[SystemZ] Replicate the load with most uses in buildVector()

[jonpa]

Summary

Iterate over all elements and count the number of uses among them for each used load. Then make sure to REPLICATE the load which has the most uses in order to minimize the number of needed element insertions.

[jonpa]

This only affects one file on SPEC:

update.s

vlvgf          :                   18                   12       -6
vrepf          :                    8                    5       -3
vlrepf         :                    3                    6       +3
l              :                  163                  160       -3
vlvgp          :                    6                    3       -3
lg             :                  320                  319       -1
Spill|Reload   :                  478                  477       -1

[uweigand]

LGTM, thanks!

[jonpa]

r346637.

lib/Target/SystemZ/SystemZISelLowering.cpp

for (unsigned I = 0; I < NumElements; ++I)
       if (!Constants[I].getNode())
         Constants[I] = DAG.getUNDEF(Elems[I].getValueType());
     Result = DAG.getBuildVector(VT, DL, Constants);
   } else {
     // Otherwise try to use VLREP or VLVGP to start the sequence in order to
     // avoid a false dependency on any previous contents of the vector
     // register.
 
-    // Use a VLREP if at least one element is a load.
-    unsigned LoadElIdx = UINT_MAX;
+    // Use a VLREP if at least one element is a load. Make sure to replicate
+    // the load with the most elements having its value.
+    std::map<const SDNode*, unsigned> UseCounts;
+    SDNode *LoadMaxUses = nullptr;
     for (unsigned I = 0; I < NumElements; ++I)
       if (Elems[I].getOpcode() == ISD::LOAD &&
           cast<LoadSDNode>(Elems[I])->isUnindexed()) {
-        LoadElIdx = I;
-        break;
-      }
-    if (LoadElIdx != UINT_MAX) {
-      Result = DAG.getNode(SystemZISD::REPLICATE, DL, VT, Elems[LoadElIdx]);
-      Done[LoadElIdx] = true;
-      ReplicatedVal = Elems[LoadElIdx];
+        SDNode *Ld = Elems[I].getNode();
+        UseCounts[Ld]++;
+        if (LoadMaxUses == nullptr || UseCounts[LoadMaxUses] < UseCounts[Ld])
+          LoadMaxUses = Ld;
+      }
+    if (LoadMaxUses != nullptr) {
+      ReplicatedVal = SDValue(LoadMaxUses, 0);
+      Result = DAG.getNode(SystemZISD::REPLICATE, DL, VT, ReplicatedVal);
     } else {
       // Try to use VLVGP.
       unsigned I1 = NumElements / 2 - 1;
       unsigned I2 = NumElements - 1;
       bool Def1 = !Elems[I1].isUndef();
       bool Def2 = !Elems[I2].isUndef();
       if (Def1 || Def2) {
         SDValue Elem1 = Elems[Def1 ? I1 : I2];

test/CodeGen/SystemZ/vec-move-20.ll

+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
+
+; Test a vector which is built with elements from two loads replicates the
+; load with most elements having its value.
+
+; CHECK:      vlef
+; CHECK-NOT:  vlvgf
+
+define void @update(i32* %src1, i32* %src2, <4 x i32>* %dst) {
+bb:
+  %tmp = load i32, i32* %src1
+  %tmp1 = load i32, i32* %src2
+  %tmp2 = insertelement <4 x i32> undef, i32 %tmp, i32 0
+  %tmp3 = insertelement <4 x i32> %tmp2, i32 %tmp1, i32 1
+  %tmp4 = insertelement <4 x i32> %tmp3, i32 %tmp1, i32 2
+  %tmp5 = insertelement <4 x i32> %tmp4, i32 %tmp1, i32 3
+  store <4 x i32> %tmp5, <4 x i32>* %dst
+  ret void
+}