atrosinenko added a comment.
@sepavloff
Thank you for the test cases. Looks like it is worth completely rewriting the
three tests as table-driven tests, so for example adding four cases for
`[+-]Inf / [+-]0.0` would be almost as easy as testing only one of them
(something similar is already implemented for `udivmod?i4_test.c`). The
differences between adjacent test cases would be more visually obvious, as well.
================
Comment at: compiler-rt/test/builtins/Unit/divdf3_test.c:80
+ // divisor is 1.0 as UQ1.31
+ if (test__divdf3(0x1.0p+0, 0x1.00000001p+0, UINT64_C(0x3fefffffffe00000)))
return 1;
----------------
sepavloff wrote:
> Is 0x1.00000001p+0 equal to 1.0 in UQ1.31?
Divisor is `1.(31 zeroes)1` after restoring the implicit bit, so it is
**truncated** to 1.0 as UQ1.31. Instead of counting bits carefully, it would
probably be better to add several tests with the `1` bit shifted 1-2 places
left/right as well as if the divisor is round up instead of truncating - //just
in case//. :) So, with table-driven test it would probably be simpler to not
make extra assumptions on the implementation.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D84932/new/
https://reviews.llvm.org/D84932
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
