|
From: John R. <jr...@bi...> - 2011-10-05 14:01:35
|
How should I enhance the valgrind regression testing so that running
on ARMv5 skips the tests which use floating-point hardware?
For memcheck and 'none': badjump2, fprw, origin2-not-quite,
origin6-fp, sh-mem-random, vcpu_fbench, vcpu_fnfns, neon128,
neon64, v6intARM, v6intThumb, v6media, vf
Similarly, atomic_incs requires opcodes that ARMv5 does not implement.
For instance, in configure.in I detect armv5, armv6, armv7 and set
one of:
AC_DEFINE(ARM_ARCH_V7,1,"Defined for v7 architectures")
AC_DEFINE(ARM_ARCH_V6,1,"Defined for v6 architectures")
AC_DEFINE(ARM_ARCH_V5TE,1,"Defined for v5te architectures")
How should I propagate this to "do not run neon128" ?
--
|
|
From: Tom H. <to...@co...> - 2011-10-05 14:13:55
|
On 05/10/11 15:03, John Reiser wrote: > How should I enhance the valgrind regression testing so that running > on ARMv5 skips the tests which use floating-point hardware? > For memcheck and 'none': badjump2, fprw, origin2-not-quite, > origin6-fp, sh-mem-random, vcpu_fbench, vcpu_fnfns, neon128, > neon64, v6intARM, v6intThumb, v6media, vf > Similarly, atomic_incs requires opcodes that ARMv5 does not implement. Well the normal method is to write a program in the tests directory (see for example tests/x86_amd64_features.c) and then a prereq directive in the vgtest file. See, for example: none/tests/x86/insn_sse2.vgtest which has: prereq: ../../../tests/x86_amd64_features x86-sse2 Tom -- Tom Hughes (to...@co...) http://compton.nu/ |