Hi,
Could I please get reviews for this 8u backport of an optimization fix
pertaining to the hotspot copy of fdlibm? Historically,
sharedRuntimeTrans.cpp and sharedRuntimeTrig.cpp have been compiled
with optimization turned off on x86/x86_64. As JDK-8210416 will be
backported to JDK 8u fixing optimization for fdlibm itself, so should
be the hotspot copy in OpenJDK 8u. The build system for hotspot in JDK
8u is very different, so the original fix for JDK 12 doesn't apply.
It's a complete rewrite. Please see this webrev which shows the
prerequisite patch. It adds the FDLIBM_CFLAGS logic:
http://cr.openjdk.java.net/~sgehwolf/webrevs/JDK-8210416/jdk8/02/
It also fixes a bug on ppc64{,le} where optimization is being turned on
in hotspot (-O3), but not providing the -ffp-contract=off flag (or
older equivalent).
Before (on ppc64le):
/bin/g++ -DLINUX -D_GNU_SOURCE -DPPC64 -DPRODUCT -I.
-I/home/openjdk/jdk8u/hotspot/src/share/vm/prims
-I/home/openjdk/jdk8u/hotspot/src/share/vm
-I/home/openjdk/jdk8u/hotspot/src/share/vm/precompiled
-I/home/openjdk/jdk8u/hotspot/src/cpu/ppc/vm
-I/home/openjdk/jdk8u/hotspot/src/os_cpu/linux_ppc/vm
-I/home/openjdk/jdk8u/hotspot/src/os/linux/vm
-I/home/openjdk/jdk8u/hotspot/src/os/posix/vm -I../generated
-DHOTSPOT_RELEASE_VERSION="\"25.212-b04\"" -DHOTSPOT_BUILD_TARGET="\"product\""
-DHOTSPOT_BUILD_USER="\"openjdk\"" -DHOTSPOT_LIB_ARCH=\"ppc64le\"
-DHOTSPOT_VM_DISTRO="\"OpenJDK\"" -DTARGET_OS_FAMILY_linux -DTARGET_ARCH_ppc
-DTARGET_ARCH_MODEL_ppc_64 -DTARGET_OS_ARCH_linux_ppc
-DTARGET_OS_ARCH_MODEL_linux_ppc_64 -DTARGET_COMPILER_gcc -DCOMPILER2 -fPIC
-fno-rtti -fno-exceptions -D_REENTRANT -fcheck-new -fvisibility=hidden -m64
-pipe -fno-strict-aliasing -fno-omit-frame-pointer -O3 -g -D_LP64=1
-DVM_LITTLE_ENDIAN -DABI_ELFv2 -mcpu=power7 -mtune=power8
-minsert-sched-nops=regroup_exact -mno-multiple -mno-string -Werror
-Wpointer-arith -Wsign-compare -Wundef -Wunused-function -Wunused-value
-std=gnu++98 -c -MMD -MP -MF ../generated/dependencies/sharedRuntimeTrig.o.d
-fpch-deps -o sharedRuntimeTrig.o
/home/openjdk/jdk8u/hotspot/src/share/vm/runtime/sharedRuntimeTrig.cpp
After (on ppc64le; Note -ffp-contract=off flag):
/bin/g++ -DLINUX -D_GNU_SOURCE -DPPC64 -DPRODUCT -I.
-I/home/openjdk/jdk8u/hotspot/src/share/vm/prims
-I/home/openjdk/jdk8u/hotspot/src/share/vm
-I/home/openjdk/jdk8u/hotspot/src/share/vm/precompiled
-I/home/openjdk/jdk8u/hotspot/src/cpu/ppc/vm
-I/home/openjdk/jdk8u/hotspot/src/os_cpu/linux_ppc/vm
-I/home/openjdk/jdk8u/hotspot/src/os/linux/vm
-I/home/openjdk/jdk8u/hotspot/src/os/posix/vm -I../generated
-DHOTSPOT_RELEASE_VERSION="\"25.212-b04\"" -DHOTSPOT_BUILD_TARGET="\"product\""
-DHOTSPOT_BUILD_USER="\"openjdk\"" -DHOTSPOT_LIB_ARCH=\"ppc64le\"
-DHOTSPOT_VM_DISTRO="\"OpenJDK\"" -DTARGET_OS_FAMILY_linux -DTARGET_ARCH_ppc
-DTARGET_ARCH_MODEL_ppc_64 -DTARGET_OS_ARCH_linux_ppc
-DTARGET_OS_ARCH_MODEL_linux_ppc_64 -DTARGET_COMPILER_gcc -DCOMPILER2 -fPIC
-fno-rtti -fno-exceptions -D_REENTRANT -fcheck-new -fvisibility=hidden -m64
-pipe -fno-strict-aliasing -fno-omit-frame-pointer -O3 -ffp-contract=off
-D_LP64=1 -DVM_LITTLE_ENDIAN -DABI_ELFv2 -mcpu=power7 -mtune=power8
-minsert-sched-nops=regroup_exact -mno-multiple -mno-string -Werror
-Wpointer-arith -Wsign-compare -Wundef -Wunused-function -Wunused-value
-std=gnu++98 -c -MMD -MP -MF ../generated/dependencies/sharedRuntimeTrig.o.d
-fpch-deps -o sharedRuntimeTrig.o
/home/openjdk/jdk8u/hotspot/src/share/vm/runtime/sharedRuntimeTrig.cpp
Bug: https://bugs.openjdk.java.net/browse/JDK-8210425
webrev: http://cr.openjdk.java.net/~sgehwolf/webrevs/JDK-8210425/jdk8/02/webrev
Testing: Builds on x86_64 (server/Zero). Manual inspection of compile
log of sharedRuntimeTrans.cpp/sharedRuntimeTrig.cpp. Same for ppc64le.
Note that we've been using this patch downstream for a while now. We
haven't seen any issues so far.
Thanks,
Severin
