[AArch64] add missing pattern for insert_subvector undef

sebpop · sebpop · commit 33bdb3e0e6ff · 2018-03-07T22:07:13.000Z
The attached testcase started failing after the patch to define
isExtractSubvectorCheap with the following pattern mismatch:

ISEL: Starting pattern match
  Initial Opcode index to 85068
    Match failed at index 85076
    LLVM ERROR: Cannot select: t47: v8i16 = insert_subvector undef:v8i16, t43, Constant:i64&lt;0&gt;

The code generated from llvm/lib/Target/AArch64/AArch64InstrInfo.td

def : Pat&lt;(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)&gt;;

is in ninja/lib/Target/AArch64/AArch64GenDAGISel.inc
At the location of the error it is:
/* 85076*/    OPC_CheckChild2Type, MVT::i32,

And it failed to match the type of operand 2.
Adding another def-pat for i64 fixes the failed def-pat error:

def : Pat&lt;(insert_subvector undef, (v4i16 FPR64:$src), (i64 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)&gt;;

llvm-svn: 326949
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -6183,20 +6183,25 @@ def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
 
 // A 64-bit subvector insert to the first 128-bit vector position
 // is a subregister copy that needs no instruction.
-def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
-def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
-          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+multiclass InsertSubvectorUndef<ValueType Ty> {
+  def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+  def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (Ty 0)),
+            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
+}
+
+defm : InsertSubvectorUndef<i32>;
+defm : InsertSubvectorUndef<i64>;
 
 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
 // or v2f32.
diff --git a/llvm/test/CodeGen/AArch64/aarch64-insert-subvector-undef.ll b/llvm/test/CodeGen/AArch64/aarch64-insert-subvector-undef.ll
@@ -0,0 +1,21 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+neon < %s
+
+; Check that this does not ICE.
+
+@d = common dso_local local_unnamed_addr global <4 x i16> zeroinitializer, align 8
+
+define <8 x i16> @c(i32 %e) {
+entry:
+  %0 = load <4 x i16>, <4 x i16>* @d, align 8
+  %vminv = tail call i32 @llvm.aarch64.neon.uminv.i32.v4i16(<4 x i16> %0)
+  %1 = trunc i32 %vminv to i16
+  %vecinit3 = insertelement <4 x i16> <i16 undef, i16 undef, i16 0, i16 0>, i16 %1, i32 1
+  %call = tail call <8 x i16> @c(i32 0) #3
+  %vgetq_lane = extractelement <8 x i16> %call, i32 0
+  %vset_lane = insertelement <4 x i16> %vecinit3, i16 %vgetq_lane, i32 0
+  %call4 = tail call i32 bitcast (i32 (...)* @k to i32 (<4 x i16>)*)(<4 x i16> %vset_lane) #3
+  ret <8 x i16> undef
+}
+
+declare i32 @llvm.aarch64.neon.uminv.i32.v4i16(<4 x i16>)
+declare i32 @k(...)
