Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

Richard Biener writes: > On Tue, 4 Jul 2023, Richard Sandiford wrote: > >> Richard Biener writes: >> > On Thu, 29 Jun 2023, Richard Biener wrote: >> > >> >> On Thu, 29 Jun 2023, Richard Sandiford wrote: >> >> >> >> > Richard Biener writes: >> >> > > With applying loop masking to epilogues on x8

Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

On Tue, 4 Jul 2023, Richard Sandiford wrote: > Richard Biener writes: > > On Thu, 29 Jun 2023, Richard Biener wrote: > > > >> On Thu, 29 Jun 2023, Richard Sandiford wrote: > >> > >> > Richard Biener writes: > >> > > With applying loop masking to epilogues on x86_64 AVX512 we see > >> > > some s

Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

Richard Biener writes: > On Thu, 29 Jun 2023, Richard Biener wrote: > >> On Thu, 29 Jun 2023, Richard Sandiford wrote: >> >> > Richard Biener writes: >> > > With applying loop masking to epilogues on x86_64 AVX512 we see >> > > some significant performance regressions when evaluating SPEC CPU 20

Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

On Thu, 29 Jun 2023, Richard Biener wrote: > On Thu, 29 Jun 2023, Richard Sandiford wrote: > > > Richard Biener writes: > > > With applying loop masking to epilogues on x86_64 AVX512 we see > > > some significant performance regressions when evaluating SPEC CPU 2017 > > > that are caused by stor

Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

On Thu, 29 Jun 2023, Richard Sandiford wrote: > Richard Biener writes: > > With applying loop masking to epilogues on x86_64 AVX512 we see > > some significant performance regressions when evaluating SPEC CPU 2017 > > that are caused by store-to-load forwarding fails across outer > > loop iterati

Re: [PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

Richard Biener writes: > With applying loop masking to epilogues on x86_64 AVX512 we see > some significant performance regressions when evaluating SPEC CPU 2017 > that are caused by store-to-load forwarding fails across outer > loop iterations when the inner loop does not iterate. Consider > >

[PATCH][RFC] target/110456 - avoid loop masking with zero distance dependences

With applying loop masking to epilogues on x86_64 AVX512 we see some significant performance regressions when evaluating SPEC CPU 2017 that are caused by store-to-load forwarding fails across outer loop iterations when the inner loop does not iterate. Consider for (j = 0; j < m; ++j) for (i