Browse Source

Split off fast function recording to lj_ffrecord.c.

Mike Pall 14 years ago
parent
commit
e7f8cc964e
7 changed files with 888 additions and 825 deletions
  1. 2 1
      src/Makefile
  2. 12 9
      src/Makefile.dep
  3. 803 0
      src/lj_ffrecord.c
  4. 17 0
      src/lj_ffrecord.h
  5. 27 815
      src/lj_record.c
  6. 26 0
      src/lj_record.h
  7. 1 0
      src/ljamalg.c

+ 2 - 1
src/Makefile

@@ -326,7 +326,8 @@ LJCORE_O= lj_gc.o lj_err.o lj_char.o lj_bc.o lj_obj.o \
 	  lj_lex.o lj_parse.o \
 	  lj_ir.o lj_opt_mem.o lj_opt_fold.o lj_opt_narrow.o \
 	  lj_opt_dce.o lj_opt_loop.o \
-	  lj_mcode.o lj_snap.o lj_record.o lj_asm.o lj_trace.o lj_gdbjit.o \
+	  lj_mcode.o lj_snap.o lj_record.o lj_ffrecord.o \
+	  lj_asm.o lj_trace.o lj_gdbjit.o \
 	  lj_ctype.o lj_cdata.o lj_cconv.o lj_cparse.o \
 	  lj_lib.o lj_alloc.o lib_aux.o \
 	  $(LJLIB_O) lib_init.o

+ 12 - 9
src/Makefile.dep

@@ -71,6 +71,10 @@ lj_dispatch.o: lj_dispatch.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
 lj_err.o: lj_err.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h lj_err.h \
  lj_errmsg.h lj_str.h lj_tab.h lj_func.h lj_state.h lj_frame.h lj_bc.h \
  lj_trace.h lj_jit.h lj_ir.h lj_dispatch.h lj_traceerr.h lj_vm.h
+lj_ffrecord.o: lj_ffrecord.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
+ lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_frame.h lj_bc.h lj_ff.h \
+ lj_ffdef.h lj_ir.h lj_jit.h lj_iropt.h lj_trace.h lj_dispatch.h \
+ lj_traceerr.h lj_record.h lj_vm.h lj_recdef.h
 lj_func.o: lj_func.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h lj_gc.h \
  lj_func.h lj_trace.h lj_jit.h lj_ir.h lj_dispatch.h lj_bc.h \
  lj_traceerr.h lj_vm.h
@@ -112,10 +116,9 @@ lj_parse.o: lj_parse.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_func.h lj_state.h \
  lj_bc.h lj_lex.h lj_parse.h lj_vm.h lj_vmevent.h
 lj_record.o: lj_record.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
- lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_state.h lj_frame.h \
- lj_bc.h lj_ff.h lj_ffdef.h lj_ir.h lj_jit.h lj_iropt.h lj_trace.h \
- lj_dispatch.h lj_traceerr.h lj_record.h lj_snap.h lj_asm.h lj_vm.h \
- lj_recdef.h
+ lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_frame.h lj_bc.h lj_ff.h \
+ lj_ffdef.h lj_ir.h lj_jit.h lj_iropt.h lj_trace.h lj_dispatch.h \
+ lj_traceerr.h lj_record.h lj_ffrecord.h lj_snap.h lj_vm.h
 lj_snap.o: lj_snap.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h lj_gc.h \
  lj_state.h lj_frame.h lj_bc.h lj_ir.h lj_jit.h lj_iropt.h lj_trace.h \
  lj_dispatch.h lj_traceerr.h lj_snap.h lj_target.h lj_target_*.h
@@ -148,9 +151,9 @@ ljamalg.o: ljamalg.c lua.h luaconf.h lauxlib.h lj_gc.c lj_obj.h lj_def.h \
  lj_cconv.c lj_cparse.c lj_cparse.h lj_lib.c lj_lib.h lj_ir.c lj_iropt.h \
  lj_opt_mem.c lj_opt_fold.c lj_folddef.h lj_opt_narrow.c lj_opt_dce.c \
  lj_opt_loop.c lj_snap.h lj_mcode.c lj_mcode.h lj_snap.c lj_target.h \
- lj_target_*.h lj_record.c lj_record.h lj_asm.h lj_recdef.h lj_asm.c \
- lj_trace.c lj_gdbjit.h lj_gdbjit.c lj_alloc.c lib_aux.c lib_base.c \
- lualib.h lj_libdef.h lib_math.c lib_string.c lib_table.c lib_io.c \
- lib_os.c lib_package.c lib_debug.c lib_bit.c lib_jit.c lib_ffi.c \
- lib_init.c
+ lj_target_*.h lj_record.c lj_record.h lj_ffrecord.h lj_ffrecord.c \
+ lj_recdef.h lj_asm.c lj_asm.h lj_trace.c lj_gdbjit.h lj_gdbjit.c \
+ lj_alloc.c lib_aux.c lib_base.c lualib.h lj_libdef.h lib_math.c \
+ lib_string.c lib_table.c lib_io.c lib_os.c lib_package.c lib_debug.c \
+ lib_bit.c lib_jit.c lib_ffi.c lib_init.c
 luajit.o: luajit.c lua.h luaconf.h lauxlib.h lualib.h luajit.h lj_arch.h

+ 803 - 0
src/lj_ffrecord.c

@@ -0,0 +1,803 @@
+/*
+** Fast function call recorder.
+** Copyright (C) 2005-2010 Mike Pall. See Copyright Notice in luajit.h
+*/
+
+#define lj_ffrecord_c
+#define LUA_CORE
+
+#include "lj_obj.h"
+
+#if LJ_HASJIT
+
+#include "lj_err.h"
+#include "lj_str.h"
+#include "lj_tab.h"
+#include "lj_frame.h"
+#include "lj_bc.h"
+#include "lj_ff.h"
+#include "lj_ir.h"
+#include "lj_jit.h"
+#include "lj_iropt.h"
+#include "lj_trace.h"
+#include "lj_record.h"
+#include "lj_dispatch.h"
+#include "lj_vm.h"
+
+/* Some local macros to save typing. Undef'd at the end. */
+#define IR(ref)			(&J->cur.ir[(ref)])
+
+/* Pass IR on to next optimization in chain (FOLD). */
+#define emitir(ot, a, b)	(lj_ir_set(J, (ot), (a), (b)), lj_opt_fold(J))
+
+/* -- Fast function recording handlers ------------------------------------ */
+
+/* Conventions for fast function call handlers:
+**
+** The argument slots start at J->base[0]. All of them are guaranteed to be
+** valid and type-specialized references. J->base[J->maxslot] is set to 0
+** as a sentinel. The runtime argument values start at rd->argv[0].
+**
+** In general fast functions should check for presence of all of their
+** arguments and for the correct argument types. Some simplifications
+** are allowed if the interpreter throws instead. But even if recording
+** is aborted, the generated IR must be consistent (no zero-refs).
+**
+** The number of results in rd->nres is set to 1. Handlers that return
+** a different number of results need to override it. A negative value
+** prevents return processing (e.g. for pending calls).
+**
+** Results need to be stored starting at J->base[0]. Return processing
+** moves them to the right slots later.
+**
+** The per-ffid auxiliary data is the value of the 2nd part of the
+** LJLIB_REC() annotation. This allows handling similar functionality
+** in a common handler.
+*/
+
+/* Data used by handlers to record a fast function. */
+typedef struct RecordFFData {
+  TValue *argv;		/* Runtime argument values. */
+  ptrdiff_t nres;	/* Number of returned results (defaults to 1). */
+  uint32_t data;	/* Per-ffid auxiliary data (opcode, literal etc.). */
+} RecordFFData;
+
+/* Type of handler to record a fast function. */
+typedef void (LJ_FASTCALL *RecordFunc)(jit_State *J, RecordFFData *rd);
+
+/* Get runtime value of int argument. */
+static int32_t argv2int(jit_State *J, TValue *o)
+{
+  if (!tvisnum(o) && !(tvisstr(o) && lj_str_tonum(strV(o), o)))
+    lj_trace_err(J, LJ_TRERR_BADTYPE);
+  return lj_num2bit(numV(o));
+}
+
+/* Get runtime value of string argument. */
+static GCstr *argv2str(jit_State *J, TValue *o)
+{
+  if (LJ_LIKELY(tvisstr(o))) {
+    return strV(o);
+  } else {
+    GCstr *s;
+    if (!tvisnum(o))
+      lj_trace_err(J, LJ_TRERR_BADTYPE);
+    s = lj_str_fromnum(J->L, &o->n);
+    setstrV(J->L, o, s);
+    return s;
+  }
+}
+
+/* Return number of results wanted by caller. */
+static ptrdiff_t results_wanted(jit_State *J)
+{
+  TValue *frame = J->L->base-1;
+  if (frame_islua(frame))
+    return (ptrdiff_t)bc_b(frame_pc(frame)[-1]) - 1;
+  else
+    return -1;
+}
+
+/* Throw error for unsupported variant of fast function. */
+LJ_NORET static void recff_nyiu(jit_State *J)
+{
+  setfuncV(J->L, &J->errinfo, J->fn);
+  lj_trace_err_info(J, LJ_TRERR_NYIFFU);
+}
+
+/* Fallback handler for all fast functions that are not recorded (yet). */
+static void LJ_FASTCALL recff_nyi(jit_State *J, RecordFFData *rd)
+{
+  setfuncV(J->L, &J->errinfo, J->fn);
+  lj_trace_err_info(J, LJ_TRERR_NYIFF);
+  UNUSED(rd);
+}
+
+/* C functions can have arbitrary side-effects and are not recorded (yet). */
+static void LJ_FASTCALL recff_c(jit_State *J, RecordFFData *rd)
+{
+  setfuncV(J->L, &J->errinfo, J->fn);
+  lj_trace_err_info(J, LJ_TRERR_NYICF);
+  UNUSED(rd);
+}
+
+/* -- Base library fast functions ----------------------------------------- */
+
+static void LJ_FASTCALL recff_assert(jit_State *J, RecordFFData *rd)
+{
+  /* Arguments already specialized. The interpreter throws for nil/false. */
+  rd->nres = J->maxslot;  /* Pass through all arguments. */
+}
+
+static void LJ_FASTCALL recff_type(jit_State *J, RecordFFData *rd)
+{
+  /* Arguments already specialized. Result is a constant string. Neat, huh? */
+  IRType t = tref_isinteger(J->base[0]) ? IRT_NUM : tref_type(J->base[0]);
+  J->base[0] = lj_ir_kstr(J, strV(&J->fn->c.upvalue[t]));
+  UNUSED(rd);
+}
+
+static void LJ_FASTCALL recff_getmetatable(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  if (tr) {
+    RecordIndex ix;
+    ix.tab = tr;
+    copyTV(J->L, &ix.tabv, &rd->argv[0]);
+    if (lj_record_mm_lookup(J, &ix, MM_metatable))
+      J->base[0] = ix.mobj;
+    else
+      J->base[0] = ix.mt;
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_setmetatable(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  TRef mt = J->base[1];
+  if (tref_istab(tr) && (tref_istab(mt) || (mt && tref_isnil(mt)))) {
+    TRef fref, mtref;
+    RecordIndex ix;
+    ix.tab = tr;
+    copyTV(J->L, &ix.tabv, &rd->argv[0]);
+    lj_record_mm_lookup(J, &ix, MM_metatable); /* Guard for no __metatable. */
+    fref = emitir(IRT(IR_FREF, IRT_PTR), tr, IRFL_TAB_META);
+    mtref = tref_isnil(mt) ? lj_ir_knull(J, IRT_TAB) : mt;
+    emitir(IRT(IR_FSTORE, IRT_TAB), fref, mtref);
+    if (!tref_isnil(mt))
+      emitir(IRT(IR_TBAR, IRT_TAB), tr, 0);
+    J->base[0] = tr;
+    J->needsnap = 1;
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_rawget(jit_State *J, RecordFFData *rd)
+{
+  RecordIndex ix;
+  ix.tab = J->base[0]; ix.key = J->base[1];
+  if (tref_istab(ix.tab) && ix.key) {
+    ix.val = 0; ix.idxchain = 0;
+    settabV(J->L, &ix.tabv, tabV(&rd->argv[0]));
+    copyTV(J->L, &ix.keyv, &rd->argv[1]);
+    J->base[0] = lj_record_idx(J, &ix);
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_rawset(jit_State *J, RecordFFData *rd)
+{
+  RecordIndex ix;
+  ix.tab = J->base[0]; ix.key = J->base[1]; ix.val = J->base[2];
+  if (tref_istab(ix.tab) && ix.key && ix.val) {
+    ix.idxchain = 0;
+    settabV(J->L, &ix.tabv, tabV(&rd->argv[0]));
+    copyTV(J->L, &ix.keyv, &rd->argv[1]);
+    copyTV(J->L, &ix.valv, &rd->argv[2]);
+    lj_record_idx(J, &ix);
+    /* Pass through table at J->base[0] as result. */
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_rawequal(jit_State *J, RecordFFData *rd)
+{
+  TRef tra = J->base[0];
+  TRef trb = J->base[1];
+  if (tra && trb) {
+    int diff = lj_record_objcmp(J, tra, trb, &rd->argv[0], &rd->argv[1]);
+    J->base[0] = diff ? TREF_FALSE : TREF_TRUE;
+  }  /* else: Interpreter will throw. */
+}
+
+/* Determine mode of select() call. */
+int32_t lj_ffrecord_select_mode(jit_State *J, TRef tr, TValue *tv)
+{
+  if (tref_isstr(tr) && *strVdata(tv) == '#') {  /* select('#', ...) */
+    if (strV(tv)->len == 1) {
+      emitir(IRT(IR_EQ, IRT_STR), tr, lj_ir_kstr(J, strV(tv)));
+    } else {
+      TRef trptr = emitir(IRT(IR_STRREF, IRT_PTR), tr, 0);
+      TRef trchar = emitir(IRT(IR_XLOAD, IRT_U8), trptr, IRXLOAD_READONLY);
+      emitir(IRT(IR_EQ, IRT_INT), trchar, lj_ir_kint(J, '#'));
+    }
+    return 0;
+  } else {  /* select(n, ...) */
+    int32_t start = argv2int(J, tv);
+    if (start == 0) lj_trace_err(J, LJ_TRERR_BADTYPE);  /* A bit misleading. */
+    return start;
+  }
+}
+
+static void LJ_FASTCALL recff_select(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  if (tr) {
+    ptrdiff_t start = lj_ffrecord_select_mode(J, tr, &rd->argv[0]);
+    if (start == 0) {  /* select('#', ...) */
+      J->base[0] = lj_ir_kint(J, J->maxslot - 1);
+    } else if (tref_isk(tr)) {  /* select(k, ...) */
+      ptrdiff_t n = (ptrdiff_t)J->maxslot;
+      if (start < 0) start += n;
+      else if (start > n) start = n;
+      rd->nres = n - start;
+      if (start >= 1) {
+	ptrdiff_t i;
+	for (i = 0; i < n - start; i++)
+	  J->base[i] = J->base[start+i];
+      }  /* else: Interpreter will throw. */
+    } else {
+      recff_nyiu(J);
+    }
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_tonumber(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  if (tref_isnumber_str(tr)) {
+    TRef base = J->base[1];
+    if (base) {
+      base = lj_ir_toint(J, base);
+      if (!tref_isk(base) || IR(tref_ref(base))->i != 10)
+	recff_nyiu(J);
+    }
+    if (tref_isstr(tr)) {
+      TValue tmp;
+      if (!lj_str_tonum(strV(&rd->argv[0]), &tmp))
+	recff_nyiu(J);  /* Would need an inverted STRTO for this case. */
+      tr = emitir(IRTG(IR_STRTO, IRT_NUM), tr, 0);
+    }
+  } else {
+    tr = TREF_NIL;
+  }
+  J->base[0] = tr;
+  UNUSED(rd);
+}
+
+static TValue *recff_metacall_cp(lua_State *L, lua_CFunction dummy, void *ud)
+{
+  jit_State *J = (jit_State *)ud;
+  lj_record_tailcall(J, 0, 1);
+  UNUSED(L); UNUSED(dummy);
+  return NULL;
+}
+
+static int recff_metacall(jit_State *J, RecordFFData *rd, MMS mm)
+{
+  RecordIndex ix;
+  ix.tab = J->base[0];
+  copyTV(J->L, &ix.tabv, &rd->argv[0]);
+  if (lj_record_mm_lookup(J, &ix, mm)) {  /* Has metamethod? */
+    int errcode;
+    /* Temporarily insert metamethod below object. */
+    J->base[1] = J->base[0];
+    J->base[0] = ix.mobj;
+    copyTV(J->L, &rd->argv[1], &rd->argv[0]);
+    copyTV(J->L, &rd->argv[0], &ix.mobjv);
+    /* Need to protect lj_record_tailcall because it may throw. */
+    errcode = lj_vm_cpcall(J->L, NULL, J, recff_metacall_cp);
+    /* Always undo Lua stack changes to avoid confusing the interpreter. */
+    copyTV(J->L, &rd->argv[0], &rd->argv[1]);
+    if (errcode)
+      lj_err_throw(J->L, errcode);  /* Propagate errors. */
+    rd->nres = -1;  /* Pending call. */
+    return 1;  /* Tailcalled to metamethod. */
+  }
+  return 0;
+}
+
+static void LJ_FASTCALL recff_tostring(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  if (tref_isstr(tr)) {
+    /* Ignore __tostring in the string base metatable. */
+    /* Pass on result in J->base[0]. */
+  } else if (!recff_metacall(J, rd, MM_tostring)) {
+    if (tref_isnumber(tr)) {
+      J->base[0] = emitir(IRT(IR_TOSTR, IRT_STR), tr, 0);
+    } else if (tref_ispri(tr)) {
+      J->base[0] = lj_ir_kstr(J, strV(&J->fn->c.upvalue[tref_type(tr)]));
+    } else {
+      recff_nyiu(J);
+    }
+  }
+}
+
+static void LJ_FASTCALL recff_ipairs_aux(jit_State *J, RecordFFData *rd)
+{
+  RecordIndex ix;
+  ix.tab = J->base[0];
+  if (tref_istab(ix.tab)) {
+    if (!tvisnum(&rd->argv[1]))  /* No support for string coercion. */
+      lj_trace_err(J, LJ_TRERR_BADTYPE);
+    setnumV(&ix.keyv, numV(&rd->argv[1])+(lua_Number)1);
+    settabV(J->L, &ix.tabv, tabV(&rd->argv[0]));
+    ix.val = 0; ix.idxchain = 0;
+    ix.key = lj_ir_toint(J, J->base[1]);
+    J->base[0] = ix.key = emitir(IRTI(IR_ADD), ix.key, lj_ir_kint(J, 1));
+    J->base[1] = lj_record_idx(J, &ix);
+    rd->nres = tref_isnil(J->base[1]) ? 0 : 2;
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_ipairs(jit_State *J, RecordFFData *rd)
+{
+#ifdef LUAJIT_ENABLE_LUA52COMPAT
+  if (!recff_metacall(J, rd, MM_ipairs))
+#endif
+  {
+    TRef tab = J->base[0];
+    if (tref_istab(tab)) {
+      J->base[0] = lj_ir_kfunc(J, funcV(&J->fn->c.upvalue[0]));
+      J->base[1] = tab;
+      J->base[2] = lj_ir_kint(J, 0);
+      rd->nres = 3;
+    }  /* else: Interpreter will throw. */
+  }
+}
+
+static void LJ_FASTCALL recff_pcall(jit_State *J, RecordFFData *rd)
+{
+  if (J->maxslot >= 1) {
+    lj_record_call(J, 0, J->maxslot - 1);
+    rd->nres = -1;  /* Pending call. */
+  }  /* else: Interpreter will throw. */
+}
+
+static TValue *recff_xpcall_cp(lua_State *L, lua_CFunction dummy, void *ud)
+{
+  jit_State *J = (jit_State *)ud;
+  lj_record_call(J, 1, J->maxslot - 2);
+  UNUSED(L); UNUSED(dummy);
+  return NULL;
+}
+
+static void LJ_FASTCALL recff_xpcall(jit_State *J, RecordFFData *rd)
+{
+  if (J->maxslot >= 2) {
+    TValue argv0, argv1;
+    TRef tmp;
+    int errcode;
+    /* Swap function and traceback. */
+    tmp = J->base[0]; J->base[0] = J->base[1]; J->base[1] = tmp;
+    copyTV(J->L, &argv0, &rd->argv[0]);
+    copyTV(J->L, &argv1, &rd->argv[1]);
+    copyTV(J->L, &rd->argv[0], &argv1);
+    copyTV(J->L, &rd->argv[1], &argv0);
+    /* Need to protect lj_record_call because it may throw. */
+    errcode = lj_vm_cpcall(J->L, NULL, J, recff_xpcall_cp);
+    /* Always undo Lua stack swap to avoid confusing the interpreter. */
+    copyTV(J->L, &rd->argv[0], &argv0);
+    copyTV(J->L, &rd->argv[1], &argv1);
+    if (errcode)
+      lj_err_throw(J->L, errcode);  /* Propagate errors. */
+    rd->nres = -1;  /* Pending call. */
+  }  /* else: Interpreter will throw. */
+}
+
+/* -- Math library fast functions ----------------------------------------- */
+
+static void LJ_FASTCALL recff_math_abs(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  J->base[0] = emitir(IRTN(IR_ABS), tr, lj_ir_knum_abs(J));
+  UNUSED(rd);
+}
+
+/* Record rounding functions math.floor and math.ceil. */
+static void LJ_FASTCALL recff_math_round(jit_State *J, RecordFFData *rd)
+{
+  if (!tref_isinteger(J->base[0]))  /* Pass through integers unmodified. */
+    J->base[0] = emitir(IRTN(IR_FPMATH), lj_ir_tonum(J, J->base[0]), rd->data);
+  /* Note: result is integral (or NaN/Inf), but may not fit into an integer. */
+}
+
+/* Record unary math.* functions, mapped to IR_FPMATH opcode. */
+static void LJ_FASTCALL recff_math_unary(jit_State *J, RecordFFData *rd)
+{
+  J->base[0] = emitir(IRTN(IR_FPMATH), lj_ir_tonum(J, J->base[0]), rd->data);
+}
+
+/* Record binary math.* functions math.atan2 and math.ldexp. */
+static void LJ_FASTCALL recff_math_binary(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  J->base[0] = emitir(IRTN(rd->data), tr, lj_ir_tonum(J, J->base[1]));
+}
+
+/* Record math.asin, math.acos, math.atan. */
+static void LJ_FASTCALL recff_math_atrig(jit_State *J, RecordFFData *rd)
+{
+  TRef y = lj_ir_tonum(J, J->base[0]);
+  TRef x = lj_ir_knum_one(J);
+  uint32_t ffid = rd->data;
+  if (ffid != FF_math_atan) {
+    TRef tmp = emitir(IRTN(IR_MUL), y, y);
+    tmp = emitir(IRTN(IR_SUB), x, tmp);
+    tmp = emitir(IRTN(IR_FPMATH), tmp, IRFPM_SQRT);
+    if (ffid == FF_math_asin) { x = tmp; } else { x = y; y = tmp; }
+  }
+  J->base[0] = emitir(IRTN(IR_ATAN2), y, x);
+}
+
+static void LJ_FASTCALL recff_math_htrig(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  J->base[0] = lj_ir_call(J, rd->data, tr);
+}
+
+static void LJ_FASTCALL recff_math_modf(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = J->base[0];
+  if (tref_isinteger(tr)) {
+    J->base[0] = tr;
+    J->base[1] = lj_ir_kint(J, 0);
+  } else {
+    TRef trt;
+    tr = lj_ir_tonum(J, tr);
+    trt = emitir(IRTN(IR_FPMATH), tr, IRFPM_TRUNC);
+    J->base[0] = trt;
+    J->base[1] = emitir(IRTN(IR_SUB), tr, trt);
+  }
+  rd->nres = 2;
+}
+
+static void LJ_FASTCALL recff_math_degrad(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  TRef trm = lj_ir_knum(J, numV(&J->fn->c.upvalue[0]));
+  J->base[0] = emitir(IRTN(IR_MUL), tr, trm);
+  UNUSED(rd);
+}
+
+static void LJ_FASTCALL recff_math_pow(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  if (!tref_isnumber_str(J->base[1]))
+    lj_trace_err(J, LJ_TRERR_BADTYPE);
+  J->base[0] = lj_opt_narrow_pow(J, tr, J->base[1], &rd->argv[1]);
+  UNUSED(rd);
+}
+
+static void LJ_FASTCALL recff_math_minmax(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tonum(J, J->base[0]);
+  uint32_t op = rd->data;
+  BCReg i;
+  for (i = 1; J->base[i] != 0; i++)
+    tr = emitir(IRTN(op), tr, lj_ir_tonum(J, J->base[i]));
+  J->base[0] = tr;
+}
+
+static void LJ_FASTCALL recff_math_random(jit_State *J, RecordFFData *rd)
+{
+  GCudata *ud = udataV(&J->fn->c.upvalue[0]);
+  TRef tr, one;
+  lj_ir_kgc(J, obj2gco(ud), IRT_UDATA);  /* Prevent collection. */
+  tr = lj_ir_call(J, IRCALL_lj_math_random_step, lj_ir_kptr(J, uddata(ud)));
+  one = lj_ir_knum_one(J);
+  tr = emitir(IRTN(IR_SUB), tr, one);
+  if (J->base[0]) {
+    TRef tr1 = lj_ir_tonum(J, J->base[0]);
+    if (J->base[1]) {  /* d = floor(d*(r2-r1+1.0)) + r1 */
+      TRef tr2 = lj_ir_tonum(J, J->base[1]);
+      tr2 = emitir(IRTN(IR_SUB), tr2, tr1);
+      tr2 = emitir(IRTN(IR_ADD), tr2, one);
+      tr = emitir(IRTN(IR_MUL), tr, tr2);
+      tr = emitir(IRTN(IR_FPMATH), tr, IRFPM_FLOOR);
+      tr = emitir(IRTN(IR_ADD), tr, tr1);
+    } else {  /* d = floor(d*r1) + 1.0 */
+      tr = emitir(IRTN(IR_MUL), tr, tr1);
+      tr = emitir(IRTN(IR_FPMATH), tr, IRFPM_FLOOR);
+      tr = emitir(IRTN(IR_ADD), tr, one);
+    }
+  }
+  J->base[0] = tr;
+  UNUSED(rd);
+}
+
+/* -- Bit library fast functions ------------------------------------------ */
+
+/* Record unary bit.tobit, bit.bnot, bit.bswap. */
+static void LJ_FASTCALL recff_bit_unary(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tobit(J, J->base[0]);
+  J->base[0] = (rd->data == IR_TOBIT) ? tr : emitir(IRTI(rd->data), tr, 0);
+}
+
+/* Record N-ary bit.band, bit.bor, bit.bxor. */
+static void LJ_FASTCALL recff_bit_nary(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tobit(J, J->base[0]);
+  uint32_t op = rd->data;
+  BCReg i;
+  for (i = 1; J->base[i] != 0; i++)
+    tr = emitir(IRTI(op), tr, lj_ir_tobit(J, J->base[i]));
+  J->base[0] = tr;
+}
+
+/* Record bit shifts. */
+static void LJ_FASTCALL recff_bit_shift(jit_State *J, RecordFFData *rd)
+{
+  TRef tr = lj_ir_tobit(J, J->base[0]);
+  TRef tsh = lj_ir_tobit(J, J->base[1]);
+  if (!(rd->data < IR_BROL ? LJ_TARGET_MASKSHIFT : LJ_TARGET_MASKROT) &&
+      !tref_isk(tsh))
+    tsh = emitir(IRTI(IR_BAND), tsh, lj_ir_kint(J, 31));
+  J->base[0] = emitir(IRTI(rd->data), tr, tsh);
+}
+
+/* -- String library fast functions --------------------------------------- */
+
+static void LJ_FASTCALL recff_string_len(jit_State *J, RecordFFData *rd)
+{
+  J->base[0] = emitir(IRTI(IR_FLOAD), lj_ir_tostr(J, J->base[0]), IRFL_STR_LEN);
+  UNUSED(rd);
+}
+
+/* Handle string.byte (rd->data = 0) and string.sub (rd->data = 1). */
+static void LJ_FASTCALL recff_string_range(jit_State *J, RecordFFData *rd)
+{
+  TRef trstr = lj_ir_tostr(J, J->base[0]);
+  TRef trlen = emitir(IRTI(IR_FLOAD), trstr, IRFL_STR_LEN);
+  TRef tr0 = lj_ir_kint(J, 0);
+  TRef trstart, trend;
+  GCstr *str = argv2str(J, &rd->argv[0]);
+  int32_t start, end;
+  if (rd->data) {  /* string.sub(str, start [,end]) */
+    start = argv2int(J, &rd->argv[1]);
+    trstart = lj_ir_toint(J, J->base[1]);
+    trend = J->base[2];
+    if (tref_isnil(trend)) {
+      trend = lj_ir_kint(J, -1);
+      end = -1;
+    } else {
+      trend = lj_ir_toint(J, trend);
+      end = argv2int(J, &rd->argv[2]);
+    }
+  } else {  /* string.byte(str, [,start [,end]]) */
+    if (J->base[1]) {
+      start = argv2int(J, &rd->argv[1]);
+      trstart = lj_ir_toint(J, J->base[1]);
+      trend = J->base[2];
+      if (tref_isnil(trend)) {
+	trend = trstart;
+	end = start;
+      } else {
+	trend = lj_ir_toint(J, trend);
+	end = argv2int(J, &rd->argv[2]);
+      }
+    } else {
+      trend = trstart = lj_ir_kint(J, 1);
+      end = start = 1;
+    }
+  }
+  if (end < 0) {
+    emitir(IRTGI(IR_LT), trend, tr0);
+    trend = emitir(IRTI(IR_ADD), emitir(IRTI(IR_ADD), trlen, trend),
+		   lj_ir_kint(J, 1));
+    end = end+(int32_t)str->len+1;
+  } else if ((MSize)end <= str->len) {
+    emitir(IRTGI(IR_ULE), trend, trlen);
+  } else {
+    emitir(IRTGI(IR_GT), trend, trlen);
+    end = (int32_t)str->len;
+    trend = trlen;
+  }
+  if (start < 0) {
+    emitir(IRTGI(IR_LT), trstart, tr0);
+    trstart = emitir(IRTI(IR_ADD), trlen, trstart);
+    start = start+(int32_t)str->len;
+    emitir(start < 0 ? IRTGI(IR_LT) : IRTGI(IR_GE), trstart, tr0);
+    if (start < 0) {
+      trstart = tr0;
+      start = 0;
+    }
+  } else {
+    if (start == 0) {
+      emitir(IRTGI(IR_EQ), trstart, tr0);
+      trstart = tr0;
+    } else {
+      trstart = emitir(IRTI(IR_ADD), trstart, lj_ir_kint(J, -1));
+      emitir(IRTGI(IR_GE), trstart, tr0);
+      start--;
+    }
+  }
+  if (rd->data) {  /* Return string.sub result. */
+    if (end - start >= 0) {
+      /* Also handle empty range here, to avoid extra traces. */
+      TRef trptr, trslen = emitir(IRTI(IR_SUB), trend, trstart);
+      emitir(IRTGI(IR_GE), trslen, tr0);
+      trptr = emitir(IRT(IR_STRREF, IRT_PTR), trstr, trstart);
+      J->base[0] = emitir(IRT(IR_SNEW, IRT_STR), trptr, trslen);
+    } else {  /* Range underflow: return empty string. */
+      emitir(IRTGI(IR_LT), trend, trstart);
+      J->base[0] = lj_ir_kstr(J, lj_str_new(J->L, strdata(str), 0));
+    }
+  } else {  /* Return string.byte result(s). */
+    ptrdiff_t i, len = end - start;
+    if (len > 0) {
+      TRef trslen = emitir(IRTI(IR_SUB), trend, trstart);
+      emitir(IRTGI(IR_EQ), trslen, lj_ir_kint(J, (int32_t)len));
+      if (J->baseslot + len > LJ_MAX_JSLOTS)
+	lj_trace_err_info(J, LJ_TRERR_STACKOV);
+      rd->nres = len;
+      for (i = 0; i < len; i++) {
+	TRef tmp = emitir(IRTI(IR_ADD), trstart, lj_ir_kint(J, (int32_t)i));
+	tmp = emitir(IRT(IR_STRREF, IRT_PTR), trstr, tmp);
+	J->base[i] = emitir(IRT(IR_XLOAD, IRT_U8), tmp, IRXLOAD_READONLY);
+      }
+    } else {  /* Empty range or range underflow: return no results. */
+      emitir(IRTGI(IR_LE), trend, trstart);
+      rd->nres = 0;
+    }
+  }
+}
+
+/* -- Table library fast functions ---------------------------------------- */
+
+static void LJ_FASTCALL recff_table_getn(jit_State *J, RecordFFData *rd)
+{
+  if (tref_istab(J->base[0]))
+    J->base[0] = lj_ir_call(J, IRCALL_lj_tab_len, J->base[0]);
+  /* else: Interpreter will throw. */
+  UNUSED(rd);
+}
+
+static void LJ_FASTCALL recff_table_remove(jit_State *J, RecordFFData *rd)
+{
+  TRef tab = J->base[0];
+  rd->nres = 0;
+  if (tref_istab(tab)) {
+    if (!J->base[1] || tref_isnil(J->base[1])) {  /* Simple pop: t[#t] = nil */
+      TRef trlen = lj_ir_call(J, IRCALL_lj_tab_len, tab);
+      GCtab *t = tabV(&rd->argv[0]);
+      MSize len = lj_tab_len(t);
+      emitir(IRTGI(len ? IR_NE : IR_EQ), trlen, lj_ir_kint(J, 0));
+      if (len) {
+	RecordIndex ix;
+	ix.tab = tab;
+	ix.key = trlen;
+	settabV(J->L, &ix.tabv, t);
+	setintV(&ix.keyv, len);
+	ix.idxchain = 0;
+	if (results_wanted(J) != 0) {  /* Specialize load only if needed. */
+	  ix.val = 0;
+	  J->base[0] = lj_record_idx(J, &ix);  /* Load previous value. */
+	  rd->nres = 1;
+	  /* Assumes ix.key/ix.tab is not modified for raw lj_record_idx(). */
+	}
+	ix.val = TREF_NIL;
+	lj_record_idx(J, &ix);  /* Remove value. */
+      }
+    } else {  /* Complex case: remove in the middle. */
+      recff_nyiu(J);
+    }
+  }  /* else: Interpreter will throw. */
+}
+
+static void LJ_FASTCALL recff_table_insert(jit_State *J, RecordFFData *rd)
+{
+  RecordIndex ix;
+  ix.tab = J->base[0];
+  ix.val = J->base[1];
+  rd->nres = 0;
+  if (tref_istab(ix.tab) && ix.val) {
+    if (!J->base[2]) {  /* Simple push: t[#t+1] = v */
+      TRef trlen = lj_ir_call(J, IRCALL_lj_tab_len, ix.tab);
+      GCtab *t = tabV(&rd->argv[0]);
+      ix.key = emitir(IRTI(IR_ADD), trlen, lj_ir_kint(J, 1));
+      settabV(J->L, &ix.tabv, t);
+      setintV(&ix.keyv, lj_tab_len(t) + 1);
+      ix.idxchain = 0;
+      lj_record_idx(J, &ix);  /* Set new value. */
+    } else {  /* Complex case: insert in the middle. */
+      recff_nyiu(J);
+    }
+  }  /* else: Interpreter will throw. */
+}
+
+/* -- I/O library fast functions ------------------------------------------ */
+
+/* Get FILE* for I/O function. Any I/O error aborts recording, so there's
+** no need to encode the alternate cases for any of the guards.
+*/
+static TRef recff_io_fp(jit_State *J, uint32_t id)
+{
+  TRef tr, ud, fp;
+  if (id) {  /* io.func() */
+    tr = lj_ir_kptr(J, &J2G(J)->gcroot[id]);
+    ud = emitir(IRT(IR_XLOAD, IRT_UDATA), tr, 0);
+  } else {  /* fp:method() */
+    ud = J->base[0];
+    if (!tref_isudata(ud))
+      lj_trace_err(J, LJ_TRERR_BADTYPE);
+    tr = emitir(IRT(IR_FLOAD, IRT_U8), ud, IRFL_UDATA_UDTYPE);
+    emitir(IRTGI(IR_EQ), tr, lj_ir_kint(J, UDTYPE_IO_FILE));
+  }
+  fp = emitir(IRT(IR_FLOAD, IRT_LIGHTUD), ud, IRFL_UDATA_FILE);
+  emitir(IRTG(IR_NE, IRT_LIGHTUD), fp, lj_ir_knull(J, IRT_LIGHTUD));
+  return fp;
+}
+
+static void LJ_FASTCALL recff_io_write(jit_State *J, RecordFFData *rd)
+{
+  TRef fp = recff_io_fp(J, rd->data);
+  TRef zero = lj_ir_kint(J, 0);
+  TRef one = lj_ir_kint(J, 1);
+  ptrdiff_t i = rd->data == 0 ? 1 : 0;
+  for (; J->base[i]; i++) {
+    TRef str = lj_ir_tostr(J, J->base[i]);
+    TRef buf = emitir(IRT(IR_STRREF, IRT_PTR), str, zero);
+    TRef len = emitir(IRTI(IR_FLOAD), str, IRFL_STR_LEN);
+    if (tref_isk(len) && IR(tref_ref(len))->i == 1) {
+      TRef tr = emitir(IRT(IR_XLOAD, IRT_U8), buf, IRXLOAD_READONLY);
+      tr = lj_ir_call(J, IRCALL_fputc, tr, fp);
+      if (results_wanted(J) != 0)  /* Check result only if not ignored. */
+	emitir(IRTGI(IR_NE), tr, lj_ir_kint(J, -1));
+    } else {
+      TRef tr = lj_ir_call(J, IRCALL_fwrite, buf, one, len, fp);
+      if (results_wanted(J) != 0)  /* Check result only if not ignored. */
+	emitir(IRTGI(IR_EQ), tr, len);
+    }
+  }
+  J->base[0] = TREF_TRUE;
+}
+
+static void LJ_FASTCALL recff_io_flush(jit_State *J, RecordFFData *rd)
+{
+  TRef fp = recff_io_fp(J, rd->data);
+  TRef tr = lj_ir_call(J, IRCALL_fflush, fp);
+  if (results_wanted(J) != 0)  /* Check result only if not ignored. */
+    emitir(IRTGI(IR_EQ), tr, lj_ir_kint(J, 0));
+  J->base[0] = TREF_TRUE;
+}
+
+/* -- Record calls to fast functions -------------------------------------- */
+
+#include "lj_recdef.h"
+
+static uint32_t recdef_lookup(GCfunc *fn)
+{
+  if (fn->c.ffid < sizeof(recff_idmap)/sizeof(recff_idmap[0]))
+    return recff_idmap[fn->c.ffid];
+  else
+    return 0;
+}
+
+/* Record entry to a fast function or C function. */
+void lj_ffrecord_func(jit_State *J)
+{
+  RecordFFData rd;
+  uint32_t m = recdef_lookup(J->fn);
+  rd.data = m & 0xff;
+  rd.nres = 1;  /* Default is one result. */
+  rd.argv = J->L->base;
+  J->base[J->maxslot] = 0;  /* Mark end of arguments. */
+  (recff_func[m >> 8])(J, &rd);  /* Call recff_* handler. */
+  if (rd.nres >= 0)
+    lj_record_ret(J, 0, rd.nres);
+}
+
+#undef IR
+#undef emitir
+
+#endif

+ 17 - 0
src/lj_ffrecord.h

@@ -0,0 +1,17 @@
+/*
+** Fast function call recorder.
+** Copyright (C) 2005-2010 Mike Pall. See Copyright Notice in luajit.h
+*/
+
+#ifndef _LJ_FFRECORD_H
+#define _LJ_FFRECORD_H
+
+#include "lj_obj.h"
+#include "lj_jit.h"
+
+#if LJ_HASJIT
+LJ_FUNC int32_t lj_ffrecord_select_mode(jit_State *J, TRef tr, TValue *tv);
+LJ_FUNC void lj_ffrecord_func(jit_State *J);
+#endif
+
+#endif

File diff suppressed because it is too large
+ 27 - 815
src/lj_record.c


+ 26 - 0
src/lj_record.h

@@ -10,6 +10,32 @@
 #include "lj_jit.h"
 
 #if LJ_HASJIT
+/* Context for recording an indexed load/store. */
+typedef struct RecordIndex {
+  TValue tabv;		/* Runtime value of table (or indexed object). */
+  TValue keyv;		/* Runtime value of key. */
+  TValue valv;		/* Runtime value of stored value. */
+  TValue mobjv;		/* Runtime value of metamethod object. */
+  GCtab *mtv;		/* Runtime value of metatable object. */
+  cTValue *oldv;	/* Runtime value of previously stored value. */
+  TRef tab;		/* Table (or indexed object) reference. */
+  TRef key;		/* Key reference. */
+  TRef val;		/* Value reference for a store or 0 for a load. */
+  TRef mt;		/* Metatable reference. */
+  TRef mobj;		/* Metamethod object reference. */
+  int idxchain;		/* Index indirections left or 0 for raw lookup. */
+} RecordIndex;
+
+LJ_FUNC int lj_record_objcmp(jit_State *J, TRef a, TRef b,
+			     cTValue *av, cTValue *bv);
+
+LJ_FUNC void lj_record_call(jit_State *J, BCReg func, ptrdiff_t nargs);
+LJ_FUNC void lj_record_tailcall(jit_State *J, BCReg func, ptrdiff_t nargs);
+LJ_FUNC void lj_record_ret(jit_State *J, BCReg rbase, ptrdiff_t gotresults);
+
+LJ_FUNC int lj_record_mm_lookup(jit_State *J, RecordIndex *ix, MMS mm);
+LJ_FUNC TRef lj_record_idx(jit_State *J, RecordIndex *ix);
+
 LJ_FUNC void lj_record_ins(jit_State *J);
 LJ_FUNC void lj_record_setup(jit_State *J);
 #endif

+ 1 - 0
src/ljamalg.c

@@ -54,6 +54,7 @@
 #include "lj_mcode.c"
 #include "lj_snap.c"
 #include "lj_record.c"
+#include "lj_ffrecord.c"
 #include "lj_asm.c"
 #include "lj_trace.c"
 #include "lj_gdbjit.c"

Some files were not shown because too many files changed in this diff