diff --git a/std/uni.d b/std/uni.d
index ac89ba0eb42..4345e549cae 100644
--- a/std/uni.d
+++ b/std/uni.d
@@ -6930,8 +6930,8 @@ unittest
     than $(LREF icmp). However keep in mind the warning below.)
 
     Params:
-        str1 = a string or a $(D ForwardRange) of $(D dchar)s
-        str2 = a string or a $(D ForwardRange) of $(D dchar)s
+        str1 = a string
+        str2 = a string
 
     Returns:
         An $(D int) that is 0 if the strings match,
@@ -6947,17 +6947,16 @@ unittest
         $(LREF icmp)
         $(XREF_PACK algorithm,comparison,cmp)
 +/
-int sicmp(S1, S2)(S1 str1, S2 str2)
-    if (isForwardRange!S1 && is(Unqual!(ElementType!S1) == dchar)
-    && isForwardRange!S2 && is(Unqual!(ElementType!S2) == dchar))
+int sicmp(S1, S2)(S1 str1, S2 str2) if (isSomeString!S1 && isSomeString!S2)
 {
     alias sTable = simpleCaseTable;
+    import std.utf : decode;
+
     size_t ridx=0;
     foreach (dchar lhs; str1)
     {
         if (ridx == str2.length)
             return 1;
-        import std.utf : decode;
         dchar rhs = decode(str2, ridx);
         int diff = lhs - rhs;
         if (!diff)
@@ -6991,7 +6990,8 @@ int sicmp(S1, S2)(S1 str1, S2 str2)
 }
 
 ///
-unittest{
+unittest
+{
     assert(sicmp("Август", "авгусТ") == 0);
     // Greek also works as long as there is no 1:M mapping in sight
     assert(sicmp("ΌΎ", "όύ") == 0);
@@ -7060,6 +7060,15 @@ private int fullCasedCmp(Range)(dchar lhs, dchar rhs, ref Range rtail)
     The cost of $(D icmp) being pedantically correct is
     slightly worse performance.
     )
+
+    Returns:
+        An $(D int) that is 0 if the strings match,
+        <0 if $(D str1) is lexicographically "less" than $(D str2),
+        >0 if $(D str1) is lexicographically "greater" than $(D str2)
+
+    See_Also:
+        $(LREF sicmp)
+        $(XREF_PACK algorithm,comparison,cmp)
 +/
 int icmp(S1, S2)(S1 str1, S2 str2)
     if (isForwardRange!S1 && is(Unqual!(ElementType!S1) == dchar)
@@ -7094,11 +7103,24 @@ int icmp(S1, S2)(S1 str1, S2 str2)
 }
 
 ///
-unittest{
+unittest
+{
     assert(icmp("Rußland", "Russland") == 0);
     assert(icmp("ᾩ -> \u1F70\u03B9", "\u1F61\u03B9 -> ᾲ") == 0);
 }
 
+/**
+ * By using $(XREF utf, byUTF) and its aliases, GC allocations via auto-decoding
+ * and thrown exceptions can be avoided, making `icmp` `@safe @nogc nothrow pure`.
+ */
+@safe @nogc nothrow pure unittest
+{
+    import std.utf : byDchar;
+
+    assert(icmp("Rußland".byDchar, "Russland".byDchar) == 0);
+    assert(icmp("ᾩ -> \u1F70\u03B9".byDchar, "\u1F61\u03B9 -> ᾲ".byDchar) == 0);
+}
+
 // overloads for the most common cases to reduce compile time
 @safe pure /*TODO nothrow*/
 {