[SVE][CodeGen] Fix bug in DAGCombiner::reduceBuildVecToShuffle

When trying to reduce a BUILD_VECTOR to a SHUFFLE_VECTOR it's
important that we carefully check the vector types that led to
that BUILD_VECTOR. In the test I have attached to this commit
there is a case where the results of two SVE faddv instructions
are being stored to consecutive memory locations. With my fix,
as part of merging those stores we discover that each BUILD_VECTOR
element came from an extract of a SVE vector element and
therefore bail out.

Differential Revision: https://reviews.llvm.org/D82564

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 4817c0e..2568f66 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -18287,6 +18287,9 @@ SDValue DAGCombiner::reduceBuildVecToShuffle(SDNode *N) {
       return SDValue();
     SDValue ExtractedFromVec = Op.getOperand(0);
 
+    if (ExtractedFromVec.getValueType().isScalableVector())
+      return SDValue();
+
     const APInt &ExtractIdx = Op.getConstantOperandAPInt(1);
     if (ExtractIdx.uge(ExtractedFromVec.getValueType().getVectorNumElements()))
       return SDValue();

diff --git a/llvm/test/CodeGen/AArch64/sve-merging-stores.ll b/llvm/test/CodeGen/AArch64/sve-merging-stores.ll
new file mode 100644 (file)
index 0000000..66a526f
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-merging-stores.ll
@@ -0,0 +1,32 @@
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+%complex = type { { double, double } }
+
+; Function Attrs: argmemonly nounwind readonly
+declare <vscale x 2 x double> @llvm.aarch64.sve.tuple.get.nxv2f64.nxv4f64(<vscale x 4 x double>, i32 immarg) #3
+
+; Function Attrs: argmemonly nounwind readonly
+declare <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1>, double*) #3
+
+; Function Attrs: nounwind readnone
+declare double @llvm.aarch64.sve.faddv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>) #2
+
+define void @foo1(%complex* %outval, <vscale x 2 x i1> %pred, double *%inptr) {
+; CHECK-LABEL: foo1:
+; CHECK: ld2d { z0.d, z1.d }, p0/z, [x1]
+; CHECK-NEXT: faddv d2, p0, z0.d
+; CHECK-NEXT: faddv d0, p0, z1.d
+; CHECK-NEXT: mov v2.d[1], v0.d[0]
+; CHECK-NEXT: str q2, [x0]
+  %realp = getelementptr inbounds %complex, %complex* %outval, i64 0, i32 0, i32 0
+  %imagp = getelementptr inbounds %complex, %complex* %outval, i64 0, i32 0, i32 1
+  %1 = call <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1> %pred, double* nonnull %inptr)
+  %2 = call <vscale x 2 x double> @llvm.aarch64.sve.tuple.get.nxv2f64.nxv4f64(<vscale x 4 x double> %1, i32 0)
+  %3 = call double @llvm.aarch64.sve.faddv.nxv2f64(<vscale x 2 x i1> %pred, <vscale x 2 x double> %2)
+  %4 = call <vscale x 2 x double> @llvm.aarch64.sve.tuple.get.nxv2f64.nxv4f64(<vscale x 4 x double> %1, i32 1)
+  %5 = call double @llvm.aarch64.sve.faddv.nxv2f64(<vscale x 2 x i1> %pred, <vscale x 2 x double> %4)
+  store double %3, double* %realp, align 8
+  store double %5, double* %imagp, align 8
+  ret void
+}
+