2015-09-05 07:47:34 +08:00
|
|
|
; RUN: opt < %s -simplifycfg -S | FileCheck %s
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
; ModuleID = 'cppeh-simplify.cpp'
|
|
|
|
target datalayout = "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
|
|
|
|
target triple = "x86_64-pc-windows-msvc18.0.0"
|
|
|
|
|
|
|
|
|
|
|
|
; This case arises when two objects with empty destructors are cleaned up.
|
|
|
|
;
|
|
|
|
; void f1() {
|
|
|
|
; S a;
|
|
|
|
; S b;
|
|
|
|
; g();
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case, both cleanup pads can be eliminated and the invoke can be
|
|
|
|
; converted to a call.
|
|
|
|
;
|
|
|
|
; CHECK: define void @f1()
|
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: call void @g()
|
|
|
|
; CHECK: ret void
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: }
|
|
|
|
;
|
|
|
|
define void @f1() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
invoke void @g() to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
ret void
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %entry
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
|
|
|
cleanupret from %0 unwind label %ehcleanup.1
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
ehcleanup.1: ; preds = %ehcleanup
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%1 = cleanuppad within none []
|
|
|
|
cleanupret from %1 unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
; This case arises when an object with an empty destructor must be cleaned up
|
|
|
|
; outside of a try-block and an object with a non-empty destructor must be
|
|
|
|
; cleaned up within the try-block.
|
|
|
|
;
|
|
|
|
; void f2() {
|
|
|
|
; S a;
|
|
|
|
; try {
|
|
|
|
; S2 b;
|
|
|
|
; g();
|
|
|
|
; } catch (...) {}
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case, the outermost cleanup pad can be eliminated and the catch block
|
|
|
|
; should unwind to the caller (that is, exception handling continues with the
|
|
|
|
; parent frame of the caller).
|
|
|
|
;
|
|
|
|
; CHECK: define void @f2()
|
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: ehcleanup:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: cleanuppad within none
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: call void @"\01??1S2@@QEAA@XZ"(%struct.S2* %b)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catch.dispatch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: catchswitch within none [label %catch] unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: catchpad
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catchret
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: }
|
|
|
|
;
|
|
|
|
define void @f2() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
%b = alloca %struct.S2, align 1
|
|
|
|
invoke void @g() to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
br label %try.cont
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %entry
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
2015-09-05 07:39:40 +08:00
|
|
|
call void @"\01??1S2@@QEAA@XZ"(%struct.S2* %b)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.dispatch: ; preds = %ehcleanup
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind label %ehcleanup.1
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%1 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
|
|
|
catchret from %1 to label %catchret.dest
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catchret.dest: ; preds = %catch
|
|
|
|
br label %try.cont
|
|
|
|
|
|
|
|
try.cont: ; preds = %catchret.dest, %invoke.cont
|
|
|
|
ret void
|
|
|
|
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
ehcleanup.1:
|
|
|
|
%2 = cleanuppad within none []
|
|
|
|
cleanupret from %2 unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
; This case arises when an object with a non-empty destructor must be cleaned up
|
|
|
|
; outside of a try-block and an object with an empty destructor must be cleaned
|
|
|
|
; within the try-block.
|
|
|
|
;
|
|
|
|
; void f3() {
|
|
|
|
; S2 a;
|
|
|
|
; try {
|
|
|
|
; S b;
|
|
|
|
; g();
|
|
|
|
; } catch (...) {}
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case the inner cleanup pad should be eliminated and the invoke of g()
|
|
|
|
; should unwind directly to the catchpad.
|
|
|
|
;
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-LABEL: define void @f3()
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: to label %try.cont unwind label %catch.dispatch
|
|
|
|
; CHECK: catch.dispatch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-NEXT: catchswitch within none [label %catch] unwind label %ehcleanup.1
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: catchpad within %cs1 [i8* null, i32 64, i8* null]
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catchret
|
|
|
|
; CHECK: ehcleanup.1:
|
|
|
|
; CHECK: cleanuppad
|
|
|
|
; CHECK: call void @"\01??1S2@@QEAA@XZ"(%struct.S2* %a)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: cleanupret from %cp3 unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: }
|
|
|
|
;
|
|
|
|
define void @f3() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
%a = alloca %struct.S2, align 1
|
|
|
|
invoke void @g() to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
br label %try.cont
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %entry
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
|
|
|
cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.dispatch: ; preds = %ehcleanup
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind label %ehcleanup.1
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cp2 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
|
|
|
catchret from %cp2 to label %catchret.dest
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catchret.dest: ; preds = %catch
|
|
|
|
br label %try.cont
|
|
|
|
|
|
|
|
try.cont: ; preds = %catchret.dest, %invoke.cont
|
|
|
|
ret void
|
|
|
|
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
ehcleanup.1:
|
|
|
|
%cp3 = cleanuppad within none []
|
2015-09-05 07:39:40 +08:00
|
|
|
call void @"\01??1S2@@QEAA@XZ"(%struct.S2* %a)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
cleanupret from %cp3 unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
; This case arises when an object with an empty destructor may require cleanup
|
|
|
|
; from either inside or outside of a try-block.
|
|
|
|
;
|
|
|
|
; void f4() {
|
|
|
|
; S a;
|
|
|
|
; g();
|
|
|
|
; try {
|
|
|
|
; g();
|
|
|
|
; } catch (...) {}
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case, the cleanuppad should be eliminated, the invoke outside of the
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; catch block should be converted to a call (that is, that is, exception
|
|
|
|
; handling continues with the parent frame of the caller).)
|
2015-09-05 07:39:40 +08:00
|
|
|
;
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-LABEL: define void @f4()
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: call void @g
|
|
|
|
; Note: The cleanuppad simplification will insert an unconditional branch here
|
|
|
|
; but it will be eliminated, placing the following invoke in the entry BB.
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: to label %try.cont unwind label %catch.dispatch
|
|
|
|
; CHECK: catch.dispatch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: catchswitch within none [label %catch] unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catch:
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK: catchpad
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: catchret
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: }
|
|
|
|
;
|
|
|
|
define void @f4() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
invoke void @g()
|
|
|
|
to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
invoke void @g()
|
|
|
|
to label %try.cont unwind label %catch.dispatch
|
|
|
|
|
|
|
|
catch.dispatch: ; preds = %invoke.cont
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind label %ehcleanup
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
|
|
|
catchret from %0 to label %try.cont
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
try.cont: ; preds = %catch, %invoke.cont
|
|
|
|
ret void
|
|
|
|
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
ehcleanup:
|
|
|
|
%cp2 = cleanuppad within none []
|
|
|
|
cleanupret from %cp2 unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
; This case tests simplification of an otherwise empty cleanup pad that contains
|
|
|
|
; a PHI node.
|
|
|
|
;
|
|
|
|
; int f6() {
|
|
|
|
; int state = 1;
|
|
|
|
; try {
|
|
|
|
; S a;
|
|
|
|
; g();
|
|
|
|
; state = 2;
|
|
|
|
; g();
|
|
|
|
; } catch (...) {
|
|
|
|
; return state;
|
|
|
|
; }
|
|
|
|
; return 0;
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case, the cleanup pad should be eliminated and the PHI node in the
|
|
|
|
; cleanup pad should be sunk into the catch dispatch block.
|
|
|
|
;
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-LABEL: define i32 @f6()
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: invoke.cont:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK-NOT: ehcleanup:
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: catch.dispatch:
|
|
|
|
; CHECK: %state.0 = phi i32 [ 2, %invoke.cont ], [ 1, %entry ]
|
|
|
|
; CHECK: }
|
|
|
|
define i32 @f6() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
invoke void @g()
|
|
|
|
to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
invoke void @g()
|
|
|
|
to label %return unwind label %ehcleanup
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %invoke.cont, %entry
|
|
|
|
%state.0 = phi i32 [ 2, %invoke.cont ], [ 1, %entry ]
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
|
|
|
cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.dispatch: ; preds = %ehcleanup
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%1 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
|
|
|
catchret from %1 to label %return
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
return: ; preds = %invoke.cont, %catch
|
|
|
|
%retval.0 = phi i32 [ %state.0, %catch ], [ 0, %invoke.cont ]
|
|
|
|
ret i32 %retval.0
|
|
|
|
}
|
|
|
|
|
|
|
|
; This case tests another variation of simplification of an otherwise empty
|
|
|
|
; cleanup pad that contains a PHI node.
|
|
|
|
;
|
|
|
|
; int f7() {
|
|
|
|
; int state = 1;
|
|
|
|
; try {
|
|
|
|
; g();
|
|
|
|
; state = 2;
|
|
|
|
; S a;
|
|
|
|
; g();
|
|
|
|
; state = 3;
|
|
|
|
; g();
|
|
|
|
; } catch (...) {
|
|
|
|
; return state;
|
|
|
|
; }
|
|
|
|
; return 0;
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; In this case, the cleanup pad should be eliminated and the PHI node in the
|
|
|
|
; cleanup pad should be merged with the PHI node in the catch dispatch block.
|
|
|
|
;
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-LABEL: define i32 @f7()
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: invoke.cont:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: invoke.cont.1:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK-NOT: ehcleanup:
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: catch.dispatch:
|
|
|
|
; CHECK: %state.1 = phi i32 [ 1, %entry ], [ 3, %invoke.cont.1 ], [ 2, %invoke.cont ]
|
|
|
|
; CHECK: }
|
|
|
|
define i32 @f7() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
invoke void @g()
|
|
|
|
to label %invoke.cont unwind label %catch.dispatch
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
invoke void @g()
|
|
|
|
to label %invoke.cont.1 unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont.1: ; preds = %invoke.cont
|
|
|
|
invoke void @g()
|
|
|
|
to label %return unwind label %ehcleanup
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %invoke.cont.1, %invoke.cont
|
|
|
|
%state.0 = phi i32 [ 3, %invoke.cont.1 ], [ 2, %invoke.cont ]
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
|
|
|
cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.dispatch: ; preds = %ehcleanup, %entry
|
|
|
|
%state.1 = phi i32 [ %state.0, %ehcleanup ], [ 1, %entry ]
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%1 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
|
|
|
catchret from %1 to label %return
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
return: ; preds = %invoke.cont.1, %catch
|
|
|
|
%retval.0 = phi i32 [ %state.1, %catch ], [ 0, %invoke.cont.1 ]
|
|
|
|
ret i32 %retval.0
|
|
|
|
}
|
|
|
|
|
|
|
|
; This case tests a scenario where an empty cleanup pad is not dominated by all
|
|
|
|
; of the predecessors of its successor, but the successor references a PHI node
|
|
|
|
; in the empty cleanup pad.
|
|
|
|
;
|
|
|
|
; Conceptually, the case being modeled is something like this:
|
|
|
|
;
|
|
|
|
; int f8() {
|
|
|
|
; int x = 1;
|
|
|
|
; try {
|
|
|
|
; S a;
|
|
|
|
; g();
|
|
|
|
; x = 2;
|
|
|
|
; retry:
|
|
|
|
; g();
|
|
|
|
; return
|
|
|
|
; } catch (...) {
|
|
|
|
; use_x(x);
|
|
|
|
; }
|
|
|
|
; goto retry;
|
|
|
|
; }
|
|
|
|
;
|
|
|
|
; While that C++ syntax isn't legal, the IR below is.
|
|
|
|
;
|
|
|
|
; In this case, the PHI node that is sunk from ehcleanup to catch.dispatch
|
|
|
|
; should have an incoming value entry for path from 'foo' that references the
|
|
|
|
; PHI node itself.
|
|
|
|
;
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
; CHECK-LABEL: define void @f8()
|
2015-09-05 07:39:40 +08:00
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK: invoke.cont:
|
|
|
|
; CHECK: invoke void @g()
|
|
|
|
; CHECK-NOT: ehcleanup:
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: catch.dispatch:
|
|
|
|
; CHECK: %x = phi i32 [ 2, %invoke.cont ], [ 1, %entry ], [ %x, %catch.cont ]
|
|
|
|
; CHECK: }
|
|
|
|
define void @f8() personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
|
|
|
|
entry:
|
|
|
|
invoke void @g()
|
|
|
|
to label %invoke.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
invoke.cont: ; preds = %entry
|
|
|
|
invoke void @g()
|
|
|
|
to label %return unwind label %ehcleanup
|
|
|
|
|
|
|
|
ehcleanup: ; preds = %invoke.cont, %entry
|
|
|
|
%x = phi i32 [ 2, %invoke.cont ], [ 1, %entry ]
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%0 = cleanuppad within none []
|
|
|
|
cleanupret from %0 unwind label %catch.dispatch
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.dispatch: ; preds = %ehcleanup, %catch.cont
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%cs1 = catchswitch within none [label %catch] unwind to caller
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch: ; preds = %catch.dispatch
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
%1 = catchpad within %cs1 [i8* null, i32 u0x40, i8* null]
|
2015-09-05 07:39:40 +08:00
|
|
|
call void @use_x(i32 %x)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 13:38:55 +08:00
|
|
|
catchret from %1 to label %catch.cont
|
2015-09-05 07:39:40 +08:00
|
|
|
|
|
|
|
catch.cont: ; preds = %catch
|
|
|
|
invoke void @g()
|
|
|
|
to label %return unwind label %catch.dispatch
|
|
|
|
|
|
|
|
return: ; preds = %invoke.cont, %catch.cont
|
|
|
|
ret void
|
|
|
|
}
|
2016-05-21 13:12:32 +08:00
|
|
|
; CHECK-LABEL: define i32 @f9()
|
|
|
|
; CHECK: entry:
|
|
|
|
; CHECK: invoke void @"\01??1S2@@QEAA@XZ"(
|
|
|
|
; CHECK-NOT: cleanuppad
|
|
|
|
; CHECK: catch.dispatch:
|
|
|
|
; CHECK: }
|
|
|
|
define i32 @f9() personality i32 (...)* @__CxxFrameHandler3 {
|
|
|
|
entry:
|
|
|
|
%s = alloca i8, align 1
|
2017-04-11 04:18:21 +08:00
|
|
|
call void @llvm.lifetime.start.p0i8(i64 1, i8* nonnull %s)
|
2016-05-21 13:12:32 +08:00
|
|
|
%bc = bitcast i8* %s to %struct.S2*
|
|
|
|
invoke void @"\01??1S2@@QEAA@XZ"(%struct.S2* %bc)
|
|
|
|
to label %try.cont unwind label %ehcleanup
|
|
|
|
|
|
|
|
ehcleanup:
|
|
|
|
%cleanup.pad = cleanuppad within none []
|
2017-04-11 04:18:21 +08:00
|
|
|
call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull %s)
|
2016-05-21 13:12:32 +08:00
|
|
|
cleanupret from %cleanup.pad unwind label %catch.dispatch
|
|
|
|
|
|
|
|
catch.dispatch:
|
|
|
|
%catch.switch = catchswitch within none [label %catch] unwind to caller
|
|
|
|
|
|
|
|
catch:
|
|
|
|
%catch.pad = catchpad within %catch.switch [i8* null, i32 0, i8* null]
|
|
|
|
catchret from %catch.pad to label %try.cont
|
|
|
|
|
|
|
|
try.cont:
|
|
|
|
ret i32 0
|
|
|
|
}
|
2015-09-05 07:39:40 +08:00
|
|
|
|
2016-06-05 07:50:03 +08:00
|
|
|
; CHECK-LABEL: define void @f10(
|
|
|
|
define void @f10(i32 %V) personality i32 (...)* @__CxxFrameHandler3 {
|
|
|
|
entry:
|
|
|
|
invoke void @g()
|
|
|
|
to label %unreachable unwind label %cleanup
|
|
|
|
; CHECK: call void @g()
|
|
|
|
; CHECK-NEXT: unreachable
|
|
|
|
|
|
|
|
unreachable:
|
|
|
|
unreachable
|
|
|
|
|
|
|
|
cleanup:
|
|
|
|
%cp = cleanuppad within none []
|
|
|
|
switch i32 %V, label %cleanupret1 [
|
|
|
|
i32 0, label %cleanupret2
|
|
|
|
]
|
|
|
|
|
|
|
|
cleanupret1:
|
|
|
|
cleanupret from %cp unwind to caller
|
|
|
|
|
|
|
|
cleanupret2:
|
|
|
|
cleanupret from %cp unwind to caller
|
|
|
|
}
|
|
|
|
|
2015-09-05 07:39:40 +08:00
|
|
|
%struct.S = type { i8 }
|
|
|
|
%struct.S2 = type { i8 }
|
|
|
|
declare void @"\01??1S2@@QEAA@XZ"(%struct.S2*)
|
|
|
|
declare void @g()
|
|
|
|
declare void @use_x(i32 %x)
|
|
|
|
|
|
|
|
declare i32 @__CxxFrameHandler3(...)
|
|
|
|
|
2017-04-11 04:18:21 +08:00
|
|
|
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture)
|
|
|
|
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture)
|