[CUDA] Make __GCC_ATOMIC_XXX_LOCK_FREE macros the same on host/device.

[jlebar]

Summary

This fixes a bug where we were unable to compile the following CUDA
file with libstdc++ (didn't try libc++):

#include <future>
void foo() { std::shared_future<int> x; }

The problem is that <future> only defines std::shared_future if
__GCC_ATOMIC_INT_LOCK_FREE > 1. When we compiled this file for device,
the macro was set to 1, and then the class didn't exist at all.

[tra]

LGTM

clang/lib/Basic/Targets.cpp

NVPTXTargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
     ProcessIDType = HostTarget->getProcessIDType();
 
     UseBitFieldTypeAlignment = HostTarget->useBitFieldTypeAlignment();
     UseZeroLengthBitfieldAlignment =
         HostTarget->useZeroLengthBitfieldAlignment();
     UseExplicitBitFieldAlignment = HostTarget->useExplicitBitFieldAlignment();
     ZeroLengthBitfieldBoundary = HostTarget->getZeroLengthBitfieldBoundary();
 
+    // This is a bit of a lie, but it controls __GCC_ATOMIC_XXX_LOCK_FREE, and
+    // we need those macros to be identical on host and device, because (among
+    // other things) they affect which standard library classes are defined, and
+    // we need all classes to be defined on both the host and device.
+    MaxAtomicInlineWidth = HostTarget->getMaxAtomicInlineWidth();
+
     // Properties intentionally not copied from host:
     // - LargeArrayMinWidth, LargeArrayAlign: Not visible across the
     //   host/device boundary.
     // - SuitableAlign: Not visible across the host/device boundary, and may
     //   correctly be different on host/device, e.g. if host has wider vector
     //   types than device.
     // - LongDoubleWidth, LongDoubleAlign: nvptx's long double type is the same
     //   as its double type, but that's not necessarily true on the host.

clang/test/Preprocessor/cuda-types.cu

-// Check that types, widths, etc. match on the host and device sides of CUDA
-// compilations.  Note that we filter out long double, as this is intentionally
-// different on host and device.
+// Check that types, widths, __GCC_ATOMIC* macros, etc. match on the host and
+// device sides of CUDA compilations.  Note that we filter out long double, as
+// this is intentionally different on host and device.
+//
+// FIXME: We really should make __GCC_HAVE_SYNC_COMPARE_AND_SWAP identical on
+// host and device, but architecturally this is difficult at the moment.
 
-// RUN: %clang --cuda-host-only -nocudainc -target i386-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/i386-host-defines
-// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target i386-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/i386-device-defines
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)' %T/i386-host-defines   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/i386-host-defines-filtered
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)' %T/i386-device-defines | grep -v '__LDBL\|_LONG_DOUBLE' > %T/i386-device-defines-filtered
+// RUN: %clang --cuda-host-only -nocudainc -target i386-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/i386-host-defines-filtered
+// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target i386-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/i386-device-defines-filtered
 // RUN: diff %T/i386-host-defines-filtered %T/i386-device-defines-filtered
 
-// RUN: %clang --cuda-host-only -nocudainc -target x86_64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/x86_64-host-defines
-// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target x86_64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/x86_64-device-defines
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF\|WIDTH\)' %T/x86_64-host-defines   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/x86_64-host-defines-filtered
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF\|WIDTH\)' %T/x86_64-device-defines | grep -v '__LDBL\|_LONG_DOUBLE' > %T/x86_64-device-defines-filtered
+// RUN: %clang --cuda-host-only -nocudainc -target x86_64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/x86_64-host-defines-filtered
+// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target x86_64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/x86_64-device-defines-filtered
 // RUN: diff %T/x86_64-host-defines-filtered %T/x86_64-device-defines-filtered
 
-// RUN: %clang --cuda-host-only -nocudainc -target powerpc64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/powerpc64-host-defines
-// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target powerpc64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null > %T/powerpc64-device-defines
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF\|WIDTH\)' %T/powerpc64-host-defines   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/powerpc64-host-defines-filtered
-// RUN: grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF\|WIDTH\)' %T/powerpc64-device-defines | grep -v '__LDBL\|_LONG_DOUBLE' > %T/powerpc64-device-defines-filtered
+// RUN: %clang --cuda-host-only -nocudainc -target powerpc64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/powerpc64-host-defines-filtered
+// RUN: %clang --cuda-device-only -nocudainc -nocudalib -target powerpc64-unknown-linux-gnu -x cuda -E -dM -o - /dev/null \
+// RUN:   | grep 'define __[^ ]*\(TYPE\|MAX\|SIZEOF|WIDTH\)\|define __GCC_ATOMIC' \
+// RUN:   | grep -v '__LDBL\|_LONG_DOUBLE' > %T/powerpc64-device-defines-filtered
 // RUN: diff %T/powerpc64-host-defines-filtered %T/powerpc64-device-defines-filtered