Add floatingpoint category

floatingpoint/README.markdown

@@ -0,0 +1,15 @@
+A few cases have been observed in the wild, where build results
+depended on the CPU type, because floating point operations
+would yield different accuracy.
+
+This might also produce deviations between i586 and x86\_64 builds.
+
+In some of these cases the goal of reproducibility may conflict with performance or accuracy.
+
+# Seen in the wild:
+
+* [fluidsynth](https://github.com/FluidSynth/fluidsynth/pull/512) fused-multiply-add used in glibc-2.29 pow if available
+
+* [calibre](https://github.com/kovidgoyal/calibre/blob/fab8c8f2d4d8c0c1b2046c2fbfd204189191c4c5/src/calibre/linux.py#L1171) `.png` files vary from SSE4.1 ?
+
+* [piglit](https://rb.zq1.de/compare.factory-20181023/piglit-compare.out) - already fixed?