Hi, Commit r12-7687 exposed one miss optimization chance in function rs6000_maybe_emit_maxc_minc, for now it only considers comparison codes GE/GT/LE/LT, but it can support more variants with codes UNLT/UNLE/UNGT/UNGE by reversing them into the equivalent ones with GE/GT/LE/LT.
Bootstrapped and regtested on powerpc64-linux-gnu P8 and powerpc64le-linux-gnu P9 and P10. Is it ok for trunk? v1: https://gcc.gnu.org/pipermail/gcc-patches/2022-March/592204.html v2: Add more comments to provide better readability as Segher suggested. BR, Kewen ----- PR target/105002 gcc/ChangeLog: * config/rs6000/rs6000.cc (rs6000_maybe_emit_maxc_minc): Support more comparison codes UNLT/UNLE/UNGT/UNGE. --- gcc/config/rs6000/rs6000.cc | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc index f41b8f740ba..40db13e6498 100644 --- a/gcc/config/rs6000/rs6000.cc +++ b/gcc/config/rs6000/rs6000.cc @@ -15838,11 +15838,26 @@ rs6000_maybe_emit_maxc_minc (rtx dest, rtx op, rtx true_cond, rtx false_cond) rtx op1 = XEXP (op, 1); machine_mode compare_mode = GET_MODE (op0); machine_mode result_mode = GET_MODE (dest); - bool max_p = false; + bool max_p; if (result_mode != compare_mode) return false; + /* See the comments of this function, it simply expects GE/GT/LE/LT in + the checks, but for the reversible equivalent UNLT/UNLE/UNGT/UNGE, + we need to do the reversions first to make the following checks + support fewer cases, like: + + (a UNLT b) ? op1 : op2 => (a >= b) ? op2 : op1; + (a UNLE b) ? op1 : op2 => (a > b) ? op2 : op1; + (a UNGT b) ? op1 : op2 => (a <= b) ? op2 : op1; + (a UNGE b) ? op1 : op2 => (a < b) ? op2 : op1; */ + if (code == UNGE || code == UNGT || code == UNLE || code == UNLT) + { + code = reverse_condition_maybe_unordered (code); + std::swap (true_cond, false_cond); + } + if (code == GE || code == GT) max_p = true; else if (code == LE || code == LT) -- 2.27.0
