From: Roger Sayle Date: Sat, 7 Feb 2026 08:04:40 +0000 (+0000) Subject: PR tree-optimization/123958: FMA vs pow(x,2.0) [vs errno] X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=30886aeca75076bfd50b0fff8d76942603bee763;p=thirdparty%2Fgcc.git PR tree-optimization/123958: FMA vs pow(x,2.0) [vs errno] This is my proposed solution to PR123958 (and PR124002) which is a regression exposed by my recent change to avoid expanding pow(x,2.0) to x*x with -fmath-errno (the default) when we can't guarantee that errno shouldn't be updated. The problem is that the logic to convert pow(x,2.0) was also duplicated (but unused) in tree-ssa-math-opts where it's intended to perform this conversion in order to expose fused-multiply-add instructions when supported by the target. The issue is that this "vestigial" code has bit-rotten over the years, and incorrectly updates vdefs when changing vops, tiggering an ICE. My pragmatic solution to this is to simply delete the problematic code; the decision of whether pow(x,2.0) should be expanded is left to the earlier pow_expand pass (which is what it's designed for), and the later FMA pass can make use of any resulting FP multiplications. Not only does this avoid the PHI related ICE, but also fixes the original PR (on updating errno) on targets with FMA, e.g. aarch64 and recent x86_64 architectures (such as -march=znver3). 2026-02-07 Roger Sayle gcc/ChangeLog PR middle-end/123826 PR tree-optimization/123958 PR c++/124002 * tree-ssa-math-opts.cc (math_opts_dom_walker::after_dom_children): Delete code that (mis)handled conversion of pow(x,2.0) to x*x. gcc/testsuite/ChangeLog PR middle-end/123826 PR tree-optimization/123958 PR c++/124002 * g++.target/i386/pr124002.C: New test case. * gcc.target/i386/pr123958.c: Likewise. * gcc.dg/errno-4.c: Likewise. --- diff --git a/gcc/testsuite/g++.target/i386/pr124002.C b/gcc/testsuite/g++.target/i386/pr124002.C new file mode 100644 index 00000000000..033cd3de870 --- /dev/null +++ b/gcc/testsuite/g++.target/i386/pr124002.C @@ -0,0 +1,15 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -march=znver3" } */ + +extern "C" double pow(double, double); +double sigmoid_x; +double sigmoid() { + if (sigmoid_x) + return sigmoid_x; + return 0; +} +double der_sigmoid() { + double tmp = sigmoid(); + return tmp - pow(tmp, 2); +} + diff --git a/gcc/testsuite/gcc.dg/errno-4.c b/gcc/testsuite/gcc.dg/errno-4.c new file mode 100644 index 00000000000..d7b80c1ca41 --- /dev/null +++ b/gcc/testsuite/gcc.dg/errno-4.c @@ -0,0 +1,35 @@ +/* PR middle-end/123826 */ +/* { dg-do run } */ +/* { dg-options "-O2" } */ + +#ifdef __NO_MATH_ERRNO__ +int main() { return 0; } +#else +#include +#include +#include + +double foo(double x) +{ + return x + pow(x, 2.0); +} + +int main() +{ +#ifdef math_errhandling +#ifdef MATH_ERRNO + if ((math_errhandling & MATH_ERRNO) == 0) + return 0; +#else + if ((math_errhandling & 1) == 0) + return 0; +#endif +#endif + + errno = 0; + double x = foo(DBL_MAX); + if (errno != ERANGE) + __builtin_abort (); + return 0; +} +#endif diff --git a/gcc/testsuite/gcc.target/i386/pr123958.c b/gcc/testsuite/gcc.target/i386/pr123958.c new file mode 100644 index 00000000000..75064e880b4 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr123958.c @@ -0,0 +1,18 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -march=znver3" } */ + +float fnk00xX_rae1_1; +double pow(double, double); +double sqrt(double); +void write_r4(float *); +void fnk00xX() { + int i; + float rs2; + double ds2; + for (; i; i += 1) { + sqrt(i); + ds2 = ds2 + pow(fnk00xX_rae1_1, 2.0); + rs2 = ds2; + } + write_r4(&rs2); +} diff --git a/gcc/tree-ssa-math-opts.cc b/gcc/tree-ssa-math-opts.cc index 1655b68cb9c..6d69711fffb 100644 --- a/gcc/tree-ssa-math-opts.cc +++ b/gcc/tree-ssa-math-opts.cc @@ -6623,25 +6623,6 @@ math_opts_dom_walker::after_dom_children (basic_block bb) { switch (gimple_call_combined_fn (stmt)) { - CASE_CFN_POW: - if (gimple_call_lhs (stmt) - && TREE_CODE (gimple_call_arg (stmt, 1)) == REAL_CST - && real_equal (&TREE_REAL_CST (gimple_call_arg (stmt, 1)), - &dconst2) - && convert_mult_to_fma (stmt, - gimple_call_arg (stmt, 0), - gimple_call_arg (stmt, 0), - &fma_state)) - { - unlink_stmt_vdef (stmt); - if (gsi_remove (&gsi, true) - && gimple_purge_dead_eh_edges (bb)) - *m_cfg_changed_p = true; - release_defs (stmt); - continue; - } - break; - case CFN_COND_MUL: if (convert_mult_to_fma (stmt, gimple_call_arg (stmt, 1),