#include <exec_scope.h>
◆ RegisterReflection()
| static void tvm::tirx::ScopeIdDefNode::RegisterReflection |
( |
| ) |
|
|
inlinestatic |
◆ TVM_FFI_DECLARE_OBJECT_INFO_FINAL()
| tvm::tirx::ScopeIdDefNode::TVM_FFI_DECLARE_OBJECT_INFO_FINAL |
( |
"tirx.ScopeIdDef" |
, |
|
|
ScopeIdDefNode |
, |
|
|
ffi::Object |
|
|
) |
| |
◆ _type_s_eq_hash_kind
| constexpr TVMFFISEqHashKind tvm::tirx::ScopeIdDefNode::_type_s_eq_hash_kind = kTVMFFISEqHashKindTreeNode |
|
staticconstexpr |
◆ def_ids
| ffi::Array<Var> tvm::tirx::ScopeIdDefNode::def_ids |
◆ extents
| ffi::Optional<ffi::Array<PrimExpr> > tvm::tirx::ScopeIdDefNode::extents |
The extents of the ScopeId.
NullOpt means the extent is deferred: the user wrote e.g. bx = T.cta_id() without specifying the extent, and the value will be inferred from sibling ScopeIdDefs at LowerTIRx entry via the verifier's BFS closure. Deferred form requires def_ids.size() == 1 (single axis only – multi-axis defers have no well-defined recovery).
Explicit (Some) form preserves the per-axis shape, e.g. [3, 4, 5] for T.cta_id([3, 4, 5]).
◆ preferred_extents
| ffi::Optional<ffi::Array<PrimExpr> > tvm::tirx::ScopeIdDefNode::preferred_extents |
Optional preferred extents (cluster→cta only). Maps to cudaLaunchAttributePreferredClusterDimension (CUDA 12.8+).
◆ scope
The (parent, cur) binding of this scope id as a closed enum.
The documentation for this class was generated from the following file: