Merge branch 'master' into getsymboladdress

[ROCm/hip commit: 8610128c3e]

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

@@ -34,6 +34,7 @@ THE SOFTWARE.
 #pragma clang diagnostic ignored "-Wunused-variable"
 
 __device__ void double_precision_intrinsics() {
+#if defined OCML_BASIC_ROUNDED_OPERATIONS
     __dadd_rd(0.0, 1.0);
     __dadd_rn(0.0, 1.0);
     __dadd_ru(0.0, 1.0);
@@ -62,6 +63,7 @@ __device__ void double_precision_intrinsics() {
     __fma_rn(1.0, 2.0, 3.0);
     __fma_ru(1.0, 2.0, 3.0);
     __fma_rz(1.0, 2.0, 3.0);
+#endif
 }
 
 __global__ void compileDoublePrecisionIntrinsics(int ignored) {

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]);

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

@@ -39,6 +39,7 @@ __device__ void single_precision_intrinsics() {
     __cosf(0.0f);
     __exp10f(0.0f);
     __expf(0.0f);
+#if defined OCML_BASIC_ROUNDED_OPERATIONS
     __fadd_rd(0.0f, 1.0f);
     __fadd_rn(0.0f, 1.0f);
     __fadd_ru(0.0f, 1.0f);
@@ -47,7 +48,9 @@ __device__ void single_precision_intrinsics() {
     __fdiv_rn(4.0f, 2.0f);
     __fdiv_ru(4.0f, 2.0f);
     __fdiv_rz(4.0f, 2.0f);
+#endif
     __fdividef(4.0f, 2.0f);
+#if defined OCML_BASIC_ROUNDED_OPERATIONS
     __fmaf_rd(1.0f, 2.0f, 3.0f);
     __fmaf_rn(1.0f, 2.0f, 3.0f);
     __fmaf_ru(1.0f, 2.0f, 3.0f);
@@ -69,6 +72,7 @@ __device__ void single_precision_intrinsics() {
     __fsub_rn(2.0f, 1.0f);
     __fsub_ru(2.0f, 1.0f);
     __fsub_rz(2.0f, 1.0f);
+#endif
     __log10f(1.0f);
     __log2f(1.0f);
     __logf(1.0f);