diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
--- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -843,6 +843,37 @@
   if (Instruction *I = foldVecTruncToExtElt(CI, *this))
     return I;
 
+  /// Whenever an element is extracted from a vector, and then truncated,
+  /// canonicalize by converting it to a bitcast followed by an
+  /// extractelement.
+  ///
+  /// Example (little endian):
+  ///   trunc (extractelement <4 x i64> %X, 0) to i32
+  ///   --->
+  ///   extractelement <8 x i32> (bitcast <4 x i64> %X to <8 x i32>), i32 0
+  if (auto *ExtElt = dyn_cast<ExtractElementInst>(CI.getOperand(0))) {
+    Value *VecOp = ExtElt->getVectorOperand();
+    Type *VecOpTy = VecOp->getType();
+    unsigned DestScalarSize = DestTy->getScalarSizeInBits();
+    unsigned VecOpScalarSize = VecOpTy->getScalarSizeInBits();
+    unsigned BitCastNumElts =
+        VecOpTy->getVectorNumElements() * VecOpScalarSize / DestScalarSize;
+
+    ConstantInt *Cst = nullptr;
+    if (match(ExtElt->getIndexOperand(), m_ConstantInt(Cst))) {
+
+      unsigned VecOpIdx = Cst->getZExtValue();
+      unsigned NewIdx =
+          DL.isBigEndian()
+              ? (VecOpIdx + 1) * VecOpScalarSize / DestScalarSize - 1
+              : VecOpIdx * VecOpScalarSize / DestScalarSize;
+
+      Type *BitCastTo = VectorType::get(DestTy, BitCastNumElts);
+      Value *BitCast = Builder.CreateBitCast(VecOp, BitCastTo);
+      return ExtractElementInst::Create(BitCast, Builder.getInt32(NewIdx));
+    }
+  }
+
   return nullptr;
 }
 
diff --git a/llvm/test/Transforms/InstCombine/pr45314_be.ll b/llvm/test/Transforms/InstCombine/pr45314_be.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/pr45314_be.ll
@@ -0,0 +1,69 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+
+define i32 @shrinkExtractElt_i64_to_i32_0(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_0(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 1
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 0
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i32 @shrinkExtractElt_i64_to_i32_1(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_1(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 3
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 1
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i32 @shrinkExtractElt_i64_to_i32_2(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_2(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 5
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 2
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_0(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_0(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 3
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 0
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_1(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_1(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 7
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 1
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_2(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_2(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 11
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 2
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}
diff --git a/llvm/test/Transforms/InstCombine/pr45314_le.ll b/llvm/test/Transforms/InstCombine/pr45314_le.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/pr45314_le.ll
@@ -0,0 +1,69 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+
+define i32 @shrinkExtractElt_i64_to_i32_0(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_0(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 0
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 0
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i32 @shrinkExtractElt_i64_to_i32_1(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_1(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 2
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 1
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i32 @shrinkExtractElt_i64_to_i32_2(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i32_2(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 4
+; CHECK-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i32 2
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_0(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_0(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 0
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 0
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_1(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_1(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 4
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 1
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}
+
+define i16 @shrinkExtractElt_i64_to_i16_2(<3 x i64> %x) {
+; CHECK-LABEL: @shrinkExtractElt_i64_to_i16_2(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <12 x i16>
+; CHECK-NEXT:    [[T:%.*]] = extractelement <12 x i16> [[TMP1]], i32 8
+; CHECK-NEXT:    ret i16 [[T]]
+;
+  %e = extractelement <3 x i64> %x, i16 2
+  %t = trunc i64 %e to i16
+  ret i16 %t
+}