wip/threadingbuildingblocks: Work around aggressive DSE causing a SEGV under g++.
Switch to clang, which doesn't do the aggressive DSE in this case. The Makefiles aleady assume we're using clang, in any case. See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71388 SEGV crash is like (from PrusaSlicer): parent=parent@entry=0x0, context=context@entry=0x79966ba4ab40 <tbb::internal::the_dummy_context>) at ../../src/tbb/scheduler.cpp:352 genuine=genuine@entry=true) at ../../src/tbb/scheduler.cpp:99 m=..., this=0x799632fe7e80) at ../../src/tbb/custom_scheduler.h:55 at ../../src/tbb/custom_scheduler.h:140
This commit is contained in:
parent
65ce81589c
commit
e590769db4
|
@ -20,6 +20,15 @@ USE_TOOLS+= gmake pax
|
|||
BUILD_TARGET= default
|
||||
TEST_TARGET= test
|
||||
|
||||
# Work around old TBB bug, we'll force using clang++, which appears to
|
||||
# have less agressive Dead-Store Elimination (DSE).
|
||||
# The makefiles already assume we're using clang, and hence fail to pass
|
||||
# "-flifetime-dse=1" to g++.
|
||||
# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71388
|
||||
PKGSRC_COMPILER= clang
|
||||
PKG_CC= clang
|
||||
PKG_CXX= clang++
|
||||
|
||||
.include "../../mk/bsd.prefs.mk"
|
||||
|
||||
.if ${OPSYS} == Linux
|
||||
|
|
Loading…
Reference in New Issue