96 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes='hardware-loops<force-hardware-loops;hardware-loop-decrement=1;hardware-loop-counter-bitwidth=32>' -S | FileCheck %s3 4define arm_aapcs_vfpcc void @test(ptr noalias nocapture readonly %off, ptr noalias nocapture %data, ptr noalias nocapture %dst, i32 %n) {5; CHECK-LABEL: @test(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[CMP252:%.*]] = icmp sgt i32 [[N:%.*]], 08; CHECK-NEXT: br i1 [[CMP252]], label [[FOR_COND1_PREHEADER_US:%.*]], label [[FOR_COND_CLEANUP:%.*]]9; CHECK: for.cond1.preheader.us:10; CHECK-NEXT: [[I_057_US:%.*]] = phi i32 [ [[INC29_US:%.*]], [[FOR_COND_CLEANUP14_US:%.*]] ], [ 0, [[ENTRY:%.*]] ]11; CHECK-NEXT: [[MUL_US:%.*]] = mul i32 [[I_057_US]], [[N]]12; CHECK-NEXT: call void @llvm.set.loop.iterations.i32(i32 [[N]])13; CHECK-NEXT: br label [[FOR_BODY4_US:%.*]]14; CHECK: for.body4.us:15; CHECK-NEXT: [[J_053_US:%.*]] = phi i32 [ 0, [[FOR_COND1_PREHEADER_US]] ], [ [[INC_US:%.*]], [[FOR_BODY4_US]] ]16; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i16, ptr [[OFF:%.*]], i32 [[J_053_US]]17; CHECK-NEXT: [[L2:%.*]] = load i16, ptr [[ARRAYIDX_US]], align 218; CHECK-NEXT: [[ARRAYIDX5_US:%.*]] = getelementptr inbounds i16, ptr [[DATA:%.*]], i32 [[J_053_US]]19; CHECK-NEXT: [[L3:%.*]] = load i16, ptr [[ARRAYIDX5_US]], align 220; CHECK-NEXT: [[ADD_US:%.*]] = add i16 [[L3]], [[L2]]21; CHECK-NEXT: [[ADD8_US:%.*]] = add i32 [[J_053_US]], [[MUL_US]]22; CHECK-NEXT: [[ARRAYIDX9_US:%.*]] = getelementptr inbounds i16, ptr [[DATA]], i32 [[ADD8_US]]23; CHECK-NEXT: store i16 [[ADD_US]], ptr [[ARRAYIDX9_US]], align 224; CHECK-NEXT: [[INC_US]] = add nuw nsw i32 [[J_053_US]], 125; CHECK-NEXT: [[TMP0:%.*]] = call i1 @llvm.loop.decrement.i32(i32 1)26; CHECK-NEXT: br i1 [[TMP0]], label [[FOR_BODY4_US]], label [[FOR_BODY15_US_PREHEADER:%.*]]27; CHECK: for.body15.us.preheader:28; CHECK-NEXT: [[J10_055_US_PH:%.*]] = phi i32 [ 0, [[FOR_BODY4_US]] ]29; CHECK-NEXT: call void @llvm.set.loop.iterations.i32(i32 [[N]])30; CHECK-NEXT: br label [[FOR_BODY15_US:%.*]]31; CHECK: for.body15.us:32; CHECK-NEXT: [[J10_055_US:%.*]] = phi i32 [ [[INC26_US:%.*]], [[FOR_BODY15_US]] ], [ [[J10_055_US_PH]], [[FOR_BODY15_US_PREHEADER]] ]33; CHECK-NEXT: [[ARRAYIDX16_US:%.*]] = getelementptr inbounds i16, ptr [[OFF]], i32 [[J10_055_US]]34; CHECK-NEXT: [[L0:%.*]] = load i16, ptr [[ARRAYIDX16_US]], align 235; CHECK-NEXT: [[ARRAYIDX18_US:%.*]] = getelementptr inbounds i16, ptr [[DATA]], i32 [[J10_055_US]]36; CHECK-NEXT: [[L1:%.*]] = load i16, ptr [[ARRAYIDX18_US]], align 237; CHECK-NEXT: [[ADD20_US:%.*]] = add i16 [[L1]], [[L0]]38; CHECK-NEXT: [[ADD23_US:%.*]] = add i32 [[J10_055_US]], [[MUL_US]]39; CHECK-NEXT: [[ARRAYIDX24_US:%.*]] = getelementptr inbounds i16, ptr [[DST:%.*]], i32 [[ADD23_US]]40; CHECK-NEXT: store i16 [[ADD20_US]], ptr [[ARRAYIDX24_US]], align 241; CHECK-NEXT: [[INC26_US]] = add nuw nsw i32 [[J10_055_US]], 142; CHECK-NEXT: [[TMP1:%.*]] = call i1 @llvm.loop.decrement.i32(i32 1)43; CHECK-NEXT: br i1 [[TMP1]], label [[FOR_BODY15_US]], label [[FOR_COND_CLEANUP14_US]]44; CHECK: for.cond.cleanup14.us:45; CHECK-NEXT: [[INC29_US]] = add nuw i32 [[I_057_US]], 146; CHECK-NEXT: [[EXITCOND94:%.*]] = icmp eq i32 [[INC29_US]], [[N]]47; CHECK-NEXT: br i1 [[EXITCOND94]], label [[FOR_COND_CLEANUP]], label [[FOR_COND1_PREHEADER_US]]48; CHECK: for.cond.cleanup:49; CHECK-NEXT: ret void50;51entry:52 %cmp252 = icmp sgt i32 %n, 053 br i1 %cmp252, label %for.cond1.preheader.us, label %for.cond.cleanup54 55for.cond1.preheader.us: ; preds = %entry, %for.cond.cleanup14.us56 %i.057.us = phi i32 [ %inc29.us, %for.cond.cleanup14.us ], [ 0, %entry ]57 %mul.us = mul i32 %i.057.us, %n58 br label %for.body4.us59 60for.body4.us: ; preds = %for.body4.us, %for.cond1.preheader.us61 %j.053.us = phi i32 [ 0, %for.cond1.preheader.us ], [ %inc.us, %for.body4.us ]62 %arrayidx.us = getelementptr inbounds i16, ptr %off, i32 %j.053.us63 %l2 = load i16, ptr %arrayidx.us, align 264 %arrayidx5.us = getelementptr inbounds i16, ptr %data, i32 %j.053.us65 %l3 = load i16, ptr %arrayidx5.us, align 266 %add.us = add i16 %l3, %l267 %add8.us = add i32 %j.053.us, %mul.us68 %arrayidx9.us = getelementptr inbounds i16, ptr %data, i32 %add8.us69 store i16 %add.us, ptr %arrayidx9.us, align 270 %inc.us = add nuw nsw i32 %j.053.us, 171 %exitcond = icmp eq i32 %inc.us, %n72 br i1 %exitcond, label %for.body15.us, label %for.body4.us73 74for.body15.us: ; preds = %for.body4.us, %for.body15.us75 %j10.055.us = phi i32 [ %inc26.us, %for.body15.us ], [ 0, %for.body4.us ]76 %arrayidx16.us = getelementptr inbounds i16, ptr %off, i32 %j10.055.us77 %l0 = load i16, ptr %arrayidx16.us, align 278 %arrayidx18.us = getelementptr inbounds i16, ptr %data, i32 %j10.055.us79 %l1 = load i16, ptr %arrayidx18.us, align 280 %add20.us = add i16 %l1, %l081 %add23.us = add i32 %j10.055.us, %mul.us82 %arrayidx24.us = getelementptr inbounds i16, ptr %dst, i32 %add23.us83 store i16 %add20.us, ptr %arrayidx24.us, align 284 %inc26.us = add nuw nsw i32 %j10.055.us, 185 %exitcond93 = icmp eq i32 %inc26.us, %n86 br i1 %exitcond93, label %for.cond.cleanup14.us, label %for.body15.us87 88for.cond.cleanup14.us: ; preds = %for.body15.us89 %inc29.us = add nuw i32 %i.057.us, 190 %exitcond94 = icmp eq i32 %inc29.us, %n91 br i1 %exitcond94, label %for.cond.cleanup, label %for.cond1.preheader.us92 93for.cond.cleanup: ; preds = %for.cond.cleanup14.us, %entry94 ret void95}96