+    |    lbz TMP3, TAB:RB->marked
-+    |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
-+    |1:
-+    |  ld CARG1, NODE:TMP2->key
-+    |    ld CARG2, NODE:TMP2->val
-+    |     ld NODE:TMP1, NODE:TMP2->next
-+    |  checkstr CARG1; bne >5
-+    |   cmpd CARG1, STR:RC; bne >5
-+    |    checknil_noclear CARG2; beq >4	// Key found, but nil value?
-+    |2:
-+    |  andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
-+    |    stfd f14, NODE:TMP2->val
-+    |  bne >7
-+    |3:
-+    |  ins_next
-+    |
-+    |4:  // Check for __newindex if previous value is nil.
-+    |  ld TAB:TMP1, TAB:RB->metatable
-+    |  cmpldi TAB:TMP1, 0
-+    |  beq <2				// No metatable: done.
-+    |  lbz TMP0, TAB:TMP1->nomm
-+    |  andi. TMP0, TMP0, 1<<MM_newindex
-+    |  bne <2				// 'no __newindex' flag set: done.
-+    |  b ->vmeta_tsets
-+    |
-+    |5:  // Follow hash chain.
-+    |  cmpldi NODE:TMP1, 0
-+    |   mr NODE:TMP2, NODE:TMP1
-+    |  bne <1
-+    |  // End of hash chain: key not found, add a new one.
-+    |
-+    |  // But check for __newindex first.
-+    |  ld TAB:TMP1, TAB:RB->metatable
-+    |   la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
-+    |   std PC, SAVE_PC
-+    |   mr CARG1, L
-+    |  cmpldi TAB:TMP1, 0
-+    |   std BASE, L->base
-+    |  beq >6				// No metatable: continue.
-+    |  lbz TMP0, TAB:TMP1->nomm
-+    |  andi. TMP0, TMP0, 1<<MM_newindex
-+    |  beq ->vmeta_tsets		// 'no __newindex' flag NOT set: check.
-+    |6:
-+    |  li TMP0, LJ_TSTR
-+    |  set_oper_type TMP0, TMP0
-+    |  add TMP0, TMP0, STR:RC
-+    |   mr CARG2, TAB:RB
-+    |  std TMP0, 0(CARG3)
-+    |  bl extern lj_tab_newkey		// (lua_State *L, GCtab *t, TValue *k)
-+    |  // Returns TValue *.
-+    |  ld BASE, L->base
-+    |  stfd f14, 0(CRET1)
-+    |  b <3				// No 2nd write barrier needed.
-+    |
-+    |7:  // Possible table write barrier for the value. Skip valiswhite check.
-+    |  barrierback TAB:RB, TMP3, TMP0
-+    |  b <3
-+    break;
-+  case BC_TSETB:
-+    |  // RA = src*8, RB = table*8, RC = index*8
-+    |  ldux CARG1, RB, BASE
-+    |   srdi TMP0, RC, 3
-+    |   ld TAB:RB, 8(RB)
-+    |  checktab CARG1; bne ->vmeta_tsetb
-+    |  lwz TMP1, TAB:CARG1->asize
-+    |   ld TMP2, TAB:CARG1->array
-+    |    lbz TMP3, TAB:CARG1->marked
-+    |  cmpld TMP0, TMP1
-+    |   lfdx f14, BASE, RA
-+    |  bge ->vmeta_tsetb
-+    |  ldx TMP1, TMP2, RC
-+    |  checknil_noclear TMP1; beq >5
-+    |1:
-+    |  andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
-+    |   stfdx f14, TMP2, RC
-+    |  bne >7
-+    |2:
-+    |  ins_next
-+    |
-+    |5:  // Check for __newindex if previous value is nil.
-+    |  ld TAB:TMP1, TAB:CARG1->metatable
-+    |  cmpldi TAB:TMP1, 0
-+    |  beq <1				// No metatable: done.
-+    |  lbz TMP1, TAB:TMP1->nomm
-+    |  andi. TMP1, TMP1, 1<<MM_newindex
-+    |  bne <1				// 'no __newindex' flag set: done.
-+    |  b ->vmeta_tsetb			// Caveat: preserve TMP0!
-+    |
-+    |7:  // Possible table write barrier for the value. Skip valiswhite check.
-+    |  barrierback TAB:CARG1, TMP3, TMP0
-+    |  b <2
-+    break;
-+  case BC_TSETR:
-+    |  // RA = dst*8, RB = table*8, RC = key*8
-+    |  add RB, BASE, RB
-+    |  ld TAB:CARG2, 0(RB)
-+    |  clear_field TAB:CARG2
-+    |  add RC, BASE, RC
-+    |    lbz TMP3, TAB:CARG2->marked
-+    |  lwz TMP0, TAB:CARG2->asize
-+    |  ld CARG3, 0(RC)
-+    |  clear_field CARG3
-+    |   ld TMP1, TAB:CARG2->array
-+    |  andi. TMP2, TMP3, LJ_GC_BLACK	// isblack(table)
-+    |  bne >7
-+    |2:
-+    |  cmpld TMP0, CARG3
-+    |   sldi TMP2, CARG3, 3
-+    |   lfdx f14, BASE, RA
-+    |  ble ->vmeta_tsetr		// In array part?
-+    |  ins_next1
-+    |   stfdx f14, TMP1, TMP2
-+    |  ins_next2
-+    |
-+    |7:  // Possible table write barrier for the value. Skip valiswhite check.
-+    |  barrierback TAB:CARG2, TMP3, TMP2
-+    |  b <2
-+    break;
-+
-+
-+  case BC_TSETM:
-+    |  // RA = base*8 (table at base-1), RD = num_const*8 (start index)
-+    |  add RA, BASE, RA
-+    |1:
-+    |   add TMP3, KBASE, RD
-+    |  ld TAB:CARG2, -8(RA)		// Guaranteed to be a table.
-+    |  clear_field TAB:CARG2
-+    |    addic. TMP0, MULTRES, -8
-+    |   ld TMP3, 0(TMP3)		// Integer constant needs to be cleaned
-+    |    srdi CARG3, TMP0, 3
-+    |   clear_field TMP3
-+    |    beq >4				// Nothing to copy?
-+    |  add CARG3, CARG3, TMP3
-+    |  lwz TMP2, TAB:CARG2->asize
-+    |   sldi TMP1, TMP3, 3
-+    |    lbz TMP3, TAB:CARG2->marked
-+    |  cmpld CARG3, TMP2
-+    |   add TMP2, RA, TMP0
-+    |   ld TMP0, TAB:CARG2->array
-+    |  bgt >5
-+    |   add TMP1, TMP1, TMP0
-+    |    andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
-+    |3:  // Copy result slots to table.
-+    |   lfd f0, 0(RA)
-+    |  addi RA, RA, 8
-+    |  cmpd cr1, RA, TMP2
-+    |   stfd f0, 0(TMP1)
-+    |    addi TMP1, TMP1, 8
-+    |  blt cr1, <3
-+    |  bne >7
-+    |4:
-+    |  ins_next
-+    |
-+    |5:  // Need to resize array part.
-+    |   std BASE, L->base
-+    |  mr CARG1, L
-+    |   std PC, SAVE_PC
-+    |  mr SAVE0, RD
-+    |  bl extern lj_tab_reasize		// (lua_State *L, GCtab *t, int nasize)
-+    |  // Must not reallocate the stack.
-+    |  mr RD, SAVE0
-+    |  b <1
-+    |
-+    |7:  // Possible table write barrier for any value. Skip valiswhite check.
-+    |  barrierback TAB:CARG2, TMP3, TMP0
-+    |  b <4
-+    break;
-+
-+  /* -- Calls and vararg handling ----------------------------------------- */
-+
-+  case BC_CALLM:
-+    |  // RA = base*8, (RB = (nresults+1)*8,) RC = extra_nargs*8
-+    |  add NARGS8:RC, NARGS8:RC, MULTRES
-+    |  // Fall through. Assumes BC_CALL follows.
-+    break;
-+  case BC_CALL:
-+    |  // RA = base*8, (RB = (nresults+1)*8,) RC = (nargs+1)*8
-+    |  mr TMP2, BASE			// needed on vmeta_call!
-+    |  ldux LFUNC:RB, BASE, RA
-+    |    subi NARGS8:RC, NARGS8:RC, 8
-+    |   addi BASE, BASE, 16
-+    |  checkfunc RB; bne ->vmeta_call
-+    |  ins_call
-+    break;
-+
-+  case BC_CALLMT:
-+    |  // RA = base*8, (RB = 0,) RC = extra_nargs*8
-+    |  add NARGS8:RC, NARGS8:RC, MULTRES
-+    |  // Fall through. Assumes BC_CALLT follows.
-+    break;
-+  case BC_CALLT:
-+    |  // RA = base*8, (RB = 0,) RC = (nargs+1)*8
-+    |  ldux TMP0, RA, BASE
-+    |   ld LFUNC:RB, 0(RA)
-+    |   clear_field RB
-+    |    subi NARGS8:RC, NARGS8:RC, 8
-+    |    ld TMP1, FRAME_PC(BASE)
-+    |  checkfunc TMP0
-+    |   addi RA, RA, 16
-+    |  bne ->vmeta_callt
-+    |->BC_CALLT_Z:
-+    |  andi. TMP0, TMP1, FRAME_TYPE	// Caveat: preserve cr0 until the crand.
-+    |   lbz TMP3, LFUNC:RB->ffid
-+    |    xori TMP2, TMP1, FRAME_VARG
-+    |    cmpldi cr1, NARGS8:RC, 0
-+    |  bne >7
-+    |1:
-+    |  std LFUNC:RB, FRAME_FUNC(BASE)	// Copy function down, but keep PC.
-+    |  li TMP2, 0
-+    |   cmpldi cr7, TMP3, 1		// (> FF_C) Calling a fast function?
-+    |    beq cr1, >3
-+    |2:
-+    |  addi TMP3, TMP2, 8
-+    |   lfdx f0, RA, TMP2
-+    |  cmpld cr1, TMP3, NARGS8:RC
-+    |   stfdx f0, BASE, TMP2
-+    |  mr TMP2, TMP3
-+    |  bne cr1, <2
-+    |3:
-+    |  crand 4*cr0+eq, 4*cr0+eq, 4*cr7+gt
-+    |  beq >5
-+    |4:
-+    |  ins_callt
-+    |
-+    |5:  // Tailcall to a fast function with a Lua frame below.
-+    |  lwz INS, -4(TMP1)
-+    |  decode_RA8 RA, INS
-+    |  sub TMP1, BASE, RA
-+    |  ld LFUNC:TMP1, FRAME_FUNC-16(TMP1)
-+    |  clear_field TMP1
-+    |  ld TMP1, LFUNC:TMP1->pc
-+    |  ld KBASE, PC2PROTO(k)(TMP1)	// Need to prepare KBASE.
-+    |  b <4
-+    |
-+    |7:  // Tailcall from a vararg function.
-+    |  andi. TMP0, TMP2, FRAME_TYPEP
-+    |  bne <1				// Vararg frame below?
-+    |  sub BASE, BASE, TMP2		// Relocate BASE down.
-+    |  ld TMP1, FRAME_PC(BASE)
-+    |  andi. TMP0, TMP1, FRAME_TYPE
-+    |  b <1
-+    break;
-+
-+  case BC_ITERC:
-+    |  // RA = base*8, (RB = (nresults+1)*8, RC = (nargs+1)*8 ((2+1)*8))
-+    |  mr TMP2, BASE
-+    |  add BASE, BASE, RA
-+    |  ld LFUNC:RB, -24(BASE)
-+    |   ld CARG1, -16(BASE)
-+    |   ld CARG2, -8(BASE)
-+    |  std LFUNC:RB, 0(BASE)		// Copy callable.
-+    |  checkfunc LFUNC:RB
-+    |     li NARGS8:RC, 16		// Iterators get 2 arguments.
-+    |    std CARG2, 24(BASE)		// Copy state.
-+    |    stdu CARG1, 16(BASE)		// Copy control var.
-+    |  bne ->vmeta_call
-+    |  ins_call
-+    break;
-+
-+  case BC_ITERN:
-+    |  // RA = base*8, (RB = (nresults+1)*8, RC = (nargs+1)*8 (2+1)*8)
-+    |.if JIT
-+    |  // NYI: add hotloop, record BC_ITERN.
-+    |.endif
-+    |  add RA, BASE, RA
-+    |  ld TAB:RB, -16(RA)
-+    |  clear_field TAB:RB
-+    |  lwz RC, -8(RA)			// Get index from control var.
-+    |  lwz TMP0, TAB:RB->asize
-+    |  ld TMP1, TAB:RB->array
-+    |   addi PC, PC, 4
-+    |1:  // Traverse array part.
-+    |  cmpld RC, TMP0
-+    |   sldi TMP3, RC, 3
-+    |  bge >5				// Index points after array part?
-+    |  ldx TMP2, TMP1, TMP3
-+    |   lfdx f0, TMP1, TMP3
-+    |  checknil_noclear TMP2
-+    |     lwz INS, -4(PC)
-+    |  beq >4
-+    |   add RC, RC, TISNUM
-+    |   std RC, 0(RA)
-+    |    addi RC, RC, 1
-+    |     addis TMP3, PC, -(BCBIAS_J*4 >> 16)
-+    |  stfd f0, 8(RA)
-+    |     decode_RD4 TMP1, INS
-+    |    std RC, -8(RA)			// Update control var.
-+    |     add PC, TMP1, TMP3
-+    |3:
-+    |  ins_next
-+    |
-+    |4:  // Skip holes in array part.
-+    |  addi RC, RC, 1
-+    |  b <1
-+    |
-+    |5:  // Traverse hash part.
-+    |  lwz TMP1, TAB:RB->hmask
-+    |  sub RC, RC, TMP0
-+    |   ld TMP2, TAB:RB->node
-+    |6:
-+    |  cmpld RC, TMP1			// End of iteration? Branch to ITERL+1.
-+    |   sldi TMP3, RC, 5
-+    |  bgty <3
-+    |   sldi RB, RC, 3
-+    |   sub TMP3, TMP3, RB
-+    |  ldx RB, TMP2, TMP3
-+    |  lfdx f0, TMP2, TMP3
-+    |   add NODE:TMP3, TMP2, TMP3
-+    |  checknil_noclear RB
-+    |     lwz INS, -4(PC)
-+    |  beq >7
-+    |   lfd f1, NODE:TMP3->key
-+    |     addis TMP2, PC, -(BCBIAS_J*4 >> 16)
-+    |  stfd f0, 8(RA)
-+    |    add RC, RC, TMP0
-+    |     decode_RD4 TMP1, INS
-+    |   stfd f1, 0(RA)
-+    |    addi RC, RC, 1
-+    |     add PC, TMP1, TMP2
-+    |    std RC, -8(RA)			// Update control var.
-+    |  b <3
-+    |
-+    |7:  // Skip holes in hash part.
-+    |  addi RC, RC, 1
-+    |  b <6
-+    break;
-+
-+  case BC_ISNEXT:
-+    |  // RA = base*8, RD = target (points to ITERN)
-+    |  add RA, BASE, RA
-+    |  ld TMP1, -24(RA)
-+    |  get_oper_type TMP0, TMP1
-+    |  cmpdi cr1, TMP0, LJ_TFUNC
-+    |   ld TMP2, -16(RA)
-+    |   get_oper_type TMP2, TMP2
-+    |   cmpdi cr0, TMP2, LJ_TTAB
-+    |    ld TMP3, -8(RA)
-+    |    get_oper_type TMP3, TMP3
-+    |    cmpdi cr6, TMP3, LJ_TNIL
-+    |
-+    |  bne cr1, >5
-+    |
-+    |  clear_field TMP1
-+    |  lbz TMP1, CFUNC:TMP1->ffid
-+    |   crand 4*cr0+eq, 4*cr0+eq, 4*cr6+eq
-+    |  cmpwi cr7, TMP1, FF_next_N
-+    |    srdi TMP0, RD, 1
-+    |  crand 4*cr0+eq, 4*cr0+eq, 4*cr7+eq
-+    |    add TMP3, PC, TMP0
-+    |  bne cr0, >5
-+    |  lus TMP1, 0xfffe
-+    |  ori TMP1, TMP1, 0x7fff
-+    |  sldi TMP1, TMP1, 32		// Initialize control var.
-+    |  std TMP1, -8(RA)
-+    |    addis PC, TMP3, -(BCBIAS_J*4 >> 16)
-+    |1:
-+    |  ins_next
-+    |5:  // Despecialize bytecode if any of the checks fail.
-+    |  li TMP0, BC_JMP
-+    |   li TMP1, BC_ITERC
-+    |  stb TMP0, -4(PC)
-+    |    addis PC, TMP3, -(BCBIAS_J*4 >> 16)
-+    |   stb TMP1, 0(PC)
-+    |  b <1
-+    break;
-+
-+  case BC_VARG:
-+    |  // RA = base*8, RB = (nresults+1)*8, RC = numparams*8
-+    |  ld TMP0, FRAME_PC(BASE)
-+    |  add RC, BASE, RC
-+    |   add RA, BASE, RA
-+    |  addi RC, RC, FRAME_VARG
-+    |   add TMP2, RA, RB
-+    |  subi TMP3, BASE, 16		// TMP3 = vtop
-+    |  sub RC, RC, TMP0			// RC = vbase
-+    |  // Note: RC may now be even _above_ BASE if nargs was < numparams.
-+    |  cmpldi cr1, RB, 0
-+    |   sub. TMP1, TMP3, RC
-+    |  beq cr1, >5			// Copy all varargs?
-+    |   subi TMP2, TMP2, 16
-+    |   ble >2				// No vararg slots?
-+    |1:  // Copy vararg slots to destination slots.
-+    |  lfd f0, 0(RC)
-+    |   addi RC, RC, 8
-+    |  stfd f0, 0(RA)
-+    |  cmpld RA, TMP2
-+    |   cmpld cr1, RC, TMP3
-+    |  bge >3				// All destination slots filled?
-+    |    addi RA, RA, 8
-+    |   blt cr1, <1			// More vararg slots?
-+    |2:  // Fill up remainder with nil.
-+    |  std TISNIL, 0(RA)
-+    |  cmpld RA, TMP2
-+    |   addi RA, RA, 8
-+    |  blt <2
-+    |3:
-+    |  ins_next
-+    |
-+    |5:  // Copy all varargs.
-+    |  ld TMP0, L->maxstack
-+    |   li MULTRES, 8			// MULTRES = (0+1)*8
-+    |  bley <3				// No vararg slots?
-+    |  add TMP2, RA, TMP1
-+    |  cmpld TMP2, TMP0
-+    |   addi MULTRES, TMP1, 8
-+    |  bgt >7
-+    |6:
-+    |  lfd f0, 0(RC)
-+    |   addi RC, RC, 8
-+    |  stfd f0, 0(RA)
-+    |  cmpld RC, TMP3
-+    |   addi RA, RA, 8
-+    |  blt <6				// More vararg slots?
-+    |  b <3
-+    |
-+    |7:  // Grow stack for varargs.
-+    |  mr CARG1, L
-+    |   std RA, L->top
-+    |  sub SAVE0, RC, BASE		// Need delta, because BASE may change.
-+    |   std BASE, L->base
-+    |  sub RA, RA, BASE
-+    |   std PC, SAVE_PC
-+    |  srdi CARG2, TMP1, 3
-+    |  bl extern lj_state_growstack	// (lua_State *L, int n)
-+    |  ld BASE, L->base
-+    |  add RA, BASE, RA
-+    |  add RC, BASE, SAVE0
-+    |  subi TMP3, BASE, 16
-+    |  b <6
-+    break;
-+
-+  /* -- Returns ----------------------------------------------------------- */
-+
-+  case BC_RETM:
-+    |  // RA = results*8, RD = extra_nresults*8
-+    |  add RD, RD, MULTRES		// MULTRES >= 8, so RD >= 8.
-+    |  // Fall through. Assumes BC_RET follows.
-+    break;
-+
-+  case BC_RET:
-+    |  // RA = results*8, RD = (nresults+1)*8
-+    |  ld PC, FRAME_PC(BASE)
-+    |   add RA, BASE, RA
-+    |    mr MULTRES, RD
-+    |1:
-+    |  andi. TMP0, PC, FRAME_TYPE
-+    |   xori TMP1, PC, FRAME_VARG
-+    |  bne ->BC_RETV_Z
-+    |
-+    |->BC_RET_Z:
-+    |  // BASE = base, RA = resultptr, RD = (nresults+1)*8, PC = return
-+    |   lwz INS, -4(PC)
-+    |  cmpdi RD, 8
-+    |   subi TMP2, BASE, 16
-+    |   subi RC, RD, 8
-+    |   decode_RB8 RB, INS
-+    |  beq >3
-+    |   li TMP1, 0
-+    |2:
-+    |  addi TMP3, TMP1, 8
-+    |   lfdx f0, RA, TMP1
-+    |  cmpd TMP3, RC
-+    |   stfdx f0, TMP2, TMP1
-+    |  beq >3
-+    |  addi TMP1, TMP3, 8
-+    |   lfdx f1, RA, TMP3
-+    |  cmpd TMP1, RC
-+    |   stfdx f1, TMP2, TMP3
-+    |  bne <2
-+    |3:
-+    |5:
-+    |  cmpld RB, RD
-+    |   decode_RA8 RA, INS
-+    |  bgt >6
-+    |   sub BASE, TMP2, RA
-+    |  ld LFUNC:TMP1, FRAME_FUNC(BASE)
-+    |  clear_field TMP1
-+    |  ins_next1
-+    |  ld TMP1, LFUNC:TMP1->pc
-+    |  ld KBASE, PC2PROTO(k)(TMP1)
-+    |  ins_next2
-+    |
-+    |6:  // Fill up results with nil.
-+    |  subi TMP1, RD, 8
-+    |   addi RD, RD, 8
-+    |  stdx TISNIL, TMP2, TMP1
-+    |  b <5
-+    |
-+    |->BC_RETV_Z:  // Non-standard return case.
-+    |  andi. TMP2, TMP1, FRAME_TYPEP
-+    |  bne ->vm_return
-+    |  // Return from vararg function: relocate BASE down.
-+    |  sub BASE, BASE, TMP1
-+    |  ld PC, FRAME_PC(BASE)
-+    |  b <1
-+    break;
-+
-+  case BC_RET0: case BC_RET1:
-+    |  // RA = results*8, RD = (nresults+1)*8
-+    |  ld PC, FRAME_PC(BASE)
-+    |   add RA, BASE, RA
-+    |    mr MULTRES, RD
-+    |  andi. TMP0, PC, FRAME_TYPE
-+    |   xori TMP1, PC, FRAME_VARG
-+    |  bney ->BC_RETV_Z
-+    |
-+    |  lwz INS, -4(PC)
-+    |   subi TMP2, BASE, 16
-+    |  decode_RB8 RB, INS
-+    if (op == BC_RET1) {
-+      |  lfd f0, 0(RA)
-+      |  stfd f0, 0(TMP2)
-+    }
-+    |5:
-+    |  cmpld RB, RD
-+    |   decode_RA8 RA, INS
-+    |  bgt >6
-+    |   sub BASE, TMP2, RA
-+    |   ld LFUNC:TMP1, FRAME_FUNC(BASE)
-+    |   clear_field TMP1
-+    |  ins_next1
-+    |  ld TMP1, LFUNC:TMP1->pc
-+    |  ld KBASE, PC2PROTO(k)(TMP1)
-+    |  ins_next2
-+    |
-+    |6:  // Fill up results with nil.
-+    |  subi TMP1, RD, 8
-+    |   addi RD, RD, 8
-+    |  stdx TISNIL, TMP2, TMP1
-+    |  b <5
-+    break;
-+
-+  /* -- Loops and branches ------------------------------------------------ */
-+
-+  case BC_FORL:
-+    |.if JIT
-+    |  hotloop
-+    |.endif
-+    |  // Fall through. Assumes BC_IFORL follows.
-+    break;
-+
-+  case BC_JFORI:
-+  case BC_JFORL:
-+#if !LJ_HASJIT
-+    break;
-+#endif
-+  case BC_FORI:
-+  case BC_IFORL:
-+    |  // RA = base*8, RD = target (after end of loop or start of loop)
-+    vk = (op == BC_IFORL || op == BC_JFORL);
-+    |  // Integer loop.
-+    |  ldux CARG1, RA, BASE
-+    |   get_oper_type TMP1, CARG1
-+    |   checknum cr0, CARG1
-+    if (vk) {
-+      |   ld CARG3, FORL_STEP*8(RA)
-+      |   checknum CARG3
-+      |  bne >9
-+      |  // Need to check overflow for (a<<32) + (b<<32).
-+      |  rldicr TMP0, CARG1, 32, 31
-+      |  rldicr TMP2, CARG3, 32, 31
-+      |  add CARG1, CARG1, CARG3
-+      |  addo. TMP0, TMP0, TMP2
-+      |    cmpdi cr6, CARG3, 0
-+      |   ld CARG2, FORL_STOP*8(RA)
-+      |    checknum CARG2
-+      |  bso >6
-+      |4:
-+      |  std CARG1, FORL_IDX*8(RA)
-+      |   checknum CARG1
-+    } else {
-+      |  ld CARG3, FORL_STEP*8(RA)
-+      |    get_oper_type TMP3, CARG3
-+      |    checknum cr7, CARG3
-+      |  ld CARG2, FORL_STOP*8(RA)
-+      |    get_oper_type TMP2, CARG2
-+      |    checknum cr1, CARG2
-+      |  crand 4*cr0+eq, 4*cr0+eq, 4*cr7+eq
-+      |  crand 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
-+      |    cmpdi cr6, CARG3, 0
-+      |  bne >9
-+    }
-+    |    blt cr6, >5
-+    |  cmpd CARG1, CARG2
-+    |1:
-+    if (op != BC_JFORL) {
-+      |  srdi RD, RD, 1
-+    }
-+    |   add_oper_type CARG1, TISNUM
-+    |   std CARG1, FORL_EXT*8(RA)
-+    if (op != BC_JFORL) {
-+      |  add RD, PC, RD
-+    }
-+    if (op == BC_FORI) {
-+      |  bgt >3  // See FP loop below.
-+    } else if (op == BC_JFORI) {
-+      |  addis PC, RD, -(BCBIAS_J*4 >> 16)
-+      |  bley >7
-+    } else if (op == BC_IFORL) {
-+      |  bgt >2
-+      |  addis PC, RD, -(BCBIAS_J*4 >> 16)
-+    } else {
-+      |  bley =>BC_JLOOP
-+    }
-+    |2:
-+    |  ins_next
-+    |5:  // Invert check for negative step.
-+    |  cmpd CARG2, CARG1
-+    |  b <1
-+    if (vk) {
-+      |6:  // Potential overflow.
-+      |  checkov <4			// Ignore unrelated overflow.
-+      |  b <2
-+    }
-+    if (vk) {
-+      |9:  // FP loop.
-+      |  lfd f1, FORL_IDX*8(RA)
-+      |  lfd f3, FORL_STEP*8(RA)
-+      |  lfd f2, FORL_STOP*8(RA)
-+      |   ld TMP3, FORL_STEP*8(RA)
-+      |  fadd f1, f1, f3
-+      |  stfd f1, FORL_IDX*8(RA)
-+    } else {
-+      |9:  // FP loop.
-+      |   lfd f1, FORL_IDX*8(RA)
-+      |  crand 4*cr0+lt, 4*cr0+lt, 4*cr7+lt
-+      |  crand 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
-+      |   lfd f2, FORL_STOP*8(RA)
-+      |  bge ->vmeta_for
-+    }
-+    |  cmpdi cr6, TMP3, 0
-+    if (op != BC_JFORL) {
-+      |  srdi RD, RD, 1
-+    }
-+    |   stfd f1, FORL_EXT*8(RA)
-+    if (op != BC_JFORL) {
-+      |  add RD, PC, RD
-+    }
-+    |  fcmpu cr0, f1, f2
-+    if (op == BC_JFORI) {
-+      |  addis PC, RD, -(BCBIAS_J*4 >> 16)
-+    }
-+    |  blt cr6, >5
-+    if (op == BC_FORI) {
-+      |  bgt >3
-+    } else if (op == BC_IFORL) {
-+      |  bgty <2
-+      |1:
-+      |  addis PC, RD, -(BCBIAS_J*4 >> 16)
-+    } else if (op == BC_JFORI) {
-+      |  bley >7
-+    } else {
-+      |  bley =>BC_JLOOP
-+    }
-+    |  b <2
-+    |5:  // Negative step.
-+    if (op == BC_FORI) {
-+      |  bge <2
-+      |3:  // Used by integer loop, too.
-+      |  addis PC, RD, -(BCBIAS_J*4 >> 16)
-+    } else if (op == BC_IFORL) {
-+      |  bgey <1
-+    } else if (op == BC_JFORI) {
-+      |  bgey >7
-+    } else {
-+      |  bgey =>BC_JLOOP
-+    }
-+    |  b <2
-+    if (op == BC_JFORI) {
-+      |7:
-+      |  lwz INS, -4(PC)
-+      |  decode_RD8 RD, INS
-+      |  b =>BC_JLOOP
-+    }
-+    break;
-+
-+  case BC_ITERL:
-+    |.if JIT
-+    |  hotloop
-+    |.endif
-+    |  // Fall through. Assumes BC_IITERL follows.
-+    break;
-+
-+  case BC_JITERL:
-+#if !LJ_HASJIT
-+    break;
-+#endif
-+  case BC_IITERL:
-+    |  // RA = base*8, RD = target
-+    |  ldux TMP1, RA, BASE
-+    |  checknil_noclear TMP1; beq >1	// Stop if iterator returned nil.
-+    if (op == BC_JITERL) {
-+      |   std TMP1, -8(RA)
-+      |  b =>BC_JLOOP
-+    } else {
-+      |  branch_RD			// Otherwise save control var + branch.
-+      |   std TMP1, -8(RA)
-+    }
-+    |1:
-+    |  ins_next
-+    break;
-+
-+  case BC_LOOP:
-+    |  // RA = base*8, RD = target (loop extent)
-+    |  // Note: RA/RD is only used by trace recorder to determine scope/extent
-+    |  // This opcode does NOT jump, it's only purpose is to detect a hot loop.
-+    |.if JIT
-+    |  hotloop
-+    |.endif
-+    |  // Fall through. Assumes BC_ILOOP follows.
-+    break;
-+
-+  case BC_ILOOP:
-+    |  // RA = base*8, RD = target (loop extent)
-+    |  ins_next
-+    break;
-+
-+  case BC_JLOOP:
-+    |.if JIT
-+    |  NYI
-+    |.endif
-+    break;
-+
-+  case BC_JMP:
-+    |  // RA = base*8 (only used by trace recorder), RD = target
-+    |  branch_RD
-+    |  ins_next
-+    break;
-+
-+  /* -- Function headers -------------------------------------------------- */
-+
-+  case BC_FUNCF:
-+    |.if JIT
-+    |  hotcall
-+    |.endif
-+  case BC_FUNCV:  /* NYI: compiled vararg functions. */
-+    |  // Fall through. Assumes BC_IFUNCF/BC_IFUNCV follow.
-+    break;
-+
-+  case BC_JFUNCF:
-+#if !LJ_HASJIT
-+    break;
-+#endif
-+  case BC_IFUNCF:
-+    |  // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8
-+    |  ld TMP2, L->maxstack
-+    |   lbz TMP1, -4+PC2PROTO(numparams)(PC)
-+    |    ld KBASE, -4+PC2PROTO(k)(PC)
-+    |  cmpld RA, TMP2
-+    |   sldi TMP1, TMP1, 3
-+    |  bgt ->vm_growstack_l
-+    if (op != BC_JFUNCF) {
-+      |  ins_next1
-+    }
-+    |2:
-+    |  cmpld NARGS8:RC, TMP1		// Check for missing parameters.
-+    |  blt >3
-+    if (op == BC_JFUNCF) {
-+      |  decode_RD8 RD, INS
-+      |  b =>BC_JLOOP
-+    } else {
-+      |  ins_next2
-+    }
-+    |
-+    |3:  // Clear missing parameters.
-+    |  stdx TISNIL, BASE, NARGS8:RC
-+    |  addi NARGS8:RC, NARGS8:RC, 8
-+    |  b <2
-+    break;
-+
-+  case BC_JFUNCV:
-+#if !LJ_HASJIT
-+    break;
-+#endif
-+    |  NYI  // NYI: compiled vararg functions
-+    break;  /* NYI: compiled vararg functions. */
-+
-+  case BC_IFUNCV:
-+    |  // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8
-+    |  ld TMP2, L->maxstack
-+    |   add TMP1, BASE, RC
-+    |  add RA, RA, RC
-+    |   std LFUNC:RB, 0(TMP1)		// Store (untagged) copy of LFUNC.
-+    |   addi TMP3, RC, 16+FRAME_VARG
-+    |    ld KBASE, -4+PC2PROTO(k)(PC)
-+    |  cmpld RA, TMP2
-+    |   std TMP3, 8(TMP1)		// Store delta + FRAME_VARG.
-+    |  bge ->vm_growstack_l
-+    |  lbz TMP2, -4+PC2PROTO(numparams)(PC)
-+    |   mr RA, BASE
-+    |   mr RC, TMP1
-+    |   addi TMP1, TMP1, 16
-+    |  ins_next1
-+    |  cmpdi TMP2, 0
-+    |   mr BASE, TMP1
-+    |  beq >3
-+    |1:
-+    |  cmpld RA, RC			// Less args than parameters?
-+    |   ld TMP0, 0(RA)
-+    |  bge >4
-+    |    std TISNIL, 0(RA)		// Clear old fixarg slot (help the GC).
-+    |    addi RA, RA, 8
-+    |2:
-+    |  addic. TMP2, TMP2, -1
-+    |   std TMP0, 0(TMP1)
-+    |    addi TMP1, TMP1, 8
-+    |  bne <1
-+    |3:
-+    |  ins_next2
-+    |
-+    |4:  // Clear missing parameters.
-+    |  mr TMP0, TISNIL
-+    |  b <2
-+    break;
-+
-+  case BC_FUNCC:
-+  case BC_FUNCCW:
-+    |  // BASE = new base, RA = BASE+framesize*8, RB = CFUNC, RC = nargs*8
-+    if (op == BC_FUNCC) {
-+      |  ld RD, CFUNC:RB->f
-+    } else {
-+      |  ld RD, DISPATCH_GL(wrapf)(DISPATCH)
-+    }
-+    |   add TMP1, RA, NARGS8:RC
-+    |   ld TMP2, L->maxstack
-+    |    add RC, BASE, NARGS8:RC
-+    |   std BASE, L->base
-+    |   cmpld TMP1, TMP2
-+    |    std RC, L->top
-+    |     li_vmstate C
-+    |  mtctr RD				// RD is r12, the function linkage register
-+    if (op == BC_FUNCCW) {
-+      |  ld CARG2, CFUNC:RB->f
-+    }
-+    |  mr CARG1, L
-+    |   bgt ->vm_growstack_c		// Need to grow stack.
-+    |     st_vmstate
-+    |  bctrl				// (lua_State *L [, lua_CFunction f])
-+    |  // Returns nresults.
-+    |  ld BASE, L->base
-+    |   ld TOCREG, SAVE_TOC
-+    |   sldi RD, CRET1, 3
-+    |  ld TMP1, L->top
-+    |    li_vmstate INTERP
-+    |  ld PC, FRAME_PC(BASE)		// Fetch PC of caller.
-+    |    std L, DISPATCH_GL(cur_L)(DISPATCH)
-+    |   sub RA, TMP1, RD		// RA = L->top - nresults*8
-+    |    st_vmstate
-+    |  b ->vm_returnc
-+    break;
-+
-+  /* ---------------------------------------------------------------------- */
-+
-+  default:
-+    fprintf(stderr, "Error: undefined opcode BC_%s\n", bc_names[op]);
-+    exit(2);
-+    break;
-+  }
-+}
-+
-+static int build_backend(BuildCtx *ctx)
-+{
-+  int op;
-+
-+  dasm_growpc(Dst, BC__MAX);
-+
-+  build_subroutines(ctx);
-+
-+  |.code_op
-+  for (op = 0; op < BC__MAX; op++)
-+    build_ins(ctx, (BCOp)op, op);
-+
-+  return BC__MAX;
-+}
-+
-+/* Emit pseudo frame-info for all assembler functions. */
-+static void emit_asm_debug(BuildCtx *ctx)
-+{
-+  int fcofs = (int)((uint8_t *)ctx->glob[GLOB_vm_ffi_call] - ctx->code);
-+  int i, lr_offset = -16 >> 2;
-+  switch (ctx->mode) {
-+  case BUILD_elfasm:
-+    fprintf(ctx->fp, "\t.section .debug_frame,\"\",@progbits\n");
-+    fprintf(ctx->fp,
-+	".Lframe0:\n"			/* Common Information Entry (CIE) */
-+	"\t.long .LECIE0-.LSCIE0\n"	/* length */
-+	".LSCIE0:\n"
-+	"\t.long 0xffffffff\n"		/* CIE_Id */
-+	"\t.byte 0x1\n"			/* Version */
-+	"\t.string \"\"\n"		/* augmentation */
-+	"\t.uleb128 0x1\n"		/* code_alignment_factor */
-+	"\t.sleb128 -4\n"		/* data_alignment_factor */
-+	"\t.byte 65\n"			/* return_address_register (LR) */
-+	"\t.byte 0xc\n\t.uleb128 1\n\t.uleb128 0\n"	/* DW_CFA_def_cfa */
-+	"\t.align 2\n"
-+	".LECIE0:\n\n");
-+    fprintf(ctx->fp,
-+	".LSFDE0:\n"			/* Frame Description Entry (FDE) */
-+	"\t.long .LEFDE0-.LASFDE0\n"	/* length */
-+	".LASFDE0:\n"
-+	"\t.long .Lframe0\n"		/* CIE_ptr */
-+	"\t.long .Lbegin\n"		/* initial_location */
-+	"\t.long %d\n"			/* address_range */
-+	"\t.byte 0xe\n\t.uleb128 %d\n"	/* DW_CFA_def_cfa_offset */
-+	/* DW_CFA_offset_extended_sf */
-+	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 %d\n"
-+	/* DW_CFA_offset_extended */
-+	"\t.byte 0x5\n\t.uleb128 70\n\t.uleb128 55\n",
-+	fcofs, CFRAME_SIZE, lr_offset);
-+    for (i = 14; i <= 31; i++)
-+      fprintf(ctx->fp,
-+	"\t.byte %d\n\t.uleb128 %d\n"	/* DW_CFA_offset from r14 to r31 */
-+	"\t.byte %d\n\t.uleb128 %d\n",	/* DW_CFA_offset from f14 to f31 */
-+	0x80+i, 38+2*(31-i), 0x80+32+i, 2+2*(31-i));
-+    fprintf(ctx->fp,
-+	"\t.align 2\n"
-+	".LEFDE0:\n\n");
-+#if LJ_HASFFI
-+    fprintf(ctx->fp,
-+	".LSFDE1:\n"			/* Frame Description Entry (FDE) */
-+	"\t.long .LEFDE1-.LASFDE1\n"	/* length */
-+	".LASFDE1:\n"
-+	"\t.long .Lframe0\n"		/* CIE_ptr */
-+	"\t.long lj_vm_ffi_call\n"	/* initial_location */
-+	"\t.long %d\n"			/* address_range */
-+	/* DW_CFA_offset_extended_sf */
-+	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 %d\n"
-+	"\t.byte 0x8e\n\t.uleb128 2\n"	/* DW_CFA_offset */
-+	"\t.byte 0xd\n\t.uleb128 0xe\n"	/* DW_CFA_def_cfa_register */
-+	"\t.align 2\n"
-+	".LEFDE1:\n\n", (int)ctx->codesz - fcofs, lr_offset);
-+#endif
-+#if !LJ_NO_UNWIND
-+    fprintf(ctx->fp, "\t.section .eh_frame,\"a\",@progbits\n");
-+    fprintf(ctx->fp,
-+	".Lframe1:\n"
-+	"\t.long .LECIE1-.LSCIE1\n"	/* length */
-+	".LSCIE1:\n"			/* Common Information Entry (CIE) */
-+	"\t.long 0\n"			/* CIE_Id */
-+	"\t.byte 0x1\n"			/* Version */
-+	"\t.string \"zPR\"\n"		/* augmentation string */
-+	"\t.uleb128 0x1\n"		/* code_alignment_factor */
-+	"\t.sleb128 -4\n"		/* data_alignment_factor */
-+	"\t.byte 65\n"			/* return_address_register (LR) */
-+	"\t.uleb128 6\n"			/* augmentation length */
-+	"\t.byte 0x1b\n"			/* pcrel|sdata4 */
-+	"\t.long lj_err_unwind_dwarf-.\n"
-+	"\t.byte 0x1b\n"			/* pcrel|sdata4 */
-+	"\t.byte 0xc\n\t.uleb128 1\n\t.uleb128 0\n"	/* DW_CFA_def_cfa */
-+	"\t.align 2\n"
-+	".LECIE1:\n\n");
-+    fprintf(ctx->fp,
-+	".LSFDE2:\n"
-+	"\t.long .LEFDE2-.LASFDE2\n"
-+	".LASFDE2:\n"
-+	"\t.long .LASFDE2-.Lframe1\n"
-+	"\t.long .Lbegin-.\n"
-+	"\t.long %d\n"
-+	"\t.uleb128 0\n"			/* augmentation length */
-+	"\t.byte 0xe\n\t.uleb128 %d\n"		/* DW_CFA_def_cfa_offset */
-+	/* DW_CFA_offset_extended_sf */
-+	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 %d\n"
-+	/* DW_CFA_offset_extended */
-+	"\t.byte 0x5\n\t.uleb128 70\n\t.uleb128 55\n",
-+	fcofs, CFRAME_SIZE, lr_offset);
-+    for (i = 14; i <= 31; i++)
-+      fprintf(ctx->fp,
-+	"\t.byte %d\n\t.uleb128 %d\n"	/* DW_CFA_offset from r14 to r31 */
-+	"\t.byte %d\n\t.uleb128 %d\n",	/* DW_CFA_offset from f14 to f31 */
-+	0x80+i, 38+2*(31-i), 0x80+32+i, 2+2*(31-i));
-+    fprintf(ctx->fp,
-+	"\t.align 2\n"
-+	".LEFDE2:\n\n");
-+#if LJ_HASFFI
-+    fprintf(ctx->fp,
-+	".Lframe2:\n"
-+	"\t.long .LECIE2-.LSCIE2\n"
-+	".LSCIE2:\n"			/* Common Information Entry (CIE) */
-+	"\t.long 0\n"			/* CIE_Id */
-+	"\t.byte 0x1\n"			/* Version */
-+	"\t.string \"zR\"\n"		/* augmentation string */
-+	"\t.uleb128 0x1\n"		/* code_alignment_factor */
-+	"\t.sleb128 -4\n"		/* data_alignment_factor */
-+	"\t.byte 65\n"			/* return_address_register (LR) */
-+	"\t.uleb128 1\n"			/* augmentation length */
-+	"\t.byte 0x1b\n"			/* pcrel|sdata4 */
-+	"\t.byte 0xc\n\t.uleb128 1\n\t.uleb128 0\n"	/* DW_CFA_def_cfa */
-+	"\t.align 2\n"
-+	".LECIE2:\n\n");
-+    fprintf(ctx->fp,
-+	".LSFDE3:\n"
-+	"\t.long .LEFDE3-.LASFDE3\n"
-+	".LASFDE3:\n"
-+	"\t.long .LASFDE3-.Lframe2\n"
-+	"\t.long lj_vm_ffi_call-.\n"
-+	"\t.long %d\n"
-+	"\t.uleb128 0\n"			/* augmentation length */
-+	/* DW_CFA_offset_extended_sf */
-+	"\t.byte 0x11\n\t.uleb128 65\n\t.sleb128 %d\n"
-+	"\t.byte 0x8e\n\t.uleb128 2\n"	/* DW_CFA_offset */
-+	"\t.byte 0xd\n\t.uleb128 0xe\n"	/* DW_CFA_def_cfa_register */
-+	"\t.align 2\n"
-+	".LEFDE3:\n\n", (int)ctx->codesz - fcofs, lr_offset);
-+#endif
-+#endif
-+    break;
-+  default:
-+    break;
-+  }
-+}
--- 
-2.12.2
-