]> Git Repo - qemu.git/blobdiff - target-ppc/fpu_helper.c
progress: Allow regressing progress
[qemu.git] / target-ppc / fpu_helper.c
index 49cf09a53f8ba75a21073c7f1b9c7673b3917ced..9f2d53d7478a230fb83a66275636f3ae9b12748c 100644 (file)
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 #include "cpu.h"
-#include "helper.h"
+#include "exec/helper-proto.h"
+
+#define float64_snan_to_qnan(x) ((x) | 0x0008000000000000ULL)
+#define float32_snan_to_qnan(x) ((x) | 0x00400000)
 
 /*****************************************************************************/
 /* Floating point operations helpers */
@@ -50,65 +53,72 @@ static inline int isden(float64 d)
     return ((u.ll >> 52) & 0x7FF) == 0;
 }
 
-uint32_t helper_compute_fprf(CPUPPCState *env, uint64_t arg, uint32_t set_fprf)
+static inline int ppc_float32_get_unbiased_exp(float32 f)
+{
+    return ((f >> 23) & 0xFF) - 127;
+}
+
+static inline int ppc_float64_get_unbiased_exp(float64 f)
+{
+    return ((f >> 52) & 0x7FF) - 1023;
+}
+
+void helper_compute_fprf(CPUPPCState *env, uint64_t arg)
 {
     CPU_DoubleU farg;
     int isneg;
-    int ret;
+    int fprf;
 
     farg.ll = arg;
     isneg = float64_is_neg(farg.d);
     if (unlikely(float64_is_any_nan(farg.d))) {
         if (float64_is_signaling_nan(farg.d)) {
             /* Signaling NaN: flags are undefined */
-            ret = 0x00;
+            fprf = 0x00;
         } else {
             /* Quiet NaN */
-            ret = 0x11;
+            fprf = 0x11;
         }
     } else if (unlikely(float64_is_infinity(farg.d))) {
         /* +/- infinity */
         if (isneg) {
-            ret = 0x09;
+            fprf = 0x09;
         } else {
-            ret = 0x05;
+            fprf = 0x05;
         }
     } else {
         if (float64_is_zero(farg.d)) {
             /* +/- zero */
             if (isneg) {
-                ret = 0x12;
+                fprf = 0x12;
             } else {
-                ret = 0x02;
+                fprf = 0x02;
             }
         } else {
             if (isden(farg.d)) {
                 /* Denormalized numbers */
-                ret = 0x10;
+                fprf = 0x10;
             } else {
                 /* Normalized numbers */
-                ret = 0x00;
+                fprf = 0x00;
             }
             if (isneg) {
-                ret |= 0x08;
+                fprf |= 0x08;
             } else {
-                ret |= 0x04;
+                fprf |= 0x04;
             }
         }
     }
-    if (set_fprf) {
-        /* We update FPSCR_FPRF */
-        env->fpscr &= ~(0x1F << FPSCR_FPRF);
-        env->fpscr |= ret << FPSCR_FPRF;
-    }
-    /* We just need fpcc to update Rc1 */
-    return ret & 0xF;
+    /* We update FPSCR_FPRF */
+    env->fpscr &= ~(0x1F << FPSCR_FPRF);
+    env->fpscr |= fprf << FPSCR_FPRF;
 }
 
 /* Floating-point invalid operations exception */
 static inline uint64_t fload_invalid_op_excp(CPUPPCState *env, int op,
                                              int set_fpcc)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
     uint64_t ret = 0;
     int ve;
 
@@ -145,7 +155,7 @@ static inline uint64_t fload_invalid_op_excp(CPUPPCState *env, int op,
         }
         /* We must update the target FPR before raising the exception */
         if (ve != 0) {
-            env->exception_index = POWERPC_EXCP_PROGRAM;
+            cs->exception_index = POWERPC_EXCP_PROGRAM;
             env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_VXVC;
             /* Update the floating-point enabled exception summary */
             env->fpscr |= 1 << FPSCR_FEX;
@@ -184,7 +194,7 @@ static inline uint64_t fload_invalid_op_excp(CPUPPCState *env, int op,
     /* Update the floating-point invalid operation summary */
     env->fpscr |= 1 << FPSCR_VX;
     /* Update the floating-point exception summary */
-    env->fpscr |= 1 << FPSCR_FX;
+    env->fpscr |= FP_FX;
     if (ve != 0) {
         /* Update the floating-point enabled exception summary */
         env->fpscr |= 1 << FPSCR_FEX;
@@ -201,7 +211,7 @@ static inline void float_zero_divide_excp(CPUPPCState *env)
     env->fpscr |= 1 << FPSCR_ZX;
     env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
     /* Update the floating-point exception summary */
-    env->fpscr |= 1 << FPSCR_FX;
+    env->fpscr |= FP_FX;
     if (fpscr_ze != 0) {
         /* Update the floating-point enabled exception summary */
         env->fpscr |= 1 << FPSCR_FEX;
@@ -214,15 +224,17 @@ static inline void float_zero_divide_excp(CPUPPCState *env)
 
 static inline void float_overflow_excp(CPUPPCState *env)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
+
     env->fpscr |= 1 << FPSCR_OX;
     /* Update the floating-point exception summary */
-    env->fpscr |= 1 << FPSCR_FX;
+    env->fpscr |= FP_FX;
     if (fpscr_oe != 0) {
         /* XXX: should adjust the result */
         /* Update the floating-point enabled exception summary */
         env->fpscr |= 1 << FPSCR_FEX;
         /* We must update the target FPR before raising the exception */
-        env->exception_index = POWERPC_EXCP_PROGRAM;
+        cs->exception_index = POWERPC_EXCP_PROGRAM;
         env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
     } else {
         env->fpscr |= 1 << FPSCR_XX;
@@ -232,29 +244,33 @@ static inline void float_overflow_excp(CPUPPCState *env)
 
 static inline void float_underflow_excp(CPUPPCState *env)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
+
     env->fpscr |= 1 << FPSCR_UX;
     /* Update the floating-point exception summary */
-    env->fpscr |= 1 << FPSCR_FX;
+    env->fpscr |= FP_FX;
     if (fpscr_ue != 0) {
         /* XXX: should adjust the result */
         /* Update the floating-point enabled exception summary */
         env->fpscr |= 1 << FPSCR_FEX;
         /* We must update the target FPR before raising the exception */
-        env->exception_index = POWERPC_EXCP_PROGRAM;
+        cs->exception_index = POWERPC_EXCP_PROGRAM;
         env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
     }
 }
 
 static inline void float_inexact_excp(CPUPPCState *env)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
+
     env->fpscr |= 1 << FPSCR_XX;
     /* Update the floating-point exception summary */
-    env->fpscr |= 1 << FPSCR_FX;
+    env->fpscr |= FP_FX;
     if (fpscr_xe != 0) {
         /* Update the floating-point enabled exception summary */
         env->fpscr |= 1 << FPSCR_FEX;
         /* We must update the target FPR before raising the exception */
-        env->exception_index = POWERPC_EXCP_PROGRAM;
+        cs->exception_index = POWERPC_EXCP_PROGRAM;
         env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
     }
 }
@@ -306,6 +322,7 @@ void helper_fpscr_clrbit(CPUPPCState *env, uint32_t bit)
 
 void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
     int prev;
 
     prev = (env->fpscr >> bit) & 1;
@@ -313,31 +330,31 @@ void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
     if (prev == 0) {
         switch (bit) {
         case FPSCR_VX:
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_ve) {
                 goto raise_ve;
             }
             break;
         case FPSCR_OX:
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_oe) {
                 goto raise_oe;
             }
             break;
         case FPSCR_UX:
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_ue) {
                 goto raise_ue;
             }
             break;
         case FPSCR_ZX:
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_ze) {
                 goto raise_ze;
             }
             break;
         case FPSCR_XX:
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_xe) {
                 goto raise_xe;
             }
@@ -352,7 +369,7 @@ void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
         case FPSCR_VXSQRT:
         case FPSCR_VXCVI:
             env->fpscr |= 1 << FPSCR_VX;
-            env->fpscr |= 1 << FPSCR_FX;
+            env->fpscr |= FP_FX;
             if (fpscr_ve != 0) {
                 goto raise_ve;
             }
@@ -429,7 +446,7 @@ void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
             /* Update the floating-point enabled exception summary */
             env->fpscr |= 1 << FPSCR_FEX;
             /* We have to update Rc1 before raising the exception */
-            env->exception_index = POWERPC_EXCP_PROGRAM;
+            cs->exception_index = POWERPC_EXCP_PROGRAM;
             break;
         }
     }
@@ -437,6 +454,7 @@ void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
 
 void helper_store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
     target_ulong prev, new;
     int i;
 
@@ -458,7 +476,7 @@ void helper_store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
     }
     if ((fpscr_ex & fpscr_eex) != 0) {
         env->fpscr |= 1 << FPSCR_FEX;
-        env->exception_index = POWERPC_EXCP_PROGRAM;
+        cs->exception_index = POWERPC_EXCP_PROGRAM;
         /* XXX: we should compute it properly */
         env->error_code = POWERPC_EXCP_FP;
     } else {
@@ -474,6 +492,7 @@ void store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
 
 void helper_float_check_status(CPUPPCState *env)
 {
+    CPUState *cs = CPU(ppc_env_get_cpu(env));
     int status = get_float_exception_flags(&env->fp_status);
 
     if (status & float_flag_divbyzero) {
@@ -486,11 +505,11 @@ void helper_float_check_status(CPUPPCState *env)
         float_inexact_excp(env);
     }
 
-    if (env->exception_index == POWERPC_EXCP_PROGRAM &&
+    if (cs->exception_index == POWERPC_EXCP_PROGRAM &&
         (env->error_code & POWERPC_EXCP_FP)) {
         /* Differred floating-point exception after target FPR update */
         if (msr_fe0 != 0 || msr_fe1 != 0) {
-            helper_raise_exception_err(env, env->exception_index,
+            helper_raise_exception_err(env, cs->exception_index,
                                        env->error_code);
         }
     }
@@ -600,109 +619,59 @@ uint64_t helper_fdiv(CPUPPCState *env, uint64_t arg1, uint64_t arg2)
     return farg1.ll;
 }
 
-/* fctiw - fctiw. */
-uint64_t helper_fctiw(CPUPPCState *env, uint64_t arg)
-{
-    CPU_DoubleU farg;
 
-    farg.ll = arg;
-
-    if (unlikely(float64_is_signaling_nan(farg.d))) {
-        /* sNaN conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN |
-                                        POWERPC_EXCP_FP_VXCVI, 1);
-    } else if (unlikely(float64_is_quiet_nan(farg.d) ||
-                        float64_is_infinity(farg.d))) {
-        /* qNan / infinity conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);
-    } else {
-        farg.ll = float64_to_int32(farg.d, &env->fp_status);
-        /* XXX: higher bits are not supposed to be significant.
-         *     to make tests easier, return the same as a real PowerPC 750
-         */
-        farg.ll |= 0xFFF80000ULL << 32;
-    }
-    return farg.ll;
-}
-
-/* fctiwz - fctiwz. */
-uint64_t helper_fctiwz(CPUPPCState *env, uint64_t arg)
-{
-    CPU_DoubleU farg;
-
-    farg.ll = arg;
-
-    if (unlikely(float64_is_signaling_nan(farg.d))) {
-        /* sNaN conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN |
-                                        POWERPC_EXCP_FP_VXCVI, 1);
-    } else if (unlikely(float64_is_quiet_nan(farg.d) ||
-                        float64_is_infinity(farg.d))) {
-        /* qNan / infinity conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);
-    } else {
-        farg.ll = float64_to_int32_round_to_zero(farg.d, &env->fp_status);
-        /* XXX: higher bits are not supposed to be significant.
-         *     to make tests easier, return the same as a real PowerPC 750
-         */
-        farg.ll |= 0xFFF80000ULL << 32;
-    }
-    return farg.ll;
-}
-
-#if defined(TARGET_PPC64)
-/* fcfid - fcfid. */
-uint64_t helper_fcfid(CPUPPCState *env, uint64_t arg)
-{
-    CPU_DoubleU farg;
-
-    farg.d = int64_to_float64(arg, &env->fp_status);
-    return farg.ll;
-}
-
-/* fctid - fctid. */
-uint64_t helper_fctid(CPUPPCState *env, uint64_t arg)
-{
-    CPU_DoubleU farg;
-
-    farg.ll = arg;
-
-    if (unlikely(float64_is_signaling_nan(farg.d))) {
-        /* sNaN conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN |
-                                        POWERPC_EXCP_FP_VXCVI, 1);
-    } else if (unlikely(float64_is_quiet_nan(farg.d) ||
-                        float64_is_infinity(farg.d))) {
-        /* qNan / infinity conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);
-    } else {
-        farg.ll = float64_to_int64(farg.d, &env->fp_status);
-    }
-    return farg.ll;
-}
-
-/* fctidz - fctidz. */
-uint64_t helper_fctidz(CPUPPCState *env, uint64_t arg)
-{
-    CPU_DoubleU farg;
-
-    farg.ll = arg;
-
-    if (unlikely(float64_is_signaling_nan(farg.d))) {
-        /* sNaN conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN |
-                                        POWERPC_EXCP_FP_VXCVI, 1);
-    } else if (unlikely(float64_is_quiet_nan(farg.d) ||
-                        float64_is_infinity(farg.d))) {
-        /* qNan / infinity conversion */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);
-    } else {
-        farg.ll = float64_to_int64_round_to_zero(farg.d, &env->fp_status);
-    }
-    return farg.ll;
-}
+#define FPU_FCTI(op, cvt, nanval)                                      \
+uint64_t helper_##op(CPUPPCState *env, uint64_t arg)                   \
+{                                                                      \
+    CPU_DoubleU farg;                                                  \
+                                                                       \
+    farg.ll = arg;                                                     \
+    farg.ll = float64_to_##cvt(farg.d, &env->fp_status);               \
+                                                                       \
+    if (unlikely(env->fp_status.float_exception_flags)) {              \
+        if (float64_is_any_nan(arg)) {                                 \
+            fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);      \
+            if (float64_is_signaling_nan(arg)) {                       \
+                fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 1); \
+            }                                                          \
+            farg.ll = nanval;                                          \
+        } else if (env->fp_status.float_exception_flags &              \
+                   float_flag_invalid) {                               \
+            fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);      \
+        }                                                              \
+        helper_float_check_status(env);                                \
+    }                                                                  \
+    return farg.ll;                                                    \
+ }
 
-#endif
+FPU_FCTI(fctiw, int32, 0x80000000U)
+FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U)
+FPU_FCTI(fctiwu, uint32, 0x00000000U)
+FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U)
+FPU_FCTI(fctid, int64, 0x8000000000000000ULL)
+FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL)
+FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL)
+FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL)
+
+#define FPU_FCFI(op, cvtr, is_single)                      \
+uint64_t helper_##op(CPUPPCState *env, uint64_t arg)       \
+{                                                          \
+    CPU_DoubleU farg;                                      \
+                                                           \
+    if (is_single) {                                       \
+        float32 tmp = cvtr(arg, &env->fp_status);          \
+        farg.d = float32_to_float64(tmp, &env->fp_status); \
+    } else {                                               \
+        farg.d = cvtr(arg, &env->fp_status);               \
+    }                                                      \
+    helper_float_check_status(env);                        \
+    return farg.ll;                                        \
+}
+
+FPU_FCFI(fcfid, int64_to_float64, 0)
+FPU_FCFI(fcfids, int64_to_float32, 1)
+FPU_FCFI(fcfidu, uint64_to_float64, 0)
+FPU_FCFI(fcfidus, uint64_to_float32, 1)
 
 static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg,
                               int rounding_mode)
@@ -713,24 +682,28 @@ static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg,
 
     if (unlikely(float64_is_signaling_nan(farg.d))) {
         /* sNaN round */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN |
-                                        POWERPC_EXCP_FP_VXCVI, 1);
-    } else if (unlikely(float64_is_quiet_nan(farg.d) ||
-                        float64_is_infinity(farg.d))) {
-        /* qNan / infinity round */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 1);
+        fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 1);
+        farg.ll = arg | 0x0008000000000000ULL;
     } else {
+        int inexact = get_float_exception_flags(&env->fp_status) &
+                      float_flag_inexact;
         set_float_rounding_mode(rounding_mode, &env->fp_status);
         farg.ll = float64_round_to_int(farg.d, &env->fp_status);
         /* Restore rounding mode from FPSCR */
         fpscr_set_rounding_mode(env);
+
+        /* fri* does not set FPSCR[XX] */
+        if (!inexact) {
+            env->fp_status.float_exception_flags &= ~float_flag_inexact;
+        }
     }
+    helper_float_check_status(env);
     return farg.ll;
 }
 
 uint64_t helper_frin(CPUPPCState *env, uint64_t arg)
 {
-    return do_fri(env, arg, float_round_nearest_even);
+    return do_fri(env, arg, float_round_ties_away);
 }
 
 uint64_t helper_friz(CPUPPCState *env, uint64_t arg)
@@ -946,14 +919,16 @@ uint64_t helper_fsqrt(CPUPPCState *env, uint64_t arg)
 
     farg.ll = arg;
 
-    if (unlikely(float64_is_neg(farg.d) && !float64_is_zero(farg.d))) {
-        /* Square root of a negative nonzero number */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
-    } else {
+    if (unlikely(float64_is_any_nan(farg.d))) {
         if (unlikely(float64_is_signaling_nan(farg.d))) {
-            /* sNaN square root */
+            /* sNaN reciprocal square root */
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 1);
+            farg.ll = float64_snan_to_qnan(farg.ll);
         }
+    } else if (unlikely(float64_is_neg(farg.d) && !float64_is_zero(farg.d))) {
+        /* Square root of a negative nonzero number */
+        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
+    } else {
         farg.d = float64_sqrt(farg.d, &env->fp_status);
     }
     return farg.ll;
@@ -997,23 +972,23 @@ uint64_t helper_fres(CPUPPCState *env, uint64_t arg)
 uint64_t helper_frsqrte(CPUPPCState *env, uint64_t arg)
 {
     CPU_DoubleU farg;
-    float32 f32;
 
     farg.ll = arg;
 
-    if (unlikely(float64_is_neg(farg.d) && !float64_is_zero(farg.d))) {
-        /* Reciprocal square root of a negative nonzero number */
-        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
-    } else {
+    if (unlikely(float64_is_any_nan(farg.d))) {
         if (unlikely(float64_is_signaling_nan(farg.d))) {
             /* sNaN reciprocal square root */
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 1);
+            farg.ll = float64_snan_to_qnan(farg.ll);
         }
+    } else if (unlikely(float64_is_neg(farg.d) && !float64_is_zero(farg.d))) {
+        /* Reciprocal square root of a negative nonzero number */
+        farg.ll = fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
+    } else {
         farg.d = float64_sqrt(farg.d, &env->fp_status);
         farg.d = float64_div(float64_one, farg.d, &env->fp_status);
-        f32 = float64_to_float32(farg.d, &env->fp_status);
-        farg.d = float32_to_float64(f32, &env->fp_status);
     }
+
     return farg.ll;
 }
 
@@ -1033,6 +1008,73 @@ uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
     }
 }
 
+uint32_t helper_ftdiv(uint64_t fra, uint64_t frb)
+{
+    int fe_flag = 0;
+    int fg_flag = 0;
+
+    if (unlikely(float64_is_infinity(fra) ||
+                 float64_is_infinity(frb) ||
+                 float64_is_zero(frb))) {
+        fe_flag = 1;
+        fg_flag = 1;
+    } else {
+        int e_a = ppc_float64_get_unbiased_exp(fra);
+        int e_b = ppc_float64_get_unbiased_exp(frb);
+
+        if (unlikely(float64_is_any_nan(fra) ||
+                     float64_is_any_nan(frb))) {
+            fe_flag = 1;
+        } else if ((e_b <= -1022) || (e_b >= 1021)) {
+            fe_flag = 1;
+        } else if (!float64_is_zero(fra) &&
+                   (((e_a - e_b) >= 1023) ||
+                    ((e_a - e_b) <= -1021) ||
+                    (e_a <= -970))) {
+            fe_flag = 1;
+        }
+
+        if (unlikely(float64_is_zero_or_denormal(frb))) {
+            /* XB is not zero because of the above check and */
+            /* so must be denormalized.                      */
+            fg_flag = 1;
+        }
+    }
+
+    return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
+}
+
+uint32_t helper_ftsqrt(uint64_t frb)
+{
+    int fe_flag = 0;
+    int fg_flag = 0;
+
+    if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) {
+        fe_flag = 1;
+        fg_flag = 1;
+    } else {
+        int e_b = ppc_float64_get_unbiased_exp(frb);
+
+        if (unlikely(float64_is_any_nan(frb))) {
+            fe_flag = 1;
+        } else if (unlikely(float64_is_zero(frb))) {
+            fe_flag = 1;
+        } else if (unlikely(float64_is_neg(frb))) {
+            fe_flag = 1;
+        } else if (!float64_is_zero(frb) && (e_b <= (-1022+52))) {
+            fe_flag = 1;
+        }
+
+        if (unlikely(float64_is_zero_or_denormal(frb))) {
+            /* XB is not zero because of the above check and */
+            /* therefore must be denormalized.               */
+            fg_flag = 1;
+        }
+    }
+
+    return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
+}
+
 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
                   uint32_t crfD)
 {
@@ -1735,11 +1777,19 @@ typedef union _ppc_vsr_t {
     float64 f64[2];
 } ppc_vsr_t;
 
+#if defined(HOST_WORDS_BIGENDIAN)
+#define VsrW(i) u32[i]
+#define VsrD(i) u64[i]
+#else
+#define VsrW(i) u32[3-(i)]
+#define VsrD(i) u64[1-(i)]
+#endif
+
 static void getVSR(int n, ppc_vsr_t *vsr, CPUPPCState *env)
 {
     if (n < 32) {
-        vsr->f64[0] = env->fpr[n];
-        vsr->u64[1] = env->vsr[n];
+        vsr->VsrD(0) = env->fpr[n];
+        vsr->VsrD(1) = env->vsr[n];
     } else {
         vsr->u64[0] = env->avr[n-32].u64[0];
         vsr->u64[1] = env->avr[n-32].u64[1];
@@ -1749,8 +1799,8 @@ static void getVSR(int n, ppc_vsr_t *vsr, CPUPPCState *env)
 static void putVSR(int n, ppc_vsr_t *vsr, CPUPPCState *env)
 {
     if (n < 32) {
-        env->fpr[n] = vsr->f64[0];
-        env->vsr[n] = vsr->u64[1];
+        env->fpr[n] = vsr->VsrD(0);
+        env->vsr[n] = vsr->VsrD(1);
     } else {
         env->avr[n-32].u64[0] = vsr->u64[0];
         env->avr[n-32].u64[1] = vsr->u64[1];
@@ -1765,7 +1815,7 @@ static void putVSR(int n, ppc_vsr_t *vsr, CPUPPCState *env)
  *   op    - operation (add or sub)
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp)                    \
@@ -1782,44 +1832,44 @@ void helper_##name(CPUPPCState *env, uint32_t opcode)                        \
     for (i = 0; i < nels; i++) {                                             \
         float_status tstat = env->fp_status;                                 \
         set_float_exception_flags(0, &tstat);                                \
-        xt.fld[i] = tp##_##op(xa.fld[i], xb.fld[i], &tstat);                 \
+        xt.fld = tp##_##op(xa.fld, xb.fld, &tstat);                          \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
                                                                              \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {    \
-            if (tp##_is_infinity(xa.fld[i]) && tp##_is_infinity(xb.fld[i])) {\
+            if (tp##_is_infinity(xa.fld) && tp##_is_infinity(xb.fld)) {      \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXISI, sfprf);    \
-            } else if (tp##_is_signaling_nan(xa.fld[i]) ||                   \
-                       tp##_is_signaling_nan(xb.fld[i])) {                   \
+            } else if (tp##_is_signaling_nan(xa.fld) ||                      \
+                       tp##_is_signaling_nan(xb.fld)) {                      \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);   \
             }                                                                \
         }                                                                    \
                                                                              \
         if (r2sp) {                                                          \
-            xt.fld[i] = helper_frsp(env, xt.fld[i]);                         \
+            xt.fld = helper_frsp(env, xt.fld);                               \
         }                                                                    \
                                                                              \
         if (sfprf) {                                                         \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                      \
+            helper_compute_fprf(env, xt.fld);                                \
         }                                                                    \
     }                                                                        \
     putVSR(xT(opcode), &xt, env);                                            \
     helper_float_check_status(env);                                          \
 }
 
-VSX_ADD_SUB(xsadddp, add, 1, float64, f64, 1, 0)
-VSX_ADD_SUB(xsaddsp, add, 1, float64, f64, 1, 1)
-VSX_ADD_SUB(xvadddp, add, 2, float64, f64, 0, 0)
-VSX_ADD_SUB(xvaddsp, add, 4, float32, f32, 0, 0)
-VSX_ADD_SUB(xssubdp, sub, 1, float64, f64, 1, 0)
-VSX_ADD_SUB(xssubsp, sub, 1, float64, f64, 1, 1)
-VSX_ADD_SUB(xvsubdp, sub, 2, float64, f64, 0, 0)
-VSX_ADD_SUB(xvsubsp, sub, 4, float32, f32, 0, 0)
+VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0)
+VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1)
+VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0)
+VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0)
+VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0)
+VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1)
+VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0)
+VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_MUL - VSX floating point multiply
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp)                              \
@@ -1836,25 +1886,25 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     for (i = 0; i < nels; i++) {                                             \
         float_status tstat = env->fp_status;                                 \
         set_float_exception_flags(0, &tstat);                                \
-        xt.fld[i] = tp##_mul(xa.fld[i], xb.fld[i], &tstat);                  \
+        xt.fld = tp##_mul(xa.fld, xb.fld, &tstat);                           \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
                                                                              \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {    \
-            if ((tp##_is_infinity(xa.fld[i]) && tp##_is_zero(xb.fld[i])) ||  \
-                (tp##_is_infinity(xb.fld[i]) && tp##_is_zero(xa.fld[i]))) {  \
+            if ((tp##_is_infinity(xa.fld) && tp##_is_zero(xb.fld)) ||        \
+                (tp##_is_infinity(xb.fld) && tp##_is_zero(xa.fld))) {        \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXIMZ, sfprf);    \
-            } else if (tp##_is_signaling_nan(xa.fld[i]) ||                   \
-                       tp##_is_signaling_nan(xb.fld[i])) {                   \
+            } else if (tp##_is_signaling_nan(xa.fld) ||                      \
+                       tp##_is_signaling_nan(xb.fld)) {                      \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);   \
             }                                                                \
         }                                                                    \
                                                                              \
         if (r2sp) {                                                          \
-            xt.fld[i] = helper_frsp(env, xt.fld[i]);                         \
+            xt.fld = helper_frsp(env, xt.fld);                               \
         }                                                                    \
                                                                              \
         if (sfprf) {                                                         \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                      \
+            helper_compute_fprf(env, xt.fld);                                \
         }                                                                    \
     }                                                                        \
                                                                              \
@@ -1862,16 +1912,16 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     helper_float_check_status(env);                                          \
 }
 
-VSX_MUL(xsmuldp, 1, float64, f64, 1, 0)
-VSX_MUL(xsmulsp, 1, float64, f64, 1, 1)
-VSX_MUL(xvmuldp, 2, float64, f64, 0, 0)
-VSX_MUL(xvmulsp, 4, float32, f32, 0, 0)
+VSX_MUL(xsmuldp, 1, float64, VsrD(0), 1, 0)
+VSX_MUL(xsmulsp, 1, float64, VsrD(0), 1, 1)
+VSX_MUL(xvmuldp, 2, float64, VsrD(i), 0, 0)
+VSX_MUL(xvmulsp, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_DIV - VSX floating point divide
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
 #define VSX_DIV(op, nels, tp, fld, sfprf, r2sp)                               \
@@ -1888,27 +1938,27 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
     for (i = 0; i < nels; i++) {                                              \
         float_status tstat = env->fp_status;                                  \
         set_float_exception_flags(0, &tstat);                                 \
-        xt.fld[i] = tp##_div(xa.fld[i], xb.fld[i], &tstat);                   \
+        xt.fld = tp##_div(xa.fld, xb.fld, &tstat);                            \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags;  \
                                                                               \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {     \
-            if (tp##_is_infinity(xa.fld[i]) && tp##_is_infinity(xb.fld[i])) { \
+            if (tp##_is_infinity(xa.fld) && tp##_is_infinity(xb.fld)) {       \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXIDI, sfprf);     \
-            } else if (tp##_is_zero(xa.fld[i]) &&                             \
-                tp##_is_zero(xb.fld[i])) {                                    \
+            } else if (tp##_is_zero(xa.fld) &&                                \
+                tp##_is_zero(xb.fld)) {                                       \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXZDZ, sfprf);     \
-            } else if (tp##_is_signaling_nan(xa.fld[i]) ||                    \
-                tp##_is_signaling_nan(xb.fld[i])) {                           \
+            } else if (tp##_is_signaling_nan(xa.fld) ||                       \
+                tp##_is_signaling_nan(xb.fld)) {                              \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);    \
             }                                                                 \
         }                                                                     \
                                                                               \
         if (r2sp) {                                                           \
-            xt.fld[i] = helper_frsp(env, xt.fld[i]);                          \
+            xt.fld = helper_frsp(env, xt.fld);                                \
         }                                                                     \
                                                                               \
         if (sfprf) {                                                          \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                       \
+            helper_compute_fprf(env, xt.fld);                                 \
         }                                                                     \
     }                                                                         \
                                                                               \
@@ -1916,19 +1966,19 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
     helper_float_check_status(env);                                           \
 }
 
-VSX_DIV(xsdivdp, 1, float64, f64, 1, 0)
-VSX_DIV(xsdivsp, 1, float64, f64, 1, 1)
-VSX_DIV(xvdivdp, 2, float64, f64, 0, 0)
-VSX_DIV(xvdivsp, 4, float32, f32, 0, 0)
+VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0)
+VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1)
+VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0)
+VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_RE  - VSX floating point reciprocal estimate
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
-#define VSX_RE(op, nels, tp, fld, sfprf)                                      \
+#define VSX_RE(op, nels, tp, fld, sfprf, r2sp)                                \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
 {                                                                             \
     ppc_vsr_t xt, xb;                                                         \
@@ -1939,12 +1989,17 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
     helper_reset_fpstatus(env);                                               \
                                                                               \
     for (i = 0; i < nels; i++) {                                              \
-        if (unlikely(tp##_is_signaling_nan(xb.fld[i]))) {                     \
+        if (unlikely(tp##_is_signaling_nan(xb.fld))) {                        \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);    \
         }                                                                     \
-        xt.fld[i] = tp##_div(tp##_one, xb.fld[i], &env->fp_status);           \
+        xt.fld = tp##_div(tp##_one, xb.fld, &env->fp_status);                 \
+                                                                              \
+        if (r2sp) {                                                           \
+            xt.fld = helper_frsp(env, xt.fld);                                \
+        }                                                                     \
+                                                                              \
         if (sfprf) {                                                          \
-            helper_compute_fprf(env, xt.fld[0], sfprf);                       \
+            helper_compute_fprf(env, xt.fld);                                 \
         }                                                                     \
     }                                                                         \
                                                                               \
@@ -1952,18 +2007,19 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
     helper_float_check_status(env);                                           \
 }
 
-VSX_RE(xsredp, 1, float64, f64, 1)
-VSX_RE(xvredp, 2, float64, f64, 0)
-VSX_RE(xvresp, 4, float32, f32, 0)
+VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0)
+VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1)
+VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0)
+VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_SQRT - VSX floating point square root
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
-#define VSX_SQRT(op, nels, tp, fld, sfprf)                                   \
+#define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp)                             \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
 {                                                                            \
     ppc_vsr_t xt, xb;                                                        \
@@ -1976,19 +2032,23 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     for (i = 0; i < nels; i++) {                                             \
         float_status tstat = env->fp_status;                                 \
         set_float_exception_flags(0, &tstat);                                \
-        xt.fld[i] = tp##_sqrt(xb.fld[i], &tstat);                            \
+        xt.fld = tp##_sqrt(xb.fld, &tstat);                                  \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
                                                                              \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {    \
-            if (tp##_is_neg(xb.fld[i]) && !tp##_is_zero(xb.fld[i])) {        \
+            if (tp##_is_neg(xb.fld) && !tp##_is_zero(xb.fld)) {              \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, sfprf);   \
-            } else if (tp##_is_signaling_nan(xb.fld[i])) {                   \
+            } else if (tp##_is_signaling_nan(xb.fld)) {                      \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);   \
             }                                                                \
         }                                                                    \
                                                                              \
+        if (r2sp) {                                                          \
+            xt.fld = helper_frsp(env, xt.fld);                               \
+        }                                                                    \
+                                                                             \
         if (sfprf) {                                                         \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                      \
+            helper_compute_fprf(env, xt.fld);                                \
         }                                                                    \
     }                                                                        \
                                                                              \
@@ -1996,18 +2056,19 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     helper_float_check_status(env);                                          \
 }
 
-VSX_SQRT(xssqrtdp, 1, float64, f64, 1)
-VSX_SQRT(xvsqrtdp, 2, float64, f64, 0)
-VSX_SQRT(xvsqrtsp, 4, float32, f32, 0)
+VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0)
+VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1)
+VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0)
+VSX_SQRT(xvsqrtsp, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_RSQRTE - VSX floating point reciprocal square root estimate
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   sfprf - set FPRF
  */
-#define VSX_RSQRTE(op, nels, tp, fld, sfprf)                                 \
+#define VSX_RSQRTE(op, nels, tp, fld, sfprf, r2sp)                           \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
 {                                                                            \
     ppc_vsr_t xt, xb;                                                        \
@@ -2020,20 +2081,24 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     for (i = 0; i < nels; i++) {                                             \
         float_status tstat = env->fp_status;                                 \
         set_float_exception_flags(0, &tstat);                                \
-        xt.fld[i] = tp##_sqrt(xb.fld[i], &tstat);                            \
-        xt.fld[i] = tp##_div(tp##_one, xt.fld[i], &tstat);                   \
+        xt.fld = tp##_sqrt(xb.fld, &tstat);                                  \
+        xt.fld = tp##_div(tp##_one, xt.fld, &tstat);                         \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
                                                                              \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {    \
-            if (tp##_is_neg(xb.fld[i]) && !tp##_is_zero(xb.fld[i])) {        \
+            if (tp##_is_neg(xb.fld) && !tp##_is_zero(xb.fld)) {              \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, sfprf);   \
-            } else if (tp##_is_signaling_nan(xb.fld[i])) {                   \
+            } else if (tp##_is_signaling_nan(xb.fld)) {                      \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);   \
             }                                                                \
         }                                                                    \
                                                                              \
+        if (r2sp) {                                                          \
+            xt.fld = helper_frsp(env, xt.fld);                               \
+        }                                                                    \
+                                                                             \
         if (sfprf) {                                                         \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                      \
+            helper_compute_fprf(env, xt.fld);                                \
         }                                                                    \
     }                                                                        \
                                                                              \
@@ -2041,25 +2106,16 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     helper_float_check_status(env);                                          \
 }
 
-VSX_RSQRTE(xsrsqrtedp, 1, float64, f64, 1)
-VSX_RSQRTE(xvrsqrtedp, 2, float64, f64, 0)
-VSX_RSQRTE(xvrsqrtesp, 4, float32, f32, 0)
-
-static inline int ppc_float32_get_unbiased_exp(float32 f)
-{
-    return ((f >> 23) & 0xFF) - 127;
-}
-
-static inline int ppc_float64_get_unbiased_exp(float64 f)
-{
-    return ((f >> 52) & 0x7FF) - 1023;
-}
+VSX_RSQRTE(xsrsqrtedp, 1, float64, VsrD(0), 1, 0)
+VSX_RSQRTE(xsrsqrtesp, 1, float64, VsrD(0), 1, 1)
+VSX_RSQRTE(xvrsqrtedp, 2, float64, VsrD(i), 0, 0)
+VSX_RSQRTE(xvrsqrtesp, 4, float32, VsrW(i), 0, 0)
 
 /* VSX_TDIV - VSX floating point test for divide
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   emin  - minimum unbiased exponent
  *   emax  - maximum unbiased exponent
  *   nbits - number of fraction bits
@@ -2076,28 +2132,28 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     getVSR(xB(opcode), &xb, env);                                       \
                                                                         \
     for (i = 0; i < nels; i++) {                                        \
-        if (unlikely(tp##_is_infinity(xa.fld[i]) ||                     \
-                     tp##_is_infinity(xb.fld[i]) ||                     \
-                     tp##_is_zero(xb.fld[i]))) {                        \
+        if (unlikely(tp##_is_infinity(xa.fld) ||                        \
+                     tp##_is_infinity(xb.fld) ||                        \
+                     tp##_is_zero(xb.fld))) {                           \
             fe_flag = 1;                                                \
             fg_flag = 1;                                                \
         } else {                                                        \
-            int e_a = ppc_##tp##_get_unbiased_exp(xa.fld[i]);           \
-            int e_b = ppc_##tp##_get_unbiased_exp(xb.fld[i]);           \
+            int e_a = ppc_##tp##_get_unbiased_exp(xa.fld);              \
+            int e_b = ppc_##tp##_get_unbiased_exp(xb.fld);              \
                                                                         \
-            if (unlikely(tp##_is_any_nan(xa.fld[i]) ||                  \
-                         tp##_is_any_nan(xb.fld[i]))) {                 \
+            if (unlikely(tp##_is_any_nan(xa.fld) ||                     \
+                         tp##_is_any_nan(xb.fld))) {                    \
                 fe_flag = 1;                                            \
             } else if ((e_b <= emin) || (e_b >= (emax-2))) {            \
                 fe_flag = 1;                                            \
-            } else if (!tp##_is_zero(xa.fld[i]) &&                      \
+            } else if (!tp##_is_zero(xa.fld) &&                         \
                        (((e_a - e_b) >= emax) ||                        \
                         ((e_a - e_b) <= (emin+1)) ||                    \
                          (e_a <= (emin+nbits)))) {                      \
                 fe_flag = 1;                                            \
             }                                                           \
                                                                         \
-            if (unlikely(tp##_is_zero_or_denormal(xb.fld[i]))) {        \
+            if (unlikely(tp##_is_zero_or_denormal(xb.fld))) {           \
                 /* XB is not zero because of the above check and */     \
                 /* so must be denormalized.                      */     \
                 fg_flag = 1;                                            \
@@ -2108,15 +2164,15 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
 }
 
-VSX_TDIV(xstdivdp, 1, float64, f64, -1022, 1023, 52)
-VSX_TDIV(xvtdivdp, 2, float64, f64, -1022, 1023, 52)
-VSX_TDIV(xvtdivsp, 4, float32, f32, -126, 127, 23)
+VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52)
+VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52)
+VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23)
 
 /* VSX_TSQRT - VSX floating point test for square root
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   emin  - minimum unbiased exponent
  *   emax  - maximum unbiased exponent
  *   nbits - number of fraction bits
@@ -2133,25 +2189,25 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     getVSR(xB(opcode), &xb, env);                                       \
                                                                         \
     for (i = 0; i < nels; i++) {                                        \
-        if (unlikely(tp##_is_infinity(xb.fld[i]) ||                     \
-                     tp##_is_zero(xb.fld[i]))) {                        \
+        if (unlikely(tp##_is_infinity(xb.fld) ||                        \
+                     tp##_is_zero(xb.fld))) {                           \
             fe_flag = 1;                                                \
             fg_flag = 1;                                                \
         } else {                                                        \
-            int e_b = ppc_##tp##_get_unbiased_exp(xb.fld[i]);           \
+            int e_b = ppc_##tp##_get_unbiased_exp(xb.fld);              \
                                                                         \
-            if (unlikely(tp##_is_any_nan(xb.fld[i]))) {                 \
+            if (unlikely(tp##_is_any_nan(xb.fld))) {                    \
                 fe_flag = 1;                                            \
-            } else if (unlikely(tp##_is_zero(xb.fld[i]))) {             \
+            } else if (unlikely(tp##_is_zero(xb.fld))) {                \
                 fe_flag = 1;                                            \
-            } else if (unlikely(tp##_is_neg(xb.fld[i]))) {              \
+            } else if (unlikely(tp##_is_neg(xb.fld))) {                 \
                 fe_flag = 1;                                            \
-            } else if (!tp##_is_zero(xb.fld[i]) &&                      \
+            } else if (!tp##_is_zero(xb.fld) &&                         \
                       (e_b <= (emin+nbits))) {                          \
                 fe_flag = 1;                                            \
             }                                                           \
                                                                         \
-            if (unlikely(tp##_is_zero_or_denormal(xb.fld[i]))) {        \
+            if (unlikely(tp##_is_zero_or_denormal(xb.fld))) {           \
                 /* XB is not zero because of the above check and */     \
                 /* therefore must be denormalized.               */     \
                 fg_flag = 1;                                            \
@@ -2162,21 +2218,21 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
 }
 
-VSX_TSQRT(xstsqrtdp, 1, float64, f64, -1022, 52)
-VSX_TSQRT(xvtsqrtdp, 2, float64, f64, -1022, 52)
-VSX_TSQRT(xvtsqrtsp, 4, float32, f32, -126, 23)
+VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52)
+VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52)
+VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23)
 
 /* VSX_MADD - VSX floating point muliply/add variations
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   maddflgs - flags for the float*muladd routine that control the
  *           various forms (madd, msub, nmadd, nmsub)
  *   afrm  - A form (1=A, 0=M)
  *   sfprf - set FPRF
  */
-#define VSX_MADD(op, nels, tp, fld, maddflgs, afrm, sfprf)                    \
+#define VSX_MADD(op, nels, tp, fld, maddflgs, afrm, sfprf, r2sp)              \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
 {                                                                             \
     ppc_vsr_t xt_in, xa, xb, xt_out;                                          \
@@ -2202,32 +2258,47 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
     for (i = 0; i < nels; i++) {                                              \
         float_status tstat = env->fp_status;                                  \
         set_float_exception_flags(0, &tstat);                                 \
-        xt_out.fld[i] = tp##_muladd(xa.fld[i], b->fld[i], c->fld[i],          \
-                                     maddflgs, &tstat);                       \
+        if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\
+            /* Avoid double rounding errors by rounding the intermediate */   \
+            /* result to odd.                                            */   \
+            set_float_rounding_mode(float_round_to_zero, &tstat);             \
+            xt_out.fld = tp##_muladd(xa.fld, b->fld, c->fld,                  \
+                                       maddflgs, &tstat);                     \
+            xt_out.fld |= (get_float_exception_flags(&tstat) &                \
+                              float_flag_inexact) != 0;                       \
+        } else {                                                              \
+            xt_out.fld = tp##_muladd(xa.fld, b->fld, c->fld,                  \
+                                        maddflgs, &tstat);                    \
+        }                                                                     \
         env->fp_status.float_exception_flags |= tstat.float_exception_flags;  \
                                                                               \
         if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {     \
-            if (tp##_is_signaling_nan(xa.fld[i]) ||                           \
-                tp##_is_signaling_nan(b->fld[i]) ||                           \
-                tp##_is_signaling_nan(c->fld[i])) {                           \
+            if (tp##_is_signaling_nan(xa.fld) ||                              \
+                tp##_is_signaling_nan(b->fld) ||                              \
+                tp##_is_signaling_nan(c->fld)) {                              \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, sfprf);    \
                 tstat.float_exception_flags &= ~float_flag_invalid;           \
             }                                                                 \
-            if ((tp##_is_infinity(xa.fld[i]) && tp##_is_zero(b->fld[i])) ||   \
-                (tp##_is_zero(xa.fld[i]) && tp##_is_infinity(b->fld[i]))) {   \
-                xt_out.fld[i] = float64_to_##tp(fload_invalid_op_excp(env,    \
+            if ((tp##_is_infinity(xa.fld) && tp##_is_zero(b->fld)) ||         \
+                (tp##_is_zero(xa.fld) && tp##_is_infinity(b->fld))) {         \
+                xt_out.fld = float64_to_##tp(fload_invalid_op_excp(env,       \
                     POWERPC_EXCP_FP_VXIMZ, sfprf), &env->fp_status);          \
                 tstat.float_exception_flags &= ~float_flag_invalid;           \
             }                                                                 \
             if ((tstat.float_exception_flags & float_flag_invalid) &&         \
-                ((tp##_is_infinity(xa.fld[i]) ||                              \
-                  tp##_is_infinity(b->fld[i])) &&                             \
-                  tp##_is_infinity(c->fld[i]))) {                             \
+                ((tp##_is_infinity(xa.fld) ||                                 \
+                  tp##_is_infinity(b->fld)) &&                                \
+                  tp##_is_infinity(c->fld))) {                                \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXISI, sfprf);     \
             }                                                                 \
         }                                                                     \
+                                                                              \
+        if (r2sp) {                                                           \
+            xt_out.fld = helper_frsp(env, xt_out.fld);                        \
+        }                                                                     \
+                                                                              \
         if (sfprf) {                                                          \
-            helper_compute_fprf(env, xt_out.fld[i], sfprf);                   \
+            helper_compute_fprf(env, xt_out.fld);                             \
         }                                                                     \
     }                                                                         \
     putVSR(xT(opcode), &xt_out, env);                                         \
@@ -2239,32 +2310,41 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                           \
 #define NMADD_FLGS float_muladd_negate_result
 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result)
 
-VSX_MADD(xsmaddadp, 1, float64, f64, MADD_FLGS, 1, 1)
-VSX_MADD(xsmaddmdp, 1, float64, f64, MADD_FLGS, 0, 1)
-VSX_MADD(xsmsubadp, 1, float64, f64, MSUB_FLGS, 1, 1)
-VSX_MADD(xsmsubmdp, 1, float64, f64, MSUB_FLGS, 0, 1)
-VSX_MADD(xsnmaddadp, 1, float64, f64, NMADD_FLGS, 1, 1)
-VSX_MADD(xsnmaddmdp, 1, float64, f64, NMADD_FLGS, 0, 1)
-VSX_MADD(xsnmsubadp, 1, float64, f64, NMSUB_FLGS, 1, 1)
-VSX_MADD(xsnmsubmdp, 1, float64, f64, NMSUB_FLGS, 0, 1)
-
-VSX_MADD(xvmaddadp, 2, float64, f64, MADD_FLGS, 1, 0)
-VSX_MADD(xvmaddmdp, 2, float64, f64, MADD_FLGS, 0, 0)
-VSX_MADD(xvmsubadp, 2, float64, f64, MSUB_FLGS, 1, 0)
-VSX_MADD(xvmsubmdp, 2, float64, f64, MSUB_FLGS, 0, 0)
-VSX_MADD(xvnmaddadp, 2, float64, f64, NMADD_FLGS, 1, 0)
-VSX_MADD(xvnmaddmdp, 2, float64, f64, NMADD_FLGS, 0, 0)
-VSX_MADD(xvnmsubadp, 2, float64, f64, NMSUB_FLGS, 1, 0)
-VSX_MADD(xvnmsubmdp, 2, float64, f64, NMSUB_FLGS, 0, 0)
-
-VSX_MADD(xvmaddasp, 4, float32, f32, MADD_FLGS, 1, 0)
-VSX_MADD(xvmaddmsp, 4, float32, f32, MADD_FLGS, 0, 0)
-VSX_MADD(xvmsubasp, 4, float32, f32, MSUB_FLGS, 1, 0)
-VSX_MADD(xvmsubmsp, 4, float32, f32, MSUB_FLGS, 0, 0)
-VSX_MADD(xvnmaddasp, 4, float32, f32, NMADD_FLGS, 1, 0)
-VSX_MADD(xvnmaddmsp, 4, float32, f32, NMADD_FLGS, 0, 0)
-VSX_MADD(xvnmsubasp, 4, float32, f32, NMSUB_FLGS, 1, 0)
-VSX_MADD(xvnmsubmsp, 4, float32, f32, NMSUB_FLGS, 0, 0)
+VSX_MADD(xsmaddadp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 0)
+VSX_MADD(xsmaddmdp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 0)
+VSX_MADD(xsmsubadp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 0)
+VSX_MADD(xsmsubmdp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 0)
+VSX_MADD(xsnmaddadp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 0)
+VSX_MADD(xsnmaddmdp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 0)
+VSX_MADD(xsnmsubadp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 0)
+VSX_MADD(xsnmsubmdp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 0)
+
+VSX_MADD(xsmaddasp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 1)
+VSX_MADD(xsmaddmsp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 1)
+VSX_MADD(xsmsubasp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 1)
+VSX_MADD(xsmsubmsp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 1)
+VSX_MADD(xsnmaddasp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 1)
+VSX_MADD(xsnmaddmsp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 1)
+VSX_MADD(xsnmsubasp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 1)
+VSX_MADD(xsnmsubmsp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 1)
+
+VSX_MADD(xvmaddadp, 2, float64, VsrD(i), MADD_FLGS, 1, 0, 0)
+VSX_MADD(xvmaddmdp, 2, float64, VsrD(i), MADD_FLGS, 0, 0, 0)
+VSX_MADD(xvmsubadp, 2, float64, VsrD(i), MSUB_FLGS, 1, 0, 0)
+VSX_MADD(xvmsubmdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0, 0)
+VSX_MADD(xvnmaddadp, 2, float64, VsrD(i), NMADD_FLGS, 1, 0, 0)
+VSX_MADD(xvnmaddmdp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0, 0)
+VSX_MADD(xvnmsubadp, 2, float64, VsrD(i), NMSUB_FLGS, 1, 0, 0)
+VSX_MADD(xvnmsubmdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0, 0)
+
+VSX_MADD(xvmaddasp, 4, float32, VsrW(i), MADD_FLGS, 1, 0, 0)
+VSX_MADD(xvmaddmsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0, 0)
+VSX_MADD(xvmsubasp, 4, float32, VsrW(i), MSUB_FLGS, 1, 0, 0)
+VSX_MADD(xvmsubmsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0, 0)
+VSX_MADD(xvnmaddasp, 4, float32, VsrW(i), NMADD_FLGS, 1, 0, 0)
+VSX_MADD(xvnmaddmsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0, 0)
+VSX_MADD(xvnmsubasp, 4, float32, VsrW(i), NMSUB_FLGS, 1, 0, 0)
+VSX_MADD(xvnmsubmsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0, 0)
 
 #define VSX_SCALAR_CMP(op, ordered)                                      \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                      \
@@ -2275,10 +2355,10 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                      \
     getVSR(xA(opcode), &xa, env);                                        \
     getVSR(xB(opcode), &xb, env);                                        \
                                                                          \
-    if (unlikely(float64_is_any_nan(xa.f64[0]) ||                        \
-                 float64_is_any_nan(xb.f64[0]))) {                       \
-        if (float64_is_signaling_nan(xa.f64[0]) ||                       \
-            float64_is_signaling_nan(xb.f64[0])) {                       \
+    if (unlikely(float64_is_any_nan(xa.VsrD(0)) ||                       \
+                 float64_is_any_nan(xb.VsrD(0)))) {                      \
+        if (float64_is_signaling_nan(xa.VsrD(0)) ||                      \
+            float64_is_signaling_nan(xb.VsrD(0))) {                      \
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0);       \
         }                                                                \
         if (ordered) {                                                   \
@@ -2286,9 +2366,10 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                      \
         }                                                                \
         cc = 1;                                                          \
     } else {                                                             \
-        if (float64_lt(xa.f64[0], xb.f64[0], &env->fp_status)) {         \
+        if (float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status)) {       \
             cc = 8;                                                      \
-        } else if (!float64_le(xa.f64[0], xb.f64[0], &env->fp_status)) { \
+        } else if (!float64_le(xa.VsrD(0), xb.VsrD(0),                   \
+                               &env->fp_status)) { \
             cc = 4;                                                      \
         } else {                                                         \
             cc = 2;                                                      \
@@ -2305,15 +2386,12 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                      \
 VSX_SCALAR_CMP(xscmpodp, 1)
 VSX_SCALAR_CMP(xscmpudp, 0)
 
-#define float64_snan_to_qnan(x) ((x) | 0x0008000000000000ul)
-#define float32_snan_to_qnan(x) ((x) | 0x00400000)
-
 /* VSX_MAX_MIN - VSX floating point maximum/minimum
  *   name  - instruction mnemonic
  *   op    - operation (max or min)
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  */
 #define VSX_MAX_MIN(name, op, nels, tp, fld)                                  \
 void helper_##name(CPUPPCState *env, uint32_t opcode)                         \
@@ -2326,9 +2404,9 @@ void helper_##name(CPUPPCState *env, uint32_t opcode)                         \
     getVSR(xT(opcode), &xt, env);                                             \
                                                                               \
     for (i = 0; i < nels; i++) {                                              \
-        xt.fld[i] = tp##_##op(xa.fld[i], xb.fld[i], &env->fp_status);         \
-        if (unlikely(tp##_is_signaling_nan(xa.fld[i]) ||                      \
-                     tp##_is_signaling_nan(xb.fld[i]))) {                     \
+        xt.fld = tp##_##op(xa.fld, xb.fld, &env->fp_status);                  \
+        if (unlikely(tp##_is_signaling_nan(xa.fld) ||                         \
+                     tp##_is_signaling_nan(xb.fld))) {                        \
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0);            \
         }                                                                     \
     }                                                                         \
@@ -2337,18 +2415,18 @@ void helper_##name(CPUPPCState *env, uint32_t opcode)                         \
     helper_float_check_status(env);                                           \
 }
 
-VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, f64)
-VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, f64)
-VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, f32)
-VSX_MAX_MIN(xsmindp, minnum, 1, float64, f64)
-VSX_MAX_MIN(xvmindp, minnum, 2, float64, f64)
-VSX_MAX_MIN(xvminsp, minnum, 4, float32, f32)
+VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0))
+VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i))
+VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i))
+VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0))
+VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i))
+VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i))
 
 /* VSX_CMP - VSX floating point compare
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   cmp   - comparison operation
  *   svxvc - set VXVC bit
  */
@@ -2365,23 +2443,23 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                       \
     getVSR(xT(opcode), &xt, env);                                         \
                                                                           \
     for (i = 0; i < nels; i++) {                                          \
-        if (unlikely(tp##_is_any_nan(xa.fld[i]) ||                        \
-                     tp##_is_any_nan(xb.fld[i]))) {                       \
-            if (tp##_is_signaling_nan(xa.fld[i]) ||                       \
-                tp##_is_signaling_nan(xb.fld[i])) {                       \
+        if (unlikely(tp##_is_any_nan(xa.fld) ||                           \
+                     tp##_is_any_nan(xb.fld))) {                          \
+            if (tp##_is_signaling_nan(xa.fld) ||                          \
+                tp##_is_signaling_nan(xb.fld)) {                          \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0);    \
             }                                                             \
             if (svxvc) {                                                  \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXVC, 0);      \
             }                                                             \
-            xt.fld[i] = 0;                                                \
+            xt.fld = 0;                                                   \
             all_true = 0;                                                 \
         } else {                                                          \
-            if (tp##_##cmp(xb.fld[i], xa.fld[i], &env->fp_status) == 1) { \
-                xt.fld[i] = -1;                                           \
+            if (tp##_##cmp(xb.fld, xa.fld, &env->fp_status) == 1) {       \
+                xt.fld = -1;                                              \
                 all_false = 0;                                            \
             } else {                                                      \
-                xt.fld[i] = 0;                                            \
+                xt.fld = 0;                                               \
                 all_true = 0;                                             \
             }                                                             \
         }                                                                 \
@@ -2394,18 +2472,12 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                       \
     helper_float_check_status(env);                                       \
  }
 
-VSX_CMP(xvcmpeqdp, 2, float64, f64, eq, 0)
-VSX_CMP(xvcmpgedp, 2, float64, f64, le, 1)
-VSX_CMP(xvcmpgtdp, 2, float64, f64, lt, 1)
-VSX_CMP(xvcmpeqsp, 4, float32, f32, eq, 0)
-VSX_CMP(xvcmpgesp, 4, float32, f32, le, 1)
-VSX_CMP(xvcmpgtsp, 4, float32, f32, lt, 1)
-
-#if defined(HOST_WORDS_BIGENDIAN)
-#define JOFFSET 0
-#else
-#define JOFFSET 1
-#endif
+VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0)
+VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1)
+VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1)
+VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0)
+VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1)
+VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1)
 
 /* VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion
  *   op    - instruction mnemonic
@@ -2426,7 +2498,6 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                \
     getVSR(xT(opcode), &xt, env);                                  \
                                                                    \
     for (i = 0; i < nels; i++) {                                   \
-        int j = 2*i + JOFFSET;                                     \
         xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status);        \
         if (unlikely(stp##_is_signaling_nan(xb.sfld))) {           \
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0); \
@@ -2434,7 +2505,7 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                \
         }                                                          \
         if (sfprf) {                                               \
             helper_compute_fprf(env, ttp##_to_float64(xt.tfld,     \
-                                &env->fp_status), sfprf);          \
+                                &env->fp_status));                 \
         }                                                          \
     }                                                              \
                                                                    \
@@ -2442,10 +2513,26 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                \
     helper_float_check_status(env);                                \
 }
 
-VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, f64[i], f32[j], 1)
-VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, f32[j], f64[i], 1)
-VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, f64[i], f32[j], 0)
-VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, f32[j], f64[i], 0)
+VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1)
+VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1)
+VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2*i), 0)
+VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2*i), VsrD(i), 0)
+
+uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb)
+{
+    float_status tstat = env->fp_status;
+    set_float_exception_flags(0, &tstat);
+
+    return (uint64_t)float64_to_float32(xb, &tstat) << 32;
+}
+
+uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb)
+{
+    float_status tstat = env->fp_status;
+    set_float_exception_flags(0, &tstat);
+
+    return float32_to_float64(xb >> 32, &tstat);
+}
 
 /* VSX_CVT_FP_TO_INT - VSX floating point to integer conversion
  *   op    - instruction mnemonic
@@ -2454,10 +2541,9 @@ VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, f32[j], f64[i], 0)
  *   ttp   - target type (int32, uint32, int64 or uint64)
  *   sfld  - source vsr_t field
  *   tfld  - target vsr_t field
- *   jdef  - definition of the j index (i or 2*i)
  *   rnan  - resulting NaN
  */
-#define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, jdef, rnan)        \
+#define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan)              \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
 {                                                                            \
     ppc_vsr_t xt, xb;                                                        \
@@ -2467,7 +2553,6 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     getVSR(xT(opcode), &xt, env);                                            \
                                                                              \
     for (i = 0; i < nels; i++) {                                             \
-        int j = jdef;                                                        \
         if (unlikely(stp##_is_any_nan(xb.sfld))) {                           \
             if (stp##_is_signaling_nan(xb.sfld)) {                           \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0);       \
@@ -2475,7 +2560,8 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 0);            \
             xt.tfld = rnan;                                                  \
         } else {                                                             \
-            xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status);              \
+            xt.tfld = stp##_to_##ttp##_round_to_zero(xb.sfld,                \
+                          &env->fp_status);                                  \
             if (env->fp_status.float_exception_flags & float_flag_invalid) { \
                 fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, 0);        \
             }                                                                \
@@ -2486,27 +2572,23 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                          \
     helper_float_check_status(env);                                          \
 }
 
-VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, f64[j], u64[i], i, \
-                  0x8000000000000000ul)
-VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, f64[i], u32[j], \
-                  2*i + JOFFSET, 0x80000000l)
-VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, f64[j], u64[i], i, 0ul)
-VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, f64[i], u32[j], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, f64[j], u64[i], i, \
-                  0x8000000000000000ul)
-VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, f64[i], u32[j], \
-                  2*i + JOFFSET, 0x80000000l)
-VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, f64[j], u64[i], i, 0ul)
-VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, f64[i], u32[j], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, f32[j], u64[i], \
-                  2*i + JOFFSET, 0x8000000000000000ul)
-VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, f32[j], u32[j], i, \
-                  0x80000000l)
-VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, f32[j], u64[i], \
-                  2*i + JOFFSET, 0ul)
-VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, f32[j], u32[i], i, 0)
+VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \
+                  0x8000000000000000ULL)
+VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \
+                  0x80000000U)
+VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL)
+VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U)
+VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \
+                  0x8000000000000000ULL)
+VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2*i), \
+                  0x80000000U)
+VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL)
+VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2*i), 0U)
+VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2*i), VsrD(i), \
+                  0x8000000000000000ULL)
+VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U)
+VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2*i), VsrD(i), 0ULL)
+VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U)
 
 /* VSX_CVT_INT_TO_FP - VSX integer to floating point conversion
  *   op    - instruction mnemonic
@@ -2518,7 +2600,7 @@ VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, f32[j], u32[i], i, 0)
  *   jdef  - definition of the j index (i or 2*i)
  *   sfprf - set FPRF
  */
-#define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, jdef, sfprf)  \
+#define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp)  \
 void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
 {                                                                       \
     ppc_vsr_t xt, xb;                                                   \
@@ -2528,10 +2610,12 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     getVSR(xT(opcode), &xt, env);                                       \
                                                                         \
     for (i = 0; i < nels; i++) {                                        \
-        int j = jdef;                                                   \
         xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status);             \
+        if (r2sp) {                                                     \
+            xt.tfld = helper_frsp(env, xt.tfld);                        \
+        }                                                               \
         if (sfprf) {                                                    \
-            helper_compute_fprf(env, xt.tfld, sfprf);                   \
+            helper_compute_fprf(env, xt.tfld);                          \
         }                                                               \
     }                                                                   \
                                                                         \
@@ -2539,20 +2623,18 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                     \
     helper_float_check_status(env);                                     \
 }
 
-VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, u64[j], f64[i], i, 1)
-VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, u64[j], f64[i], i, 1)
-VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, u64[j], f64[i], i, 0)
-VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, u64[j], f64[i], i, 0)
-VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, u32[j], f64[i], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, u32[j], f64[i], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, u64[i], f32[j], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, u64[i], f32[j], \
-                  2*i + JOFFSET, 0)
-VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, u32[j], f32[i], i, 0)
-VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, u32[j], f32[i], i, 0)
+VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0)
+VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0)
+VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1)
+VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1)
+VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2*i), VsrD(i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2*i), VsrD(i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2*i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2*i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0)
+VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0)
 
 /* For "use current rounding mode", define a value that will not be one of
  * the existing rounding model enums.
@@ -2564,7 +2646,7 @@ VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, u32[j], f32[i], i, 0)
  *   op    - instruction mnemonic
  *   nels  - number of elements (1, 2 or 4)
  *   tp    - type (float32 or float64)
- *   fld   - vsr_t field (f32 or f64)
+ *   fld   - vsr_t field (VsrD(*) or VsrW(*))
  *   rmode - rounding mode
  *   sfprf - set FPRF
  */
@@ -2581,14 +2663,14 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                    \
     }                                                                  \
                                                                        \
     for (i = 0; i < nels; i++) {                                       \
-        if (unlikely(tp##_is_signaling_nan(xb.fld[i]))) {              \
+        if (unlikely(tp##_is_signaling_nan(xb.fld))) {                 \
             fload_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, 0);     \
-            xt.fld[i] = tp##_snan_to_qnan(xb.fld[i]);                  \
+            xt.fld = tp##_snan_to_qnan(xb.fld);                        \
         } else {                                                       \
-            xt.fld[i] = tp##_round_to_int(xb.fld[i], &env->fp_status); \
+            xt.fld = tp##_round_to_int(xb.fld, &env->fp_status);       \
         }                                                              \
         if (sfprf) {                                                   \
-            helper_compute_fprf(env, xt.fld[i], sfprf);                \
+            helper_compute_fprf(env, xt.fld);                          \
         }                                                              \
     }                                                                  \
                                                                        \
@@ -2604,20 +2686,31 @@ void helper_##op(CPUPPCState *env, uint32_t opcode)                    \
     helper_float_check_status(env);                                    \
 }
 
-VSX_ROUND(xsrdpi, 1, float64, f64, float_round_nearest_even, 1)
-VSX_ROUND(xsrdpic, 1, float64, f64, FLOAT_ROUND_CURRENT, 1)
-VSX_ROUND(xsrdpim, 1, float64, f64, float_round_down, 1)
-VSX_ROUND(xsrdpip, 1, float64, f64, float_round_up, 1)
-VSX_ROUND(xsrdpiz, 1, float64, f64, float_round_to_zero, 1)
+VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_nearest_even, 1)
+VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1)
+VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1)
+VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1)
+VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1)
 
-VSX_ROUND(xvrdpi, 2, float64, f64, float_round_nearest_even, 0)
-VSX_ROUND(xvrdpic, 2, float64, f64, FLOAT_ROUND_CURRENT, 0)
-VSX_ROUND(xvrdpim, 2, float64, f64, float_round_down, 0)
-VSX_ROUND(xvrdpip, 2, float64, f64, float_round_up, 0)
-VSX_ROUND(xvrdpiz, 2, float64, f64, float_round_to_zero, 0)
+VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_nearest_even, 0)
+VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0)
+VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0)
+VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0)
+VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0)
 
-VSX_ROUND(xvrspi, 4, float32, f32, float_round_nearest_even, 0)
-VSX_ROUND(xvrspic, 4, float32, f32, FLOAT_ROUND_CURRENT, 0)
-VSX_ROUND(xvrspim, 4, float32, f32, float_round_down, 0)
-VSX_ROUND(xvrspip, 4, float32, f32, float_round_up, 0)
-VSX_ROUND(xvrspiz, 4, float32, f32, float_round_to_zero, 0)
+VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_nearest_even, 0)
+VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0)
+VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0)
+VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0)
+VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0)
+
+uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb)
+{
+    helper_reset_fpstatus(env);
+
+    uint64_t xt = helper_frsp(env, xb);
+
+    helper_compute_fprf(env, xt);
+    helper_float_check_status(env);
+    return xt;
+}
This page took 0.089674 seconds and 4 git commands to generate.