GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 19:59:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add EIGEN_HAS_INTRINSIC_INT128 macro

Browse Source 
Add a new EIGEN_HAS_INTRINSIC_INT128 macro, and use this instead of __SIZEOF_INT128__. This fixes related issues with TensorIntDiv.h when building with Clang for Windows, where support for 128-bit integer arithmetic is advertised but broken in practice.
This commit is contained in:
Rasmus Munk Larsen 2019-11-06 14:24:33 -08:00
parent ee404667e2
commit cc3d0e6a40
2 changed files with 17 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Eigen/src/Core/util/Macros.h
View File

@ -757,6 +757,21 @@
#endif #endif
#endif #endif
// Does the compiler support the __int128 and __uint128_t extensions for 128-bit
// integer arithmetic?
//
// Clang and GCC define __SIZEOF_INT128__ when these extensions are supported,
// but we avoid using them in certain cases:
//
// * Building using Clang for Windows, where the Clang runtime library has
// 128-bit support only on LP64 architectures, but Windows is LLP64.
#ifndef EIGEN_HAS_BUILTIN_INT128
#if defined(__SIZEOF_INT128__) && !(EIGEN_OS_WIN && EIGEN_COMP_CLANG)
#define EIGEN_HAS_BUILTIN_INT128 1
#else
#define EIGEN_HAS_BUILTIN_INT128 0
#endif
#endif
//------------------------------------------------------------------------------------------ //------------------------------------------------------------------------------------------
// Preprocessor programming helpers // Preprocessor programming helpers

4
unsupported/Eigen/CXX11/src/Tensor/TensorIntDiv.h
View File

@ -105,7 +105,7 @@ namespace {
return __umul64hi(a, b); return __umul64hi(a, b);
#elif defined(SYCL_DEVICE_ONLY) #elif defined(SYCL_DEVICE_ONLY)
return cl::sycl::mul_hi(a, static_cast<uint64_t>(b)); return cl::sycl::mul_hi(a, static_cast<uint64_t>(b));
#elif defined(__SIZEOF_INT128__) #elif EIGEN_HAS_BUILTIN_INT128
__uint128_t v = static_cast<__uint128_t>(a) * static_cast<__uint128_t>(b); __uint128_t v = static_cast<__uint128_t>(a) * static_cast<__uint128_t>(b);
return static_cast<uint64_t>(v >> 64); return static_cast<uint64_t>(v >> 64);
#else #else
@ -124,7 +124,7 @@ namespace {
template <typename T> template <typename T>
struct DividerHelper<64, T> { struct DividerHelper<64, T> {
static EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE uint64_t computeMultiplier(const int log_div, const T divider) { static EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE uint64_t computeMultiplier(const int log_div, const T divider) {
#if defined(__SIZEOF_INT128__) && !defined(EIGEN_GPU_COMPILE_PHASE) && !defined(SYCL_DEVICE_ONLY) #if EIGEN_HAS_BUILTIN_INT128 && !defined(EIGEN_GPU_COMPILE_PHASE) && !defined(SYCL_DEVICE_ONLY)
return static_cast<uint64_t>((static_cast<__uint128_t>(1) << (64+log_div)) / static_cast<__uint128_t>(divider) - (static_cast<__uint128_t>(1) << 64) + 1); return static_cast<uint64_t>((static_cast<__uint128_t>(1) << (64+log_div)) / static_cast<__uint128_t>(divider) - (static_cast<__uint128_t>(1) << 64) + 1);
#else #else
const uint64_t shift = 1ULL << log_div; const uint64_t shift = 1ULL << log_div;