On Fri, 20 Dec 2024 12:52:07 GMT, Magnus Ihse Bursie <[email protected]> wrote:
>> Hi please consider. >> >> This adds below to hs_err: >> >> Floating point state: >> fcsr=1 >> Floating point registers: >> f0=0xffffffff44a72000 | 1.84467e+19 >> f1=0xffffffff44a72000 | 1.84467e+19 >> .... >> f31=0xffffffff44a72000 | 1.84467e+19 >> >> Vector state: >> vstart=0x0000000000000000 >> vl=0x0000000000000020 >> vtype=0x0000000000000000 >> vcsr=0x0000000000000000 >> vlenb=0x0000000000000020 >> Vector registers: >> v0=0x0101010101010101010101010101010101010101010101010101010101010101 >> .... >> v31=0x0101010101010101010101010101010101010101010101010101010101010101 >> >> >> To get vector the headers need to include those structures, hence build >> files hackery. >> This means if you compile on a kernel without RVV support the error handler >> will lack support for it. >> We don't care about RVV option as carshing in native may still use vector >> even if the jit do not. >> >> I'm doubt full about the printing as double for fp regs, maybe that should >> be removed. >> >> Local testing, running t1 over weekend. >> >> Thanks, Robbin > > make/hotspot/lib/CompileJvm.gmk line 139: > >> 137: >> 138: ifeq ($(call isTargetCpu, riscv64), true) >> 139: JVM_CFLAGS += $(RVV_CFLAGS) > > The new flag should be added to the JVM_CFLAGS at configure time. No need to > export it to spec.gmk and do it at runtime. The flag is different for the build JVM and the JVM. If I just set in configure time it seem like both OPENJDK_BUILD_JVM_CFLAGS and JVM_CFLAGS have the flag. Which is incorrect, as only OPENJDK_BUILD_JVM_CFLAGS in this case should have it. Am I doing something wrong? I mirrored SVE_FLAGS which do this... ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/22845#discussion_r1893933496
