[PATCH] D127681: [RISCV] Disable matchSplatAsGather for i1 vectors to prevent creating illegal nodes.

Craig Topper via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Jun 13 12:08:31 PDT 2022


craig.topper updated this revision to Diff 436505.
craig.topper added a comment.

Use getVectorElementType() instead of getScalarType().


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D127681/new/

https://reviews.llvm.org/D127681

Files:
  llvm/lib/Target/RISCV/RISCVISelLowering.cpp
  llvm/test/CodeGen/RISCV/rvv/vsplats-i1.ll


Index: llvm/test/CodeGen/RISCV/rvv/vsplats-i1.ll
===================================================================
--- llvm/test/CodeGen/RISCV/rvv/vsplats-i1.ll
+++ llvm/test/CodeGen/RISCV/rvv/vsplats-i1.ll
@@ -191,3 +191,24 @@
   %splat = shufflevector <vscale x 16 x i1> %head, <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer
   ret <vscale x 16 x i1> %splat
 }
+
+define <vscale x 4 x i1> @splat_idx_nxv4i32(<vscale x 4 x i1> %v, i64 %idx) {
+; CHECK-LABEL: splat_idx_nxv4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a1, zero, e8, mf2, ta, mu
+; CHECK-NEXT:    vmv.v.i v8, 0
+; CHECK-NEXT:    vmerge.vim v8, v8, 1, v0
+; CHECK-NEXT:    vsetivli zero, 1, e8, mf2, ta, mu
+; CHECK-NEXT:    vslidedown.vx v8, v8, a0
+; CHECK-NEXT:    vmv.x.s a0, v8
+; CHECK-NEXT:    andi a0, a0, 1
+; CHECK-NEXT:    vsetvli a1, zero, e8, mf2, ta, mu
+; CHECK-NEXT:    vmv.v.x v8, a0
+; CHECK-NEXT:    vmsne.vi v0, v8, 0
+; CHECK-NEXT:    ret
+  %x = extractelement <vscale x 4 x i1> %v, i64 %idx
+  %ins = insertelement <vscale x 4 x i1> poison, i1 %x, i32 0
+  %splat = shufflevector <vscale x 4 x i1> %ins, <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
+  ret <vscale x 4 x i1> %splat
+}
+
Index: llvm/lib/Target/RISCV/RISCVISelLowering.cpp
===================================================================
--- llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -1936,7 +1936,9 @@
     return SDValue();
   SDValue Vec = SplatVal.getOperand(0);
   // Only perform this optimization on vectors of the same size for simplicity.
-  if (Vec.getValueType() != VT)
+  // Don't perform this optimization for i1 vectors.
+  // FIXME: Support i1 vectors, maybe by promoting to i1?
+  if (Vec.getValueType() != VT || VT.getVectorElementType() == MVT::i1)
     return SDValue();
   SDValue Idx = SplatVal.getOperand(1);
   // The index must be a legal type.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D127681.436505.patch
Type: text/x-patch
Size: 1922 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220613/0a554daf/attachment.bin>


More information about the llvm-commits mailing list