From b5b33e113434be909e8a6d7b93824196fb6925c0 Mon Sep 17 00:00:00 2001 From: Wilco Dijkstra Date: Mon, 5 Mar 2018 14:40:55 +0000 Subject: [PATCH] [AArch64] PR84114: Avoid reassociating FMA As discussed in the PR, the reassociation phase runs before FMAs are formed and so can significantly reduce FMA opportunities. Although reassociation could be switched off, it helps in many cases, so a better alternative is to only avoid reassociation of floating point additions. This fixes the testcase and gives 1% speedup on SPECFP2017, fixing the performance regression. gcc/ PR tree-optimization/84114 * config/aarch64/aarch64.c (aarch64_reassociation_width) Avoid reassociation of FLOAT_MODE addition. From-SVN: r258248 --- gcc/ChangeLog | 6 ++++++ gcc/config/aarch64/aarch64.c | 7 ++++--- 2 files changed, 10 insertions(+), 3 deletions(-) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 28e31de036f..eeb36339911 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,9 @@ +2018-03-05 Wilco Dijkstra + + PR tree-optimization/84114 + * config/aarch64/aarch64.c (aarch64_reassociation_width) + Avoid reassociation of FLOAT_MODE addition. + 2018-03-05 Olga Makhotina * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET, diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c index 2f98a21acf1..07c55b132a7 100644 --- a/gcc/config/aarch64/aarch64.c +++ b/gcc/config/aarch64/aarch64.c @@ -1094,15 +1094,16 @@ aarch64_min_divisions_for_recip_mul (machine_mode mode) return aarch64_tune_params.min_div_recip_mul_df; } +/* Return the reassociation width of treeop OPC with mode MODE. */ static int -aarch64_reassociation_width (unsigned opc ATTRIBUTE_UNUSED, - machine_mode mode) +aarch64_reassociation_width (unsigned opc, machine_mode mode) { if (VECTOR_MODE_P (mode)) return aarch64_tune_params.vec_reassoc_width; if (INTEGRAL_MODE_P (mode)) return aarch64_tune_params.int_reassoc_width; - if (FLOAT_MODE_P (mode)) + /* Avoid reassociating floating point addition so we emit more FMAs. */ + if (FLOAT_MODE_P (mode) && opc != PLUS_EXPR) return aarch64_tune_params.fp_reassoc_width; return 1; } -- 2.11.4.GIT