path: root/src/buildvm_ppc.dasc

|// Low-level VM code for PowerPC CPUs.
|// Bytecode interpreter, fast functions and helper functions.
|// Copyright (C) 2005-2010 Mike Pall. See Copyright Notice in luajit.h
|
|.arch ppc
|.section code_op, code_sub
|
|.actionlist build_actionlist
|.globals GLOB_
|.globalnames globnames
|.externnames extnames
|
|.if not SPE
|.error "No support for plain PowerPC CPUs (yet)"
|.endif
|
|// Note: The ragged indentation of the instructions is intentional.
|//       The starting columns indicate data dependencies.
|
|//-----------------------------------------------------------------------
|
|// Fixed register assignments for the interpreter.
|// Don't use: r1 = sp, r2 and r13 = reserved and/or small data area ptr
|
|// The following must be C callee-save (but BASE is often refetched).
|.define BASE,		r14	// Base of current Lua stack frame.
|.define KBASE,		r15	// Constants of current Lua function.
|.define PC,		r16	// Next PC.
|.define DISPATCH,	r17	// Opcode dispatch table.
|.define LREG,		r18	// Register holding lua_State (also in SAVE_L).
|
|// Constants for vectorized type-comparisons (hi+low GPR). C callee-save.
|.define TISNUM,	r21
|.if SPE
|.define TISSTR,	r22
|.define TISTAB,	r23
|.define TISFUNC,	r24
|.define TISNIL,	r25
|.endif
|
|// The following temporaries are not saved across C calls, except for RA.
|.define RA,		r19	// Callee-save.
|.define RB,		r10
|.define RC,		r11
|.define RD,		r12
|.define INS,		r7	// Overlaps CARG5.
|
|.define TMP0,		r0
|.define TMP1,		r8
|.define TMP2,		r9
|.define TMP3,		r6	// Overlaps CARG4.
|
|// Saved temporaries.
|.define SAVE0,		r20
|
|// Calling conventions.
|.define CARG1,		r3
|.define CARG2,		r4
|.define CARG3,		r5
|.define CARG4,		r6	// Overlaps TMP3.
|.define CARG5,		r7	// Overlaps INS.
|
|.define CRET1,		r3
|.define CRET2,		r4
|
|// Stack layout while in interpreter. Must match with lj_frame.h.
|.if SPE
|.define SAVE_LR,	180(sp)
|.define CFRAME_SPACE,	176	// Delta for sp.
|// Back chain for sp:	176(sp)	<-- sp entering interpreter
|.define SAVE_r31,	168(sp)	// 64 bit register saves.
|.define SAVE_r30,	160(sp)
|.define SAVE_r29,	152(sp)
|.define SAVE_r28,	144(sp)
|.define SAVE_r27,	136(sp)
|.define SAVE_r26,	128(sp)
|.define SAVE_r25,	120(sp)
|.define SAVE_r24,	112(sp)
|.define SAVE_r23,	104(sp)
|.define SAVE_r22,	96(sp)
|.define SAVE_r21,	88(sp)
|.define SAVE_r20,	80(sp)
|.define SAVE_r19,	72(sp)
|.define SAVE_r18,	64(sp)
|.define SAVE_r17,	56(sp)
|.define SAVE_r16,	48(sp)
|.define SAVE_r15,	40(sp)
|.define SAVE_r14,	32(sp)
|.define SAVE_ERRF,	28(sp)	// 32 bit C frame info.
|.define SAVE_NRES,	24(sp)
|.define SAVE_CFRAME,	20(sp)
|.define SAVE_L,	16(sp)
|.define SAVE_PC,	12(sp)
|.define SAVE_MULTRES,	8(sp)
|// Next frame lr:	4(sp)
|// Back chain for sp:	0(sp)	<-- sp while in interpreter
|
|.macro save_, reg; evstdd reg, SAVE_..reg; .endmacro
|.macro rest_, reg; evldd reg, SAVE_..reg; .endmacro
|.endif
|
|.macro saveregs
|  stwu sp, -CFRAME_SPACE(sp)
|  save_ r14; save_ r15; save_ r16; save_ r17; save_ r18; save_ r19
|  mflr r0
|  save_ r20; save_ r21; save_ r22; save_ r23; save_ r24; save_ r25
|  stw  r0, SAVE_LR
|  save_ r26; save_ r27; save_ r28; save_ r29; save_ r30; save_ r31
|.endmacro
|
|.macro restoreregs
|  lwz r0, SAVE_LR
|  rest_ r14; rest_ r15; rest_ r16; rest_ r17; rest_ r18; rest_ r19
|  mtlr r0
|  rest_ r20; rest_ r21; rest_ r22; rest_ r23; rest_ r24; rest_ r25
|  rest_ r26; rest_ r27; rest_ r28; rest_ r29; rest_ r30; rest_ r31
|  addi sp, sp, CFRAME_SPACE
|.endmacro
|
|// Type definitions. Some of these are only used for documentation.
|.type L,		lua_State,	LREG
|.type GL,		global_State
|.type TVALUE,		TValue
|.type GCOBJ,		GCobj
|.type STR,		GCstr
|.type TAB,		GCtab
|.type LFUNC,		GCfuncL
|.type CFUNC,		GCfuncC
|.type PROTO,		GCproto
|.type UPVAL,		GCupval
|.type NODE,		Node
|.type NARGS8,		int
|.type TRACE,		GCtrace
|
|//-----------------------------------------------------------------------
|
|// These basic macros should really be part of DynASM.
|.macro srwi, rx, ry, n; rlwinm rx, ry, 32-n, n, 31; .endmacro
|.macro slwi, rx, ry, n; rlwinm rx, ry, n, 0, 31-n; .endmacro
|.macro subi, rx, ry, i; addi rx, ry, -i; .endmacro
|
|// Trap for not-yet-implemented parts.
|.macro NYI; tw 4, sp, sp; .endmacro
|
|//-----------------------------------------------------------------------
|
|// Access to frame relative to BASE.
|.define FRAME_PC,	-8
|.define FRAME_FUNC,	-4
|
|// Instruction decode.
|.macro decode_OP4, dst, ins; rlwinm dst, ins, 2, 22, 29; .endmacro
|.macro decode_RA8, dst, ins; rlwinm dst, ins, 27, 21, 28; .endmacro
|.macro decode_RB8, dst, ins; rlwinm dst, ins, 11, 21, 28; .endmacro
|.macro decode_RC8, dst, ins; rlwinm dst, ins, 19, 21, 28; .endmacro
|.macro decode_RD8, dst, ins; rlwinm dst, ins, 19, 13, 28; .endmacro
|
|.macro decode_OP1, dst, ins; rlwinm dst, ins, 0, 24, 31; .endmacro
|.macro decode_RD4, dst, ins; rlwinm dst, ins, 18, 14, 29; .endmacro
|
|// Instruction decode+dispatch.
|.macro ins_NEXT
|  lwz INS, 0(PC)
|   addi PC, PC, 4
|  decode_OP4 TMP0, INS
|   decode_RB8 RB, INS
|  lwzx TMP0, DISPATCH, TMP0
|   decode_RD8 RD, INS
|   decode_RC8 RC, INS
|  mtctr TMP0
|   decode_RA8 RA, INS
|  bctr
|.endmacro
|
|// Instruction footer.
|.if 1
|  // Replicated dispatch. Less unpredictable branches, but higher I-Cache use.
|  .define ins_next, ins_NEXT
|  .define ins_next_, ins_NEXT
|.else
|  // Common dispatch. Lower I-Cache use, only one (very) unpredictable branch.
|  // Affects only certain kinds of benchmarks (and only with -j off).
|  .macro ins_next
|    b ->ins_next
|  .endmacro
|  .macro ins_next_
|  ->ins_next:
|    ins_NEXT
|  .endmacro
|.endif
|
|// Call decode and dispatch.
|.macro ins_callt
|  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC
|  lwz PC, LFUNC:RB->pc
|  lwz INS, 0(PC)
|   addi PC, PC, 4
|  decode_OP4 TMP0, INS
|   decode_RA8 RA, INS
|  lwzx TMP0, DISPATCH, TMP0
|   add RA, RA, BASE
|  mtctr TMP0
|  bctr
|.endmacro
|
|.macro ins_call
|  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, PC = caller PC
|  stw PC, FRAME_PC(BASE)
|  ins_callt
|.endmacro
|
|//-----------------------------------------------------------------------
|
|// Macros to test operand types.
|.if SPE
|.macro checknum, reg; evcmpltu reg, TISNUM; .endmacro
|.macro checkstr, reg; evcmpeq reg, TISSTR; .endmacro
|.macro checktab, reg; evcmpeq reg, TISTAB; .endmacro
|.macro checkfunc, reg; evcmpeq reg, TISFUNC; .endmacro
|.macro checknil, reg; evcmpeq reg, TISNIL; .endmacro
|.macro checkok, label; blt label; .endmacro
|.macro checkfail, label; bge label; .endmacro
|.macro checkanyfail, label; bns label; .endmacro
|.endif
|
|.macro branch_RD
|  srwi TMP0, RD, 1
|  add PC, PC, TMP0
|  addis PC, PC, -(BCBIAS_J*4 >> 16)
|.endmacro
|
|// Assumes DISPATCH is relative to GL.
#define DISPATCH_GL(field)	(GG_DISP2G + (int)offsetof(global_State, field))
#define DISPATCH_J(field)	(GG_DISP2J + (int)offsetof(jit_State, field))
|
#define PC2PROTO(field)  ((int)offsetof(GCproto, field)-(int)sizeof(GCproto))
|
|.macro hotloop
|  NYI
|.endmacro
|
|.macro hotcall
|  NYI
|.endmacro
|
|// Set current VM state. Uses TMP0.
|.macro li_vmstate, st; li TMP0, ~LJ_VMST_..st; .endmacro
|.macro st_vmstate; stw TMP0, DISPATCH_GL(vmstate)(DISPATCH); .endmacro
|
|// Move table write barrier back. Overwrites mark and tmp.
|.macro barrierback, tab, mark, tmp
|  lwz tmp, DISPATCH_GL(gc.grayagain)(DISPATCH)
|  // Assumes LJ_GC_BLACK is 0x04.
|   rlwinm mark, mark, 0, 30, 28		// black2gray(tab)
|  stw tab, DISPATCH_GL(gc.grayagain)(DISPATCH)
|   stb mark, tab->marked
|  stw tmp, tab->gclist
|.endmacro
|
|//-----------------------------------------------------------------------

/* Generate subroutines used by opcodes and other parts of the VM. */
/* The .code_sub section should be last to help static branch prediction. */
static void build_subroutines(BuildCtx *ctx)
{
  |.code_sub
  |
  |//-----------------------------------------------------------------------
  |//-- Return handling ----------------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |->vm_returnp:
  |  // See vm_return. Also: TMP2 = previous base.
  |  andi. TMP0, PC, FRAME_P
  |   evsplati TMP1, LJ_TTRUE
  |  beq ->cont_dispatch
  |
  |  // Return from pcall or xpcall fast func.
  |  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.
  |
  |->vm_returnc:
  |  andi. TMP0, PC, FRAME_TYPE
  |   addi RD, RD, 8			// RD = (nresults+1)*8.
  |   stw RD, SAVE_MULTRES
  |  beq ->BC_RET_Z			// Handle regular return to Lua.
  |
  |->vm_return:
  |  // BASE = base, RA = resultptr, RD/MULTRES = (nresults+1)*8, PC = return
  |  // TMP0 = PC & FRAME_TYPE
  |  cmpwi TMP0, FRAME_C
  |   rlwinm TMP2, PC, 0, 0, 28
  |    li_vmstate C
  |   sub TMP2, BASE, TMP2		// TMP2 = previous base.
  |  bne ->vm_returnp
  |
  |  addic. TMP1, RD, -8
  |   stw TMP2, L->base
  |   lwz TMP2, SAVE_NRES
  |    subi BASE, BASE, 8
  |    st_vmstate
  |   slwi TMP2, TMP2, 3
  |  beq >2
  |1:
  |  addic. TMP1, TMP1, -8
  |   evldd TMP0, 0(RA)
  |    addi RA, RA, 8
  |   evstdd TMP0, 0(BASE)
  |    addi BASE, BASE, 8
  |  bne <1
  |
  |2:
  |  cmpw TMP2, RD			// More/less results wanted?
  |  bne >6
  |3:
  |  stw BASE, L->top			// Store new top.
  |
  |->vm_leave_cp:
  |  lwz TMP0, SAVE_CFRAME		// Restore previous C frame.
  |   li CRET1, 0			// Ok return status for vm_pcall.
  |  stw TMP0, L->cframe
  |
  |->vm_leave_unw:
  |  restoreregs
  |  blr
  |
  |6:
  |  ble >7				// Less results wanted?
  |  // More results wanted. Check stack size and fill up results with nil.
  |  lwz TMP1, L->maxstack
  |  cmplw BASE, TMP1
  |  bge >8
  |  evstdd TISNIL, 0(BASE)
  |  addi RD, RD, 8
  |  addi BASE, BASE, 8
  |  b <2
  |
  |7:  // Less results wanted.
  |   sub TMP0, RD, TMP2
  |  cmpwi TMP2, 0			// LUA_MULTRET+1 case?
  |   sub TMP0, BASE, TMP0		// Subtract the difference.
  |  iseleq BASE, BASE, TMP0		// Either keep top or shrink it.
  |  b <3
  |
  |8:  // Corner case: need to grow stack for filling up results.
  |  // This can happen if:
  |  // - A C function grows the stack (a lot).
  |  // - The GC shrinks the stack in between.
  |  // - A return back from a lua_call() with (high) nresults adjustment.
  |  stw BASE, L->top			// Save current top held in BASE (yes).
  |   mr SAVE0, RD
  |  mr CARG2, TMP2
  |  mr CARG1, L
  |  bl extern lj_state_growstack	// (lua_State *L, int n)
  |    lwz TMP2, SAVE_NRES
  |   mr RD, SAVE0
  |    slwi TMP2, TMP2, 3
  |  lwz BASE, L->top			// Need the (realloced) L->top in BASE.
  |  b <2
  |
  |->vm_unwind_c:			// Unwind C stack, return from vm_pcall.
  |  // (void *cframe, int errcode)
  |  mr sp, CARG1
  |  mr CRET1, CARG2
  |->vm_unwind_c_eh:			// Landing pad for external unwinder.
  |  lwz L, SAVE_L
  |   li TMP0, ~LJ_VMST_C
  |  lwz GL:TMP1, L->glref
  |   stw TMP0, GL:TMP1->vmstate
  |  b ->vm_leave_unw
  |
  |->vm_unwind_ff:			// Unwind C stack, return from ff pcall.
  |  // (void *cframe)
  |  rlwinm sp, CARG1, 0, 0, 29
  |->vm_unwind_ff_eh:			// Landing pad for external unwinder.
  |  lwz L, SAVE_L
  |  lwz BASE, L->base
  |  li TMP1, LJ_TFALSE
  |   lwz DISPATCH, L->glref		// Setup pointer to dispatch table.
  |    li_vmstate INTERP
  |  lwz PC, FRAME_PC(BASE)		// Fetch PC of previous frame.
  |  subi RA, BASE, 8			// Results start at BASE-8.
  |   addi DISPATCH, DISPATCH, GG_G2DISP
  |  stw TMP1, 0(RA)			// Prepend false to error message.
  |  li RD, 16				// 2 results: false + error message.
  |    st_vmstate
  |  b ->vm_returnc
  |
  |//-----------------------------------------------------------------------
  |//-- Grow stack for calls -----------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |->vm_growstack_c:			// Grow stack for C function.
  |  li CARG2, LUA_MINSTACK
  |  b >2
  |
  |->vm_growstack_l:			// Grow stack for Lua function.
  |  // BASE = new base, RA = BASE+framesize*8, RC = nargs*8, PC = first PC
  |  add RC, BASE, RC
  |   sub RA, RA, BASE
  |  stw BASE, L->base
  |   addi PC, PC, 4			// Must point after first instruction.
  |  stw RC, L->top
  |   srwi CARG2, RA, 3
  |2:
  |  // L->base = new base, L->top = top
  |  mr CARG1, L
  |  bl extern lj_state_growstack	// (lua_State *L, int n)
  |  lwz BASE, L->base
  |  lwz RC, L->top
  |  lwz LFUNC:RB, FRAME_FUNC(BASE)
  |  sub RC, RC, BASE
  |  // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC
  |  ins_callt				// Just retry the call.
  |
  |//-----------------------------------------------------------------------
  |//-- Entry points into the assembler VM ---------------------------------
  |//-----------------------------------------------------------------------
  |
  |->vm_resume:				// Setup C frame and resume thread.
  |  NYI
  |
  |->vm_pcall:				// Setup protected C frame and enter VM.
  |  // (lua_State *L, TValue *base, int nres1, ptrdiff_t ef)
  |  saveregs
  |  li PC, FRAME_CP
  |  stw CARG4, SAVE_ERRF
  |  b >1
  |
  |->vm_call:				// Setup C frame and enter VM.
  |  // (lua_State *L, TValue *base, int nres1)
  |  saveregs
  |  li PC, FRAME_C
  |
  |1:  // Entry point for vm_pcall above (PC = ftype).
  |  lwz TMP1, L:CARG1->cframe
  |   stw CARG3, SAVE_NRES
  |    mr L, CARG1
  |   stw CARG1, SAVE_L
  |    mr BASE, CARG2
  |  stw sp, L->cframe			// Add our C frame to cframe chain.
  |    lwz DISPATCH, L->glref		// Setup pointer to dispatch table.
  |   stw CARG1, SAVE_PC		// Any value outside of bytecode is ok.
  |  stw TMP1, SAVE_CFRAME
  |    addi DISPATCH, DISPATCH, GG_G2DISP
  |
  |3:  // Entry point for vm_cpcall/vm_resume (BASE = base, PC = ftype).
  |  lwz TMP2, L->base			// TMP2 = old base (used in vmeta_call).
  |    evsplati TISNUM, LJ_TISNUM+1	// Setup type comparison constants.
  |   lwz TMP1, L->top
  |    evsplati TISFUNC, LJ_TFUNC
  |  add PC, PC, BASE
  |    li_vmstate INTERP
  |    evsplati TISTAB, LJ_TTAB
  |  sub PC, PC, TMP2			// PC = frame delta + frame type
  |    st_vmstate
  |    evsplati TISSTR, LJ_TSTR
  |   sub NARGS8:RC, TMP1, BASE
  |    evsplati TISNIL, LJ_TNIL
  |
  |->vm_call_dispatch:
  |  // TMP2 = old base, BASE = new base, RC = nargs*8, PC = caller PC
  |  // NYI: reschedule.
  |  li TMP0, -8
  |  evlddx LFUNC:RB, BASE, TMP0
  |  checkfunc LFUNC:RB
  |  checkfail ->vmeta_call		// Ensure KBASE defined and != BASE.
  |
  |->vm_call_dispatch_f:
  |  ins_call
  |  // BASE = new base, RC = nargs*8
  |
  |->vm_cpcall:				// Setup protected C frame, call C.
  |  // (lua_State *L, lua_CFunction func, void *ud, lua_CPFunction cp)
  |  saveregs
  |  mr L, CARG1
  |  mtctr CARG4
  |   lwz TMP0, L:CARG1->stack
  |  stw CARG1, SAVE_L
  |   lwz TMP1, L->top
  |  stw CARG1, SAVE_PC			// Any value outside of bytecode is ok.
  |   sub TMP0, TMP0, TMP1		// Compute -savestack(L, L->top).
  |    lwz TMP1, L->cframe
  |    stw sp, L->cframe		// Add our C frame to cframe chain.
  |  li TMP2, 0
  |   stw TMP0, SAVE_NRES		// Neg. delta means cframe w/o frame.
  |  stw TMP2, SAVE_ERRF		// No error function.
  |    stw TMP1, SAVE_CFRAME
  |  bctrl			// (lua_State *L, lua_CFunction func, void *ud)
  |  mr. BASE, CRET1
  |   lwz DISPATCH, L->glref		// Setup pointer to dispatch table.
  |    li PC, FRAME_CP
  |   addi DISPATCH, DISPATCH, GG_G2DISP
  |  bne <3				// Else continue with the call.
  |  b ->vm_leave_cp			// No base? Just remove C frame.
  |
  |//-----------------------------------------------------------------------
  |//-- Metamethod handling ------------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |//-- Continuation dispatch ----------------------------------------------
  |
  |->cont_dispatch:
  |  NYI
  |
  |->cont_cat:
  |  NYI
  |
  |//-- Table indexing metamethods -----------------------------------------
  |
  |->vmeta_tgets1:
  |->vmeta_tgets:
  |  NYI
  |
  |->vmeta_tgetb:
  |  NYI
  |
  |->vmeta_tgetv:
  |  NYI
  |
  |//-----------------------------------------------------------------------
  |
  |->vmeta_tsets1:
  |->vmeta_tsets:
  |  NYI
  |
  |->vmeta_tsetb:
  |  NYI
  |
  |->vmeta_tsetv:
  |  NYI
  |
  |//-- Comparison metamethods ---------------------------------------------
  |
  |->vmeta_comp:
  |  NYI
  |->cont_nop:
  |  NYI
  |
  |->cont_ra:
  |  NYI
  |
  |->cont_condt:
  |  NYI
  |
  |->cont_condf:
  |  NYI
  |
  |->vmeta_equal:
  |  NYI
  |
  |//-- Arithmetic metamethods ---------------------------------------------
  |
  |->vmeta_arith_vn:
  |  NYI
  |
  |->vmeta_arith_nv:
  |  NYI
  |
  |->vmeta_unm:
  |  NYI
  |
  |->vmeta_arith_vv:
  |  NYI
  |
  |->vmeta_binop:
  |  NYI
  |
  |->vmeta_len:
  |  NYI
  |
  |//-- Call metamethod ----------------------------------------------------
  |
  |->vmeta_call_ra:
  |  NYI
  |
  |->vmeta_call:			// Resolve and call __call metamethod.
  |  NYI
  |
  |->vmeta_callt:			// Resolve __call for BC_CALLT.
  |  NYI
  |
  |//-- Argument coercion for 'for' statement ------------------------------
  |
  |->vmeta_for:
  |  NYI
  |
  |//-----------------------------------------------------------------------
  |//-- Fast functions -----------------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |.macro .ffunc, name
  |->ff_ .. name:
  |.endmacro
  |
  |.macro .ffunc_1, name
  |->ff_ .. name:
  |  NYI
  |.endmacro
  |
  |.macro .ffunc_2, name
  |->ff_ .. name:
  |  NYI
  |.endmacro
  |
  |.macro .ffunc_n, name
  |  .ffunc_1 name
  |  NYI
  |.endmacro
  |
  |.macro .ffunc_nn, name
  |  .ffunc_2 name
  |  NYI
  |.endmacro
  |
  |.macro ffgccheck
  |  NYI
  |.endmacro
  |
  |//-- Base library: checks -----------------------------------------------
  |
  |.ffunc assert
  |  NYI
  |
  |.ffunc type
  |  NYI
  |
  |//-- Base library: getters and setters ---------------------------------
  |
  |.ffunc_1 getmetatable
  |  NYI
  |
  |.ffunc_2 setmetatable
  |  NYI
  |
  |.ffunc_2 rawget
  |  NYI
  |
  |//-- Base library: conversions ------------------------------------------
  |
  |.ffunc tonumber
  |  NYI
  |
  |.ffunc_1 tostring
  |  NYI
  |
  |//-- Base library: iterators -------------------------------------------
  |
  |.ffunc_1 next
  |  NYI
  |
  |->fff_res2:
  |  NYI
  |
  |.ffunc_1 pairs
  |  NYI
  |
  |.ffunc_1 ipairs_aux
  |  NYI
  |
  |->fff_res0:
  |  NYI
  |
  |.ffunc_1 ipairs
  |  NYI
  |
  |//-- Base library: catch errors ----------------------------------------
  |
  |.ffunc_1 pcall
  |  NYI
  |
  |.ffunc_2 xpcall
  |  NYI
  |
  |//-- Coroutine library --------------------------------------------------
  |
  |.macro coroutine_resume_wrap, resume
  |.if resume
  |.ffunc_1 coroutine_resume
  |.else
  |.ffunc coroutine_wrap_aux
  |.endif
  |  NYI
  |.endmacro
  |
  |  coroutine_resume_wrap 1		// coroutine.resume
  |  coroutine_resume_wrap 0		// coroutine.wrap
  |
  |.ffunc coroutine_yield
  |  NYI
  |
  |//-- Math library -------------------------------------------------------
  |
  |.ffunc_n math_abs
  |  NYI
  |  // Fallthrough.
  |
  |->fff_res1:
  |  NYI
  |->fff_res:
  |  NYI
  |
  |.macro math_extern, func
  |  .ffunc math_ .. func
  |  NYI
  |.endmacro
  |
  |.macro math_extern2, func
  |  .ffunc math_ .. func
  |  NYI
  |.endmacro
  |
  |  math_extern floor
  |  math_extern ceil
  |
  |  math_extern sqrt
  |  math_extern log
  |  math_extern log10
  |  math_extern exp
  |  math_extern sin
  |  math_extern cos
  |  math_extern tan
  |  math_extern asin
  |  math_extern acos
  |  math_extern atan
  |  math_extern sinh
  |  math_extern cosh
  |  math_extern tanh
  |  math_extern2 pow
  |  math_extern2 atan2
  |  math_extern2 fmod
  |
  |->ff_math_deg:
  |.ffunc_1 math_rad
  |  NYI
  |
  |.ffunc_nn math_ldexp;	NYI
  |.ffunc_n math_frexp;		NYI
  |.ffunc_n math_modf;		NYI
  |
  |.macro math_minmax, name, cmpop
  |  .ffunc_1 name
  |  NYI
  |.endmacro
  |
  |  math_minmax math_min, efdtstlt
  |  math_minmax math_max, efdtstgt
  |
  |//-- String library -----------------------------------------------------
  |
  |.ffunc_1 string_len
  |  NYI
  |
  |.ffunc string_byte			// Only handle the 1-arg case here.
  |  NYI
  |
  |.ffunc string_char			// Only handle the 1-arg case here.
  |  NYI
  |
  |->fff_newstr:
  |  NYI
  |
  |.ffunc string_sub
  |  NYI
  |
  |->fff_emptystr:  // Range underflow.
  |  NYI
  |
  |.ffunc_2 string_rep			// Only handle the 1-char case inline.
  |  NYI
  |
  |.ffunc_1 string_reverse
  |  NYI
  |
  |.macro ffstring_case, name, lo, hi
  |  .ffunc_1 name
  |  NYI
  |.endmacro
  |
  |ffstring_case string_lower, 0x41, 0x5a
  |ffstring_case string_upper, 0x61, 0x7a
  |
  |//-- Table library ------------------------------------------------------
  |
  |.ffunc_1 table_getn
  |  NYI
  |
  |//-- Bit library --------------------------------------------------------
  |
  |.ffunc_n bit_tobit
  |  NYI
  |
  |.macro .ffunc_bit, name
  |  .ffunc_n name
  |  NYI
  |.endmacro
  |
  |.macro .ffunc_bit_op, name, ins
  |  .ffunc_bit name
  |  NYI
  |.endmacro
  |
  |.ffunc_bit_op bit_band, and
  |.ffunc_bit_op bit_bor, or
  |.ffunc_bit_op bit_bxor, xor
  |
  |.ffunc_bit bit_bswap
  |  NYI
  |
  |.ffunc_bit bit_bnot
  |  NYI
  |
  |->fff_resbit:
  |  NYI
  |
  |->fff_fallback_bit_op:
  |  NYI
  |
  |.macro .ffunc_bit_sh, name, ins
  |  .ffunc_nn name
  |  NYI
  |.endmacro
  |
  |.ffunc_bit_sh bit_lshift, shl
  |.ffunc_bit_sh bit_rshift, shr
  |.ffunc_bit_sh bit_arshift, sar
  |.ffunc_bit_sh bit_rol, rol
  |.ffunc_bit_sh bit_ror, ror
  |
  |//-----------------------------------------------------------------------
  |
  |->fff_fallback:			// Call fast function fallback handler.
  |  NYI
  |
  |->fff_gcstep:			// Call GC step function.
  |  NYI
  |
  |//-----------------------------------------------------------------------
  |//-- Special dispatch targets -------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |->vm_record:				// Dispatch target for recording phase.
#if LJ_HASJIT
  |  NYI
#endif
  |
  |->vm_rethook:			// Dispatch target for return hooks.
  |  NYI
  |
  |->vm_inshook:			// Dispatch target for instr/line hooks.
  |  NYI
  |
  |->cont_hook:				// Continue from hook yield.
  |  NYI
  |
  |->vm_hotloop:			// Hot loop counter underflow.
#if LJ_HASJIT
  |  NYI
#endif
  |
  |->vm_callhook:			// Dispatch target for call hooks.
  |  NYI
  |
  |->vm_hotcall:			// Hot call counter underflow.
#if LJ_HASJIT
  |  NYI
#endif
  |
  |//-----------------------------------------------------------------------
  |//-- Trace exit handler -------------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |->vm_exit_handler:
#if LJ_HASJIT
  |  NYI
#endif
  |->vm_exit_interp:
#if LJ_HASJIT
  |  NYI
#endif
  |
  |//-----------------------------------------------------------------------
  |//-- Math helper functions ----------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |// FP value rounding. Called by math.floor/math.ceil fast functions
  |// and from JIT code.
  |//
  |// This can be inlined if the CPU has the frin/friz/frip/frim instructions.
  |// The alternative hard-float approaches have a deep dependency chain.
  |// The resulting latency is at least 3x-7x the double-precision FP latency
  |// (e500v2: 6cy, e600: 5cy, Cell: 10cy) or around 20-70 cycles.
  |//
  |// The soft-float approach is tedious, but much faster (e500v2: ~11cy/~6cy).
  |// However it relies on a fast way to transfer the FP value to GPRs
  |// (e500v2: 0cy for lo-word, 1cy for hi-word).
  |//
  |.macro vm_round, name, mode
  |  // Used temporaries: TMP0, TMP1, TMP2, TMP3.
  |->name:				// Input: CARG2, output: CRET2
  |  evmergehi CARG1, CARG2, CARG2
  |->name.._hilo:
  |  // Input: CARG1 (hi), CARG2 (hi, lo), output: CRET2
  |  rlwinm TMP2, CARG1, 12, 21, 31
  |  addic. TMP2, TMP2, -1023		// exp = exponent(x) - 1023
  |   li TMP1, -1
  |  cmplwi cr1, TMP2, 51		// 0 <= exp < 51?
  |   subfic TMP0, TMP2, 52
  |  bgt cr1, >1
  |   lus TMP3, 0xfff0
  |  slw TMP0, TMP1, TMP0		// lomask = -1 << (52-exp)
  |   sraw TMP1, TMP3, TMP2		// himask = (int32_t)0xfff00000 >> exp
  |.if mode == 2		// trunc(x):
  |  evmergelo TMP0, TMP1, TMP0
  |  evand CRET2, CARG2, TMP0		// hi &= himask, lo &= lomask
  |.else
  |  andc TMP2, CARG2, TMP0
  |   andc TMP3, CARG1, TMP1
  |  or TMP2, TMP2, TMP3		// ztest = (hi&~himask) | (lo&~lomask)
  |   srawi TMP3, CARG1, 31		// signmask = (int32_t)hi >> 31
  |.if mode == 0		// floor(x):
  |  and. TMP2, TMP2, TMP3		// iszero = ((ztest & signmask) == 0)
  |.else			// ceil(x):
  |  andc. TMP2, TMP2, TMP3		// iszero = ((ztest & ~signmask) == 0)
  |.endif
  |  and CARG2, CARG2, TMP0		// lo &= lomask
  |  and CARG1, CARG1, TMP1		// hi &= himask
  |   subc TMP0, CARG2, TMP0
  |  iseleq TMP0, CARG2, TMP0		// lo = iszero ? lo : lo-lomask
  |   sube TMP1, CARG1, TMP1
  |  iseleq TMP1, CARG1, TMP1		// hi = iszero ? hi : hi-himask+carry
  |  evmergelo CRET2, TMP1, TMP0
  |.endif
  |  blr
  |1:
  |  bgtlr				// Already done if >=2^52, +-inf or nan.
  |.if mode == 2		// trunc(x):
  |  rlwinm TMP1, CARG1, 0, 0, 0	// hi = sign(x)
  |  li TMP0, 0
  |  evmergelo CRET2, TMP1, TMP0
  |.else
  |  rlwinm TMP2, CARG1, 0, 1, 31
  |  srawi TMP0, CARG1, 31		// signmask = (int32_t)hi >> 31
  |  or TMP2, TMP2, CARG2		// ztest = abs(hi) | lo
  |   lus TMP1, 0x3ff0
  |.if mode == 0		// floor(x):
  |  and. TMP2, TMP2, TMP0		// iszero = ((ztest & signmask) == 0)
  |.else			// ceil(x):
  |  andc. TMP2, TMP2, TMP0		// iszero = ((ztest & ~signmask) == 0)
  |.endif
  |   li TMP0, 0
  |  iseleq TMP1, r0, TMP1
  |  rlwimi CARG1, TMP1, 0, 1, 31	// hi = sign(x) | (iszero ? 0.0 : 1.0)
  |  evmergelo CRET2, CARG1, TMP0
  |.endif
  |  blr
  |.endmacro
  |
  |  vm_round vm_floor, 0
  |  vm_round vm_ceil,  1
#if LJ_HASJIT
  |  vm_round vm_trunc, 2
#else
  |->vm_trunc:
  |->vm_trunc_hilo:
#endif
  |
  |->vm_powi:
#if LJ_HASJIT
  |  NYI
#endif
  |
  |->vm_foldfpm:
#if LJ_HASJIT
  |  NYI
#endif
  |
  |// Callable from C: double lj_vm_foldarith(double x, double y, int op)
  |// Compute x op y for basic arithmetic operators (+ - * / % ^ and unary -)
  |// and basic math functions. ORDER ARITH
  |->vm_foldarith:
  |  evmergelo CARG2, CARG1, CARG2
  |   cmplwi CARG5, 1
  |  evmergelo CARG4, CARG3, CARG4
  |   beq >1; bgt >2
  |  efdadd CRET2, CARG2, CARG4; evmergehi CRET1, CRET2, CRET2; blr
  |1:
  |  efdsub CRET2, CARG2, CARG4; evmergehi CRET1, CRET2, CRET2; blr
  |2:
  |   cmplwi CARG5, 3; beq >1; bgt >2
  |  efdmul CRET2, CARG2, CARG4; evmergehi CRET1, CRET2, CRET2; blr
  |1:
  |  efddiv CRET2, CARG2, CARG4; evmergehi CRET1, CRET2, CRET2; blr
  |2:
  |   cmplwi CARG5, 5; beq >1; bgt >2
  |  evmr CARG3, CARG2; efddiv CRET2, CARG2, CARG4
  |  mflr SAVE0; bl ->vm_floor; mtlr SAVE0
  |  efdmul CRET2, CRET2, CARG4; efdsub CRET2, CARG3, CRET2
  |  evmergehi CRET1, CRET2, CRET2; blr
  |1:
  |  b extern pow
  |2:
  |   cmplwi CARG5, 7; beq >1; bgt >2
  |  xoris CARG1, CARG1, 0x8000; blr
  |1:
  |  rlwinm CARG1, CARG1, 0, 1, 31; blr
  |2:
  |  NYI  // Other operations only needed by JIT compiler.
  |
  |//-----------------------------------------------------------------------
  |//-- Miscellaneous functions --------------------------------------------
  |//-----------------------------------------------------------------------
  |
  |//-----------------------------------------------------------------------
}

/* Generate the code for a single instruction. */
static void build_ins(BuildCtx *ctx, BCOp op, int defop)
{
  int vk = 0;
  |=>defop:

  switch (op) {

  /* -- Comparison ops ---------------------------------------------------- */

  /* Remember: all ops branch for a true comparison, fall through otherwise. */

  case BC_ISLT: case BC_ISGE: case BC_ISLE: case BC_ISGT:
    |  // RA = src1*8, RD = src2*8, JMP with RD = target
    |  evlddx TMP0, BASE, RA
    |   addi PC, PC, 4
    |  evlddx TMP1, BASE, RD
    |   addis TMP3, PC, -(BCBIAS_J*4 >> 16)
    |   lwz INS, -4(PC)
    |  evmergehi RB, TMP0, TMP1
    |   decode_RD4 TMP2, INS
    |  checknum RB
    |   add TMP2, TMP2, TMP3
    |  checkanyfail ->vmeta_comp
    if (op == BC_ISLT || op == BC_ISGE) {
      |  efdcmplt TMP0, TMP1
    } else {
      |  efdcmpgt TMP0, TMP1
    }
    if (op == BC_ISLT || op == BC_ISGT) {
      |  iselgt PC, TMP2, PC
    } else {
      |  iselgt PC, PC, TMP2
    }
    |  ins_next
    break;

  case BC_ISEQV: case BC_ISNEV:
    vk = op == BC_ISEQV;
    |  // RA = src1*8, RD = src2*8, JMP with RD = target
    |  evlddx TMP0, BASE, RA
    |   addi PC, PC, 4
    |  evlddx TMP1, BASE, RD
    |   addis TMP3, PC, -(BCBIAS_J*4 >> 16)
    |   lwz INS, -4(PC)
    |  evmergehi RB, TMP0, TMP1
    |   decode_RD4 TMP2, INS
    |  checknum RB
    |   add TMP2, TMP2, TMP3
    |  checkanyfail >5
    |  efdcmpeq TMP0, TMP1
    if (vk) {
      |  iselgt PC, TMP2, PC
    } else {
      |  iselgt PC, PC, TMP2
    }
    |1:
    |  ins_next
    |
    |5:  // Either or both types are not numbers.
    |  evcmpeq TMP0, TMP1
    |   not TMP3, RB
    |   cmplwi cr1, TMP3, ~LJ_TISPRI		// Primitive?
    |  crorc 4*cr7+lt, 4*cr0+so, 4*cr0+lt	// 1: Same tv or different type.
    |   cmplwi cr6, TMP3, ~LJ_TISTABUD		// Table or userdata?
    |  crandc 4*cr7+gt, 4*cr0+lt, 4*cr1+gt	// 2: Same type and primitive.
    |   mr SAVE0, PC
    if (vk) {
      |  isel PC, TMP2, PC, 4*cr7+gt
    } else {
      |  isel TMP2, PC, TMP2, 4*cr7+gt
    }
    |  cror 4*cr7+lt, 4*cr7+lt, 4*cr7+gt	// 1 or 2.
    if (vk) {
      |  isel PC, TMP2, PC, 4*cr0+so
    } else {
      |  isel PC, PC, TMP2, 4*cr0+so
    }
    |  blt cr7, <1			// Done if 1 or 2.
    |  blt cr6, <1			// Done if not tab/ud.
    |
    |  // Different tables or userdatas. Need to check __eq metamethod.
    |  // Field metatable must be at same offset for GCtab and GCudata!
    |  lwz TAB:TMP3, TAB:TMP1->metatable
    |   li RB, 1-vk			// ne = 0 or 1.
    |  cmplwi TAB:TMP3, 0
    |  beq <1				// No metatable?
    |  lbz TMP0, TAB:TMP3->nomm
    |  andi. TMP0, TMP0, 1<<MM_eq
    |  bne <1				// Or 'no __eq' flag set?
    |  mr PC, SAVE0			// Restore old PC.
    |  b ->vmeta_equal			// Handle __eq metamethod.
    break;

  case BC_ISEQS: case BC_ISNES:
    vk = op == BC_ISEQS;
    |  // RA = src*8, RD = str_const*8 (~), JMP with RD = target
    |  evlddx TMP0, BASE, RA
    |   srwi RD, RD, 1
    |    lwz INS, 0(PC)
    |   subfic RD, RD, -4
    |    addi PC, PC, 4
    |   lwzx STR:TMP1, KBASE, RD	// KBASE-4-str_const*4
    |    addis TMP3, PC, -(BCBIAS_J*4 >> 16)
    |    decode_RD4 TMP2, INS
    |   evmergelo STR:TMP1, TISSTR, STR:TMP1
    |    add TMP2, TMP2, TMP3
    |  evcmpeq TMP0, STR:TMP1
    if (vk) {
      |  isel PC, TMP2, PC, 4*cr0+so
    } else {
      |  isel PC, PC, TMP2, 4*cr0+so
    }
    |  ins_next
    break;

  case BC_ISEQN: case BC_ISNEN:
    vk = op == BC_ISEQN;
    |  // RA = src*8, RD = num_const*8, JMP with RD = target
    |  evlddx TMP0, BASE, RA
    |   addi PC, PC, 4
    |  evlddx TMP1, KBASE, RD
    |   addis TMP3, PC, -(BCBIAS_J*4 >> 16)
    |   lwz INS, -4(PC)
    |  efdcmpeq TMP0, TMP1		// NYI: avoid comparison with NaN.
    |   decode_RD4 TMP2, INS
    |   add TMP2, TMP2, TMP3
    if (vk) {
      |  iselgt PC, TMP2, PC
    } else {
      |  iselgt PC, PC, TMP2
    }
    |  ins_next
    break;

  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
    |   srwi TMP1, RD, 3
    |    lwz INS, 0(PC)
    |    addi PC, PC, 4
    |   not TMP1, TMP1
    |    addis TMP3, PC, -(BCBIAS_J*4 >> 16)
    |  cmplw TMP0, TMP1
    |    decode_RD4 TMP2, INS
    |    add TMP2, TMP2, TMP3
    if (vk) {
      |  iseleq PC, TMP2, PC
    } else {
      |  iseleq PC, PC, TMP2
    }
    |  ins_next
    break;

  /* -- Unary test and copy ops ------------------------------------------- */

  case BC_ISTC: case BC_ISFC: case BC_IST: case BC_ISF:
    |  // RA = dst*8 or unused, RD = src*8, JMP with RD = target
    |  evlddx TMP0, BASE, RD
    |   evaddw TMP1, TISNIL, TISNIL	// Synthesize LJ_TFALSE.
    |   lwz INS, 0(PC)
    |  evcmpltu TMP0, TMP1
    |   addi PC, PC, 4
    if (op == BC_IST || op == BC_ISF) {
      |  addis TMP3, PC, -(BCBIAS_J*4 >> 16)
      |  decode_RD4 TMP2, INS
      |  add TMP2, TMP2, TMP3
      if (op == BC_IST) {
	|  isellt PC, TMP2, PC
      } else {
	|  isellt PC, PC, TMP2
      }
    } else {
      if (op == BC_ISTC) {
	|  checkfail >1
      } else {
	|  checkok >1
      }
      |  addis PC, PC, -(BCBIAS_J*4 >> 16)
      |  decode_RD4 TMP2, INS
      |   evstddx TMP0, BASE, RA
      |  add PC, PC, TMP2
      |1:
    }
    |  ins_next
    break;

  /* -- Unary ops --------------------------------------------------------- */

  case BC_MOV:
    |  // RA = dst*8, RD = src*8
    |  evlddx TMP0, BASE, RD
    |  evstddx TMP0, BASE, RA
    |  ins_next_
    break;
  case BC_NOT:
    |  // RA = dst*8, RD = src*8
    |  lwzx TMP0, BASE, RD
    |  subfic TMP1, TMP0, LJ_TTRUE
    |  adde TMP0, TMP0, TMP1
    |  stwx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_UNM:
    |  // RA = dst*8, RD = src*8
    |  evlddx TMP0, BASE, RD
    |   lus TMP1, 0x8000
    |   li TMP2, 0
    |  checknum TMP0
    |   evmergelo TMP1, TMP1, TMP2
    |  checkfail ->vmeta_unm
    |  evxor TMP0, TMP0, TMP1
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_LEN:
    |  // RA = dst*8, RD = src*8
    |  evlddx CARG1, BASE, RD
    |  checkstr CARG1
    |  checkfail >2
    |  lwz CRET1, STR:CARG1->len
    |1:
    |  efdcfsi TMP0, CRET1
    |  evstddx TMP0, BASE, RA
    |  ins_next
    |2:
    |  checktab CARG1
    |  checkfail ->vmeta_len
    |  bl extern lj_tab_len		// (GCtab *t)
    |  // Returns uint32_t (but less than 2^31).
    |  b <1
    break;

  /* -- Binary ops -------------------------------------------------------- */

    |.macro ins_arithpre, t0, t1
    |  // RA = dst*8, RB = src1*8, RC = src2*8 | num_const*8
    ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN);
    ||switch (vk) {
    ||case 0:
    |   evlddx t0, BASE, RB
    |    checknum t0
    |   evlddx t1, KBASE, RC
    |    checkfail ->vmeta_arith_vn
    ||  break;
    ||case 1:
    |   evlddx t1, BASE, RB
    |    checknum t1
    |   evlddx t0, KBASE, RC
    |    checkfail ->vmeta_arith_nv
    ||  break;
    ||default:
    |   evlddx t0, BASE, RB
    |   evlddx t1, BASE, RC
    |    evmergehi TMP2, t0, t1
    |    checknum TMP2
    |    checkanyfail ->vmeta_arith_vv
    ||  break;
    ||}
    |.endmacro
    |
    |.macro ins_arith, ins
    |  ins_arithpre TMP0, TMP1
    |  ins TMP0, TMP0, TMP1
    |  evstddx TMP0, BASE, RA
    |  ins_next
    |.endmacro

  case BC_ADDVN: case BC_ADDNV: case BC_ADDVV:
    |  ins_arith efdadd
    break;
  case BC_SUBVN: case BC_SUBNV: case BC_SUBVV:
    |  ins_arith efdsub
    break;
  case BC_MULVN: case BC_MULNV: case BC_MULVV:
    |  ins_arith efdmul
    break;
  case BC_DIVVN: case BC_DIVNV: case BC_DIVVV:
    |  ins_arith efddiv
    break;
  case BC_MODVN:
    |  ins_arithpre RD, SAVE0
    |->BC_MODVN_Z:
    |  efddiv CARG2, RD, SAVE0
    |  bl ->vm_floor			// floor(b/c)
    |  efdmul TMP0, CRET2, SAVE0
    |  efdsub TMP0, RD, TMP0		// b - floor(b/c)*c
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_MODNV: case BC_MODVV:
    |  ins_arithpre RD, SAVE0
    |  b ->BC_MODVN_Z			// Avoid 3 copies. It's slow anyway.
    break;
  case BC_POW:
    |  evlddx CARG2, BASE, RB
    |  evlddx CARG4, BASE, RC
    |  evmergehi CARG1, CARG4, CARG2
    |  checknum CARG1
    |   evmergehi CARG3, CARG4, CARG4
    |  checkanyfail ->vmeta_arith_vv
    |  bl extern pow
    |  evmergelo CRET2, CRET1, CRET2
    |  evstddx CRET2, BASE, RA
    |  ins_next
    break;

  case BC_CAT:
    |  NYI
    break;

  /* -- Constant ops ------------------------------------------------------ */

  case BC_KSTR:
    |  // RA = dst*8, RD = str_const*8 (~)
    |  srwi TMP1, RD, 1
    |  subfic TMP1, TMP1, -4
    |  lwzx TMP0, KBASE, TMP1		// KBASE-4-str_const*4
    |  evmergelo TMP0, TISSTR, TMP0
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_KSHORT:
    |  // RA = dst*8, RD = int16_literal*8
    |  srwi TMP1, RD, 3
    |  extsh TMP1, TMP1
    |  efdcfsi TMP0, TMP1
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_KNUM:
    |  // RA = dst*8, RD = num_const*8
    |  evlddx TMP0, KBASE, RD
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_KPRI:
    |  // RA = dst*8, RD = primitive_type*8 (~)
    |  srwi TMP1, RD, 3
    |  not TMP0, TMP1
    |  stwx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_KNIL:
    |  // RA = base*8, RD = end*8
    |  evstddx TISNIL, BASE, RA
    |   addi RA, RA, 8
    |1:
    |  evstddx TISNIL, BASE, RA
    |  cmpw RA, RD
    |   addi RA, RA, 8
    |  blt <1
    |  ins_next
    break;

  /* -- Upvalue and function ops ------------------------------------------ */

  case BC_UGET:
    |  // RA = dst*8, RD = uvnum*8
    |  lwz LFUNC:RB, FRAME_FUNC(BASE)
    |   srwi RD, RD, 1
    |   addi RD, RD, offsetof(GCfuncL, uvptr)
    |  lwzx UPVAL:RB, LFUNC:RB, RD
    |  lwz TMP1, UPVAL:RB->v
    |  evldd TMP0, 0(TMP1)
    |  evstddx TMP0, BASE, RA
    |  ins_next
    break;
  case BC_USETV:
    |  // RA = uvnum*8, RD = src*8
    |  lwz LFUNC:RB, FRAME_FUNC(BASE)
    |    srwi RA, RA, 1
    |    addi RA, RA, offsetof(GCfuncL, uvptr)
    |   evlddx TMP1, BASE, RD
    |  lwzx UPVAL:RB, LFUNC:RB, RA
    |  lbz TMP3, UPVAL:RB->marked
    |   lwz CARG2, UPVAL:RB->v
    |  andi. TMP3, TMP3, LJ_GC_BLACK	// isblack(uv)
    |    lbz TMP2, UPVAL:RB->closed
    |   evmergehi TMP0, TMP1, TMP1
    |   evstdd TMP1, 0(CARG2)
    |    cmplwi cr1, TMP2, 0
    |  cror 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
    |   subi TMP0, TMP0, LJ_TISNUM
    |  bne >2				// Upvalue is closed and black?
    |1:
    |  ins_next
    |
    |2:  // Check if new value is collectable.
    |  cmplwi TMP0, LJ_TISGCV - LJ_TISNUM
    |  bge <1				// tvisgcv(v)
    |  lbz TMP3, GCOBJ:TMP1->gch.marked
    |  andi. TMP3, TMP3, LJ_GC_WHITES	// iswhite(v)
    |   la CARG1, GG_DISP2G(DISPATCH)
    |  // Crossed a write barrier. Move the barrier forward.
    |  bnel extern lj_gc_barrieruv	// (global_State *g, TValue *tv)
    |  b <1
    break;
  case BC_USETS:
    |  // RA = uvnum*8, RD = str_const*8 (~)
    |  lwz LFUNC:RB, FRAME_FUNC(BASE)
    |   srwi TMP1, RD, 1
    |    srwi RA, RA, 1
    |   subfic TMP1, TMP1, -4
    |    addi RA, RA, offsetof(GCfuncL, uvptr)
    |   lwzx STR:TMP1, KBASE, TMP1	// KBASE-4-str_const*4
    |  lwzx UPVAL:RB, LFUNC:RB, RA
    |   evmergelo STR:TMP1, TISSTR, STR:TMP1
    |  lbz TMP3, UPVAL:RB->marked
    |   lwz CARG2, UPVAL:RB->v
    |  andi. TMP3, TMP3, LJ_GC_BLACK	// isblack(uv)
    |   lbz TMP3, STR:TMP1->marked
    |   lbz TMP2, UPVAL:RB->closed
    |   evstdd STR:TMP1, 0(CARG2)
    |  bne >2
    |1:
    |  ins_next
    |
    |2:  // Check if string is white and ensure upvalue is closed.
    |  andi. TMP3, TMP3, LJ_GC_WHITES	// iswhite(str)
    |   cmplwi cr1, TMP2, 0
    |  crand 4*cr0+eq, 4*cr0+eq, 4*cr1+eq
    |   la CARG1, GG_DISP2G(DISPATCH)
    |  // Crossed a write barrier. Move the barrier forward.
    |  bnel extern lj_gc_barrieruv	// (global_State *g, TValue *tv)
    |  b <1
    break;
  case BC_USETN:
    |  // RA = uvnum*8, RD = num_const*8
    |  lwz LFUNC:RB, FRAME_FUNC(BASE)
    |   srwi RA, RA, 1
    |   addi RA, RA, offsetof(GCfuncL, uvptr)
    |    evlddx TMP0, KBASE, RD
    |  lwzx UPVAL:RB, LFUNC:RB, RA
    |  lwz TMP1, UPVAL:RB->v
    |  evstdd TMP0, 0(TMP1)
    |  ins_next
    break;
  case BC_USETP:
    |  // RA = uvnum*8, RD = primitive_type*8 (~)
    |  lwz LFUNC:RB, FRAME_FUNC(BASE)
    |   srwi RA, RA, 1
    |   addi RA, RA, offsetof(GCfuncL, uvptr)
    |    srwi TMP0, RD, 3
    |  lwzx UPVAL:RB, LFUNC:RB, RA
    |    not TMP0, TMP0
    |  lwz TMP1, UPVAL:RB->v
    |  stw TMP0, 0(TMP1)
    |  ins_next
    break;

  case BC_UCLO:
    |  // RA = level*8, RD = target
    |  lwz TMP1, L->openupval
    |  branch_RD			// Do this first since RD is not saved.
    |   stw BASE, L->base
    |  cmplwi TMP1, 0
    |   mr CARG1, L
    |  beq >1
    |   add CARG2, BASE, RA
    |  bl extern lj_func_closeuv	// (lua_State *L, TValue *level)
    |  lwz BASE, L->base
    |1:
    |  ins_next
    break;

  case BC_FNEW:
    |  // RA = dst*8, RD = proto_const*8 (~) (holding function prototype)
    |  srwi TMP1, RD, 1
    |   stw BASE, L->base
    |  subfic TMP1, TMP1, -4
    |   stw PC, SAVE_PC
    |  lwzx CARG2, KBASE, TMP1		// KBASE-4-tab_const*4
    |   mr CARG1, L
    |  lwz CARG3, FRAME_FUNC(BASE)
    |  // (lua_State *L, GCproto *pt, GCfuncL *parent)
    |  bl extern lj_func_newL_gc
    |  // Returns GCfuncL *.
    |  lwz BASE, L->base
    |  evmergelo LFUNC:CRET1, TISFUNC, LFUNC:CRET1
    |  evstddx LFUNC:CRET1, BASE, RA
    |  ins_next
    break;

  /* -- Table ops --------------------------------------------------------- */

  case BC_TNEW:
  case BC_TDUP:
    |  // RA = dst*8, RD = (hbits|asize)*8 | tab_const*8 (~)
    |  lwz TMP0, DISPATCH_GL(gc.total)(DISPATCH)
    |   mr CARG1, L
    |  lwz TMP1, DISPATCH_GL(gc.threshold)(DISPATCH)
    |   stw BASE, L->base
    |  cmplw TMP0, TMP1
    |   stw PC, SAVE_PC
    |  bge >5
    |1:
    if (op == BC_TNEW) {
      |  rlwinm CARG2, RD, 29, 21, 31
      |  rlwinm CARG3, RD, 18, 27, 31
      |  cmpwi CARG2, 0x7ff
      |   li TMP1, 0x801
      |  iseleq CARG2, TMP1, CARG2
      |  bl extern lj_tab_new  // (lua_State *L, int32_t asize, uint32_t hbits)
      |  // Returns Table *.
    } else {
      |  srwi TMP1, RD, 1
      |  subfic TMP1, TMP1, -4
      |  lwzx CARG2, KBASE, TMP1		// KBASE-4-tab_const*4
      |  bl extern lj_tab_dup  // (lua_State *L, Table *kt)
      |  // Returns Table *.
    }
    |  lwz BASE, L->base
    |  evmergelo TAB:CRET1, TISTAB, TAB:CRET1
    |  evstddx TAB:CRET1, BASE, RA
    |  ins_next
    |5:
    |  mr SAVE0, RD
    |  bl extern lj_gc_step_fixtop  // (lua_State *L)
    |  mr RD, SAVE0
    |  mr CARG1, L
    |  b <1
    break;

  case BC_GGET:
    |  // RA = dst*8, RD = str_const*8 (~)
  case BC_GSET:
    |  // RA = src*8, RD = str_const*8 (~)
    |  lwz LFUNC:TMP2, FRAME_FUNC(BASE)
    |   srwi TMP1, RD, 1
    |  lwz TAB:RB, LFUNC:TMP2->env
    |   subfic TMP1, TMP1, -4
    |   lwzx STR:RC, KBASE, TMP1	// KBASE-4-str_const*4
    if (op == BC_GGET) {
      |  b ->BC_TGETS_Z
    } else {
      |  b ->BC_TSETS_Z
    }
    break;

  case BC_TGETV:
    |  // RA = dst*8, RB = table*8, RC = key*8
    |  evlddx TAB:RB, BASE, RB
    |   evlddx RC, BASE, RC
    |  checktab TAB:RB
    |  checkfail ->vmeta_tgetv
    |  checknum RC
    |  checkfail >5
    |  // Convert number key to integer
    |  efdctsi TMP2, RC
    |   lwz TMP0, TAB:RB->asize
    |  efdcfsi TMP1, TMP2
    |   cmplw cr0, TMP0, TMP2
    |  efdcmpeq cr1, RC, TMP1
    |   lwz TMP1, TAB:RB->array
    |  crand 4*cr0+gt, 4*cr0+gt, 4*cr1+gt
    |   slwi TMP2, TMP2, 3
    |  ble ->vmeta_tgetv		// Integer key and in array part?
    |  evlddx TMP1, TMP1, TMP2
    |  checknil TMP1
    |  checkok >2
    |1:
    |  evstddx TMP1, BASE, RA
    |  ins_next
    |
    |2:  // Check for __index if table value is nil.
    |  lwz TAB:TMP2, TAB:RB->metatable
    |  cmplwi TAB:TMP2, 0
    |  beq <1				// No metatable: done.
    |  lbz TMP0, TAB:TMP2->nomm
    |  andi. TMP0, TMP0, 1<<MM_index
    |  bne <1				// 'no __index' flag set: done.
    |  b ->vmeta_tgetv
    |
    |5:
    |  checkstr STR:RC			// String key?
    |  checkok ->BC_TGETS_Z
    |  b ->vmeta_tgetv
    break;
  case BC_TGETS:
    |  // RA = dst*8, RB = table*8, RC = str_const*8 (~)
    |  evlddx TAB:RB, BASE, RB
    |   srwi TMP1, RC, 1
    |  checktab TAB:RB
    |   subfic TMP1, TMP1, -4
    |   lwzx STR:RC, KBASE, TMP1	// KBASE-4-str_const*4
    |  checkfail ->vmeta_tgets1
    |->BC_TGETS_Z:
    |  // TAB:RB = GCtab *, STR:RC = GCstr *, RA = dst*8
    |  lwz TMP0, TAB:RB->hmask
    |  lwz TMP1, STR:RC->hash
    |  lwz NODE:TMP2, TAB:RB->node
    |   evmergelo STR:RC, TISSTR, STR:RC
    |  and TMP1, TMP1, TMP0		// idx = str->hash & tab->hmask
    |  slwi TMP0, TMP1, 5
    |  slwi TMP1, TMP1, 3
    |  sub TMP1, TMP0, TMP1
    |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
    |1:
    |  evldd TMP0, NODE:TMP2->key
    |   evldd TMP1, NODE:TMP2->val
    |  evcmpeq TMP0, STR:RC
    |  checkanyfail >4
    |   checknil TMP1
    |   checkok >5			// Key found, but nil value?
    |3:
    |   evstddx TMP1, BASE, RA
    |  ins_next
    |
    |4:  // Follow hash chain.
    |  lwz NODE:TMP2, NODE:TMP2->next
    |  cmplwi NODE:TMP2, 0
    |  bne <1
    |  // End of hash chain: key not found, nil result.
    |   evmr TMP1, TISNIL
    |
    |5:  // Check for __index if table value is nil.
    |  lwz TAB:TMP2, TAB:RB->metatable
    |  cmplwi TAB:TMP2, 0
    |  beq <3				// No metatable: done.
    |  lbz TMP0, TAB:TMP2->nomm
    |  andi. TMP0, TMP0, 1<<MM_index
    |  bne <3				// 'no __index' flag set: done.
    |  b ->vmeta_tgets
    break;
  case BC_TGETB:
    |  // RA = dst*8, RB = table*8, RC = index*8
    |  evlddx TAB:RB, BASE, RB
    |   srwi TMP0, RC, 3
    |  checktab TAB:RB
    |  checkfail ->vmeta_tgetb
    |  lwz TMP1, TAB:RB->asize
    |   lwz TMP2, TAB:RB->array
    |  cmplw TMP0, TMP1
    |  bge ->vmeta_tgetb
    |  evlddx TMP1, TMP2, RC
    |  checknil TMP1
    |  checkok >5
    |1:
    |  evstddx TMP1, BASE, RA
    |  ins_next
    |
    |5:  // Check for __index if table value is nil.
    |  lwz TAB:TMP2, TAB:RB->metatable
    |  cmplwi TAB:TMP2, 0
    |  beq <1				// No metatable: done.
    |  lbz TMP2, TAB:TMP2->nomm
    |  andi. TMP2, TMP2, 1<<MM_index
    |  bne <1				// 'no __index' flag set: done.
    |  b ->vmeta_tgetb			// Caveat: preserve TMP0!
    break;

  case BC_TSETV:
    |  // RA = src*8, RB = table*8, RC = key*8
    |  evlddx TAB:RB, BASE, RB
    |   evlddx RC, BASE, RC
    |  checktab TAB:RB
    |  checkfail ->vmeta_tsetv
    |  checknum RC
    |  checkfail >5
    |  // Convert number key to integer
    |  efdctsi TMP2, RC
    |    evlddx SAVE0, BASE, RA
    |   lwz TMP0, TAB:RB->asize
    |  efdcfsi TMP1, TMP2
    |   cmplw cr0, TMP0, TMP2
    |  efdcmpeq cr1, RC, TMP1
    |   lwz TMP1, TAB:RB->array
    |  crand 4*cr0+gt, 4*cr0+gt, 4*cr1+gt
    |   slwi TMP2, TMP2, 3
    |  ble ->vmeta_tsetv		// Integer key and in array part?
    |  evlddx TMP0, TMP1, TMP2
    |  checknil TMP0
    |  checkok >2
    |1:
    |  evstddx SAVE0, TMP1, TMP2
    |  ins_next
    |
    |2:  // Check for __newindex if previous value is nil.
    |  lwz TAB:TMP3, TAB:RB->metatable
    |  cmplwi TAB:TMP3, 0
    |  beq <1				// No metatable: done.
    |  lbz TMP0, TAB:TMP3->nomm
    |  andi. TMP0, TMP0, 1<<MM_newindex
    |  bne <1				// 'no __newindex' flag set: done.
    |  b ->vmeta_tsetv
    |
    |5:
    |  checkstr STR:RC			// String key?
    |  checkok ->BC_TSETS_Z
    |  b ->vmeta_tsetv
    break;
  case BC_TSETS:
    |  // RA = src*8, RB = table*8, RC = str_const*8 (~)
    |  evlddx TAB:RB, BASE, RB
    |   srwi TMP1, RC, 1
    |  checktab TAB:RB
    |   subfic TMP1, TMP1, -4
    |   lwzx STR:RC, KBASE, TMP1	// KBASE-4-str_const*4
    |  checkfail ->vmeta_tsets1
    |->BC_TSETS_Z:
    |  // TAB:RB = GCtab *, STR:RC = GCstr *, RA = src*8
    |  lwz TMP0, TAB:RB->hmask
    |  lwz TMP1, STR:RC->hash
    |  lwz NODE:TMP2, TAB:RB->node
    |    li TMP3, 0
    |   evmergelo STR:RC, TISSTR, STR:RC
    |    stb TMP3, TAB:RB->nomm		// Clear metamethod cache.
    |  and TMP1, TMP1, TMP0		// idx = str->hash & tab->hmask
    |    evlddx SAVE0, BASE, RA
    |  slwi TMP0, TMP1, 5
    |  slwi TMP1, TMP1, 3
    |  sub TMP1, TMP0, TMP1
    |    lbz TMP3, TAB:RB->marked
    |  add NODE:TMP2, NODE:TMP2, TMP1	// node = tab->node + (idx*32-idx*8)
    |1:
    |  evldd TMP0, NODE:TMP2->key
    |   evldd TMP1, NODE:TMP2->val
    |  evcmpeq TMP0, STR:RC
    |  checkanyfail >5
    |   checknil TMP1
    |   checkok >4			// Key found, but nil value?
    |2:
    |  andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
    |    evstdd SAVE0, NODE:TMP2->val
    |  bne >7
    |3:
    |  ins_next
    |
    |4:  // Check for __newindex if previous value is nil.
    |  lwz TAB:TMP1, TAB:RB->metatable
    |  cmplwi TAB:TMP1, 0
    |  beq <2				// No metatable: done.
    |  lbz TMP0, TAB:TMP1->nomm
    |  andi. TMP0, TMP0, 1<<MM_newindex
    |  bne <2				// 'no __newindex' flag set: done.
    |  b ->vmeta_tsets
    |
    |5:  // Follow hash chain.
    |  lwz NODE:TMP2, NODE:TMP2->next
    |  cmplwi NODE:TMP2, 0
    |  bne <1
    |  // End of hash chain: key not found, add a new one.
    |
    |  // But check for __newindex first.
    |  lwz TAB:TMP1, TAB:RB->metatable
    |   la CARG3, DISPATCH_GL(tmptv)(DISPATCH)
    |   stw PC, SAVE_PC
    |   mr CARG1, L
    |  cmplwi TAB:TMP1, 0
    |   stw BASE, L->base
    |  beq >6				// No metatable: continue.
    |  lbz TMP0, TAB:TMP1->nomm
    |  andi. TMP0, TMP0, 1<<MM_newindex
    |  beq ->vmeta_tsets		// 'no __newindex' flag NOT set: check.
    |6:
    |  mr CARG2, TAB:RB
    |  evstdd STR:RC, 0(CARG3)
    |  bl extern lj_tab_newkey		// (lua_State *L, GCtab *t, TValue *k)
    |  // Returns TValue *.
    |  lwz BASE, L->base
    |  evstdd SAVE0, 0(CRET1)
    |  b <3				// No 2nd write barrier needed.
    |
    |7:  // Possible table write barrier for the value. Skip valiswhite check.
    |  barrierback TAB:RB, TMP3, TMP0
    |  b <3
    break;
  case BC_TSETB:
    |  // RA = src*8, RB = table*8, RC = index*8
    |  evlddx TAB:RB, BASE, RB
    |   srwi TMP0, RC, 3
    |  checktab TAB:RB
    |  checkfail ->vmeta_tsetb
    |  lwz TMP1, TAB:RB->asize
    |   lwz TMP2, TAB:RB->array
    |    lbz TMP3, TAB:RB->marked
    |  cmplw TMP0, TMP1
    |   evlddx SAVE0, BASE, RA
    |  bge ->vmeta_tsetb
    |  evlddx TMP1, TMP2, RC
    |  checknil TMP1
    |  checkok >5
    |1:
    |  andi. TMP0, TMP3, LJ_GC_BLACK	// isblack(table)
    |   evstddx SAVE0, TMP2, RC
    |  bne >7
    |2:
    |  ins_next
    |
    |5:  // Check for __newindex if previous value is nil.
    |  lwz TAB:TMP1, TAB:RB->metatable
    |  cmplwi TAB:TMP1, 0
    |  beq <1				// No metatable: done.
    |  lbz TMP1, TAB:TMP1->nomm
    |  andi. TMP1, TMP1, 1<<MM_newindex
    |  bne <1				// 'no __newindex' flag set: done.
    |  b ->vmeta_tsetb			// Caveat: preserve TMP0!
    |
    |7:  // Possible table write barrier for the value. Skip valiswhite check.
    |  barrierback TAB:RB, TMP3, TMP0
    |  b <2
    break;

  case BC_TSETM:
    |  NYI
    break;

  /* -- Calls and vararg handling ----------------------------------------- */

  case BC_CALLM:
    |  // RA = base*8, (RB = (nresults+1)*8,) RC = extra_nargs*8
    |  lwz TMP0, SAVE_MULTRES
    |  add NARGS8:RC, NARGS8:RC, TMP0
    |  // Fall through. Assumes BC_CALL follows.
    break;
  case BC_CALL:
    |  // RA = base*8, (RB = (nresults+1)*8,) RC = (nargs+1)*8
    |  evlddx LFUNC:RB, BASE, RA
    |   mr TMP2, BASE
    |   add BASE, BASE, RA
    |    subi NARGS8:RC, NARGS8:RC, 8
    |  checkfunc LFUNC:RB
    |   addi BASE, BASE, 8
    |  checkfail ->vmeta_call
    |  ins_call
    break;

  case BC_CALLMT:
    |  // RA = base*8, (RB = 0,) RC = extra_nargs*8
    |  lwz TMP0, SAVE_MULTRES
    |  add NARGS8:RC, NARGS8:RC, TMP0
    |  // Fall through. Assumes BC_CALLT follows.
    break;
  case BC_CALLT:
    |  // RA = base*8, (RB = 0,) RC = (nargs+1)*8
    |  evlddx LFUNC:RB, BASE, RA
    |   add RA, BASE, RA
    |    lwz TMP1, FRAME_PC(BASE)
    |    subi NARGS8:RC, NARGS8:RC, 8
    |  checkfunc LFUNC:RB
    |   addi RA, RA, 8
    |  checkfail ->vmeta_callt
    |->BC_CALLT_Z:
    |  andi. TMP0, TMP1, FRAME_TYPE	// Caveat: preserve cr0 until the crand.
    |   lbz TMP3, LFUNC:RB->ffid
    |    xori TMP2, TMP1, FRAME_VARG
    |    cmplwi cr1, NARGS8:RC, 0
    |  bne >7
    |1:
    |  stw LFUNC:RB, FRAME_FUNC(BASE)	// Copy function down, but keep PC.
    |  li TMP2, 0
    |   cmplwi cr7, TMP3, 1		// (> FF_C) Calling a fast function?
    |    beq cr1, >3
    |2:
    |  addi TMP3, TMP2, 8
    |   evlddx TMP0, RA, TMP2
    |  cmplw cr1, TMP3, NARGS8:RC
    |   evstddx TMP0, BASE, TMP2
    |  mr TMP2, TMP3
    |  bne cr1, <2
    |3:
    |  crand 4*cr0+eq, 4*cr0+eq, 4*cr7+gt
    |  beq >5
    |4:
    |  ins_callt
    |
    |5:  // Tailcall to a fast function with a Lua frame below.
    |  lwz INS, -4(TMP1)
    |  decode_RA8 RA, INS
    |  sub TMP1, BASE, RA
    |  lwz LFUNC:TMP1, FRAME_FUNC(TMP1)
    |  lwz TMP1, LFUNC:TMP1->pc
    |  lwz KBASE, PC2PROTO(k)(TMP1)	// Need to prepare KBASE.
    |  b <4
    |
    |7:  // Tailcall from a vararg function.
    |  andi. TMP0, TMP2, FRAME_TYPEP
    |  bne <1				// Vararg frame below?
    |  sub BASE, BASE, TMP2		// Relocate BASE down.
    |  lwz TMP1, FRAME_PC(BASE)
    |  andi. TMP0, TMP1, FRAME_TYPE
    |  b <1
    break;

  case BC_ITERC:
    |  // RA = base*8, (RB = (nresults+1)*8,) RC = (nargs+1)*8 ((2+1)*8)
    |  subi RA, RA, 24			// evldd doesn't support neg. offsets.
    |   mr TMP2, BASE
    |  evlddx LFUNC:RB, BASE, RA
    |   add BASE, BASE, RA
    |   evldd TMP0, 8(BASE)
    |    evldd TMP1, 16(BASE)
    |  evstdd LFUNC:RB, 24(BASE)	// Copy callable.
    |  checkfunc LFUNC:RB
    |   evstdd TMP0, 32(BASE)		// Copy state.
    |     li NARGS8:RC, 16		// Iterators get 2 arguments.
    |    evstdd TMP1, 40(BASE)		// Copy control var.
    |     addi BASE, BASE, 32
    |  checkfail ->vmeta_call
    |  ins_call
    break;

  case BC_VARG:
    |  NYI
    break;

  /* -- Returns ----------------------------------------------------------- */

  case BC_RETM:
    |  // RA = results*8, RD = extra_nresults*8
    |  lwz TMP0, SAVE_MULTRES
    |  add RD, RD, TMP0			// SAVE_MULTRES >= 8, so RD >= 8.
    |  // Fall through. Assumes BC_RET follows.
    break;

  case BC_RET:
    |  // RA = results*8, RD = (nresults+1)*8
    |  lwz PC, FRAME_PC(BASE)
    |   add RA, BASE, RA
    |    stw RD, SAVE_MULTRES
    |1:
    |  andi. TMP0, PC, FRAME_TYPE
    |   xori TMP1, PC, FRAME_VARG
    |  bne ->BC_RETV_Z
    |
    |->BC_RET_Z:
    |  // BASE = base, RA = resultptr, RD = (nresults+1)*8, PC = return
    |   lwz INS, -4(PC)
    |  cmpwi RD, 8
    |   subi TMP2, BASE, 8
    |   decode_RB8 RB, INS
    |  beq >3
    |   li TMP1, 0
    |2:
    |  addi TMP3, TMP1, 8
    |   evlddx TMP0, RA, TMP1
    |  cmpw TMP3, RD
    |   evstddx TMP0, TMP2, TMP1
    |  beq >3
    |  addi TMP1, TMP3, 8
    |   evlddx TMP0, RA, TMP3
    |  cmpw TMP1, RD
    |   evstddx TMP0, TMP2, TMP3
    |  bne <2
    |3:
    |5:
    |  cmplw RB, RD
    |   decode_RA8 RA, INS
    |  bgt >6
    |   sub BASE, TMP2, RA
    |  lwz LFUNC:TMP1, FRAME_FUNC(BASE)
    |  lwz TMP1, LFUNC:TMP1->pc
    |  lwz KBASE, PC2PROTO(k)(TMP1)
    |  ins_next
    |
    |6:  // Fill up results with nil.
    |  subi TMP1, RD, 8
    |   addi RD, RD, 8
    |  evstddx TISNIL, TMP2, TMP1
    |  b <5
    |
    |->BC_RETV_Z:  // Non-standard return case.
    |  andi. TMP2, TMP1, FRAME_TYPEP
    |  bne ->vm_return
    |  // Return from vararg function: relocate BASE down.
    |  sub BASE, BASE, TMP1
    |  lwz PC, FRAME_PC(BASE)
    |  b <1
    break;

  case BC_RET0: case BC_RET1:
    |  // RA = results*8, RD = (nresults+1)*8
    |  lwz PC, FRAME_PC(BASE)
    |   add RA, BASE, RA
    |    stw RD, SAVE_MULTRES
    |  andi. TMP0, PC, FRAME_TYPE
    |   xori TMP1, PC, FRAME_VARG
    |  bne ->BC_RETV_Z
    |
    |  lwz INS, -4(PC)
    |   subi TMP2, BASE, 8
    |  decode_RB8 RB, INS
    if (op == BC_RET1) {
      |  evldd TMP0, 0(RA)
      |  evstdd TMP0, 0(TMP2)
    }
    |5:
    |  cmplw RB, RD
    |   decode_RA8 RA, INS
    |  bgt >6
    |   sub BASE, TMP2, RA
    |  lwz LFUNC:TMP1, FRAME_FUNC(BASE)
    |  lwz TMP1, LFUNC:TMP1->pc
    |  lwz KBASE, PC2PROTO(k)(TMP1)
    |  ins_next
    |
    |6:  // Fill up results with nil.
    |  subi TMP1, RD, 8
    |   addi RD, RD, 8
    |  evstddx TISNIL, TMP2, TMP1
    |  b <5
    break;

  /* -- Loops and branches ------------------------------------------------ */

  case BC_FORL:
#if LJ_HASJIT
    |  hotloop
#endif
    |  // Fall through. Assumes BC_IFORL follows.
    break;

  case BC_JFORI:
  case BC_JFORL:
#if !LJ_HASJIT
    break;
#endif
  case BC_FORI:
  case BC_IFORL:
    |  // RA = base*8, RD = target (after end of loop or start of loop)
    vk = (op == BC_IFORL || op == BC_JFORL);
    |  add RA, BASE, RA
    |  evldd TMP1, FORL_IDX*8(RA)
    |  evldd TMP2, FORL_STOP*8(RA)
    |  evldd TMP3, FORL_STEP*8(RA)
    if (!vk) {
      |  evcmpgtu cr0, TMP1, TISNUM
      |  evcmpgtu cr1, TMP2, TISNUM
      |  evcmpgtu cr7, TMP3, TISNUM
      |  cror 4*cr0+lt, 4*cr0+lt, 4*cr1+lt
      |  cror 4*cr0+lt, 4*cr0+lt, 4*cr7+lt
      |  blt ->vmeta_for
    }
    if (vk) {
      |  efdadd TMP1, TMP1, TMP3
    }
    if (vk) {
      |  evstdd TMP1, FORL_IDX*8(RA)
    }
    |   evcmpgts TMP3, TISNIL
    |  evstdd TMP1, FORL_EXT*8(RA)
    |   bge >2
    |  efdcmpgt TMP1, TMP2
    |1:
    if (op != BC_JFORL) {
      |  srwi RD, RD, 1
      |  add RD, PC, RD
      if (op == BC_JFORI) {
	|  addis PC, RD, -(BCBIAS_J*4 >> 16)
      } else {
	|  addis RD, RD, -(BCBIAS_J*4 >> 16)
      }
    }
    if (op == BC_FORI) {
      |  iselgt PC, RD, PC
    } else if (op == BC_IFORL) {
      |  iselgt PC, PC, RD
    } else {
      |  ble =>BC_JLOOP
    }
    |  ins_next
    |2:
    |  efdcmpgt TMP2, TMP1
    |  b <1
    break;

  case BC_ITERL:
#if LJ_HASJIT
    |  hotloop
#endif
    |  // Fall through. Assumes BC_IITERL follows.
    break;

  case BC_JITERL:
#if !LJ_HASJIT
    break;
#endif
  case BC_IITERL:
    |  // RA = base*8, RD = target
    |  evlddx TMP1, BASE, RA
    |   subi RA, RA, 8
    |  checknil TMP1
    |  checkok >1			// Stop if iterator returned nil.
    if (op == BC_JITERL) {
      |  NYI
    } else {
      |  branch_RD			// Otherwise save control var + branch.
      |  evstddx TMP1, BASE, RA
    }
    |1:
    |  ins_next
    break;

  case BC_LOOP:
    |  // RA = base*8, RD = target (loop extent)
    |  // Note: RA/RD is only used by trace recorder to determine scope/extent
    |  // This opcode does NOT jump, it's only purpose is to detect a hot loop.
#if LJ_HASJIT
    |  hotloop
#endif
    |  // Fall through. Assumes BC_ILOOP follows.
    break;

  case BC_ILOOP:
    |  // RA = base*8, RD = target (loop extent)
    |  ins_next
    break;

  case BC_JLOOP:
#if LJ_HASJIT
    |  NYI
#endif
    break;

  case BC_JMP:
    |  // RA = base*8 (only used by trace recorder), RD = target
    |  branch_RD
    |  ins_next
    break;

  /* -- Function headers -------------------------------------------------- */

  case BC_FUNCF:
#if LJ_HASJIT
    |  hotcall
#endif
  case BC_FUNCV:  /* NYI: compiled vararg functions. */
    |  // Fall through. Assumes BC_IFUNCF/BC_IFUNCV follow.
    break;

  case BC_JFUNCF:
#if !LJ_HASJIT
    break;
#endif
  case BC_IFUNCF:
    |  // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8
    |  lwz TMP2, L->maxstack
    |   lbz TMP1, -4+PC2PROTO(numparams)(PC)
    |    lwz KBASE, -4+PC2PROTO(k)(PC)
    |  cmplw RA, TMP2
    |   slwi TMP1, TMP1, 3
    |  bgt ->vm_growstack_l
    |2:
    |  cmplw NARGS8:RC, TMP1		// Check for missing parameters.
    |  ble >3
    if (op == BC_JFUNCF) {
      |  NYI
    } else {
      |  ins_next
    }
    |
    |3:  // Clear missing parameters.
    |  evstddx TISNIL, BASE, NARGS8:RC
    |  addi NARGS8:RC, NARGS8:RC, 8
    |  b <2
    break;

  case BC_JFUNCV:
#if !LJ_HASJIT
    break;
#endif
    |  NYI  // NYI: compiled vararg functions
    break;  /* NYI: compiled vararg functions. */

  case BC_IFUNCV:
    |  // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8
    |  lwz TMP2, L->maxstack
    |   add TMP1, BASE, RC
    |  add TMP0, RA, RC
    |   stw LFUNC:RB, 4(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.
    |  bge ->vm_growstack_l
    |  lbz TMP2, -4+PC2PROTO(numparams)(PC)
    |   mr RA, BASE
    |   mr RC, TMP1
    |  cmpwi TMP2, 0
    |   addi BASE, TMP1, 8
    |  beq >3
    |1:
    |  cmplw RA, RC			// Less args than parameters?
    |   evldd TMP0, 0(RA)
    |  bge >4
    |    evstdd TISNIL, 0(RA)		// Clear old fixarg slot (help the GC).
    |    addi RA, RA, 8
    |2:
    |  addic. TMP2, TMP2, -1
    |   evstdd TMP0, 8(TMP1)
    |    addi TMP1, TMP1, 8
    |  bne <1
    |3:
    |  ins_next
    |
    |4:  // Clear missing parameters.
    |  evmr TMP0, TISNIL
    |  b <2
    break;

  case BC_FUNCC:
  case BC_FUNCCW:
    |  // BASE = new base, RA = BASE+framesize*8, RB = CFUNC, RC = nargs*8
    if (op == BC_FUNCC) {
      |  lwz TMP0, CFUNC:RB->f
    } else {
      |  lwz TMP0, DISPATCH_GL(wrapf)(DISPATCH)
    }
    |   add TMP1, RA, NARGS8:RC
    |   lwz TMP2, L->maxstack
    |    add RC, BASE, NARGS8:RC
    |   stw BASE, L->base
    |  mtctr TMP0
    |   cmplw TMP1, TMP2
    |    stw RC, L->top
    |     li_vmstate C
    if (op == BC_FUNCCW) {
      |  lwz CARG2, CFUNC:RB->f
    }
    |  mr CARG1, L
    |   bgt ->vm_growstack_c		// Need to grow stack.
    |     st_vmstate
    |  bctrl				// (lua_State *L [, lua_CFunction f])
    |  // Returns nresults.
    |  lwz TMP1, L->top
    |   slwi RD, CRET1, 3
    |  lwz BASE, L->base
    |    li_vmstate INTERP
    |  lwz PC, FRAME_PC(BASE)		// Fetch PC of caller.
    |   sub RA, TMP1, RD		// RA = L->top - nresults*8
    |    st_vmstate
    |  b ->vm_returnc
    break;

  /* ---------------------------------------------------------------------- */

  default:
    fprintf(stderr, "Error: undefined opcode BC_%s\n", bc_names[op]);
    exit(2);
    break;
  }
}

static int build_backend(BuildCtx *ctx)
{
  int op;

  dasm_growpc(Dst, BC__MAX);

  build_subroutines(ctx);

  |.code_op
  for (op = 0; op < BC__MAX; op++)
    build_ins(ctx, (BCOp)op, op);

  return BC__MAX;
}

/* Emit pseudo frame-info for all assembler functions. */
static void emit_asm_debug(BuildCtx *ctx)
{
  /* NYI */
  UNUSED(ctx);
}