mirror of
https://github.com/llvm/llvm-project.git
synced 2025-04-29 14:16:07 +00:00
b9808e5660
Loop unrolling tends to produce chains of `%x1 = add %x0, 1; %x2 = add %x1, 1; ...` with one add per unrolled iteration. This patch simplifies these adds to `%xN = add %x0, N` directly during unrolling, rather than waiting for InstCombine to do so. The motivation for this is that having a single add (rather than an add chain) on the induction variable makes it a simple recurrence, which we specially recognize in a number of places. This allows InstCombine to directly perform folds with that knowledge, instead of first folding the add chains, and then doing other folds in another InstCombine iteration. Due to the reduced number of InstCombine iterations, this also results in a small compile-time improvement. Differential Revision: https://reviews.llvm.org/D153540
190 lines
12 KiB
LLVM
190 lines
12 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -passes=loop-unroll -unroll-runtime -unroll-allow-remainder -unroll-count=4 -unroll-remainder -S %s -o - | FileCheck %s
|
|
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
target triple = "x86_64-unknown-linux-gnu"
|
|
|
|
@b = common local_unnamed_addr global i32 0, align 4, !dbg !0
|
|
@a = common local_unnamed_addr global ptr null, align 8, !dbg !6
|
|
|
|
; Test that loop remainder unrolling doesn't corrupt debuginfo. This example
|
|
; used to cause an assert, but also test that the unrolled backwards branches
|
|
; have the same DILocation.
|
|
|
|
define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
|
|
;
|
|
; CHECK-LABEL: @func_c(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[DOTPR:%.*]] = load i32, ptr @b, align 4, !dbg [[DBG17:![0-9]+]], !tbaa [[TBAA20:![0-9]+]]
|
|
; CHECK-NEXT: [[TOBOOL1:%.*]] = icmp eq i32 [[DOTPR]], 0, !dbg [[DBG24:![0-9]+]]
|
|
; CHECK-NEXT: br i1 [[TOBOOL1]], label [[FOR_END:%.*]], label [[FOR_BODY_LR_PH:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body.lr.ph:
|
|
; CHECK-NEXT: [[A_PROMOTED:%.*]] = load ptr, ptr @a, align 8, !dbg [[DBG25:![0-9]+]], !tbaa [[TBAA26:![0-9]+]]
|
|
; CHECK-NEXT: [[TMP0:%.*]] = sub i32 -2, [[DOTPR]], !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[TMP0]], -2, !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[DOTPR]], [[TMP1]], !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[TMP3:%.*]] = lshr i32 [[TMP0]], 1, !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[TMP4:%.*]] = add nuw i32 [[TMP3]], 1, !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[XTRAITER:%.*]] = and i32 [[TMP4]], 3, !dbg [[DBG24]]
|
|
; CHECK-NEXT: [[LCMP_MOD:%.*]] = icmp ne i32 [[XTRAITER]], 0, !dbg [[DBG24]]
|
|
; CHECK-NEXT: br i1 [[LCMP_MOD]], label [[FOR_BODY_PROL_PREHEADER:%.*]], label [[FOR_BODY_PROL_LOOPEXIT:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body.prol.preheader:
|
|
; CHECK-NEXT: br label [[FOR_BODY_PROL:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body.prol:
|
|
; CHECK-NEXT: [[ARRAYIDX_PROL:%.*]] = getelementptr inbounds i32, ptr [[A_PROMOTED]], i64 1, !dbg [[DBG28:![0-9]+]]
|
|
; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_PROL]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_PROL:%.*]] = sext i32 [[TMP5]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP6:%.*]] = inttoptr i64 [[CONV_PROL]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ADD_PROL:%.*]] = add nsw i32 [[DOTPR]], 2, !dbg [[DBG29:![0-9]+]]
|
|
; CHECK-NEXT: [[PROL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]], !dbg [[DBG24]]
|
|
; CHECK-NEXT: br i1 [[PROL_ITER_CMP]], label [[FOR_BODY_PROL_1:%.*]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body.prol.1:
|
|
; CHECK-NEXT: [[ARRAYIDX_PROL_1:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_PROL_1]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_PROL_1:%.*]] = sext i32 [[TMP7]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP8:%.*]] = inttoptr i64 [[CONV_PROL_1]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ADD_PROL_1:%.*]] = add nsw i32 [[DOTPR]], 4, !dbg [[DBG29]]
|
|
; CHECK-NEXT: [[PROL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]], !dbg [[DBG24]]
|
|
; CHECK-NEXT: br i1 [[PROL_ITER_CMP_1]], label [[FOR_BODY_PROL_2:%.*]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]], !dbg [[DBG24]]
|
|
; CHECK: for.body.prol.2:
|
|
; CHECK-NEXT: [[ARRAYIDX_PROL_2:%.*]] = getelementptr inbounds i32, ptr [[TMP8]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_PROL_2]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_PROL_2:%.*]] = sext i32 [[TMP9]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[CONV_PROL_2]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ADD_PROL_2:%.*]] = add nsw i32 [[DOTPR]], 6, !dbg [[DBG29]]
|
|
; CHECK-NEXT: br label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]]
|
|
; CHECK: for.body.prol.loopexit.unr-lcssa:
|
|
; CHECK-NEXT: [[DOTLCSSA_UNR_PH:%.*]] = phi ptr [ [[TMP6]], [[FOR_BODY_PROL]] ], [ [[TMP8]], [[FOR_BODY_PROL_1]] ], [ [[TMP10]], [[FOR_BODY_PROL_2]] ]
|
|
; CHECK-NEXT: [[DOTUNR_PH:%.*]] = phi ptr [ [[TMP6]], [[FOR_BODY_PROL]] ], [ [[TMP8]], [[FOR_BODY_PROL_1]] ], [ [[TMP10]], [[FOR_BODY_PROL_2]] ]
|
|
; CHECK-NEXT: [[DOTUNR1_PH:%.*]] = phi i32 [ [[ADD_PROL]], [[FOR_BODY_PROL]] ], [ [[ADD_PROL_1]], [[FOR_BODY_PROL_1]] ], [ [[ADD_PROL_2]], [[FOR_BODY_PROL_2]] ]
|
|
; CHECK-NEXT: br label [[FOR_BODY_PROL_LOOPEXIT]], !dbg [[DBG24]]
|
|
; CHECK: for.body.prol.loopexit:
|
|
; CHECK-NEXT: [[DOTLCSSA_UNR:%.*]] = phi ptr [ undef, [[FOR_BODY_LR_PH]] ], [ [[DOTLCSSA_UNR_PH]], [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]] ]
|
|
; CHECK-NEXT: [[DOTUNR:%.*]] = phi ptr [ [[A_PROMOTED]], [[FOR_BODY_LR_PH]] ], [ [[DOTUNR_PH]], [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]] ]
|
|
; CHECK-NEXT: [[DOTUNR1:%.*]] = phi i32 [ [[DOTPR]], [[FOR_BODY_LR_PH]] ], [ [[DOTUNR1_PH]], [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]] ]
|
|
; CHECK-NEXT: [[TMP11:%.*]] = icmp ult i32 [[TMP3]], 3, !dbg [[DBG24]]
|
|
; CHECK-NEXT: br i1 [[TMP11]], label [[FOR_COND_FOR_END_CRIT_EDGE:%.*]], label [[FOR_BODY_LR_PH_NEW:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body.lr.ph.new:
|
|
; CHECK-NEXT: br label [[FOR_BODY:%.*]], !dbg [[DBG24]]
|
|
; CHECK: for.body:
|
|
; CHECK-NEXT: [[TMP12:%.*]] = phi ptr [ [[DOTUNR]], [[FOR_BODY_LR_PH_NEW]] ], [ [[TMP21:%.*]], [[FOR_BODY]] ], !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP13:%.*]] = phi i32 [ [[DOTUNR1]], [[FOR_BODY_LR_PH_NEW]] ], [ [[ADD_3:%.*]], [[FOR_BODY]] ]
|
|
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP12]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV:%.*]] = sext i32 [[TMP14]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP15:%.*]] = inttoptr i64 [[CONV]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_1:%.*]] = sext i32 [[TMP16]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP17:%.*]] = inttoptr i64 [[CONV_1]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[TMP17]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_2:%.*]] = sext i32 [[TMP18]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP19:%.*]] = inttoptr i64 [[CONV_2]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[TMP19]], i64 1, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: [[CONV_3:%.*]] = sext i32 [[TMP20]] to i64, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP21]] = inttoptr i64 [[CONV_3]] to ptr, !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[ADD_3]] = add nsw i32 [[TMP13]], 8, !dbg [[DBG29]]
|
|
; CHECK-NEXT: [[TOBOOL_3:%.*]] = icmp eq i32 [[ADD_3]], 0, !dbg [[DBG24]]
|
|
; CHECK-NEXT: br i1 [[TOBOOL_3]], label [[FOR_COND_FOR_END_CRIT_EDGE_UNR_LCSSA:%.*]], label [[FOR_BODY]], !dbg [[DBG24]], !llvm.loop [[LOOP30:![0-9]+]]
|
|
; CHECK: for.cond.for.end_crit_edge.unr-lcssa:
|
|
; CHECK-NEXT: [[DOTLCSSA_PH:%.*]] = phi ptr [ [[TMP21]], [[FOR_BODY]] ]
|
|
; CHECK-NEXT: br label [[FOR_COND_FOR_END_CRIT_EDGE]], !dbg [[DBG24]]
|
|
; CHECK: for.cond.for.end_crit_edge:
|
|
; CHECK-NEXT: [[DOTLCSSA:%.*]] = phi ptr [ [[DOTLCSSA_UNR]], [[FOR_BODY_PROL_LOOPEXIT]] ], [ [[DOTLCSSA_PH]], [[FOR_COND_FOR_END_CRIT_EDGE_UNR_LCSSA]] ], !dbg [[DBG28]]
|
|
; CHECK-NEXT: [[TMP22:%.*]] = add i32 [[TMP2]], 2, !dbg [[DBG24]]
|
|
; CHECK-NEXT: store ptr [[DOTLCSSA]], ptr @a, align 8, !dbg [[DBG25]], !tbaa [[TBAA26]]
|
|
; CHECK-NEXT: store i32 [[TMP22]], ptr @b, align 4, !dbg [[DBG33:![0-9]+]], !tbaa [[TBAA20]]
|
|
; CHECK-NEXT: br label [[FOR_END]], !dbg [[DBG24]]
|
|
; CHECK: for.end:
|
|
; CHECK-NEXT: ret i32 undef, !dbg [[DBG34:![0-9]+]]
|
|
;
|
|
entry:
|
|
%.pr = load i32, ptr @b, align 4, !dbg !17, !tbaa !20
|
|
%tobool1 = icmp eq i32 %.pr, 0, !dbg !24
|
|
br i1 %tobool1, label %for.end, label %for.body.lr.ph, !dbg !24
|
|
|
|
for.body.lr.ph:
|
|
%a.promoted = load ptr, ptr @a, align 8, !dbg !25, !tbaa !26
|
|
%0 = sub i32 -2, %.pr, !dbg !24
|
|
%1 = and i32 %0, -2, !dbg !24
|
|
%2 = add i32 %.pr, %1, !dbg !24
|
|
br label %for.body, !dbg !24
|
|
|
|
for.body:
|
|
%3 = phi ptr [ %a.promoted, %for.body.lr.ph ], [ %6, %for.body ], !dbg !28
|
|
%4 = phi i32 [ %.pr, %for.body.lr.ph ], [ %add, %for.body ]
|
|
%arrayidx = getelementptr inbounds i32, ptr %3, i64 1, !dbg !28
|
|
%5 = load i32, ptr %arrayidx, align 4, !dbg !28, !tbaa !20
|
|
%conv = sext i32 %5 to i64, !dbg !28
|
|
%6 = inttoptr i64 %conv to ptr, !dbg !28
|
|
%add = add nsw i32 %4, 2, !dbg !29
|
|
%tobool = icmp eq i32 %add, 0, !dbg !24
|
|
br i1 %tobool, label %for.cond.for.end_crit_edge, label %for.body, !dbg !24, !llvm.loop !30
|
|
|
|
for.cond.for.end_crit_edge:
|
|
%7 = add i32 %2, 2, !dbg !24
|
|
store ptr %6, ptr @a, align 8, !dbg !25, !tbaa !26
|
|
store i32 %7, ptr @b, align 4, !dbg !32, !tbaa !20
|
|
br label %for.end, !dbg !24
|
|
|
|
for.end:
|
|
ret i32 undef, !dbg !33
|
|
}
|
|
|
|
define void @func_d() local_unnamed_addr #1 !dbg !34 {
|
|
;
|
|
; CHECK-LABEL: @func_d(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: ret void, !dbg [[DBG38:![0-9]+]]
|
|
;
|
|
entry:
|
|
ret void, !dbg !37
|
|
}
|
|
|
|
attributes #0 = { norecurse nounwind uwtable }
|
|
attributes #0 = { norecurse nounwind readnone uwtable }
|
|
|
|
!llvm.dbg.cu = !{!2}
|
|
!llvm.module.flags = !{!10, !11, !12}
|
|
!llvm.ident = !{!13}
|
|
|
|
!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
|
|
!1 = distinct !DIGlobalVariable(name: "b", scope: !2, file: !3, line: 2, type: !9, isLocal: false, isDefinition: true)
|
|
!2 = distinct !DICompileUnit(language: DW_LANG_C99, file: !3, producer: "clang version 6.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, globals: !5)
|
|
!3 = !DIFile(filename: "loop.c", directory: "/work/projects/src/tests/unroll-debug-info")
|
|
!4 = !{}
|
|
!5 = !{!6, !0}
|
|
!6 = !DIGlobalVariableExpression(var: !7, expr: !DIExpression())
|
|
!7 = distinct !DIGlobalVariable(name: "a", scope: !2, file: !3, line: 1, type: !8, isLocal: false, isDefinition: true)
|
|
!8 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !9, size: 64)
|
|
!9 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
|
|
!10 = !{i32 2, !"Dwarf Version", i32 4}
|
|
!11 = !{i32 2, !"Debug Info Version", i32 3}
|
|
!12 = !{i32 1, !"wchar_size", i32 4}
|
|
!13 = !{!"clang version 6.0.0"}
|
|
!14 = distinct !DISubprogram(name: "c", scope: !3, file: !3, line: 3, type: !15, isLocal: false, isDefinition: true, scopeLine: 3, isOptimized: true, unit: !2, retainedNodes: !4)
|
|
!15 = !DISubroutineType(types: !16)
|
|
!16 = !{!9}
|
|
!17 = !DILocation(line: 4, column: 12, scope: !18)
|
|
!18 = distinct !DILexicalBlock(scope: !19, file: !3, line: 4, column: 5)
|
|
!19 = distinct !DILexicalBlock(scope: !14, file: !3, line: 4, column: 5)
|
|
!20 = !{!21, !21, i64 0}
|
|
!21 = !{!"int", !22, i64 0}
|
|
!22 = !{!"omnipotent char", !23, i64 0}
|
|
!23 = !{!"Simple C/C++ TBAA"}
|
|
!24 = !DILocation(line: 4, column: 5, scope: !19)
|
|
!25 = !DILocation(line: 5, column: 13, scope: !18)
|
|
!26 = !{!27, !27, i64 0}
|
|
!27 = !{!"any pointer", !22, i64 0}
|
|
!28 = !DILocation(line: 5, column: 15, scope: !18)
|
|
!29 = !DILocation(line: 4, column: 21, scope: !18)
|
|
!30 = distinct !{!30, !24, !31}
|
|
!31 = !DILocation(line: 5, column: 18, scope: !19)
|
|
!32 = !DILocation(line: 4, column: 17, scope: !18)
|
|
!33 = !DILocation(line: 6, column: 1, scope: !14)
|
|
!34 = distinct !DISubprogram(name: "d", scope: !3, file: !3, line: 7, type: !35, isLocal: false, isDefinition: true, scopeLine: 7, isOptimized: true, unit: !2, retainedNodes: !4)
|
|
!35 = !DISubroutineType(types: !36)
|
|
!36 = !{null}
|
|
!37 = !DILocation(line: 7, column: 11, scope: !34)
|