Removed cpu specification from .ll test; Added check for simplify form of loop before rotation

Author: Marek Sedlacek

llvm/lib/Transforms/Utils/LoopUnroll.cpp

@@ -487,11 +487,14 @@ llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
 
   LoopUnrollResult Result = LoopUnrollResult::Unmodified;
 
-  if (ULO.Runtime && SE) {
+  // Rotate loop if it makes the exit count from the latch computable (for
+  // later unrolling).
+  // The check for LoopSimplify form is done so that after the rotation this
+  // check does not fail in UnrollRuntimeLoopRemainder and the rotation is not
+  // redundant.
+  if (ULO.Runtime && SE && L->isLoopSimplifyForm()) {
     BasicBlock *OrigHeader = L->getHeader();
     BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator());
-    // Rotate loop if it makes the exit count from the latch computable (for
-    // later unrolling).
     if (BI && !BI->isUnconditional() &&
         isa<SCEVCouldNotCompute>(SE->getExitCount(L, L->getLoopLatch())) &&
         !isa<SCEVCouldNotCompute>(SE->getExitCount(L, OrigHeader))) {

…nroll/X86/runtime-unroll-after-rotate.ll …oopUnroll/runtime-unroll-after-rotate.llllvm/test/Transforms/LoopUnroll/X86/runtime-unroll-after-rotate.ll renamed to llvm/test/Transforms/LoopUnroll/runtime-unroll-after-rotate.ll llvm/test/Transforms/LoopUnroll/X86/runtime-unroll-after-rotate.ll renamed to llvm/test/Transforms/LoopUnroll/runtime-unroll-after-rotate.ll

@@ -1,18 +1,18 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
-; RUN: opt --passes=loop-unroll -unroll-runtime-other-exit-predictable=1 -S %s | FileCheck %s
+; RUN: opt --passes=loop-unroll -unroll-runtime=true -unroll-runtime-other-exit-predictable=1 -S %s | FileCheck %s
 target triple = "x86_64-unknown-linux-gnu"
 
 define void @test(i64 %0, ptr %1) #0 {
 ; CHECK-LABEL: define void @test(
-; CHECK-SAME: i64 [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-SAME: i64 [[TMP0:%.*]], ptr [[TMP1:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[B1:%.*]] = icmp eq i64 [[TMP0]], 0
 ; CHECK-NEXT:    br i1 [[B1]], label %[[AFTER:.*]], label %[[BODY_LR_PH:.*]]
 ; CHECK:       [[BODY_LR_PH]]:
 ; CHECK-NEXT:    [[TMP5:%.*]] = sub i64 0, [[TMP0]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = freeze i64 [[TMP5]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[TMP2]], -1
-; CHECK-NEXT:    [[XTRAITER:%.*]] = and i64 [[TMP2]], 3
+; CHECK-NEXT:    [[XTRAITER:%.*]] = and i64 [[TMP2]], 7
 ; CHECK-NEXT:    [[LCMP_MOD:%.*]] = icmp ne i64 [[XTRAITER]], 0
 ; CHECK-NEXT:    br i1 [[LCMP_MOD]], label %[[BODY_PROL_PREHEADER:.*]], label %[[BODY_PROL_LOOPEXIT:.*]]
 ; CHECK:       [[BODY_PROL_PREHEADER]]:
@@ -35,22 +35,30 @@ define void @test(i64 %0, ptr %1) #0 {
 ; CHECK-NEXT:    br label %[[BODY_PROL_LOOPEXIT]]
 ; CHECK:       [[BODY_PROL_LOOPEXIT]]:
 ; CHECK-NEXT:    [[A2_UNR:%.*]] = phi i64 [ [[TMP0]], %[[BODY_LR_PH]] ], [ [[A2_UNR_PH]], %[[BODY_PROL_LOOPEXIT_UNR_LCSSA]] ]
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i64 [[TMP3]], 3
-; CHECK-NEXT:    br i1 [[TMP4]], label %[[HEADER_AFTER_CRIT_EDGE:.*]], label %[[BODY_LR_PH_NEW:.*]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ult i64 [[TMP3]], 7
+; CHECK-NEXT:    br i1 [[TMP6]], label %[[HEADER_AFTER_CRIT_EDGE:.*]], label %[[BODY_LR_PH_NEW:.*]]
 ; CHECK:       [[BODY_LR_PH_NEW]]:
 ; CHECK-NEXT:    br label %[[BODY:.*]]
 ; CHECK:       [[HEADER:.*]]:
 ; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT:.*]], label %[[HEADER_1:.*]]
 ; CHECK:       [[HEADER_1]]:
 ; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_2:.*]]
 ; CHECK:       [[HEADER_2]]:
-; CHECK-NEXT:    [[C_7:%.*]] = add i64 [[A2:%.*]], 4
 ; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_3:.*]]
 ; CHECK:       [[HEADER_3]]:
+; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_4:.*]]
+; CHECK:       [[HEADER_4]]:
+; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_5:.*]]
+; CHECK:       [[HEADER_5]]:
+; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_6:.*]]
+; CHECK:       [[HEADER_6]]:
+; CHECK-NEXT:    [[C_7:%.*]] = add i64 [[A2:%.*]], 8
+; CHECK-NEXT:    br i1 false, label %[[END_LOOPEXIT]], label %[[HEADER_7:.*]]
+; CHECK:       [[HEADER_7]]:
 ; CHECK-NEXT:    [[B_7:%.*]] = icmp eq i64 [[C_7]], 0
 ; CHECK-NEXT:    br i1 [[B_7]], label %[[HEADER_AFTER_CRIT_EDGE_UNR_LCSSA:.*]], label %[[BODY]]
 ; CHECK:       [[BODY]]:
-; CHECK-NEXT:    [[A2]] = phi i64 [ [[A2_UNR]], %[[BODY_LR_PH_NEW]] ], [ [[C_7]], %[[HEADER_3]] ]
+; CHECK-NEXT:    [[A2]] = phi i64 [ [[A2_UNR]], %[[BODY_LR_PH_NEW]] ], [ [[C_7]], %[[HEADER_7]] ]
 ; CHECK-NEXT:    [[D:%.*]] = load i32, ptr [[TMP1]], align 4
 ; CHECK-NEXT:    [[E:%.*]] = icmp eq i32 [[D]], 0
 ; CHECK-NEXT:    br i1 [[E]], label %[[END_LOOPEXIT]], label %[[HEADER]]
@@ -92,7 +100,6 @@ after:
 
 declare void @foo()
 
-attributes #0 = { "tune-cpu"="generic" }
 ;.
 ; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]}
 ; CHECK: [[META1]] = !{!"llvm.loop.unroll.disable"}