[PowerPC] Improve BUILD_VECTOR of 4 i32s

[lei]

Summary

Currently, for this node:

vector int test(int a, int b, int c, int d) {
  return (vector int) { a, b, c, d };
}

we get this on Power9:

mtvsrdd 34, 5, 3
mtvsrdd 35, 6, 4
vmrgow 2, 3, 2

and this on Power8:

mtvsrwz 0, 3
mtvsrwz 1, 5
mtvsrwz 2, 4
mtvsrwz 3, 6
xxmrghd 34, 1, 0
xxmrghd 35, 3, 2
vmrgow 2, 3, 2

This can be improved to this on LE Power9:

rldimi 3, 4, 32, 0
rldimi 5, 6, 32, 0
mtvsrdd 34, 5, 3

and this on LE Power8

rldimi 3, 4, 32, 0
rldimi 5, 6, 32, 0
mtvsrd 34, 3
mtvsrd 35, 5
xxpermdi 34, 35, 34, 0

This patch updates the TD pattern to generate the optimized sequence for both Power8 and Power9 on LE and BE.

[nemanjai]

LGTM.

llvm/lib/Target/PowerPC/PPCInstrVSX.td

let AddedComplexity = 400 in {
 
   let Predicates = [IsBigEndian, HasDirectMove, NoP9Vector] in {
     // Big endian integer vectors using direct moves.
     def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
               (v2i64 (XXPERMDI
                         (COPY_TO_REGCLASS (MTVSRD $A), VSRC),
                         (COPY_TO_REGCLASS (MTVSRD $B), VSRC), 0))>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
-              (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC),
-                                   (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC), 0),
-                      (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC),
-                                   (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC), 0))>;
+              (XXPERMDI
+                (COPY_TO_REGCLASS
+                  (MTVSRD (RLDIMI AnyExts.B, AnyExts.A, 32, 0)), VSRC),
+                (COPY_TO_REGCLASS
+                  (MTVSRD (RLDIMI AnyExts.D, AnyExts.C, 32, 0)), VSRC), 0)>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
               (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
   }
 
   let Predicates = [IsLittleEndian, HasDirectMove, NoP9Vector] in {
     // Little endian integer vectors using direct moves.
     def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
               (v2i64 (XXPERMDI
                         (COPY_TO_REGCLASS (MTVSRD $B), VSRC),
                         (COPY_TO_REGCLASS (MTVSRD $A), VSRC), 0))>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
-              (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC),
-                                   (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC), 0),
-                      (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC),
-                                   (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 0))>;
+              (XXPERMDI
+                (COPY_TO_REGCLASS
+                  (MTVSRD (RLDIMI AnyExts.C, AnyExts.D, 32, 0)), VSRC),
+                (COPY_TO_REGCLASS
+                  (MTVSRD (RLDIMI AnyExts.A, AnyExts.B, 32, 0)), VSRC), 0)>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
               (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
   }
 
   let Predicates = [HasP9Vector] in {
     // Endianness-neutral patterns for const splats with ISA 3.0 instructions.
     def : Pat<(v4i32 (scalar_to_vector i32:$A)),
               (v4i32 (MTVSRWS $A))>;
let AddedComplexity = 400 in {
 
   let Predicates = [IsISA3_0, HasDirectMove, IsBigEndian] in {
     def : Pat<(i64 (extractelt v2i64:$A, 1)),
               (i64 (MFVSRLD $A))>;
     // Better way to build integer vectors if we have MTVSRDD. Big endian.
     def : Pat<(v2i64 (build_vector i64:$rB, i64:$rA)),
               (v2i64 (MTVSRDD $rB, $rA))>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
-              (VMRGOW
-                (v4i32 (COPY_TO_REGCLASS (MTVSRDD AnyExts.A, AnyExts.C), VSRC)),
-                (v4i32
-                  (COPY_TO_REGCLASS (MTVSRDD AnyExts.B, AnyExts.D), VSRC)))>;
+              (MTVSRDD
+                (RLDIMI AnyExts.B, AnyExts.A, 32, 0),
+                (RLDIMI AnyExts.D, AnyExts.C, 32, 0))>;
   }
 
   let Predicates = [IsISA3_0, HasDirectMove, IsLittleEndian] in {
     def : Pat<(i64 (extractelt v2i64:$A, 0)),
               (i64 (MFVSRLD $A))>;
     // Better way to build integer vectors if we have MTVSRDD. Little endian.
     def : Pat<(v2i64 (build_vector i64:$rA, i64:$rB)),
               (v2i64 (MTVSRDD $rB, $rA))>;
     def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
-              (VMRGOW
-                (v4i32 (COPY_TO_REGCLASS (MTVSRDD AnyExts.D, AnyExts.B), VSRC)),
-                (v4i32
-                  (COPY_TO_REGCLASS (MTVSRDD AnyExts.C, AnyExts.A), VSRC)))>;
+              (MTVSRDD
+                (RLDIMI AnyExts.C, AnyExts.D, 32, 0),
+                (RLDIMI AnyExts.A, AnyExts.B, 32, 0))>;
   }
   // P9 Altivec instructions that can be used to build vectors.
   // Adding them to PPCInstrVSX.td rather than PPCAltivecVSX.td to compete
   // with complexities of existing build vector patterns in this file.
   let Predicates = [HasP9Altivec, IsLittleEndian] in {
     def : Pat<(v2i64 (build_vector WordToDWord.LE_A0, WordToDWord.LE_A1)),
               (v2i64 (VEXTSW2D $A))>;
     def : Pat<(v2i64 (build_vector HWordToDWord.LE_A0, HWordToDWord.LE_A1)),

llvm/test/CodeGen/PowerPC/build-vector-tests.ll

entry:
   %vecinit1 = insertelement <4 x i32> %vecinit, i32 %b, i32 1
   %vecinit2 = insertelement <4 x i32> %vecinit1, i32 %c, i32 2
   %vecinit3 = insertelement <4 x i32> %vecinit2, i32 %d, i32 3
   ret <4 x i32> %vecinit3
 ; P9BE-LABEL: fromRegsi
 ; P9LE-LABEL: fromRegsi
 ; P8BE-LABEL: fromRegsi
 ; P8LE-LABEL: fromRegsi
-; P9BE-DAG: mtvsrdd [[REG1:v[0-9]+]], r3, r5
-; P9BE-DAG: mtvsrdd [[REG2:v[0-9]+]], r4, r6
-; P9BE: vmrgow v2, [[REG1]], [[REG2]]
-; P9BE: blr
-; P9LE-DAG: mtvsrdd [[REG1:v[0-9]+]], r5, r3
-; P9LE-DAG: mtvsrdd [[REG2:v[0-9]+]], r6, r4
-; P9LE: vmrgow v2, [[REG2]], [[REG1]]
-; P9LE: blr
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG1:[0-9]+]], r3
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG2:[0-9]+]], r4
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG3:[0-9]+]], r5
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG4:[0-9]+]], r6
-; P8BE-DAG: xxmrghd [[REG5:v[0-9]+]], {{[v][s]*}}[[REG1]], {{[v][s]*}}[[REG3]]
-; P8BE-DAG: xxmrghd [[REG6:v[0-9]+]], {{[v][s]*}}[[REG2]], {{[v][s]*}}[[REG4]]
-; P8BE: vmrgow v2, [[REG5]], [[REG6]]
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG1:[0-9]+]], r3
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG2:[0-9]+]], r4
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG3:[0-9]+]], r5
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG4:[0-9]+]], r6
-; P8LE: xxmrghd [[REG5:v[0-9]+]], {{[v][s]*}}[[REG3]], {{[v][s]*}}[[REG1]]
-; P8LE: xxmrghd [[REG6:v[0-9]+]], {{[v][s]*}}[[REG4]], {{[v][s]*}}[[REG2]]
-; P8LE: vmrgow v2, [[REG6]], [[REG5]]
+; P9BE-DAG: rldimi r6, r5, 32, 0
+; P9BE-DAG: rldimi r4, r3, 32, 0
+; P9BE: mtvsrdd v2, r4, r6
+; P9BE: blr
+; P9LE-DAG: rldimi r3, r4, 32, 0
+; P9LE-DAG: rldimi r5, r6, 32, 0
+; P9LE: mtvsrdd v2, r5, r3
+; P9LE: blr
+; P8BE-DAG: rldimi r6, r5, 32, 0
+; P8BE-DAG: rldimi r4, r3, 32, 0
+; P8BE-DAG: mtvsrd f[[REG1:[0-9]+]], r6
+; P8BE-DAG: mtvsrd f[[REG2:[0-9]+]], r4
+; P8BE-DAG: xxmrghd v2, vs[[REG2]], vs[[REG1]]
+; P8BE: blr
+; P8LE-DAG: rldimi r3, r4, 32, 0
+; P8LE-DAG: rldimi r5, r6, 32, 0
+; P8LE-DAG: mtvsrd f[[REG1:[0-9]+]], r3
+; P8LE-DAG: mtvsrd f[[REG2:[0-9]+]], r5
+; P8LE-DAG: xxmrghd v2, vs[[REG2]], vs[[REG1]]
+; P8LE: blr
 }
 
 ; Function Attrs: norecurse nounwind readnone
 define <4 x i32> @fromDiffConstsi() {
 entry:
   ret <4 x i32> <i32 242, i32 -113, i32 889, i32 19>
 ; P9BE-LABEL: fromDiffConstsi
 ; P9LE-LABEL: fromDiffConstsi

 ; P9BE-LABEL: fromRandMemConsi
 ; P9LE-LABEL: fromRandMemConsi
 ; P8BE-LABEL: fromRandMemConsi
 ; P8LE-LABEL: fromRandMemConsi
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
+; P9BE: rldimi
+; P9BE: rldimi
 ; P9BE: mtvsrdd
-; P9BE: mtvsrdd
-; P9BE: vmrgow
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
+; P9LE: rldimi
+; P9LE: rldimi
 ; P9LE: mtvsrdd
-; P9LE: mtvsrdd
-; P9LE: vmrgow
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: xxmrghd
+; P8BE: rldimi
+; P8BE: rldimi
+; P8BE: mtvsrd
+; P8BE: mtvsrd
 ; P8BE: xxmrghd
-; P8BE: vmrgow
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: xxmrghd
+; P8LE: rldimi
+; P8LE: rldimi
+; P8LE: mtvsrd
+; P8LE: mtvsrd
 ; P8LE: xxmrghd
-; P8LE: vmrgow
 }
 
 ; Function Attrs: norecurse nounwind readonly
 define <4 x i32> @fromRandMemVari(i32* nocapture readonly %arr, i32 signext %elem) {
 entry:
   %add = add nsw i32 %elem, 4
   %idxprom = sext i32 %add to i64
   %arrayidx = getelementptr inbounds i32, i32* %arr, i64 %idxprom

 ; P9LE-LABEL: fromRandMemVari
 ; P8BE-LABEL: fromRandMemVari
 ; P8LE-LABEL: fromRandMemVari
 ; P9BE: sldi r4, r4, 2
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
+; P9BE: rldimi
+; P9BE: rldimi
 ; P9BE: mtvsrdd
-; P9BE: mtvsrdd
-; P9BE: vmrgow
 ; P9LE: sldi r4, r4, 2
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
+; P9LE: rldimi
+; P9LE: rldimi
 ; P9LE: mtvsrdd
-; P9LE: mtvsrdd
-; P9LE: vmrgow
 ; P8BE: sldi r4, r4, 2
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: xxmrghd
+; P8BE: rldimi
+; P8BE: rldimi
+; P8BE: mtvsrd
+; P8BE: mtvsrd
 ; P8BE: xxmrghd
-; P8BE: vmrgow
 ; P8LE: sldi r4, r4, 2
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: xxmrghd
+; P8LE: rldimi
+; P8LE: rldimi
+; P8LE: mtvsrd
+; P8LE: mtvsrd
 ; P8LE: xxmrghd
-; P8LE: vmrgow
 }
 
 ; Function Attrs: norecurse nounwind readnone
 define <4 x i32> @spltRegVali(i32 signext %val) {
 entry:
   %splat.splatinsert = insertelement <4 x i32> undef, i32 %val, i32 0
   %splat.splat = shufflevector <4 x i32> %splat.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
   ret <4 x i32> %splat.splat
entry:
   %vecinit1 = insertelement <4 x i32> %vecinit, i32 %b, i32 1
   %vecinit2 = insertelement <4 x i32> %vecinit1, i32 %c, i32 2
   %vecinit3 = insertelement <4 x i32> %vecinit2, i32 %d, i32 3
   ret <4 x i32> %vecinit3
 ; P9BE-LABEL: fromRegsui
 ; P9LE-LABEL: fromRegsui
 ; P8BE-LABEL: fromRegsui
 ; P8LE-LABEL: fromRegsui
-; P9BE-DAG: mtvsrdd [[REG1:v[0-9]+]], r3, r5
-; P9BE-DAG: mtvsrdd [[REG2:v[0-9]+]], r4, r6
-; P9BE: vmrgow v2, [[REG1]], [[REG2]]
-; P9BE: blr
-; P9LE-DAG: mtvsrdd [[REG1:v[0-9]+]], r5, r3
-; P9LE-DAG: mtvsrdd [[REG2:v[0-9]+]], r6, r4
-; P9LE: vmrgow v2, [[REG2]], [[REG1]]
-; P9LE: blr
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG1:[0-9]+]], r3
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG2:[0-9]+]], r4
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG3:[0-9]+]], r5
-; P8BE-DAG: mtvsrwz {{[vf]}}[[REG4:[0-9]+]], r6
-; P8BE-DAG: xxmrghd [[REG5:v[0-9]+]], {{[v][s]*}}[[REG1]], {{[v][s]*}}[[REG3]]
-; P8BE-DAG: xxmrghd [[REG6:v[0-9]+]], {{[v][s]*}}[[REG2]], {{[v][s]*}}[[REG4]]
-; P8BE: vmrgow v2, [[REG5]], [[REG6]]
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG1:[0-9]+]], r3
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG2:[0-9]+]], r4
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG3:[0-9]+]], r5
-; P8LE-DAG: mtvsrwz {{[vf]}}[[REG4:[0-9]+]], r6
-; P8LE: xxmrghd [[REG5:v[0-9]+]], {{[v][s]*}}[[REG3]], {{[v][s]*}}[[REG1]]
-; P8LE: xxmrghd [[REG6:v[0-9]+]], {{[v][s]*}}[[REG4]], {{[v][s]*}}[[REG2]]
-; P8LE: vmrgow v2, [[REG6]], [[REG5]]
+; P9BE-DAG: rldimi r6, r5, 32, 0
+; P9BE-DAG: rldimi r4, r3, 32, 0
+; P9BE: mtvsrdd v2, r4, r6
+; P9BE: blr
+; P9LE-DAG: rldimi r3, r4, 32, 0
+; P9LE-DAG: rldimi r5, r6, 32, 0
+; P9LE: mtvsrdd v2, r5, r3
+; P9LE: blr
+; P8BE-DAG: rldimi r6, r5, 32, 0
+; P8BE-DAG: rldimi r4, r3, 32, 0
+; P8BE-DAG: mtvsrd f[[REG1:[0-9]+]], r6
+; P8BE-DAG: mtvsrd f[[REG2:[0-9]+]], r4
+; P8BE-DAG: xxmrghd v2, vs[[REG2]], vs[[REG1]]
+; P8BE: blr
+; P8LE-DAG: rldimi r3, r4, 32, 0
+; P8LE-DAG: rldimi r5, r6, 32, 0
+; P8LE-DAG: mtvsrd f[[REG1:[0-9]+]], r3
+; P8LE-DAG: mtvsrd f[[REG2:[0-9]+]], r5
+; P8LE-DAG: xxmrghd v2, vs[[REG2]], vs[[REG1]]
+; P8LE: blr
 }
 
 ; Function Attrs: norecurse nounwind readnone
 define <4 x i32> @fromDiffConstsui() {
 entry:
   ret <4 x i32> <i32 242, i32 -113, i32 889, i32 19>
 ; P9BE-LABEL: fromDiffConstsui
 ; P9LE-LABEL: fromDiffConstsui

 ; P9BE-LABEL: fromRandMemConsui
 ; P9LE-LABEL: fromRandMemConsui
 ; P8BE-LABEL: fromRandMemConsui
 ; P8LE-LABEL: fromRandMemConsui
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
+; P9BE: rldimi
+; P9BE: rldimi
 ; P9BE: mtvsrdd
-; P9BE: mtvsrdd
-; P9BE: vmrgow
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
+; P9LE: rldimi
+; P9LE: rldimi
 ; P9LE: mtvsrdd
-; P9LE: mtvsrdd
-; P9LE: vmrgow
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: xxmrghd
+; P8BE: rldimi
+; P8BE: rldimi
+; P8BE: mtvsrd
+; P8BE: mtvsrd
 ; P8BE: xxmrghd
-; P8BE: vmrgow
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: xxmrghd
+; P8LE: rldimi
+; P8LE: rldimi
+; P8LE: mtvsrd
+; P8LE: mtvsrd
 ; P8LE: xxmrghd
-; P8LE: vmrgow
 }
 
 ; Function Attrs: norecurse nounwind readonly
 define <4 x i32> @fromRandMemVarui(i32* nocapture readonly %arr, i32 signext %elem) {
 entry:
   %add = add nsw i32 %elem, 4
   %idxprom = sext i32 %add to i64
   %arrayidx = getelementptr inbounds i32, i32* %arr, i64 %idxprom

 ; P9LE-LABEL: fromRandMemVarui
 ; P8BE-LABEL: fromRandMemVarui
 ; P8LE-LABEL: fromRandMemVarui
 ; P9BE: sldi r4, r4, 2
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
 ; P9BE: lwz
+; P9BE: rldimi
+; P9BE: rldimi
 ; P9BE: mtvsrdd
-; P9BE: mtvsrdd
-; P9BE: vmrgow
 ; P9LE: sldi r4, r4, 2
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
 ; P9LE: lwz
+; P9LE: rldimi
+; P9LE: rldimi
 ; P9LE: mtvsrdd
-; P9LE: mtvsrdd
-; P9LE: vmrgow
 ; P8BE: sldi r4, r4, 2
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
 ; P8BE: lwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: mtvsrwz
-; P8BE: xxmrghd
+; P8BE: rldimi
+; P8BE: rldimi
+; P8BE: mtvsrd
+; P8BE: mtvsrd
 ; P8BE: xxmrghd
-; P8BE: vmrgow
 ; P8LE: sldi r4, r4, 2
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
 ; P8LE: lwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: mtvsrwz
-; P8LE: xxmrghd
+; P8LE: rldimi
+; P8LE: rldimi
+; P8LE: mtvsrd
+; P8LE: mtvsrd
 ; P8LE: xxmrghd
-; P8LE: vmrgow
 }
 
 ; Function Attrs: norecurse nounwind readnone
 define <4 x i32> @spltRegValui(i32 zeroext %val) {
 entry:
   %splat.splatinsert = insertelement <4 x i32> undef, i32 %val, i32 0
   %splat.splat = shufflevector <4 x i32> %splat.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
   ret <4 x i32> %splat.splat