104 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=nary-reassociate -S | FileCheck %s3 4declare i32 @llvm.smax.i32(i32 %a, i32 %b)5declare i64 @llvm.umin.i64(i64, i64)6 7; This is a negative test. We should not optimize if intermediate result8; has a use outside of optimizable pattern. In other words %smax2 has one9; use from %smax3 and side use from %res2.10define i32 @smax_test1(i32 %a, i32 %b, i32 %c) {11; CHECK-LABEL: @smax_test1(12; CHECK-NEXT: [[C1:%.*]] = icmp sgt i32 [[A:%.*]], [[B:%.*]]13; CHECK-NEXT: [[SMAX1:%.*]] = select i1 [[C1]], i32 [[A]], i32 [[B]]14; CHECK-NEXT: [[C2:%.*]] = icmp sgt i32 [[B]], [[C:%.*]]15; CHECK-NEXT: [[SMAX2:%.*]] = select i1 [[C2]], i32 [[B]], i32 [[C]]16; CHECK-NEXT: [[C3:%.*]] = icmp sgt i32 [[SMAX2]], [[A]]17; CHECK-NEXT: [[SMAX3:%.*]] = select i1 [[C3]], i32 [[SMAX2]], i32 [[A]]18; CHECK-NEXT: [[RES:%.*]] = add i32 [[SMAX1]], [[SMAX3]]19; CHECK-NEXT: [[RES2:%.*]] = add i32 [[RES]], [[SMAX2]]20; CHECK-NEXT: ret i32 [[RES]]21;22 %c1 = icmp sgt i32 %a, %b23 %smax1 = select i1 %c1, i32 %a, i32 %b24 %c2 = icmp sgt i32 %b, %c25 %smax2 = select i1 %c2, i32 %b, i32 %c26 %c3 = icmp sgt i32 %smax2, %a27 %smax3 = select i1 %c3, i32 %smax2, i32 %a28 %res = add i32 %smax1, %smax329 %res2 = add i32 %res, %smax230 ret i32 %res31}32 33; This is a negative test. It similar to the previous one34; but a bit more complex. In particular after first iteration35; e10 is replaced with %e10.nary = call i64 @llvm.umin.i64(i64 %e5, i64 %e).36; No more reassociation should be applied to %e10.nary since37; %e5 has side use in %e6.38define void @test2(i64 %arg) {39; CHECK-LABEL: @test2(40; CHECK-NEXT: bb:41; CHECK-NEXT: [[E:%.*]] = sub i64 undef, 042; CHECK-NEXT: [[E1:%.*]] = sub i64 [[ARG:%.*]], 043; CHECK-NEXT: [[E2:%.*]] = call i64 @llvm.umin.i64(i64 [[E]], i64 [[E1]])44; CHECK-NEXT: [[E3:%.*]] = call i64 @llvm.umin.i64(i64 [[E2]], i64 16384)45; CHECK-NEXT: [[E4:%.*]] = sub i64 [[ARG]], 046; CHECK-NEXT: [[E5:%.*]] = call i64 @llvm.umin.i64(i64 [[E4]], i64 16384)47; CHECK-NEXT: [[E6:%.*]] = icmp ugt i64 [[E5]], 048; CHECK-NEXT: [[E7:%.*]] = sub i64 undef, 049; CHECK-NEXT: [[E10_NARY:%.*]] = call i64 @llvm.umin.i64(i64 [[E5]], i64 [[E7]])50; CHECK-NEXT: unreachable51;52bb:53 %e = sub i64 undef, 054 %e1 = sub i64 %arg, 055 %e2 = call i64 @llvm.umin.i64(i64 %e, i64 %e1)56 %e3 = call i64 @llvm.umin.i64(i64 %e2, i64 16384)57 %e4 = sub i64 %arg, 058 %e5 = call i64 @llvm.umin.i64(i64 %e4, i64 16384)59 %e6 = icmp ugt i64 %e5, 060 %e7 = sub i64 undef, 061 %e8 = sub i64 %arg, 062 %e9 = call i64 @llvm.umin.i64(i64 %e7, i64 %e8)63 %e10 = call i64 @llvm.umin.i64(i64 %e9, i64 16384)64 unreachable65}66 67; Make sure we don't fall into infinte loop optimizing %sel5.68; The subtle thing is that %sel3 is min/max as well and69; there is "unexpected" reassociation coming from SCEV Expander70; during %sel5 rewrite. That results in a new chain of min/max71; which is matched on the next iteration.72define i32 @nary_infinite_loop_minmax(i32 %d0, i32 %d1, i32 %d2, i32 %d3) {73; CHECK-LABEL: @nary_infinite_loop_minmax(74; CHECK-NEXT: [[CMP0:%.*]] = icmp slt i32 [[D2:%.*]], [[D1:%.*]]75; CHECK-NEXT: [[SEL0:%.*]] = select i1 [[CMP0]], i32 [[D1]], i32 [[D2]]76; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 [[D3:%.*]], [[D0:%.*]]77; CHECK-NEXT: [[SEL1:%.*]] = select i1 [[CMP1]], i32 [[D0]], i32 [[D3]]78; CHECK-NEXT: [[CMP2:%.*]] = icmp slt i32 [[SEL1]], [[SEL0]]79; CHECK-NEXT: [[SEL2:%.*]] = select i1 [[CMP2]], i32 [[SEL1]], i32 [[SEL0]]80; CHECK-NEXT: [[CMP3:%.*]] = icmp slt i32 [[D3]], [[D0]]81; CHECK-NEXT: [[SEL3:%.*]] = select i1 [[CMP3]], i32 [[D0]], i32 [[D3]]82; CHECK-NEXT: [[SEL5_NARY:%.*]] = call i32 @llvm.smax.i32(i32 [[SEL0]], i32 [[SEL3]])83; CHECK-NEXT: ret i32 [[SEL5_NARY]]84;85 %cmp0 = icmp slt i32 %d2, %d186 %sel0 = select i1 %cmp0, i32 %d1, i32 %d287 88 %cmp1 = icmp slt i32 %d3, %d089 %sel1 = select i1 %cmp1, i32 %d0, i32 %d390 91 %cmp2 = icmp slt i32 %sel1, %sel092 %sel2 = select i1 %cmp2, i32 %sel1, i32 %sel093 94 %cmp3 = icmp slt i32 %d3, %d095 %sel3 = select i1 %cmp3, i32 %d0, i32 %d396 97 %cmp4 = icmp slt i32 %sel3, %d298 %sel4 = select i1 %cmp4, i32 %d2, i32 %sel399 100 %cmp5 = icmp slt i32 %sel4, %d1101 %sel5 = select i1 %cmp5, i32 %d1, i32 %sel4102 ret i32 %sel5103}104