minor optimization when extracting from index 0

Author: tangaac

llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td

@@ -1640,6 +1640,18 @@ defm : PairInsertExtractPatV8<v8f32, f32>;
 defm : PairInsertExtractPatV4<v4i64, GRLenVT>;
 defm : PairInsertExtractPatV4<v4f64, f64>;
 
+foreach imm = 0...7 in {
+  def : Pat<(vector_insert v8i32:$xd, (GRLenVT(vector_extract v8i32:$xj, 0)),
+                uimm3:$imm),
+            (XVINSVE0_W v8i32:$xd, v8i32:$xj, uimm3:$imm)>;
+}
+
+foreach imm = 0...3 in {
+  def : Pat<(vector_insert v4i64:$xd, (GRLenVT(vector_extract v4i64:$xj, 0)),
+                uimm2:$imm),
+            (XVINSVE0_D v4i64:$xd, v4i64:$xj, uimm2:$imm)>;
+}
+
 def : Pat<(vector_insert v8i32:$xd,
               (GRLenVT(vector_extract v8i32:$xj, uimm3:$imm1)), uimm3:$imm2),
           (XVINSVE0_W v8i32:$xd, (XVPICKVE_W v8i32:$xj, uimm3:$imm1),

llvm/test/CodeGen/LoongArch/lasx/ir-instruction/insert-extract-element.ll

@@ -39,6 +39,18 @@ entry:
   ret <8 x i32> %c
 }
 
+
+define <8 x i32> @insert_extract0_v8i32(<8 x i32> %a) nounwind {
+; CHECK-LABEL: insert_extract0_v8i32:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvinsve0.w $xr0, $xr0, 1
+; CHECK-NEXT:    ret
+entry:
+  %b = extractelement <8 x i32> %a, i32 0
+  %c = insertelement <8 x i32> %a, i32 %b, i32 1
+  ret <8 x i32> %c
+}
+
 define <8 x float> @insert_extract_v8f32(<8 x float> %a) nounwind {
 ; CHECK-LABEL: insert_extract_v8f32:
 ; CHECK:       # %bb.0: # %entry
@@ -63,6 +75,17 @@ entry:
   ret <4 x i64> %c
 }
 
+define <4 x i64> @insert_extract0_v4i64(<4 x i64> %a) nounwind {
+; CHECK-LABEL: insert_extract0_v4i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvinsve0.d $xr0, $xr0, 1
+; CHECK-NEXT:    ret
+entry:
+  %b = extractelement <4 x i64> %a, i32 0
+  %c = insertelement <4 x i64> %a, i64 %b, i32 1
+  ret <4 x i64> %c
+}
+
 define <4 x double> @insert_extract_v4f64(<4 x double> %a) nounwind {
 ; CHECK-LABEL: insert_extract_v4f64:
 ; CHECK:       # %bb.0: # %entry