tstellar / rpms / llvm

Forked from rpms/llvm 3 years ago
Clone
Blob Blame History Raw
From 88ad713b81c2f51dd8405b251f9825b0bca6e57d Mon Sep 17 00:00:00 2001
From: Nemanja Ivanovic <nemanja.i.ibm@gmail.com>
Date: Thu, 2 Aug 2018 00:03:22 +0000
Subject: [PATCH] [PowerPC] Do not round values prior to converting to integer

Adding the FP_ROUND nodes when combining FP_TO_[SU]INT of elements
feeding a BUILD_VECTOR into an FP_TO_[SU]INT of the built vector
loses precision. This patch removes the code that adds these nodes
to true f64 operands. It also adds patterns required to ensure
the code is still vectorized rather than converting individual
elements and inserting into a vector.

Fixes https://bugs.llvm.org/show_bug.cgi?id=38342

Differential Revision: https://reviews.llvm.org/D50121

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338658 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/PowerPC/PPCISelLowering.cpp     |  22 +-
 lib/Target/PowerPC/PPCInstrVSX.td          |  86 +++++++
 test/CodeGen/PowerPC/build-vector-tests.ll | 357 +++++++++++++----------------
 3 files changed, 258 insertions(+), 207 deletions(-)

diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index f622b05..527ec5a 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -11560,6 +11560,14 @@ SDValue PPCTargetLowering::DAGCombineExtBoolTrunc(SDNode *N,
       ShiftCst);
 }
 
+// Is this an extending load from an f32 to an f64?
+static bool isFPExtLoad(SDValue Op) {
+  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Op.getNode()))
+    return LD->getExtensionType() == ISD::EXTLOAD &&
+      Op.getValueType() == MVT::f64;
+  return false;
+}
+
 /// \brief Reduces the number of fp-to-int conversion when building a vector.
 ///
 /// If this vector is built out of floating to integer conversions,
@@ -11594,11 +11602,18 @@ combineElementTruncationToVectorTruncation(SDNode *N,
     SmallVector<SDValue, 4> Ops;
     EVT TargetVT = N->getValueType(0);
     for (int i = 0, e = N->getNumOperands(); i < e; ++i) {
-      if (N->getOperand(i).getOpcode() != PPCISD::MFVSR)
+      SDValue NextOp = N->getOperand(i);
+      if (NextOp.getOpcode() != PPCISD::MFVSR)
         return SDValue();
-      unsigned NextConversion = N->getOperand(i).getOperand(0).getOpcode();
+      unsigned NextConversion = NextOp.getOperand(0).getOpcode();
       if (NextConversion != FirstConversion)
         return SDValue();
+      // If we are converting to 32-bit integers, we need to add an FP_ROUND.
+      // This is not valid if the input was originally double precision. It is
+      // also not profitable to do unless this is an extending load in which
+      // case doing this combine will allow us to combine consecutive loads.
+      if (Is32Bit && !isFPExtLoad(NextOp.getOperand(0).getOperand(0)))
+        return SDValue();
       if (N->getOperand(i) != FirstInput)
         IsSplat = false;
     }
@@ -11612,8 +11627,9 @@ combineElementTruncationToVectorTruncation(SDNode *N,
     // Now that we know we have the right type of node, get its operands
     for (int i = 0, e = N->getNumOperands(); i < e; ++i) {
       SDValue In = N->getOperand(i).getOperand(0);
-      // For 32-bit values, we need to add an FP_ROUND node.
       if (Is32Bit) {
+        // For 32-bit values, we need to add an FP_ROUND node (if we made it
+        // here, we know that all inputs are extending loads so this is safe).
         if (In.isUndef())
           Ops.push_back(DAG.getUNDEF(SrcVT));
         else {
diff --git a/lib/Target/PowerPC/PPCInstrVSX.td b/lib/Target/PowerPC/PPCInstrVSX.td
index 6f71978..1f48473 100644
--- a/lib/Target/PowerPC/PPCInstrVSX.td
+++ b/lib/Target/PowerPC/PPCInstrVSX.td
@@ -3100,6 +3100,17 @@ def DblToFlt {
   dag B1 = (f32 (fpround (f64 (extractelt v2f64:$B, 1))));
 }
 
+def ExtDbl {
+  dag A0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 0))))));
+  dag A1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 1))))));
+  dag B0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 0))))));
+  dag B1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 1))))));
+  dag A0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 0))))));
+  dag A1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 1))))));
+  dag B0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 0))))));
+  dag B1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 1))))));
+}
+
 def ByteToWord {
   dag LE_A0 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 0)), i8));
   dag LE_A1 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 4)), i8));
@@ -3177,9 +3188,15 @@ def FltToULong {
 }
 def DblToInt {
   dag A = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$A))));
+  dag B = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$B))));
+  dag C = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$C))));
+  dag D = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$D))));
 }
 def DblToUInt {
   dag A = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$A))));
+  dag B = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$B))));
+  dag C = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$C))));
+  dag D = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$D))));
 }
 def DblToLong {
   dag A = (i64 (PPCmfvsr (f64 (PPCfctidz f64:$A))));
@@ -3218,6 +3235,47 @@ def MrgFP {
   dag BAlToFlt = (XVCVDPSP (XXPERMDI $B, $A, 3));
 }
 
+// Word-element merge dags - conversions from f64 to i32 merged into vectors.
+def MrgWords {
+  // For big endian, we merge low and hi doublewords (A, B).
+  dag A0B0 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 0));
+  dag A1B1 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 3));
+  dag CVA1B1S = (v4i32 (XVCVDPSXWS A1B1));
+  dag CVA0B0S = (v4i32 (XVCVDPSXWS A0B0));
+  dag CVA1B1U = (v4i32 (XVCVDPUXWS A1B1));
+  dag CVA0B0U = (v4i32 (XVCVDPUXWS A0B0));
+
+  // For little endian, we merge low and hi doublewords (B, A).
+  dag B1A1 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 0));
+  dag B0A0 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 3));
+  dag CVB1A1S = (v4i32 (XVCVDPSXWS B1A1));
+  dag CVB0A0S = (v4i32 (XVCVDPSXWS B0A0));
+  dag CVB1A1U = (v4i32 (XVCVDPUXWS B1A1));
+  dag CVB0A0U = (v4i32 (XVCVDPUXWS B0A0));
+
+  // For big endian, we merge hi doublewords of (A, C) and (B, D), convert
+  // then merge.
+  dag AC = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$A, VSRC),
+                            (COPY_TO_REGCLASS f64:$C, VSRC), 0));
+  dag BD = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$B, VSRC),
+                            (COPY_TO_REGCLASS f64:$D, VSRC), 0));
+  dag CVACS = (v4i32 (XVCVDPSXWS AC));
+  dag CVBDS = (v4i32 (XVCVDPSXWS BD));
+  dag CVACU = (v4i32 (XVCVDPUXWS AC));
+  dag CVBDU = (v4i32 (XVCVDPUXWS BD));
+
+  // For little endian, we merge hi doublewords of (D, B) and (C, A), convert
+  // then merge.
+  dag DB = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$D, VSRC),
+                            (COPY_TO_REGCLASS f64:$B, VSRC), 0));
+  dag CA = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$C, VSRC),
+                            (COPY_TO_REGCLASS f64:$A, VSRC), 0));
+  dag CVDBS = (v4i32 (XVCVDPSXWS DB));
+  dag CVCAS = (v4i32 (XVCVDPSXWS CA));
+  dag CVDBU = (v4i32 (XVCVDPUXWS DB));
+  dag CVCAU = (v4i32 (XVCVDPUXWS CA));
+}
+
 // Patterns for BUILD_VECTOR nodes.
 def NoP9Vector : Predicate<"!PPCSubTarget->hasP9Vector()">;
 let AddedComplexity = 400 in {
@@ -3286,6 +3344,20 @@ let AddedComplexity = 400 in {
     def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
                                    DblToFlt.B0, DblToFlt.B1)),
               (v4f32 (VMRGEW MrgFP.ABhToFlt, MrgFP.ABlToFlt))>;
+
+    // Convert 4 doubles to a vector of ints.
+    def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
+                                   DblToInt.C, DblToInt.D)),
+              (v4i32 (VMRGEW MrgWords.CVACS, MrgWords.CVBDS))>;
+    def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
+                                   DblToUInt.C, DblToUInt.D)),
+              (v4i32 (VMRGEW MrgWords.CVACU, MrgWords.CVBDU))>;
+    def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
+                                   ExtDbl.B0S, ExtDbl.B1S)),
+              (v4i32 (VMRGEW MrgWords.CVA0B0S, MrgWords.CVA1B1S))>;
+    def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
+                                   ExtDbl.B0U, ExtDbl.B1U)),
+              (v4i32 (VMRGEW MrgWords.CVA0B0U, MrgWords.CVA1B1U))>;
   }
 
   let Predicates = [IsLittleEndian, HasVSX] in {
@@ -3300,6 +3372,20 @@ let AddedComplexity = 400 in {
     def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
                                    DblToFlt.B0, DblToFlt.B1)),
               (v4f32 (VMRGEW MrgFP.BAhToFlt, MrgFP.BAlToFlt))>;
+
+    // Convert 4 doubles to a vector of ints.
+    def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
+                                   DblToInt.C, DblToInt.D)),
+              (v4i32 (VMRGEW MrgWords.CVDBS, MrgWords.CVCAS))>;
+    def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
+                                   DblToUInt.C, DblToUInt.D)),
+              (v4i32 (VMRGEW MrgWords.CVDBU, MrgWords.CVCAU))>;
+    def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
+                                   ExtDbl.B0S, ExtDbl.B1S)),
+              (v4i32 (VMRGEW MrgWords.CVB1A1S, MrgWords.CVB0A0S))>;
+    def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
+                                   ExtDbl.B0U, ExtDbl.B1U)),
+              (v4i32 (VMRGEW MrgWords.CVB1A1U, MrgWords.CVB0A0U))>;
   }
 
   let Predicates = [HasDirectMove] in {
diff --git a/test/CodeGen/PowerPC/build-vector-tests.ll b/test/CodeGen/PowerPC/build-vector-tests.ll
index 16b562b..3785b2a 100644
--- a/test/CodeGen/PowerPC/build-vector-tests.ll
+++ b/test/CodeGen/PowerPC/build-vector-tests.ll
@@ -119,8 +119,8 @@
 ;vector int spltCnstConvftoi() {                                              //
 ;  return (vector int) 4.74f;                                                 //
 ;}                                                                            //
-;// P8: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws                         //
-;// P9: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvdpsxws                         //
+;// P8: 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                                   //
+;// P9: 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                                   //
 ;vector int fromRegsConvftoi(float a, float b, float c, float d) {            //
 ;  return (vector int) { a, b, c, d };                                        //
 ;}                                                                            //
@@ -139,15 +139,15 @@
 ;vector int fromDiffMemConsDConvftoi(float *ptr) {                            //
 ;  return (vector int) { ptr[3], ptr[2], ptr[1], ptr[0] };                    //
 ;}                                                                            //
-;// P8: 4 x lxsspx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws             //
-;// P9: 4 x lxssp, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws              //
+;// P8: 4 x lxsspx, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                       //
+;// P9: 4 x lxssp, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                        //
 ;// Note: if the consecutive loads learns to handle pre-inc, this can be:     //
 ;//       sldi 2, load, xvcvspuxws                                            //
 ;vector int fromDiffMemVarAConvftoi(float *arr, int elem) {                   //
 ;  return (vector int) { arr[elem], arr[elem+1], arr[elem+2], arr[elem+3] };  //
 ;}                                                                            //
-;// P8: 4 x lxsspx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws             //
-;// P9: 4 x lxssp, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws              //
+;// P8: 4 x lxsspx, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                       //
+;// P9: 4 x lxssp, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                        //
 ;// Note: if the consecutive loads learns to handle pre-inc, this can be:     //
 ;//       sldi 2, 2 x load, vperm, xvcvspuxws                                 //
 ;vector int fromDiffMemVarDConvftoi(float *arr, int elem) {                   //
@@ -168,8 +168,8 @@
 ;vector int spltCnstConvdtoi() {                                              //
 ;  return (vector int) 4.74;                                                  //
 ;}                                                                            //
-;// P8: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws                         //
-;// P9: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws                         //
+;// P8: 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                                   //
+;// P9: 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                                   //
 ;vector int fromRegsConvdtoi(double a, double b, double c, double d) {        //
 ;  return (vector int) { a, b, c, d };                                        //
 ;}                                                                            //
@@ -178,25 +178,23 @@
 ;vector int fromDiffConstsConvdtoi() {                                        //
 ;  return (vector int) { 24.46, 234., 988.19, 422.39 };                       //
 ;}                                                                            //
-;// P8: 2 x lxvd2x, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvdpsp, vmrgew,      //
-;//     xvcvspsxws                                                            //
-;// P9: 2 x lxvx, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvdpsp, vmrgew,        //
-;//     xvcvspsxws                                                            //
+;// P8: 2 x lxvd2x, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvspsxws, vmrgew     //
+;// P9: 2 x lxvx, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvspsxws, vmrgew       //
 ;vector int fromDiffMemConsAConvdtoi(double *ptr) {                           //
 ;  return (vector int) { ptr[0], ptr[1], ptr[2], ptr[3] };                    //
 ;}                                                                            //
-;// P8: 4 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws              //
-;// P9: 4 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws                //
+;// P8: 4 x lxsdx, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                        //
+;// P9: 4 x lfd, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                          //
 ;vector int fromDiffMemConsDConvdtoi(double *ptr) {                           //
 ;  return (vector int) { ptr[3], ptr[2], ptr[1], ptr[0] };                    //
 ;}                                                                            //
-;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws       //
-;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws         //
+;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                 //
+;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                   //
 ;vector int fromDiffMemVarAConvdtoi(double *arr, int elem) {                  //
 ;  return (vector int) { arr[elem], arr[elem+1], arr[elem+2], arr[elem+3] };  //
 ;}                                                                            //
-;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws       //
-;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspsxws         //
+;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                 //
+;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvspsxws, vmrgew                   //
 ;vector int fromDiffMemVarDConvdtoi(double *arr, int elem) {                  //
 ;  return (vector int) { arr[elem], arr[elem-1], arr[elem-2], arr[elem-3] };  //
 ;}                                                                            //
@@ -296,8 +294,8 @@
 ;vector unsigned int spltCnstConvftoui() {                                    //
 ;  return (vector unsigned int) 4.74f;                                        //
 ;}                                                                            //
-;// P8: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws                         //
-;// P9: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws                         //
+;// P8: 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                                   //
+;// P9: 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                                   //
 ;vector unsigned int fromRegsConvftoui(float a, float b, float c, float d) {  //
 ;  return (vector unsigned int) { a, b, c, d };                               //
 ;}                                                                            //
@@ -316,16 +314,16 @@
 ;vector unsigned int fromDiffMemConsDConvftoui(float *ptr) {                  //
 ;  return (vector unsigned int) { ptr[3], ptr[2], ptr[1], ptr[0] };           //
 ;}                                                                            //
-;// P8: lfsux, 3 x lxsspx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws      //
-;// P9: lfsux, 3 x lfs, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws         //
+;// P8: lfsux, 3 x lxsspx, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                //
+;// P9: lfsux, 3 x lfs, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                   //
 ;// Note: if the consecutive loads learns to handle pre-inc, this can be:     //
 ;//       sldi 2, load, xvcvspuxws                                            //
 ;vector unsigned int fromDiffMemVarAConvftoui(float *arr, int elem) {         //
 ;  return (vector unsigned int) { arr[elem], arr[elem+1],                     //
 ;                                 arr[elem+2], arr[elem+3] };                 //
 ;}                                                                            //
-;// P8: lfsux, 3 x lxsspx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws      //
-;// P9: lfsux, 3 x lfs, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws         //
+;// P8: lfsux, 3 x lxsspx, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                //
+;// P9: lfsux, 3 x lfs, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                   //
 ;// Note: if the consecutive loads learns to handle pre-inc, this can be:     //
 ;//       sldi 2, 2 x load, vperm, xvcvspuxws                                 //
 ;vector unsigned int fromDiffMemVarDConvftoui(float *arr, int elem) {         //
@@ -347,8 +345,8 @@
 ;vector unsigned int spltCnstConvdtoui() {                                    //
 ;  return (vector unsigned int) 4.74;                                         //
 ;}                                                                            //
-;// P8: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws                         //
-;// P9: 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws                         //
+;// P8: 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                                   //
+;// P9: 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                                   //
 ;vector unsigned int fromRegsConvdtoui(double a, double b,                    //
 ;                                      double c, double d) {                  //
 ;  return (vector unsigned int) { a, b, c, d };                               //
@@ -358,25 +356,24 @@
 ;vector unsigned int fromDiffConstsConvdtoui() {                              //
 ;  return (vector unsigned int) { 24.46, 234., 988.19, 422.39 };              //
 ;}                                                                            //
-;// P8: 2 x lxvd2x, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvdpsp, vmrgew,      //
-;//     xvcvspuxws                                                            //
-;// P9: 2 x lxvx, xxmrgld, xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws          //
+;// P8: 2 x lxvd2x, 2 x xxswapd, xxmrgld, xxmrghd, 2 x xvcvspuxws, vmrgew     //
+;// P9: 2 x lxvx, xxmrgld, xxmrghd, 2 x xvcvspuxws, vmrgew                    //
 ;vector unsigned int fromDiffMemConsAConvdtoui(double *ptr) {                 //
 ;  return (vector unsigned int) { ptr[0], ptr[1], ptr[2], ptr[3] };           //
 ;}                                                                            //
-;// P8: 4 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws              //
-;// P9: 4 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws                //
+;// P8: 4 x lxsdx, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                        //
+;// P9: 4 x lfd, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                          //
 ;vector unsigned int fromDiffMemConsDConvdtoui(double *ptr) {                 //
 ;  return (vector unsigned int) { ptr[3], ptr[2], ptr[1], ptr[0] };           //
 ;}                                                                            //
-;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws       //
-;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws         //
+;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                 //
+;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                   //
 ;vector unsigned int fromDiffMemVarAConvdtoui(double *arr, int elem) {        //
 ;  return (vector unsigned int) { arr[elem], arr[elem+1],                     //
 ;                                 arr[elem+2], arr[elem+3] };                 //
 ;}                                                                            //
-;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws       //
-;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvdpsp, vmrgew, xvcvspuxws         //
+;// P8: lfdux, 3 x lxsdx, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                 //
+;// P9: lfdux, 3 x lfd, 2 x xxmrghd, 2 x xvcvspuxws, vmrgew                   //
 ;vector unsigned int fromDiffMemVarDConvdtoui(double *arr, int elem) {        //
 ;  return (vector unsigned int) { arr[elem], arr[elem-1],                     //
 ;                                 arr[elem-2], arr[elem-3] };                 //
@@ -1253,28 +1250,24 @@ entry:
 ; P8LE-LABEL: fromRegsConvftoi
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P9BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9BE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9BE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9BE: vmrgew v2, [[REG3]], [[REG4]]
-; P9BE: xvcvspsxws v2, v2
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P9LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9LE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9LE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9LE: vmrgew v2, [[REG4]], [[REG3]]
-; P9LE: xvcvspsxws v2, v2
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P8BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8BE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8BE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8BE: vmrgew v2, [[REG3]], [[REG4]]
-; P8BE: xvcvspsxws v2, v2
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P8LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8LE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8LE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8LE: vmrgew v2, [[REG4]], [[REG3]]
-; P8LE: xvcvspsxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
@@ -1529,28 +1522,24 @@ entry:
 ; P8LE-LABEL: fromRegsConvdtoi
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P9BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9BE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9BE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9BE: vmrgew v2, [[REG3]], [[REG4]]
-; P9BE: xvcvspsxws v2, v2
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P9LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9LE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9LE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9LE: vmrgew v2, [[REG4]], [[REG3]]
-; P9LE: xvcvspsxws v2, v2
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P8BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8BE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8BE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8BE: vmrgew v2, [[REG3]], [[REG4]]
-; P8BE: xvcvspsxws v2, v2
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P8LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8LE-DAG: xvcvdpsxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8LE-DAG: xvcvdpsxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8LE: vmrgew v2, [[REG4]], [[REG3]]
-; P8LE: xvcvspsxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
@@ -1592,36 +1581,32 @@ entry:
 ; P9BE-DAG: lxv [[REG2:[vs0-9]+]], 16(r3)
 ; P9BE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG1]], [[REG2]]
 ; P9BE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG1]], [[REG2]]
-; P9BE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P9BE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P9BE-DAG: xvcvdpsxws [[REG5:[vs0-9]+]], [[REG3]]
+; P9BE-DAG: xvcvdpsxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P9BE: vmrgew v2, [[REG6]], [[REG5]]
-; P9BE: xvcvspsxws v2, v2
 ; P9LE-DAG: lxv [[REG1:[vs0-9]+]], 0(r3)
 ; P9LE-DAG: lxv [[REG2:[vs0-9]+]], 16(r3)
 ; P9LE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG2]], [[REG1]]
 ; P9LE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG2]], [[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P9LE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P9LE-DAG: xvcvdpsxws [[REG5:[vs0-9]+]], [[REG3]]
+; P9LE-DAG: xvcvdpsxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P9LE: vmrgew v2, [[REG6]], [[REG5]]
-; P9LE: xvcvspsxws v2, v2
 ; P8BE: lxvd2x [[REG1:[vs0-9]+]], 0, r3
 ; P8BE: lxvd2x [[REG2:[vs0-9]+]], r3, r4
 ; P8BE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG1]], [[REG2]]
 ; P8BE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG1]], [[REG2]]
-; P8BE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P8BE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P8BE-DAG: xvcvdpsxws [[REG5:[vs0-9]+]], [[REG3]]
+; P8BE-DAG: xvcvdpsxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P8BE: vmrgew v2, [[REG6]], [[REG5]]
-; P8BE: xvcvspsxws v2, v2
 ; P8LE: lxvd2x [[REG1:[vs0-9]+]], 0, r3
 ; P8LE: lxvd2x [[REG2:[vs0-9]+]], r3, r4
 ; P8LE-DAG: xxswapd [[REG3:[vs0-9]+]], [[REG1]]
 ; P8LE-DAG: xxswapd [[REG4:[vs0-9]+]], [[REG2]]
 ; P8LE-DAG: xxmrgld [[REG5:[vs0-9]+]], [[REG4]], [[REG3]]
 ; P8LE-DAG: xxmrghd [[REG6:[vs0-9]+]], [[REG4]], [[REG3]]
-; P8LE-DAG: xvcvdpsp [[REG7:[vs0-9]+]], [[REG5]]
-; P8LE-DAG: xvcvdpsp [[REG8:[vs0-9]+]], [[REG6]]
+; P8LE-DAG: xvcvdpsxws [[REG7:[vs0-9]+]], [[REG5]]
+; P8LE-DAG: xvcvdpsxws [[REG8:[vs0-9]+]], [[REG6]]
 ; P8LE: vmrgew v2, [[REG8]], [[REG7]]
-; P8LE: xvcvspsxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -1653,40 +1638,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspsxws v2
+; P9BE: xvcvdpsxws
+; P9BE: xvcvdpsxws
+; P9BE: vmrgew v2
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspsxws v2
+; P9LE: xvcvdpsxws
+; P9LE: xvcvdpsxws
+; P9LE: vmrgew v2
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspsxws v2
+; P8BE: xvcvdpsxws
+; P8BE: xvcvdpsxws
+; P8BE: vmrgew v2
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspsxws v2
+; P8LE: xvcvdpsxws
+; P8LE: xvcvdpsxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -1726,40 +1707,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspsxws v2
+; P9BE: xvcvdpsxws
+; P9BE: xvcvdpsxws
+; P9BE: vmrgew v2
 ; P9LE: lfdux
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspsxws v2
+; P9LE: xvcvdpsxws
+; P9LE: xvcvdpsxws
+; P9LE: vmrgew v2
 ; P8BE: lfdux
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspsxws v2
+; P8BE: xvcvdpsxws
+; P8BE: xvcvdpsxws
+; P8BE: vmrgew v2
 ; P8LE: lfdux
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspsxws v2
+; P8LE: xvcvdpsxws
+; P8LE: xvcvdpsxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -1799,40 +1776,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspsxws v2
+; P9BE: xvcvdpsxws
+; P9BE: xvcvdpsxws
+; P9BE: vmrgew v2
 ; P9LE: lfdux
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspsxws v2
+; P9LE: xvcvdpsxws
+; P9LE: xvcvdpsxws
+; P9LE: vmrgew v2
 ; P8BE: lfdux
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspsxws v2
+; P8BE: xvcvdpsxws
+; P8BE: xvcvdpsxws
+; P8BE: vmrgew v2
 ; P8LE: lfdux
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspsxws v2
+; P8LE: xvcvdpsxws
+; P8LE: xvcvdpsxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
@@ -2413,28 +2386,24 @@ entry:
 ; P8LE-LABEL: fromRegsConvftoui
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P9BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9BE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9BE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9BE: vmrgew v2, [[REG3]], [[REG4]]
-; P9BE: xvcvspuxws v2, v2
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P9LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9LE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9LE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9LE: vmrgew v2, [[REG4]], [[REG3]]
-; P9LE: xvcvspuxws v2, v2
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P8BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8BE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8BE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8BE: vmrgew v2, [[REG3]], [[REG4]]
-; P8BE: xvcvspuxws v2, v2
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P8LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8LE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8LE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8LE: vmrgew v2, [[REG4]], [[REG3]]
-; P8LE: xvcvspuxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
@@ -2689,28 +2658,24 @@ entry:
 ; P8LE-LABEL: fromRegsConvdtoui
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P9BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P9BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9BE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9BE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9BE: vmrgew v2, [[REG3]], [[REG4]]
-; P9BE: xvcvspuxws v2, v2
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P9LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P9LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P9LE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P9LE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P9LE: vmrgew v2, [[REG4]], [[REG3]]
-; P9LE: xvcvspuxws v2, v2
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs1, vs3
 ; P8BE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs2, vs4
-; P8BE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8BE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8BE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8BE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8BE: vmrgew v2, [[REG3]], [[REG4]]
-; P8BE: xvcvspuxws v2, v2
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG1:[0-9]+]], vs3, vs1
 ; P8LE-DAG: xxmrghd {{[vs]+}}[[REG2:[0-9]+]], vs4, vs2
-; P8LE-DAG: xvcvdpsp [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
-; P8LE-DAG: xvcvdpsp [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
+; P8LE-DAG: xvcvdpuxws [[REG3:v[0-9]+]], {{[vs]+}}[[REG1]]
+; P8LE-DAG: xvcvdpuxws [[REG4:v[0-9]+]], {{[vs]+}}[[REG2]]
 ; P8LE: vmrgew v2, [[REG4]], [[REG3]]
-; P8LE: xvcvspuxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
@@ -2752,36 +2717,32 @@ entry:
 ; P9BE-DAG: lxv [[REG2:[vs0-9]+]], 16(r3)
 ; P9BE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG1]], [[REG2]]
 ; P9BE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG1]], [[REG2]]
-; P9BE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P9BE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P9BE-DAG: xvcvdpuxws [[REG5:[vs0-9]+]], [[REG3]]
+; P9BE-DAG: xvcvdpuxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P9BE: vmrgew v2, [[REG6]], [[REG5]]
-; P9BE: xvcvspuxws v2, v2
 ; P9LE-DAG: lxv [[REG1:[vs0-9]+]], 0(r3)
 ; P9LE-DAG: lxv [[REG2:[vs0-9]+]], 16(r3)
-; P9LE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG2]], [[REG1]]
 ; P9LE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG2]], [[REG1]]
-; P9LE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P9LE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P9LE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG2]], [[REG1]]
+; P9LE-DAG: xvcvdpuxws [[REG5:[vs0-9]+]], [[REG3]]
+; P9LE-DAG: xvcvdpuxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P9LE: vmrgew v2, [[REG6]], [[REG5]]
-; P9LE: xvcvspuxws v2, v2
 ; P8BE: lxvd2x [[REG1:[vs0-9]+]], 0, r3
 ; P8BE: lxvd2x [[REG2:[vs0-9]+]], r3, r4
 ; P8BE-DAG: xxmrgld [[REG3:[vs0-9]+]], [[REG1]], [[REG2]]
 ; P8BE-DAG: xxmrghd [[REG4:[vs0-9]+]], [[REG1]], [[REG2]]
-; P8BE-DAG: xvcvdpsp [[REG5:[vs0-9]+]], [[REG3]]
-; P8BE-DAG: xvcvdpsp [[REG6:[vs0-9]+]], [[REG4]]
+; P8BE-DAG: xvcvdpuxws [[REG5:[vs0-9]+]], [[REG3]]
+; P8BE-DAG: xvcvdpuxws [[REG6:[vs0-9]+]], [[REG4]]
 ; P8BE: vmrgew v2, [[REG6]], [[REG5]]
-; P8BE: xvcvspuxws v2, v2
 ; P8LE: lxvd2x [[REG1:[vs0-9]+]], 0, r3
 ; P8LE: lxvd2x [[REG2:[vs0-9]+]], r3, r4
 ; P8LE-DAG: xxswapd [[REG3:[vs0-9]+]], [[REG1]]
 ; P8LE-DAG: xxswapd [[REG4:[vs0-9]+]], [[REG2]]
 ; P8LE-DAG: xxmrgld [[REG5:[vs0-9]+]], [[REG4]], [[REG3]]
 ; P8LE-DAG: xxmrghd [[REG6:[vs0-9]+]], [[REG4]], [[REG3]]
-; P8LE-DAG: xvcvdpsp [[REG7:[vs0-9]+]], [[REG5]]
-; P8LE-DAG: xvcvdpsp [[REG8:[vs0-9]+]], [[REG6]]
+; P8LE-DAG: xvcvdpuxws [[REG7:[vs0-9]+]], [[REG5]]
+; P8LE-DAG: xvcvdpuxws [[REG8:[vs0-9]+]], [[REG6]]
 ; P8LE: vmrgew v2, [[REG8]], [[REG7]]
-; P8LE: xvcvspuxws v2, v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -2813,40 +2774,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspuxws v2
+; P9BE: xvcvdpuxws
+; P9BE: xvcvdpuxws
+; P9BE: vmrgew v2
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspuxws v2
+; P9LE: xvcvdpuxws
+; P9LE: xvcvdpuxws
+; P9LE: vmrgew v2
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspuxws v2
+; P8BE: xvcvdpuxws
+; P8BE: xvcvdpuxws
+; P8BE: vmrgew v2
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspuxws v2
+; P8LE: xvcvdpuxws
+; P8LE: xvcvdpuxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -2886,40 +2843,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspuxws v2
+; P9BE: xvcvdpuxws
+; P9BE: xvcvdpuxws
+; P9BE: vmrgew v2
 ; P9LE: lfdux
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspuxws v2
+; P9LE: xvcvdpuxws
+; P9LE: xvcvdpuxws
+; P9LE: vmrgew v2
 ; P8BE: lfdux
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspuxws v2
+; P8BE: xvcvdpuxws
+; P8BE: xvcvdpuxws
+; P8BE: vmrgew v2
 ; P8LE: lfdux
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspuxws v2
+; P8LE: xvcvdpuxws
+; P8LE: xvcvdpuxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readonly
@@ -2959,40 +2912,36 @@ entry:
 ; P9BE: lfd
 ; P9BE: xxmrghd
 ; P9BE: xxmrghd
-; P9BE: xvcvdpsp
-; P9BE: xvcvdpsp
-; P9BE: vmrgew
-; P9BE: xvcvspuxws v2
+; P9BE: xvcvdpuxws
+; P9BE: xvcvdpuxws
+; P9BE: vmrgew v2
 ; P9LE: lfdux
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: lfd
 ; P9LE: xxmrghd
 ; P9LE: xxmrghd
-; P9LE: xvcvdpsp
-; P9LE: xvcvdpsp
-; P9LE: vmrgew
-; P9LE: xvcvspuxws v2
+; P9LE: xvcvdpuxws
+; P9LE: xvcvdpuxws
+; P9LE: vmrgew v2
 ; P8BE: lfdux
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: lxsdx
 ; P8BE: xxmrghd
 ; P8BE: xxmrghd
-; P8BE: xvcvdpsp
-; P8BE: xvcvdpsp
-; P8BE: vmrgew
-; P8BE: xvcvspuxws v2
+; P8BE: xvcvdpuxws
+; P8BE: xvcvdpuxws
+; P8BE: vmrgew v2
 ; P8LE: lfdux
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: lxsdx
 ; P8LE: xxmrghd
 ; P8LE: xxmrghd
-; P8LE: xvcvdpsp
-; P8LE: xvcvdpsp
-; P8LE: vmrgew
-; P8LE: xvcvspuxws v2
+; P8LE: xvcvdpuxws
+; P8LE: xvcvdpuxws
+; P8LE: vmrgew v2
 }
 
 ; Function Attrs: norecurse nounwind readnone
-- 
1.8.3.1