summaryrefslogtreecommitdiff
path: root/src/cmd/5g/ggen.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/cmd/5g/ggen.c')
-rw-r--r--src/cmd/5g/ggen.c195
1 files changed, 126 insertions, 69 deletions
diff --git a/src/cmd/5g/ggen.c b/src/cmd/5g/ggen.c
index 040c3d2a9..fb32c2f36 100644
--- a/src/cmd/5g/ggen.c
+++ b/src/cmd/5g/ggen.c
@@ -9,78 +9,114 @@
#include "gg.h"
#include "opt.h"
-static Prog* appendp(Prog*, int, int, int, int32, int, int, int32);
+static Prog* appendpp(Prog*, int, int, int, int32, int, int, int32);
+static Prog *zerorange(Prog *p, vlong frame, vlong lo, vlong hi, uint32 *r0);
void
-defframe(Prog *ptxt, Bvec *bv)
+defframe(Prog *ptxt)
{
- int i, j, first;
- uint32 frame;
- Prog *p, *p1;
-
+ uint32 frame, r0;
+ Prog *p;
+ vlong hi, lo;
+ NodeList *l;
+ Node *n;
+
// fill in argument size
ptxt->to.type = D_CONST2;
ptxt->to.offset2 = rnd(curfn->type->argwid, widthptr);
// fill in final stack size
- if(stksize > maxstksize)
- maxstksize = stksize;
- frame = rnd(maxstksize+maxarg, widthptr);
+ frame = rnd(stksize+maxarg, widthptr);
ptxt->to.offset = frame;
- maxstksize = 0;
-
- // insert code to clear pointered part of the frame,
+
+ // insert code to contain ambiguously live variables
// so that garbage collector only sees initialized values
// when it looks for pointers.
p = ptxt;
- while(p->link->as == AFUNCDATA || p->link->as == APCDATA || p->link->as == ATYPE)
- p = p->link;
- if(stkzerosize >= 8*widthptr) {
- p = appendp(p, AMOVW, D_CONST, NREG, 0, D_REG, 0, 0);
- p = appendp(p, AADD, D_CONST, NREG, 4+frame-stkzerosize, D_REG, 1, 0);
+ lo = hi = 0;
+ r0 = 0;
+ for(l=curfn->dcl; l != nil; l = l->next) {
+ n = l->n;
+ if(!n->needzero)
+ continue;
+ if(n->class != PAUTO)
+ fatal("needzero class %d", n->class);
+ if(n->type->width % widthptr != 0 || n->xoffset % widthptr != 0 || n->type->width == 0)
+ fatal("var %lN has size %d offset %d", n, (int)n->type->width, (int)n->xoffset);
+ if(lo != hi && n->xoffset + n->type->width >= lo - 2*widthptr) {
+ // merge with range we already have
+ lo = rnd(n->xoffset, widthptr);
+ continue;
+ }
+ // zero old range
+ p = zerorange(p, frame, lo, hi, &r0);
+
+ // set new range
+ hi = n->xoffset + n->type->width;
+ lo = n->xoffset;
+ }
+ // zero final range
+ zerorange(p, frame, lo, hi, &r0);
+}
+
+static Prog*
+zerorange(Prog *p, vlong frame, vlong lo, vlong hi, uint32 *r0)
+{
+ vlong cnt, i;
+ Prog *p1;
+ Node *f;
+
+ cnt = hi - lo;
+ if(cnt == 0)
+ return p;
+ if(*r0 == 0) {
+ p = appendpp(p, AMOVW, D_CONST, NREG, 0, D_REG, 0, 0);
+ *r0 = 1;
+ }
+ if(cnt < 4*widthptr) {
+ for(i = 0; i < cnt; i += widthptr)
+ p = appendpp(p, AMOVW, D_REG, 0, 0, D_OREG, REGSP, 4+frame+lo+i);
+ } else if(cnt <= 128*widthptr) {
+ p = appendpp(p, AADD, D_CONST, NREG, 4+frame+lo, D_REG, 1, 0);
p->reg = REGSP;
- p = appendp(p, AADD, D_CONST, NREG, stkzerosize, D_REG, 2, 0);
+ p = appendpp(p, ADUFFZERO, D_NONE, NREG, 0, D_OREG, NREG, 0);
+ f = sysfunc("duffzero");
+ naddr(f, &p->to, 1);
+ afunclit(&p->to, f);
+ p->to.offset = 4*(128-cnt/widthptr);
+ } else {
+ p = appendpp(p, AADD, D_CONST, NREG, 4+frame+lo, D_REG, 1, 0);
+ p->reg = REGSP;
+ p = appendpp(p, AADD, D_CONST, NREG, cnt, D_REG, 2, 0);
p->reg = 1;
- p1 = p = appendp(p, AMOVW, D_REG, 0, 0, D_OREG, 1, 4);
+ p1 = p = appendpp(p, AMOVW, D_REG, 0, 0, D_OREG, 1, 4);
p->scond |= C_PBIT;
- p = appendp(p, ACMP, D_REG, 1, 0, D_NONE, 0, 0);
+ p = appendpp(p, ACMP, D_REG, 1, 0, D_NONE, 0, 0);
p->reg = 2;
- p = appendp(p, ABNE, D_NONE, NREG, 0, D_BRANCH, NREG, 0);
+ p = appendpp(p, ABNE, D_NONE, NREG, 0, D_BRANCH, NREG, 0);
patch(p, p1);
- } else {
- first = 1;
- j = (stkptrsize - stkzerosize)/widthptr * 2;
- for(i=0; i<stkzerosize; i+=widthptr) {
- if(bvget(bv, j) || bvget(bv, j+1)) {
- if(first) {
- p = appendp(p, AMOVW, D_CONST, NREG, 0, D_REG, 0, 0);
- first = 0;
- }
- p = appendp(p, AMOVW, D_REG, 0, 0, D_OREG, REGSP, 4+frame-stkzerosize+i);
- }
- j += 2;
- }
}
+ return p;
}
-static Prog*
-appendp(Prog *p, int as, int ftype, int freg, int32 foffset, int ttype, int treg, int32 toffset)
-{
- Prog *q;
-
- q = mal(sizeof(*q));
- clearp(q);
- q->as = as;
- q->lineno = p->lineno;
- q->from.type = ftype;
- q->from.reg = freg;
- q->from.offset = foffset;
- q->to.type = ttype;
- q->to.reg = treg;
- q->to.offset = toffset;
- q->link = p->link;
- p->link = q;
- return q;
+static Prog*
+appendpp(Prog *p, int as, int ftype, int freg, int32 foffset, int ttype, int treg, int32 toffset)
+{
+ Prog *q;
+
+ q = mal(sizeof(*q));
+ clearp(q);
+ q->as = as;
+ q->lineno = p->lineno;
+ q->from.type = ftype;
+ q->from.reg = freg;
+ q->from.offset = foffset;
+ q->to.type = ttype;
+ q->to.reg = treg;
+ q->to.offset = toffset;
+ q->link = p->link;
+ p->link = q;
+ return q;
}
// Sweep the prog list to mark any used nodes.
@@ -88,13 +124,13 @@ void
markautoused(Prog* p)
{
for (; p; p = p->link) {
- if (p->as == ATYPE)
+ if (p->as == ATYPE || p->as == AVARDEF || p->as == AVARKILL)
continue;
- if (p->from.name == D_AUTO && p->from.node)
+ if (p->from.node)
p->from.node->used = 1;
- if (p->to.name == D_AUTO && p->to.node)
+ if (p->to.node)
p->to.node->used = 1;
}
}
@@ -110,6 +146,16 @@ fixautoused(Prog* p)
*lp = p->link;
continue;
}
+ if ((p->as == AVARDEF || p->as == AVARKILL) && p->to.node && !p->to.node->used) {
+ // Cannot remove VARDEF instruction, because - unlike TYPE handled above -
+ // VARDEFs are interspersed with other code, and a jump might be using the
+ // VARDEF as a target. Replace with a no-op instead. A later pass will remove
+ // the no-ops.
+ p->to.type = D_NONE;
+ p->to.node = N;
+ p->as = ANOP;
+ continue;
+ }
if (p->from.name == D_AUTO && p->from.node)
p->from.offset += p->from.node->stkdelta;
@@ -245,7 +291,9 @@ ginscall(Node *f, int proc)
nodconst(&con, types[TINT32], 0);
p = gins(ACMP, &con, N);
p->reg = 0;
- patch(gbranch(ABNE, T, -1), retpc);
+ p = gbranch(ABEQ, T, +1);
+ cgen_ret(N);
+ patch(p, pc);
}
break;
}
@@ -459,16 +507,16 @@ cgen_ret(Node *n)
{
Prog *p;
- genlist(n->list); // copy out args
- if(hasdefer || curfn->exit) {
- gjmp(retpc);
- return;
- }
+ if(n != N)
+ genlist(n->list); // copy out args
+ if(hasdefer)
+ ginscall(deferreturn, 0);
+ genlist(curfn->exit);
p = gins(ARET, N, N);
- if(n->op == ORETJMP) {
+ if(n != N && n->op == ORETJMP) {
p->to.name = D_EXTERN;
p->to.type = D_CONST;
- p->to.sym = n->left->sym;
+ p->to.sym = linksym(n->left->sym);
}
}
@@ -816,14 +864,13 @@ void
clearfat(Node *nl)
{
uint32 w, c, q;
- Node dst, nc, nz, end;
+ Node dst, nc, nz, end, r0, r1, *f;
Prog *p, *pl;
/* clear a fat object */
if(debug['g'])
dump("\nclearfat", nl);
-
w = nl->type->width;
// Avoid taking the address for simple enough types.
if(componentgen(N, nl))
@@ -832,13 +879,17 @@ clearfat(Node *nl)
c = w % 4; // bytes
q = w / 4; // quads
- regalloc(&dst, types[tptr], N);
+ r0.op = OREGISTER;
+ r0.val.u.reg = REGALLOC_R0;
+ r1.op = OREGISTER;
+ r1.val.u.reg = REGALLOC_R0 + 1;
+ regalloc(&dst, types[tptr], &r1);
agen(nl, &dst);
nodconst(&nc, types[TUINT32], 0);
- regalloc(&nz, types[TUINT32], 0);
+ regalloc(&nz, types[TUINT32], &r0);
cgen(&nc, &nz);
- if(q >= 4) {
+ if(q > 128) {
regalloc(&end, types[tptr], N);
p = gins(AMOVW, &dst, &end);
p->from.type = D_CONST;
@@ -855,6 +906,12 @@ clearfat(Node *nl)
patch(gbranch(ABNE, T, 0), pl);
regfree(&end);
+ } else if(q >= 4) {
+ f = sysfunc("duffzero");
+ p = gins(ADUFFZERO, N, f);
+ afunclit(&p->to, f);
+ // 4 and 128 = magic constants: see ../../pkg/runtime/asm_arm.s
+ p->to.offset = 4*(128-q);
} else
while(q > 0) {
p = gins(AMOVW, &nz, &dst);
@@ -901,7 +958,7 @@ expandchecks(Prog *firstp)
p1->link = p->link;
p->link = p1;
p1->lineno = p->lineno;
- p1->loc = 9999;
+ p1->pc = 9999;
p1->as = AMOVW;
p1->from.type = D_REG;
p1->from.reg = reg;
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-17 20:07:18 +0000