From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 1666) id 7DEF938518B9; Fri, 18 Nov 2022 07:37:46 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 7DEF938518B9 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1668757066; bh=ovqBE01orpzIOgBqYziA/RWVDthgjNXTleohBlRIInU=; h=From:To:Subject:Date:From; b=xVTO+BMIEkXwIe/joH5YJh6mJRxknunC4OxlRZrQVOuc7Vh2x+Xu8m1Xqm0WE710t aAy0rCr1URb6M0R2lTqADnDfVqWDo6E0zZfu1FtBbVGyuR5upMMuAubTDg6zIoIT1g RgS5MPzTBsJJvkocXZOgplKRvH+D6aAANtyhFKaw= MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="utf-8" From: Richard Biener To: gcc-cvs@gcc.gnu.org Subject: [gcc r13-4137] tree-optimization/107647 - avoid FMA from SLP with -ffp-contract=off X-Act-Checkin: gcc X-Git-Author: Richard Biener X-Git-Refname: refs/heads/master X-Git-Oldrev: f0024bfb228f94e60e06dc32a4983e40a9b90be5 X-Git-Newrev: c5df8392c5848c0462558f41cdf6eab31db301cf Message-Id: <20221118073746.7DEF938518B9@sourceware.org> Date: Fri, 18 Nov 2022 07:37:46 +0000 (GMT) List-Id: https://gcc.gnu.org/g:c5df8392c5848c0462558f41cdf6eab31db301cf commit r13-4137-gc5df8392c5848c0462558f41cdf6eab31db301cf Author: Richard Biener Date: Thu Nov 17 09:43:31 2022 +0100 tree-optimization/107647 - avoid FMA from SLP with -ffp-contract=off Only with -ffp-contract=fast we can synthesize FMA operations like vfmaddsub231ps, so properly guard the transform in SLP pattern detection. PR tree-optimization/107647 * tree-vect-slp-patterns.cc (addsub_pattern::recognize): Only allow FMA generation with -ffp-contract=fast for FP types. (complex_mul_pattern::matches): Likewise. * gcc.target/i386/pr107647.c: New testcase. Diff: --- gcc/testsuite/gcc.target/i386/pr107647.c | 17 +++++++++++++++++ gcc/tree-vect-slp-patterns.cc | 15 +++++++++++---- 2 files changed, 28 insertions(+), 4 deletions(-) diff --git a/gcc/testsuite/gcc.target/i386/pr107647.c b/gcc/testsuite/gcc.target/i386/pr107647.c new file mode 100644 index 00000000000..45fcb55d698 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr107647.c @@ -0,0 +1,17 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -ffp-contract=off -mavx2 -mfma" } */ + +void cscal(int n, float da_r, float *x) +{ + for (int i = 0; i < n; i += 4) + { + float temp0 = da_r * x[i] - x[i+1]; + float temp1 = da_r * x[i+2] - x[i+3]; + x[i+1] = da_r * x[i+1] + x[i]; + x[i+3] = da_r * x[i+3] + x[i+2]; + x[i] = temp0; + x[i+2] = temp1; + } +} + +/* { dg-final { scan-assembler-not "fma" } } */ diff --git a/gcc/tree-vect-slp-patterns.cc b/gcc/tree-vect-slp-patterns.cc index dc694b8e531..122d697a809 100644 --- a/gcc/tree-vect-slp-patterns.cc +++ b/gcc/tree-vect-slp-patterns.cc @@ -1035,8 +1035,11 @@ complex_mul_pattern::matches (complex_operation_t op, auto_vec left_op, right_op; slp_tree add0 = NULL; - /* Check if we may be a multiply add. */ + /* Check if we may be a multiply add. It's only valid to form FMAs + with -ffp-contract=fast. */ if (!mul0 + && (flag_fp_contract_mode == FP_CONTRACT_FAST + || !FLOAT_TYPE_P (SLP_TREE_VECTYPE (l0node[0]))) && vect_match_expression_p (l0node[0], PLUS_EXPR)) { auto vals = SLP_TREE_CHILDREN (l0node[0]); @@ -1501,9 +1504,13 @@ addsub_pattern::recognize (slp_tree_to_load_perm_map_t *, } /* Now we have either { -, +, -, + ... } (!l0add_p) or { +, -, +, - ... } - (l0add_p), see whether we have FMA variants. */ - if (!l0add_p - && vect_match_expression_p (SLP_TREE_CHILDREN (l0node)[0], MULT_EXPR)) + (l0add_p), see whether we have FMA variants. We can only form FMAs + if allowed via -ffp-contract=fast. */ + if (flag_fp_contract_mode != FP_CONTRACT_FAST + && FLOAT_TYPE_P (SLP_TREE_VECTYPE (l0node))) + ; + else if (!l0add_p + && vect_match_expression_p (SLP_TREE_CHILDREN (l0node)[0], MULT_EXPR)) { /* (c * d) -+ a */ if (vect_pattern_validate_optab (IFN_VEC_FMADDSUB, node))