[PowerPC] Fix handling of undefs in the PPC::isSplatShuffleMask query

[w2yehia]

Currently, the query assumes that a single undef bit implies the rest of the EltSize - 1 bits are undefs, but that's not always true.
e.g. isSplatShuffleMask( <0,1,2,3,4,5,6,7,undef,undef,undef,undef,0,1,2,3>, 8) should return false.

[llvmbot]

@llvm/pr-subscribers-backend-powerpc

Author: Wael Yehia (w2yehia)

Changes

Currently, the query assumes that a single undef bit implies the rest of the EltSize - 1 bits are undefs, but that's not always true.
e.g. isSplatShuffleMask( <0,1,2,3,4,5,6,7,undef,undef,undef,undef,0,1,2,3>, 8) should return false.

---

Full diff: https://github.com/llvm/llvm-project/pull/145149.diff

2 Files Affected:

• (modified) llvm/lib/Target/PowerPC/PPCISelLowering.cpp (+7-2)
• (added) llvm/test/CodeGen/PowerPC/pr141642.ll (+13)

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index f502d8570425a..d1b17f227a684 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -2242,8 +2242,13 @@ bool PPC::isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize) {
       return false;
 
   for (unsigned i = EltSize, e = 16; i != e; i += EltSize) {
-    if (N->getMaskElt(i) < 0) continue;
-    for (unsigned j = 0; j != EltSize; ++j)
+    // An UNDEF element is a sequence of UNDEF bits.
+    if (N->getMaskElt(i) < 0) {
+      for (unsigned j = 1; j != EltSize; ++j)
+        if (N->getMaskElt(i+j) >= 0)
+          return false;
+    }
+    else for (unsigned j = 0; j != EltSize; ++j)
       if (N->getMaskElt(i+j) != N->getMaskElt(j))
         return false;
   }
diff --git a/llvm/test/CodeGen/PowerPC/pr141642.ll b/llvm/test/CodeGen/PowerPC/pr141642.ll
new file mode 100644
index 0000000000000..38a7065747867
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/pr141642.ll
@@ -0,0 +1,13 @@
+; RUN: llc -mcpu=pwr8 -mtriple=powerpc64le-unknown-linux-gnu -O0 -debug-only=selectiondag -o - < %s 2>&1 | \
+; RUN:  FileCheck %s
+; CHECK-NOT: lxvdsx
+; CHECK-NOT: LD_SPLAT
+
+define weak_odr dso_local void @unpack(ptr noalias noundef %packed_in) local_unnamed_addr {
+entry:
+  %ld = load <2 x i32>, ptr %packed_in, align 2
+  %shuf = shufflevector <2 x i32> %ld, <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 0>
+  %ie = insertelement <4 x i32> %shuf, i32 7, i32 2
+  store <4 x i32> %shuf, ptr %packed_in, align 2
+  ret void
+}

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff HEAD~1 HEAD --extensions cpp -- llvm/lib/Target/PowerPC/PPCISelLowering.cpp

View the diff from clang-format here.

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index d1b17f227..967ebc4cb 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -2245,12 +2245,12 @@ bool PPC::isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize) {
     // An UNDEF element is a sequence of UNDEF bits.
     if (N->getMaskElt(i) < 0) {
       for (unsigned j = 1; j != EltSize; ++j)
-        if (N->getMaskElt(i+j) >= 0)
+        if (N->getMaskElt(i + j) >= 0)
+          return false;
+    } else
+      for (unsigned j = 0; j != EltSize; ++j)
+        if (N->getMaskElt(i + j) != N->getMaskElt(j))
           return false;
-    }
-    else for (unsigned j = 0; j != EltSize; ++j)
-      if (N->getMaskElt(i+j) != N->getMaskElt(j))
-        return false;
   }
   return true;
 }