diff options
author | Mike Pall <mike> | 2011-11-20 17:56:47 +0100 |
---|---|---|
committer | Mike Pall <mike> | 2011-11-20 17:56:47 +0100 |
commit | f8a4769fb2bd398d526dfe0bdd6814875a2fa39e (patch) | |
tree | 23153ce6405908dd4c6d40d38749254e7ca0cc6e /src/lj_snap.c | |
parent | dc2a39e46d9498c475eaf9ad7c4a8ae61a73094a (diff) | |
download | luajit-f8a4769fb2bd398d526dfe0bdd6814875a2fa39e.tar.gz luajit-f8a4769fb2bd398d526dfe0bdd6814875a2fa39e.tar.bz2 luajit-f8a4769fb2bd398d526dfe0bdd6814875a2fa39e.zip |
Keep maximum frame extent in snap->topslot.
Diffstat (limited to 'src/lj_snap.c')
-rw-r--r-- | src/lj_snap.c | 40 |
1 files changed, 15 insertions, 25 deletions
diff --git a/src/lj_snap.c b/src/lj_snap.c index a2025d88..89f73982 100644 --- a/src/lj_snap.c +++ b/src/lj_snap.c | |||
@@ -63,7 +63,8 @@ static MSize snapshot_slots(jit_State *J, SnapEntry *map, BCReg nslots) | |||
63 | if (ref) { | 63 | if (ref) { |
64 | SnapEntry sn = SNAP_TR(s, tr); | 64 | SnapEntry sn = SNAP_TR(s, tr); |
65 | IRIns *ir = IR(ref); | 65 | IRIns *ir = IR(ref); |
66 | if (ir->o == IR_SLOAD && ir->op1 == s && ref > retf) { | 66 | if (!(sn & (SNAP_CONT|SNAP_FRAME)) && |
67 | ir->o == IR_SLOAD && ir->op1 == s && ref > retf) { | ||
67 | /* No need to snapshot unmodified non-inherited slots. */ | 68 | /* No need to snapshot unmodified non-inherited slots. */ |
68 | if (!(ir->op2 & IRSLOAD_INHERIT)) | 69 | if (!(ir->op2 & IRSLOAD_INHERIT)) |
69 | continue; | 70 | continue; |
@@ -81,16 +82,19 @@ static MSize snapshot_slots(jit_State *J, SnapEntry *map, BCReg nslots) | |||
81 | } | 82 | } |
82 | 83 | ||
83 | /* Add frame links at the end of the snapshot. */ | 84 | /* Add frame links at the end of the snapshot. */ |
84 | static void snapshot_framelinks(jit_State *J, SnapEntry *map) | 85 | static BCReg snapshot_framelinks(jit_State *J, SnapEntry *map) |
85 | { | 86 | { |
86 | cTValue *frame = J->L->base - 1; | 87 | cTValue *frame = J->L->base - 1; |
87 | cTValue *lim = J->L->base - J->baseslot; | 88 | cTValue *lim = J->L->base - J->baseslot; |
89 | cTValue *ftop = frame + funcproto(frame_func(frame))->framesize; | ||
88 | MSize f = 0; | 90 | MSize f = 0; |
89 | map[f++] = SNAP_MKPC(J->pc); /* The current PC is always the first entry. */ | 91 | map[f++] = SNAP_MKPC(J->pc); /* The current PC is always the first entry. */ |
90 | while (frame > lim) { /* Backwards traversal of all frames above base. */ | 92 | while (frame > lim) { /* Backwards traversal of all frames above base. */ |
91 | if (frame_islua(frame)) { | 93 | if (frame_islua(frame)) { |
92 | map[f++] = SNAP_MKPC(frame_pc(frame)); | 94 | map[f++] = SNAP_MKPC(frame_pc(frame)); |
93 | frame = frame_prevl(frame); | 95 | frame = frame_prevl(frame); |
96 | if (frame + funcproto(frame_func(frame))->framesize > ftop) | ||
97 | ftop = frame + funcproto(frame_func(frame))->framesize; | ||
94 | } else if (frame_iscont(frame)) { | 98 | } else if (frame_iscont(frame)) { |
95 | map[f++] = SNAP_MKFTSZ(frame_ftsz(frame)); | 99 | map[f++] = SNAP_MKFTSZ(frame_ftsz(frame)); |
96 | map[f++] = SNAP_MKPC(frame_contpc(frame)); | 100 | map[f++] = SNAP_MKPC(frame_contpc(frame)); |
@@ -102,6 +106,7 @@ static void snapshot_framelinks(jit_State *J, SnapEntry *map) | |||
102 | } | 106 | } |
103 | } | 107 | } |
104 | lua_assert(f == (MSize)(1 + J->framedepth)); | 108 | lua_assert(f == (MSize)(1 + J->framedepth)); |
109 | return (BCReg)(ftop - lim); | ||
105 | } | 110 | } |
106 | 111 | ||
107 | /* Take a snapshot of the current stack. */ | 112 | /* Take a snapshot of the current stack. */ |
@@ -114,7 +119,7 @@ static void snapshot_stack(jit_State *J, SnapShot *snap, MSize nsnapmap) | |||
114 | lj_snap_grow_map(J, nsnapmap + nslots + (MSize)J->framedepth+1); | 119 | lj_snap_grow_map(J, nsnapmap + nslots + (MSize)J->framedepth+1); |
115 | p = &J->cur.snapmap[nsnapmap]; | 120 | p = &J->cur.snapmap[nsnapmap]; |
116 | nent = snapshot_slots(J, p, nslots); | 121 | nent = snapshot_slots(J, p, nslots); |
117 | snapshot_framelinks(J, p + nent); | 122 | snap->topslot = (uint8_t)snapshot_framelinks(J, p + nent); |
118 | snap->mapofs = (uint16_t)nsnapmap; | 123 | snap->mapofs = (uint16_t)nsnapmap; |
119 | snap->ref = (IRRef1)J->cur.nins; | 124 | snap->ref = (IRRef1)J->cur.nins; |
120 | snap->nent = (uint8_t)nent; | 125 | snap->nent = (uint8_t)nent; |
@@ -338,7 +343,6 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr) | |||
338 | SnapEntry *map = &T->snapmap[snap->mapofs]; | 343 | SnapEntry *map = &T->snapmap[snap->mapofs]; |
339 | SnapEntry *flinks = &T->snapmap[snap_nextofs(T, snap)-1]; | 344 | SnapEntry *flinks = &T->snapmap[snap_nextofs(T, snap)-1]; |
340 | int32_t ftsz0; | 345 | int32_t ftsz0; |
341 | BCReg nslots = snap->nslots; | ||
342 | TValue *frame; | 346 | TValue *frame; |
343 | BloomFilter rfilt = snap_renamefilter(T, snapno); | 347 | BloomFilter rfilt = snap_renamefilter(T, snapno); |
344 | const BCIns *pc = snap_pc(map[nent]); | 348 | const BCIns *pc = snap_pc(map[nent]); |
@@ -348,9 +352,9 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr) | |||
348 | setcframe_pc(cframe_raw(L->cframe), pc+1); | 352 | setcframe_pc(cframe_raw(L->cframe), pc+1); |
349 | 353 | ||
350 | /* Make sure the stack is big enough for the slots from the snapshot. */ | 354 | /* Make sure the stack is big enough for the slots from the snapshot. */ |
351 | if (LJ_UNLIKELY(L->base + nslots > tvref(L->maxstack))) { | 355 | if (LJ_UNLIKELY(L->base + snap->topslot > tvref(L->maxstack))) { |
352 | L->top = curr_topL(L); | 356 | L->top = curr_topL(L); |
353 | lj_state_growstack(L, nslots - curr_proto(L)->framesize); | 357 | lj_state_growstack(L, snap->topslot - curr_proto(L)->framesize); |
354 | } | 358 | } |
355 | 359 | ||
356 | /* Fill stack slots with data from the registers and spill slots. */ | 360 | /* Fill stack slots with data from the registers and spill slots. */ |
@@ -364,27 +368,9 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr) | |||
364 | IRIns *ir = &T->ir[ref]; | 368 | IRIns *ir = &T->ir[ref]; |
365 | if (irref_isk(ref)) { /* Restore constant slot. */ | 369 | if (irref_isk(ref)) { /* Restore constant slot. */ |
366 | lj_ir_kvalue(L, o, ir); | 370 | lj_ir_kvalue(L, o, ir); |
367 | if ((sn & (SNAP_CONT|SNAP_FRAME))) { | ||
368 | /* Overwrite tag with frame link. */ | ||
369 | o->fr.tp.ftsz = s != 0 ? (int32_t)*flinks-- : ftsz0; | ||
370 | if ((sn & SNAP_FRAME)) { | ||
371 | GCfunc *fn = ir_kfunc(ir); | ||
372 | if (isluafunc(fn)) { | ||
373 | MSize framesize = funcproto(fn)->framesize; | ||
374 | L->base = ++o; | ||
375 | if (LJ_UNLIKELY(o + framesize > tvref(L->maxstack))) { | ||
376 | ptrdiff_t fsave = savestack(L, frame); | ||
377 | L->top = o; | ||
378 | lj_state_growstack(L, framesize); /* Grow again. */ | ||
379 | frame = restorestack(L, fsave); | ||
380 | } | ||
381 | } | ||
382 | } | ||
383 | } | ||
384 | } else if (!(sn & SNAP_NORESTORE)) { | 371 | } else if (!(sn & SNAP_NORESTORE)) { |
385 | IRType1 t = ir->t; | 372 | IRType1 t = ir->t; |
386 | RegSP rs = ir->prev; | 373 | RegSP rs = ir->prev; |
387 | lua_assert(!(sn & (SNAP_CONT|SNAP_FRAME))); | ||
388 | if (LJ_UNLIKELY(bloomtest(rfilt, ref))) | 374 | if (LJ_UNLIKELY(bloomtest(rfilt, ref))) |
389 | rs = snap_renameref(T, snapno, ref, rs); | 375 | rs = snap_renameref(T, snapno, ref, rs); |
390 | if (ra_hasspill(regsp_spill(rs))) { /* Restore from spill slot. */ | 376 | if (ra_hasspill(regsp_spill(rs))) { /* Restore from spill slot. */ |
@@ -438,10 +424,14 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr) | |||
438 | (uint32_t)ex->gpr[regsp_reg(rs)-RID_MIN_GPR]; | 424 | (uint32_t)ex->gpr[regsp_reg(rs)-RID_MIN_GPR]; |
439 | } | 425 | } |
440 | } | 426 | } |
427 | if ((sn & (SNAP_CONT|SNAP_FRAME))) { /* Overwrite tag with frame link. */ | ||
428 | o->fr.tp.ftsz = s != 0 ? (int32_t)*flinks-- : ftsz0; | ||
429 | L->base = o+1; | ||
430 | } | ||
441 | } | 431 | } |
442 | switch (bc_op(*pc)) { | 432 | switch (bc_op(*pc)) { |
443 | case BC_CALLM: case BC_CALLMT: case BC_RETM: case BC_TSETM: | 433 | case BC_CALLM: case BC_CALLMT: case BC_RETM: case BC_TSETM: |
444 | L->top = frame + nslots; | 434 | L->top = frame + snap->nslots; |
445 | break; | 435 | break; |
446 | default: | 436 | default: |
447 | L->top = curr_topL(L); | 437 | L->top = curr_topL(L); |