Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
; REQUIRES: object-emission
|
|
|
|
|
|
|
|
|
|
; RUN: llc -mtriple=i386-linux-gnu -filetype=obj -relocation-model=pic %s -o /dev/null
|
|
|
|
|
|
|
|
|
|
; Derived from the test case in PR20367, there's nothing really positive to
|
|
|
|
|
; test here (hence no FileCheck, etc). All that was wrong is that the debug info
|
|
|
|
|
; intrinsics (introduced by inlining) in 'f1' were causing codegen to crash, but
|
|
|
|
|
; since 'f1' is a nodebug function, there's no positive outcome to confirm, just
|
|
|
|
|
; that debug info doesn't get in the way/cause a crash.
|
|
|
|
|
|
|
|
|
|
; The test case isn't particularly well reduced/tidy, but as simple as I could
|
|
|
|
|
; get the C++ source. I assume the complexity is mostly just about producing a
|
|
|
|
|
; certain amount of register pressure, so it might be able to be simplified/made
|
|
|
|
|
; more uniform.
|
|
|
|
|
|
|
|
|
|
; Generated from:
|
|
|
|
|
; $ clang-tot -cc1 -triple i386 -emit-obj -g -O3 repro.cpp
|
|
|
|
|
; void sink(const void *);
|
|
|
|
|
; int source();
|
|
|
|
|
; void f3(int);
|
|
|
|
|
;
|
|
|
|
|
; extern bool b;
|
|
|
|
|
;
|
|
|
|
|
; struct string {
|
|
|
|
|
; unsigned *mem;
|
|
|
|
|
; };
|
|
|
|
|
;
|
|
|
|
|
; extern string &str;
|
|
|
|
|
;
|
|
|
|
|
; inline __attribute__((always_inline)) void s2(string *lhs) { sink(lhs->mem); }
|
|
|
|
|
; inline __attribute__((always_inline)) void f() {
|
|
|
|
|
; string str2;
|
|
|
|
|
; s2(&str2);
|
|
|
|
|
; sink(&str2);
|
|
|
|
|
; }
|
|
|
|
|
; void __attribute__((nodebug)) f1() {
|
|
|
|
|
; for (int iter = 0; iter != 2; ++iter) {
|
|
|
|
|
; f();
|
|
|
|
|
; sink(str.mem);
|
|
|
|
|
; if (b) return;
|
|
|
|
|
; }
|
|
|
|
|
; }
|
|
|
|
|
|
|
|
|
|
%struct.string = type { i32* }
|
|
|
|
|
|
|
|
|
|
@str = external constant %struct.string*
|
|
|
|
|
@b = external global i8
|
|
|
|
|
|
|
|
|
|
; Function Attrs: nounwind
|
|
|
|
|
define void @_Z2f1v() #0 {
|
|
|
|
|
entry:
|
|
|
|
|
%str2.i = alloca %struct.string, align 4
|
|
|
|
|
%0 = bitcast %struct.string* %str2.i to i8*, !dbg !26
|
|
|
|
|
%1 = load %struct.string** @str, align 4
|
|
|
|
|
%mem = getelementptr inbounds %struct.string* %1, i32 0, i32 0
|
|
|
|
|
br label %for.body
|
|
|
|
|
|
|
|
|
|
for.body: ; preds = %for.body, %entry
|
|
|
|
|
%iter.02 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
|
|
|
|
|
call void @llvm.lifetime.start(i64 4, i8* %0), !dbg !26
|
2014-10-03 20:01:09 +00:00
|
|
|
call void @llvm.dbg.value(metadata !{%struct.string* %str2.i}, i64 0, metadata !16, metadata !{metadata !"0x102"}) #3, !dbg !26
|
|
|
|
|
call void @llvm.dbg.value(metadata !{%struct.string* %str2.i}, i64 0, metadata !27, metadata !{metadata !"0x102"}) #3, !dbg !29
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
call void @_Z4sinkPKv(i8* undef) #3, !dbg !29
|
|
|
|
|
call void @_Z4sinkPKv(i8* %0) #3, !dbg !30
|
|
|
|
|
call void @llvm.lifetime.end(i64 4, i8* %0), !dbg !31
|
|
|
|
|
%2 = load i32** %mem, align 4, !tbaa !32
|
|
|
|
|
%3 = bitcast i32* %2 to i8*
|
|
|
|
|
call void @_Z4sinkPKv(i8* %3) #3
|
|
|
|
|
%4 = load i8* @b, align 1, !tbaa !37, !range !39
|
|
|
|
|
%tobool = icmp ne i8 %4, 0
|
|
|
|
|
%inc = add nsw i32 %iter.02, 1
|
|
|
|
|
%cmp = icmp eq i32 %inc, 2
|
|
|
|
|
%or.cond = or i1 %tobool, %cmp
|
|
|
|
|
br i1 %or.cond, label %for.end, label %for.body
|
|
|
|
|
|
|
|
|
|
for.end: ; preds = %for.body
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare void @_Z4sinkPKv(i8*) #1
|
|
|
|
|
|
|
|
|
|
; Function Attrs: nounwind readnone
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
declare void @llvm.dbg.value(metadata, i64, metadata, metadata) #2
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
|
|
|
|
|
; Function Attrs: nounwind
|
|
|
|
|
declare void @llvm.lifetime.start(i64, i8* nocapture) #3
|
|
|
|
|
|
|
|
|
|
; Function Attrs: nounwind
|
|
|
|
|
declare void @llvm.lifetime.end(i64, i8* nocapture) #3
|
|
|
|
|
|
|
|
|
|
attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-realign-stack" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
|
|
|
|
attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-realign-stack" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
|
|
|
|
attributes #2 = { nounwind readnone }
|
|
|
|
|
attributes #3 = { nounwind }
|
|
|
|
|
|
|
|
|
|
!llvm.dbg.cu = !{!0}
|
|
|
|
|
!llvm.module.flags = !{!23, !24}
|
|
|
|
|
!llvm.ident = !{!25}
|
|
|
|
|
|
2014-10-03 20:01:09 +00:00
|
|
|
!0 = metadata !{metadata !"0x11\004\00clang version 3.5.0 \001\00\000\00\001", metadata !1, metadata !2, metadata !3, metadata !10, metadata !2, metadata !2} ; [ DW_TAG_compile_unit ] [/tmp/dbginfo/<stdin>] [DW_LANG_C_plus_plus]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!1 = metadata !{metadata !"<stdin>", metadata !"/tmp/dbginfo"}
|
|
|
|
|
!2 = metadata !{}
|
|
|
|
|
!3 = metadata !{metadata !4}
|
2014-10-03 20:01:09 +00:00
|
|
|
!4 = metadata !{metadata !"0x13\00string\007\0032\0032\000\000\000", metadata !5, null, null, metadata !6, null, null, metadata !"_ZTS6string"} ; [ DW_TAG_structure_type ] [string] [line 7, size 32, align 32, offset 0] [def] [from ]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!5 = metadata !{metadata !"repro.cpp", metadata !"/tmp/dbginfo"}
|
|
|
|
|
!6 = metadata !{metadata !7}
|
2014-10-03 20:01:09 +00:00
|
|
|
!7 = metadata !{metadata !"0xd\00mem\008\0032\0032\000\000", metadata !5, metadata !"_ZTS6string", metadata !8} ; [ DW_TAG_member ] [mem] [line 8, size 32, align 32, offset 0] [from ]
|
|
|
|
|
!8 = metadata !{metadata !"0xf\00\000\0032\0032\000\000", null, null, metadata !9} ; [ DW_TAG_pointer_type ] [line 0, size 32, align 32, offset 0] [from unsigned int]
|
|
|
|
|
!9 = metadata !{metadata !"0x24\00unsigned int\000\0032\0032\000\000\007", null, null} ; [ DW_TAG_base_type ] [unsigned int] [line 0, size 32, align 32, offset 0, enc DW_ATE_unsigned]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!10 = metadata !{metadata !11, metadata !17}
|
2014-10-03 20:01:09 +00:00
|
|
|
!11 = metadata !{metadata !"0x2e\00f\00f\00_Z1fv\0014\000\001\000\006\00256\001\0014", metadata !5, metadata !12, metadata !13, null, null, null, null, metadata !15} ; [ DW_TAG_subprogram ] [line 14] [def] [f]
|
|
|
|
|
!12 = metadata !{metadata !"0x29", metadata !5} ; [ DW_TAG_file_type ] [/tmp/dbginfo/repro.cpp]
|
|
|
|
|
!13 = metadata !{metadata !"0x15\00\000\000\000\000\000\000", i32 0, null, null, metadata !14, null, null, null} ; [ DW_TAG_subroutine_type ] [line 0, size 0, align 0, offset 0] [from ]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!14 = metadata !{null}
|
|
|
|
|
!15 = metadata !{metadata !16}
|
2014-10-03 20:01:09 +00:00
|
|
|
!16 = metadata !{metadata !"0x100\00str2\0015\000", metadata !11, metadata !12, metadata !"_ZTS6string"} ; [ DW_TAG_auto_variable ] [str2] [line 15]
|
|
|
|
|
!17 = metadata !{metadata !"0x2e\00s2\00s2\00_Z2s2P6string\0013\000\001\000\006\00256\001\0013", metadata !5, metadata !12, metadata !18, null, null, null, null, metadata !21} ; [ DW_TAG_subprogram ] [line 13] [def] [s2]
|
|
|
|
|
!18 = metadata !{metadata !"0x15\00\000\000\000\000\000\000", i32 0, null, null, metadata !19, null, null, null} ; [ DW_TAG_subroutine_type ] [line 0, size 0, align 0, offset 0] [from ]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!19 = metadata !{null, metadata !20}
|
2014-10-03 20:01:09 +00:00
|
|
|
!20 = metadata !{metadata !"0xf\00\000\0032\0032\000\000", null, null, metadata !"_ZTS6string"} ; [ DW_TAG_pointer_type ] [line 0, size 32, align 32, offset 0] [from _ZTS6string]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!21 = metadata !{metadata !22}
|
2014-10-03 20:01:09 +00:00
|
|
|
!22 = metadata !{metadata !"0x101\00lhs\0016777229\000", metadata !17, metadata !12, metadata !20} ; [ DW_TAG_arg_variable ] [lhs] [line 13]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!23 = metadata !{i32 2, metadata !"Dwarf Version", i32 4}
|
2014-10-03 20:01:09 +00:00
|
|
|
!24 = metadata !{i32 2, metadata !"Debug Info Version", i32 2}
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!25 = metadata !{metadata !"clang version 3.5.0 "}
|
|
|
|
|
!26 = metadata !{i32 15, i32 0, metadata !11, null}
|
2014-10-03 20:01:09 +00:00
|
|
|
!27 = metadata !{metadata !"0x101\00lhs\0016777229\000", metadata !17, metadata !12, metadata !20, metadata !28} ; [ DW_TAG_arg_variable ] [lhs] [line 13]
|
Recommit r212203: Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.
Reverted by Eric Christopher (Thanks!) in r212203 after Bob Wilson
reported LTO issues. Duncan Exon Smith and Aditya Nandakumar helped
provide a reduced reproduction, though the failure wasn't too hard to
guess, and even easier with the example to confirm.
The assertion that the subprogram metadata associated with an
llvm::Function matches the scope data referenced by the DbgLocs on the
instructions in that function is not valid under LTO. In LTO, a C++
inline function might exist in multiple CUs and the subprogram metadata
nodes will refer to the same llvm::Function. In this case, depending on
the order of the CUs, the first intance of the subprogram metadata may
not be the one referenced by the instructions in that function and the
assertion will fail.
A test case (test/DebugInfo/cross-cu-linkonce-distinct.ll) is added, the
assertion removed and a comment added to explain this situation.
This was then reverted again in r213581 as it caused PR20367. The root
cause of this was the early exit in LiveDebugVariables meant that
spurious DBG_VALUE intrinsics that referenced dead variables were not
removed, causing an assertion/crash later on. The fix is to have
LiveDebugVariables strip all DBG_VALUE intrinsics in functions without
debug info as they're not needed anyway. Test case added to cover this
situation (that occurs when a debug-having function is inlined into a
nodebug function) in test/DebugInfo/X86/nodebug_with_debug_loc.ll
Original commit message:
If a function isn't actually in a CU's subprogram list in the debug info
metadata, ignore all the DebugLocs and don't try to build scopes, track
variables, etc.
While this is possibly a minor optimization, it's also a correctness fix
for an incoming patch that will add assertions to LexicalScopes and the
debug info verifier to ensure that all scope chains lead to debug info
for the current function.
Fix up a few test cases that had broken/incomplete debug info that could
violate this constraint.
Add a test case where this occurs by design (inlining a
debug-info-having function in an attribute nodebug function - we want
this to work because /if/ the nodebug function is then inlined into a
debug-info-having function, it should be fine (and will work fine - we
just stitch the scopes up as usual), but should the inlining not happen
we need to not assert fail either).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213952 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 16:10:16 +00:00
|
|
|
!28 = metadata !{i32 16, i32 0, metadata !11, null}
|
|
|
|
|
!29 = metadata !{i32 13, i32 0, metadata !17, metadata !28}
|
|
|
|
|
!30 = metadata !{i32 17, i32 0, metadata !11, null}
|
|
|
|
|
!31 = metadata !{i32 18, i32 0, metadata !11, null}
|
|
|
|
|
!32 = metadata !{metadata !33, metadata !34, i64 0}
|
|
|
|
|
!33 = metadata !{metadata !"_ZTS6string", metadata !34, i64 0}
|
|
|
|
|
!34 = metadata !{metadata !"any pointer", metadata !35, i64 0}
|
|
|
|
|
!35 = metadata !{metadata !"omnipotent char", metadata !36, i64 0}
|
|
|
|
|
!36 = metadata !{metadata !"Simple C/C++ TBAA"}
|
|
|
|
|
!37 = metadata !{metadata !38, metadata !38, i64 0}
|
|
|
|
|
!38 = metadata !{metadata !"bool", metadata !35, i64 0}
|
|
|
|
|
!39 = metadata !{i8 0, i8 2}
|
