Re: [PATCH v4] aarch64: New RTL optimization pass avoid-store-forwarding.

2023-12-05 Thread Manos Anagnostakis
Hi Richard, the patch is working correctly with the four lines deleted and "get_addr" on "load_mem" inlined on "rtx_equal_for_cselib_1" call. Also changed store_info to str_info to avoid a warning of duplicate names on bootstrap (one definition rule). Rebased on top of your sme2 changes and submi

Re: [PATCH v4] aarch64: New RTL optimization pass avoid-store-forwarding.

2023-12-04 Thread Manos Anagnostakis
Στις Δευ 4 Δεκ 2023, 21:22 ο χρήστης Richard Sandiford < richard.sandif...@arm.com> έγραψε: > Manos Anagnostakis writes: > > This is an RTL pass that detects store forwarding from stores to larger > loads (load pairs). > > > > This optimization is SPEC2017-driven and was found to be beneficial fo

Re: [PATCH v4] aarch64: New RTL optimization pass avoid-store-forwarding.

2023-12-04 Thread Richard Sandiford
Manos Anagnostakis writes: > Στις Δευ 4 Δεκ 2023, 21:22 ο χρήστης Richard Sandiford < > richard.sandif...@arm.com> έγραψε: > >> Manos Anagnostakis writes: >> > This is an RTL pass that detects store forwarding from stores to larger >> loads (load pairs). >> > >> > This optimization is SPEC2017-dr

Re: [PATCH v4] aarch64: New RTL optimization pass avoid-store-forwarding.

2023-12-04 Thread Richard Sandiford
Manos Anagnostakis writes: > This is an RTL pass that detects store forwarding from stores to larger loads > (load pairs). > > This optimization is SPEC2017-driven and was found to be beneficial for some > benchmarks, > through testing on ampere1/ampere1a machines. > > For example, it can transf

[PATCH v4] aarch64: New RTL optimization pass avoid-store-forwarding.

2023-12-04 Thread Manos Anagnostakis
This is an RTL pass that detects store forwarding from stores to larger loads (load pairs). This optimization is SPEC2017-driven and was found to be beneficial for some benchmarks, through testing on ampere1/ampere1a machines. For example, it can transform cases like str d5, [sp, #320] fmul d