shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/DebugInfo/unrolled-loop-remainder.ll
Philip Reames 37ead201e6 [runtime-unroll] Use incrementing IVs instead of decrementing ones 
This is one of those wonderful "in theory X doesn't matter, but in practice is does" changes. In this particular case, we shift the IVs inserted by the runtime unroller to clamp iteration count of the loops* from decrementing to incrementing.

Why does this matter?  A couple of reasons:
* SCEV doesn't have a native subtract node.  Instead, all subtracts (A - B) are represented as A + -1 * B and drops any flags invalidated by such.  As a result, SCEV is slightly less good at reasoning about edge cases involving decrementing addrecs than incrementing ones.  (You can see this in the inferred flags in some of the test cases.)
* Other parts of the optimizer produce incrementing IVs, and they're common in idiomatic source language.  We do have support for reversing IVs, but in general if we produce one of each, the pair will persist surprisingly far through the optimizer before being coalesced.  (You can see this looking at nearby phis in the test cases.)

Note that if the hardware prefers decrementing (i.e. zero tested) loops, LSR should convert back immediately before codegen.

* Mostly irrelevant detail: The main loop of the prolog case is handled independently and will simple use the original IV with a changed start value.  We could in theory use this scheme for all iteration clamping, but that's a larger and more invasive change.
2021-11-12 15:44:58 -08:00

193 lines
12 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -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 i32* 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, i32* @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 i32*, i32** @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, i32* [[A_PROMOTED]], i64 1, !dbg [[DBG28:![0-9]+]]
; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[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 i32*, !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, i32* [[TMP6]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[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 i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD_PROL_1:%.*]] = add nsw i32 [[ADD_PROL]], 2, !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, i32* [[TMP8]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP9:%.*]] = load i32, i32* [[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 i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD_PROL_2:%.*]] = add nsw i32 [[ADD_PROL_1]], 2, !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 i32* [ [[TMP6]], [[FOR_BODY_PROL]] ], [ [[TMP8]], [[FOR_BODY_PROL_1]] ], [ [[TMP10]], [[FOR_BODY_PROL_2]] ]
; CHECK-NEXT: [[DOTUNR_PH:%.*]] = phi i32* [ [[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 i32* [ undef, [[FOR_BODY_LR_PH]] ], [ [[DOTLCSSA_UNR_PH]], [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]] ]
; CHECK-NEXT: [[DOTUNR:%.*]] = phi i32* [ [[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 i32* [ [[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, i32* [[TMP12]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP14:%.*]] = load i32, i32* [[ARRAYIDX]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
; CHECK-NEXT: [[CONV:%.*]] = sext i32 [[TMP14]] to i64, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP15:%.*]] = inttoptr i64 [[CONV]] to i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP13]], 2, !dbg [[DBG29]]
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP16:%.*]] = load i32, i32* [[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 i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[ADD]], 2, !dbg [[DBG29]]
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[TMP17]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP18:%.*]] = load i32, i32* [[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 i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[ADD_1]], 2, !dbg [[DBG29]]
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[TMP19]], i64 1, !dbg [[DBG28]]
; CHECK-NEXT: [[TMP20:%.*]] = load i32, i32* [[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 i32*, !dbg [[DBG28]]
; CHECK-NEXT: [[ADD_3]] = add nsw i32 [[ADD_2]], 2, !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 i32* [ [[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 i32* [ [[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 i32* [[DOTLCSSA]], i32** @a, align 8, !dbg [[DBG25]], !tbaa [[TBAA26]]
; CHECK-NEXT: store i32 [[TMP22]], i32* @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, i32* @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 i32*, i32** @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 i32* [ %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, i32* %3, i64 1, !dbg !28
%5 = load i32, i32* %arrayidx, align 4, !dbg !28, !tbaa !20
%conv = sext i32 %5 to i64, !dbg !28
%6 = inttoptr i64 %conv to i32*, !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 i32* %6, i32** @a, align 8, !dbg !25, !tbaa !26
store i32 %7, i32* @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)
Reference in New Issue View Git Blame Copy Permalink