commit 600a0099a1eb2335a3f9563534c112e11817002b

Author: Vadim Barkov <vbrkov@gmail.com>

Date:   Fri Oct 5 13:51:49 2018 +0300

    Bug 385411 s390x: Add z13 vector floating point support

    This adds support for the z/Architecture vector FP instructions that were

    introduced with z13.

    The patch was contributed by Vadim Barkov, with some clean-up and minor

    adjustments by Andreas Arnez.

diff --git a/VEX/priv/guest_s390_defs.h b/VEX/priv/guest_s390_defs.h

index 3bfecbe..d72cc9f 100644

--- a/VEX/priv/guest_s390_defs.h

+++ b/VEX/priv/guest_s390_defs.h

@@ -281,7 +281,11 @@ enum {

    S390_VEC_OP_VMALH = 13,

    S390_VEC_OP_VCH = 14,

    S390_VEC_OP_VCHL = 15,

-   S390_VEC_OP_LAST = 16 // supposed to be the last element in enum

+   S390_VEC_OP_VFCE = 16,

+   S390_VEC_OP_VFCH = 17,

+   S390_VEC_OP_VFCHE = 18,

+   S390_VEC_OP_VFTCI = 19,

+   S390_VEC_OP_LAST = 20 // supposed to be the last element in enum

 } s390x_vec_op_t;

 /* Arguments of s390x_dirtyhelper_vec_op(...) which are packed into one

@@ -300,8 +304,10 @@ typedef union {

       unsigned int m4 : 4;        // field m4 of insn or zero if it's missing

       unsigned int m5 : 4;        // field m5 of insn or zero if it's missing

+      unsigned int m6 : 4;        // field m6 of insn or zero if it's missing

+      unsigned int i3 : 12;       // field i3 of insn or zero if it's missing

       unsigned int read_only: 1;  // don't write result to Guest State

-      unsigned int reserved : 27; // reserved for future

+      unsigned int reserved : 11; // reserved for future

    };

    ULong serialized;

 } s390x_vec_op_details_t;

diff --git a/VEX/priv/guest_s390_helpers.c b/VEX/priv/guest_s390_helpers.c

index d9773e7..5877743 100644

--- a/VEX/priv/guest_s390_helpers.c

+++ b/VEX/priv/guest_s390_helpers.c

@@ -2498,6 +2498,10 @@ s390x_dirtyhelper_vec_op(VexGuestS390XState *guest_state,

       {0xe7, 0xa9}, /* VMALH */

       {0xe7, 0xfb}, /* VCH */

       {0xe7, 0xf9}, /* VCHL */

+      {0xe7, 0xe8}, /* VFCE */

+      {0xe7, 0xeb}, /* VFCH */

+      {0xe7, 0xea}, /* VFCHE */

+      {0xe7, 0x4a}  /* VFTCI */

    };

    union {

@@ -2525,6 +2529,28 @@ s390x_dirtyhelper_vec_op(VexGuestS390XState *guest_state,

         unsigned int rxb : 4;

         unsigned int op2 : 8;

       } VRRd;

+      struct {

+         UInt op1 : 8;

+         UInt v1  : 4;

+         UInt v2  : 4;

+         UInt v3  : 4;

+         UInt     : 4;

+         UInt m6  : 4;

+         UInt m5  : 4;

+         UInt m4  : 4;

+         UInt rxb : 4;

+         UInt op2 : 8;

+      } VRRc;

+      struct {

+         UInt op1 : 8;

+         UInt v1  : 4;

+         UInt v2  : 4;

+         UInt i3  : 12;

+         UInt m5  : 4;

+         UInt m4  : 4;

+         UInt rxb : 4;

+         UInt op2 : 8;

+      } VRIe;

       UChar bytes[6];

    } the_insn;

@@ -2578,6 +2604,27 @@ s390x_dirtyhelper_vec_op(VexGuestS390XState *guest_state,

       the_insn.VRRd.m6 = d->m5;

       break;

+   case S390_VEC_OP_VFCE:

+   case S390_VEC_OP_VFCH:

+   case S390_VEC_OP_VFCHE:

+      the_insn.VRRc.v1 = 1;

+      the_insn.VRRc.v2 = 2;

+      the_insn.VRRc.v3 = 3;

+      the_insn.VRRc.rxb = 0b1110;

+      the_insn.VRRc.m4 = d->m4;

+      the_insn.VRRc.m5 = d->m5;

+      the_insn.VRRc.m6 = d->m6;

+      break;

+

+   case S390_VEC_OP_VFTCI:

+      the_insn.VRIe.v1 = 1;

+      the_insn.VRIe.v2 = 2;

+      the_insn.VRIe.rxb = 0b1100;

+      the_insn.VRIe.i3 = d->i3;

+      the_insn.VRIe.m4 = d->m4;

+      the_insn.VRIe.m5 = d->m5;

+      break;

+

    default:

       vex_printf("operation = %d\n", d->op);

       vpanic("s390x_dirtyhelper_vec_op: unknown operation");

diff --git a/VEX/priv/guest_s390_toIR.c b/VEX/priv/guest_s390_toIR.c

index 50a5a41..1c4ac39 100644

--- a/VEX/priv/guest_s390_toIR.c

+++ b/VEX/priv/guest_s390_toIR.c

@@ -86,6 +86,7 @@ typedef enum {

    S390_DECODE_UNKNOWN_INSN,

    S390_DECODE_UNIMPLEMENTED_INSN,

    S390_DECODE_UNKNOWN_SPECIAL_INSN,

+   S390_DECODE_SPECIFICATION_EXCEPTION,

    S390_DECODE_ERROR

 } s390_decode_t;

@@ -421,6 +422,26 @@ yield_if(IRExpr *condition)

                     S390X_GUEST_OFFSET(guest_IA)));

 }

+/* Convenience macro to yield a specification exception if the given condition

+   is not met.  Used to pass this type of decoding error up through the call

+   chain. */

+#define s390_insn_assert(mnm, cond)             \

+   do {                                         \

+      if (!(cond)) {                            \

+         dis_res->whatNext = Dis_StopHere;      \

+         dis_res->jk_StopHere = Ijk_NoDecode;   \

+         return (mnm);                          \

+      }                                         \

+   } while (0)

+

+/* Convenience function to check for a specification exception. */

+static Bool

+is_specification_exception(void)

+{

+   return (dis_res->whatNext == Dis_StopHere &&

+           dis_res->jk_StopHere == Ijk_NoDecode);

+}

+

 static __inline__ IRExpr *get_fpr_dw0(UInt);

 static __inline__ void    put_fpr_dw0(UInt, IRExpr *);

 static __inline__ IRExpr *get_dpr_dw0(UInt);

@@ -1770,6 +1791,11 @@ s390_vr_get_type(const UChar m)

 /* Determine if Zero Search (ZS) flag is set in m field */

 #define s390_vr_is_zs_set(m) (((m) & 0b0010) != 0)

+/* Check if the "Single-Element-Control" bit is set.

+   Used in vector FP instructions.

+ */

+#define s390_vr_is_single_element_control_set(m) (((m) & 0x8) != 0)

+

 /* Generates arg1 < arg2 (or arg1 <= arg2 if allow_equal == True) expression.

    Arguments must have V128 type and are treated as unsigned 128-bit numbers.

 */

@@ -2001,12 +2027,14 @@ s390_vr_offset_by_index(UInt archreg,IRType type, UChar index)

       return vr_offset(archreg) + sizeof(UShort) * index;

    case Ity_I32:

+   case Ity_F32:

       if(index > 3) {

          goto invalidIndex;

       }

       return vr_offset(archreg) + sizeof(UInt) * index;

    case Ity_I64:

+   case Ity_F64:

       if(index > 1) {

          goto invalidIndex;

       }

@@ -2237,8 +2265,8 @@ encode_bfp_rounding_mode(UChar mode)

    case S390_BFP_ROUND_PER_FPC:

       rm = get_bfp_rounding_mode_from_fpc();

       break;

-   case S390_BFP_ROUND_NEAREST_AWAY:  /* not supported */

-   case S390_BFP_ROUND_PREPARE_SHORT: /* not supported */

+   case S390_BFP_ROUND_NEAREST_AWAY:  rm = mkU32(Irrm_NEAREST_TIE_AWAY_0); break;

+   case S390_BFP_ROUND_PREPARE_SHORT: rm = mkU32(Irrm_PREPARE_SHORTER); break;

    case S390_BFP_ROUND_NEAREST_EVEN:  rm = mkU32(Irrm_NEAREST); break;

    case S390_BFP_ROUND_ZERO:          rm = mkU32(Irrm_ZERO);    break;

    case S390_BFP_ROUND_POSINF:        rm = mkU32(Irrm_PosINF);  break;

@@ -3524,6 +3552,26 @@ s390_format_VRI_VVIM(const HChar *(*irgen)(UChar v1, UChar v3, UShort i2, UChar

       s390_disasm(ENC5(MNM, VR, VR, UINT, UINT), mnm, v1, v3, i2, m4);

 }

+static void

+s390_format_VRI_VVIMM(const HChar *(*irgen)(UChar v1, UChar v2, UShort i3,

+                                            UChar m4, UChar m5),

+                      UChar v1, UChar v2, UShort i3, UChar m4, UChar m5,

+                      UChar rxb)

+{

+   const HChar *mnm;

+

+   if (!s390_host_has_vx) {

+      emulation_failure(EmFail_S390X_vx);

+      return;

+   }

+

+   v1 = s390_vr_getVRindex(v1, 1, rxb);

+   v2 = s390_vr_getVRindex(v2, 2, rxb);

+   mnm = irgen(v1, v2, i3, m4, m5);

+

+   if (vex_traceflags & VEX_TRACE_FE)

+      s390_disasm(ENC6(MNM, VR, VR, UINT, UINT, UINT), mnm, v1, v2, i3, m4, m5);

+}

 static void

 s390_format_VRS_RRDVM(const HChar *(*irgen)(UChar r1, IRTemp op2addr, UChar v3,

@@ -3680,7 +3728,7 @@ s390_format_VRV_VVRDMT(const HChar *(*irgen)(UChar v1, IRTemp op2addr, UChar m3)

 static void

-s390_format_VRRd_VVVVMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar v3,

+s390_format_VRR_VVVVMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar v3,

                                               UChar v4, UChar m5, UChar m6),

                         UChar v1, UChar v2, UChar v3, UChar v4, UChar m5,

                         UChar m6, UChar rxb)

@@ -3794,6 +3842,92 @@ s390_format_VRRd_VVVVM(const HChar *(*irgen)(UChar v1, UChar v2, UChar v3,

 }

+static void

+s390_format_VRRa_VVMMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar m3,

+                                             UChar m4, UChar m5),

+                       UChar v1, UChar v2, UChar m3, UChar m4, UChar m5,

+                       UChar rxb)

+{

+   const HChar *mnm;

+

+   if (!s390_host_has_vx) {

+      emulation_failure(EmFail_S390X_vx);

+      return;

+   }

+

+   v1 = s390_vr_getVRindex(v1, 1, rxb);

+   v2 = s390_vr_getVRindex(v2, 2, rxb);

+   mnm = irgen(v1, v2, m3, m4, m5);

+

+   if (vex_traceflags & VEX_TRACE_FE)

+      s390_disasm(ENC6(MNM, VR, VR, UINT, UINT, UINT), mnm, v1, v2, m3, m4, m5);

+}

+

+static void

+s390_format_VRRa_VVVMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar v3,

+                                             UChar m4, UChar m5),

+                       UChar v1, UChar v2, UChar v3, UChar m4, UChar m5,

+                       UChar rxb)

+{

+   const HChar *mnm;

+

+   if (!s390_host_has_vx) {

+      emulation_failure(EmFail_S390X_vx);

+      return;

+   }

+

+   v1 = s390_vr_getVRindex(v1, 1, rxb);

+   v2 = s390_vr_getVRindex(v2, 2, rxb);

+   v3 = s390_vr_getVRindex(v3, 3, rxb);

+   mnm = irgen(v1, v2, v3, m4, m5);

+

+   if (vex_traceflags & VEX_TRACE_FE)

+      s390_disasm(ENC6(MNM, VR, VR, VR, UINT, UINT), mnm, v1, v2, v3, m4, m5);

+}

+

+static void

+s390_format_VRRa_VVMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar m3,

+                                            UChar m4),

+                       UChar v1, UChar v2, UChar m3, UChar m4, UChar rxb)

+{

+   const HChar *mnm;

+

+   if (!s390_host_has_vx) {

+      emulation_failure(EmFail_S390X_vx);

+      return;

+   }

+

+   v1 = s390_vr_getVRindex(v1, 1, rxb);

+   v2 = s390_vr_getVRindex(v2, 2, rxb);

+   mnm = irgen(v1, v2, m3, m4);

+

+   if (vex_traceflags & VEX_TRACE_FE)

+      s390_disasm(ENC5(MNM, VR, VR, UINT, UINT), mnm, v1, v2, m3, m4);

+}

+

+static void

+s390_format_VRRa_VVVMMM(const HChar *(*irgen)(UChar v1, UChar v2, UChar v3,

+                                              UChar m4, UChar m5, UChar m6),

+                        UChar v1, UChar v2, UChar v3, UChar m4, UChar m5,

+                        UChar m6, UChar rxb)

+{

+   const HChar *mnm;

+

+   if (!s390_host_has_vx) {

+      emulation_failure(EmFail_S390X_vx);

+      return;

+   }

+

+   v1 = s390_vr_getVRindex(v1, 1, rxb);

+   v2 = s390_vr_getVRindex(v2, 2, rxb);

+   v3 = s390_vr_getVRindex(v3, 3, rxb);

+   mnm = irgen(v1, v2, v3, m4, m5, m6);

+

+   if (vex_traceflags & VEX_TRACE_FE)

+      s390_disasm(ENC6(MNM, VR, VR, VR, UINT, UINT),

+                  mnm, v1, v2, v3, m4, m5, m6);

+}

+

 /*------------------------------------------------------------*/

 /*--- Build IR for opcodes                                 ---*/

 /*------------------------------------------------------------*/

@@ -17895,6 +18029,575 @@ s390_irgen_VMALH(UChar v1, UChar v2, UChar v3, UChar v4, UChar m5)

    return "vmalh";

 }

+static void

+s390_vector_fp_convert(IROp op, IRType fromType, IRType toType,

+                       UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   Bool isSingleElementOp = s390_vr_is_single_element_control_set(m4);

+   UChar maxIndex = isSingleElementOp ? 0 : 1;

+

+   /* For Iop_F32toF64 we do this:

+      f32[0] -> f64[0]

+      f32[2] -> f64[1]

+

+      For Iop_F64toF32 we do this:

+      f64[0] -> f32[0]

+      f64[1] -> f32[2]

+

+      The magic below with scaling factors is used to achieve the logic

+      described above.

+   */

+   const UChar sourceIndexScaleFactor = (op == Iop_F32toF64) ? 2 : 1;

+   const UChar destinationIndexScaleFactor = (op == Iop_F64toF32) ? 2 : 1;

+

+   const Bool isUnary = (op == Iop_F32toF64);

+   for (UChar i = 0; i <= maxIndex; i++) {

+      IRExpr* argument = get_vr(v2, fromType, i * sourceIndexScaleFactor);

+      IRExpr* result;

+      if (!isUnary) {

+         result = binop(op,

+                        mkexpr(encode_bfp_rounding_mode(m5)),

+                        argument);

+      } else {

+         result = unop(op, argument);

+      }

+      put_vr(v1, toType, i * destinationIndexScaleFactor, result);

+   }

+

+   if (isSingleElementOp) {

+      put_vr_dw1(v1, mkU64(0));

+   }

+}

+

+static const HChar *

+s390_irgen_VCDG(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vcdg", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      emulation_warning(EmWarn_S390X_fpext_rounding);

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_I64StoF64, Ity_I64, Ity_F64, v1, v2, m3, m4, m5);

+

+   return "vcdg";

+}

+

+static const HChar *

+s390_irgen_VCDLG(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vcdlg", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      emulation_warning(EmWarn_S390X_fpext_rounding);

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_I64UtoF64, Ity_I64, Ity_F64, v1, v2, m3, m4, m5);

+

+   return "vcdlg";

+}

+

+static const HChar *

+s390_irgen_VCGD(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vcgd", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      emulation_warning(EmWarn_S390X_fpext_rounding);

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_F64toI64S, Ity_F64, Ity_I64, v1, v2, m3, m4, m5);

+

+   return "vcgd";

+}

+

+static const HChar *

+s390_irgen_VCLGD(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vclgd", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      emulation_warning(EmWarn_S390X_fpext_rounding);

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_F64toI64U, Ity_F64, Ity_I64, v1, v2, m3, m4, m5);

+

+   return "vclgd";

+}

+

+static const HChar *

+s390_irgen_VFI(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfi", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      emulation_warning(EmWarn_S390X_fpext_rounding);

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_RoundF64toInt, Ity_F64, Ity_F64,

+                          v1, v2, m3, m4, m5);

+

+   return "vcgld";

+}

+

+static const HChar *

+s390_irgen_VLDE(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vlde", m3 == 2);

+

+   s390_vector_fp_convert(Iop_F32toF64, Ity_F32, Ity_F64, v1, v2, m3, m4, m5);

+

+   return "vlde";

+}

+

+static const HChar *

+s390_irgen_VLED(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vled", m3 == 3);

+

+   if (!s390_host_has_fpext && m5 != S390_BFP_ROUND_PER_FPC) {

+      m5 = S390_BFP_ROUND_PER_FPC;

+   }

+

+   s390_vector_fp_convert(Iop_F64toF32, Ity_F64, Ity_F32, v1, v2, m3, m4, m5);

+

+   return "vled";

+}

+

+static const HChar *

+s390_irgen_VFPSO(UChar v1, UChar v2, UChar m3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfpso", m3 == 3);

+

+   IRExpr* result;

+   switch (m5) {

+   case 0: {

+      /* Invert sign */

+      if (!s390_vr_is_single_element_control_set(m4)) {

+         result = unop(Iop_Neg64Fx2, get_vr_qw(v2));

+      }

+      else {

+         result = binop(Iop_64HLtoV128,

+                        unop(Iop_ReinterpF64asI64,

+                             unop(Iop_NegF64, get_vr(v2, Ity_F64, 0))),

+                        mkU64(0));

+      }

+      break;

+   }

+

+   case 1: {

+      /* Set sign to negative */

+      IRExpr* highHalf = mkU64(0x8000000000000000ULL);

+      if (!s390_vr_is_single_element_control_set(m4)) {

+         IRExpr* lowHalf = highHalf;

+         IRExpr* mask = binop(Iop_64HLtoV128, highHalf, lowHalf);

+         result = binop(Iop_OrV128, get_vr_qw(v2), mask);

+      }

+      else {

+         result = binop(Iop_64HLtoV128,

+                        binop(Iop_Or64, get_vr_dw0(v2), highHalf),

+                        mkU64(0ULL));

+      }

+

+      break;

+   }

+

+   case 2: {

+      /* Set sign to positive */

+      if (!s390_vr_is_single_element_control_set(m4)) {

+         result = unop(Iop_Abs64Fx2, get_vr_qw(v2));

+      }

+      else {

+         result = binop(Iop_64HLtoV128,

+                        unop(Iop_ReinterpF64asI64,

+                             unop(Iop_AbsF64, get_vr(v2, Ity_F64, 0))),

+                        mkU64(0));

+      }

+

+      break;

+   }

+

+   default:

+      vpanic("s390_irgen_VFPSO: Invalid m5 value");

+   }

+

+   put_vr_qw(v1, result);

+   if (s390_vr_is_single_element_control_set(m4)) {

+      put_vr_dw1(v1, mkU64(0ULL));

+   }

+

+   return "vfpso";

+}

+

+static void s390x_vec_fp_binary_op(IROp generalOp, IROp singleElementOp,

+                                   UChar v1, UChar v2, UChar v3, UChar m4,

+                                   UChar m5)

+{

+   IRExpr* result;

+   if (!s390_vr_is_single_element_control_set(m5)) {

+      result = triop(generalOp, get_bfp_rounding_mode_from_fpc(),

+                     get_vr_qw(v2), get_vr_qw(v3));

+   } else {

+      IRExpr* highHalf = triop(singleElementOp,

+                               get_bfp_rounding_mode_from_fpc(),

+                               get_vr(v2, Ity_F64, 0),

+                               get_vr(v3, Ity_F64, 0));

+      result = binop(Iop_64HLtoV128, unop(Iop_ReinterpF64asI64, highHalf),

+                     mkU64(0ULL));

+   }

+

+   put_vr_qw(v1, result);

+}

+

+static void s390x_vec_fp_unary_op(IROp generalOp, IROp singleElementOp,

+                                  UChar v1, UChar v2, UChar m3, UChar m4)

+{

+   IRExpr* result;

+   if (!s390_vr_is_single_element_control_set(m4)) {

+      result = binop(generalOp, get_bfp_rounding_mode_from_fpc(),

+                     get_vr_qw(v2));

+   }

+   else {

+      IRExpr* highHalf = binop(singleElementOp,

+                               get_bfp_rounding_mode_from_fpc(),

+                               get_vr(v2, Ity_F64, 0));

+      result = binop(Iop_64HLtoV128, unop(Iop_ReinterpF64asI64, highHalf),

+                     mkU64(0ULL));

+   }

+

+   put_vr_qw(v1, result);

+}

+

+

+static void

+s390_vector_fp_mulAddOrSub(IROp singleElementOp,

+                           UChar v1, UChar v2, UChar v3, UChar v4,

+                           UChar m5, UChar m6)

+{

+   Bool isSingleElementOp = s390_vr_is_single_element_control_set(m5);

+   IRTemp irrm_temp = newTemp(Ity_I32);

+   assign(irrm_temp, get_bfp_rounding_mode_from_fpc());

+   IRExpr* irrm = mkexpr(irrm_temp);

+   IRExpr* result;

+   IRExpr* highHalf = qop(singleElementOp,

+                          irrm,

+                          get_vr(v2, Ity_F64, 0),

+                          get_vr(v3, Ity_F64, 0),

+                          get_vr(v4, Ity_F64, 0));

+

+   if (isSingleElementOp) {

+      result = binop(Iop_64HLtoV128, unop(Iop_ReinterpF64asI64, highHalf),

+                     mkU64(0ULL));

+   } else {

+      IRExpr* lowHalf = qop(singleElementOp,

+                            irrm,

+                            get_vr(v2, Ity_F64, 1),

+                            get_vr(v3, Ity_F64, 1),

+                            get_vr(v4, Ity_F64, 1));

+      result = binop(Iop_64HLtoV128, unop(Iop_ReinterpF64asI64, highHalf),

+                     unop(Iop_ReinterpF64asI64, lowHalf));

+   }

+

+   put_vr_qw(v1, result);

+}

+

+static const HChar *

+s390_irgen_VFA(UChar v1, UChar v2, UChar v3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfa", m4 == 3);

+   s390x_vec_fp_binary_op(Iop_Add64Fx2, Iop_AddF64, v1, v2, v3, m4, m5);

+   return "vfa";

+}

+

+static const HChar *

+s390_irgen_VFS(UChar v1, UChar v2, UChar v3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfs", m4 == 3);

+   s390x_vec_fp_binary_op(Iop_Sub64Fx2, Iop_SubF64, v1, v2, v3, m4, m5);

+   return "vfs";

+}

+

+static const HChar *

+s390_irgen_VFM(UChar v1, UChar v2, UChar v3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfm", m4 == 3);

+   s390x_vec_fp_binary_op(Iop_Mul64Fx2, Iop_MulF64, v1, v2, v3, m4, m5);

+   return "vfm";

+}

+

+static const HChar *

+s390_irgen_VFD(UChar v1, UChar v2, UChar v3, UChar m4, UChar m5)

+{

+   s390_insn_assert("vfd", m4 == 3);

+   s390x_vec_fp_binary_op(Iop_Div64Fx2, Iop_DivF64, v1, v2, v3, m4, m5);

+   return "vfd";

+}

+

+static const HChar *

+s390_irgen_VFSQ(UChar v1, UChar v2, UChar m3, UChar m4)

+{

+   s390_insn_assert("vfsq", m3 == 3);

+   s390x_vec_fp_unary_op(Iop_Sqrt64Fx2, Iop_SqrtF64, v1, v2, m3, m4);

+

+   return "vfsq";

+}

+

+static const HChar *

+s390_irgen_VFMA(UChar v1, UChar v2, UChar v3, UChar v4, UChar m5, UChar m6)

+{

+   s390_insn_assert("vfma", m6 == 3);

+   s390_vector_fp_mulAddOrSub(Iop_MAddF64, v1, v2, v3, v4, m5, m6);

+   return "vfma";

+}

+

+static const HChar *

+s390_irgen_VFMS(UChar v1, UChar v2, UChar v3, UChar v4, UChar m5, UChar m6)

+{

+   s390_insn_assert("vfms", m6 == 3);

+   s390_vector_fp_mulAddOrSub(Iop_MSubF64, v1, v2, v3, v4, m5, m6);

+   return "vfms";

+}

+

+static const HChar *

+s390_irgen_WFC(UChar v1, UChar v2, UChar m3, UChar m4)

+{

+   s390_insn_assert("wfc", m3 == 3);

+   s390_insn_assert("wfc", m4 == 0);

+

+   IRTemp cc_vex = newTemp(Ity_I32);

+   assign(cc_vex, binop(Iop_CmpF64,

+                        get_vr(v1, Ity_F64, 0), get_vr(v2, Ity_F64, 0)));

+

+   IRTemp cc_s390 = newTemp(Ity_I32);

+   assign(cc_s390, convert_vex_bfpcc_to_s390(cc_vex));

+   s390_cc_thunk_put1(S390_CC_OP_SET, cc_s390, False);

+

+   return "wfc";

+}

+

+static const HChar *

+s390_irgen_WFK(UChar v1, UChar v2, UChar m3, UChar m4)

+{

+   s390_irgen_WFC(v1, v2, m3, m4);

+

+   return "wfk";

+}

+

+static const HChar *

+s390_irgen_VFCE(UChar v1, UChar v2, UChar v3, UChar m4, UChar m5, UChar m6)

+{

+   s390_insn_assert("vfce", m4 == 3);

+

+   Bool isSingleElementOp = s390_vr_is_single_element_control_set(m5);

+   if (!s390_vr_is_cs_set(m6)) {

+      if (!isSingleElementOp) {

+         put_vr_qw(v1, binop(Iop_CmpEQ64Fx2, get_vr_qw(v2), get_vr_qw(v3)));

+      } else {

+         IRExpr* comparisonResult = binop(Iop_CmpF64, get_vr(v2, Ity_F64, 0),

+                                          get_vr(v3, Ity_F64, 0));

+         IRExpr* result = mkite(binop(Iop_CmpEQ32, comparisonResult,

+                                      mkU32(Ircr_EQ)),

+                                mkU64(0xffffffffffffffffULL),

+                                mkU64(0ULL));

+         put_vr_qw(v1, binop(Iop_64HLtoV128, result, mkU64(0ULL)));

+      }

+   } else {

+      IRDirty* d;

+      IRTemp cc = newTemp(Ity_I64);

+

+      s390x_vec_op_details_t details = { .serialized = 0ULL };

+      details.op = S390_VEC_OP_VFCE;

+      details.v1 = v1;

+      details.v2 = v2;

+      details.v3 = v3;

+      details.m4 = m4;

+      details.m5 = m5;

+      details.m6 = m6;

+

+      d = unsafeIRDirty_1_N(cc, 0, "s390x_dirtyhelper_vec_op",

+                            &s390x_dirtyhelper_vec_op,

+                            mkIRExprVec_2(IRExpr_GSPTR(),

+                                          mkU64(details.serialized)));

+

+      const UChar elementSize = isSingleElementOp ? sizeof(ULong) : sizeof(V128);