search

riscv-software-src / riscv-tests

Other
872 stars 449 forks source link

fmin sNaN propagation #99

Open LeeMooreImperas opened 6 years ago

LeeMooreImperas commented 6 years ago

Hi I have noted that the behavior of the sNaN propagation as per the test for rv64uf-p-fmin has changed since the release of v2.2 of the ISA Specification In the 2.2 spec an sNaN on any input of FMIN/FMAX should generate a canonical NaN This is no longer the expectaion with either the current test or spike.

In order to have a set of tests and a version of spike which reflects the specification as per 2.2, how is this to be done ? as far as I can see there are no 'releases' bound to the version of the ISA, I would like to have a set of tests and golden reference simulator which matches the 2.2 spec.

Thx Lee

aswaterman commented 6 years ago

Sorry for not versioning the releases; we should start doing that. Indeed Spike and the tests correspond to v2.3-draft, which defines FMIN to behave like the IEEE 754 function minimumNumber. (You can find the up-to-date spec here: https://github.com/riscv/riscv-isa-manual)

On Fri, Nov 10, 2017 at 8:53 AM, LeeMooreImperas notifications@github.com wrote:

Hi I have noted that the behavior of the sNaN propagation as per the test for rv64uf-p-fmin has changed since the release of v2.2 of the ISA Specification In the 2.2 spec an sNaN on any input of FMIN/FMAX should generate a canonical NaN This is no longer the expectaion with either the current test or spike.

In order to have a set of tests and a version of spike which reflects the specification as per 2.2, how is this to be done ? as far as I can see there are no 'releases' bound to the version of the ISA, I would like to have a set of tests and golden reference simulator which matches the 2.2 spec.

Thx Lee

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/riscv/riscv-tests/issues/99, or mute the thread https://github.com/notifications/unsubscribe-auth/AA-7wlqQIyno0fypzIk9oqJ_oVQVb-Zaks5s1H-SgaJpZM4QZ1H- .

LeeMooreImperas commented 6 years ago

Hi Andrew, Thanks for the explaination. Is it possible for me to do a checkout of spike and riscv-tests, on a particular date, in which the date will correspond to the ISA spec as per ver 2.2 Thx Lee

aswaterman commented 6 years ago

If you look at the git log for the file riscv/insns/fmin_s.h in Spike, you should be able to find the commit where this behavior changed. The date of that commit will probably match the date that the test program changed, too.