[AArch64][GlobalISel] Look through COPY and G_BITCAST while selecting fcvtl2 (fpext)

It tackles some regressions introduced in
https://reviews.llvm.org/D144670.

Author: dzhidzhoev

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

@@ -6777,11 +6777,29 @@ AArch64InstructionSelector::selectExtractHigh(MachineOperand &Root) const {
   MachineRegisterInfo &MRI =
       Root.getParent()->getParent()->getParent()->getRegInfo();
 
-  MachineInstr *Extract = getDefIgnoringCopies(Root.getReg(), MRI);
-  if (Extract && Extract->getOpcode() == TargetOpcode::G_UNMERGE_VALUES &&
-      Root.getReg() == Extract->getOperand(1).getReg()) {
-    Register ExtReg = Extract->getOperand(2).getReg();
-    return {{[=](MachineInstrBuilder &MIB) { MIB.addUse(ExtReg); }}};
+  auto Extract = getDefSrcRegIgnoringCopies(Root.getReg(), MRI);
+  while (Extract && Extract->MI->getOpcode() == TargetOpcode::G_BITCAST &&
+         STI.isLittleEndian())
+    Extract =
+        getDefSrcRegIgnoringCopies(Extract->MI->getOperand(1).getReg(), MRI);
+  if (!Extract)
+    return std::nullopt;
+
+  if (Extract->MI->getOpcode() == TargetOpcode::G_UNMERGE_VALUES) {
+    if (Extract->Reg == Extract->MI->getOperand(1).getReg()) {
+      Register ExtReg = Extract->MI->getOperand(2).getReg();
+      return {{[=](MachineInstrBuilder &MIB) { MIB.addUse(ExtReg); }}};
+    }
+  }
+  if (Extract->MI->getOpcode() == TargetOpcode::G_EXTRACT_VECTOR_ELT) {
+    LLT SrcTy = MRI.getType(Extract->MI->getOperand(1).getReg());
+    auto LaneIdx = getIConstantVRegValWithLookThrough(
+        Extract->MI->getOperand(2).getReg(), MRI);
+    if (LaneIdx && SrcTy == LLT::fixed_vector(2, 64) &&
+        LaneIdx->Value.getSExtValue() == 1) {
+      Register ExtReg = Extract->MI->getOperand(1).getReg();
+      return {{[=](MachineInstrBuilder &MIB) { MIB.addUse(ExtReg); }}};
+    }
   }
 
   return std::nullopt;

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

@@ -46,8 +46,7 @@ define <2 x double> @test_vcvt_high_v1f64_f32_bitcast(<4 x float> %x) nounwind r
 ;
 ; GISEL-LABEL: test_vcvt_high_v1f64_f32_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.2d, v0.2s
+; GISEL-NEXT:    fcvtl2 v0.2d, v0.4s
 ; GISEL-NEXT:    ret
   %bc1 = bitcast <4 x float> %x to <2 x double>
   %ext = shufflevector <2 x double> %bc1, <2 x double> undef, <1 x i32> <i32 1>
@@ -64,8 +63,7 @@ define <2 x double> @test_vcvt_high_v1i64_f32_bitcast(<2 x i64> %x) nounwind rea
 ;
 ; GISEL-LABEL: test_vcvt_high_v1i64_f32_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.2d, v0.2s
+; GISEL-NEXT:    fcvtl2 v0.2d, v0.4s
 ; GISEL-NEXT:    ret
   %ext = shufflevector <2 x i64> %x, <2 x i64> undef, <1 x i32> <i32 1>
   %bc2 = bitcast <1 x i64> %ext to <2 x float>
@@ -97,8 +95,7 @@ define <2 x double> @test_vcvt_high_v4i16_f32_bitcast(<8 x i16> %x) nounwind rea
 ;
 ; GISEL-LABEL: test_vcvt_high_v4i16_f32_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.2d, v0.2s
+; GISEL-NEXT:    fcvtl2 v0.2d, v0.4s
 ; GISEL-NEXT:    ret
   %ext = shufflevector <8 x i16> %x, <8 x i16> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
   %bc2 = bitcast <4 x i16> %ext to <2 x float>
@@ -114,8 +111,7 @@ define <2 x double> @test_vcvt_high_v8i8_f32_bitcast(<16 x i8> %x) nounwind read
 ;
 ; GISEL-LABEL: test_vcvt_high_v8i8_f32_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.2d, v0.2s
+; GISEL-NEXT:    fcvtl2 v0.2d, v0.4s
 ; GISEL-NEXT:    ret
   %ext = shufflevector <16 x i8> %x, <16 x i8> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %bc2 = bitcast <8 x i8> %ext to <2 x float>
@@ -131,8 +127,7 @@ define <4 x float> @test_vcvt_high_v1i64_f16_bitcast(<2 x i64> %x) nounwind read
 ;
 ; GISEL-LABEL: test_vcvt_high_v1i64_f16_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.4s, v0.4h
+; GISEL-NEXT:    fcvtl2 v0.4s, v0.8h
 ; GISEL-NEXT:    ret
   %ext = shufflevector <2 x i64> %x, <2 x i64> undef, <1 x i32> <i32 1>
   %bc2 = bitcast <1 x i64> %ext to <4 x half>
@@ -148,8 +143,7 @@ define <4 x float> @test_vcvt_high_v2i32_f16_bitcast(<4 x i32> %x) nounwind read
 ;
 ; GISEL-LABEL: test_vcvt_high_v2i32_f16_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.4s, v0.4h
+; GISEL-NEXT:    fcvtl2 v0.4s, v0.8h
 ; GISEL-NEXT:    ret
   %ext = shufflevector <4 x i32> %x, <4 x i32> undef, <2 x i32> <i32 2, i32 3>
   %bc2 = bitcast <2 x i32> %ext to <4 x half>
@@ -181,8 +175,7 @@ define <4 x float> @test_vcvt_high_v8i8_f16_bitcast(<16 x i8> %x) nounwind readn
 ;
 ; GISEL-LABEL: test_vcvt_high_v8i8_f16_bitcast:
 ; GISEL:       // %bb.0:
-; GISEL-NEXT:    mov d0, v0[1]
-; GISEL-NEXT:    fcvtl v0.4s, v0.4h
+; GISEL-NEXT:    fcvtl2 v0.4s, v0.8h
 ; GISEL-NEXT:    ret
   %ext = shufflevector <16 x i8> %x, <16 x i8> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %bc2 = bitcast <8 x i8> %ext to <4 x half>