diff --git a/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch b/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch new file mode 100644 index 0000000..6be719a --- /dev/null +++ b/0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch @@ -0,0 +1,3521 @@ +From: "Rodrigo R. Galvao" +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 +--- + 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/0006-Fix-register-allocation-bug-in-arm64.patch b/0006-Fix-register-allocation-bug-in-arm64.patch new file mode 100644 index 0000000..fb5b730 --- /dev/null +++ b/0006-Fix-register-allocation-bug-in-arm64.patch @@ -0,0 +1,29 @@ +From: Jason Teplitz +Date: Mon, 9 Oct 2017 23:03:09 +0000 +Subject: Fix register allocation bug in arm64 + +--- + src/lj_asm_arm64.h | 3 +-- + 1 file changed, 1 insertion(+), 2 deletions(-) + +diff --git a/src/lj_asm_arm64.h b/src/lj_asm_arm64.h +index 8fd92e7..549f8a6 100644 +--- a/src/lj_asm_arm64.h ++++ b/src/lj_asm_arm64.h +@@ -871,7 +871,7 @@ static void asm_hrefk(ASMState *as, IRIns *ir) + int bigofs = !emit_checkofs(A64I_LDRx, ofs); + RegSet allow = RSET_GPR; + Reg dest = (ra_used(ir) || bigofs) ? ra_dest(as, ir, RSET_GPR) : RID_NONE; +- Reg node = ra_alloc1(as, ir->op1, allow); ++ Reg node = ra_alloc1(as, ir->op1, ra_hasreg(dest) ? rset_clear(allow, dest) : allow); + Reg key = ra_scratch(as, rset_clear(allow, node)); + Reg idx = node; + uint64_t k; +@@ -879,7 +879,6 @@ static void asm_hrefk(ASMState *as, IRIns *ir) + rset_clear(allow, key); + if (bigofs) { + idx = dest; +- rset_clear(allow, dest); + kofs = (int32_t)offsetof(Node, key); + } else if (ra_hasreg(dest)) { + emit_opk(as, A64I_ADDx, dest, node, ofs, allow); diff --git a/texlive.changes b/texlive.changes index 5c1e67c..5ac4750 100644 --- a/texlive.changes +++ b/texlive.changes @@ -1,3 +1,12 @@ +------------------------------------------------------------------- +Wed Apr 29 09:56:08 UTC 2020 - Dr. Werner Fink + +- Add patch 0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch + *Support luajit on ppc64/ppc64le +- Add patch 0006-Fix-register-allocation-bug-in-arm64.patch + * luajit fix for arm64 +- Modify spec file to build luajit based programs on ppc64* + ------------------------------------------------------------------- Wed Apr 22 11:20:57 UTC 2020 - Dr. Werner Fink diff --git a/texlive.spec b/texlive.spec index 4976fc3..335551b 100644 --- a/texlive.spec +++ b/texlive.spec @@ -19,7 +19,7 @@ %define texlive_version 2020 %define texlive_previous 2019 %define texlive_release 20200327 -%define texlive_noarch 171 +%define texlive_noarch 172 %define texlive_source texlive-20200327-source %define __perl_requires %{nil} @@ -42,7 +42,7 @@ %ifarch %ix86 x86_64 ppc %arml aarch64 mips %global with_LuaJIT 1 %endif -%bcond_with LuaJIT +%bcond_without LuaJIT # # poppler -- use system wide libpoppler @@ -283,6 +283,10 @@ Patch61: source-poppler0.83.0.patch Patch62: source-psutils-kpathsea.dif # PATCH-FIX-UPSTREAM source-poppler0.86.0.patch Patch64: source-poppler0.86.0.patch +# PATCH-FIX-SUSE Support luajit on ppc64/ppc64le +Patch104: 0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch +# PATCH-FIX-SUSE Support luajit fix for arm64 +Patch106: 0006-Fix-register-allocation-bug-in-arm64.patch Prefix: %{_bindir} Provides: pdfjam = %{version} Obsoletes: pdfjam < %{version} @@ -3802,6 +3806,10 @@ This package is required by the package texlive-biber-bin. %patch18 -p0 -b .a2p %patch19 -p0 -b .dvipng %patch20 -p0 -b .missed +pushd libs/luajit/LuaJIT-src/ +%patch104 -p1 -b .ppc64 +%patch106 -p1 -b .arm64 +popd %patch0 -p0 -b .p0 %if %{with buildbiber} pushd ../*biber-*/