1; RUN: opt < %s -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -S | FileCheck %s 2 3; Test case for PR44488. Checks that the correct predicates are created for 4; branches where true and false successors are equal. See the checks involving 5; CMP1 and CMP2. 6 7@v_38 = global i16 12061, align 1 8@v_39 = global i16 11333, align 1 9 10define i16 @test_true_and_false_branch_equal() { 11; CHECK-LABEL: @test_true_and_false_branch_equal( 12; CHECK-LABEL: vector.body: 13; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], [[PRED_SREM_CONTINUE2:%.*]] ] 14; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 [[INDEX]] to i16 15; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i16 99, [[TMP0]] 16; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i16> undef, i16 [[OFFSET_IDX]], i32 0 17; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i16> [[BROADCAST_SPLATINSERT]], <2 x i16> undef, <2 x i32> zeroinitializer 18; CHECK-NEXT: [[INDUCTION:%.*]] = add <2 x i16> [[BROADCAST_SPLAT]], <i16 0, i16 1> 19; CHECK-NEXT: [[TMP1:%.*]] = add i16 [[OFFSET_IDX]], 0 20; CHECK-NEXT: [[TMP2:%.*]] = load i16, i16* @v_38, align 1 21; CHECK-NEXT: [[TMP3:%.*]] = load i16, i16* @v_38, align 1 22; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i16> undef, i16 [[TMP2]], i32 0 23; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i16> [[TMP4]], i16 [[TMP3]], i32 1 24; CHECK-NEXT: [[CMP1:%.*]] = icmp eq <2 x i16> [[TMP5]], <i16 32767, i16 32767> 25; CHECK-NEXT: [[CMP2:%.*]] = icmp eq <2 x i16> [[TMP5]], zeroinitializer 26; CHECK-NEXT: [[NOT_CMP2:%.*]] = xor <2 x i1> [[CMP2]], <i1 true, i1 true> 27; CHECK-NEXT: [[PRED1:%.*]] = extractelement <2 x i1> [[NOT_CMP2]], i32 0 28; CHECK-NEXT: br i1 [[PRED1]], label [[PRED_SREM_IF:%.*]], label [[PRED_SREM_CONTINUE:%.*]] 29; CHECK: pred.srem.if: 30; CHECK-NEXT: [[TMP14:%.*]] = srem i16 5786, [[TMP2]] 31; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i16> undef, i16 [[TMP14]], i32 0 32; CHECK-NEXT: br label [[PRED_SREM_CONTINUE]] 33; CHECK: pred.srem.continue: 34; CHECK-NEXT: [[TMP16:%.*]] = phi <2 x i16> [ undef, %vector.body ], [ [[TMP15]], [[PRED_SREM_IF]] ] 35; CHECK-NEXT: [[PRED2:%.*]] = extractelement <2 x i1> [[NOT_CMP2]], i32 1 36; CHECK-NEXT: br i1 [[PRED2]], label [[PRED_SREM_IF1:%.*]], label [[PRED_SREM_CONTINUE2]] 37; CHECK: pred.srem.if1: 38; CHECK-NEXT: [[TMP18:%.*]] = srem i16 5786, [[TMP3]] 39; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x i16> [[TMP16]], i16 [[TMP18]], i32 1 40; CHECK-NEXT: br label [[PRED_SREM_CONTINUE2]] 41; CHECK: pred.srem.continue2: 42; CHECK-NEXT: [[TMP20:%.*]] = phi <2 x i16> [ [[TMP16]], [[PRED_SREM_CONTINUE]] ], [ [[TMP19]], [[PRED_SREM_IF1]] ] 43; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[CMP2]], <2 x i16> <i16 5786, i16 5786>, <2 x i16> [[TMP20]] 44; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 0 45; CHECK-NEXT: store i16 [[TMP22]], i16* @v_39, align 1 46; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 1 47; CHECK-NEXT: store i16 [[TMP23]], i16* @v_39, align 1 48; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 2 49; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], 12 50; CHECK-NEXT: br i1 [[TMP24]], label %middle.block, label %vector.body, !llvm.loop !0 51; 52entry: 53 br label %for.body 54 55for.body: ; preds = %entry, %for.latch 56 %i.07 = phi i16 [ 99, %entry ], [ %inc7, %for.latch ] 57 %lv = load i16, i16* @v_38, align 1 58 %cmp1 = icmp eq i16 %lv, 32767 59 br i1 %cmp1, label %cond.end, label %cond.end 60 61cond.end: ; preds = %for.body, %for.body 62 %cmp2 = icmp eq i16 %lv, 0 63 br i1 %cmp2, label %for.latch, label %cond.false4 64 65cond.false4: ; preds = %cond.end 66 %rem = srem i16 5786, %lv 67 br label %for.latch 68 69for.latch: ; preds = %cond.end, %cond.false4 70 %cond6 = phi i16 [ %rem, %cond.false4 ], [ 5786, %cond.end ] 71 store i16 %cond6, i16* @v_39, align 1 72 %inc7 = add nsw i16 %i.07, 1 73 %cmp = icmp slt i16 %inc7, 111 74 br i1 %cmp, label %for.body, label %exit 75 76exit: ; preds = %for.latch 77 %rv = load i16, i16* @v_39, align 1 78 ret i16 %rv 79} 80