Hi Hongxu On 10/11/2018 11:01 AM, Hongxu Jia wrote: > On 2018年10月11日 08:21, Anuj Mittal wrote:
>> Is it possible that the value of -march is also being defined from >> somewhere within dpdk leading to a mismatch? > > Hi Anuj, > > I think so, since var-DPDK_TARGET_MACHINE is `default' which mismatch > with `-march=**' in var-HOST_CC_ARCH > > For qemux86-64, while var-DPDK_TARGET_MACHINE is `default', its > `-march=corei7', > it mismatch with `-march=core2' in var-HOST_CC_ARCH > > [grep corei -rn mk] > mk/machine/native/rte.vars.mk:35:# so, set the compilation target to be > a corei7, minimum target with SSE4.2. > mk/machine/native/rte.vars.mk:38: MACHINE_CFLAGS = -march=corei7 > mk/machine/default/rte.vars.mk:30:MACHINE_CFLAGS += -march=corei7 > mk/machine/wsm/rte.vars.mk:30:MACHINE_CFLAGS = -march=corei7 -maes -mpclmul > mk/machine/snb/rte.vars.mk:30:MACHINE_CFLAGS = -march=corei7-avx > mk/machine/nhm/rte.vars.mk:30:MACHINE_CFLAGS = -march=corei > [grep corei -rn mk] > > For default, its MACHINE_CFLAGS is ` -march=corei7', and set > `DPDK_TARGET_MACHINE = 'corei5'' > does not work, no mk/machine/corei5/rte.vars.mk to support it > > I think this fix is not appropriate, please drop it. My suggestion is remove > var-HOST_CC_ARCH from do_compile, V2 incoming > I think we shouldn't be building with -march=corei7 (which is what it will do if you remove HOST_CC_ARCH) if the underlying hardware doesn't support it. It probably wouldn't even work since default value is core2duo for qemu in tune-core2.inc which doesn't support SSE 4.2 (which is a requirement for dpdk). In my opinion, we should keep HOST_CC_ARCH and anyone using dpdk should make sure that they define COMPATIBLE_MACHINE and DPDK_TARGET_MACHINE properly depending on the arch that they are building for. Does that make sense? Thanks, Anuj -- _______________________________________________ meta-intel mailing list [email protected] https://lists.yoctoproject.org/listinfo/meta-intel
