diff options
author | Antonio Sanchez <cantonios@google.com> | 2020-11-17 15:32:44 -0800 |
---|---|---|
committer | Antonio Sánchez <cantonios@google.com> | 2020-11-18 20:32:35 +0000 |
commit | 17268b155d54422f1294130c0fb8c178757d911a (patch) | |
tree | 2be3d541729f3e9be6a180a58270bae10156df4f | |
parent | 41d5d5334b8a4e364dfd88dcd91f6cd38834b8ed (diff) |
Add bit_cast for half/bfloat to/from uint16_t, fix TensorRandom
The existing `TensorRandom.h` implementation makes the assumption that
`half` (`bfloat16`) has a `uint16_t` member `x` (`value`), which is not
always true. This currently fails on arm64, where `x` has type `__fp16`.
Added `bit_cast` specializations to allow casting to/from `uint16_t`
for both `half` and `bfloat16`. Also added tests in
`half_float`, `bfloat16_float`, and `cxx11_tensor_random` to catch
these errors in the future.
-rw-r--r-- | Eigen/src/Core/arch/Default/BFloat16.h | 24 | ||||
-rw-r--r-- | Eigen/src/Core/arch/Default/Half.h | 30 | ||||
-rw-r--r-- | test/bfloat16_float.cpp | 101 | ||||
-rw-r--r-- | test/half_float.cpp | 124 | ||||
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorRandom.h | 47 | ||||
-rw-r--r-- | unsupported/test/cxx11_tensor_random.cpp | 18 |
6 files changed, 195 insertions, 149 deletions
diff --git a/Eigen/src/Core/arch/Default/BFloat16.h b/Eigen/src/Core/arch/Default/BFloat16.h index 63ceace1e..6f81fe382 100644 --- a/Eigen/src/Core/arch/Default/BFloat16.h +++ b/Eigen/src/Core/arch/Default/BFloat16.h @@ -69,7 +69,7 @@ struct bfloat16 : public bfloat16_impl::bfloat16_base { template<class T> explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16(const T& val) : bfloat16_impl::bfloat16_base(bfloat16_impl::float_to_bfloat16_rtne<internal::is_integral<T>::value>(static_cast<float>(val))) {} - + explicit EIGEN_DEVICE_FUNC bfloat16(float f) : bfloat16_impl::bfloat16_base(bfloat16_impl::float_to_bfloat16_rtne<false>(f)) {} @@ -88,7 +88,7 @@ struct bfloat16 : public bfloat16_impl::bfloat16_base { // +0.0 and -0.0 become false, everything else becomes true. return (value & 0x7fff) != 0; } -#endif +#endif }; } // namespace Eigen @@ -272,10 +272,14 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw truncate_to_bfloat16(const return output; } -EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw raw_uint16_to_bfloat16(unsigned short value) { +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw raw_uint16_to_bfloat16(numext::uint16_t value) { return __bfloat16_raw(value); } +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR numext::uint16_t raw_bfloat16_as_uint16(const __bfloat16_raw& bf) { + return bf.value; +} + // float_to_bfloat16_rtne template specialization that does not make any // assumption about the value of its function argument (ff). template <> @@ -454,7 +458,7 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<fals // float_to_bfloat16_rtne template specialization that assumes that its function // argument (ff) is either a normal floating point number, or +/-infinity, or // zero. Used to improve the runtime performance of conversion from an integer -// type to bfloat16. +// type to bfloat16. template <> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<true>(float ff) { #if (defined(EIGEN_HAS_CUDA_BF16) && defined(EIGEN_HAS_HIP_BF16)) @@ -691,7 +695,17 @@ bool (isfinite)(const Eigen::bfloat16& h) { return (bfloat16_impl::isfinite)(h); } -} // namespace numext +template <> +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::bfloat16 bit_cast<Eigen::bfloat16, uint16_t>(const uint16_t& src) { + return Eigen::bfloat16(Eigen::bfloat16_impl::raw_uint16_to_bfloat16(src)); +} + +template <> +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC uint16_t bit_cast<uint16_t, Eigen::bfloat16>(const Eigen::bfloat16& src) { + return Eigen::bfloat16_impl::raw_bfloat16_as_uint16(src); +} + +} // namespace numext } // namespace Eigen #endif // EIGEN_BFLOAT16_H diff --git a/Eigen/src/Core/arch/Default/Half.h b/Eigen/src/Core/arch/Default/Half.h index 0bc1e9d19..fda38bcb0 100644 --- a/Eigen/src/Core/arch/Default/Half.h +++ b/Eigen/src/Core/arch/Default/Half.h @@ -494,6 +494,19 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw raw_uint16_to_h #endif } +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR numext::uint16_t raw_half_as_uint16(const __half_raw& h) { + // HIP/CUDA/Default have a member 'x' of type uint16_t. + // For ARM64 native half, the member 'x' is of type __fp16, so we need to bit-cast. + // For SYCL, cl::sycl::half is _Float16, so cast directly. +#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC) + return numext::bit_cast<numext::uint16_t>(h.x); +#elif defined(SYCL_DEVICE_ONLY) + return numext::bit_cast<numext::uint16_t>(h); +#else + return h.x; +#endif +} + union float32_bits { unsigned int u; float f; @@ -812,10 +825,11 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half __ldg(const Eigen::half* ptr) } #endif -#if defined(EIGEN_GPU_COMPILE_PHASE) namespace Eigen { namespace numext { +#if defined(EIGEN_GPU_COMPILE_PHASE) + template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE bool(isnan)(const Eigen::half& h) { return (half_impl::isnan)(h); @@ -830,8 +844,20 @@ template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE bool(isfinite)(const Eigen::half& h) { return (half_impl::isfinite)(h); } + +#endif + +template <> +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half bit_cast<Eigen::half, uint16_t>(const uint16_t& src) { + return Eigen::half(Eigen::half_impl::raw_uint16_to_half(src)); +} + +template <> +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC uint16_t bit_cast<uint16_t, Eigen::half>(const Eigen::half& src) { + return Eigen::half_impl::raw_half_as_uint16(src); +} + } // namespace numext } // namespace Eigen -#endif #endif // EIGEN_HALF_H diff --git a/test/bfloat16_float.cpp b/test/bfloat16_float.cpp index 09df2b2f2..fc648dfec 100644 --- a/test/bfloat16_float.cpp +++ b/test/bfloat16_float.cpp @@ -13,6 +13,9 @@ #include <Eigen/src/Core/arch/Default/BFloat16.h> +#define VERIFY_BFLOAT16_BITS_EQUAL(h, bits) \ + VERIFY_IS_EQUAL((numext::bit_cast<numext::uint16_t>(h)), (static_cast<numext::uint16_t>(bits))) + // Make sure it's possible to forward declare Eigen::bfloat16 namespace Eigen { struct bfloat16; @@ -58,31 +61,45 @@ void test_conversion() { using Eigen::bfloat16_impl::__bfloat16_raw; + // Round-trip casts + VERIFY_IS_EQUAL( + numext::bit_cast<bfloat16>(numext::bit_cast<numext::uint16_t>(bfloat16(1.0f))), + bfloat16(1.0f)); + VERIFY_IS_EQUAL( + numext::bit_cast<bfloat16>(numext::bit_cast<numext::uint16_t>(bfloat16(0.5f))), + bfloat16(0.5f)); + VERIFY_IS_EQUAL( + numext::bit_cast<bfloat16>(numext::bit_cast<numext::uint16_t>(bfloat16(-0.33333f))), + bfloat16(-0.33333f)); + VERIFY_IS_EQUAL( + numext::bit_cast<bfloat16>(numext::bit_cast<numext::uint16_t>(bfloat16(0.0f))), + bfloat16(0.0f)); + // Conversion from float. - VERIFY_IS_EQUAL(bfloat16(1.0f).value, 0x3f80); - VERIFY_IS_EQUAL(bfloat16(0.5f).value, 0x3f00); - VERIFY_IS_EQUAL(bfloat16(0.33333f).value, 0x3eab); - VERIFY_IS_EQUAL(bfloat16(3.38e38f).value, 0x7f7e); - VERIFY_IS_EQUAL(bfloat16(3.40e38f).value, 0x7f80); // Becomes infinity. + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(1.0f), 0x3f80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f), 0x3f00); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.33333f), 0x3eab); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3.38e38f), 0x7f7e); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3.40e38f), 0x7f80); // Becomes infinity. // Verify round-to-nearest-even behavior. float val1 = static_cast<float>(bfloat16(__bfloat16_raw(0x3c00))); float val2 = static_cast<float>(bfloat16(__bfloat16_raw(0x3c01))); float val3 = static_cast<float>(bfloat16(__bfloat16_raw(0x3c02))); - VERIFY_IS_EQUAL(bfloat16(0.5f * (val1 + val2)).value, 0x3c00); - VERIFY_IS_EQUAL(bfloat16(0.5f * (val2 + val3)).value, 0x3c02); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f * (val1 + val2)), 0x3c00); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f * (val2 + val3)), 0x3c02); // Conversion from int. - VERIFY_IS_EQUAL(bfloat16(-1).value, 0xbf80); - VERIFY_IS_EQUAL(bfloat16(0).value, 0x0000); - VERIFY_IS_EQUAL(bfloat16(1).value, 0x3f80); - VERIFY_IS_EQUAL(bfloat16(2).value, 0x4000); - VERIFY_IS_EQUAL(bfloat16(3).value, 0x4040); - VERIFY_IS_EQUAL(bfloat16(12).value, 0x4140); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(-1), 0xbf80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(1), 0x3f80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(2), 0x4000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3), 0x4040); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(12), 0x4140); // Conversion from bool. - VERIFY_IS_EQUAL(bfloat16(false).value, 0x0000); - VERIFY_IS_EQUAL(bfloat16(true).value, 0x3f80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(false), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(true), 0x3f80); // Conversion to bool VERIFY_IS_EQUAL(static_cast<bool>(bfloat16(3)), true); @@ -102,8 +119,8 @@ void test_conversion() VERIFY_IS_EQUAL(bfloat16(0.0f), bfloat16(0.0f)); VERIFY_IS_EQUAL(bfloat16(-0.0f), bfloat16(0.0f)); VERIFY_IS_EQUAL(bfloat16(-0.0f), bfloat16(-0.0f)); - VERIFY_IS_EQUAL(bfloat16(0.0f).value, 0x0000); - VERIFY_IS_EQUAL(bfloat16(-0.0f).value, 0x8000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.0f), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(-0.0f), 0x8000); // Flush denormals to zero for (float denorm = -std::numeric_limits<float>::denorm_min(); @@ -117,16 +134,16 @@ void test_conversion() VERIFY_IS_EQUAL(bfloat16(denorm), false); if (std::signbit(denorm)) { - VERIFY_IS_EQUAL(bf_trunc.value, 0x8000); + VERIFY_BFLOAT16_BITS_EQUAL(bf_trunc, 0x8000); } else { - VERIFY_IS_EQUAL(bf_trunc.value, 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bf_trunc, 0x0000); } bfloat16 bf_round = Eigen::bfloat16_impl::float_to_bfloat16_rtne<false>(denorm); VERIFY_IS_EQUAL(static_cast<float>(bf_round), 0.0f); if (std::signbit(denorm)) { - VERIFY_IS_EQUAL(bf_round.value, 0x8000); + VERIFY_BFLOAT16_BITS_EQUAL(bf_round, 0x8000); } else { - VERIFY_IS_EQUAL(bf_round.value, 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bf_round, 0x0000); } } @@ -231,33 +248,35 @@ void test_conversion() VERIFY((numext::isinf)(bfloat16(__bfloat16_raw(0x7f80)))); VERIFY((numext::isnan)(bfloat16(__bfloat16_raw(0x7fc0)))); - VERIFY_IS_EQUAL(bfloat16(BinaryToFloat(0x0, 0xff, 0x40, 0x0)).value, 0x7fc0); - VERIFY_IS_EQUAL(bfloat16(BinaryToFloat(0x1, 0xff, 0x40, 0x0)).value, 0xffc0); - VERIFY_IS_EQUAL(Eigen::bfloat16_impl::truncate_to_bfloat16( - BinaryToFloat(0x0, 0xff, 0x40, 0x0)) - .value, - 0x7fc0); - VERIFY_IS_EQUAL(Eigen::bfloat16_impl::truncate_to_bfloat16( - BinaryToFloat(0x1, 0xff, 0x40, 0x0)) - .value, - 0xffc0); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(BinaryToFloat(0x0, 0xff, 0x40, 0x0)), 0x7fc0); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(BinaryToFloat(0x1, 0xff, 0x40, 0x0)), 0xffc0); + VERIFY_BFLOAT16_BITS_EQUAL(Eigen::bfloat16_impl::truncate_to_bfloat16( + BinaryToFloat(0x0, 0xff, 0x40, 0x0)), + 0x7fc0); + VERIFY_BFLOAT16_BITS_EQUAL(Eigen::bfloat16_impl::truncate_to_bfloat16( + BinaryToFloat(0x1, 0xff, 0x40, 0x0)), + 0xffc0); } void test_numtraits() { - std::cout << "epsilon = " << NumTraits<bfloat16>::epsilon() << " (0x" << std::hex << NumTraits<bfloat16>::epsilon().value << ")" << std::endl; - std::cout << "highest = " << NumTraits<bfloat16>::highest() << " (0x" << std::hex << NumTraits<bfloat16>::highest().value << ")" << std::endl; - std::cout << "lowest = " << NumTraits<bfloat16>::lowest() << " (0x" << std::hex << NumTraits<bfloat16>::lowest().value << ")" << std::endl; - std::cout << "min = " << (std::numeric_limits<bfloat16>::min)() << " (0x" << std::hex << (std::numeric_limits<bfloat16>::min)().value << ")" << std::endl; - std::cout << "denorm min = " << (std::numeric_limits<bfloat16>::denorm_min)() << " (0x" << std::hex << (std::numeric_limits<bfloat16>::denorm_min)().value << ")" << std::endl; - std::cout << "infinity = " << NumTraits<bfloat16>::infinity() << " (0x" << std::hex << NumTraits<bfloat16>::infinity().value << ")" << std::endl; - std::cout << "quiet nan = " << NumTraits<bfloat16>::quiet_NaN() << " (0x" << std::hex << NumTraits<bfloat16>::quiet_NaN().value << ")" << std::endl; - std::cout << "signaling nan = " << std::numeric_limits<bfloat16>::signaling_NaN() << " (0x" << std::hex << std::numeric_limits<bfloat16>::signaling_NaN().value << ")" << std::endl; + std::cout << "epsilon = " << NumTraits<bfloat16>::epsilon() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<bfloat16>::epsilon()) << ")" << std::endl; + std::cout << "highest = " << NumTraits<bfloat16>::highest() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<bfloat16>::highest()) << ")" << std::endl; + std::cout << "lowest = " << NumTraits<bfloat16>::lowest() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<bfloat16>::lowest()) << ")" << std::endl; + std::cout << "min = " << (std::numeric_limits<bfloat16>::min)() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>((std::numeric_limits<bfloat16>::min)()) << ")" << std::endl; + std::cout << "denorm min = " << (std::numeric_limits<bfloat16>::denorm_min)() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>((std::numeric_limits<bfloat16>::denorm_min)()) << ")" << std::endl; + std::cout << "infinity = " << NumTraits<bfloat16>::infinity() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<bfloat16>::infinity()) << ")" << std::endl; + std::cout << "quiet nan = " << NumTraits<bfloat16>::quiet_NaN() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<bfloat16>::quiet_NaN()) << ")" << std::endl; + std::cout << "signaling nan = " << std::numeric_limits<bfloat16>::signaling_NaN() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(std::numeric_limits<bfloat16>::signaling_NaN()) << ")" << std::endl; VERIFY(NumTraits<bfloat16>::IsSigned); - VERIFY_IS_EQUAL( std::numeric_limits<bfloat16>::infinity().value, bfloat16(std::numeric_limits<float>::infinity()).value ); - VERIFY_IS_EQUAL( std::numeric_limits<bfloat16>::quiet_NaN().value, bfloat16(std::numeric_limits<float>::quiet_NaN()).value ); + VERIFY_IS_EQUAL( + numext::bit_cast<numext::uint16_t>(std::numeric_limits<bfloat16>::infinity()), + numext::bit_cast<numext::uint16_t>(bfloat16(std::numeric_limits<float>::infinity())) ); + VERIFY_IS_EQUAL( + numext::bit_cast<numext::uint16_t>(std::numeric_limits<bfloat16>::quiet_NaN()), + numext::bit_cast<numext::uint16_t>(bfloat16(std::numeric_limits<float>::quiet_NaN())) ); VERIFY( (std::numeric_limits<bfloat16>::min)() > bfloat16(0.f) ); VERIFY( (std::numeric_limits<bfloat16>::denorm_min)() > bfloat16(0.f) ); VERIFY_IS_EQUAL( (std::numeric_limits<bfloat16>::denorm_min)()/bfloat16(2), bfloat16(0.f) ); diff --git a/test/half_float.cpp b/test/half_float.cpp index b301b371d..cf6df547a 100644 --- a/test/half_float.cpp +++ b/test/half_float.cpp @@ -11,6 +11,9 @@ #include <Eigen/src/Core/arch/Default/Half.h> +#define VERIFY_HALF_BITS_EQUAL(h, bits) \ + VERIFY_IS_EQUAL((numext::bit_cast<numext::uint16_t>(h)), (static_cast<numext::uint16_t>(bits))) + // Make sure it's possible to forward declare Eigen::half namespace Eigen { struct half; @@ -22,75 +25,51 @@ void test_conversion() { using Eigen::half_impl::__half_raw; - // We don't use a uint16_t raw member x if the platform has native Arm __fp16 - // support -#if !defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC) - // Conversion from float. - VERIFY_IS_EQUAL(half(1.0f).x, 0x3c00); - VERIFY_IS_EQUAL(half(0.5f).x, 0x3800); - VERIFY_IS_EQUAL(half(0.33333f).x, 0x3555); - VERIFY_IS_EQUAL(half(0.0f).x, 0x0000); - VERIFY_IS_EQUAL(half(-0.0f).x, 0x8000); - VERIFY_IS_EQUAL(half(65504.0f).x, 0x7bff); - VERIFY_IS_EQUAL(half(65536.0f).x, 0x7c00); // Becomes infinity. - - // Denormals. - VERIFY_IS_EQUAL(half(-5.96046e-08f).x, 0x8001); - VERIFY_IS_EQUAL(half(5.96046e-08f).x, 0x0001); - VERIFY_IS_EQUAL(half(1.19209e-07f).x, 0x0002); - - // Verify round-to-nearest-even behavior. - float val1 = float(half(__half_raw(0x3c00))); - float val2 = float(half(__half_raw(0x3c01))); - float val3 = float(half(__half_raw(0x3c02))); - VERIFY_IS_EQUAL(half(0.5f * (val1 + val2)).x, 0x3c00); - VERIFY_IS_EQUAL(half(0.5f * (val2 + val3)).x, 0x3c02); + // Round-trip bit-cast with uint16. + VERIFY_IS_EQUAL( + numext::bit_cast<half>(numext::bit_cast<numext::uint16_t>(half(1.0f))), + half(1.0f)); + VERIFY_IS_EQUAL( + numext::bit_cast<half>(numext::bit_cast<numext::uint16_t>(half(0.5f))), + half(0.5f)); + VERIFY_IS_EQUAL( + numext::bit_cast<half>(numext::bit_cast<numext::uint16_t>(half(-0.33333f))), + half(-0.33333f)); + VERIFY_IS_EQUAL( + numext::bit_cast<half>(numext::bit_cast<numext::uint16_t>(half(0.0f))), + half(0.0f)); - // Conversion from int. - VERIFY_IS_EQUAL(half(-1).x, 0xbc00); - VERIFY_IS_EQUAL(half(0).x, 0x0000); - VERIFY_IS_EQUAL(half(1).x, 0x3c00); - VERIFY_IS_EQUAL(half(2).x, 0x4000); - VERIFY_IS_EQUAL(half(3).x, 0x4200); - - // Conversion from bool. - VERIFY_IS_EQUAL(half(false).x, 0x0000); - VERIFY_IS_EQUAL(half(true).x, 0x3c00); -#endif - -#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC) - // Conversion from float. - VERIFY_IS_EQUAL(half(1.0f).x, __fp16(1.0f)); - VERIFY_IS_EQUAL(half(0.5f).x, __fp16(0.5f)); - VERIFY_IS_EQUAL(half(0.33333f).x, __fp16(0.33333f)); - VERIFY_IS_EQUAL(half(0.0f).x, __fp16(0.0f)); - VERIFY_IS_EQUAL(half(-0.0f).x, __fp16(-0.0f)); - VERIFY_IS_EQUAL(half(65504.0f).x, __fp16(65504.0f)); - VERIFY_IS_EQUAL(half(65536.0f).x, __fp16(65536.0f)); // Becomes infinity. + // Conversion from float. + VERIFY_HALF_BITS_EQUAL(half(1.0f), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(0.5f), 0x3800); + VERIFY_HALF_BITS_EQUAL(half(0.33333f), 0x3555); + VERIFY_HALF_BITS_EQUAL(half(0.0f), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(-0.0f), 0x8000); + VERIFY_HALF_BITS_EQUAL(half(65504.0f), 0x7bff); + VERIFY_HALF_BITS_EQUAL(half(65536.0f), 0x7c00); // Becomes infinity. // Denormals. - VERIFY_IS_EQUAL(half(-5.96046e-08f).x, __fp16(-5.96046e-08f)); - VERIFY_IS_EQUAL(half(5.96046e-08f).x, __fp16(5.96046e-08f)); - VERIFY_IS_EQUAL(half(1.19209e-07f).x, __fp16(1.19209e-07f)); + VERIFY_HALF_BITS_EQUAL(half(-5.96046e-08f), 0x8001); + VERIFY_HALF_BITS_EQUAL(half(5.96046e-08f), 0x0001); + VERIFY_HALF_BITS_EQUAL(half(1.19209e-07f), 0x0002); // Verify round-to-nearest-even behavior. float val1 = float(half(__half_raw(0x3c00))); float val2 = float(half(__half_raw(0x3c01))); float val3 = float(half(__half_raw(0x3c02))); - VERIFY_IS_EQUAL(half(0.5f * (val1 + val2)).x, __fp16(0.5f * (val1 + val2))); - VERIFY_IS_EQUAL(half(0.5f * (val2 + val3)).x, __fp16(0.5f * (val2 + val3))); + VERIFY_HALF_BITS_EQUAL(half(0.5f * (val1 + val2)), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(0.5f * (val2 + val3)), 0x3c02); // Conversion from int. - VERIFY_IS_EQUAL(half(-1).x, __fp16(-1)); - VERIFY_IS_EQUAL(half(0).x, __fp16(0)); - VERIFY_IS_EQUAL(half(1).x, __fp16(1)); - VERIFY_IS_EQUAL(half(2).x, __fp16(2)); - VERIFY_IS_EQUAL(half(3).x, __fp16(3)); + VERIFY_HALF_BITS_EQUAL(half(-1), 0xbc00); + VERIFY_HALF_BITS_EQUAL(half(0), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(1), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(2), 0x4000); + VERIFY_HALF_BITS_EQUAL(half(3), 0x4200); // Conversion from bool. - VERIFY_IS_EQUAL(half(false).x, __fp16(false)); - VERIFY_IS_EQUAL(half(true).x, __fp16(true)); -#endif + VERIFY_HALF_BITS_EQUAL(half(false), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(true), 0x3c00); // Conversion to float. VERIFY_IS_EQUAL(float(half(__half_raw(0x0000))), 0.0f); @@ -143,24 +122,27 @@ void test_conversion() void test_numtraits() { - std::cout << "epsilon = " << NumTraits<half>::epsilon() << " (0x" << std::hex << NumTraits<half>::epsilon().x << ")" << std::endl; - std::cout << "highest = " << NumTraits<half>::highest() << " (0x" << std::hex << NumTraits<half>::highest().x << ")" << std::endl; - std::cout << "lowest = " << NumTraits<half>::lowest() << " (0x" << std::hex << NumTraits<half>::lowest().x << ")" << std::endl; - std::cout << "min = " << (std::numeric_limits<half>::min)() << " (0x" << std::hex << half((std::numeric_limits<half>::min)()).x << ")" << std::endl; - std::cout << "denorm min = " << (std::numeric_limits<half>::denorm_min)() << " (0x" << std::hex << half((std::numeric_limits<half>::denorm_min)()).x << ")" << std::endl; - std::cout << "infinity = " << NumTraits<half>::infinity() << " (0x" << std::hex << NumTraits<half>::infinity().x << ")" << std::endl; - std::cout << "quiet nan = " << NumTraits<half>::quiet_NaN() << " (0x" << std::hex << NumTraits<half>::quiet_NaN().x << ")" << std::endl; - std::cout << "signaling nan = " << std::numeric_limits<half>::signaling_NaN() << " (0x" << std::hex << std::numeric_limits<half>::signaling_NaN().x << ")" << std::endl; + std::cout << "epsilon = " << NumTraits<half>::epsilon() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<half>::epsilon()) << ")" << std::endl; + std::cout << "highest = " << NumTraits<half>::highest() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<half>::highest()) << ")" << std::endl; + std::cout << "lowest = " << NumTraits<half>::lowest() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<half>::lowest()) << ")" << std::endl; + std::cout << "min = " << (std::numeric_limits<half>::min)() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(half((std::numeric_limits<half>::min)())) << ")" << std::endl; + std::cout << "denorm min = " << (std::numeric_limits<half>::denorm_min)() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(half((std::numeric_limits<half>::denorm_min)())) << ")" << std::endl; + std::cout << "infinity = " << NumTraits<half>::infinity() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<half>::infinity()) << ")" << std::endl; + std::cout << "quiet nan = " << NumTraits<half>::quiet_NaN() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(NumTraits<half>::quiet_NaN()) << ")" << std::endl; + std::cout << "signaling nan = " << std::numeric_limits<half>::signaling_NaN() << " (0x" << std::hex << numext::bit_cast<numext::uint16_t>(std::numeric_limits<half>::signaling_NaN()) << ")" << std::endl; VERIFY(NumTraits<half>::IsSigned); - VERIFY_IS_EQUAL( std::numeric_limits<half>::infinity().x, half(std::numeric_limits<float>::infinity()).x ); + VERIFY_IS_EQUAL( + numext::bit_cast<numext::uint16_t>(std::numeric_limits<half>::infinity()), + numext::bit_cast<numext::uint16_t>(half(std::numeric_limits<float>::infinity())) ); + VERIFY_IS_EQUAL( + numext::bit_cast<numext::uint16_t>(std::numeric_limits<half>::quiet_NaN()), + numext::bit_cast<numext::uint16_t>(half(std::numeric_limits<float>::quiet_NaN())) ); + VERIFY_IS_EQUAL( + numext::bit_cast<numext::uint16_t>(std::numeric_limits<half>::signaling_NaN()), + numext::bit_cast<numext::uint16_t>(half(std::numeric_limits<float>::signaling_NaN())) ); -// If we have a native fp16 types this becomes a nan == nan comparision so we have to disable it -#if !defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC) - VERIFY_IS_EQUAL( std::numeric_limits<half>::quiet_NaN().x, half(std::numeric_limits<float>::quiet_NaN()).x ); - VERIFY_IS_EQUAL( std::numeric_limits<half>::signaling_NaN().x, half(std::numeric_limits<float>::signaling_NaN()).x ); -#endif VERIFY( (std::numeric_limits<half>::min)() > half(0.f) ); VERIFY( (std::numeric_limits<half>::denorm_min)() > half(0.f) ); VERIFY( (std::numeric_limits<half>::min)()/half(2) > half(0.f) ); diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorRandom.h b/unsupported/Eigen/CXX11/src/Tensor/TensorRandom.h index ea286fee1..13450e1a7 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorRandom.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorRandom.h @@ -91,24 +91,21 @@ T RandomToTypeUniform(uint64_t* state, uint64_t stream) { template <> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half RandomToTypeUniform<Eigen::half>(uint64_t* state, uint64_t stream) { - Eigen::half result; - // Generate 10 random bits for the mantissa + // Generate 10 random bits for the mantissa, merge with exponent. unsigned rnd = PCG_XSH_RS_generator(state, stream); - result.x = static_cast<uint16_t>(rnd & 0x3ffu); - // Set the exponent - result.x |= (static_cast<uint16_t>(15) << 10); + const uint16_t half_bits = static_cast<uint16_t>(rnd & 0x3ffu) | (static_cast<uint16_t>(15) << 10); + Eigen::half result = Eigen::numext::bit_cast<Eigen::half>(half_bits); // Return the final result return result - Eigen::half(1.0f); } template <> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::bfloat16 RandomToTypeUniform<Eigen::bfloat16>(uint64_t* state, uint64_t stream) { - Eigen::bfloat16 result; - // Generate 7 random bits for the mantissa + + // Generate 7 random bits for the mantissa, merge with exponent. unsigned rnd = PCG_XSH_RS_generator(state, stream); - result.value = static_cast<uint16_t>(rnd & 0x7fu); - // Set the exponent - result.value |= (static_cast<uint16_t>(127) << 7); + const uint16_t half_bits = static_cast<uint16_t>(rnd & 0x7fu) | (static_cast<uint16_t>(127) << 7); + Eigen::bfloat16 result = Eigen::numext::bit_cast<Eigen::bfloat16>(half_bits); // Return the final result return result - Eigen::bfloat16(1.0f); } @@ -169,19 +166,19 @@ template <typename T> class UniformRandomGenerator { uint64_t seed = 0) { m_state = PCG_XSH_RS_state(seed); #ifdef EIGEN_USE_SYCL - // In SYCL it is not possible to build PCG_XSH_RS_state in one step. + // In SYCL it is not possible to build PCG_XSH_RS_state in one step. // Therefor, we need two step to initializate the m_state. // IN SYCL, the constructor of the functor is s called on the CPU - // and we get the clock seed here from the CPU. However, This seed is + // and we get the clock seed here from the CPU. However, This seed is //the same for all the thread. As unlike CUDA, the thread.ID, BlockID, etc is not a global function. // and only available on the Operator() function (which is called on the GPU). - // Thus for CUDA (((CLOCK + global_thread_id)* 6364136223846793005ULL) + 0xda3e39cb94b95bdbULL) is passed to each thread - // but for SYCL ((CLOCK * 6364136223846793005ULL) + 0xda3e39cb94b95bdbULL) is passed to each thread and each thread adds - // the (global_thread_id* 6364136223846793005ULL) for itself only once, in order to complete the construction - // similar to CUDA Therefore, the thread Id injection is not available at this stage. - //However when the operator() is called the thread ID will be avilable. So inside the opeator, - // we add the thrreadID, BlockId,... (which is equivalent of i) - //to the seed and construct the unique m_state per thead similar to cuda. + // Thus for CUDA (((CLOCK + global_thread_id)* 6364136223846793005ULL) + 0xda3e39cb94b95bdbULL) is passed to each thread + // but for SYCL ((CLOCK * 6364136223846793005ULL) + 0xda3e39cb94b95bdbULL) is passed to each thread and each thread adds + // the (global_thread_id* 6364136223846793005ULL) for itself only once, in order to complete the construction + // similar to CUDA Therefore, the thread Id injection is not available at this stage. + //However when the operator() is called the thread ID will be avilable. So inside the opeator, + // we add the thrreadID, BlockId,... (which is equivalent of i) + //to the seed and construct the unique m_state per thead similar to cuda. m_exec_once =false; #endif } @@ -282,16 +279,16 @@ template <typename T> class NormalRandomGenerator { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE NormalRandomGenerator(uint64_t seed = 0) { m_state = PCG_XSH_RS_state(seed); #ifdef EIGEN_USE_SYCL - // In SYCL it is not possible to build PCG_XSH_RS_state in one step. + // In SYCL it is not possible to build PCG_XSH_RS_state in one step. // Therefor, we need two steps to initializate the m_state. // IN SYCL, the constructor of the functor is s called on the CPU - // and we get the clock seed here from the CPU. However, This seed is + // and we get the clock seed here from the CPU. However, This seed is //the same for all the thread. As unlike CUDA, the thread.ID, BlockID, etc is not a global function. // and only available on the Operator() function (which is called on the GPU). - // Therefore, the thread Id injection is not available at this stage. However when the operator() - //is called the thread ID will be avilable. So inside the opeator, - // we add the thrreadID, BlockId,... (which is equivalent of i) - //to the seed and construct the unique m_state per thead similar to cuda. + // Therefore, the thread Id injection is not available at this stage. However when the operator() + //is called the thread ID will be avilable. So inside the opeator, + // we add the thrreadID, BlockId,... (which is equivalent of i) + //to the seed and construct the unique m_state per thead similar to cuda. m_exec_once =false; #endif } diff --git a/unsupported/test/cxx11_tensor_random.cpp b/unsupported/test/cxx11_tensor_random.cpp index 4740d5811..b9d4c5584 100644 --- a/unsupported/test/cxx11_tensor_random.cpp +++ b/unsupported/test/cxx11_tensor_random.cpp @@ -11,9 +11,10 @@ #include <Eigen/CXX11/Tensor> +template<typename Scalar> static void test_default() { - Tensor<float, 1> vec(6); + Tensor<Scalar, 1> vec(6); vec.setRandom(); // Fixme: we should check that the generated numbers follow a uniform @@ -23,10 +24,11 @@ static void test_default() } } +template<typename Scalar> static void test_normal() { - Tensor<float, 1> vec(6); - vec.setRandom<Eigen::internal::NormalRandomGenerator<float>>(); + Tensor<Scalar, 1> vec(6); + vec.template setRandom<Eigen::internal::NormalRandomGenerator<Scalar>>(); // Fixme: we should check that the generated numbers follow a gaussian // distribution instead. @@ -72,7 +74,13 @@ static void test_custom() EIGEN_DECLARE_TEST(cxx11_tensor_random) { - CALL_SUBTEST(test_default()); - CALL_SUBTEST(test_normal()); + CALL_SUBTEST((test_default<float>())); + CALL_SUBTEST((test_normal<float>())); + CALL_SUBTEST((test_default<double>())); + CALL_SUBTEST((test_normal<double>())); + CALL_SUBTEST((test_default<Eigen::half>())); + CALL_SUBTEST((test_normal<Eigen::half>())); + CALL_SUBTEST((test_default<Eigen::bfloat16>())); + CALL_SUBTEST((test_normal<Eigen::bfloat16>())); CALL_SUBTEST(test_custom()); } |