[AArch64][GISel] Lower EXT of 0 to a COPY

This allows us to select G_SHUFFLE_VECTOR with identity masks (possibly
including undef elements), but avoid the actual EXT instruction if the shift
amount is 0.

llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp

@@ -457,11 +457,15 @@ void applyShuffleVectorPseudo(MachineInstr &MI,
 /// for the imported tablegen patterns to work.
 void applyEXT(MachineInstr &MI, ShuffleVectorPseudo &MatchInfo) {
   MachineIRBuilder MIRBuilder(MI);
-  // Tablegen patterns expect an i32 G_CONSTANT as the final op.
-  auto Cst =
-      MIRBuilder.buildConstant(LLT::scalar(32), MatchInfo.SrcOps[2].getImm());
-  MIRBuilder.buildInstr(MatchInfo.Opc, {MatchInfo.Dst},
-                        {MatchInfo.SrcOps[0], MatchInfo.SrcOps[1], Cst});
+  if (MatchInfo.SrcOps[2].getImm() == 0)
+    MIRBuilder.buildCopy(MatchInfo.Dst, MatchInfo.SrcOps[0]);
+  else {
+    // Tablegen patterns expect an i32 G_CONSTANT as the final op.
+    auto Cst =
+        MIRBuilder.buildConstant(LLT::scalar(32), MatchInfo.SrcOps[2].getImm());
+    MIRBuilder.buildInstr(MatchInfo.Opc, {MatchInfo.Dst},
+                          {MatchInfo.SrcOps[0], MatchInfo.SrcOps[1], Cst});
+  }
   MI.eraseFromParent();
 }
 

llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-lowering-shuffle-splat.mir

@@ -196,9 +196,8 @@ body:             |
     ; CHECK-NEXT: [[DEF:%[0-9]+]]:_(<2 x s64>) = G_IMPLICIT_DEF
     ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
     ; CHECK-NEXT: [[IVEC:%[0-9]+]]:_(<2 x s64>) = G_INSERT_VECTOR_ELT [[DEF]], [[COPY]](s64), [[C]](s32)
-    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
-    ; CHECK-NEXT: [[EXT:%[0-9]+]]:_(<2 x s64>) = G_EXT [[IVEC]], [[DEF]], [[C1]](s32)
-    ; CHECK-NEXT: $q0 = COPY [[EXT]](<2 x s64>)
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY [[IVEC]](<2 x s64>)
+    ; CHECK-NEXT: $q0 = COPY [[COPY1]](<2 x s64>)
     ; CHECK-NEXT: RET_ReallyLR implicit $q0
     %0:_(s64) = COPY $x0
     %2:_(<2 x s64>) = G_IMPLICIT_DEF

llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-lowering-zip.mir

@@ -194,10 +194,8 @@ body:             |
     ; CHECK: liveins: $q0, $q1
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $q1
-    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
-    ; CHECK-NEXT: [[EXT:%[0-9]+]]:_(<2 x s64>) = G_EXT [[COPY]], [[COPY1]], [[C]](s32)
-    ; CHECK-NEXT: $q0 = COPY [[EXT]](<2 x s64>)
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY [[COPY]](<2 x s64>)
+    ; CHECK-NEXT: $q0 = COPY [[COPY1]](<2 x s64>)
     ; CHECK-NEXT: RET_ReallyLR implicit $q0
     %0:_(<2 x s64>) = COPY $q0
     %1:_(<2 x s64>) = COPY $q1

llvm/test/CodeGen/AArch64/arm64-dup.ll

@@ -555,7 +555,6 @@ define <2 x i32> @dup_const4_ext(<4 x i32> %A) nounwind {
 ; CHECK-GI-NEXT:    adrp x8, .LCPI39_0
 ; CHECK-GI-NEXT:    ldr q1, [x8, :lo12:.LCPI39_0]
 ; CHECK-GI-NEXT:    add.4s v0, v0, v1
-; CHECK-GI-NEXT:    ext.16b v0, v0, v0, #0
 ; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
 ; CHECK-GI-NEXT:    ret
   %tmp1 = add <4 x i32> %A, <i32 8421377, i32 8421377, i32 8421377, i32 8421377>

llvm/test/CodeGen/AArch64/arm64-neon-add-pairwise.ll

@@ -137,9 +137,8 @@ define i32 @addp_v4i32(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-GI-LABEL: addp_v4i32:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.4s, v0.4s, v1.4s
-; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #0
-; CHECK-GI-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
-; CHECK-GI-NEXT:    addp v0.2s, v1.2s, v0.2s
+; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-GI-NEXT:    addp v0.2s, v0.2s, v1.2s
 ; CHECK-GI-NEXT:    rev64 v1.2s, v0.2s
 ; CHECK-GI-NEXT:    add v0.2s, v0.2s, v1.2s
 ; CHECK-GI-NEXT:    fmov w0, s0
@@ -165,9 +164,8 @@ define <4 x i16> @addp_v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; CHECK-GI-LABEL: addp_v8i16:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.8h, v0.8h, v1.8h
-; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #0
-; CHECK-GI-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
-; CHECK-GI-NEXT:    addp v0.4h, v1.4h, v0.4h
+; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-GI-NEXT:    addp v0.4h, v0.4h, v1.4h
 ; CHECK-GI-NEXT:    ret
   %1 = add <8 x i16> %a, %b
   %2 = shufflevector <8 x i16> %1, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -187,9 +185,8 @@ define <8 x i8> @addp_v16i8(<16 x i8> %a, <16 x i8> %b) {
 ; CHECK-GI-LABEL: addp_v16i8:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.16b, v0.16b, v1.16b
-; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #0
-; CHECK-GI-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
-; CHECK-GI-NEXT:    addp v0.8b, v1.8b, v0.8b
+; CHECK-GI-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-GI-NEXT:    addp v0.8b, v0.8b, v1.8b
 ; CHECK-GI-NEXT:    ret
   %1 = add <16 x i8> %a, %b
   %2 = shufflevector <16 x i8> %1, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>

llvm/test/CodeGen/AArch64/arm64-vabs.ll

@@ -1614,18 +1614,11 @@ declare float @llvm.fabs.f32(float) nounwind readnone
 declare double @llvm.fabs.f64(double) nounwind readnone
 
 define <2 x i64> @uabdl_from_extract_dup(<4 x i32> %lhs, i32 %rhs) {
-; CHECK-SD-LABEL: uabdl_from_extract_dup:
-; CHECK-SD:       // %bb.0:
-; CHECK-SD-NEXT:    dup.2s v1, w0
-; CHECK-SD-NEXT:    uabdl.2d v0, v0, v1
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: uabdl_from_extract_dup:
-; CHECK-GI:       // %bb.0:
-; CHECK-GI-NEXT:    dup.2s v1, w0
-; CHECK-GI-NEXT:    ext.16b v0, v0, v0, #0
-; CHECK-GI-NEXT:    uabdl.2d v0, v0, v1
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: uabdl_from_extract_dup:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    dup.2s v1, w0
+; CHECK-NEXT:    uabdl.2d v0, v0, v1
+; CHECK-NEXT:    ret
   %rhsvec.tmp = insertelement <2 x i32> undef, i32 %rhs, i32 0
   %rhsvec = insertelement <2 x i32> %rhsvec.tmp, i32 %rhs, i32 1
   %lhs.high = shufflevector <4 x i32> %lhs, <4 x i32> undef, <2 x i32> <i32 0, i32 1>
@@ -1656,18 +1649,11 @@ define <2 x i64> @uabdl2_from_extract_dup(<4 x i32> %lhs, i32 %rhs) {
 }
 
 define <2 x i64> @sabdl_from_extract_dup(<4 x i32> %lhs, i32 %rhs) {
-; CHECK-SD-LABEL: sabdl_from_extract_dup:
-; CHECK-SD:       // %bb.0:
-; CHECK-SD-NEXT:    dup.2s v1, w0
-; CHECK-SD-NEXT:    sabdl.2d v0, v0, v1
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: sabdl_from_extract_dup:
-; CHECK-GI:       // %bb.0:
-; CHECK-GI-NEXT:    dup.2s v1, w0
-; CHECK-GI-NEXT:    ext.16b v0, v0, v0, #0
-; CHECK-GI-NEXT:    sabdl.2d v0, v0, v1
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: sabdl_from_extract_dup:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    dup.2s v1, w0
+; CHECK-NEXT:    sabdl.2d v0, v0, v1
+; CHECK-NEXT:    ret
   %rhsvec.tmp = insertelement <2 x i32> undef, i32 %rhs, i32 0
   %rhsvec = insertelement <2 x i32> %rhsvec.tmp, i32 %rhs, i32 1
   %lhs.high = shufflevector <4 x i32> %lhs, <4 x i32> undef, <2 x i32> <i32 0, i32 1>