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.

tests/src/deviceLib/hipDoublePrecisionIntrinsics.cpp

@@ -47,12 +47,10 @@ __device__ void double_precision_intrinsics() {
     __dmul_rn(1.0, 2.0);
     __dmul_ru(1.0, 2.0);
     __dmul_rz(1.0, 2.0);
-#endif
     __drcp_rd(2.0);
     __drcp_rn(2.0);
     __drcp_ru(2.0);
     __drcp_rz(2.0);
-#if defined OCML_BASIC_ROUNDED_OPERATIONS
     __dsqrt_rd(4.0);
     __dsqrt_rn(4.0);
     __dsqrt_ru(4.0);

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]);

tests/src/deviceLib/hipSinglePrecisionIntrinsics.cpp

@@ -59,13 +59,11 @@ __device__ void single_precision_intrinsics() {
     __fmul_rn(1.0f, 2.0f);
     __fmul_ru(1.0f, 2.0f);
     __fmul_rz(1.0f, 2.0f);
-#endif
     __frcp_rd(2.0f);
     __frcp_rn(2.0f);
     __frcp_ru(2.0f);
     __frcp_rz(2.0f);
     __frsqrt_rn(4.0f);
-#if defined OCML_BASIC_ROUNDED_OPERATIONS
     __fsqrt_rd(4.0f);
     __fsqrt_rn(4.0f);
     __fsqrt_ru(4.0f);