[flang-commits] [flang] [flang] Correct MIN/MAX bug with DO CONCURRENT and REDUCE (PR #196708)

[Michael Klemm via flang-commits]

https://github.com/mjklemm created https://github.com/llvm/llvm-project/pull/196708

Flang used a the same reduction name for MIN/MAX with meant that if both happened, the compiler would use the first reduction operation it saw for both MIN/MAX.  This is now corrected.

>From 2a3d69893f9a47f094e76eae6126d5e64433dddb Mon Sep 17 00:00:00 2001
From: Michael Klemm <michael.klemm at amd.com>
Date: Sat, 9 May 2026 12:22:19 +0200
Subject: [PATCH] [flang] Correct MIN/MAX bug with DO CONCURRENT and REDUCE

Flang used a the same reduction name for MIN/MAX with meant that if both
happened, the compiler would use the first reduction operation it saw
for both MIN/MAX.  This is now corrected.
---
 flang/lib/Lower/Support/ReductionProcessor.cpp | 18 ++++++++++++++++--
 flang/test/Lower/loops3.f90                    |  2 +-
 2 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/flang/lib/Lower/Support/ReductionProcessor.cpp b/flang/lib/Lower/Support/ReductionProcessor.cpp
index d5387f7a59118..b3a27736d1616 100644
--- a/flang/lib/Lower/Support/ReductionProcessor.cpp
+++ b/flang/lib/Lower/Support/ReductionProcessor.cpp
@@ -194,9 +194,23 @@ ReductionProcessor::getReductionName(ReductionIdentifier redId,
   case ReductionIdentifier::NEQV:
     reductionName = "neqv_reduction";
     break;
-  default:
-    reductionName = "other_reduction";
+  case ReductionIdentifier::MAX:
+    reductionName = "max_reduction";
+    break;
+  case ReductionIdentifier::MIN:
+    reductionName = "min_reduction";
+    break;
+  case ReductionIdentifier::IAND:
+    reductionName = "iand_reduction";
     break;
+  case ReductionIdentifier::IOR:
+    reductionName = "ior_reduction";
+    break;
+  case ReductionIdentifier::IEOR:
+    reductionName = "ieor_reduction";
+    break;
+  default:
+    llvm_unreachable("unsupported reduction identifier");
   }
 
   return getReductionName(reductionName, kindMap, ty, isByRef);
diff --git a/flang/test/Lower/loops3.f90 b/flang/test/Lower/loops3.f90
index 5df3c4fd93703..9c295af66c510 100644
--- a/flang/test/Lower/loops3.f90
+++ b/flang/test/Lower/loops3.f90
@@ -12,7 +12,7 @@ subroutine loop_test
 
   ! CHECK: %[[M:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFloop_testEm"}
   ! CHECK: %[[SUM:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFloop_testEsum"}
-  ! CHECK: fir.do_concurrent.loop ({{.*}}) = ({{.*}}) to ({{.*}}) step ({{.*}}) local(@_QFloop_testEtmp_private_i32 %{{.*}} -> %{{.*}} : !fir.ref<i32>) reduce(@add_reduction_i32 #fir.reduce_attr<add> %[[SUM]]#0 -> %{{.*}}, @other_reduction_f32 #fir.reduce_attr<max> %[[M]]#0 -> %{{.*}} : !fir.ref<i32>, !fir.ref<f32>) {
+  ! CHECK: fir.do_concurrent.loop ({{.*}}) = ({{.*}}) to ({{.*}}) step ({{.*}}) local(@_QFloop_testEtmp_private_i32 %{{.*}} -> %{{.*}} : !fir.ref<i32>) reduce(@add_reduction_i32 #fir.reduce_attr<add> %[[SUM]]#0 -> %{{.*}}, @max_reduction_f32 #fir.reduce_attr<max> %[[M]]#0 -> %{{.*}} : !fir.ref<i32>, !fir.ref<f32>) {
   ! CHECK: %[[TMP:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFloop_testEtmp"}
   ! CHECK: %[[SUM_INNER:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFloop_testEsum"}
   ! CHECK: %[[M_INNER:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFloop_testEm"}