soft float support

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@1336 c046a42c-6fe2-441c-8c8c-71466251a162

9 files changed, 139 insertions, 231 deletions

diff --git a/dyngen-exec.h b/dyngen-exec.h
index d8ecc02ffb..9b39f59b56 100644
--- a/dyngen-exec.h
+++ b/dyngen-exec.h
@@ -62,17 +62,6 @@ extern int fprintf(FILE *, const char *, ...);
 extern int printf(const char *, ...);
 #undef NULL
 #define NULL 0
-#if defined(_BSD) && !defined(__APPLE__)
-#include <ieeefp.h>
-
-#define FE_TONEAREST   FP_RN
-#define FE_DOWNWARD    FP_RM
-#define FE_UPWARD      FP_RP
-#define FE_TOWARDZERO  FP_RZ
-#define fesetround(x)  fpsetround(x)
-#else
-#include <fenv.h>
-#endif
 
 #ifdef __i386__
 #define AREG0 "ebp"
diff --git a/target-i386/cpu.h b/target-i386/cpu.h
index 154d8bfdae..ce65c34306 100644
--- a/target-i386/cpu.h
+++ b/target-i386/cpu.h
@@ -36,6 +36,8 @@
 
 #include "cpu-defs.h"
 
+#include "softfloat.h"
+
 #if defined(__i386__) && !defined(CONFIG_SOFTMMU)
 #define USE_CODE_COPY
 #endif
@@ -332,14 +334,14 @@ enum {
     CC_OP_NB,
 };
 
-#if (defined(__i386__) || defined(__x86_64__)) && !defined(_BSD)
+#ifdef FLOATX80
 #define USE_X86LDOUBLE
 #endif
 
 #ifdef USE_X86LDOUBLE
-typedef long double CPU86_LDouble;
+typedef floatx80 CPU86_LDouble;
 #else
-typedef double CPU86_LDouble;
+typedef float64 CPU86_LDouble;
 #endif
 
 typedef struct SegmentCache {
@@ -354,8 +356,8 @@ typedef union {
     uint16_t _w[8];
     uint32_t _l[4];
     uint64_t _q[2];
-    float _s[4];
-    double _d[2];
+    float32 _s[4];
+    float64 _d[2];
 } XMMReg;
 
 typedef union {
@@ -441,6 +443,7 @@ typedef struct CPUX86State {
     } fpregs[8];
 
     /* emulator internal variables */
+    float_status fp_status;
     CPU86_LDouble ft0;
     union {
 	float f;
@@ -449,6 +452,7 @@ typedef struct CPUX86State {
         int64_t i64;
     } fp_convert;
     
+    float_status sse_status;
     uint32_t mxcsr;
     XMMReg xmm_regs[CPU_NB_REGS];
     XMMReg xmm_t0;
diff --git a/target-i386/exec.h b/target-i386/exec.h
index c4e59c4235..137774f5a3 100644
--- a/target-i386/exec.h
+++ b/target-i386/exec.h
@@ -139,44 +139,6 @@ extern int loglevel;
 #include "cpu.h"
 #include "exec-all.h"
 
-/* XXX: add a generic FPU library */
-
-static inline double float32_to_float64(float a)
-{
-    return a;
-}
-
-static inline float float64_to_float32(double a)
-{
-    return a;
-}
-
-#if defined(__powerpc__)
-/* better to call an helper on ppc */
-float int32_to_float32(int32_t a);
-double int32_to_float64(int32_t a);
-#else
-static inline float int32_to_float32(int32_t a)
-{
-    return (float)a;
-}
-
-static inline double int32_to_float64(int32_t a)
-{
-    return (double)a;
-}
-#endif
-
-static inline float int64_to_float32(int64_t a)
-{
-    return (float)a;
-}
-
-static inline double int64_to_float64(int64_t a)
-{
-    return (double)a;
-}
-
 typedef struct CCTable {
     int (*compute_all)(void); /* return all the flags */
     int (*compute_c)(void);  /* return the C flag */
@@ -358,9 +320,11 @@ static inline void stfl(target_ulong ptr, float v)
 
 #ifdef USE_X86LDOUBLE
 /* use long double functions */
-#define lrint lrintl
-#define llrint llrintl
-#define fabs fabsl
+#define floatx_to_int32 floatx80_to_int32
+#define floatx_to_int64 floatx80_to_int64
+#define floatx_abs floatx80_abs
+#define floatx_chs floatx80_chs
+#define floatx_round_to_int floatx80_round_to_int
 #define sin sinl
 #define cos cosl
 #define sqrt sqrtl
@@ -370,17 +334,14 @@ static inline void stfl(target_ulong ptr, float v)
 #define atan2 atan2l
 #define floor floorl
 #define ceil ceill
-#define rint rintl
-#endif
-
-#if !defined(_BSD)
-extern int lrint(CPU86_LDouble x);
-extern int64_t llrint(CPU86_LDouble x);
 #else
-#define lrint(d)		((int)rint(d))
-#define llrint(d)		((int)rint(d))
+#define floatx_to_int32 float64_to_int32
+#define floatx_to_int64 float64_to_int64
+#define floatx_abs float64_abs
+#define floatx_chs float64_chs
+#define floatx_round_to_int float64_round_to_int
 #endif
-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);
@@ -390,7 +351,6 @@ 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
@@ -400,13 +360,6 @@ extern CPU86_LDouble rint(CPU86_LDouble x);
 
 #define MAXTAN 9223372036854775808.0
 
-#ifdef __arm__
-/* we have no way to do correct rounding - a FPU emulator is needed */
-#define FE_DOWNWARD   FE_TONEAREST
-#define FE_UPWARD     FE_TONEAREST
-#define FE_TOWARDZERO FE_TONEAREST
-#endif
-
 #ifdef USE_X86LDOUBLE
 
 /* only for x86 */
@@ -596,6 +549,7 @@ float approx_rsqrt(float a);
 float approx_rcp(float a);
 double helper_sqrt(double a);
 int fpu_isnan(double a);
+void update_fp_status(void);
 
 extern const uint8_t parity_table[256];
 extern const uint8_t rclw_table[32];
diff --git a/target-i386/helper.c b/target-i386/helper.c
index 53b85d3aaa..64cb51ab53 100644
--- a/target-i386/helper.c
+++ b/target-i386/helper.c
@@ -2541,13 +2541,11 @@ void helper_fbld_ST0_A0(void)
 
 void helper_fbst_ST0_A0(void)
 {
-    CPU86_LDouble tmp;
     int v;
     target_ulong mem_ref, mem_end;
     int64_t val;
 
-    tmp = rint(ST0);
-    val = (int64_t)tmp;
+    val = floatx_to_int64(ST0, &env->fp_status);
     mem_ref = A0;
     mem_end = mem_ref + 9;
     if (val < 0) {
@@ -2740,29 +2738,7 @@ void helper_fsincos(void)
 
 void helper_frndint(void)
 {
-    CPU86_LDouble a;
-
-    a = ST0;
-#ifdef __arm__
-    switch(env->fpuc & RC_MASK) {
-    default:
-    case RC_NEAR:
-        asm("rndd %0, %1" : "=f" (a) : "f"(a));
-        break;
-    case RC_DOWN:
-        asm("rnddm %0, %1" : "=f" (a) : "f"(a));
-        break;
-    case RC_UP:
-        asm("rnddp %0, %1" : "=f" (a) : "f"(a));
-        break;
-    case RC_CHOP:
-        asm("rnddz %0, %1" : "=f" (a) : "f"(a));
-        break;
-    }
-#else
-    a = rint(a);
-#endif
-    ST0 = a;
+    ST0 = floatx_round_to_int(ST0, &env->fp_status);
 }
 
 void helper_fscale(void)
@@ -3263,25 +3239,43 @@ float approx_rcp(float a)
     return 1.0 / a;
 }
 
-/* XXX: find a better solution */
-double helper_sqrt(double a)
+void update_fp_status(void)
 {
-    return sqrt(a);
-}
+    int rnd_type;
 
-/* XXX: move that to another file */
-#if defined(__powerpc__)
-/* better to call an helper on ppc */
-float int32_to_float32(int32_t a)
-{
-    return (float)a;
-}
-
-double int32_to_float64(int32_t a)
-{
-    return (double)a;
-}
+    /* set rounding mode */
+    switch(env->fpuc & RC_MASK) {
+    default:
+    case RC_NEAR:
+        rnd_type = float_round_nearest_even;
+        break;
+    case RC_DOWN:
+        rnd_type = float_round_down;
+        break;
+    case RC_UP:
+        rnd_type = float_round_up;
+        break;
+    case RC_CHOP:
+        rnd_type = float_round_to_zero;
+        break;
+    }
+    set_float_rounding_mode(rnd_type, &env->fp_status);
+#ifdef FLOATX80
+    switch((env->fpuc >> 8) & 3) {
+    case 0:
+        rnd_type = 32;
+        break;
+    case 2:
+        rnd_type = 64;
+        break;
+    case 3:
+    default:
+        rnd_type = 80;
+        break;
+    }
+    set_floatx80_rounding_precision(rnd_type, &env->fp_status);
 #endif
+}
 
 #if !defined(CONFIG_USER_ONLY) 
 
diff --git a/target-i386/op.c b/target-i386/op.c
index dc0dceadfd..bbc2b19aed 100644
--- a/target-i386/op.c
+++ b/target-i386/op.c
@@ -1598,26 +1598,6 @@ CCTable cc_table[CC_OP_NB] = {
    functions comes from the LGPL'ed x86 emulator found in the Willows
    TWIN windows emulator. */
 
-#if defined(__powerpc__)
-extern CPU86_LDouble copysign(CPU86_LDouble, CPU86_LDouble);
-
-/* correct (but slow) PowerPC rint() (glibc version is incorrect) */
-double qemu_rint(double x)
-{
-    double y = 4503599627370496.0;
-    if (fabs(x) >= y)
-        return x;
-    if (x < 0) 
-        y = -y;
-    y = (x + y) - y;
-    if (y == 0.0)
-        y = copysign(y, x);
-    return y;
-}
-
-#define rint qemu_rint
-#endif
-
 /* fp load FT0 */
 
 void OPPROTO op_flds_FT0_A0(void)
@@ -1866,7 +1846,7 @@ void OPPROTO op_fist_ST0_A0(void)
     int val;
 
     d = ST0;
-    val = lrint(d);
+    val = floatx_to_int32(d, &env->fp_status);
     if (val != (int16_t)val)
         val = -32768;
     stw(A0, val);
@@ -1883,7 +1863,7 @@ void OPPROTO op_fistl_ST0_A0(void)
     int val;
 
     d = ST0;
-    val = lrint(d);
+    val = floatx_to_int32(d, &env->fp_status);
     stl(A0, val);
     FORCE_RET();
 }
@@ -1898,7 +1878,7 @@ void OPPROTO op_fistll_ST0_A0(void)
     int64_t val;
 
     d = ST0;
-    val = llrint(d);
+    val = floatx_to_int64(d, &env->fp_status);
     stq(A0, val);
     FORCE_RET();
 }
@@ -2101,12 +2081,12 @@ void OPPROTO op_fdivr_STN_ST0(void)
 /* misc FPU operations */
 void OPPROTO op_fchs_ST0(void)
 {
-    ST0 = -ST0;
+    ST0 = floatx_chs(ST0);
 }
 
 void OPPROTO op_fabs_ST0(void)
 {
-    ST0 = fabs(ST0);
+    ST0 = floatx_abs(ST0);
 }
 
 void OPPROTO op_fxam_ST0(void)
@@ -2251,25 +2231,8 @@ void OPPROTO op_fnstcw_A0(void)
 
 void OPPROTO op_fldcw_A0(void)
 {
-    int rnd_type;
     env->fpuc = lduw(A0);
-    /* set rounding mode */
-    switch(env->fpuc & RC_MASK) {
-    default:
-    case RC_NEAR:
-        rnd_type = FE_TONEAREST;
-        break;
-    case RC_DOWN:
-        rnd_type = FE_DOWNWARD;
-        break;
-    case RC_UP:
-        rnd_type = FE_UPWARD;
-        break;
-    case RC_CHOP:
-        rnd_type = FE_TOWARDZERO;
-        break;
-    }
-    fesetround(rnd_type);
+    update_fp_status();
 }
 
 void OPPROTO op_fclex(void)
diff --git a/target-i386/ops_sse.h b/target-i386/ops_sse.h
index 8472a34d35..be1d68bfb7 100644
--- a/target-i386/ops_sse.h
+++ b/target-i386/ops_sse.h
@@ -654,10 +654,10 @@ void OPPROTO op_ ## name ## ps (void)\
     Reg *d, *s;\
     d = (Reg *)((char *)env + PARAM1);\
     s = (Reg *)((char *)env + PARAM2);\
-    d->XMM_S(0) = F(d->XMM_S(0), s->XMM_S(0));\
-    d->XMM_S(1) = F(d->XMM_S(1), s->XMM_S(1));\
-    d->XMM_S(2) = F(d->XMM_S(2), s->XMM_S(2));\
-    d->XMM_S(3) = F(d->XMM_S(3), s->XMM_S(3));\
+    d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
+    d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
+    d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
+    d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 }\
 \
 void OPPROTO op_ ## name ## ss (void)\
@@ -665,15 +665,15 @@ void OPPROTO op_ ## name ## ss (void)\
     Reg *d, *s;\
     d = (Reg *)((char *)env + PARAM1);\
     s = (Reg *)((char *)env + PARAM2);\
-    d->XMM_S(0) = F(d->XMM_S(0), s->XMM_S(0));\
+    d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 }\
 void OPPROTO op_ ## name ## pd (void)\
 {\
     Reg *d, *s;\
     d = (Reg *)((char *)env + PARAM1);\
     s = (Reg *)((char *)env + PARAM2);\
-    d->XMM_D(0) = F(d->XMM_D(0), s->XMM_D(0));\
-    d->XMM_D(1) = F(d->XMM_D(1), s->XMM_D(1));\
+    d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
+    d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 }\
 \
 void OPPROTO op_ ## name ## sd (void)\
@@ -681,16 +681,16 @@ void OPPROTO op_ ## name ## sd (void)\
     Reg *d, *s;\
     d = (Reg *)((char *)env + PARAM1);\
     s = (Reg *)((char *)env + PARAM2);\
-    d->XMM_D(0) = F(d->XMM_D(0), s->XMM_D(0));\
+    d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 }
 
-#define FPU_ADD(a, b) (a) + (b)
-#define FPU_SUB(a, b) (a) - (b)
-#define FPU_MUL(a, b) (a) * (b)
-#define FPU_DIV(a, b) (a) / (b)
-#define FPU_MIN(a, b) (a) < (b) ? (a) : (b)
-#define FPU_MAX(a, b) (a) > (b) ? (a) : (b)
-#define FPU_SQRT(a, b) helper_sqrt(b)
+#define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
+#define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
+#define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
+#define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
+#define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
+#define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
+#define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 
 SSE_OP_S(add, FPU_ADD)
 SSE_OP_S(sub, FPU_SUB)
@@ -710,8 +710,8 @@ void OPPROTO op_cvtps2pd(void)
     s = (Reg *)((char *)env + PARAM2);
     s0 = s->XMM_S(0);
     s1 = s->XMM_S(1);
-    d->XMM_D(0) = float32_to_float64(s0);
-    d->XMM_D(1) = float32_to_float64(s1);
+    d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
+    d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 }
 
 void OPPROTO op_cvtpd2ps(void)
@@ -719,8 +719,8 @@ void OPPROTO op_cvtpd2ps(void)
     Reg *d, *s;
     d = (Reg *)((char *)env + PARAM1);
     s = (Reg *)((char *)env + PARAM2);
-    d->XMM_S(0) = float64_to_float32(s->XMM_D(0));
-    d->XMM_S(1) = float64_to_float32(s->XMM_D(1));
+    d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
+    d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
     d->Q(1) = 0;
 }
 
@@ -729,7 +729,7 @@ void OPPROTO op_cvtss2sd(void)
     Reg *d, *s;
     d = (Reg *)((char *)env + PARAM1);
     s = (Reg *)((char *)env + PARAM2);
-    d->XMM_D(0) = float32_to_float64(s->XMM_S(0));
+    d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 }
 
 void OPPROTO op_cvtsd2ss(void)
@@ -737,7 +737,7 @@ void OPPROTO op_cvtsd2ss(void)
     Reg *d, *s;
     d = (Reg *)((char *)env + PARAM1);
     s = (Reg *)((char *)env + PARAM2);
-    d->XMM_S(0) = float64_to_float32(s->XMM_D(0));
+    d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 }
 
 /* integer to float */
@@ -745,10 +745,10 @@ void OPPROTO op_cvtdq2ps(void)
 {
     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->XMM_S(0) = int32_to_float32(s->XMM_L(0));
-    d->XMM_S(1) = int32_to_float32(s->XMM_L(1));
-    d->XMM_S(2) = int32_to_float32(s->XMM_L(2));
-    d->XMM_S(3) = int32_to_float32(s->XMM_L(3));
+    d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
+    d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
+    d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
+    d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 }
 
 void OPPROTO op_cvtdq2pd(void)
@@ -758,49 +758,49 @@ void OPPROTO op_cvtdq2pd(void)
     int32_t l0, l1;
     l0 = (int32_t)s->XMM_L(0);
     l1 = (int32_t)s->XMM_L(1);
-    d->XMM_D(0) = int32_to_float64(l0);
-    d->XMM_D(1) = int32_to_float64(l1);
+    d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
+    d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 }
 
 void OPPROTO op_cvtpi2ps(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
-    d->XMM_S(0) = int32_to_float32(s->MMX_L(0));
-    d->XMM_S(1) = int32_to_float32(s->MMX_L(1));
+    d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
+    d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 }
 
 void OPPROTO op_cvtpi2pd(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
-    d->XMM_D(0) = int32_to_float64(s->MMX_L(0));
-    d->XMM_D(1) = int32_to_float64(s->MMX_L(1));
+    d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
+    d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 }
 
 void OPPROTO op_cvtsi2ss(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
-    d->XMM_S(0) = int32_to_float32(T0);
+    d->XMM_S(0) = int32_to_float32(T0, &env->sse_status);
 }
 
 void OPPROTO op_cvtsi2sd(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
-    d->XMM_D(0) = int32_to_float64(T0);
+    d->XMM_D(0) = int32_to_float64(T0, &env->sse_status);
 }
 
 #ifdef TARGET_X86_64
 void OPPROTO op_cvtsq2ss(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
-    d->XMM_S(0) = int64_to_float32(T0);
+    d->XMM_S(0) = int64_to_float32(T0, &env->sse_status);
 }
 
 void OPPROTO op_cvtsq2sd(void)
 {
     XMMReg *d = (Reg *)((char *)env + PARAM1);
-    d->XMM_D(0) = int64_to_float64(T0);
+    d->XMM_D(0) = int64_to_float64(T0, &env->sse_status);
 }
 #endif
 
@@ -809,18 +809,18 @@ void OPPROTO op_cvtps2dq(void)
 {
     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->XMM_L(0) = lrint(s->XMM_S(0));
-    d->XMM_L(1) = lrint(s->XMM_S(1));
-    d->XMM_L(2) = lrint(s->XMM_S(2));
-    d->XMM_L(3) = lrint(s->XMM_S(3));
+    d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
+    d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
+    d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
+    d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 }
 
 void OPPROTO op_cvtpd2dq(void)
 {
     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->XMM_L(0) = lrint(s->XMM_D(0));
-    d->XMM_L(1) = lrint(s->XMM_D(1));
+    d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
+    d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
     d->XMM_Q(1) = 0;
 }
 
@@ -828,41 +828,41 @@ void OPPROTO op_cvtps2pi(void)
 {
     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->MMX_L(0) = lrint(s->XMM_S(0));
-    d->MMX_L(1) = lrint(s->XMM_S(1));
+    d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
+    d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 }
 
 void OPPROTO op_cvtpd2pi(void)
 {
     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->MMX_L(0) = lrint(s->XMM_D(0));
-    d->MMX_L(1) = lrint(s->XMM_D(1));
+    d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
+    d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 }
 
 void OPPROTO op_cvtss2si(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int32_t)lrint(s->XMM_S(0));
+    T0 = float32_to_int32(s->XMM_S(0), &env->sse_status);
 }
 
 void OPPROTO op_cvtsd2si(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int32_t)lrint(s->XMM_D(0));
+    T0 = float64_to_int32(s->XMM_D(0), &env->sse_status);
 }
 
 #ifdef TARGET_X86_64
 void OPPROTO op_cvtss2sq(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = llrint(s->XMM_S(0));
+    T0 = float32_to_int64(s->XMM_S(0), &env->sse_status);
 }
 
 void OPPROTO op_cvtsd2sq(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = llrint(s->XMM_D(0));
+    T0 = float64_to_int64(s->XMM_D(0), &env->sse_status);
 }
 #endif
 
@@ -871,18 +871,18 @@ void OPPROTO op_cvttps2dq(void)
 {
     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->XMM_L(0) = (int32_t)s->XMM_S(0);
-    d->XMM_L(1) = (int32_t)s->XMM_S(1);
-    d->XMM_L(2) = (int32_t)s->XMM_S(2);
-    d->XMM_L(3) = (int32_t)s->XMM_S(3);
+    d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
+    d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
+    d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
+    d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 }
 
 void OPPROTO op_cvttpd2dq(void)
 {
     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->XMM_L(0) = (int32_t)s->XMM_D(0);
-    d->XMM_L(1) = (int32_t)s->XMM_D(1);
+    d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
+    d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
     d->XMM_Q(1) = 0;
 }
 
@@ -890,41 +890,41 @@ void OPPROTO op_cvttps2pi(void)
 {
     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->MMX_L(0) = (int32_t)(s->XMM_S(0));
-    d->MMX_L(1) = (int32_t)(s->XMM_S(1));
+    d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
+    d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 }
 
 void OPPROTO op_cvttpd2pi(void)
 {
     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
-    d->MMX_L(0) = (int32_t)(s->XMM_D(0));
-    d->MMX_L(1) = (int32_t)(s->XMM_D(1));
+    d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
+    d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 }
 
 void OPPROTO op_cvttss2si(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int32_t)(s->XMM_S(0));
+    T0 = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 }
 
 void OPPROTO op_cvttsd2si(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int32_t)(s->XMM_D(0));
+    T0 = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 }
 
 #ifdef TARGET_X86_64
 void OPPROTO op_cvttss2sq(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int64_t)(s->XMM_S(0));
+    T0 = float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 }
 
 void OPPROTO op_cvttsd2sq(void)
 {
     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
-    T0 = (int64_t)(s->XMM_D(0));
+    T0 = float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 }
 #endif
 
diff --git a/target-sparc/cpu.h b/target-sparc/cpu.h
index b07e2afcf4..67a6e3e465 100644
--- a/target-sparc/cpu.h
+++ b/target-sparc/cpu.h
@@ -15,6 +15,8 @@
 
 #include "cpu-defs.h"
 
+#include "softfloat.h"
+
 /*#define EXCP_INTERRUPT 0x100*/
 
 /* trap definitions */
@@ -150,6 +152,7 @@ typedef struct CPUSPARCState {
     /* temporary float registers */
     float ft0, ft1, ft2;
     double dt0, dt1, dt2;
+    float_status fp_status;
 #if defined(TARGET_SPARC64)
     target_ulong t0, t1, t2;
 #endif
diff --git a/target-sparc/op_helper.c b/target-sparc/op_helper.c
index 9cb3de49ac..40ab90156e 100644
--- a/target-sparc/op_helper.c
+++ b/target-sparc/op_helper.c
@@ -1,5 +1,3 @@
-#include <math.h>
-#include <fenv.h>
 #include "exec.h"
 
 //#define DEBUG_MMU
@@ -24,17 +22,17 @@ void do_fitod(void)
 
 void do_fabss(void)
 {
-    FT0 = fabsf(FT1);
+    FT0 = float32_abs(FT1);
 }
 
 void do_fsqrts(void)
 {
-    FT0 = sqrtf(FT1);
+    FT0 = float32_sqrt(FT1, &env->fp_status);
 }
 
 void do_fsqrtd(void)
 {
-    DT0 = sqrt(DT1);
+    DT0 = float64_sqrt(DT1, &env->fp_status);
 }
 
 void do_fcmps (void)
@@ -252,20 +250,22 @@ void helper_rett()
 
 void helper_ldfsr(void)
 {
+    int rnd_mode;
     switch (env->fsr & FSR_RD_MASK) {
     case FSR_RD_NEAREST:
-	fesetround(FE_TONEAREST);
+        rnd_mode = float_round_nearest_even;
 	break;
     case FSR_RD_ZERO:
-	fesetround(FE_TOWARDZERO);
+        rnd_mode = float_round_to_zero;
 	break;
     case FSR_RD_POS:
-	fesetround(FE_UPWARD);
+        rnd_mode = float_round_up;
 	break;
     case FSR_RD_NEG:
-	fesetround(FE_DOWNWARD);
+        rnd_mode = float_round_down;
 	break;
     }
+    set_float_rounding_mode(rnd_mode, &env->fp_status);
 }
 
 void cpu_get_fp64(uint64_t *pmant, uint16_t *pexp, double f)
diff --git a/vl.c b/vl.c
index a815a50fa8..9f8b8a6b99 100644
--- a/vl.c
+++ b/vl.c
@@ -2271,6 +2271,7 @@ int cpu_load(QEMUFile *f, void *opaque, int version_id)
     }
 
     env->fpuc = fpuc;
+    /* XXX: restore FPU round state */
     env->fpstt = (fpus >> 11) & 7;
     env->fpus = fpus & ~0x3800;
     fptag ^= 0xff;