[llvm] [LV] Vectorize fmin/fmax reductions (PR #198300)

[Nashe Mncube via llvm-commits]

================
@@ -87,6 +87,50 @@ exit:
   ret i64 %res
 }
 
+; fmax last-index with strict predicate should not vectorize: FP strict
+; predicates still cannot use the first-argmax transform.
+define i64 @test_fmax_last_idx_strict(ptr %src, i64 %n) {
+; CHECK-LABEL: define i64 @test_fmax_last_idx_strict(
+; CHECK-SAME: ptr [[SRC:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[MAX_IDX:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[MAX_IDX_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[MAX_VAL:%.*]] = phi float [ f0xFF7FFFFF, %[[ENTRY]] ], [ [[MAX_VAL_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr float, ptr [[SRC]], i64 [[IV]]
+; CHECK-NEXT:    [[L:%.*]] = load float, ptr [[GEP]], align 4
+; CHECK-NEXT:    [[CMP:%.*]] = fcmp nnan olt float [[MAX_VAL]], [[L]]
+; CHECK-NEXT:    [[MAX_VAL_NEXT]] = call nnan float @llvm.maxnum.f32(float [[L]], float [[MAX_VAL]])
+; CHECK-NEXT:    [[MAX_IDX_NEXT]] = select i1 [[CMP]], i64 [[IV]], i64 [[MAX_IDX]]
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    [[RES:%.*]] = phi i64 [ [[MAX_IDX_NEXT]], %[[LOOP]] ]
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %max.idx = phi i64 [ 0, %entry ], [ %max.idx.next, %loop ]
+  %max.val = phi float [ 0xC7EFFFFFE0000000, %entry ], [ %max.val.next, %loop ]
+  %gep = getelementptr float, ptr %src, i64 %iv
+  %l = load float, ptr %gep
+  %cmp = fcmp nnan olt float %max.val, %l
+  %max.val.next = call nnan float @llvm.maxnum.f32(float %l, float %max.val)
----------------
nasherm wrote:

Done

https://github.com/llvm/llvm-project/pull/198300