<div dir="ltr">This broke bootstrap builds, I reverted it for now in r283747.</div><div class="gmail_extra"><br><div class="gmail_quote">On Sat, Oct 8, 2016 at 6:16 PM, Justin Lebar via cfe-commits <span dir="ltr"><<a href="mailto:cfe-commits@lists.llvm.org" target="_blank">cfe-commits@lists.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Author: jlebar<br>
Date: Sat Oct 8 17:16:12 2016<br>
New Revision: 283680<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=283680&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project?rev=283680&view=rev</a><br>
Log:<br>
[CUDA] Support <complex> and std::min/max on the device.<br>
<br>
Summary:<br>
We do this by wrapping <complex> and <algorithm>.<br>
<br>
Tests are in the test-suite.<br>
<br>
Reviewers: tra<br>
<br>
Subscribers: jhen, beanz, cfe-commits, mgorny<br>
<br>
Differential Revision: <a href="https://reviews.llvm.org/D24979" rel="noreferrer" target="_blank">https://reviews.llvm.org/<wbr>D24979</a><br>
<br>
Added:<br>
cfe/trunk/lib/Headers/__clang_<wbr>cuda_complex_builtins.h<br>
cfe/trunk/lib/Headers/cuda_<wbr>wrappers/<br>
cfe/trunk/lib/Headers/cuda_<wbr>wrappers/algorithm<br>
cfe/trunk/lib/Headers/cuda_<wbr>wrappers/complex<br>
Modified:<br>
cfe/trunk/lib/Driver/<wbr>ToolChains.cpp<br>
cfe/trunk/lib/Headers/<wbr>CMakeLists.txt<br>
cfe/trunk/lib/Headers/__clang_<wbr>cuda_runtime_wrapper.h<br>
<br>
Modified: cfe/trunk/lib/Driver/<wbr>ToolChains.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Driver/ToolChains.cpp?rev=283680&r1=283679&r2=283680&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Driver/<wbr>ToolChains.cpp?rev=283680&r1=<wbr>283679&r2=283680&view=diff</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Driver/<wbr>ToolChains.cpp (original)<br>
+++ cfe/trunk/lib/Driver/<wbr>ToolChains.cpp Sat Oct 8 17:16:12 2016<br>
@@ -4694,6 +4694,15 @@ void Linux::<wbr>AddClangCXXStdlibIncludeArgs<br>
<br>
void Linux::AddCudaIncludeArgs(<wbr>const ArgList &DriverArgs,<br>
ArgStringList &CC1Args) const {<br>
+ if (!DriverArgs.hasArg(options::<wbr>OPT_nobuiltininc)) {<br>
+ // Add cuda_wrappers/* to our system include path. This lets us wrap<br>
+ // standard library headers.<br>
+ SmallString<128> P(getDriver().ResourceDir);<br>
+ llvm::sys::path::append(P, "include");<br>
+ llvm::sys::path::append(P, "cuda_wrappers");<br>
+ addSystemInclude(DriverArgs, CC1Args, P);<br>
+ }<br>
+<br>
if (DriverArgs.hasArg(options::<wbr>OPT_nocudainc))<br>
return;<br>
<br>
<br>
Modified: cfe/trunk/lib/Headers/<wbr>CMakeLists.txt<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/CMakeLists.txt?rev=283680&r1=283679&r2=283680&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Headers/<wbr>CMakeLists.txt?rev=283680&r1=<wbr>283679&r2=283680&view=diff</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Headers/<wbr>CMakeLists.txt (original)<br>
+++ cfe/trunk/lib/Headers/<wbr>CMakeLists.txt Sat Oct 8 17:16:12 2016<br>
@@ -24,10 +24,13 @@ set(files<br>
bmiintrin.h<br>
__clang_cuda_builtin_vars.h<br>
__clang_cuda_cmath.h<br>
+ __clang_cuda_complex_builtins.<wbr>h<br>
__clang_cuda_intrinsics.h<br>
__clang_cuda_math_forward_<wbr>declares.h<br>
__clang_cuda_runtime_wrapper.h<br>
cpuid.h<br>
+ cuda_wrappers/algorithm<br>
+ cuda_wrappers/complex<br>
clflushoptintrin.h<br>
emmintrin.h<br>
f16cintrin.h<br>
<br>
Added: cfe/trunk/lib/Headers/__clang_<wbr>cuda_complex_builtins.h<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/__clang_cuda_complex_builtins.h?rev=283680&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Headers/<wbr>__clang_cuda_complex_builtins.<wbr>h?rev=283680&view=auto</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Headers/__clang_<wbr>cuda_complex_builtins.h (added)<br>
+++ cfe/trunk/lib/Headers/__clang_<wbr>cuda_complex_builtins.h Sat Oct 8 17:16:12 2016<br>
@@ -0,0 +1,203 @@<br>
+/*===-- __clang_cuda_complex_builtins - CUDA impls of runtime complex fns ---===<br>
+ *<br>
+ * Permission is hereby granted, free of charge, to any person obtaining a copy<br>
+ * of this software and associated documentation files (the "Software"), to deal<br>
+ * in the Software without restriction, including without limitation the rights<br>
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell<br>
+ * copies of the Software, and to permit persons to whom the Software is<br>
+ * furnished to do so, subject to the following conditions:<br>
+ *<br>
+ * The above copyright notice and this permission notice shall be included in<br>
+ * all copies or substantial portions of the Software.<br>
+ *<br>
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR<br>
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,<br>
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE<br>
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER<br>
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,<br>
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN<br>
+ * THE SOFTWARE.<br>
+ *<br>
+ *===--------------------------<wbr>------------------------------<wbr>---------------===<br>
+ */<br>
+<br>
+#ifndef __CLANG_CUDA_COMPLEX_BUILTINS<br>
+#define __CLANG_CUDA_COMPLEX_BUILTINS<br>
+<br>
+// This header defines __muldc3, __mulsc3, __divdc3, and __divsc3. These are<br>
+// libgcc functions that clang assumes are available when compiling c99 complex<br>
+// operations. (These implementations come from libc++, and have been modified<br>
+// to work with CUDA.)<br>
+<br>
+extern "C" inline __device__ double _Complex __muldc3(double __a, double __b,<br>
+ double __c, double __d) {<br>
+ double __ac = __a * __c;<br>
+ double __bd = __b * __d;<br>
+ double __ad = __a * __d;<br>
+ double __bc = __b * __c;<br>
+ double _Complex z;<br>
+ __real__(z) = __ac - __bd;<br>
+ __imag__(z) = __ad + __bc;<br>
+ if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {<br>
+ int __recalc = 0;<br>
+ if (std::isinf(__a) || std::isinf(__b)) {<br>
+ __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);<br>
+ __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);<br>
+ if (std::isnan(__c))<br>
+ __c = std::copysign(0, __c);<br>
+ if (std::isnan(__d))<br>
+ __d = std::copysign(0, __d);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (std::isinf(__c) || std::isinf(__d)) {<br>
+ __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);<br>
+ __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);<br>
+ if (std::isnan(__a))<br>
+ __a = std::copysign(0, __a);<br>
+ if (std::isnan(__b))<br>
+ __b = std::copysign(0, __b);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (!__recalc && (std::isinf(__ac) || std::isinf(__bd) ||<br>
+ std::isinf(__ad) || std::isinf(__bc))) {<br>
+ if (std::isnan(__a))<br>
+ __a = std::copysign(0, __a);<br>
+ if (std::isnan(__b))<br>
+ __b = std::copysign(0, __b);<br>
+ if (std::isnan(__c))<br>
+ __c = std::copysign(0, __c);<br>
+ if (std::isnan(__d))<br>
+ __d = std::copysign(0, __d);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (__recalc) {<br>
+ // Can't use std::numeric_limits<double>::<wbr>infinity() -- that doesn't have<br>
+ // a device overload (and isn't constexpr before C++11, naturally).<br>
+ __real__(z) = __builtin_huge_valf() * (__a * __c - __b * __d);<br>
+ __imag__(z) = __builtin_huge_valf() * (__a * __d + __b * __c);<br>
+ }<br>
+ }<br>
+ return z;<br>
+}<br>
+<br>
+extern "C" inline __device__ float _Complex __mulsc3(float __a, float __b,<br>
+ float __c, float __d) {<br>
+ float __ac = __a * __c;<br>
+ float __bd = __b * __d;<br>
+ float __ad = __a * __d;<br>
+ float __bc = __b * __c;<br>
+ float _Complex z;<br>
+ __real__(z) = __ac - __bd;<br>
+ __imag__(z) = __ad + __bc;<br>
+ if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {<br>
+ int __recalc = 0;<br>
+ if (std::isinf(__a) || std::isinf(__b)) {<br>
+ __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);<br>
+ __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);<br>
+ if (std::isnan(__c))<br>
+ __c = std::copysign(0, __c);<br>
+ if (std::isnan(__d))<br>
+ __d = std::copysign(0, __d);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (std::isinf(__c) || std::isinf(__d)) {<br>
+ __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);<br>
+ __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);<br>
+ if (std::isnan(__a))<br>
+ __a = std::copysign(0, __a);<br>
+ if (std::isnan(__b))<br>
+ __b = std::copysign(0, __b);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (!__recalc && (std::isinf(__ac) || std::isinf(__bd) ||<br>
+ std::isinf(__ad) || std::isinf(__bc))) {<br>
+ if (std::isnan(__a))<br>
+ __a = std::copysign(0, __a);<br>
+ if (std::isnan(__b))<br>
+ __b = std::copysign(0, __b);<br>
+ if (std::isnan(__c))<br>
+ __c = std::copysign(0, __c);<br>
+ if (std::isnan(__d))<br>
+ __d = std::copysign(0, __d);<br>
+ __recalc = 1;<br>
+ }<br>
+ if (__recalc) {<br>
+ __real__(z) = __builtin_huge_valf() * (__a * __c - __b * __d);<br>
+ __imag__(z) = __builtin_huge_valf() * (__a * __d + __b * __c);<br>
+ }<br>
+ }<br>
+ return z;<br>
+}<br>
+<br>
+extern "C" inline __device__ double _Complex __divdc3(double __a, double __b,<br>
+ double __c, double __d) {<br>
+ int __ilogbw = 0;<br>
+ // Can't use std::max, because that's defined in <algorithm>, and we don't<br>
+ // want to pull that in for every compile. The CUDA headers define<br>
+ // ::max(float, float) and ::max(double, double), which is sufficient for us.<br>
+ double __logbw = std::logb(max(std::abs(__c), std::abs(__d)));<br>
+ if (std::isfinite(__logbw)) {<br>
+ __ilogbw = (int)__logbw;<br>
+ __c = std::scalbn(__c, -__ilogbw);<br>
+ __d = std::scalbn(__d, -__ilogbw);<br>
+ }<br>
+ double __denom = __c * __c + __d * __d;<br>
+ double _Complex z;<br>
+ __real__(z) = std::scalbn((__a * __c + __b * __d) / __denom, -__ilogbw);<br>
+ __imag__(z) = std::scalbn((__b * __c - __a * __d) / __denom, -__ilogbw);<br>
+ if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {<br>
+ if ((__denom == 0.0) && (!std::isnan(__a) || !std::isnan(__b))) {<br>
+ __real__(z) = std::copysign(__builtin_huge_<wbr>valf(), __c) * __a;<br>
+ __imag__(z) = std::copysign(__builtin_huge_<wbr>valf(), __c) * __b;<br>
+ } else if ((std::isinf(__a) || std::isinf(__b)) && std::isfinite(__c) &&<br>
+ std::isfinite(__d)) {<br>
+ __a = std::copysign(std::isinf(__a) ? 1.0 : 0.0, __a);<br>
+ __b = std::copysign(std::isinf(__b) ? 1.0 : 0.0, __b);<br>
+ __real__(z) = __builtin_huge_valf() * (__a * __c + __b * __d);<br>
+ __imag__(z) = __builtin_huge_valf() * (__b * __c - __a * __d);<br>
+ } else if (std::isinf(__logbw) && __logbw > 0.0 && std::isfinite(__a) &&<br>
+ std::isfinite(__b)) {<br>
+ __c = std::copysign(std::isinf(__c) ? 1.0 : 0.0, __c);<br>
+ __d = std::copysign(std::isinf(__d) ? 1.0 : 0.0, __d);<br>
+ __real__(z) = 0.0 * (__a * __c + __b * __d);<br>
+ __imag__(z) = 0.0 * (__b * __c - __a * __d);<br>
+ }<br>
+ }<br>
+ return z;<br>
+}<br>
+<br>
+extern "C" inline __device__ float _Complex __divsc3(float __a, float __b,<br>
+ float __c, float __d) {<br>
+ int __ilogbw = 0;<br>
+ float __logbw = std::logb(max(std::abs(__c), std::abs(__d)));<br>
+ if (std::isfinite(__logbw)) {<br>
+ __ilogbw = (int)__logbw;<br>
+ __c = std::scalbn(__c, -__ilogbw);<br>
+ __d = std::scalbn(__d, -__ilogbw);<br>
+ }<br>
+ float __denom = __c * __c + __d * __d;<br>
+ float _Complex z;<br>
+ __real__(z) = std::scalbn((__a * __c + __b * __d) / __denom, -__ilogbw);<br>
+ __imag__(z) = std::scalbn((__b * __c - __a * __d) / __denom, -__ilogbw);<br>
+ if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {<br>
+ if ((__denom == 0) && (!std::isnan(__a) || !std::isnan(__b))) {<br>
+ __real__(z) = std::copysign(__builtin_huge_<wbr>valf(), __c) * __a;<br>
+ __imag__(z) = std::copysign(__builtin_huge_<wbr>valf(), __c) * __b;<br>
+ } else if ((std::isinf(__a) || std::isinf(__b)) && std::isfinite(__c) &&<br>
+ std::isfinite(__d)) {<br>
+ __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);<br>
+ __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);<br>
+ __real__(z) = __builtin_huge_valf() * (__a * __c + __b * __d);<br>
+ __imag__(z) = __builtin_huge_valf() * (__b * __c - __a * __d);<br>
+ } else if (std::isinf(__logbw) && __logbw > 0 && std::isfinite(__a) &&<br>
+ std::isfinite(__b)) {<br>
+ __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);<br>
+ __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);<br>
+ __real__(z) = 0 * (__a * __c + __b * __d);<br>
+ __imag__(z) = 0 * (__b * __c - __a * __d);<br>
+ }<br>
+ }<br>
+ return z;<br>
+}<br>
+<br>
+#endif // __CLANG_CUDA_COMPLEX_BUILTINS<br>
<br>
Modified: cfe/trunk/lib/Headers/__clang_<wbr>cuda_runtime_wrapper.h<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/__clang_cuda_runtime_wrapper.h?rev=283680&r1=283679&r2=283680&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Headers/<wbr>__clang_cuda_runtime_wrapper.<wbr>h?rev=283680&r1=283679&r2=<wbr>283680&view=diff</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Headers/__clang_<wbr>cuda_runtime_wrapper.h (original)<br>
+++ cfe/trunk/lib/Headers/__clang_<wbr>cuda_runtime_wrapper.h Sat Oct 8 17:16:12 2016<br>
@@ -312,6 +312,7 @@ __device__ inline __cuda_builtin_gridDim<br>
<br>
#include <__clang_cuda_cmath.h><br>
#include <__clang_cuda_intrinsics.h><br>
+#include <__clang_cuda_complex_<wbr>builtins.h><br>
<br>
// curand_mtgp32_kernel helpfully redeclares blockDim and threadIdx in host<br>
// mode, giving them their "proper" types of dim3 and uint3. This is<br>
<br>
Added: cfe/trunk/lib/Headers/cuda_<wbr>wrappers/algorithm<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/cuda_wrappers/algorithm?rev=283680&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Headers/<wbr>cuda_wrappers/algorithm?rev=<wbr>283680&view=auto</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Headers/cuda_<wbr>wrappers/algorithm (added)<br>
+++ cfe/trunk/lib/Headers/cuda_<wbr>wrappers/algorithm Sat Oct 8 17:16:12 2016<br>
@@ -0,0 +1,96 @@<br>
+/*===---- complex - CUDA wrapper for <algorithm> ----------------------------==<wbr>=<br>
+ *<br>
+ * Permission is hereby granted, free of charge, to any person obtaining a copy<br>
+ * of this software and associated documentation files (the "Software"), to deal<br>
+ * in the Software without restriction, including without limitation the rights<br>
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell<br>
+ * copies of the Software, and to permit persons to whom the Software is<br>
+ * furnished to do so, subject to the following conditions:<br>
+ *<br>
+ * The above copyright notice and this permission notice shall be included in<br>
+ * all copies or substantial portions of the Software.<br>
+ *<br>
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR<br>
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,<br>
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE<br>
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER<br>
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,<br>
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN<br>
+ * THE SOFTWARE.<br>
+ *<br>
+ *===--------------------------<wbr>------------------------------<wbr>---------------===<br>
+ */<br>
+<br>
+#ifndef __CLANG_CUDA_WRAPPERS_<wbr>ALGORITHM<br>
+#define __CLANG_CUDA_WRAPPERS_<wbr>ALGORITHM<br>
+<br>
+// This header defines __device__ overloads of std::min/max, but only if we're<br>
+// <= C++11. In C++14, these functions are constexpr, and so are implicitly<br>
+// __host__ __device__.<br>
+//<br>
+// We don't support the initializer_list overloads because<br>
+// initializer_list::begin() and end() are not __host__ __device__ functions.<br>
+//<br>
+// When compiling in C++14 mode, we could force std::min/max to have different<br>
+// implementations for host and device, by declaring the device overloads<br>
+// before the constexpr overloads appear. We choose not to do this because<br>
+<br>
+// a) why write our own implementation when we can use one from the standard<br>
+// library? and<br>
+// b) libstdc++ is evil and declares min/max inside a header that is included<br>
+// *before* we include <algorithm>. So we'd have to unconditionally<br>
+// declare our __device__ overloads of min/max, but that would pollute<br>
+// things for people who choose not to include <algorithm>.<br>
+<br>
+#include_next <algorithm><br>
+<br>
+#if __cplusplus <= 201103L<br>
+<br>
+// We need to define these overloads in exactly the namespace our standard<br>
+// library uses (including the right inline namespace), otherwise they won't be<br>
+// picked up by other functions in the standard library (e.g. functions in<br>
+// <complex>). Thus the ugliness below.<br>
+#ifdef _LIBCPP_BEGIN_NAMESPACE_STD<br>
+_LIBCPP_BEGIN_NAMESPACE_STD<br>
+#else<br>
+namespace std {<br>
+#ifdef _GLIBCXX_BEGIN_NAMESPACE_<wbr>VERSION<br>
+_GLIBCXX_BEGIN_NAMESPACE_<wbr>VERSION<br>
+#endif<br>
+#endif<br>
+<br>
+template <class __T, class __Cmp><br>
+inline __device__ const __T &<br>
+max(const __T &__a, const __T &__b, __Cmp __cmp) {<br>
+ return __cmp(__a, __b) ? __b : __a;<br>
+}<br>
+<br>
+template <class __T><br>
+inline __device__ const __T &<br>
+max(const __T &__a, const __T &__b) {<br>
+ return __a < __b ? __b : __a;<br>
+}<br>
+<br>
+template <class __T, class __Cmp><br>
+inline __device__ const __T &<br>
+min(const __T &__a, const __T &__b, __Cmp __cmp) {<br>
+ return __cmp(__b, __a) ? __b : __a;<br>
+}<br>
+<br>
+template <class __T><br>
+inline __device__ const __T &<br>
+min(const __T &__a, const __T &__b) {<br>
+ return __a < __b ? __b : __a;<br>
+}<br>
+<br>
+#ifdef _LIBCPP_END_NAMESPACE_STD<br>
+_LIBCPP_END_NAMESPACE_STD<br>
+#else<br>
+#ifdef _GLIBCXX_BEGIN_NAMESPACE_<wbr>VERSION<br>
+_GLIBCXX_END_NAMESPACE_<wbr>VERSION<br>
+#endif<br>
+} // namespace std<br>
+#endif<br>
+<br>
+#endif // __cplusplus <= 201103L<br>
+#endif // __CLANG_CUDA_WRAPPERS_<wbr>ALGORITHM<br>
<br>
Added: cfe/trunk/lib/Headers/cuda_<wbr>wrappers/complex<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/cuda_wrappers/complex?rev=283680&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/cfe/trunk/lib/Headers/<wbr>cuda_wrappers/complex?rev=<wbr>283680&view=auto</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- cfe/trunk/lib/Headers/cuda_<wbr>wrappers/complex (added)<br>
+++ cfe/trunk/lib/Headers/cuda_<wbr>wrappers/complex Sat Oct 8 17:16:12 2016<br>
@@ -0,0 +1,79 @@<br>
+/*===---- complex - CUDA wrapper for <complex> ------------------------------<wbr>===<br>
+ *<br>
+ * Permission is hereby granted, free of charge, to any person obtaining a copy<br>
+ * of this software and associated documentation files (the "Software"), to deal<br>
+ * in the Software without restriction, including without limitation the rights<br>
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell<br>
+ * copies of the Software, and to permit persons to whom the Software is<br>
+ * furnished to do so, subject to the following conditions:<br>
+ *<br>
+ * The above copyright notice and this permission notice shall be included in<br>
+ * all copies or substantial portions of the Software.<br>
+ *<br>
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR<br>
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,<br>
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE<br>
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER<br>
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,<br>
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN<br>
+ * THE SOFTWARE.<br>
+ *<br>
+ *===--------------------------<wbr>------------------------------<wbr>---------------===<br>
+ */<br>
+<br>
+#pragma once<br>
+<br>
+// Wrapper around <complex> that forces its functions to be __host__<br>
+// __device__.<br>
+<br>
+// First, include host-only headers we think are likely to be included by<br>
+// <complex>, so that the pragma below only applies to <complex> itself.<br>
+#if __cplusplus >= 201103L<br>
+#include <type_traits><br>
+#endif<br>
+#include <stdexcept><br>
+#include <cmath><br>
+#include <sstream><br>
+<br>
+// Next, include our <algorithm> wrapper, to ensure that device overloads of<br>
+// std::min/max are available.<br>
+#include <algorithm><br>
+<br>
+#pragma clang force_cuda_host_device begin<br>
+<br>
+// When compiling for device, ask libstdc++ to use its own implements of<br>
+// complex functions, rather than calling builtins (which resolve to library<br>
+// functions that don't exist when compiling CUDA device code).<br>
+//<br>
+// This is a little dicey, because it causes libstdc++ to define a different<br>
+// set of overloads on host and device.<br>
+//<br>
+// // Present only when compiling for host.<br>
+// __host__ __device__ void complex<float> sin(const complex<float>& x) {<br>
+// return __builtin_csinf(x);<br>
+// }<br>
+//<br>
+// // Present when compiling for host and for device.<br>
+// template <typename T><br>
+// void __host__ __device__ complex<T> sin(const complex<T>& x) {<br>
+// return complex<T>(sin(x.real()) * cosh(x.imag()),<br>
+// cos(x.real()), sinh(x.imag()));<br>
+// }<br>
+//<br>
+// This is safe because when compiling for device, all function calls in<br>
+// __host__ code to sin() will still resolve to *something*, even if they don't<br>
+// resolve to the same function as they resolve to when compiling for host. We<br>
+// don't care that they don't resolve to the right function because we won't<br>
+// codegen this host code when compiling for device.<br>
+<br>
+#pragma push_macro("_GLIBCXX_USE_C99_<wbr>COMPLEX")<br>
+#pragma push_macro("_GLIBCXX_USE_C99_<wbr>COMPLEX_TR1")<br>
+#define _GLIBCXX_USE_C99_COMPLEX 0<br>
+#define _GLIBCXX_USE_C99_COMPLEX_TR1 0<br>
+<br>
+#include_next <complex><br>
+<br>
+#pragma pop_macro("_GLIBCXX_USE_C99_<wbr>COMPLEX_TR1")<br>
+#pragma pop_macro("_GLIBCXX_USE_C99_<wbr>COMPLEX")<br>
+<br>
+#pragma clang force_cuda_host_device end<br>
<br>
<br>
______________________________<wbr>_________________<br>
cfe-commits mailing list<br>
<a href="mailto:cfe-commits@lists.llvm.org">cfe-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/cfe-commits</a><br>
</blockquote></div><br></div>