Hi Brian, Yes, my experience is the fixed point 56k DSP family. But I’m pretty sure that floating point DSPs either don’t denormalize (most likely—not much need for DSP applications), or don’t have a penalty—DSP instruction execution time needs to be limited, better yet constant.
Nigel > On Apr 12, 2023, at 6:31 PM, brianw <bri...@audiobanshee.com> wrote: > > Thanks for the link to your articles, Nigel. > > In one of them, you mentioned, "After working on DSP chips, primarily, which > are unaffected by this issue, ..." > > Are you referring to fixed-point DSP architectures, only? ... or do the > floating point DSP chips have a solution? > > I've only designed for fixed-point DSP architectures (TMS320, which also has > float models in the family), and I've not had a chance to look at the details > of floating point DSP architectures. > > Brian > > > On Apr 11, 2023, at 5:05 PM, Nigel Redmon <earle...@earlevel.com> wrote: >> Plus, I run everything as doubles already, it takes a little time to get in >> trouble. Not much time in human terms, though. Scroll down a page or two to >> the images here: >> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.earlevel.com_main_2019_04_19_floating-2Dpoint-2Ddenormals_&d=DwIFAg&c=009klHSCxuh5AI1vNQzSO0KGjl4nbi2Q0M1QLJX9BeE&r=TRvFbpof3kTa2q5hdjI2hccynPix7hNL2n0I6DmlDy0&m=I8WhqZGA-QjDmDFFwy2SYMOokWKYx1r1tpA0MC_-47dbZP00AtlxeRV8VCq6yzKV&s=womzPquRmStkhV6FidsyoYyS4jcfxhL-a9YkTTMdQHA&e= >> >
