Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
; RUN: llc < %s -march=ppc64 -mtriple=powerpc64-unknown-linux-gnu
|
|
|
|
|
; RUN: llc < %s -march=ppc64le -mtriple=powerpc64le-unknown-linux-gnu
|
|
|
|
|
|
|
|
|
|
%"class.std::__1::locale::id.1580.4307.4610.8491" = type { %"struct.std::__1::once_flag.1579.4306.4609.8490", i32 }
|
|
|
|
|
%"struct.std::__1::once_flag.1579.4306.4609.8490" = type { i64 }
|
|
|
|
|
%"class.Foam::IOerror.1581.4308.4611.8505" = type { %"class.Foam::error.1535.4262.4565.8504", %"class.Foam::string.1530.4257.4560.8499", i32, i32 }
|
|
|
|
|
%"class.Foam::error.1535.4262.4565.8504" = type { %"class.std::exception.1523.4250.4553.8492", [36 x i8], %"class.Foam::string.1530.4257.4560.8499", %"class.Foam::string.1530.4257.4560.8499", i32, i8, i8, %"class.Foam::OStringStream.1534.4261.4564.8503"* }
|
|
|
|
|
%"class.std::exception.1523.4250.4553.8492" = type { i32 (...)** }
|
|
|
|
|
%"class.Foam::OStringStream.1534.4261.4564.8503" = type { %"class.Foam::OSstream.1533.4260.4563.8502" }
|
|
|
|
|
%"class.Foam::OSstream.1533.4260.4563.8502" = type { [50 x i8], %"class.Foam::fileName.1531.4258.4561.8500", %"class.std::__1::basic_ostream.1532.4259.4562.8501"* }
|
|
|
|
|
%"class.Foam::fileName.1531.4258.4561.8500" = type { %"class.Foam::string.1530.4257.4560.8499" }
|
|
|
|
|
%"class.std::__1::basic_ostream.1532.4259.4562.8501" = type { i32 (...)**, [148 x i8] }
|
|
|
|
|
%"class.Foam::string.1530.4257.4560.8499" = type { %"class.std::__1::basic_string.1529.4256.4559.8498" }
|
|
|
|
|
%"class.std::__1::basic_string.1529.4256.4559.8498" = type { %"class.std::__1::__compressed_pair.1528.4255.4558.8497" }
|
|
|
|
|
%"class.std::__1::__compressed_pair.1528.4255.4558.8497" = type { %"class.std::__1::__libcpp_compressed_pair_imp.1527.4254.4557.8496" }
|
|
|
|
|
%"class.std::__1::__libcpp_compressed_pair_imp.1527.4254.4557.8496" = type { %"struct.std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >::__rep.1526.4253.4556.8495" }
|
|
|
|
|
%"struct.std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >::__rep.1526.4253.4556.8495" = type { %union.anon.1525.4252.4555.8494 }
|
|
|
|
|
%union.anon.1525.4252.4555.8494 = type { %"struct.std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >::__long.1524.4251.4554.8493" }
|
|
|
|
|
%"struct.std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >::__long.1524.4251.4554.8493" = type { i64, i64, i8* }
|
|
|
|
|
|
|
|
|
|
@.str3 = external unnamed_addr constant [16 x i8], align 1
|
|
|
|
|
@_ZNSt3__15ctypeIcE2idE = external global %"class.std::__1::locale::id.1580.4307.4610.8491"
|
|
|
|
|
@_ZN4Foam12FatalIOErrorE = external global %"class.Foam::IOerror.1581.4308.4611.8505"
|
|
|
|
|
@.str204 = external unnamed_addr constant [18 x i8], align 1
|
|
|
|
|
@.str205 = external unnamed_addr constant [34 x i8], align 1
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4FoamlsERNS_7OstreamEPKc() #0
|
|
|
|
|
|
|
|
|
|
declare i32 @__gxx_personality_v0(...)
|
|
|
|
|
|
|
|
|
|
declare void @_ZNKSt3__18ios_base6getlocEv() #0
|
|
|
|
|
|
|
|
|
|
declare void @_ZNKSt3__16locale9use_facetERNS0_2idE() #0
|
|
|
|
|
|
|
|
|
|
; Function Attrs: noreturn
|
|
|
|
|
declare void @_ZNKSt3__121__basic_string_commonILb1EE20__throw_length_errorEv() #1 align 2
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4Foam6string6expandEb() #0
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4Foam8IFstreamC1ERKNS_8fileNameENS_8IOstream12streamFormatENS4_13versionNumberE() #0
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4Foam7IOerrorclEPKcS2_iRKNS_8IOstreamE() #0
|
|
|
|
|
|
|
|
|
|
declare void @_ZN4Foam7IOerror4exitEi() #0
|
|
|
|
|
|
|
|
|
|
; Function Attrs: inlinehint
|
|
|
|
|
declare void @_ZN4Foam8fileName12stripInvalidEv() #2 align 2
|
|
|
|
|
|
2015-06-18 04:52:32 +08:00
|
|
|
define void @_ZN4Foam3CSVINS_6VectorIdEEE4readEv() #0 align 2 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
entry:
|
|
|
|
|
invoke void @_ZN4Foam6string6expandEb()
|
|
|
|
|
to label %invoke.cont unwind label %lpad
|
|
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
|
br i1 undef, label %if.then.i.i.i.i176, label %_ZN4Foam6stringC2ERKS0_.exit.i
|
|
|
|
|
|
|
|
|
|
if.then.i.i.i.i176: ; preds = %invoke.cont
|
|
|
|
|
invoke void @_ZNKSt3__121__basic_string_commonILb1EE20__throw_length_errorEv()
|
|
|
|
|
to label %.noexc unwind label %lpad
|
|
|
|
|
|
|
|
|
|
.noexc: ; preds = %if.then.i.i.i.i176
|
|
|
|
|
unreachable
|
|
|
|
|
|
|
|
|
|
_ZN4Foam6stringC2ERKS0_.exit.i: ; preds = %invoke.cont
|
|
|
|
|
invoke void @_ZN4Foam8fileName12stripInvalidEv()
|
|
|
|
|
to label %invoke.cont2 unwind label %lpad.i
|
|
|
|
|
|
|
|
|
|
lpad.i: ; preds = %_ZN4Foam6stringC2ERKS0_.exit.i
|
2015-06-18 04:52:32 +08:00
|
|
|
%0 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
invoke.cont2: ; preds = %_ZN4Foam6stringC2ERKS0_.exit.i
|
|
|
|
|
invoke void @_ZN4Foam8IFstreamC1ERKNS_8fileNameENS_8IOstream12streamFormatENS4_13versionNumberE()
|
|
|
|
|
to label %invoke.cont4 unwind label %lpad3
|
|
|
|
|
|
|
|
|
|
invoke.cont4: ; preds = %invoke.cont2
|
|
|
|
|
br i1 undef, label %for.body, label %if.then
|
|
|
|
|
|
|
|
|
|
if.then: ; preds = %invoke.cont4
|
|
|
|
|
invoke void @_ZN4Foam7IOerrorclEPKcS2_iRKNS_8IOstreamE()
|
|
|
|
|
to label %invoke.cont8 unwind label %lpad5
|
|
|
|
|
|
|
|
|
|
invoke.cont8: ; preds = %if.then
|
|
|
|
|
invoke void @_ZN4FoamlsERNS_7OstreamEPKc()
|
|
|
|
|
to label %memptr.end.i unwind label %lpad5
|
|
|
|
|
|
|
|
|
|
memptr.end.i: ; preds = %invoke.cont8
|
|
|
|
|
invoke void @_ZN4Foam7IOerror4exitEi()
|
|
|
|
|
to label %if.end unwind label %lpad5
|
|
|
|
|
|
|
|
|
|
lpad: ; preds = %if.then.i.i.i.i176, %entry
|
2015-06-18 04:52:32 +08:00
|
|
|
%1 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
lpad3: ; preds = %invoke.cont2
|
2015-06-18 04:52:32 +08:00
|
|
|
%2 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
lpad5: ; preds = %memptr.end.i, %invoke.cont8, %if.then
|
2015-06-18 04:52:32 +08:00
|
|
|
%3 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
if.end: ; preds = %memptr.end.i
|
|
|
|
|
br i1 undef, label %for.body, label %vector.body
|
|
|
|
|
|
|
|
|
|
for.body: ; preds = %if.end, %invoke.cont4
|
|
|
|
|
invoke void @_ZNKSt3__18ios_base6getlocEv()
|
|
|
|
|
to label %.noexc205 unwind label %lpad19
|
|
|
|
|
|
|
|
|
|
.noexc205: ; preds = %for.body
|
|
|
|
|
invoke void @_ZNKSt3__16locale9use_facetERNS0_2idE()
|
|
|
|
|
to label %invoke.cont.i.i.i unwind label %lpad.i.i.i
|
|
|
|
|
|
|
|
|
|
invoke.cont.i.i.i: ; preds = %.noexc205
|
|
|
|
|
unreachable
|
|
|
|
|
|
|
|
|
|
lpad.i.i.i: ; preds = %.noexc205
|
2015-06-18 04:52:32 +08:00
|
|
|
%4 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
lpad19: ; preds = %for.body
|
2015-06-18 04:52:32 +08:00
|
|
|
%5 = landingpad { i8*, i32 }
|
Disable IsSub subregister assert. pr18663.
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
2014-08-01 03:50:53 +08:00
|
|
|
cleanup
|
|
|
|
|
br label %ehcleanup142
|
|
|
|
|
|
|
|
|
|
vector.body: ; preds = %vector.body, %if.end
|
|
|
|
|
%vec.phi = phi <8 x i32> [ %10, %vector.body ], [ undef, %if.end ]
|
|
|
|
|
%vec.phi1302 = phi <8 x i32> [ %11, %vector.body ], [ undef, %if.end ]
|
|
|
|
|
%vec.phi1303 = phi <8 x i32> [ %12, %vector.body ], [ undef, %if.end ]
|
|
|
|
|
%vec.phi1304 = phi <8 x i32> [ %13, %vector.body ], [ undef, %if.end ]
|
|
|
|
|
%6 = icmp sgt <8 x i32> undef, %vec.phi
|
|
|
|
|
%7 = icmp sgt <8 x i32> undef, %vec.phi1302
|
|
|
|
|
%8 = icmp sgt <8 x i32> undef, %vec.phi1303
|
|
|
|
|
%9 = icmp sgt <8 x i32> undef, %vec.phi1304
|
|
|
|
|
%10 = select <8 x i1> %6, <8 x i32> undef, <8 x i32> %vec.phi
|
|
|
|
|
%11 = select <8 x i1> %7, <8 x i32> undef, <8 x i32> %vec.phi1302
|
|
|
|
|
%12 = select <8 x i1> %8, <8 x i32> undef, <8 x i32> %vec.phi1303
|
|
|
|
|
%13 = select <8 x i1> %9, <8 x i32> undef, <8 x i32> %vec.phi1304
|
|
|
|
|
br label %vector.body
|
|
|
|
|
|
|
|
|
|
ehcleanup142: ; preds = %lpad19, %lpad.i.i.i, %lpad5, %lpad3, %lpad, %lpad.i
|
|
|
|
|
resume { i8*, i32 } undef
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
attributes #0 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
|
|
|
|
attributes #1 = { noreturn "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
|
|
|
|
attributes #2 = { inlinehint "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
|
|
|
|
|