On 10 October 2011 12:26, Christophe Lyon <christophe.l...@st.com> wrote: > On 09.10.2011 00:57, Peter Maydell wrote: >> So we weren't generating incorrect results, we were just doing >> slightly more work than we really needed, right? I'm curious >> what prompted this patch :-) >> > Exactly. And no way to expose a bug :-(
If we always generate the correct results for all inputs, then by definition it's not a bug. It's just a slightly suboptimal calculation (both here and in the ARM ARM pseudocode). > I was reading 2 revisions of the ARM ARM, and noticed erratas in the > descriptions of FPRecipEstimate and FPRSqrtEstimate. > > Sign propagation has been removed in the former, and not in the later, so I > re-read both functions carefully as well as qemu's implementation and came > to this conclusion :-) For FPRecipEstimate the change in the sign handling was fixing a genuine erratum in the pseudocode. (QEMU's code for that is correct; I recall checking at the time that we were following the amended pseudocode rather than the old version.) For FPSqrtEstimate it's just a clarity of phrasing issue. Anyway, I agree we should make this change and have tested that we still generate correct results. So: Reviewed-by: Peter Maydell <peter.mayd...@linaro.org> -- PMM
