Merge branch 'develop' into uif2-initial

include/ck/utility/f8_utils.hpp

@@ -6,8 +6,6 @@
 #include "ck/utility/data_type.hpp"
 
 // these conversions are disabled if native conversions available
-#if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
 namespace ck {
 
 // fp8 rounding modes
@@ -244,5 +242,3 @@ __host__ __device__ Y cast_from_f8(X x)
 }
 
 } // namespace ck::utils
-#endif // #if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-#endif // #if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)

include/ck/utility/math.hpp

@@ -150,28 +150,6 @@ __host__ __device__ constexpr T clamp(const T& x, const T& lowerbound, const T&
     return min(max(x, lowerbound), upperbound);
 }
 
-// disallow implicit type casting
-template <typename T>
-__device__ T exp(T x);
-
-// TODO: add f16 support using v_exp_f16
-
-template <>
-__device__ float exp<float>(float x)
-{
-    return __expf(x);
-}
-
-template <>
-__device__ double exp<double>(double x)
-{
-    return exp(x);
-}
-
-static inline __host__ float exp(float x) { return std::expf(x); }
-
-static inline __host__ double exp(double x) { return std::exp(x); }
-
 // greatest common divisor, aka highest common factor
 __host__ __device__ constexpr index_t gcd(index_t x, index_t y)
 {

include/ck/utility/statically_indexed_array_multi_index.hpp

@@ -5,6 +5,7 @@
 #define CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
 
 #include "common_header.hpp"
+#include "ck/utility/math_v2.hpp"
 
 namespace ck {