[Support] Workaround a GCC 4.8 bug on constant expression evaluation.

[hliao]

Summary

• The following stripped code trigger a gcc-4.8 bug. To work that around, move the alignment evaluation into template parameter.

// https://godbolt.org/z/58p5_X
//
#include <cstddef>
#include <cstdint>

enum { aligned = 0, unaligned = 1 };

template <typename T, int alignment> struct PickAlignment {
  enum { value = alignment == 0 ? alignof(T) : alignment };
};

template <typename ValueType, std::size_t Alignment> struct packed {
private:
  struct {
    alignas(
        PickAlignment<ValueType, Alignment>::value) char buffer[sizeof(int)];
  } Value;
};

using ule16_t = packed<uint16_t, unaligned>;

ule16_t x;

• Also, replace alignas with LLVMALIGN_AS to improve the compiler compatibility.

[MaskRay]

https://llvm.org/docs/GettingStarted.html#id7

GCC >=5.1.0 C/C++ compiler

The current recommendation is >=5.1.0...

[hliao]

Format the commit message

[hliao]

In D65452#1606342, @MaskRay wrote:

https://llvm.org/docs/GettingStarted.html#id7

GCC >=5.1.0 C/C++ compiler

The current recommendation is >=5.1.0...

But, unfortunately, there are many old systems with gcc-4.8 only, like ubuntu 16.04, centos 4.x, and other long-term systems

[hliao]

BTW, if I read that correct, even though gcc 5.1 is listed as requirement, but that page also mentioned that gcc 4.8 could compile correctly the current LLVM codebase, https://llvm.org/docs/GettingStarted.html#host-c-toolchain-both-compiler-and-standard-library. As gcc 4.8.x is widely used in many long-term maintained systems, it would be nice to compile LLVM correctly against it.

[xbolva00]

Is it worth to workaround old gcc?

Support for gcc 4.8 will be dropped soon(ish) anyway..

[hliao]

In D65452#1606400, @xbolva00 wrote:

Is it worth to workaround old gcc?

Support for gcc 4.8 will be dropped soon(ish) anyway..

I don't want to be pedantic, but will means not right now. Many long-term systems, says CentOS 7.2 (with gcc-4.8.5 and the same issue), would be EOL up to 2024.

[xbolva00]

But one can use devtoolset on centos/rhel :)

There is a llvm-dev discusion (but now dead, Google blocked it? @jfb @chandlerc) about bumping gcc versions, LLVM wants to use C++-14.

Anyway, this small change looks fine.

[hliao]

In D65452#1606444, @xbolva00 wrote:

But one can use devtoolset on centos/rhel :)

There is a llvm-dev discusion (but now dead, Google blocked it? @jfb @chandlerc) about bumping gcc versions, LLVM wants to use C++-14.

Anyway, this small change looks fine.

thanks so much!

[jfb]

In D65452#1606444, @xbolva00 wrote:

But one can use devtoolset on centos/rhel :)

There is a llvm-dev discusion (but now dead, Google blocked it? @jfb @chandlerc) about bumping gcc versions, LLVM wants to use C++-14.

Anyway, this small change looks fine.

It should happen soon indeed. It was supposed to be months ago, but things slipped a bit. The change is fine since we do still kinda support GCC 4.8 for now, though I think we should remove LLVM_ALIGNAS (I'll do so...).

llvm/trunk/include/llvm/Support/Endian.h

 }
 
 } // end namespace endian
 
 namespace detail {
 
 template<typename ValueType,
          endianness Endian,
-         std::size_t Alignment>
+         std::size_t Alignment,
+         std::size_t ALIGN = PickAlignment<ValueType, Alignment>::value>
 struct packed_endian_specific_integral {
   using value_type = ValueType;
   static constexpr endianness endian = Endian;
   static constexpr std::size_t alignment = Alignment;
 
   packed_endian_specific_integral() = default;
 
   explicit packed_endian_specific_integral(value_type val) { *this = val; }
struct packed_endian_specific_integral {
 
   packed_endian_specific_integral &operator&=(value_type newValue) {
     *this = *this & newValue;
     return *this;
   }
 
 private:
   struct {
-    alignas(PickAlignment<value_type,
-                          alignment>::value) char buffer[sizeof(value_type)];
+    LLVM_ALIGNAS(ALIGN) char buffer[sizeof(value_type)];
   } Value;
 
 public:
   struct ref {
     explicit ref(void *Ptr) : Ptr(Ptr) {}
 
     operator value_type() const {
       return endian::read<value_type, endian, alignment>(Ptr);