coroutines: Backout mandate for tailcalls at O < 2 [PR94359]

Message ID 0C4F887E-FFA8-4448-B4F0-1D9EF2C0577E@sandoe.co.uk
State New
Headers show
Series
  • coroutines: Backout mandate for tailcalls at O < 2 [PR94359]
Related show

Commit Message

Iain Sandoe April 16, 2020, 1:51 p.m.
Hi

For symmetric transfers to work with C++20 coroutines, it is
currently necessary to tail call the callee coroutine from resume
method of the caller coroutine.  However there are several targets
which don't support an indirect tail call to an arbitrary callee.

Unfortunately, the target 'function_ok_for_sibcall' is not usable
from the front end in all cases.  While it is possible to add a new
hook to cover this circumstance, IMO, it is too late in the release
cycle to be sure of getting the setting correct for all targets***.

So, this patch backs out the use of function_ok_for_sibcall () and
the mandate of CALL_EXPR_MUST_TAIL_CALL from the symmetric 
transfer.

Targets that can make indirect tail calls to arbitrary callees will
still be able to make use of the symmetric transfer (without risking
overrunning the stack) for optimization levels >= 2.

The draft standard does not mandate unlimited symmetric transfers,
so removing this is a QOI issue (albeit an important one) rather
than a correctness one.

The test is moved and adjusted so that it can be opted into by any
target that supports the necessary tailcall.

(tested on x86_64-linux/darwin, powerpc64-linux, sparc-solaris11)
OK for master?
thanks
Iain

*** that can always be revisited.

gcc/cp/ChangeLog:

2020-04-16  Iain Sandoe  <iain@sandoe.co.uk>

	* coroutines.cc (build_actor_fn): Back out use of
	targetm.function_ok_for_sibcall.  Do not mark the resume
	call as CALL_EXPR_MUST_TAIL_CALL.

gcc/testsuite/ChangeLog:

2020-04-16  Iain Sandoe  <iain@sandoe.co.uk>

	* g++.dg/coroutines/torture/symmetric-transfer-00-basic.C: Move..
	* g++.dg/coroutines/symmetric-transfer-00-basic.C: ..here and
	adjust to run at O2 for targets supporting the necessary tail
	call.
---
 gcc/cp/coroutines.cc                             | 16 ++--------------
 .../{torture => }/symmetric-transfer-00-basic.C  | 13 ++++++-------
 2 files changed, 8 insertions(+), 21 deletions(-)
 rename gcc/testsuite/g++.dg/coroutines/{torture => }/symmetric-transfer-00-basic.C (87%)

-- 
2.17.1

Comments

Nathan Sidwell April 16, 2020, 2:05 p.m. | #1
On 4/16/20 9:51 AM, Iain Sandoe wrote:
> Hi

> 

> For symmetric transfers to work with C++20 coroutines, it is

> currently necessary to tail call the callee coroutine from resume

> method of the caller coroutine.  However there are several targets

> which don't support an indirect tail call to an arbitrary callee.

> 

> Unfortunately, the target 'function_ok_for_sibcall' is not usable

> from the front end in all cases.  While it is possible to add a new

> hook to cover this circumstance, IMO, it is too late in the release

> cycle to be sure of getting the setting correct for all targets***.

> 

> So, this patch backs out the use of function_ok_for_sibcall () and

> the mandate of CALL_EXPR_MUST_TAIL_CALL from the symmetric

> transfer.

> 

> Targets that can make indirect tail calls to arbitrary callees will

> still be able to make use of the symmetric transfer (without risking

> overrunning the stack) for optimization levels >= 2.

> 

> The draft standard does not mandate unlimited symmetric transfers,

> so removing this is a QOI issue (albeit an important one) rather

> than a correctness one.

> 

> The test is moved and adjusted so that it can be opted into by any

> target that supports the necessary tailcall.

> 

> (tested on x86_64-linux/darwin, powerpc64-linux, sparc-solaris11)

> OK for master?

> thanks

> Iain


OK.  will you leave 94359 open, or is there a separate bug for the lack 
of generalized symmetric xfer?

nathan


-- 
Nathan Sidwell

Patch

diff --git a/gcc/cp/coroutines.cc b/gcc/cp/coroutines.cc
index e4ba642d527..0a8e7521c4f 100644
--- a/gcc/cp/coroutines.cc
+++ b/gcc/cp/coroutines.cc
@@ -2377,21 +2377,9 @@  build_actor_fn (location_t loc, tree coro_frame_type, tree actor, tree fnbody,
     (loc, builtin_decl_explicit (BUILT_IN_CORO_RESUME), 1, addr);
 
   /* In order to support an arbitrary number of coroutine continuations,
-     we must tail call them.  However, some targets might not support this
-     for indirect calls, or calls between DSOs.
-     FIXME: see if there's an alternate strategy for such targets.  */
-  /* Now we have the actual call, and we can mark it as a tail.  */
+     we must tail call them.  However, some targets do not support indirect
+     tail calls to arbitrary callees.  See PR94359.  */
   CALL_EXPR_TAILCALL (resume) = true;
-  /* Temporarily, switch cfun so that we can use the target hook.  */
-  push_struct_function (actor);
-  if (targetm.function_ok_for_sibcall (NULL_TREE, resume))
-    {
-      /* ... and for optimisation levels 0..1, which do not normally tail-
-	-call, mark it as requiring a tail-call for correctness.  */
-      if (optimize < 2)
-	CALL_EXPR_MUST_TAIL_CALL (resume) = true;
-    }
-  pop_cfun ();
   resume = coro_build_cvt_void_expr_stmt (resume, loc);
   add_stmt (resume);
 
diff --git a/gcc/testsuite/g++.dg/coroutines/torture/symmetric-transfer-00-basic.C b/gcc/testsuite/g++.dg/coroutines/symmetric-transfer-00-basic.C
similarity index 87%
rename from gcc/testsuite/g++.dg/coroutines/torture/symmetric-transfer-00-basic.C
rename to gcc/testsuite/g++.dg/coroutines/symmetric-transfer-00-basic.C
index 6f379c8e77a..b78ae20d9d4 100644
--- a/gcc/testsuite/g++.dg/coroutines/torture/symmetric-transfer-00-basic.C
+++ b/gcc/testsuite/g++.dg/coroutines/symmetric-transfer-00-basic.C
@@ -1,10 +1,9 @@ 
-// { dg-do run }
-// See PR94359 - some targets are unable to make general indirect tailcalls
-// for example, between different DSOs.
-// { dg-xfail-run-if "" { hppa*-*-hpux11* } }
-// { dg-xfail-run-if "" { ia64-*-linux-gnu } }
-// { dg-xfail-run-if "" { { lp64 && { powerpc*-linux-gnu } } || { *-*-aix* } } }
-// { dg-xfail-run-if "" { sparc*-*-* } }
+// See PR94359, we will need either a general solution to this, or at least
+// some hook for targets to opt in, for now the test will work on targets that
+// can do the tailcall (which would normally be available for O2+)
+
+// { dg-do run { target { i?86-*-linux-gnu x86_64-*-linux-gnu *-*-darwin* } } }
+// { dg-additional-options "-O2" }
 
 #if __has_include(<coroutine>)