Pierrick Bouvier <pierrick.bouvier@linaro.org> writes:
On 2/29/24 6:21 PM, Alex Bennée wrote:
Pierrick Bouvier <pierrick.bouvier@linaro.org> writes:
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
I did notice there is a discrepancy between what libisns and libb
report. The libb looks like an overcount so I wonder if there are some
instructions we are not picking up but I can't see where that would be.
➜ ./qemu-hppa -plugin ./tests/plugin/libinsn.so -plugin
./tests/plugin/libbb.so,inline=true -d plugin
./tests/tcg/hppa-linux-user/sha512
1..10
ok 1 - do_test(&tests[i])
ok 2 - do_test(&tests[i])
ok 3 - do_test(&tests[i])
ok 4 - do_test(&tests[i])
ok 5 - do_test(&tests[i])
ok 6 - do_test(&tests[i])
ok 7 - do_test(&tests[i])
ok 8 - do_test(&tests[i])
ok 9 - do_test(&tests[i])
ok 10 - do_test(&tests[i])
CPU0: bb's: 54282, insns: 775697
Total: bb's: 54282, insns: 775697
cpu 0 insns: 774827
total insns: 774827
Although weirdly maybe only an hppa thing. Richard?
Do you observe the exact same number if you run only one of the plugin?
bb count number of instructions in an executed block, while insn
effectively count every instructions ran.
Maybe there is hppa specifity that makes some tb exit in the middle,
thus executing less instructions than expected from bb count.
Almost certainly - I just wasn't sure what would do that on straight
line code. Probably some funky aspect of HPPA I'm not aware off ;-)
I don't know how to reproduce this test. Did you run it from a
specific docker env?
If you have docker enabled the "make check-tcg" will build and use a
container to build the test cases. If you are on debian you just need:
apt install gcc-hppa-linux-gnu libc6-dev-hppa-cross
and re-run configure.