Fix PR7156. If the sources of a REG_SEQUENCE are all IMPLICIT_DEF's. Replace it with...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103984 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/TwoAddressInstructionPass.cpp b/lib/CodeGen/TwoAddressInstructionPass.cpp
index fb9bddca25c082f5b4a9e60ffae74924c7b9d7d9..2d43fba4bfa2dc25ef4bd8e71978c81aaf23d040 100644 (file)
--- a/lib/CodeGen/TwoAddressInstructionPass.cpp
+++ b/lib/CodeGen/TwoAddressInstructionPass.cpp
@@ -1238,6 +1238,7 @@ bool TwoAddressInstructionPass::EliminateRegSequences() {
       llvm_unreachable(0);
     }
 
+    bool IsImpDef = true;
     SmallVector<unsigned, 4> RealSrcs;
     SmallSet<unsigned, 4> Seen;
     for (unsigned i = 1, e = MI->getNumOperands(); i < e; i += 2) {
@@ -1253,6 +1254,7 @@ bool TwoAddressInstructionPass::EliminateRegSequences() {
         DefMI->eraseFromParent();
         continue;
       }
+      IsImpDef = false;
 
       // Remember EXTRACT_SUBREG sources. These might be candidate for
       // coalescing.
@@ -1297,8 +1299,15 @@ bool TwoAddressInstructionPass::EliminateRegSequences() {
       UpdateRegSequenceSrcs(SrcReg, DstReg, SubIdx, MRI);
     }
 
-    DEBUG(dbgs() << "Eliminated: " << *MI);
-    MI->eraseFromParent();
+    if (IsImpDef) {
+      DEBUG(dbgs() << "Turned: " << *MI << " into an IMPLICIT_DEF");
+      MI->setDesc(TII->get(TargetOpcode::IMPLICIT_DEF));
+      for (int j = MI->getNumOperands() - 1, ee = 0; j > ee; --j)
+        MI->RemoveOperand(j);      
+    } else {
+      DEBUG(dbgs() << "Eliminated: " << *MI);
+      MI->eraseFromParent();
+    }
 
     // Try coalescing some EXTRACT_SUBREG instructions.
     CoalesceExtSubRegs(RealSrcs, DstReg);

diff --git a/test/CodeGen/ARM/reg_sequence.ll b/test/CodeGen/ARM/reg_sequence.ll
index 5bfabefa8eba5f41383dadc1336512cba394450c..cc75a13436ab3f50c282312edb9c6cf4c70e02c7 100644 (file)
--- a/test/CodeGen/ARM/reg_sequence.ll
+++ b/test/CodeGen/ARM/reg_sequence.ll
@@ -185,6 +185,50 @@ entry:
   ret void
 }
 
+; PR7156
+define arm_aapcs_vfpcc i32 @t8() nounwind {
+; CHECK: t8:
+; CHECK: vrsqrte.f32 q0, q0
+bb.nph55.bb.nph55.split_crit_edge:
+  br label %bb3
+
+bb3:                                              ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
+  br i1 undef, label %bb5, label %bb3
+
+bb5:                                              ; preds = %bb3
+  br label %bb.i25
+
+bb.i25:                                           ; preds = %bb.i25, %bb5
+  %0 = shufflevector <2 x float> undef, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
+  %1 = call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %0) nounwind ; <<4 x float>> [#uses=1]
+  %2 = fmul <4 x float> %1, undef                 ; <<4 x float>> [#uses=1]
+  %3 = fmul <4 x float> undef, %2                 ; <<4 x float>> [#uses=1]
+  %tmp26.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
+  %4 = extractelement <2 x double> %tmp26.i, i32 0 ; <double> [#uses=1]
+  %5 = bitcast double %4 to <2 x float>           ; <<2 x float>> [#uses=1]
+  %6 = extractelement <2 x float> %5, i32 1       ; <float> [#uses=1]
+  store float %6, float* undef, align 4
+  br i1 undef, label %bb6, label %bb.i25
+
+bb6:                                              ; preds = %bb.i25
+  br i1 undef, label %bb7, label %bb14
+
+bb7:                                              ; preds = %bb6
+  br label %bb.i49
+
+bb.i49:                                           ; preds = %bb.i49, %bb7
+  br i1 undef, label %bb.i19, label %bb.i49
+
+bb.i19:                                           ; preds = %bb.i19, %bb.i49
+  br i1 undef, label %exit, label %bb.i19
+
+exit:          ; preds = %bb.i19
+  unreachable
+
+bb14:                                             ; preds = %bb6
+  ret i32 0
+}
+
 declare <4 x i32> @llvm.arm.neon.vld1.v4i32(i8*) nounwind readonly
 
 declare <8 x i16> @llvm.arm.neon.vld1.v8i16(i8*) nounwind readonly
@@ -209,4 +253,6 @@ declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16(i8*, <8 x i16>,
 
 declare void @llvm.arm.neon.vst2.v4i32(i8*, <4 x i32>, <4 x i32>) nounwind
 
+declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
+
 declare void @llvm.trap() nounwind