Guard rcp rounded implementation as well

Since rcp implementations of non-default rounded versions are not correct or supported in OCML, guard them using the same macro OCML_BASIC_ROUNDED_OPERATIONS. Also update the docs and tests.


[ROCm/hip commit: 0cfaa52d15]

projects/hip/tests/src/deviceLib/hipFloatMath.cpp

@@ -38,11 +38,13 @@ __global__ void floatMath(float* In, float* Out) {
     Out[tid] = __cosf(In[tid]);
     Out[tid] = __exp10f(Out[tid]);
     Out[tid] = __expf(Out[tid]);
+#if defined OCML_BASIC_ROUNDED_OPERATIONS
     Out[tid] = __frsqrt_rn(Out[tid]);
-    //Out[tid] = __fsqrt_rd(Out[tid]);
-    //Out[tid] = __fsqrt_rn(Out[tid]);
-    //Out[tid] = __fsqrt_ru(Out[tid]);
-    //Out[tid] = __fsqrt_rz(Out[tid]);
+    Out[tid] = __fsqrt_rd(Out[tid]);
+    Out[tid] = __fsqrt_rn(Out[tid]);
+    Out[tid] = __fsqrt_ru(Out[tid]);
+    Out[tid] = __fsqrt_rz(Out[tid]);
+#endif
     Out[tid] = __log10f(Out[tid]);
     Out[tid] = __log2f(Out[tid]);
     Out[tid] = __logf(Out[tid]);