[Test] Add test showing missed branch elimination due to loop predication transform

Author: d-makogon

llvm/test/Transforms/LoopPredication/assumes.ll

@@ -0,0 +1,115 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -passes='loop-mssa(loop-predication),gvn,simplifycfg' < %s 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i32 @test0(i32* %p1, i8* %p2, i32* %p3, i8* %p4, i8* %p5, i1 %c, i32 %x) {
+; CHECK-LABEL: @test0(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[P1_1:%.*]] = getelementptr i32, i32* [[P1:%.*]], i64 1
+; CHECK-NEXT:    [[P1_2:%.*]] = getelementptr i32, i32* [[P1]], i64 2
+; CHECK-NEXT:    [[P1_3:%.*]] = getelementptr i32, i32* [[P1]], i64 3
+; CHECK-NEXT:    [[IV_1_START:%.*]] = load i32, i32* [[P1_1]], align 4, !range [[RNG0:![0-9]+]]
+; CHECK-NEXT:    [[IV_1_END:%.*]] = load i32, i32* [[P1_2]], align 4, !range [[RNG0]]
+; CHECK-NEXT:    [[IV_2_END:%.*]] = load i32, i32* [[P1_3]], align 4, !range [[RNG0]]
+; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_2_END]], [[IV_1_END]]
+; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
+; CHECK:       loop.preheader:
+; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[IV_1_END]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[TMP0]], [[IV_1_START]]
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ule i32 [[IV_2_END]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[IV_1_START]], [[IV_1_END]]
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP3]], [[TMP2]]
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[IV_1:%.*]] = phi i32 [ [[IV_1_NEXT:%.*]], [[LATCH:%.*]] ], [ [[IV_1_START]], [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    [[IV_2:%.*]] = phi i32 [ [[IV_2_NEXT:%.*]], [[LATCH]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    [[GEP_1:%.*]] = getelementptr i8, i8* [[P2:%.*]], i32 [[IV_1]]
+; CHECK-NEXT:    [[VALUE:%.*]] = load i8, i8* [[GEP_1]], align 1
+; CHECK-NEXT:    [[COND_1:%.*]] = icmp ult i32 [[IV_1]], [[IV_1_END]]
+; CHECK-NEXT:    [[WC:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP4]], [[WC]]
+; CHECK-NEXT:    br i1 [[TMP5]], label [[LOOP_NEXT:%.*]], label [[DEOPT:%.*]]
+; CHECK:       loop.next:
+; CHECK-NEXT:    br i1 [[COND_1]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
+; CHECK:       if.true:
+; CHECK-NEXT:    [[GEP_3:%.*]] = getelementptr i8, i8* [[P4:%.*]], i32 [[IV_1]]
+; CHECK-NEXT:    store i8 [[VALUE]], i8* [[GEP_3]], align 1
+; CHECK-NEXT:    br label [[LATCH]]
+; CHECK:       if.false:
+; CHECK-NEXT:    [[GEP_4:%.*]] = getelementptr i8, i8* [[P4]], i32 [[IV_2]]
+; CHECK-NEXT:    store i8 [[VALUE]], i8* [[GEP_4]], align 1
+; CHECK-NEXT:    br label [[LATCH]]
+; CHECK:       latch:
+; CHECK-NEXT:    [[IV_1_NEXT]] = add nuw nsw i32 [[IV_1]], 1
+; CHECK-NEXT:    [[IV_2_NEXT]] = add nuw nsw i32 [[IV_2]], 1
+; CHECK-NEXT:    [[LATCH_COND:%.*]] = icmp ult i32 [[IV_2]], [[IV_2_END]]
+; CHECK-NEXT:    br i1 [[LATCH_COND]], label [[LOOP]], label [[EXIT]]
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPT_RES:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPT_RES]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[RES:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_2]], [[LATCH]] ]
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+entry:
+  %p1.1 = getelementptr i32, i32* %p1, i64 1
+  %p1.2 = getelementptr i32, i32* %p1, i64 2
+  %p1.3 = getelementptr i32, i32* %p1, i64 3
+  %iv.1.start = load i32, i32* %p1.1, !range !0
+  %iv.1.end = load i32, i32* %p1.2, !range !0
+  %iv.2.end = load i32, i32* %p1.3, !range !0
+  %loop.cond = icmp ult i32 %iv.2.end, %iv.1.end
+  br i1 %loop.cond, label %loop, label %exit
+
+loop:
+  %iv.1 = phi i32 [ %iv.1.start, %entry ], [ %iv.1.next, %latch ]
+  %iv.2 = phi i32 [ 0, %entry ], [ %iv.2.next, %latch ]
+  %gep.1 = getelementptr i8, i8* %p2, i32 %iv.1
+  %value = load i8, i8* %gep.1
+  %cond.1 = icmp ult i32 %iv.1, %iv.1.end
+  %wc = call i1 @llvm.experimental.widenable.condition()
+  %explicit_guard_cond = and i1 %cond.1, %wc
+  br i1 %explicit_guard_cond, label %loop.next, label %deopt
+
+loop.next:
+  br i1 %cond.1, label %if.true, label %if.false
+
+if.true:
+  %gep.3 = getelementptr i8, i8* %p4, i32 %iv.1
+  store i8 %value, i8* %gep.3
+  br label %latch
+
+if.false:
+  %gep.4 = getelementptr i8, i8* %p4, i32 %iv.2
+  store i8 %value, i8* %gep.4
+  br label %latch
+
+latch:
+  %iv.1.next = add nuw nsw i32 %iv.1, 1
+  %iv.2.next = add nuw nsw i32 %iv.2, 1
+  %latch.cond = icmp ult i32 %iv.2, %iv.2.end
+  br i1 %latch.cond, label %loop, label %exit
+
+deopt:
+  %deopt_res = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deopt_res
+
+exit:
+  %res = phi i32 [ 0, %entry ], [ %iv.2, %latch ]
+  ret i32 %res
+}
+
+; Function Attrs: inaccessiblememonly nocallback nofree nosync nounwind willreturn
+declare void @llvm.assume(i1) #0
+
+; Function Attrs: inaccessiblememonly nocallback nofree nosync nounwind speculatable willreturn
+declare i1 @llvm.experimental.widenable.condition() #1
+
+declare i32 @llvm.experimental.deoptimize.i32(...)
+
+attributes #0 = { inaccessiblememonly nocallback nofree nosync nounwind willreturn }
+attributes #1 = { inaccessiblememonly nocallback nofree nosync nounwind speculatable willreturn }
+
+!0 = !{i32 0, i32 2147483646}