aboutsummaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
Diffstat
-rw-r--r--Eigen/Core10
-rw-r--r--Eigen/src/Core/arch/CUDA/Half.h12
-rw-r--r--Eigen/src/Core/arch/CUDA/PacketMathHalf.h2
-rw-r--r--Eigen/src/Core/util/Macros.h7
-rw-r--r--test/cuda_basic.cu2
-rw-r--r--unsupported/test/cxx11_tensor_argmax_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_cast_float16_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_complex_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_complex_cwise_ops_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_contract_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_device.cu2
-rw-r--r--unsupported/test/cxx11_tensor_of_float16_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_random_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_reduction_cuda.cu2
-rw-r--r--unsupported/test/cxx11_tensor_scan_cuda.cu2
16 files changed, 32 insertions, 23 deletions
diff --git a/Eigen/Core b/Eigen/Core
index f613ee9aa..f6fe4b0ec 100644
--- a/Eigen/Core
+++ b/Eigen/Core
@@ -22,6 +22,14 @@
#define EIGEN_CUDA_ARCH __CUDA_ARCH__
#endif
+#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
+#define EIGEN_CUDACC_VER ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
+#elif defined(__CUDACC_VER__)
+#define EIGEN_CUDACC_VER __CUDACC_VER__
+#else
+#define EIGEN_CUDACC_VER 0
+#endif
+
// Handle NVCC/CUDA/SYCL
#if defined(EIGEN_CUDACC) || defined(__SYCL_DEVICE_ONLY__)
// Do not try asserts on CUDA and SYCL!
@@ -248,7 +256,7 @@
#if defined EIGEN_CUDACC
#define EIGEN_VECTORIZE_CUDA
#include <vector_types.h>
- #if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+ #if EIGEN_CUDACC_VER >= 70500
#define EIGEN_HAS_CUDA_FP16
#endif
#endif
diff --git a/Eigen/src/Core/arch/CUDA/Half.h b/Eigen/src/Core/arch/CUDA/Half.h
index e99847e24..8cedd65ad 100644
--- a/Eigen/src/Core/arch/CUDA/Half.h
+++ b/Eigen/src/Core/arch/CUDA/Half.h
@@ -386,7 +386,7 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half abs(const half& a) {
return result;
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half exp(const half& a) {
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
return half(hexp(a));
#else
return half(::expf(float(a)));
@@ -396,7 +396,7 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half expm1(const half& a) {
return half(numext::expm1(float(a)));
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log(const half& a) {
-#if defined(EIGEN_HAS_CUDA_FP16) && defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+#if defined(EIGEN_HAS_CUDA_FP16) && EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
return half(::hlog(a));
#else
return half(::logf(float(a)));
@@ -409,7 +409,7 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log10(const half& a) {
return half(::log10f(float(a)));
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sqrt(const half& a) {
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
return half(hsqrt(a));
#else
return half(::sqrtf(float(a)));
@@ -431,14 +431,14 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tanh(const half& a) {
return half(::tanhf(float(a)));
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half floor(const half& a) {
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
return half(hfloor(a));
#else
return half(::floorf(float(a)));
#endif
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half ceil(const half& a) {
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
return half(hceil(a));
#else
return half(::ceilf(float(a)));
@@ -576,7 +576,7 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half exph(const Eigen::half& a) {
return Eigen::half(::expf(float(a)));
}
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half logh(const Eigen::half& a) {
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+#if EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
return Eigen::half(::hlog(a));
#else
return Eigen::half(::logf(float(a)));
diff --git a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
index 4a730a0e0..ba6a7f920 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -284,7 +284,7 @@ template<> __device__ EIGEN_STRONG_INLINE half2 pexpm1<half2>(const half2& a) {
return __floats2half2_rn(r1, r2);
}
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
template<> __device__ EIGEN_STRONG_INLINE
half2 plog<half2>(const half2& a) {
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 322e8d899..b63ea2697 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -413,7 +413,7 @@
// Does the compiler support variadic templates?
#ifndef EIGEN_HAS_VARIADIC_TEMPLATES
#if EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) \
- && (!defined(__NVCC__) || !EIGEN_ARCH_ARM_OR_ARM64 || (defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000) )
+ && (!defined(__NVCC__) || !EIGEN_ARCH_ARM_OR_ARM64 || (EIGEN_CUDACC_VER >= 80000) )
// ^^ Disable the use of variadic templates when compiling with versions of nvcc older than 8.0 on ARM devices:
// this prevents nvcc from crashing when compiling Eigen on Tegra X1
#define EIGEN_HAS_VARIADIC_TEMPLATES 1
@@ -429,7 +429,7 @@
#if defined(EIGEN_CUDACC)
// Const expressions are supported provided that c++11 is enabled and we're using either clang or nvcc 7.5 or above
-#if EIGEN_MAX_CPP_VER>=14 && (__cplusplus > 199711L && defined(__CUDACC_VER__) && (EIGEN_COMP_CLANG || __CUDACC_VER__ >= 70500))
+#if EIGEN_MAX_CPP_VER>=14 && (__cplusplus > 199711L && (EIGEN_COMP_CLANG || EIGEN_CUDACC_VER >= 70500))
#define EIGEN_HAS_CONSTEXPR 1
#endif
#elif EIGEN_MAX_CPP_VER>=14 && (__has_feature(cxx_relaxed_constexpr) || (defined(__cplusplus) && __cplusplus >= 201402L) || \
@@ -837,7 +837,8 @@ namespace Eigen {
// just an empty macro !
#define EIGEN_EMPTY
-#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || defined(__CUDACC_VER__)) // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
+#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || EIGEN_CUDACC_VER>0)
+ // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
using Base::operator =;
#elif EIGEN_COMP_CLANG // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
diff --git a/test/cuda_basic.cu b/test/cuda_basic.cu
index cb2e4167a..0ff13477d 100644
--- a/test/cuda_basic.cu
+++ b/test/cuda_basic.cu
@@ -20,7 +20,7 @@
#include <math_constants.h>
#include <cuda.h>
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_argmax_cuda.cu b/unsupported/test/cxx11_tensor_argmax_cuda.cu
index 653443dc5..0dfd6cfe1 100644
--- a/unsupported/test/cxx11_tensor_argmax_cuda.cu
+++ b/unsupported/test/cxx11_tensor_argmax_cuda.cu
@@ -12,7 +12,7 @@
#define EIGEN_TEST_FUNC cxx11_tensor_cuda
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu
index 88c233994..83a740e7a 100644
--- a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu
+++ b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu
@@ -13,7 +13,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_complex_cuda.cu b/unsupported/test/cxx11_tensor_complex_cuda.cu
index 87cf28920..cbff5a9b2 100644
--- a/unsupported/test/cxx11_tensor_complex_cuda.cu
+++ b/unsupported/test/cxx11_tensor_complex_cuda.cu
@@ -11,7 +11,7 @@
#define EIGEN_TEST_FUNC cxx11_tensor_complex
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_complex_cwise_ops_cuda.cu b/unsupported/test/cxx11_tensor_complex_cwise_ops_cuda.cu
index 2baf5eaad..9133fce5a 100644
--- a/unsupported/test/cxx11_tensor_complex_cwise_ops_cuda.cu
+++ b/unsupported/test/cxx11_tensor_complex_cwise_ops_cuda.cu
@@ -11,7 +11,7 @@
#define EIGEN_TEST_FUNC cxx11_tensor_complex_cwise_ops
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_contract_cuda.cu b/unsupported/test/cxx11_tensor_contract_cuda.cu
index dd68430ce..0b2f3f0f4 100644
--- a/unsupported/test/cxx11_tensor_contract_cuda.cu
+++ b/unsupported/test/cxx11_tensor_contract_cuda.cu
@@ -14,7 +14,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_cuda.cu b/unsupported/test/cxx11_tensor_cuda.cu
index 0ba9d52e9..ad8c9662f 100644
--- a/unsupported/test/cxx11_tensor_cuda.cu
+++ b/unsupported/test/cxx11_tensor_cuda.cu
@@ -12,7 +12,7 @@
#define EIGEN_TEST_FUNC cxx11_tensor_cuda
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_device.cu b/unsupported/test/cxx11_tensor_device.cu
index fde20ddf2..ae21f492a 100644
--- a/unsupported/test/cxx11_tensor_device.cu
+++ b/unsupported/test/cxx11_tensor_device.cu
@@ -13,7 +13,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_of_float16_cuda.cu b/unsupported/test/cxx11_tensor_of_float16_cuda.cu
index 908a5e5a9..0ba7657b8 100644
--- a/unsupported/test/cxx11_tensor_of_float16_cuda.cu
+++ b/unsupported/test/cxx11_tensor_of_float16_cuda.cu
@@ -13,7 +13,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_random_cuda.cu b/unsupported/test/cxx11_tensor_random_cuda.cu
index b3be199e1..94d5f4e5a 100644
--- a/unsupported/test/cxx11_tensor_random_cuda.cu
+++ b/unsupported/test/cxx11_tensor_random_cuda.cu
@@ -13,7 +13,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_reduction_cuda.cu b/unsupported/test/cxx11_tensor_reduction_cuda.cu
index 6858b43a7..fd09d013b 100644
--- a/unsupported/test/cxx11_tensor_reduction_cuda.cu
+++ b/unsupported/test/cxx11_tensor_reduction_cuda.cu
@@ -12,7 +12,7 @@
#define EIGEN_TEST_FUNC cxx11_tensor_reduction_cuda
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if dEIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
diff --git a/unsupported/test/cxx11_tensor_scan_cuda.cu b/unsupported/test/cxx11_tensor_scan_cuda.cu
index 5f146f3c9..46571cfea 100644
--- a/unsupported/test/cxx11_tensor_scan_cuda.cu
+++ b/unsupported/test/cxx11_tensor_scan_cuda.cu
@@ -13,7 +13,7 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
-#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
+#if EIGEN_CUDACC_VER >= 70500
#include <cuda_fp16.h>
#endif
#include "main.h"
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-28 20:30:23 +0000