[SYCL] Add marray support to common + some math functions

[dm-vodopyanov]

This patch adds marray support to all functions from Table 179 of SYCL 2020 spec + to functions fabs, ilogb, fmax, fmin, ldexp, pown, rootn from Table 175 + to function exp10 from Table 177.

E2E tests: intel/llvm-test-suite#1656

[dm-vodopyanov]

/verify with intel/llvm-test-suite#1656

[AlexeySachkov]

Looks like we can add one more high-level macro to generalize definitions of pown, rootn and ldexp. Similar to __SYCL_MATH_FUNCTION_2_OVERLOAD(atan2)

[dm-vodopyanov]

Basically yes, the reason I didn't do this is to stick to code style in this part of the file.

[AlexeySachkov]

I suggest to outline such tiny additions into a separate PR, because we can merge them almost right away.

[dm-vodopyanov]

/verify with intel/llvm-test-suite#1656

[KornevNikita]

/verify with intel/llvm-test-suite#1656

[KornevNikita]

@bader @AlexeySachkov lol, now clang-format suggests to get these suspicious changes back. Should I apply them or I can just add ignore-lint label?

[bader]

@bader @AlexeySachkov lol, now clang-format suggests to get these suspicious changes back. Should I apply them or I can just add ignore-lint label?

I'm against ignoring formatting issues in the code we own. If you prefer to keep existing formatting, you can disable clang-format for some parts. https://clang.llvm.org/docs/ClangFormatStyleOptions.html#disabling-formatting-on-a-piece-of-code

Some suggestions look reasonable. E.g.

@@ -182,9 +182,8 @@ __SYCL_MATH_FUNCTION_2_OVERLOAD(remainder)
 template <typename T, size_t N>
 inline __SYCL_ALWAYS_INLINE
     std::enable_if_t<__FAST_MATH_SGENFLOAT(T), marray<T, N>>
-    powr(marray<T, N> x, marray<T, N> y) __NOEXC {
-  __SYCL_MATH_FUNCTION_2_OVERLOAD_IMPL(powr)
-}
+    powr(marray<T, N> x,
+         marray<T, N> y) __NOEXC{__SYCL_MATH_FUNCTION_2_OVERLOAD_IMPL(powr)}

Do you know why clang-format gives different results? I know that the version must be the same, so I suppose there were changes in the config file. Right?

[KornevNikita]

@bader honestly I don't know.

This one definitely looks strange:

 __SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmax)
-__SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmin)
+    __SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmin)
 
 #undef __SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD
 
-template <typename T, size_t N>
-inline __SYCL_ALWAYS_INLINE
-    std::enable_if_t<detail::is_sgenfloat<T>::value, marray<T, N>>
-    ldexp(marray<T, N> x, marray<int, N> k) __NOEXC {
+        template <typename T, size_t N>
+        inline __SYCL_ALWAYS_INLINE
+    std::enable_if_t<detail::is_sgenfloat<T>::value, marray<T, N>> ldexp(
+        marray<T, N> x, marray<int, N> k) __NOEXC {

I'm okay to apply these suggestions if you think that this formatting is right.

[bader]

This should be the decision of the code owners. I've given you options on how to keep the existing formatting so you can change it, keep it, or mix these options.

 -__SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmax)
 -__SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmin)
 +    __SYCL_MATH_FUNCTION_2_SGENFLOAT_Y_OVERLOAD(fmin)

My personal view is that this change makes formatting inconsistent, so wouldn't apply it. I suppose ldexp declaration formatting is off because of this misalignment with macro.

[KornevNikita]

/verify with intel/llvm-test-suite#1656

[AlexeySachkov]

@bader @AlexeySachkov lol, now clang-format suggests to get these suspicious changes back. Should I apply them or I can just add ignore-lint label?

Current formatting looks good to me and I would prefer not to revert back to the weird one

[AlexeySachkov]

@againull, @intel/llvm-reviewers-runtime, could you please take a look?

[KornevNikita]

/verify with intel/llvm-test-suite#1656