backend: don't mixup float and int temps

copy propagation only happens when dataclasses match, register allocator
ignores hints if hint register class and instruction class differ, also
add mov between int and float regs in amd64/emit

4 files changed, 29 insertions, 11 deletions

diff --git a/amd64/emit.c b/amd64/emit.c
index 5b3a298..70386be 100644
--- a/amd64/emit.c
+++ b/amd64/emit.c
@@ -382,8 +382,8 @@ static void Xmov(uchar **pcode, enum irclass k, struct oper dst, struct oper src
       {8, PFPR, PFPR, "\xF2\x0F\x10", EN_RR}, /* MOVSD xmm, xmm */
       {8, PFPR, PMEM, "\xF2\x0F\x10", EN_RM}, /* MOVSD xmm, m64 */
       {8, PMEM, PFPR, "\xF2\x0F\x11", EN_MR}, /* MOVSS m64, xmm */
-      {4|8, PGPR, PFPR, "\x66\x0F\x6E", EN_RRX}, /* MOVD/Q r64/32, xmm */
-      {4|8, PFPR, PGPR, "\x66\x0F\x6E", EN_RR}, /* MOVD/Q xmm, r64/32 */
+      {4|8, PFPR, PGPR, "\x66\x0F\x6E", EN_RR},  /* MOVD/Q xmm, r64/32 */
+      {4|8, PGPR, PFPR, "\x66\x0F\x7E", EN_RRX}, /* MOVD/Q r64/32, xmm */
    };
    static const uchar k2off[] = {
       [KI4] = 0,
@@ -646,7 +646,7 @@ gencopy(uchar **pcode, enum irclass cls, struct block *blk, int curi, struct ope
    } else {
       struct oper src = mkimmdatregoper(val);
       if (memcmp(&dst, &src, sizeof dst) != 0)
-         Xmov(pcode, cls, dst, src);
+         Xmov(pcode, cls == KF8 && src.reg < XMM0 ? KI8 : cls, dst, src);
    }
 }
 
diff --git a/regalloc.c b/regalloc.c
index a49330d..ae81fcf 100644
--- a/regalloc.c
+++ b/regalloc.c
@@ -7,7 +7,7 @@ static regset gpregset, fpregset;
 static regset globusage;
 
 #define isfpr(reg) in_range((reg), mctarg->fpr0, mctarg->fpr0 + mctarg->nfpr - 1)
-#define isgpr(reg) (!isfpr(reg))
+#define isgpr(reg) in_range((reg), mctarg->gpr0, mctarg->gpr0 + mctarg->nfpr - 1)
 
 enum { ADEAD, AREG, ASTACK } ;
 struct alloc { ushort t : 2, a : 14; };
@@ -125,7 +125,7 @@ def(struct rega *ra, struct instr *ins, struct block *blk, int curi)
 
          if ((ins->op == Ocopy || ins->inplace) && ins->l.t == RREG) {
             int hint = ins->l.i;
-            if (rstest(ra->m.free, hint)) {
+            if (rstest(ra->m.free, hint) && kisflt(insrescls(*ins)) == isfpr(hint)) {
                take(ra, reg = hint, mkref(RTMP, var));
                assert(ra->m.regs[reg] == var && !rstest(ra->m.free, hint));
             }
@@ -301,7 +301,8 @@ use(struct rega *ra, struct block *blk, int curi, enum op op, int hint, union re
          /* for '%x = copy Rx', hint %x to use Rx */
          hint = ins->l.i;
 
-      if (hint != -1 && !rstest(excl, hint) && rstest(ra->m.free, hint)) {
+      if (hint != -1 && !rstest(excl, hint) && rstest(ra->m.free, hint)
+       && kisflt(insrescls(*ins)) == isfpr(hint)) {
          take(ra, hint, *ref);
          reg = hint;
       } else {
@@ -568,9 +569,9 @@ regalloc(struct function *fn)
    /* setup */
    if (!fpregset || !gpregset) {
       for (int i = 0; i < MAXREGS; ++i) {
-         if (in_range(i, mctarg->fpr0, mctarg->fpr0 + mctarg->nfpr - 1))
+         if (isfpr(i))
             fpregset = rsset(fpregset, i);
-         else if (in_range(i, mctarg->gpr0, mctarg->gpr0 + mctarg->ngpr - 1))
+         else if (isgpr(i))
             gpregset = rsset(gpregset, i);
       }
    }
diff --git a/ssa.c b/ssa.c
index 4a4ed4d..fc4464b 100644
--- a/ssa.c
+++ b/ssa.c
@@ -12,11 +12,17 @@ copyopt(struct function *fn)
          struct use *use, *uend;
          union ref var = mkref(RTMP, blk->ins.p[i]);
          struct instr *ins = &instrtab[var.i];
+         enum irclass k;
 
          if (ins->op == Ocopy) {
-            assert(ins->l.t != RREG);
-
-            replcuses(var, ins->l);
+            union ref arg = ins->l;
+            if (arg.t == RTMP) k = insrescls(instrtab[arg.i]);
+            else if (arg.t == RICON) k = cls2siz[ins->cls] == 4 ? KI4 : KI8;
+            else if (arg.t == RXCON) k = isnumcon(arg) ? conht[arg.i].cls : KPTR;
+            else assert(0);
+            if (ins->cls != k) continue;
+
+            replcuses(var, arg);
             *ins = mkinstr(Onop,0,);
             deluses(var.i);
          }
diff --git a/test/fib.c b/test/fib.c
index eadce03..5fe430a 100644
--- a/test/fib.c
+++ b/test/fib.c
@@ -13,6 +13,16 @@ unsigned fibr(unsigned x) {
    return fibr(x-1) + fibr(x-2);
 }
 
+double fibf(unsigned x) {
+   double r = 0., q = 1.;
+   while (x-- > 1) {
+      double s = r + q;
+      r = q;
+      q = s;
+   }
+   return q;
+}
+
 int atoi(const char *);
 int printf(const char *, ...);
 
@@ -20,6 +30,7 @@ int main(int argc, char **argv) {
    unsigned n = argv[1] ? atoi(argv[1]) : 10;
    printf("fib(%u)  = %u\n", n, fib(n));
    printf("fibr(%u) = %u\n", n, fibr(n));
+   printf("fibf(%u) = %g\n", n, fibf(n));
 }
 
 /* vim:set ts=3 sw=3 expandtab: */