199 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt %s -S -riscv-codegenprepare -mtriple=riscv64 | FileCheck %s3; RUN: opt %s -S -passes=riscv-codegenprepare -mtriple=riscv64 | FileCheck %s4 5; Make sure we convert the 4294967294 in for.body.preheader.new to -2 based on6; the upper 33 bits being zero by the dominating condition %cmp3.7define void @test2(ptr nocapture noundef %a, i32 noundef signext %n) {8; CHECK-LABEL: @test2(9; CHECK-NEXT: entry:10; CHECK-NEXT: [[CMP3:%.*]] = icmp sgt i32 [[N:%.*]], 011; CHECK-NEXT: br i1 [[CMP3]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]12; CHECK: for.body.preheader:13; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i6414; CHECK-NEXT: [[XTRAITER:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 115; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[N]], 116; CHECK-NEXT: br i1 [[TMP0]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]17; CHECK: for.body.preheader.new:18; CHECK-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[WIDE_TRIP_COUNT]], -219; CHECK-NEXT: br label [[FOR_BODY:%.*]]20; CHECK: for.cond.cleanup.loopexit.unr-lcssa:21; CHECK-NEXT: [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT_1:%.*]], [[FOR_BODY]] ]22; CHECK-NEXT: [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 023; CHECK-NEXT: br i1 [[LCMP_MOD_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY_EPIL:%.*]]24; CHECK: for.body.epil:25; CHECK-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV_UNR]]26; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX_EPIL]], align 427; CHECK-NEXT: [[ADD_EPIL:%.*]] = add nsw i32 [[TMP1]], 428; CHECK-NEXT: store i32 [[ADD_EPIL]], ptr [[ARRAYIDX_EPIL]], align 429; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]30; CHECK: for.cond.cleanup:31; CHECK-NEXT: ret void32; CHECK: for.body:33; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_1]], [[FOR_BODY]] ]34; CHECK-NEXT: [[NITER:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[FOR_BODY]] ]35; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]36; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX]], align 437; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP2]], 438; CHECK-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX]], align 439; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = or i64 [[INDVARS_IV]], 140; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]41; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 442; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP3]], 443; CHECK-NEXT: store i32 [[ADD_1]], ptr [[ARRAYIDX_1]], align 444; CHECK-NEXT: [[INDVARS_IV_NEXT_1]] = add nuw nsw i64 [[INDVARS_IV]], 245; CHECK-NEXT: [[NITER_NEXT_1]] = add i64 [[NITER]], 246; CHECK-NEXT: [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NEXT_1]], [[UNROLL_ITER]]47; CHECK-NEXT: br i1 [[NITER_NCMP_1]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]48;49entry:50 %cmp3 = icmp sgt i32 %n, 051 br i1 %cmp3, label %for.body.preheader, label %for.cond.cleanup52 53for.body.preheader: ; preds = %entry54 %wide.trip.count = zext nneg i32 %n to i6455 %xtraiter = and i64 %wide.trip.count, 156 %0 = icmp eq i32 %n, 157 br i1 %0, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body.preheader.new58 59for.body.preheader.new: ; preds = %for.body.preheader60 %unroll_iter = and i64 %wide.trip.count, 429496729461 br label %for.body62 63for.cond.cleanup.loopexit.unr-lcssa: ; preds = %for.body, %for.body.preheader64 %indvars.iv.unr = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next.1, %for.body ]65 %lcmp.mod.not = icmp eq i64 %xtraiter, 066 br i1 %lcmp.mod.not, label %for.cond.cleanup, label %for.body.epil67 68for.body.epil: ; preds = %for.cond.cleanup.loopexit.unr-lcssa69 %arrayidx.epil = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.unr70 %1 = load i32, ptr %arrayidx.epil, align 471 %add.epil = add nsw i32 %1, 472 store i32 %add.epil, ptr %arrayidx.epil, align 473 br label %for.cond.cleanup74 75for.cond.cleanup: ; preds = %for.body.epil, %for.cond.cleanup.loopexit.unr-lcssa, %entry76 ret void77 78for.body: ; preds = %for.body, %for.body.preheader.new79 %indvars.iv = phi i64 [ 0, %for.body.preheader.new ], [ %indvars.iv.next.1, %for.body ]80 %niter = phi i64 [ 0, %for.body.preheader.new ], [ %niter.next.1, %for.body ]81 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv82 %2 = load i32, ptr %arrayidx, align 483 %add = add nsw i32 %2, 484 store i32 %add, ptr %arrayidx, align 485 %indvars.iv.next = or i64 %indvars.iv, 186 %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next87 %3 = load i32, ptr %arrayidx.1, align 488 %add.1 = add nsw i32 %3, 489 store i32 %add.1, ptr %arrayidx.1, align 490 %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 291 %niter.next.1 = add i64 %niter, 292 %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter93 br i1 %niter.ncmp.1, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body94}95 96; Make sure we do not change 4294967295 to -1 here.97define i64 @bug(i32 %x) {98; CHECK-LABEL: @bug(99; CHECK-NEXT: [[A:%.*]] = sext i32 [[X:%.*]] to i64100; CHECK-NEXT: [[B:%.*]] = and i64 [[A]], 4294967295101; CHECK-NEXT: ret i64 [[B]]102;103 %a = sext i32 %x to i64104 %b = and i64 %a, 4294967295105 ret i64 %b106}107 108define i1 @widen_anyof_rdx(ptr %p, i64 %n) {109; CHECK-LABEL: @widen_anyof_rdx(110; CHECK-NEXT: entry:111; CHECK-NEXT: br label [[LOOP:%.*]]112; CHECK: loop:113; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]114; CHECK-NEXT: [[TMP0:%.*]] = phi <vscale x 4 x i8> [ zeroinitializer, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]115; CHECK-NEXT: [[AVL:%.*]] = sub i64 [[N:%.*]], [[IV]]116; CHECK-NEXT: [[EVL:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 4, i1 true)117; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[P:%.*]], i64 [[IV]]118; CHECK-NEXT: [[X:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr [[GEP]], <vscale x 4 x i1> splat (i1 true), i32 [[EVL]])119; CHECK-NEXT: [[CMP:%.*]] = icmp ne <vscale x 4 x i32> [[X]], zeroinitializer120; CHECK-NEXT: [[TMP1]] = call <vscale x 4 x i8> @llvm.vp.merge.nxv4i8(<vscale x 4 x i1> [[CMP]], <vscale x 4 x i8> splat (i8 1), <vscale x 4 x i8> [[TMP0]], i32 [[EVL]])121; CHECK-NEXT: [[TMP4:%.*]] = trunc <vscale x 4 x i8> [[TMP1]] to <vscale x 4 x i1>122; CHECK-NEXT: [[EVL_ZEXT:%.*]] = zext i32 [[EVL]] to i64123; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], [[EVL_ZEXT]]124; CHECK-NEXT: [[DONE:%.*]] = icmp sge i64 [[IV_NEXT]], [[N]]125; CHECK-NEXT: br i1 [[DONE]], label [[EXIT:%.*]], label [[LOOP]]126; CHECK: exit:127; CHECK-NEXT: [[RES:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[TMP4]])128; CHECK-NEXT: ret i1 [[RES]]129;130entry:131 br label %loop132loop:133 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]134 %phi = phi <vscale x 4 x i1> [ zeroinitializer, %entry ], [ %rec, %loop ]135 %avl = sub i64 %n, %iv136 %evl = call i32 @llvm.experimental.get.vector.length(i64 %avl, i32 4, i1 true)137 138 %gep = getelementptr i32, ptr %p, i64 %iv139 %x = call <vscale x 4 x i32> @llvm.vp.load(ptr %gep, <vscale x 4 x i1> splat (i1 true), i32 %evl)140 %cmp = icmp ne <vscale x 4 x i32> %x, zeroinitializer141 %rec = call <vscale x 4 x i1> @llvm.vp.merge(<vscale x 4 x i1> %cmp, <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i1> %phi, i32 %evl)142 143 %evl.zext = zext i32 %evl to i64144 %iv.next = add i64 %iv, %evl.zext145 %done = icmp sge i64 %iv.next, %n146 br i1 %done, label %exit, label %loop147exit:148 %res = call i1 @llvm.vector.reduce.or(<vscale x 4 x i1> %rec)149 ret i1 %res150}151 152 153define i1 @widen_anyof_rdx_use_in_loop(ptr %p, i64 %n) {154; CHECK-LABEL: @widen_anyof_rdx_use_in_loop(155; CHECK-NEXT: entry:156; CHECK-NEXT: br label [[LOOP:%.*]]157; CHECK: loop:158; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]159; CHECK-NEXT: [[TMP0:%.*]] = phi <vscale x 4 x i8> [ zeroinitializer, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]160; CHECK-NEXT: [[AVL:%.*]] = sub i64 [[N:%.*]], [[IV]]161; CHECK-NEXT: [[EVL:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 4, i1 true)162; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[P:%.*]], i64 [[IV]]163; CHECK-NEXT: [[X:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr [[GEP]], <vscale x 4 x i1> splat (i1 true), i32 [[EVL]])164; CHECK-NEXT: [[CMP:%.*]] = icmp ne <vscale x 4 x i32> [[X]], zeroinitializer165; CHECK-NEXT: [[TMP1]] = call <vscale x 4 x i8> @llvm.vp.merge.nxv4i8(<vscale x 4 x i1> [[CMP]], <vscale x 4 x i8> splat (i8 1), <vscale x 4 x i8> [[TMP0]], i32 [[EVL]])166; CHECK-NEXT: [[REC:%.*]] = trunc <vscale x 4 x i8> [[TMP1]] to <vscale x 4 x i1>167; CHECK-NEXT: store <vscale x 4 x i1> [[REC]], ptr [[GEP]], align 1168; CHECK-NEXT: [[EVL_ZEXT:%.*]] = zext i32 [[EVL]] to i64169; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], [[EVL_ZEXT]]170; CHECK-NEXT: [[DONE:%.*]] = icmp sge i64 [[IV_NEXT]], [[N]]171; CHECK-NEXT: br i1 [[DONE]], label [[EXIT:%.*]], label [[LOOP]]172; CHECK: exit:173; CHECK-NEXT: [[RES:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[REC]])174; CHECK-NEXT: ret i1 [[RES]]175;176entry:177 br label %loop178loop:179 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]180 %phi = phi <vscale x 4 x i1> [ zeroinitializer, %entry ], [ %rec, %loop ]181 %avl = sub i64 %n, %iv182 %evl = call i32 @llvm.experimental.get.vector.length(i64 %avl, i32 4, i1 true)183 184 %gep = getelementptr i32, ptr %p, i64 %iv185 %x = call <vscale x 4 x i32> @llvm.vp.load(ptr %gep, <vscale x 4 x i1> splat (i1 true), i32 %evl)186 %cmp = icmp ne <vscale x 4 x i32> %x, zeroinitializer187 %rec = call <vscale x 4 x i1> @llvm.vp.merge(<vscale x 4 x i1> %cmp, <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i1> %phi, i32 %evl)188 189 store <vscale x 4 x i1> %rec, ptr %gep190 191 %evl.zext = zext i32 %evl to i64192 %iv.next = add i64 %iv, %evl.zext193 %done = icmp sge i64 %iv.next, %n194 br i1 %done, label %exit, label %loop195exit:196 %res = call i1 @llvm.vector.reduce.or(<vscale x 4 x i1> %rec)197 ret i1 %res198}199