[CUDA] Support <complex> and std::min/max on the device.

[jlebar]

Summary

We do this by wrapping <complex> and <algorithm>.

Tests are in the test-suite. <complex.h> support to come separately.

[tra]

This looks like fix-includes and it may be somewhat shaky if users start messing with include paths. You may want to get @echristo's input on that. I' personally would prefer to force-include these files. I suspect it will not change things much as we already include a lot.

clang/lib/Driver/ToolChains.cpp

4704 ↗

(On Diff #72686)

path::append accepts multiple path parts so you can construct path in one call.

[jlebar]

I' personally would prefer to force-include these files. I suspect it will not change things much as we already include a lot.

We have already had bugs filed by users whose root cause was that we #included more things than nvcc #includes. I know exact compatibility with nvcc is not our goal, but unless we have a good reason I don't think we should break compatibility with nvcc *and* the C++ standard by force-including additional system headers.

This looks like fix-includes and it may be somewhat shaky if users start messing with include paths.

We add this include path first, so I think it should be OK? What do you think, @echristo?

[echristo]

In D24979#554378, @jlebar wrote:

I' personally would prefer to force-include these files. I suspect it will not change things much as we already include a lot.

We have already had bugs filed by users whose root cause was that we #included more things than nvcc #includes. I know exact compatibility with nvcc is not our goal, but unless we have a good reason I don't think we should break compatibility with nvcc *and* the C++ standard by force-including additional system headers.

This looks like fix-includes and it may be somewhat shaky if users start messing with include paths.

We add this include path first, so I think it should be OK? What do you think, @echristo?

It is very similar to fixincludes, but in this case I think it's probably the right thing.

If nothing else it's very easy to change our minds later.

One small wording change requested.

Thanks!

-eric

clang/lib/Headers/__clang_cuda_complex_builtins.h

1 ↗

(On Diff #72686)

Would swap libgcc for "runtime"

[jlebar]

s/libgcc/runtime/

cfe/trunk/lib/Driver/ToolChains.cpp

if (addLibStdCXXIncludePaths(IncludePath, /*Suffix*/ "", TripleStr,
                                  /*TargetMultiarchTriple*/ "",
                                  Multilib.includeSuffix(), DriverArgs, CC1Args))
       break;
   }
 }
 
 void Linux::AddCudaIncludeArgs(const ArgList &DriverArgs,
                                ArgStringList &CC1Args) const {
+  if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
+    // Add cuda_wrappers/* to our system include path.  This lets us wrap
+    // standard library headers.
+    SmallString<128> P(getDriver().ResourceDir);
+    llvm::sys::path::append(P, "include");
+    llvm::sys::path::append(P, "cuda_wrappers");
+    addSystemInclude(DriverArgs, CC1Args, P);
+  }
+
   if (DriverArgs.hasArg(options::OPT_nocudainc))
     return;
 
   if (!CudaInstallation.isValid()) {
     getDriver().Diag(diag::err_drv_no_cuda_installation);
     return;
   }
 

cfe/trunk/lib/Headers/CMakeLists.txt

set(files
   avx512vlcdintrin.h
   avx512vldqintrin.h
   avx512vlintrin.h
   avxintrin.h
   bmi2intrin.h
   bmiintrin.h
   __clang_cuda_builtin_vars.h
   __clang_cuda_cmath.h
+  __clang_cuda_complex_builtins.h
   __clang_cuda_intrinsics.h
   __clang_cuda_math_forward_declares.h
   __clang_cuda_runtime_wrapper.h
   cpuid.h
+  cuda_wrappers/algorithm
+  cuda_wrappers/complex
   clflushoptintrin.h
   emmintrin.h
   f16cintrin.h
   float.h
   fma4intrin.h
   fmaintrin.h
   fxsrintrin.h
   htmintrin.h

cfe/trunk/lib/Headers/__clang_cuda_complex_builtins.h

+/*===-- __clang_cuda_complex_builtins - CUDA impls of runtime complex fns ---===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CLANG_CUDA_COMPLEX_BUILTINS
+#define __CLANG_CUDA_COMPLEX_BUILTINS
+
+// This header defines __muldc3, __mulsc3, __divdc3, and __divsc3.  These are
+// libgcc functions that clang assumes are available when compiling c99 complex
+// operations.  (These implementations come from libc++, and have been modified
+// to work with CUDA.)
+
+extern "C" inline __device__ double _Complex __muldc3(double __a, double __b,
+                                                      double __c, double __d) {
+  double __ac = __a * __c;
+  double __bd = __b * __d;
+  double __ad = __a * __d;
+  double __bc = __b * __c;
+  double _Complex z;
+  __real__(z) = __ac - __bd;
+  __imag__(z) = __ad + __bc;
+  if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {
+    int __recalc = 0;
+    if (std::isinf(__a) || std::isinf(__b)) {
+      __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);
+      __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);
+      if (std::isnan(__c))
+        __c = std::copysign(0, __c);
+      if (std::isnan(__d))
+        __d = std::copysign(0, __d);
+      __recalc = 1;
+    }
+    if (std::isinf(__c) || std::isinf(__d)) {
+      __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);
+      __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);
+      if (std::isnan(__a))
+        __a = std::copysign(0, __a);
+      if (std::isnan(__b))
+        __b = std::copysign(0, __b);
+      __recalc = 1;
+    }
+    if (!__recalc && (std::isinf(__ac) || std::isinf(__bd) ||
+                      std::isinf(__ad) || std::isinf(__bc))) {
+      if (std::isnan(__a))
+        __a = std::copysign(0, __a);
+      if (std::isnan(__b))
+        __b = std::copysign(0, __b);
+      if (std::isnan(__c))
+        __c = std::copysign(0, __c);
+      if (std::isnan(__d))
+        __d = std::copysign(0, __d);
+      __recalc = 1;
+    }
+    if (__recalc) {
+      // Can't use std::numeric_limits<double>::infinity() -- that doesn't have
+      // a device overload (and isn't constexpr before C++11, naturally).
+      __real__(z) = __builtin_huge_valf() * (__a * __c - __b * __d);
+      __imag__(z) = __builtin_huge_valf() * (__a * __d + __b * __c);
+    }
+  }
+  return z;
+}
+
+extern "C" inline __device__ float _Complex __mulsc3(float __a, float __b,
+                                                     float __c, float __d) {
+  float __ac = __a * __c;
+  float __bd = __b * __d;
+  float __ad = __a * __d;
+  float __bc = __b * __c;
+  float _Complex z;
+  __real__(z) = __ac - __bd;
+  __imag__(z) = __ad + __bc;
+  if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {
+    int __recalc = 0;
+    if (std::isinf(__a) || std::isinf(__b)) {
+      __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);
+      __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);
+      if (std::isnan(__c))
+        __c = std::copysign(0, __c);
+      if (std::isnan(__d))
+        __d = std::copysign(0, __d);
+      __recalc = 1;
+    }
+    if (std::isinf(__c) || std::isinf(__d)) {
+      __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);
+      __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);
+      if (std::isnan(__a))
+        __a = std::copysign(0, __a);
+      if (std::isnan(__b))
+        __b = std::copysign(0, __b);
+      __recalc = 1;
+    }
+    if (!__recalc && (std::isinf(__ac) || std::isinf(__bd) ||
+                      std::isinf(__ad) || std::isinf(__bc))) {
+      if (std::isnan(__a))
+        __a = std::copysign(0, __a);
+      if (std::isnan(__b))
+        __b = std::copysign(0, __b);
+      if (std::isnan(__c))
+        __c = std::copysign(0, __c);
+      if (std::isnan(__d))
+        __d = std::copysign(0, __d);
+      __recalc = 1;
+    }
+    if (__recalc) {
+      __real__(z) = __builtin_huge_valf() * (__a * __c - __b * __d);
+      __imag__(z) = __builtin_huge_valf() * (__a * __d + __b * __c);
+    }
+  }
+  return z;
+}
+
+extern "C" inline __device__ double _Complex __divdc3(double __a, double __b,
+                                                      double __c, double __d) {
+  int __ilogbw = 0;
+  // Can't use std::max, because that's defined in <algorithm>, and we don't
+  // want to pull that in for every compile.  The CUDA headers define
+  // ::max(float, float) and ::max(double, double), which is sufficient for us.
+  double __logbw = std::logb(max(std::abs(__c), std::abs(__d)));
+  if (std::isfinite(__logbw)) {
+    __ilogbw = (int)__logbw;
+    __c = std::scalbn(__c, -__ilogbw);
+    __d = std::scalbn(__d, -__ilogbw);
+  }
+  double __denom = __c * __c + __d * __d;
+  double _Complex z;
+  __real__(z) = std::scalbn((__a * __c + __b * __d) / __denom, -__ilogbw);
+  __imag__(z) = std::scalbn((__b * __c - __a * __d) / __denom, -__ilogbw);
+  if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {
+    if ((__denom == 0.0) && (!std::isnan(__a) || !std::isnan(__b))) {
+      __real__(z) = std::copysign(__builtin_huge_valf(), __c) * __a;
+      __imag__(z) = std::copysign(__builtin_huge_valf(), __c) * __b;
+    } else if ((std::isinf(__a) || std::isinf(__b)) && std::isfinite(__c) &&
+               std::isfinite(__d)) {
+      __a = std::copysign(std::isinf(__a) ? 1.0 : 0.0, __a);
+      __b = std::copysign(std::isinf(__b) ? 1.0 : 0.0, __b);
+      __real__(z) = __builtin_huge_valf() * (__a * __c + __b * __d);
+      __imag__(z) = __builtin_huge_valf() * (__b * __c - __a * __d);
+    } else if (std::isinf(__logbw) && __logbw > 0.0 && std::isfinite(__a) &&
+               std::isfinite(__b)) {
+      __c = std::copysign(std::isinf(__c) ? 1.0 : 0.0, __c);
+      __d = std::copysign(std::isinf(__d) ? 1.0 : 0.0, __d);
+      __real__(z) = 0.0 * (__a * __c + __b * __d);
+      __imag__(z) = 0.0 * (__b * __c - __a * __d);
+    }
+  }
+  return z;
+}
+
+extern "C" inline __device__ float _Complex __divsc3(float __a, float __b,
+                                                     float __c, float __d) {
+  int __ilogbw = 0;
+  float __logbw = std::logb(max(std::abs(__c), std::abs(__d)));
+  if (std::isfinite(__logbw)) {
+    __ilogbw = (int)__logbw;
+    __c = std::scalbn(__c, -__ilogbw);
+    __d = std::scalbn(__d, -__ilogbw);
+  }
+  float __denom = __c * __c + __d * __d;
+  float _Complex z;
+  __real__(z) = std::scalbn((__a * __c + __b * __d) / __denom, -__ilogbw);
+  __imag__(z) = std::scalbn((__b * __c - __a * __d) / __denom, -__ilogbw);
+  if (std::isnan(__real__(z)) && std::isnan(__imag__(z))) {
+    if ((__denom == 0) && (!std::isnan(__a) || !std::isnan(__b))) {
+      __real__(z) = std::copysign(__builtin_huge_valf(), __c) * __a;
+      __imag__(z) = std::copysign(__builtin_huge_valf(), __c) * __b;
+    } else if ((std::isinf(__a) || std::isinf(__b)) && std::isfinite(__c) &&
+               std::isfinite(__d)) {
+      __a = std::copysign(std::isinf(__a) ? 1 : 0, __a);
+      __b = std::copysign(std::isinf(__b) ? 1 : 0, __b);
+      __real__(z) = __builtin_huge_valf() * (__a * __c + __b * __d);
+      __imag__(z) = __builtin_huge_valf() * (__b * __c - __a * __d);
+    } else if (std::isinf(__logbw) && __logbw > 0 && std::isfinite(__a) &&
+               std::isfinite(__b)) {
+      __c = std::copysign(std::isinf(__c) ? 1 : 0, __c);
+      __d = std::copysign(std::isinf(__d) ? 1 : 0, __d);
+      __real__(z) = 0 * (__a * __c + __b * __d);
+      __imag__(z) = 0 * (__b * __c - __a * __d);
+    }
+  }
+  return z;
+}
+
+#endif // __CLANG_CUDA_COMPLEX_BUILTINS

cfe/trunk/lib/Headers/__clang_cuda_runtime_wrapper.h

 }
 
 __device__ inline __cuda_builtin_gridDim_t::operator dim3() const {
   return dim3(x, y, z);
 }
 
 #include <__clang_cuda_cmath.h>
 #include <__clang_cuda_intrinsics.h>
+#include <__clang_cuda_complex_builtins.h>
 
 // curand_mtgp32_kernel helpfully redeclares blockDim and threadIdx in host
 // mode, giving them their "proper" types of dim3 and uint3.  This is
 // incompatible with the types we give in __clang_cuda_builtin_vars.h.  As as
 // hack, force-include the header (nvcc doesn't include it by default) but
 // redefine dim3 and uint3 to our builtin types.  (Thankfully dim3 and uint3 are
 // only used here for the redeclarations of blockDim and threadIdx.)
 #pragma push_macro("dim3")

cfe/trunk/lib/Headers/cuda_wrappers/algorithm

+/*===---- complex - CUDA wrapper for <algorithm> ----------------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CLANG_CUDA_WRAPPERS_ALGORITHM
+#define __CLANG_CUDA_WRAPPERS_ALGORITHM
+
+// This header defines __device__ overloads of std::min/max, but only if we're
+// <= C++11.  In C++14, these functions are constexpr, and so are implicitly
+// __host__ __device__.
+//
+// We don't support the initializer_list overloads because
+// initializer_list::begin() and end() are not __host__ __device__ functions.
+//
+// When compiling in C++14 mode, we could force std::min/max to have different
+// implementations for host and device, by declaring the device overloads
+// before the constexpr overloads appear.  We choose not to do this because
+
+//  a) why write our own implementation when we can use one from the standard
+//     library? and
+//  b) libstdc++ is evil and declares min/max inside a header that is included
+//     *before* we include <algorithm>.  So we'd have to unconditionally
+//     declare our __device__ overloads of min/max, but that would pollute
+//     things for people who choose not to include <algorithm>.
+
+#include_next <algorithm>
+
+#if __cplusplus <= 201103L
+
+// We need to define these overloads in exactly the namespace our standard
+// library uses (including the right inline namespace), otherwise they won't be
+// picked up by other functions in the standard library (e.g. functions in
+// <complex>).  Thus the ugliness below.
+#ifdef _LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_NAMESPACE_STD
+#else
+namespace std {
+#ifdef _GLIBCXX_BEGIN_NAMESPACE_VERSION
+_GLIBCXX_BEGIN_NAMESPACE_VERSION
+#endif
+#endif
+
+template <class __T, class __Cmp>
+inline __device__ const __T &
+max(const __T &__a, const __T &__b, __Cmp __cmp) {
+  return __cmp(__a, __b) ? __b : __a;
+}
+
+template <class __T>
+inline __device__ const __T &
+max(const __T &__a, const __T &__b) {
+  return __a < __b ? __b : __a;
+}
+
+template <class __T, class __Cmp>
+inline __device__ const __T &
+min(const __T &__a, const __T &__b, __Cmp __cmp) {
+  return __cmp(__b, __a) ? __b : __a;
+}
+
+template <class __T>
+inline __device__ const __T &
+min(const __T &__a, const __T &__b) {
+  return __a < __b ? __b : __a;
+}
+
+#ifdef _LIBCPP_END_NAMESPACE_STD
+_LIBCPP_END_NAMESPACE_STD
+#else
+#ifdef _GLIBCXX_BEGIN_NAMESPACE_VERSION
+_GLIBCXX_END_NAMESPACE_VERSION
+#endif
+} // namespace std
+#endif
+
+#endif // __cplusplus <= 201103L
+#endif // __CLANG_CUDA_WRAPPERS_ALGORITHM

cfe/trunk/lib/Headers/cuda_wrappers/complex

+/*===---- complex - CUDA wrapper for <complex> ------------------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#pragma once
+
+// Wrapper around <complex> that forces its functions to be __host__
+// __device__.
+
+// First, include host-only headers we think are likely to be included by
+// <complex>, so that the pragma below only applies to <complex> itself.
+#if __cplusplus >= 201103L
+#include <type_traits>
+#endif
+#include <stdexcept>
+#include <cmath>
+#include <sstream>
+
+// Next, include our <algorithm> wrapper, to ensure that device overloads of
+// std::min/max are available.
+#include <algorithm>
+
+#pragma clang force_cuda_host_device begin
+
+// When compiling for device, ask libstdc++ to use its own implements of
+// complex functions, rather than calling builtins (which resolve to library
+// functions that don't exist when compiling CUDA device code).
+//
+// This is a little dicey, because it causes libstdc++ to define a different
+// set of overloads on host and device.
+//
+//   // Present only when compiling for host.
+//   __host__ __device__ void complex<float> sin(const complex<float>& x) {
+//     return __builtin_csinf(x);
+//   }
+//
+//   // Present when compiling for host and for device.
+//   template <typename T>
+//   void __host__ __device__ complex<T> sin(const complex<T>& x) {
+//     return complex<T>(sin(x.real()) * cosh(x.imag()),
+//                       cos(x.real()), sinh(x.imag()));
+//   }
+//
+// This is safe because when compiling for device, all function calls in
+// __host__ code to sin() will still resolve to *something*, even if they don't
+// resolve to the same function as they resolve to when compiling for host.  We
+// don't care that they don't resolve to the right function because we won't
+// codegen this host code when compiling for device.
+
+#pragma push_macro("_GLIBCXX_USE_C99_COMPLEX")
+#pragma push_macro("_GLIBCXX_USE_C99_COMPLEX_TR1")
+#define _GLIBCXX_USE_C99_COMPLEX 0
+#define _GLIBCXX_USE_C99_COMPLEX_TR1 0
+
+#include_next <complex>
+
+#pragma pop_macro("_GLIBCXX_USE_C99_COMPLEX_TR1")
+#pragma pop_macro("_GLIBCXX_USE_C99_COMPLEX")
+
+#pragma clang force_cuda_host_device end