Richard Henderson <richard.hender...@linaro.org> writes:
> On 2/22/19 12:45 PM, Alex Bennée wrote: >> @@ -3526,6 +3551,8 @@ static float32 roundAndPackFloat32(flag zSign, int >> zExp, uint32_t zSig, >> case float_round_down: >> roundIncrement = zSign ? 0x7f : 0; >> break; >> + case float_round_to_odd: >> + roundIncrement = zSig & 0x80 ? 0 : 0x7f; >> default: >> abort(); > > I clearly missed a break here. > > Since this didn't kill fp-test, are we missing a float128_to_float32 > test? Ahh yes missing that: 10:11:17 [alex@zen:~/l/q/b/a/t/fp] fpu/next + ./fp-test f16_to_f32 f64_to_f32 f128_to_f32 -r odd >> Testing f16_to_f32 408 tests total. 408 tests performed. In 408 tests, no errors found in f16_to_f32. >> Testing f64_to_f32, rounding odd, tininess before rounding 768 tests total. 768 tests performed. In 768 tests, no errors found in f64_to_f32, rounding odd, tininess before rounding. >> Testing f64_to_f32, rounding odd, tininess after rounding 768 tests total. 768 tests performed. In 768 tests, no errors found in f64_to_f32, rounding odd, tininess after rounding. >> Testing f128_to_f32, rounding odd, tininess before rounding 936 tests total. fish: “./fp-test f16_to_f32 f64_to_f32…” terminated by signal SIGABRT (Abort) I'll patch it up and fix. > > > r~ -- Alex Bennée
