transform.h File Reference

TIR specific transformation passes. More...

#include <tvm/ir/transform.h>
#include <tvm/target/target.h>
#include <tvm/tir/expr.h>
#include <tvm/tir/function.h>
#include <string>
#include <vector>

Include dependency graph for transform.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Namespaces
	tvm
	Performance counters for profiling via the PAPI library.
	tvm::tir
	tvm::tir::transform

Functions
Pass	tvm::tir::transform::CreatePrimFuncPass (std::function< PrimFunc(PrimFunc, IRModule, PassContext)> pass_func, int opt_level, ffi::String name, tvm::ffi::Array< ffi::String > required, bool traceable=false)
Pass	tvm::tir::transform::VectorizeLoop (bool enable_vectorize=true)
	Lower vectorization loops. More...
Pass	tvm::tir::transform::StorageRewrite ()
	Rewrite storage allocation pattern. Moves the allocation to outer most possible scope. Trying to share space between allocations to make a static allocation plan when possible. More...
Pass	tvm::tir::transform::UnrollLoop ()
	unroll the constant loop marked by unroll. This pass also automatically attach pragma unroll tag to loops which meets the standard. More...
Pass	tvm::tir::transform::RemoveNoOp ()
	Remove No Op from the Stmt. More...
Pass	tvm::tir::transform::RewriteUnsafeSelect ()
	Detect and rewrite unsafe select that contains memory access. More...
Pass	tvm::tir::transform::Simplify ()
	Run arithmetic simplifications on the statements and expressions. More...
Pass	tvm::tir::transform::ConvertSSA ()
	Convert an IRModule to be SSA form. More...
Pass	tvm::tir::transform::InstrumentBoundCheckers ()
	Instruments bound checkers. More...
Pass	tvm::tir::transform::MakePackedAPI ()
	Transform the high-level PrimFunc to a low-level version that can be used as an API function. More...
Pass	tvm::tir::transform::MakeUnpackedAPI ()
	Transform the high-level PrimFunc to a C signature that can be used to call the operator directly. More...
Pass	tvm::tir::transform::RemapThreadAxis (ffi::Map< ffi::String, IterVar > axis_map)
	Remap the thread axis. More...
Pass	tvm::tir::transform::LowerCustomDatatypes ()
	Lower custom datatypes. More...
Pass	tvm::tir::transform::DecorateDeviceScope ()
	Decorate all the function's body as device function. More...
Pass	tvm::tir::transform::AnnotateDeviceRegions ()
	Annotate locations that should be run on the device. More...
Pass	tvm::tir::transform::SplitHostDevice ()
	Split the function into a host function and device functions. More...
Pass	tvm::tir::transform::LowerDeviceKernelLaunch ()
	Lower cross-device function calls. More...
Pass	tvm::tir::transform::SkipAssert ()
	skip assert stmt. More...
Pass	tvm::tir::transform::ThreadSync (ffi::String storage_scope)
	Insert sync between parallel read/write of shared buffers. More...
Pass	tvm::tir::transform::LowerThreadAllreduce ()
	Lower cross thread alleduce. More...
Pass	tvm::tir::transform::InferFragment ()
	Infer the TensorCore fragment infomation using tensor intrinsics. More...
Pass	tvm::tir::transform::LowerTVMBuiltin ()
	Lower builtin intrinsics. More...
Pass	tvm::tir::transform::LowerIntrin ()
	Lower the target specific function intrinsics in each of the function. More...
Pass	tvm::tir::transform::LowerWarpMemory ()
	Lower warp memory access to low-level device related function calls. More...
Pass	tvm::tir::transform::LowerDeviceStorageAccessInfo ()
	Lower attached storage access information on device. More...
Pass	tvm::tir::transform::CombineContextCall ()
	Combine context calls in the host function. More...
Pass	tvm::tir::transform::NarrowDataType (int target_bits)
	Narrow down PrimExpr datatype in stmt to target_bits. More...
Pass	tvm::tir::transform::ForceNarrowIndexToInt32 ()
	Force to narrow down indexing expressions and integer buffers to int32 dtype. More...
Pass	tvm::tir::transform::BF16ComputeLegalize ()
	Legalize bf16 compute Ops. Add a cast to fp32 before Ops, then add a cast back to bf16. More...
Pass	tvm::tir::transform::FP8ComputeLegalize (ffi::String promote_dtype="float16")
	Legalize fp8 compute Ops. Add a cast to fp16/fp32 before Ops, then add a cast back to fp8. More...
Pass	tvm::tir::transform::BF16StorageLegalize ()
	Legalize bf16 storage types to u16. More...
Pass	tvm::tir::transform::FP8StorageLegalize ()
	Legalize fp8 storage types to u8. More...
Pass	tvm::tir::transform::InlinePrivateFunctions ()
	Inline calls to private functions. More...
Pass	tvm::tir::transform::PointerValueTypeRewrite ()
	Rewrite the pointer content type of arguments, as well as Alloc internal to the function to use the most frequently accessed type for load/store to avoid pointer casting in backend when possible. More...
Pass	tvm::tir::transform::HoistIfThenElse ()
	Hoist loop-invariant IfThenElse nodes to outside the elligible loops. More...
Pass	tvm::tir::transform::HoistExpression ()
	Hoist loop-invariant expressions nodes to outside the elligible loops. More...
Pass	tvm::tir::transform::FlattenBuffer ()
	Flatten the multi-dimensional BufferLoad and BufferStore to single dimensional BufferLoad/BufferStore for the TIR not contains opaque block. More...
Pass	tvm::tir::transform::LowerVtcmAlloc ()
Pass	tvm::tir::transform::LowerAsyncDMA ()
	Lower Async TIR primitives to DMA copy and wait builtins. More...
Pass	tvm::tir::transform::CommonSubexprElimTIR (bool enable_cse_tir=true, bool identify_equiv_terms=false)
	Implements a Common Subexpression Elimination (CSE) for TIR which introduces let-in bindings for duplicated sub-expressions. More...
Pass	tvm::tir::transform::MergeSharedMemoryAllocations ()
Pass	tvm::tir::transform::ConvertForLoopsToSerial ()
	This pass is post-scheduling pass to convert all Parallel For loops to Serial ones. This is run to attain lesser memory and/or executor/backend does not support parallel launch of For loops. More...
Pass	tvm::tir::transform::UnifiedStaticMemoryPlanner ()
	This is the unified static memory planner pass that will plan for memory intra- and inter- PrimFuncs together. The pass requires all the function to be PrimFuncs including the main. More...
Pass	tvm::tir::transform::BindParams (const ffi::Array< runtime::Tensor > &constants)
Pass	tvm::tir::transform::ExtractPrimFuncConstants ()
	Pass to collect tir non-scalar constants into module's 'Constants' attribute. More...
Pass	tvm::tir::transform::RenormalizeSplitPattern ()
	Renormalize the split pattern from floordiv(floormod()) to floormod(floordiv()) More...
Pass	tvm::tir::transform::BindTarget (Target target)
	Annotate a PrimFunc with a given target. More...
Pass	tvm::tir::transform::AnnotateEntryFunc ()
	Set a PrimFunc as the entry point if it is only function in IRModule. More...
Pass	tvm::tir::transform::Filter (ffi::TypedFunction< bool(PrimFunc)> fcond)
	Filter PrimFuncs with a given condition. More...
Pass	tvm::tir::transform::InjectPTXAsyncCopy ()
	Pass to rewrite global to shared memory copy on CUDA with asyncronous copy. More...
Pass	tvm::tir::transform::InjectPTXLDG32 (bool enable_ptx_ldg32=true)
	Pass to rewrite global to local memory copy on CUDA with ldg32 instruction. More...
Pass	tvm::tir::transform::RemoveWeightLayoutRewriteBlock (bool skip_tensor_rewrite=false)
	Remove the weight layout rewrite block. More...
Pass	tvm::tir::transform::InstrumentProfileIntrinsics ()
	Insert intrinsic calls to instrument function and loop level profiling. More...
Pass	tvm::tir::transform::DefaultGPUSchedule ()
	The pass sets default thread bindings for PrimFuncs, including symbolic shape functions, allowing their build and execution on GPU devices. It examines all the blocks within the PrimFunc and conducts loop fusion, splitting, and reordering operations based on the loop extent and target information, such as the maximum thread block number and maximum thread per block. More...
Pass	tvm::tir::transform::UseAssumeToReduceBranches ()
	This pass analyzes primfunc & eliminates branch introdued due to layout specific padding. It leverages from the buffer assumptions and use the information to eliminate the branch. More...

Detailed Description

TIR specific transformation passes.