diff --git a/mach/powerpc/ncg/mach.c b/mach/powerpc/ncg/mach.c
index 31ca2d371..32e05daa3 100644
--- a/mach/powerpc/ncg/mach.c
+++ b/mach/powerpc/ncg/mach.c
@@ -106,6 +106,7 @@ char *segname[] = {
 static long savedf[32];
 static long savedi[32];
 static int savedtop;
+static unsigned long lfs_set;
 
 /* Calculate the register score of a local variable. */
 int
@@ -118,7 +119,7 @@ regscore(long offset, int size, int type, int frequency, int totype)
 			/* Don't put reg_float in reg_any. */
 			if (totype != reg_float)
 				return -1;
-			assert(size == 8);
+			assert(size == 4 || size == 8);
 			break;
 		default:
 			assert(size == 4);
@@ -155,6 +156,8 @@ i_regsave(void)
 
 	/* Set top of register save area, relative to fp. */
 	savedtop = -framesize;
+
+	lfs_set = 0;  /* empty set */
 }
 
 /* Mark a register as being saved. */
@@ -168,6 +171,8 @@ regsave(const char* regname, long offset, int size)
 		case 'f':
 			savedf[regnum] = offset;
 			framesize += 8;
+			if (size == 4)
+				lfs_set |= (1U << regnum);
 			break;
 		case 'r':
 			savedi[regnum] = offset;
@@ -224,9 +229,14 @@ f_regsave(void)
 	emit_prolog();
 	saveloadregs("stw", "stmw", "stfd");
 
+	/*
+	 * Register variables with offset >= 0 must load an argument
+	 * from that offset.
+	 */
 	for (reg = 31; reg >= 0; reg--)
 		if (savedf[reg] >= 0)
-			fprintf(codefile, "lfd f%d, %ld(fp)\n",
+			fprintf(codefile, "%s f%d, %ld(fp)\n",
+				(lfs_set & (1U << reg)) ? "lfs" : "lfd",
 				reg, savedf[reg]);
 
 	for (reg = 31; reg >= 0; reg--)
diff --git a/mach/powerpc/ncg/table b/mach/powerpc/ncg/table
index 7c6ffbd49..23b904264 100644
--- a/mach/powerpc/ncg/table
+++ b/mach/powerpc/ncg/table
@@ -77,6 +77,14 @@ REGISTERS
 	fs13("f13")=f13
 	  : FSREG.
 
+	/* reglap: reg_float may have subregister of different size */
+	fs14("f14")=f14, fs15("f15")=f15, fs16("f16")=f16, fs17("f17")=f17,
+	fs18("f18")=f18, fs19("f19")=f19, fs20("f20")=f20, fs21("f21")=f21,
+	fs22("f22")=f22, fs23("f23")=f23, fs24("f24")=f24, fs25("f25")=f25,
+	fs26("f26")=f26, fs27("f27")=f27, fs28("f28")=f28, fs29("f29")=f29,
+	fs30("f30")=f30, fs31("f31")=f31
+	  : FSREG regvar(reg_float).
+
 	lr, ctr     : SPR.
 	cr0         : CR.
 
@@ -257,7 +265,7 @@ INSTRUCTIONS
   fdiv            FREG+DLOCAL:wo, FREG:ro, FREG:ro cost(4, 35).
   fdivs           FSREG:wo, FSREG:ro, FSREG:ro cost(4, 21).
   fmr             FPR+DLOCAL:wo, FPR:ro cost(4, 5).
-  fmr             FSREG:wo, FSREG:ro cost(4, 5).
+  fmr             FSREG+LOCAL:wo, FSREG:ro cost(4, 5).
   fmul            FREG+DLOCAL:wo, FREG:ro, FREG:ro cost(4, 5).
   fmuls           FSREG:wo, FSREG:ro, FSREG:ro cost(4, 5).
   fneg            FREG+DLOCAL:wo, FREG:ro cost(4, 5).
@@ -270,9 +278,9 @@ INSTRUCTIONS
   lfd             FPR+DLOCAL:wo, IND_RC_D+IND_RL_D:ro cost(4, 5).
   lfdu            FPR:wo, IND_RC_D:ro cost(4, 5).
   lfdx            FPR+DLOCAL:wo, GPR:ro, GPR:ro cost(4, 5).
-  lfs             FSREG:wo, IND_RC_W+IND_RL_W:ro cost(4, 4).
+  lfs             FSREG+LOCAL:wo, IND_RC_W+IND_RL_W:ro cost(4, 4).
   lfsu            FSREG:wo, IND_RC_W:rw cost(4, 4).
-  lfsx            FSREG:wo, GPR:ro, GPR:ro cost(4, 4).
+  lfsx            FSREG+LOCAL:wo, GPR:ro, GPR:ro cost(4, 4).
   lha             GPR:wo, IND_RC_H_S+IND_RL_H_S:ro cost(4, 3).
   lhax            GPR:wo, GPR:ro, GPR:ro cost(4, 3).
   lhz             GPR:wo, IND_RC_H+IND_RL_H:ro cost(4, 3).
@@ -309,7 +317,7 @@ INSTRUCTIONS
   stfdu           FPR+DLOCAL:ro, IND_RC_D:rw cost(4, 4).
   stfdx           FPR:ro, GPR:ro, GPR:ro cost(4, 4).
   stfs            FSREG:ro, IND_RC_W+IND_RL_W:rw cost(4, 3).
-  stfsu           FSREG:ro, IND_RC_W:rw cost(4, 3).
+  stfsu           FSREG+LOCAL:ro, IND_RC_W:rw cost(4, 3).
   stfsx           FSREG:ro, GPR:ro, GPR:ro cost(4, 3).
   sth             GPR:ro, IND_RC_H+IND_RL_H:rw cost(4, 3).
   sthx            GPR:ro, GPR:ro, GPR:ro cost(4, 3).
@@ -328,6 +336,9 @@ MOVES
 	from GPR to GPR
 		gen mr %2, %1
 
+	from FSREG to FSREG
+		gen fmr %2, %1
+
 	from FPR to FPR+DLOCAL
 		gen fmr %2, %1
 
@@ -577,6 +588,11 @@ TESTS
 
 STACKINGRULES
 
+	from LOCAL inreg(%off)==reg_float to STACK
+		gen
+			COMMENT("stack LOCAL")
+			stfsu %1, {IND_RC_W, sp, 0-4}
+
 	from LOCAL to STACK
 		gen
 			COMMENT("stack LOCAL")
@@ -775,7 +791,7 @@ PATTERNS
 		yields {SUM_RC, %a, los($1)}
 
 	/* Load word from local */
-	pat lol inreg($1)==reg_any
+	pat lol inreg($1)==reg_any || inreg($1)==reg_float
 		yields {LOCAL, $1}
 	pat lol
 		leaving
@@ -795,6 +811,13 @@ PATTERNS
 		with ANY_BHW
 			kills regvar($1), LOCAL %off==$1
 			gen move %1, {GPRE, regvar($1)}
+	pat stl inreg($1)==reg_float
+		with exact FSREG
+			gen fmr {LOCAL, $1}, %1
+		with STACK
+			gen
+				lfs {LOCAL, $1}, {IND_RC_W, sp, 0}
+				addi sp, sp, {CONST, 4}
 	pat stl
 		leaving
 			lal $1
diff --git a/mach/proto/ncg/regvar.c b/mach/proto/ncg/regvar.c
index 4c11da2b1..934b775f8 100644
--- a/mach/proto/ncg/regvar.c
+++ b/mach/proto/ncg/regvar.c
@@ -50,7 +50,19 @@ tryreg(rvlp,typ) register struct regvar *rvlp; {
 	struct regvar *save;
 
 	if (typ != reg_any && nregvar[typ]!=0) {
-		if (machregs[rvnumbers[typ][0]].r_size!=rvlp->rv_size)
+		struct reginfo *ri = &machregs[rvnumbers[typ][0]];
+		int size, wrong;
+
+		size = ri->r_size;
+		wrong = (size!=rvlp->rv_size);
+#ifdef REGLAP
+		/* reg_float may have one subregister */
+		if (wrong && ri->r_members[0]!=0) {
+			size = machregs[ri->r_members[0]].r_size;
+			wrong = (size!=rvlp->rv_size);
+		}
+#endif /* REGLAP */
+		if (wrong)
 			score = -1;
 		else
 			score = regscore(rvlp->rv_off,
@@ -99,11 +111,29 @@ fixregvars(saveall) {
 		if (saveall) {
 			struct reginfo *rp;
 			rp= &machregs[rvnumbers[rvtyp][i]];
-			regsave(codestrings[rp->r_repr],(long)-TEM_WSIZE,rp->r_size);
+			regsave(codestrings[rp->r_repr],
+				    (long)-TEM_WSIZE,rp->r_size);
+#ifdef REGLAP
+			/* reg_float may have one subregister */
+			if (rp->r_members[0]!=0) {
+				rp= &machregs[rp->r_members[0]];
+				regsave(codestrings[rp->r_repr],
+					    (long)-TEM_WSIZE,rp->r_size);
+			}
+#endif
 		} else if(regassigned[rvtyp][i].ra_score>0) {
 			rv=regassigned[rvtyp][i].ra_rv;
 			rv->rv_reg=rvnumbers[rvtyp][i];
 			rv->rv_type = rvtyp;
+#ifdef REGLAP
+			/* reg_float may have alternate size */
+			if (machregs[rv->rv_reg].r_size != rv->rv_size) {
+				rv->rv_reg = machregs[rv->rv_reg].r_members[0];
+				assert(rv->rv_reg != 0);
+				assert(machregs[rv->rv_reg].r_size ==
+					   rv->rv_size);
+			}
+#endif
 			regsave(codestrings[machregs[rv->rv_reg].r_repr],
 				    rv->rv_off,rv->rv_size);
 		}
diff --git a/util/ncgg/expr.c b/util/ncgg/expr.c
index 77afe5774..0a6351590 100644
--- a/util/ncgg/expr.c
+++ b/util/ncgg/expr.c
@@ -18,6 +18,7 @@ static char rcsid[]= "$Id$";
 #include "expr.h"
 #include "regvar.h"
 #include <cgg_cg.h>
+#include <em_reg.h>
 #include "extern.h"
 
 extern set_t l_sets[];
@@ -276,12 +277,21 @@ expr_t iextoaddr(e) expr_t e; {
 expr_t regvar_expr(e,regtyp) expr_t e; {
 	expr_t result;
 	register i;
-	
+
 	result = make_expr(TYPREG,EX_REGVAR,i_expr(e),0);
 	for(i=0;i<SZOFSET(MAXREGS);i++)
 		result.ex_regset[i]=0;
 	for(i=0;i<nregvar[regtyp];i++)
 		BIS(result.ex_regset,rvnumbers[regtyp][i]);
+	/* reglap: float may overlap with one subregister */
+	if (reglap!=0 && regtyp==reg_float) {
+		for(i=0;i<nregvar[regtyp];i++) {
+			/* reg = first subregister */
+			int reg = l_regs[rvnumbers[regtyp][i]].ri_memb[0];
+			if (reg!=0)
+				BIS(result.ex_regset,reg);
+		}
+	}
 	return(result);
 }
 		
diff --git a/util/ncgg/output.c b/util/ncgg/output.c
index 77e3eb41f..9d2f2b8d4 100644
--- a/util/ncgg/output.c
+++ b/util/ncgg/output.c
@@ -317,7 +317,15 @@ outregs() {
 			}
 			clashlist[iclashlist++] = 0;
 		}
-		fprintf(ctable,",%d",l_regs[i].ri_rregvar>=0);
+		/*
+		 * Write .r_refcount = 1 for register variables or 0
+		 * for other registers.
+		 *
+		 * reglap: Write .r_refcount = 0 if the register
+		 *   variable has a subregister.
+		 */
+		fprintf(ctable,",%d",
+		    l_regs[i].ri_rregvar>=0 && l_regs[i].ri_memb[0]==0);
 		fprintf(ctable,"},\n");
 	}
 	fprintf(ctable,"};\n\n short clashlist[] = {\n\t");
@@ -332,6 +340,8 @@ outregvars() {
 	register i,j;
 
 	fprintf(htable,"#define REGVARS\n");
+	if (reglap!=0)
+		fprintf(htable,"#define REGLAP\n");
 	fprintf(ctable,"#include \"regvar.h\"\n");
 	fprintf(ctable,"int nregvar[4] = { ");
 	for (i=0;i<4;i++) {
diff --git a/util/ncgg/regvar.h b/util/ncgg/regvar.h
index 7961d1eb9..9a772c0f6 100644
--- a/util/ncgg/regvar.h
+++ b/util/ncgg/regvar.h
@@ -9,6 +9,7 @@
 #define DL_REGVAR	0x4
 
 extern int rvused;
+extern int reglap;
 extern int nregvar[4];
 extern int rvsize[4];
 extern int rvnumbers[4][MAXREGVAR];
diff --git a/util/ncgg/subr.c b/util/ncgg/subr.c
index 08c7c650f..89b2b0917 100644
--- a/util/ncgg/subr.c
+++ b/util/ncgg/subr.c
@@ -19,6 +19,7 @@ static char rcsid[]= "$Id$";
 #include "token.h"
 #include "regvar.h"
 #include <cgg_cg.h>
+#include <em_reg.h>
 #include "extern.h"
 
 n_proc(name) char *name; {
@@ -228,6 +229,70 @@ n_sconst(ident,val) char *ident,*val; {
 	sy_p->sy_value.syv_stringno = strlookup(val);
 }
 
+static void
+add_regvar(int rvnum, reginfo *regp, int rv)
+{
+	int overlap, wrong;
+
+	overlap = wrong = 0;
+	if (regp->ri_memb[0]!=0) {
+		/* reglap: float may overlap with one subregister */
+		if (rv==reg_float && regp->ri_memb[1]==0)
+			overlap = 1;
+		else
+			wrong = 1;
+	}
+	if (wrong)
+		error("Register variables may not have subregisters");
+
+	rvused |= ANY_REGVAR;
+	if (regp->ri_size == wordsize)
+		rvused |= SL_REGVAR;
+	else if (regp->ri_size == 2*wordsize)
+		rvused |= DL_REGVAR;
+
+	wrong = 0;
+	if (overlap) {
+		/* reglap = size of overlap float */
+		if (reglap==0)
+			reglap = regp->ri_size;
+		else if (reglap!=regp->ri_size)
+			wrong = 1;
+	} else {
+		if (nregvar[rv]==0)
+			rvsize[rv] = regp->ri_size;
+		else if (rvsize[rv]!=regp->ri_size)
+			wrong = 1;
+	}
+	if (wrong)
+		error("All register variables of one type must have the same size");
+
+	if (overlap) {
+		reginfo *member_p = &l_regs[regp->ri_memb[0]];
+		int i;
+
+		/*
+		 * reglap: Remove member_p from rvnumbers.
+		 *   Add reg_p in its place.
+		 */
+		wrong = 1;
+		for (i = 0; i < nregvar[rv]; i++) {
+			if (rvnumbers[rv][i] == regp->ri_memb[0]) {
+				rvnumbers[rv][i] = rvnum;
+				wrong = 0;
+				break;
+			}
+		}
+		if (wrong)
+			error("Register variable %s can't overlap %s",
+			    regp->ri_name, member_p->ri_name);
+	} else {
+		NEXT(nregvar[rv],MAXREGVAR,"Register variable");
+		rvnumbers[rv][nregvar[rv]-1] = rvnum;
+	}
+}
+
+void
 regline(rl,pl,rv) varinfo *rl,*pl; {
 	register varinfo *rrl,*rpl;
 	register short *sp;
@@ -251,21 +316,8 @@ regline(rl,pl,rv) varinfo *rl,*pl; {
 		regp->ri_size = thissize;
 		regp->ri_class = regclass;
 		regp->ri_rregvar = rv;
-		if (rv>=0) {
-			if (regp->ri_memb[0]!=0)
-				error("Register variables may not have subregisters");
-			rvused |= ANY_REGVAR;
-			if (regp->ri_size == wordsize)
-				rvused |= SL_REGVAR;
-			else if (regp->ri_size == 2*wordsize)
-				rvused |= DL_REGVAR;
-			if (nregvar[rv]==0)
-				rvsize[rv] = regp->ri_size;
-			else if (rvsize[rv]!=regp->ri_size)
-				error("All register variables of one type must have the same size");
-			NEXT(nregvar[rv],MAXREGVAR,"Register variable");
-			rvnumbers[rv][nregvar[rv]-1] = rrl->vi_int[0];
-		}
+		if (rv>=0)
+			add_regvar(rrl->vi_int[0], regp, rv);
 	}
 	regclass++;
 }
diff --git a/util/ncgg/var.c b/util/ncgg/var.c
index 72a3f825d..19a4da4f3 100644
--- a/util/ncgg/var.c
+++ b/util/ncgg/var.c
@@ -39,6 +39,7 @@ int maxmembers=0;
 int regclass=1;
 int maxtokensize=0;
 int rvused=0;
+int reglap=0;
 int nregvar[4];
 int rvsize[4];
 int rvnumbers[4][MAXREGVAR];