Correct ill-formed merge in earlier commit and adjust for differences with the new CUDA natural indexing mechanism.
Этот коммит содержится в:
@@ -389,9 +389,18 @@ template<typename std::common_type<
|
||||
class Coordinates {
|
||||
using R = decltype(f(0));
|
||||
|
||||
struct X { __device__ operator R() const { return f(0); } };
|
||||
struct Y { __device__ operator R() const { return f(1); } };
|
||||
struct Z { __device__ operator R() const { return f(2); } };
|
||||
struct X {
|
||||
__device__ operator R() const { return f(0); }
|
||||
__host__ operator R() const { return R{}; }
|
||||
};
|
||||
struct Y {
|
||||
__device__ operator R() const { return f(1); }
|
||||
__host__ operator R() const { return R{}; }
|
||||
};
|
||||
struct Z {
|
||||
__device__ operator R() const { return f(2); }
|
||||
__host__ operator R() const { return R{}; }
|
||||
};
|
||||
public:
|
||||
static constexpr X x{};
|
||||
static constexpr Y y{};
|
||||
|
||||
@@ -91,7 +91,6 @@ namespace hip_impl
|
||||
const dim3& dimBlocks,
|
||||
uint32_t sharedMemBytes,
|
||||
hipStream_t stream,
|
||||
<<<<<<< HEAD
|
||||
void** kernarg)
|
||||
{
|
||||
const auto it0 = functions().find(function_address);
|
||||
@@ -134,22 +133,6 @@ namespace hip_impl
|
||||
nullptr,
|
||||
kernarg);
|
||||
}
|
||||
=======
|
||||
void* locked_stream,
|
||||
const char* kernel_name,
|
||||
hc::accelerator_view* acc_v)
|
||||
{ // Precondition: acc_v is the accelerator_view associated with stream
|
||||
// which is guarded by locked_stream;
|
||||
// locked_stream is deletable.
|
||||
using L = decltype(stream->lockopen_preKernelCommand());
|
||||
|
||||
stream->lockclose_postKernelCommand(kernel_name, acc_v);
|
||||
|
||||
delete static_cast<L*>(locked_stream);
|
||||
locked_stream = nullptr;
|
||||
if(HIP_PROFILE_API) {
|
||||
MARKER_END();
|
||||
>>>>>>> e8ede28ec4f5744185b171031e537237afb7affa
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Ссылка в новой задаче
Block a user