added float support

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@15 c046a42c-6fe2-441c-8c8c-71466251a162

6 files changed, 934 insertions, 160 deletions

diff --git a/Makefile b/Makefile
index 397ddf1342..a61ab4a76a 100644
--- a/Makefile
+++ b/Makefile
@@ -8,6 +8,7 @@ LDFLAGS=-g
 LIBS=
 CC=gcc
 DEFINES=-DHAVE_BYTESWAP_H
+OP_CFLAGS=$(CFLAGS) -malign-functions=0 -mpreferred-stack-boundary=2
 endif
 
 ifeq ($(ARCH),ppc)
@@ -24,6 +25,7 @@ CRTEND=$(GCC_LIBS_DIR)/crtend.o
 LDFLAGS=-static -g -nostdlib $(CRT1) $(CRTI) $(CRTBEGIN) 
 LIBS=-L$(LIBS_DIR) -ltinyc -lgcc $(CRTEND) $(CRTN)
 DEFINES=-Dsocklen_t=int
+OP_CFLAGS=$(CFLAGS)
 endif
 
 #########################################################
@@ -31,7 +33,7 @@ endif
 DEFINES+=-D_GNU_SOURCE -DGEMU -DDOSEMU -DNO_TRACE_MSGS
 DEFINES+=-DCONFIG_PREFIX=\"/usr/local\"
 LDSCRIPT=$(ARCH).ld
-LIBS+=-ldl
+LIBS+=-ldl -lm
 
 OBJS= i386/fp87.o i386/interp_main.o i386/interp_modrm.o i386/interp_16_32.o \
       i386/interp_32_16.o i386/interp_32_32.o i386/emu-utils.o \
@@ -67,7 +69,7 @@ op-i386.h: op-i386.o dyngen
 	./dyngen -o $@ $<
 
 op-i386.o: op-i386.c opreg_template.h ops_template.h
-	$(CC) $(CFLAGS) $(DEFINES) -c -o $@ $<
+	$(CC) $(OP_CFLAGS) $(DEFINES) -c -o $@ $<
 
 %.o: %.c
 	$(CC) $(CFLAGS) $(DEFINES) -c -o $@ $<
diff --git a/TODO b/TODO
index 7ba6ab4a79..24eb3863f6 100644
--- a/TODO
+++ b/TODO
@@ -3,3 +3,4 @@
 - threads
 - fix printf for doubles (fp87.c bug ?)
 - make it self runnable (use same trick as ld.so : include its own relocator and libc)
+- better FPU comparisons (ucom/com)
diff --git a/cpu-i386.h b/cpu-i386.h
index a857efb7df..c056a46771 100644
--- a/cpu-i386.h
+++ b/cpu-i386.h
@@ -75,6 +75,16 @@ enum {
     CC_OP_NB,
 };
 
+#ifdef __i386__
+#define USE_X86LDOUBLE
+#endif
+
+#ifdef USE_X86LDOUBLE
+typedef long double CPU86_LDouble;
+#else
+typedef double CPU86_LDouble;
+#endif
+
 typedef struct CPU86State {
     /* standard registers */
     uint32_t regs[8];
@@ -91,10 +101,18 @@ typedef struct CPU86State {
     uint8_t *segs_base[6];
     uint32_t segs[6];
 
+    /* FPU state */
+    CPU86_LDouble fpregs[8];    
+    uint8_t fptags[8];   /* 0 = valid, 1 = empty */
+    unsigned int fpstt; /* top of stack index */
+    unsigned int fpus;
+    unsigned int fpuc;
+
     /* emulator internal variables */
     uint32_t t0; /* temporary t0 storage */
     uint32_t t1; /* temporary t1 storage */
     uint32_t a0; /* temporary a0 storage (address) */
+    CPU86_LDouble ft0;
 } CPU86State;
 
 static inline int ldub(void *ptr)
@@ -122,6 +140,10 @@ static inline int ldl(void *ptr)
     return *(uint32_t *)ptr;
 }
 
+static inline uint64_t ldq(void *ptr)
+{
+    return *(uint64_t *)ptr;
+}
 
 static inline void stb(void *ptr, int v)
 {
@@ -138,11 +160,40 @@ static inline void stl(void *ptr, int v)
     *(uint32_t *)ptr = v;
 }
 
+static inline void stq(void *ptr, int v)
+{
+    *(uint64_t *)ptr = v;
+}
+
+/* float access */
+
+static inline float ldfl(void *ptr)
+{
+    return *(float *)ptr;
+}
+
+static inline double ldfq(void *ptr)
+{
+    return *(double *)ptr;
+}
+
+static inline void stfl(void *ptr, float v)
+{
+    *(float *)ptr = v;
+}
+
+static inline void stfq(void *ptr, double v)
+{
+    *(double *)ptr = v;
+}
+
+#ifndef IN_OP_I386
 void port_outb(int addr, int val);
 void port_outw(int addr, int val);
 void port_outl(int addr, int val);
 int port_inb(int addr);
 int port_inw(int addr);
 int port_inl(int addr);
+#endif
 
 #endif /* CPU_I386_H */
diff --git a/dyngen.c b/dyngen.c
index ff10891b9e..ce38dcaec4 100644
--- a/dyngen.c
+++ b/dyngen.c
@@ -243,6 +243,8 @@ void gen_code(const char *name, unsigned long offset, unsigned long size,
                     if (n >= MAX_ARGS)
                         error("too many arguments in %s", name);
                     args_present[n - 1] = 1;
+                } else {
+                    fprintf(outfile, "extern char %s;\n", sym_name);
                 }
             }
         }
@@ -257,6 +259,8 @@ void gen_code(const char *name, unsigned long offset, unsigned long size,
                     if (n >= MAX_ARGS)
                         error("too many arguments in %s", name);
                     args_present[n - 1] = 1;
+                } else {
+                    fprintf(outfile, "extern char %s;\n", sym_name);
                 }
             }
         }
diff --git a/op-i386.c b/op-i386.c
index 8607cf4552..9fd4accd57 100644
--- a/op-i386.c
+++ b/op-i386.c
@@ -66,7 +66,13 @@ register struct CPU86State *env asm("l3");
 
 #define CC_SRC (env->cc_src)
 #define CC_DST (env->cc_dst)
-#define CC_OP (env->cc_op)
+#define CC_OP  (env->cc_op)
+
+/* float macros */
+#define FT0    (env->ft0)
+#define ST0    (env->fpregs[env->fpstt])
+#define ST(n)  (env->fpregs[(env->fpstt + (n)) & 7])
+#define ST1    ST(1)
 
 extern int __op_param1, __op_param2, __op_param3;
 #define PARAM1 ((long)(&__op_param1))
@@ -133,6 +139,44 @@ const uint8_t rclb_table[32] = {
     6, 7, 8, 0, 1, 2, 3, 4,
 };
 
+#ifdef USE_X86LDOUBLE
+/* an array of Intel 80-bit FP constants, to be loaded via integer ops */
+typedef unsigned short f15ld[5];
+const f15ld f15rk[] =
+{
+/*0*/	{0x0000,0x0000,0x0000,0x0000,0x0000},
+/*1*/	{0x0000,0x0000,0x0000,0x8000,0x3fff},
+/*pi*/	{0xc235,0x2168,0xdaa2,0xc90f,0x4000},
+/*lg2*/	{0xf799,0xfbcf,0x9a84,0x9a20,0x3ffd},
+/*ln2*/	{0x79ac,0xd1cf,0x17f7,0xb172,0x3ffe},
+/*l2e*/	{0xf0bc,0x5c17,0x3b29,0xb8aa,0x3fff},
+/*l2t*/	{0x8afe,0xcd1b,0x784b,0xd49a,0x4000}
+};
+#else
+/* the same, 64-bit version */
+typedef unsigned short f15ld[4];
+const f15ld f15rk[] =
+{
+#ifndef WORDS_BIGENDIAN
+/*0*/	{0x0000,0x0000,0x0000,0x0000},
+/*1*/	{0x0000,0x0000,0x0000,0x3ff0},
+/*pi*/	{0x2d18,0x5444,0x21fb,0x4009},
+/*lg2*/	{0x79ff,0x509f,0x4413,0x3fd3},
+/*ln2*/	{0x39ef,0xfefa,0x2e42,0x3fe6},
+/*l2e*/	{0x82fe,0x652b,0x1547,0x3ff7},
+/*l2t*/	{0xa371,0x0979,0x934f,0x400a}
+#else
+/*0*/   {0x0000,0x0000,0x0000,0x0000},
+/*1*/   {0x3ff0,0x0000,0x0000,0x0000},
+/*pi*/  {0x4009,0x21fb,0x5444,0x2d18},
+/*lg2*/	{0x3fd3,0x4413,0x509f,0x79ff},
+/*ln2*/	{0x3fe6,0x2e42,0xfefa,0x39ef},
+/*l2e*/	{0x3ff7,0x1547,0x652b,0x82fe},
+/*l2t*/	{0x400a,0x934f,0x0979,0xa371}
+#endif
+};
+#endif
+    
 /* n must be a constant to be efficient */
 static inline int lshift(int x, int n)
 {
@@ -866,3 +910,685 @@ CCTable cc_table[CC_OP_NB] = {
     [CC_OP_SHLW] = { compute_all_shlw, compute_c_shlw },
     [CC_OP_SHLL] = { compute_all_shll, compute_c_shll },
 };
+
+/* floating point support */
+
+#ifdef USE_X86LDOUBLE
+/* use long double functions */
+#define lrint lrintl
+#define llrint llrintl
+#define fabs fabsl
+#define sin sinl
+#define cos cosl
+#define sqrt sqrtl
+#define pow powl
+#define log logl
+#define tan tanl
+#define atan2 atan2l
+#define floor floorl
+#define ceil ceill
+#define rint rintl
+#endif
+
+extern int lrint(CPU86_LDouble x);
+extern int64_t llrint(CPU86_LDouble x);
+extern CPU86_LDouble fabs(CPU86_LDouble x);
+extern CPU86_LDouble sin(CPU86_LDouble x);
+extern CPU86_LDouble cos(CPU86_LDouble x);
+extern CPU86_LDouble sqrt(CPU86_LDouble x);
+extern CPU86_LDouble pow(CPU86_LDouble, CPU86_LDouble);
+extern CPU86_LDouble log(CPU86_LDouble x);
+extern CPU86_LDouble tan(CPU86_LDouble x);
+extern CPU86_LDouble atan2(CPU86_LDouble, CPU86_LDouble);
+extern CPU86_LDouble floor(CPU86_LDouble x);
+extern CPU86_LDouble ceil(CPU86_LDouble x);
+extern CPU86_LDouble rint(CPU86_LDouble x);
+
+#define RC_MASK         0xc00
+#define RC_NEAR		0x000
+#define RC_DOWN		0x400
+#define RC_UP		0x800
+#define RC_CHOP		0xc00
+
+#define MAXTAN 9223372036854775808.0
+
+#ifdef USE_X86LDOUBLE
+
+/* only for x86 */
+typedef union {
+    long double d;
+    struct {
+        unsigned long long lower;
+        unsigned short upper;
+    } l;
+} CPU86_LDoubleU;
+
+/* the following deal with x86 long double-precision numbers */
+#define MAXEXPD 0x7fff
+#define EXPBIAS 16383
+#define EXPD(fp)	(fp.l.upper & 0x7fff)
+#define SIGND(fp)	((fp.l.upper) & 0x8000)
+#define MANTD(fp)       (fp.l.lower)
+#define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS
+
+#else
+
+typedef {
+    double d;
+#ifndef WORDS_BIGENDIAN
+    struct {
+        unsigned long lower;
+        long upper;
+    } l;
+#else
+    struct {
+        long upper;
+        unsigned long lower;
+    } l;
+#endif
+    long long ll;
+} CPU86_LDoubleU;
+
+/* the following deal with IEEE double-precision numbers */
+#define MAXEXPD 0x7ff
+#define EXPBIAS 1023
+#define EXPD(fp)	(((fp.l.upper) >> 20) & 0x7FF)
+#define SIGND(fp)	((fp.l.upper) & 0x80000000)
+#define MANTD(fp)	(fp.ll & ((1LL << 52) - 1))
+#define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7ff << 20)) | (EXPBIAS << 20)
+#endif
+
+/* fp load FT0 */
+
+void OPPROTO op_flds_FT0_A0(void)
+{
+    FT0 = ldfl((void *)A0);
+}
+
+void OPPROTO op_fldl_FT0_A0(void)
+{
+    FT0 = ldfq((void *)A0);
+}
+
+void OPPROTO op_fild_FT0_A0(void)
+{
+    FT0 = (CPU86_LDouble)ldsw((void *)A0);
+}
+
+void OPPROTO op_fildl_FT0_A0(void)
+{
+    FT0 = (CPU86_LDouble)((int32_t)ldl((void *)A0));
+}
+
+void OPPROTO op_fildll_FT0_A0(void)
+{
+    FT0 = (CPU86_LDouble)((int64_t)ldq((void *)A0));
+}
+
+/* fp load ST0 */
+
+void OPPROTO op_flds_ST0_A0(void)
+{
+    ST0 = ldfl((void *)A0);
+}
+
+void OPPROTO op_fldl_ST0_A0(void)
+{
+    ST0 = ldfq((void *)A0);
+}
+
+void OPPROTO op_fild_ST0_A0(void)
+{
+    ST0 = (CPU86_LDouble)ldsw((void *)A0);
+}
+
+void OPPROTO op_fildl_ST0_A0(void)
+{
+    ST0 = (CPU86_LDouble)((int32_t)ldl((void *)A0));
+}
+
+void OPPROTO op_fildll_ST0_A0(void)
+{
+    ST0 = (CPU86_LDouble)((int64_t)ldq((void *)A0));
+}
+
+/* fp store */
+
+void OPPROTO op_fsts_ST0_A0(void)
+{
+    stfl((void *)A0, (float)ST0);
+}
+
+void OPPROTO op_fstl_ST0_A0(void)
+{
+    ST0 = ldfq((void *)A0);
+}
+
+void OPPROTO op_fist_ST0_A0(void)
+{
+    int val;
+    val = lrint(ST0);
+    stw((void *)A0, val);
+}
+
+void OPPROTO op_fistl_ST0_A0(void)
+{
+    int val;
+    val = lrint(ST0);
+    stl((void *)A0, val);
+}
+
+void OPPROTO op_fistll_ST0_A0(void)
+{
+    int64_t val;
+    val = llrint(ST0);
+    stq((void *)A0, val);
+}
+
+/* FPU move */
+
+static inline void fpush(void)
+{
+    env->fpstt = (env->fpstt - 1) & 7;
+    env->fptags[env->fpstt] = 0; /* validate stack entry */
+}
+
+static inline void fpop(void)
+{
+    env->fptags[env->fpstt] = 1; /* invvalidate stack entry */
+    env->fpstt = (env->fpstt + 1) & 7;
+}
+
+void OPPROTO op_fpush(void)
+{
+    fpush();
+}
+
+void OPPROTO op_fpop(void)
+{
+    fpop();
+}
+
+void OPPROTO op_fdecstp(void)
+{
+    env->fpstt = (env->fpstt - 1) & 7;
+    env->fpus &= (~0x4700);
+}
+
+void OPPROTO op_fincstp(void)
+{
+    env->fpstt = (env->fpstt + 1) & 7;
+    env->fpus &= (~0x4700);
+}
+
+void OPPROTO op_fmov_ST0_FT0(void)
+{
+    ST0 = FT0;
+}
+
+void OPPROTO op_fmov_FT0_STN(void)
+{
+    FT0 = ST(PARAM1);
+}
+
+void OPPROTO op_fmov_ST0_STN(void)
+{
+    ST0 = ST(PARAM1);
+}
+
+void OPPROTO op_fmov_STN_ST0(void)
+{
+    ST(PARAM1) = ST0;
+}
+
+void OPPROTO op_fxchg_ST0_STN(void)
+{
+    CPU86_LDouble tmp;
+    tmp = ST(PARAM1);
+    ST(PARAM1) = ST0;
+    ST0 = tmp;
+}
+
+/* FPU operations */
+
+/* XXX: handle nans */
+void OPPROTO op_fcom_ST0_FT0(void)
+{
+    env->fpus &= (~0x4500);	/* (C3,C2,C0) <-- 000 */
+    if (ST0 < FT0)
+        env->fpus |= 0x100;	/* (C3,C2,C0) <-- 001 */
+    else if (ST0 == FT0)
+        env->fpus |= 0x4000; /* (C3,C2,C0) <-- 100 */
+    FORCE_RET();
+}
+
+void OPPROTO op_fadd_ST0_FT0(void)
+{
+    ST0 += FT0;
+}
+
+void OPPROTO op_fmul_ST0_FT0(void)
+{
+    ST0 *= FT0;
+}
+
+void OPPROTO op_fsub_ST0_FT0(void)
+{
+    ST0 -= FT0;
+}
+
+void OPPROTO op_fsubr_ST0_FT0(void)
+{
+    ST0 = FT0 - ST0;
+}
+
+void OPPROTO op_fdiv_ST0_FT0(void)
+{
+    ST0 /= FT0;
+}
+
+void OPPROTO op_fdivr_ST0_FT0(void)
+{
+    ST0 = FT0 / ST0;
+}
+
+/* fp operations between STN and ST0 */
+
+void OPPROTO op_fadd_STN_ST0(void)
+{
+    ST(PARAM1) += ST0;
+}
+
+void OPPROTO op_fmul_STN_ST0(void)
+{
+    ST(PARAM1) *= ST0;
+}
+
+void OPPROTO op_fsub_STN_ST0(void)
+{
+    ST(PARAM1) -= ST0;
+}
+
+void OPPROTO op_fsubr_STN_ST0(void)
+{
+    CPU86_LDouble *p;
+    p = &ST(PARAM1);
+    *p = ST0 - *p;
+}
+
+void OPPROTO op_fdiv_STN_ST0(void)
+{
+    ST(PARAM1) /= ST0;
+}
+
+void OPPROTO op_fdivr_STN_ST0(void)
+{
+    CPU86_LDouble *p;
+    p = &ST(PARAM1);
+    *p = ST0 / *p;
+}
+
+/* misc FPU operations */
+void OPPROTO op_fchs_ST0(void)
+{
+    ST0 = -ST0;
+}
+
+void OPPROTO op_fabs_ST0(void)
+{
+    ST0 = fabs(ST0);
+}
+
+void OPPROTO op_fxam_ST0(void)
+{
+    CPU86_LDoubleU temp;
+    int expdif;
+
+    temp.d = ST0;
+
+    env->fpus &= (~0x4700);  /* (C3,C2,C1,C0) <-- 0000 */
+    if (SIGND(temp))
+        env->fpus |= 0x200; /* C1 <-- 1 */
+
+    expdif = EXPD(temp);
+    if (expdif == MAXEXPD) {
+        if (MANTD(temp) == 0)
+            env->fpus |=  0x500 /*Infinity*/;
+        else
+            env->fpus |=  0x100 /*NaN*/;
+    } else if (expdif == 0) {
+        if (MANTD(temp) == 0)
+            env->fpus |=  0x4000 /*Zero*/;
+        else
+            env->fpus |= 0x4400 /*Denormal*/;
+    } else {
+        env->fpus |= 0x400;
+    }
+    FORCE_RET();
+}
+
+void OPPROTO op_fld1_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[1];
+}
+
+void OPPROTO op_fld2t_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[6];
+}
+
+void OPPROTO op_fld2e_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[5];
+}
+
+void OPPROTO op_fldpi_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[2];
+}
+
+void OPPROTO op_fldlg2_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[3];
+}
+
+void OPPROTO op_fldln2_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[4];
+}
+
+void OPPROTO op_fldz_ST0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[0];
+}
+
+void OPPROTO op_fldz_FT0(void)
+{
+    ST0 = *(CPU86_LDouble *)&f15rk[0];
+}
+
+void helper_f2xm1(void)
+{
+    ST0 = pow(2.0,ST0) - 1.0;
+}
+
+void helper_fyl2x(void)
+{
+    CPU86_LDouble fptemp;
+    
+    fptemp = ST0;
+    if (fptemp>0.0){
+        fptemp = log(fptemp)/log(2.0);	 /* log2(ST) */
+        ST1 *= fptemp;
+        fpop();
+    } else { 
+        env->fpus &= (~0x4700);
+        env->fpus |= 0x400;
+    }
+}
+
+void helper_fptan(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if((fptemp > MAXTAN)||(fptemp < -MAXTAN)) {
+        env->fpus |= 0x400;
+    } else {
+        ST0 = tan(fptemp);
+        fpush();
+        ST0 = 1.0;
+        env->fpus &= (~0x400);  /* C2 <-- 0 */
+        /* the above code is for  |arg| < 2**52 only */
+    }
+}
+
+void helper_fpatan(void)
+{
+    CPU86_LDouble fptemp, fpsrcop;
+
+    fpsrcop = ST1;
+    fptemp = ST0;
+    ST1 = atan2(fpsrcop,fptemp);
+    fpop();
+}
+
+void helper_fxtract(void)
+{
+    CPU86_LDoubleU temp;
+    unsigned int expdif;
+
+    temp.d = ST0;
+    expdif = EXPD(temp) - EXPBIAS;
+    /*DP exponent bias*/
+    ST0 = expdif;
+    fpush();
+    BIASEXPONENT(temp);
+    ST0 = temp.d;
+}
+
+void helper_fprem1(void)
+{
+    CPU86_LDouble dblq, fpsrcop, fptemp;
+    CPU86_LDoubleU fpsrcop1, fptemp1;
+    int expdif;
+    int q;
+
+    fpsrcop = ST0;
+    fptemp = ST1;
+    fpsrcop1.d = fpsrcop;
+    fptemp1.d = fptemp;
+    expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
+    if (expdif < 53) {
+        dblq = fpsrcop / fptemp;
+        dblq = (dblq < 0.0)? ceil(dblq): floor(dblq);
+        ST0 = fpsrcop - fptemp*dblq;
+        q = (int)dblq; /* cutting off top bits is assumed here */
+        env->fpus &= (~0x4700); /* (C3,C2,C1,C0) <-- 0000 */
+				/* (C0,C1,C3) <-- (q2,q1,q0) */
+        env->fpus |= (q&0x4) << 6; /* (C0) <-- q2 */
+        env->fpus |= (q&0x2) << 8; /* (C1) <-- q1 */
+        env->fpus |= (q&0x1) << 14; /* (C3) <-- q0 */
+    } else {
+        env->fpus |= 0x400;  /* C2 <-- 1 */
+        fptemp = pow(2.0, expdif-50);
+        fpsrcop = (ST0 / ST1) / fptemp;
+        /* fpsrcop = integer obtained by rounding to the nearest */
+        fpsrcop = (fpsrcop-floor(fpsrcop) < ceil(fpsrcop)-fpsrcop)?
+            floor(fpsrcop): ceil(fpsrcop);
+        ST0 -= (ST1 * fpsrcop * fptemp);
+    }
+}
+
+void helper_fprem(void)
+{
+    CPU86_LDouble dblq, fpsrcop, fptemp;
+    CPU86_LDoubleU fpsrcop1, fptemp1;
+    int expdif;
+    int q;
+    
+    fpsrcop = ST0;
+    fptemp = ST1;
+    fpsrcop1.d = fpsrcop;
+    fptemp1.d = fptemp;
+    expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
+    if ( expdif < 53 ) {
+        dblq = fpsrcop / fptemp;
+        dblq = (dblq < 0.0)? ceil(dblq): floor(dblq);
+        ST0 = fpsrcop - fptemp*dblq;
+        q = (int)dblq; /* cutting off top bits is assumed here */
+        env->fpus &= (~0x4700); /* (C3,C2,C1,C0) <-- 0000 */
+				/* (C0,C1,C3) <-- (q2,q1,q0) */
+        env->fpus |= (q&0x4) << 6; /* (C0) <-- q2 */
+        env->fpus |= (q&0x2) << 8; /* (C1) <-- q1 */
+        env->fpus |= (q&0x1) << 14; /* (C3) <-- q0 */
+    } else {
+        env->fpus |= 0x400;  /* C2 <-- 1 */
+        fptemp = pow(2.0, expdif-50);
+        fpsrcop = (ST0 / ST1) / fptemp;
+        /* fpsrcop = integer obtained by chopping */
+        fpsrcop = (fpsrcop < 0.0)?
+            -(floor(fabs(fpsrcop))): floor(fpsrcop);
+        ST0 -= (ST1 * fpsrcop * fptemp);
+    }
+}
+
+void helper_fyl2xp1(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if ((fptemp+1.0)>0.0) {
+        fptemp = log(fptemp+1.0) / log(2.0); /* log2(ST+1.0) */
+        ST1 *= fptemp;
+        fpop();
+    } else { 
+        env->fpus &= (~0x4700);
+        env->fpus |= 0x400;
+    }
+}
+
+void helper_fsqrt(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if (fptemp<0.0) { 
+        env->fpus &= (~0x4700);  /* (C3,C2,C1,C0) <-- 0000 */
+        env->fpus |= 0x400;
+    }
+    ST0 = sqrt(fptemp);
+}
+
+void helper_fsincos(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if ((fptemp > MAXTAN)||(fptemp < -MAXTAN)) {
+        env->fpus |= 0x400;
+    } else {
+        ST0 = sin(fptemp);
+        fpush();
+        ST0 = cos(fptemp);
+        env->fpus &= (~0x400);  /* C2 <-- 0 */
+        /* the above code is for  |arg| < 2**63 only */
+    }
+}
+
+void helper_frndint(void)
+{
+    ST0 = rint(ST0);
+}
+
+void helper_fscale(void)
+{
+    CPU86_LDouble fpsrcop, fptemp;
+
+    fpsrcop = 2.0;
+    fptemp = pow(fpsrcop,ST1);
+    ST0 *= fptemp;
+}
+
+void helper_fsin(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if ((fptemp > MAXTAN)||(fptemp < -MAXTAN)) {
+        env->fpus |= 0x400;
+    } else {
+        ST0 = sin(fptemp);
+        env->fpus &= (~0x400);  /* C2 <-- 0 */
+        /* the above code is for  |arg| < 2**53 only */
+    }
+}
+
+void helper_fcos(void)
+{
+    CPU86_LDouble fptemp;
+
+    fptemp = ST0;
+    if((fptemp > MAXTAN)||(fptemp < -MAXTAN)) {
+        env->fpus |= 0x400;
+    } else {
+        ST0 = cos(fptemp);
+        env->fpus &= (~0x400);  /* C2 <-- 0 */
+        /* the above code is for  |arg5 < 2**63 only */
+    }
+}
+
+/* associated heplers to reduce generated code length and to simplify
+   relocation (FP constants are usually stored in .rodata section) */
+
+void OPPROTO op_f2xm1(void)
+{
+    helper_f2xm1();
+}
+
+void OPPROTO op_fyl2x(void)
+{
+    helper_fyl2x();
+}
+
+void OPPROTO op_fptan(void)
+{
+    helper_fptan();
+}
+
+void OPPROTO op_fpatan(void)
+{
+    helper_fpatan();
+}
+
+void OPPROTO op_fxtract(void)
+{
+    helper_fxtract();
+}
+
+void OPPROTO op_fprem1(void)
+{
+    helper_fprem1();
+}
+
+
+void OPPROTO op_fprem(void)
+{
+    helper_fprem();
+}
+
+void OPPROTO op_fyl2xp1(void)
+{
+    helper_fyl2xp1();
+}
+
+void OPPROTO op_fsqrt(void)
+{
+    helper_fsqrt();
+}
+
+void OPPROTO op_fsincos(void)
+{
+    helper_fsincos();
+}
+
+void OPPROTO op_frndint(void)
+{
+    helper_frndint();
+}
+
+void OPPROTO op_fscale(void)
+{
+    helper_fscale();
+}
+
+void OPPROTO op_fsin(void)
+{
+    helper_fsin();
+}
+
+void OPPROTO op_fcos(void)
+{
+    helper_fcos();
+}
+
diff --git a/translate-i386.c b/translate-i386.c
index 0c1b95a923..0ad24e3a40 100644
--- a/translate-i386.c
+++ b/translate-i386.c
@@ -5,6 +5,7 @@
 #include <inttypes.h>
 #include <assert.h>
 
+#define IN_OP_I386
 #include "cpu-i386.h"
 
 static uint8_t *gen_code_ptr;
@@ -39,7 +40,6 @@ typedef struct DisasContext {
     int prefix;
     int aflag, dflag;
     uint8_t *pc; /* current pc */
-    uint8_t *runtime_pc; /* current pc in the runtime generated code */
     int cc_op; /* current CC operation */
     int f_st;
 } DisasContext;
@@ -68,18 +68,6 @@ enum {
     OP_SAR = 7,
 };
 
-
-static const int fp_ops[8] = {
-#if 0
-    OP_FADDQ, OP_FMULQ, OP_CMP, OP_CMP,
-    OP_FSUBQ, OP_FSUBQ, OP_FDIVQ, OP_FDIVQ
-#endif
-};
-
-extern char cc_table, rclw_table, rclb_table;
-extern char helper_rcll_T0_T1_cc;
-extern char __udivdi3, __umoddi3;
-
 #include "op-i386.h"
 
 /* operand size */
@@ -606,6 +594,28 @@ static GenOpFunc *gen_setcc_sub[3][8] = {
     },
 };
 
+static GenOpFunc *gen_op_fp_arith_ST0_FT0[8] = {
+    gen_op_fadd_ST0_FT0,
+    gen_op_fmul_ST0_FT0,
+    gen_op_fcom_ST0_FT0,
+    gen_op_fcom_ST0_FT0,
+    gen_op_fsub_ST0_FT0,
+    gen_op_fsubr_ST0_FT0,
+    gen_op_fdiv_ST0_FT0,
+    gen_op_fdivr_ST0_FT0,
+};
+
+static GenOpFunc1 *gen_op_fp_arith_STN_ST0[8] = {
+    gen_op_fadd_STN_ST0,
+    gen_op_fmul_STN_ST0,
+    NULL,
+    NULL,
+    gen_op_fsub_STN_ST0,
+    gen_op_fsubr_STN_ST0,
+    gen_op_fdiv_STN_ST0,
+    gen_op_fdivr_STN_ST0,
+};
+
 static void gen_op(DisasContext *s1, int op, int ot, int d, int s)
 {
     if (d != OR_TMP0)
@@ -1345,12 +1355,12 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
         /**************************/
         /* push/pop */
     case 0x50 ... 0x57: /* push */
-        gen_op_mov_TN_reg[OT_LONG][0][(b & 7)]();
+        gen_op_mov_TN_reg[OT_LONG][0][b & 7]();
         gen_op_pushl_T0();
         break;
     case 0x58 ... 0x5f: /* pop */
         gen_op_popl_T0();
-        gen_op_mov_reg_T0[OT_LONG][reg]();
+        gen_op_mov_reg_T0[OT_LONG][b & 7]();
         break;
     case 0x68: /* push Iv */
     case 0x6a:
@@ -1581,7 +1591,6 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
 
         /************************/
         /* floats */
-#if 0        
     case 0xd8 ... 0xdf: 
         modrm = ldub(s->pc++);
         mod = (modrm >> 6) & 3;
@@ -1597,52 +1606,29 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             case 0x20 ... 0x27: /* fxxxl */
             case 0x30 ... 0x37: /* fixxx */
                 {
-                    int op1, swap;
-                    op1 = fp_ops[op & 7];
-
-                    swap = 0;
-                    if ((op & 7) == 5 || (op & 7) == 7)
-                        swap = 1;
+                    int op1;
+                    op1 = op & 7;
 
                     switch(op >> 4) {
                     case 0:
-                        ot = OT_LONG;
-                        is_int = 0;
+                        gen_op_flds_FT0_A0();
                         break;
                     case 1:
-                        ot = OT_LONG;
-                        is_int = 1;
+                        gen_op_fildl_FT0_A0();
                         break;
                     case 2:
-                        ot = OT_QUAD;
-                        is_int = 0;
+                        gen_op_fldl_FT0_A0();
                         break;
                     case 3:
                     default:
-                        ot = OT_WORD;
-                        is_int = 1;
+                        gen_op_fild_FT0_A0();
                         break;
                     }
                     
-                    /* if integer, needs to convert to float */
-                    if (is_int) {
-                        /* XXX: potential loss of precision if large integer */
-                        gen_ld(OP_LDUB + ot, OR_TMP0, reg_addr, offset_addr);
-                        gen_insn2(OP_I2FL, OR_FTMP0, OR_TMP0);
-                    } else {
-                        gen_ld(OP_LDUB + ot, OR_FTMP0, reg_addr, offset_addr);
-                    }
-                    if (ot != OT_QUAD)
-                        op1 += OP_FADDL - OP_FADDQ;
-
-                    if (!swap)
-                        gen_insn3(op1, OR_ST0, OR_ST0, OR_FTMP0);
-                    else
-                        gen_insn3(op1, OR_ST0, OR_FTMP0, OR_ST0);
-                        
-                    if ((op & 7) == 3) {
+                    gen_op_fp_arith_ST0_FT0[op1]();
+                    if (op1 == 3) {
                         /* fcomp needs pop */
-                        gen_insn0(OP_FPOP);
+                        gen_op_fpop();
                     }
                 }
                 break;
@@ -1659,49 +1645,47 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             case 0x3a: /* fists */
             case 0x3b: /* fistps */
                 
-                switch(op >> 4) {
-                case 0:
-                    ot = OT_LONG;
-                    is_int = 0;
-                    break;
-                case 1:
-                    ot = OT_LONG;
-                    is_int = 1;
-                    break;
-                case 2:
-                    ot = OT_QUAD;
-                    is_int = 0;
-                    break;
-                case 3:
-                default:
-                    ot = OT_WORD;
-                    is_int = 1;
-                    break;
-                }
-
                 switch(op & 7) {
                 case 0:
-                    gen_insn0(OP_FPUSH);
-                    if (is_int) {
-                        /* XXX: potential loss of precision */
-                        gen_ld(OP_LDUB + ot, OR_TMP0, reg_addr, offset_addr);
-                        gen_insn2(OP_I2FL, OR_ST0, OR_TMP0);
-                    } else {
-                        gen_ld(OP_LDUB + ot, OR_ST0, reg_addr, offset_addr);
+                    gen_op_fpush();
+                    switch(op >> 4) {
+                    case 0:
+                        gen_op_flds_ST0_A0();
+                        break;
+                    case 1:
+                        gen_op_fildl_ST0_A0();
+                        break;
+                    case 2:
+                        gen_op_fldl_ST0_A0();
+                        break;
+                    case 3:
+                    default:
+                        gen_op_fild_ST0_A0();
+                        break;
                     }
                     break;
                 default:
-                    if (is_int) {
-                        gen_insn2(OP_F2IL, OR_TMP0, OR_ST0);
-                        gen_st(OP_STB + ot, OR_TMP0, reg_addr, offset_addr);
-                    } else {
-                        gen_st(OP_STB + ot, OR_ST0, reg_addr, offset_addr);
+                    switch(op >> 4) {
+                    case 0:
+                        gen_op_fsts_ST0_A0();
+                        break;
+                    case 1:
+                        gen_op_fistl_ST0_A0();
+                        break;
+                    case 2:
+                        gen_op_fstl_ST0_A0();
+                        break;
+                    case 3:
+                    default:
+                        gen_op_fist_ST0_A0();
+                        break;
                     }
                     if ((op & 7) == 3)
-                        gen_insn0(OP_FPOP);
+                        gen_op_fpop();
                     break;
                 }
                 break;
+#if 0
             case 0x2f: /* fnstsw mem */
                 gen_insn3(OP_FNSTS, OR_TMP0, OR_ZERO, OR_ZERO);
                 gen_st(OP_STW, OR_TMP0, reg_addr, offset_addr);
@@ -1711,15 +1695,14 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             case 0x3e: /* fbstp */
                 error("float BCD not hanlded");
                 return -1;
+#endif
             case 0x3d: /* fildll */
-                gen_insn0(OP_FPUSH);
-                gen_ld(OP_LDQ, OR_TMP0, reg_addr, offset_addr);
-                gen_insn2(OP_I2FQ, OR_ST0, OR_TMP0);
+                gen_op_fpush();
+                gen_op_fildll_ST0_A0();
                 break;
             case 0x3f: /* fistpll */
-                gen_insn2(OP_F2IQ, OR_TMP0, OR_ST0);
-                gen_st(OP_STQ, OR_TMP0, reg_addr, offset_addr);
-                gen_insn0(OP_FPOP);
+                gen_op_fistll_ST0_A0();
+                gen_op_fpop();
                 break;
             default:
                 error("unhandled memory FP\n");
@@ -1727,22 +1710,19 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             }
         } else {
             /* register float ops */
-            opreg = rm + OR_ST0;
+            opreg = rm;
 
             switch(op) {
             case 0x08: /* fld sti */
-                gen_insn0(OP_FPUSH);
-                gen_mov(OR_ST0, OR_ST0 + ((rm + 1) & 7));
+                gen_op_fpush();
+                gen_op_fmov_ST0_STN((opreg + 1) & 7);
                 break;
             case 0x09: /* fxchg sti */
-                gen_mov(OR_TMP0, OR_ST0);
-                gen_mov(OR_ST0, opreg);
-                gen_mov(opreg, OR_TMP0);
+                gen_op_fxchg_ST0_STN((opreg + 1) & 7);
                 break;
             case 0x0a: /* grp d9/2 */
                 switch(rm) {
                 case 0: /* fnop */
-                    gen_insn0(OP_NOP);
                     break;
                 default:
                     error("unhandled FP GRP d9/2\n");
@@ -1752,16 +1732,17 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             case 0x0c: /* grp d9/4 */
                 switch(rm) {
                 case 0: /* fchs */
-                    gen_insn3(OP_FSUBQ, OR_ST0, OR_ZERO, OR_ST0);
+                    gen_op_fchs_ST0();
                     break;
                 case 1: /* fabs */
-                    gen_insn2(OP_FABSQ, OR_ST0, OR_ST0);
+                    gen_op_fabs_ST0();
                     break;
                 case 4: /* ftst */
-                    gen_insn3(OP_CMP, OR_ZERO, OR_ST0, OR_ZERO);
+                    gen_op_fldz_FT0();
+                    gen_op_fcom_ST0_FT0();
                     break;
                 case 5: /* fxam */
-                    gen_insn3(OP_FSPECIAL, OR_ZERO, OR_ST0, OR_ZERO);
+                    gen_op_fxam_ST0();
                     break;
                 default:
                     return -1;
@@ -1769,76 +1750,88 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
                 break;
             case 0x0d: /* grp d9/5 */
                 {
-                    if (rm == 7) {
-                        error("bad GRP d9/5");
+                    switch(rm) {
+                    case 0:
+                        gen_op_fld1_ST0();
+                        break;
+                    case 1:
+                        gen_op_fld2t_ST0();
+                        break;
+                    case 2:
+                        gen_op_fld2e_ST0();
+                        break;
+                    case 3:
+                        gen_op_fldpi_ST0();
+                        break;
+                    case 4:
+                        gen_op_fldlg2_ST0();
+                        break;
+                    case 5:
+                        gen_op_fldln2_ST0();
+                        break;
+                    case 6:
+                        gen_op_fldz_ST0();
+                        break;
+                    default:
                         return -1;
                     }
-                    /* XXX: needs constant load or symbol table */
-                    gen_insn0(OP_FPUSH);
-                    gen_ld(OP_LDQ, OR_ST0, OR_ZERO, 
-                               (rm * 8) + FLOAT_CONST_ADDR);
                 }
                 break;
             case 0x0e: /* grp d9/6 */
                 switch(rm) {
                 case 0: /* f2xm1 */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ZERO);
+                    gen_op_f2xm1();
                     break;
                 case 1: /* fyl2x */
-                    gen_insn3(OP_FSPECIAL, OR_ST1, OR_ST0, OR_ST1);
-                    gen_insn0(OP_FPOP);
+                    gen_op_fyl2x();
                     break;
                 case 2: /* fptan */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ZERO);
-                    gen_insn0(OP_FPUSH);
-                    /* load one */
-                    gen_ld(OP_LDQ, OR_ST0, OR_ZERO, 
-                               (0 * 8) + FLOAT_CONST_ADDR);
+                    gen_op_fptan();
                     break;
                 case 3: /* fpatan */
-                    gen_insn3(OP_FSPECIAL, OR_ST1, OR_ST0, OR_ST1);
-                    gen_insn0(OP_FPOP);
+                    gen_op_fpatan();
                     break;
                 case 4: /* fxtract */
-                    gen_insn0(OP_FPUSH);
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST1, OR_ZERO);
-                    gen_insn3(OP_FSPECIAL, OR_ST1, OR_ST1, OR_ZERO);
+                    gen_op_fxtract();
                     break;
                 case 5: /* fprem1 */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ST1);
+                    gen_op_fprem1();
                     break;
                 case 6: /* fdecstp */
-                    gen_insn0(OP_FPUSH);
+                    gen_op_fdecstp();
                     break;
                 default:
-                case 7: /* fdecstp */
-                    gen_insn0(OP_FPOP);
+                case 7: /* fincstp */
+                    gen_op_fincstp();
                     break;
                 }
                 break;
             case 0x0f: /* grp d9/7 */
                 switch(rm) {
                 case 0: /* fprem */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ST1);
+                    gen_op_fprem();
                     break;
                 case 1: /* fyl2xp1 */
-                    gen_insn3(OP_FSPECIAL, OR_ST1, OR_ST0, OR_ST1);
-                    gen_insn0(OP_FPOP);
+                    gen_op_fyl2xp1();
+                    break;
+                case 2: /* fsqrt */
+                    gen_op_fsqrt();
                     break;
                 case 3: /* fsincos */
-                    gen_insn0(OP_FPUSH);
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST1, OR_ZERO);
-                    gen_insn3(OP_FSPECIAL, OR_ST1, OR_ST1, OR_ZERO);
+                    gen_op_fsincos();
                     break;
                 case 5: /* fscale */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ST1);
+                    gen_op_fscale();
                     break;
-                case 2: /* fsqrt */
                 case 4: /* frndint */
+                    gen_op_frndint();
+                    break;
                 case 6: /* fsin */
+                    gen_op_fsin();
+                    break;
                 default:
                 case 7: /* fcos */
-                    gen_insn3(OP_FSPECIAL, OR_ST0, OR_ST0, OR_ZERO);
+                    gen_op_fcos();
                     break;
                 }
                 break;
@@ -1846,58 +1839,54 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             case 0x20: case 0x21: case 0x24 ... 0x27: /* fxxx sti, st */
             case 0x30: case 0x31: case 0x34 ... 0x37: /* fxxxp sti, st */
                 {
-                    int op1, swap;
+                    int op1;
                     
-                    op1 = fp_ops[op & 7];
-                    swap = 0;
-                    if ((op & 7) == 5 || (op & 7) == 7)
-                        swap = 1;
+                    op1 = op & 7;
                     if (op >= 0x20) {
-                        if (swap) 
-                            gen_insn3(op1, opreg, OR_ST0, opreg);
-                        else
-                            gen_insn3(op1, opreg, opreg, OR_ST0);
+                        gen_op_fp_arith_STN_ST0[op1](opreg);
                     } else {
-                        if (swap)
-                            gen_insn3(op1, OR_ST0, opreg, OR_ST0);
-                        else
-                            gen_insn3(op1, OR_ST0, OR_ST0, opreg);
+                        gen_op_fmov_FT0_STN(opreg);
+                        gen_op_fp_arith_ST0_FT0[op1]();
                     }
                     if (op >= 0x30)
-                        gen_insn0(OP_FPOP);
+                        gen_op_fpop();
                 }
                 break;
             case 0x02: /* fcom */
-                gen_insn3(OP_CMP, OR_ZERO, OR_ST0, opreg);
+                gen_op_fmov_FT0_STN(opreg);
+                gen_op_fcom_ST0_FT0();
                 break;
             case 0x03: /* fcomp */
-                gen_insn3(OP_CMP, OR_ZERO, OR_ST0, opreg);
-                gen_insn0(OP_FPOP);
+                gen_op_fmov_FT0_STN(opreg);
+                gen_op_fcom_ST0_FT0();
+                gen_op_fpop();
                 break;
             case 0x15: /* da/5 */
                 switch(rm) {
                 case 1: /* fucompp */
-                    gen_insn3(OP_CMP, OR_ZERO, OR_ST0, opreg);
-                    gen_insn0(OP_FPOP);
-                    gen_insn0(OP_FPOP);
+                    gen_op_fmov_FT0_STN(1);
+                    gen_op_fcom_ST0_FT0();
+                    gen_op_fpop();
+                    gen_op_fpop();
                     break;
                 default:
                     return -1;
                 }
                 break;
             case 0x2a: /* fst sti */
-                gen_mov(opreg, OR_ST0);
+                gen_op_fmov_STN_ST0(opreg);
                 break;
             case 0x2b: /* fstp sti */
-                gen_mov(opreg, OR_ST0);
-                gen_insn0(OP_FPOP);
+                gen_op_fmov_STN_ST0(opreg);
+                gen_op_fpop();
                 break;
             case 0x33: /* de/3 */
                 switch(rm) {
                 case 1: /* fcompp */
-                    gen_insn3(OP_CMP, OR_ZERO, OR_ST0, opreg);
-                    gen_insn0(OP_FPOP);
-                    gen_insn0(OP_FPOP);
+                    gen_op_fmov_FT0_STN(1);
+                    gen_op_fcom_ST0_FT0();
+                    gen_op_fpop();
+                    gen_op_fpop();
                     break;
                 default:
                     return -1;
@@ -1905,9 +1894,11 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
                 break;
             case 0x3c: /* df/4 */
                 switch(rm) {
+#if 0
                 case 0:
                     gen_insn3(OP_FNSTS, OR_EAX, OR_ZERO, OR_ZERO);
                     break;
+#endif
                 default:
                     return -1;
                 }
@@ -1918,7 +1909,6 @@ int disas_insn(DisasContext *s, uint8_t *pc_start)
             }
         }
         break;
-#endif
         /************************/
         /* string ops */
     case 0xa4: /* movsS */