maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/WinEH/wineh-demotion.ll

357 lines
10 KiB
LLVM
Raw Normal View History

[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
; RUN: opt -mtriple=x86_64-pc-windows-msvc -S -winehprepare < %s | FileCheck %s
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
declare i32 @__CxxFrameHandler3(...)
declare void @f()
declare i32 @g()
declare void @h(i32)
declare i1 @i()
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
declare void @llvm.bar() nounwind
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK-LABEL: @test1(
define void @test1(i1 %B) personality i32 (...)* @__CxxFrameHandler3 {
entry:
; Spill slot should be inserted here
; CHECK: [[Slot:%[^ ]+]] = alloca
; Can't store for %phi at these defs because the lifetimes overlap
; CHECK-NOT: store
%x = call i32 @g()
%y = call i32 @g()
br i1 %B, label %left, label %right
left:
; CHECK: left:
; CHECK-NEXT: store i32 %x, i32* [[Slot]]
; CHECK-NEXT: invoke void @f
invoke void @f()
to label %exit unwind label %merge
right:
; CHECK: right:
; CHECK-NEXT: store i32 %y, i32* [[Slot]]
; CHECK-NEXT: invoke void @f
invoke void @f()
to label %exit unwind label %merge
merge:
; CHECK: merge:
; CHECK-NOT: = phi
%phi = phi i32 [ %x, %left ], [ %y, %right ]
[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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
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
%cp = catchpad within %cs1 []
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK: catch:
; CHECK: [[Reload:%[^ ]+]] = load i32, i32* [[Slot]]
; CHECK-NEXT: call void @h(i32 [[Reload]])
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %phi) [ "funclet"(token %cp) ]
[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 %cp to label %exit
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
exit:
ret void
}
; CHECK-LABEL: @test2(
define void @test2(i1 %B) personality i32 (...)* @__CxxFrameHandler3 {
entry:
br i1 %B, label %left, label %right
left:
; Need two stores here because %x and %y interfere so they need 2 slots
; CHECK: left:
; CHECK: store i32 1, i32* [[Slot1:%[^ ]+]]
; CHECK: store i32 1, i32* [[Slot2:%[^ ]+]]
; CHECK-NEXT: invoke void @f
invoke void @f()
to label %exit unwind label %merge.inner
right:
; Need two stores here because %x and %y interfere so they need 2 slots
; CHECK: right:
; CHECK-DAG: store i32 2, i32* [[Slot1]]
; CHECK-DAG: store i32 2, i32* [[Slot2]]
; CHECK: invoke void @f
invoke void @f()
to label %exit unwind label %merge.inner
merge.inner:
; CHECK: merge.inner:
; CHECK-NOT: = phi
[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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
%x = phi i32 [ 1, %left ], [ 2, %right ]
[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.inner] unwind label %merge.outer
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catch.inner:
[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
%cpinner = catchpad within %cs1 []
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; Need just one store here because only %y is affected
; CHECK: catch.inner:
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
%z = call i32 @g() [ "funclet"(token %cpinner) ]
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK: store i32 %z
; CHECK-NEXT: invoke void @f
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
invoke void @f() [ "funclet"(token %cpinner) ]
[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
to label %catchret.inner unwind label %merge.outer
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catchret.inner:
[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 %cpinner to label %exit
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
merge.outer:
[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
%y = phi i32 [ %x, %merge.inner ], [ %z, %catch.inner ]
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK: merge.outer:
[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-NOT: = phi
; CHECK: catchswitch within none
%cs2 = catchswitch within none [label %catch.outer] unwind to caller
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catch.outer:
[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
%cpouter = catchpad within %cs2 []
; CHECK: catch.outer:
; CHECK: [[CatchPad:%[^ ]+]] = catchpad within %cs2 []
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; Need to load x and y from two different slots since they're both live
; and can have different values (if we came from catch.inner)
; CHECK-DAG: load i32, i32* [[Slot1]]
; CHECK-DAG: load i32, i32* [[Slot2]]
[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: catchret from [[CatchPad]] to label
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %x) [ "funclet"(token %cpouter) ]
call void @h(i32 %y) [ "funclet"(token %cpouter) ]
[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 %cpouter to label %exit
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
exit:
ret void
}
; test4: don't need stores for %phi.inner, as its only use is to feed %phi.outer
; %phi.outer needs stores in %left, %right, and %join
; CHECK-LABEL: @test4(
define void @test4(i1 %B) personality i32 (...)* @__CxxFrameHandler3 {
entry:
; CHECK: entry:
; CHECK: [[Slot:%[^ ]+]] = alloca
; CHECK-NEXT: br
br i1 %B, label %left, label %right
left:
; CHECK: left:
; CHECK-NOT: store
; CHECK: store i32 %l, i32* [[Slot]]
; CHECK-NEXT: invoke void @f
%l = call i32 @g()
invoke void @f()
to label %join unwind label %catchpad.inner
right:
; CHECK: right:
; CHECK-NOT: store
; CHECK: store i32 %r, i32* [[Slot]]
; CHECK-NEXT: invoke void @f
%r = call i32 @g()
invoke void @f()
to label %join unwind label %catchpad.inner
catchpad.inner:
; CHECK: catchpad.inner:
[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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
%phi.inner = phi i32 [ %l, %left ], [ %r, %right ]
[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.inner] unwind label %catchpad.outer
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catch.inner:
[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
%cp1 = catchpad within %cs1 []
catchret from %cp1 to label %join
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
join:
; CHECK: join:
; CHECK-NOT: store
; CHECK: store i32 %j, i32* [[Slot]]
; CHECK-NEXT: invoke void @f
%j = call i32 @g()
invoke void @f()
to label %exit unwind label %catchpad.outer
[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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catchpad.outer:
; CHECK: catchpad.outer:
[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
%phi.outer = phi i32 [ %phi.inner, %catchpad.inner ], [ %j, %join ]
%cs2 = catchswitch within none [label %catch.outer] unwind to caller
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catch.outer:
; CHECK: catch.outer:
; CHECK: [[Reload:%[^ ]+]] = load i32, i32* [[Slot]]
; CHECK: call void @h(i32 [[Reload]])
[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 %cs2 []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %phi.outer) [ "funclet"(token %cp2) ]
[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 %cp2 to label %exit
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
exit:
ret void
}
; CHECK-LABEL: @test5(
define void @test5() personality i32 (...)* @__CxxFrameHandler3 {
entry:
; need store for %phi.cleanup
; CHECK: entry:
; CHECK: store i32 1, i32* [[CleanupSlot:%[^ ]+]]
; CHECK-NEXT: invoke void @f
invoke void @f()
to label %invoke.cont unwind label %cleanup
invoke.cont:
; need store for %phi.cleanup
; CHECK: invoke.cont:
; CHECK-NEXT: store i32 2, i32* [[CleanupSlot]]
; CHECK-NEXT: invoke void @f
invoke void @f()
to label %invoke.cont2 unwind label %cleanup
cleanup:
; cleanup phi can be loaded at cleanup entry
; CHECK: cleanup:
[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: cleanuppad within none []
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK: [[CleanupReload:%[^ ]+]] = load i32, i32* [[CleanupSlot]]
%phi.cleanup = phi i32 [ 1, %entry ], [ 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
%cp = cleanuppad within none []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
%b = call i1 @i() [ "funclet"(token %cp) ]
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
br i1 %b, label %left, label %right
left:
; CHECK: left:
; CHECK: call void @h(i32 [[CleanupReload]]
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %phi.cleanup) [ "funclet"(token %cp) ]
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
br label %merge
right:
; CHECK: right:
; CHECK: call void @h(i32 [[CleanupReload]]
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %phi.cleanup) [ "funclet"(token %cp) ]
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
br label %merge
merge:
; need store for %phi.catch
; CHECK: merge:
; CHECK-NEXT: store i32 [[CleanupReload]], i32* [[CatchSlot:%[^ ]+]]
[WinEH] Require token linkage in EH pad/ret signatures Summary: WinEHPrepare is going to require that cleanuppad and catchpad produce values of token type which are consumed by any cleanupret or catchret exiting the pad. This change updates the signatures of those operators to require/enforce that the type produced by the pads is token type and that the rets have an appropriate argument. The catchpad argument of a `CatchReturnInst` must be a `CatchPadInst` (and similarly for `CleanupReturnInst`/`CleanupPadInst`). To accommodate that restriction, this change adds a notion of an operator constraint to both LLParser and BitcodeReader, allowing appropriate sentinels to be constructed for forward references and appropriate error messages to be emitted for illegal inputs. Also add a verifier rule (noted in LangRef) that a catchpad with a catchpad predecessor must have no other predecessors; this ensures that WinEHPrepare will see the expected linear relationship between sibling catches on the same try. Lastly, remove some superfluous/vestigial casts from instruction operand setters operating on BasicBlocks. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12108 llvm-svn: 245797
2015-08-23 08:26:33 +08:00
; CHECK-NEXT: cleanupret
[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 %cp unwind label %catchswitch
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
invoke.cont2:
; need store for %phi.catch
; CHECK: invoke.cont2:
; CHECK-NEXT: store i32 3, i32* [[CatchSlot]]
; CHECK-NEXT: invoke void @f
invoke void @f()
[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
to label %exit unwind label %catchswitch
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +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
catchswitch:
; CHECK: catchswitch:
; CHECK-NEXT: catchswitch within none
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
%phi.catch = phi i32 [ %phi.cleanup, %merge ], [ 3, %invoke.cont2 ]
[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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
catch:
; 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
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
; CHECK: [[CatchReload:%[^ ]+]] = load i32, i32* [[CatchSlot]]
; CHECK: call void @h(i32 [[CatchReload]]
[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 []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %phi.catch) [ "funclet"(token %cp2) ]
[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 %cp2 to label %exit
[WinEHPrepare] Update demotion logic Summary: Update the demotion logic in WinEHPrepare to avoid creating new cleanups by walking predecessors as necessary to insert stores for EH-pad PHIs. Also avoid creating stores for EH-pad PHIs that have no uses. The store/load placement is still pretty naive. Likely future improvements (at least for optimized compiles) include: - Share loads for related uses as possible - Coalesce non-interfering use/def-related PHIs - Store at definition point rather than each PHI pred for non-interfering lifetimes. Reviewers: rnk, majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D11955 llvm-svn: 244894
2015-08-13 22:30:10 +08:00
exit:
ret void
}
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
[WinEH] Disable most forms of demotion Now that the register allocator knows about the barriers on funclet entry and exit, testing has shown that this is unnecessary. We still demote PHIs on unsplittable blocks due to the differences between the IR CFG and the Machine CFG. llvm-svn: 253619
2015-11-20 07:23:33 +08:00
; We used to demote %x, but we don't need to anymore.
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
; CHECK-LABEL: @test6(
define void @test6() personality i32 (...)* @__CxxFrameHandler3 {
entry:
; CHECK: entry:
[WinEH] Disable most forms of demotion Now that the register allocator knows about the barriers on funclet entry and exit, testing has shown that this is unnecessary. We still demote PHIs on unsplittable blocks due to the differences between the IR CFG and the Machine CFG. llvm-svn: 253619
2015-11-20 07:23:33 +08:00
; CHECK: %x = invoke i32 @g()
; CHECK-NEXT: to label %loop unwind label %to_caller
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
%x = invoke i32 @g()
to label %loop unwind label %to_caller
to_caller:
[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
%cp1 = cleanuppad within none []
cleanupret from %cp1 unwind to caller
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
loop:
invoke void @f()
to label %loop unwind label %cleanup
cleanup:
; CHECK: cleanup:
[WinEH] Disable most forms of demotion Now that the register allocator knows about the barriers on funclet entry and exit, testing has shown that this is unnecessary. We still demote PHIs on unsplittable blocks due to the differences between the IR CFG and the Machine CFG. llvm-svn: 253619
2015-11-20 07:23:33 +08:00
; CHECK: call void @h(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
%cp2 = cleanuppad within none []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @h(i32 %x) [ "funclet"(token %cp2) ]
[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 %cp2 unwind to caller
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
}
; CHECK-LABEL: @test7(
define void @test7() personality i32 (...)* @__CxxFrameHandler3 {
entry:
; %x is an EH pad phi, so gets stored in pred here
; CHECK: entry:
; CHECK: store i32 1, i32* [[SlotX:%[^ ]+]]
; CHECK: invoke void @f()
invoke void @f()
to label %invoke.cont unwind label %catchpad
invoke.cont:
; %x is an EH pad phi, so gets stored in pred here
; CHECK: invoke.cont:
; CHECK: store i32 2, i32* [[SlotX]]
; CHECK: invoke void @f()
invoke void @f()
to label %exit unwind label %catchpad
catchpad:
; %x phi should be eliminated
; CHECK: 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-NEXT: catchswitch within none
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
%x = phi i32 [ 1, %entry ], [ 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
%cs1 = catchswitch within none [label %catch] unwind to caller
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
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: catch:
; CHECK-NEXT: %[[CatchPad:[^ ]+]] = catchpad within %cs1 []
%cp = catchpad within %cs1 []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
%b = call i1 @i() [ "funclet"(token %cp) ]
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
br i1 %b, label %left, label %right
left:
; Edge from %left to %join needs to be split so that
; the load of %x can be inserted *after* the catchret
; CHECK: left:
[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: catchret from %[[CatchPad]] to label %[[SplitLeft:[^ ]+]]
catchret from %cp to label %join
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
; CHECK: [[SplitLeft]]:
; CHECK: [[LoadX:%[^ ]+]] = load i32, i32* [[SlotX]]
; CHECK: br label %join
right:
; Edge from %right to %join needs to be split so that
; the load of %y can be inserted *after* the catchret
; CHECK: right:
[WinEH] Disable most forms of demotion Now that the register allocator knows about the barriers on funclet entry and exit, testing has shown that this is unnecessary. We still demote PHIs on unsplittable blocks due to the differences between the IR CFG and the Machine CFG. llvm-svn: 253619
2015-11-20 07:23:33 +08:00
; CHECK: %y = call i32 @g()
[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: catchret from %[[CatchPad]] to label %join
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
%y = call i32 @g() [ "funclet"(token %cp) ]
[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 %cp to label %join
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
join:
; CHECK: join:
[WinEH] Disable most forms of demotion Now that the register allocator knows about the barriers on funclet entry and exit, testing has shown that this is unnecessary. We still demote PHIs on unsplittable blocks due to the differences between the IR CFG and the Machine CFG. llvm-svn: 253619
2015-11-20 07:23:33 +08:00
; CHECK: %phi = phi i32 [ [[LoadX]], %[[SplitLeft]] ], [ %y, %right ]
[WinEHPrepare] Fix catchret successor phi demotion Summary: When demoting an SSA value that has a use on a phi and one of the phi's predecessors terminates with catchret, the edge needs to be split and the load inserted in the new block, else we'll still have a cross-funclet SSA value. Add a test for this, and for the similar case where a def to be spilled is on and invoke and a critical edge, which was already implemented but missing a test. Reviewers: majnemer Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D12065 llvm-svn: 245218
2015-08-17 21:51:37 +08:00
%phi = phi i32 [ %x, %left ], [ %y, %right ]
call void @h(i32 %phi)
br label %exit
exit:
ret void
}
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
; CHECK-LABEL: @test8(
Revert r258580 and r258581. Those commits created an artificial edge from a cleanup to a synthesized catchswitch in order to get the MSVC personality routine to execute cleanups which don't cleanupret and are not wrapped by a catchswitch. This worked well enough but is not a complete solution in situations where there the cleanup infinite loops. However, the real deal breaker behind this approach comes about from a degenerate case where the cleanup is post-dominated by unreachable *and* throws an exception. This ends poorly because the catchswitch will inadvertently catch the exception. Because of this we should go back to our previous behavior of not executing certain cleanups (identical behavior with the Itanium ABI implementation in clang, GCC and ICC). N.B. I think this could be salvaged by making the catchpad rethrow the exception and properly transforming throwing calls in the cleanup into invokes. llvm-svn: 259338
2016-02-01 11:29:38 +08:00
define void @test8() personality i32 (...)* @__CxxFrameHandler3 { entry:
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
invoke void @f()
to label %done unwind label %cleanup1
invoke void @f()
to label %done unwind label %cleanup2
done:
ret void
cleanup1:
[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: [[CleanupPad1:%[^ ]+]] = cleanuppad within none []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
; CHECK-NEXT: call void @llvm.bar()
[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: cleanupret from [[CleanupPad1]]
%cp0 = cleanuppad within none []
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
br label %cleanupexit
cleanup2:
[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 []
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
; CHECK-NEXT: call void @llvm.bar()
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
; CHECK-NEXT: unreachable
[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
%cp1 = cleanuppad within none []
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
br label %cleanupexit
cleanupexit:
[WinEH] Use operand bundles to describe call sites SimplifyCFG allows tail merging with code which terminates in unreachable which, in turn, makes it possible for an invoke to end up in a funclet which it was not originally part of. Using operand bundles on invokes allows us to determine whether or not an invoke was part of a funclet in the source program. Furthermore, it allows us to unambiguously answer questions about the legality of inlining into call sites which the personality may have trouble with. Differential Revision: http://reviews.llvm.org/D15517 llvm-svn: 255674
2015-12-16 05:27:27 +08:00
call void @llvm.bar()
[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 %cp0 unwind label %cleanup2
[WinEHPrepare] Replace unreasonable funclet terminators with unreachable It is possible to be in a situation where more than one funclet token is a valid SSA value. If we see a terminator which exits a funclet which doesn't use the funclet's token, replace it with unreachable. Differential Revision: http://reviews.llvm.org/D12074 llvm-svn: 245238
2015-08-18 04:56:39 +08:00
}