rocPRIM
|
![]() |
Namespaces | |
detail | |
Deprecated: Configuration of device-level scan primitives. | |
Classes | |
class | double_buffer< T > |
This class provides an convenient way to do double buffering. More... | |
class | future_value< T, Iter > |
Allows passing values that are not yet known at launch time as paramters to device algorithms. More... | |
struct | key_value_pair< Key_, Value_ > |
Convenience struct that allows you to store key-value pairs as a composite entity. More... | |
struct | empty_type |
Empty type used as a placeholder, usually used to flag that given template parameter should not be used. More... | |
struct | empty_binary_op |
Binary operator that takes two instances of empty_type, usually used as nop replacement for the HIP-CPU back-end. More... | |
Macros | |
#define | __AMDGCN_WAVEFRONT_SIZE 64 |
Typedefs | |
using | half = ::__half |
Half-precision floating point type. | |
using | bfloat16 = ::hip_bfloat16 |
bfloat16 floating point type | |
using | lane_mask_type = unsigned long long int |
The lane_mask_type is an integer that contains one bit per thread. More... | |
using | native_half = _Float16 |
Native half-precision floating point type. | |
using | native_bfloat16 = bfloat16 |
native bfloat16 type | |
using lane_mask_type = unsigned long long int |
The lane_mask_type is an integer that contains one bit per thread.
The total number of bits is equal to the total number of threads in a warp. Used to for warp-level operations.