[PATCH] D79854: [PowerPC] Prevent legalization loop from promoting SELECT_CC from v4i32 to v4i32

Nemanja Ivanovic via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed May 13 05:54:36 PDT 2020


nemanjai created this revision.
nemanjai added reviewers: PowerPC, hfinkel, tstellar.
Herald added subscribers: shchenz, kbarton, hiraditya.
Herald added a project: LLVM.

As reported in https://bugs.llvm.org/show_bug.cgi?id=45709 we can hit an infinite loop in legalization since we set the legalization action for `ISD::SELECT_CC` for all fixed length vector types to `Promote`. Without some different legalization action for the type being promoted to, the legalizer simply loops. Since we don't have patterns to match the node, the right legalization action should be `Expand`.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D79854

Files:
  llvm/lib/Target/PowerPC/PPCISelLowering.cpp
  llvm/test/CodeGen/PowerPC/pr45709.ll


Index: llvm/test/CodeGen/PowerPC/pr45709.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/PowerPC/pr45709.ll
@@ -0,0 +1,54 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-unknown \
+; RUN:     -mcpu=pwr6 -ppc-asm-full-reg-names -mattr=-vsx \
+; RUN:     -ppc-vsr-nums-as-vr < %s | FileCheck %s
+define dso_local void @_ZN1a1bEv(<4 x float> %in) local_unnamed_addr #0 align 2 {
+; CHECK-LABEL: _ZN1a1bEv:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    bclr 12, 4*cr5+lt, 0
+; CHECK-NEXT:  # %bb.1: # %.preheader
+; CHECK-NEXT:    addis r3, r2, .LCPI0_0 at toc@ha
+; CHECK-NEXT:    vxor v3, v3, v3
+; CHECK-NEXT:    addi r3, r3, .LCPI0_0 at toc@l
+; CHECK-NEXT:    lvx v4, 0, r3
+; CHECK-NEXT:    addi r3, r1, -48
+; CHECK-NEXT:    stvx v3, 0, r3
+; CHECK-NEXT:    addi r3, r1, -32
+; CHECK-NEXT:    vperm v2, v2, v2, v4
+; CHECK-NEXT:    stvx v2, 0, r3
+; CHECK-NEXT:    lwz r3, -48(r1)
+; CHECK-NEXT:    lwz r4, -32(r1)
+; CHECK-NEXT:    cmpw r4, r3
+; CHECK-NEXT:    bc 12, gt, .LBB0_2
+; CHECK-NEXT:    b .LBB0_3
+; CHECK-NEXT:  .LBB0_2: # %.preheader
+; CHECK-NEXT:    addi r3, r4, 0
+; CHECK-NEXT:  .LBB0_3: # %.preheader
+; CHECK-NEXT:    stw r3, -64(r1)
+; CHECK-NEXT:    addi r3, r1, -64
+; CHECK-NEXT:    lvx v2, 0, r3
+; CHECK-NEXT:    addi r3, r1, -16
+; CHECK-NEXT:    stvx v2, 0, r3
+; CHECK-NEXT:    blr
+  br i1 undef, label %7, label %1
+
+1:                                                ; preds = %1, %0
+  br i1 undef, label %2, label %1
+
+2:                                                ; preds = %1
+  %3 = shufflevector <4 x float> %in, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 0>
+  %4 = call fast <4 x float> @llvm.maxnum.v4f32(<4 x float> %3, <4 x float> zeroinitializer)
+  %5 = call fast <4 x float> @llvm.maxnum.v4f32(<4 x float> %4, <4 x float> undef)
+  %6 = extractelement <4 x float> %5, i32 0
+  br label %7
+
+7:                                                ; preds = %2, %0
+  %8 = phi float [ %6, %2 ], [ undef, %0 ]
+  %9 = fcmp fast une float %8, 0.000000e+00
+  ret void
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>) #0
+
+attributes #0 = { nounwind }
Index: llvm/lib/Target/PowerPC/PPCISelLowering.cpp
===================================================================
--- llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -711,6 +711,7 @@
         setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
       }
     }
+    setOperationAction(ISD::SELECT_CC, MVT::v4i32, Expand);
     if (!Subtarget.hasP8Vector()) {
       setOperationAction(ISD::SMAX, MVT::v2i64, Expand);
       setOperationAction(ISD::SMIN, MVT::v2i64, Expand);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D79854.263681.patch
Type: text/x-patch
Size: 2868 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200513/c7e255e1/attachment.bin>


More information about the llvm-commits mailing list