gh-mirror/mikepaul-LuaJIT
1
0
Fork 0
mirror of https://github.com/LuaJIT/LuaJIT.git synced 2025-02-07 15:14:08 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Switch to pre-initialized stacks. Drop frame clearing in interpreter.

Browse Source
This commit is contained in:
Mike Pall 2010-02-11 01:21:40 +01:00
parent bb0384c366
commit ab90b8fc2b
7 changed files with 2707 additions and 2696 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1460
src/buildvm_x64.h
View File

File diff suppressed because it is too large Load Diff

1353
src/buildvm_x64win.h
View File

File diff suppressed because it is too large Load Diff

60
src/buildvm_x86.dasc
View File

@ -414,23 +414,25 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
| mov KBASE, [PC+PC2PROTO(k)]
| mov L:RB, SAVE_L
| lea RA, [BASE+RA*8] // Top of frame.
| lea RC, [BASE+NARGS:RC*8-4] // Points to tag of 1st free slot.
| cmp RA, L:RB->maxstack
| ja ->gate_lf_growstack
|9: // Entry point from vararg setup below.
| mov RB, LJ_TNIL
|1: // Clear free slots until top of frame.
| mov [RC], RB
| mov [RC+8], RB
| add RC, 16
| cmp RC, RA
| jb <1
| movzx RA, byte [PC+PC2PROTO(numparams)]
| cmp NARGS:RC, RA // Check for missing parameters.
| jbe >3
|2:
#if LJ_HASJIT
| // NYI: Disabled, until the tracer supports recursion/upcalls/leaves.
| // hotcall RB
#endif
| ins_next
|
|3: // Clear missing parameters.
| mov dword [BASE+NARGS:RC*8-4], LJ_TNIL
| add NARGS:RC, 1
| cmp NARGS:RC, RA // Check for missing parameters.
| jbe <3
| jmp <2
|
|->gate_lv: // Call gate for vararg Lua functions.
| // RA = new base, RB = LFUNC, RC = nargs+1, (BASE = old base), PC = return
| // DISPATCH initialized
@ -443,29 +445,38 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
| movzx RB, byte [PC+PC2PROTO(framesize)]
| lea KBASE, [BASE+RB*8]
| mov L:RB, SAVE_L
| lea RC, [BASE+4]
| cmp KBASE, L:RB->maxstack
| ja ->gate_lv_growstack // Need to grow stack.
| mov RC, BASE
| movzx RB, byte [PC+PC2PROTO(numparams)]
| test RB, RB
| jz >2
|1: // Copy fixarg slots up.
|1: // Copy fixarg slots up to new frame.
| add RA, 8
| cmp RA, BASE
| jnb >2
| jnb >3 // Less args than parameters?
| mov KBASE, [RA-8]
| mov [RC-4], KBASE
| mov KBASE, [RA-4]
| mov [RC], KBASE
| mov KBASE, [RA-4]
| mov [RC+4], KBASE
| add RC, 8
| mov dword [RA-4], LJ_TNIL // Clear old fixarg slot (help the GC).
| sub RB, 1
| jnz <1
|2:
| movzx RA, byte [PC+PC2PROTO(framesize)]
| mov KBASE, [PC+PC2PROTO(k)]
| lea RA, [BASE+RA*8]
| jmp <9
#if LJ_HASJIT
| // NYI: Disabled, until the tracer supports recursion/upcalls/leaves.
| // hotcall RB
#endif
| ins_next
|
|3: // Clear missing parameters.
| mov dword [RC+4], LJ_TNIL
| add RC, 8
| sub RB, 1
| jnz <3
| jmp <2
|
|->gate_cwrap: // Call gate for wrapped C functions.
| // RA = new base, RB = CFUNC, RC = nargs+1, (BASE = old base), PC = return
@ -663,19 +674,12 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
| jmp >1
|
|->gate_lv_growstack: // Grow stack for vararg Lua function.
//XXX
| sub RC, 8
| mov BASE, RA
| mov RA, KBASE
| lea PC, [PROTO:RB+sizeof(GCproto)]
| mov L:RB, SAVE_L
| mov BASE, RA // Drop vararg frame again.
|
|->gate_lf_growstack: // Grow stack for fixarg Lua function.
| // BASE = new base, RA = requested top, RC = top (offset +4 bytes)
| // RB = L, PC = first PC of called function (or anything if C function)
| sub RC, 4 // Adjust top.
| sub RA, BASE
| shr RA, 3 // n = pt->framesize - L->top
| // BASE = new base, RC = nargs+1, RB = L, PC = first PC
| lea RC, [BASE+NARGS:RC*8-8]
| movzx RA, byte [PC+PC2PROTO(framesize)]
| add PC, 4 // Must point after first instruction.
| mov L:RB->base, BASE
| mov L:RB->top, RC

2460
src/buildvm_x86.h
View File

File diff suppressed because it is too large Load Diff

33
src/lj_gc.c
View File

@ -261,7 +261,7 @@ static void gc_traverse_proto(global_State *g, GCproto *pt)
}
/* Traverse the frame structure of a stack. */
static TValue *gc_traverse_frames(global_State *g, lua_State *th)
static MSize gc_traverse_frames(global_State *g, lua_State *th)
{
TValue *frame, *top = th->top-1;
/* Note: extra vararg frame not skipped, marks function twice (harmless). */
@ -274,32 +274,22 @@ static TValue *gc_traverse_frames(global_State *g, lua_State *th)
}
top++; /* Correct bias of -1 (frame == base-1). */
if (top > th->maxstack) top = th->maxstack;
return top;
return (MSize)(top - th->stack); /* Return minimum needed stack size. */
}
/* Traverse a thread object. */
static void gc_traverse_thread(global_State *g, lua_State *th)
{
TValue *o, *lim;
gc_markobj(g, tabref(th->env));
for (o = th->stack+1; o < th->top; o++)
TValue *o, *top = th->top;
for (o = th->stack+1; o < top; o++)
gc_marktv(g, o);
lim = gc_traverse_frames(g, th);
/* Extra cleanup required to avoid this marking problem:
**
** [aa[bb.X| X created.
** [aa[cc| GC called from (small) inner frame, X destroyed.
** [aa....X.| GC called again in (larger) outer frame, X resurrected (ouch).
**
** During GC in step 2 the stack must be cleaned up to the max. frame extent:
**
** ***| Slots cleaned
** [cc| from top of last frame
** [aa......| to max. frame extent.
*/
for (; o <= lim; o++)
setnilV(o);
lj_state_shrinkstack(th, (MSize)(lim - th->stack));
if (g->gc.state == GCSatomic) {
top = th->stack + th->stacksize;
for (; o < top; o++) /* Clear unmarked slots. */
setnilV(o);
}
gc_markobj(g, tabref(th->env));
lj_state_shrinkstack(th, gc_traverse_frames(g, th));
}
/* Propagate one gray object. Traverse it and turn it black. */
@ -524,6 +514,7 @@ static void atomic(global_State *g, lua_State *L)
{
size_t udsize;
g->gc.state = GCSatomic;
gc_mark_uv(g); /* Need to remark open upvalues (the thread may be dead). */
gc_propagate_gray(g); /* Propagate any left-overs. */

4
src/lj_gc.h
View File

@ -9,7 +9,9 @@
#include "lj_obj.h"
/* Garbage collector states. Order matters. */
enum { GCSpause, GCSpropagate, GCSsweepstring, GCSsweep, GCSfinalize };
enum {
GCSpause, GCSpropagate, GCSatomic, GCSsweepstring, GCSsweep, GCSfinalize
};
/* Bitmasks for marked field of GCobj. */
#define LJ_GC_WHITE0 0x01

33
src/lj_state.c
View File

@ -49,14 +49,18 @@
/* Resize stack slots and adjust pointers in state. */
static void resizestack(lua_State *L, MSize n)
{
TValue *oldst = L->stack;
TValue *st, *oldst = L->stack;
ptrdiff_t delta;
MSize oldsize = L->stacksize;
MSize realsize = n + 1 + LJ_STACK_EXTRA;
GCobj *up;
lua_assert((MSize)(L->maxstack-L->stack) == L->stacksize-LJ_STACK_EXTRA-1);
lj_mem_reallocvec(L, L->stack, L->stacksize, realsize, TValue);
delta = (char *)L->stack - (char *)oldst;
L->maxstack = L->stack + n;
st = L->stack;
delta = (char *)st - (char *)oldst;
L->maxstack = st + n;
while (oldsize < realsize) /* Clear new slots. */
setnilV(st + oldsize++);
L->stacksize = realsize;
L->base = (TValue *)((char *)L->base + delta);
L->top = (TValue *)((char *)L->top + delta);
@ -90,13 +94,8 @@ void LJ_FASTCALL lj_state_growstack(lua_State *L, MSize need)
if (L->stacksize > LJ_STACK_MAXEX) /* overflow while handling overflow? */
lj_err_throw(L, LUA_ERRERR);
resizestack(L, L->stacksize + (need > L->stacksize ? need : L->stacksize));
if (L->stacksize > LJ_STACK_MAXEX) {
if (curr_funcisL(L)) { /* Clear slots of incomplete Lua frame. */
TValue *top = curr_topL(L);
while (--top >= L->top) setnilV(top);
}
lj_err_msg(L, LJ_ERR_STKOV); /* ... to allow L->top = curr_topL(L). */
}
if (L->stacksize > LJ_STACK_MAXEX)
lj_err_msg(L, LJ_ERR_STKOV);
}
void LJ_FASTCALL lj_state_growstack1(lua_State *L)
@ -107,13 +106,15 @@ void LJ_FASTCALL lj_state_growstack1(lua_State *L)
/* Allocate basic stack for new state. */
static void stack_init(lua_State *L1, lua_State *L)
{
L1->stack = lj_mem_newvec(L, LJ_STACK_START + LJ_STACK_EXTRA, TValue);
TValue *st, *stend;
L1->stack = st = lj_mem_newvec(L, LJ_STACK_START + LJ_STACK_EXTRA, TValue);
L1->stacksize = LJ_STACK_START + LJ_STACK_EXTRA;
L1->top = L1->stack;
L1->maxstack = L1->stack+(L1->stacksize - LJ_STACK_EXTRA)-1;
setthreadV(L1, L1->top, L1); /* needed for curr_funcisL() on empty stack */
setnilV(L1->top); /* but clear its type */
L1->base = ++L1->top;
stend = st + L1->stacksize;
L1->maxstack = stend - LJ_STACK_EXTRA - 1;
L1->base = L1->top = st+1;
setthreadV(L1, st, L1); /* Needed for curr_funcisL() on empty stack. */
while (st < stend) /* Clear new slots. */
setnilV(st++);
}
/* -- State handling ------------------------------------------------------ */