tvm_tirx.tirx.script

Public canonical TVMScript dialect namespace.

Classes

Optional
LetAnnotation([type_spec])	Marker for an immutable Bind, created by T.let or T.let[type].

Alias	Canonical API
Bind	tvm_tirx.ir.Bind
Call	tvm_tirx.ir.Call
DataTypeImm	tvm_tirx.ir.DataTypeImm
FuncType	tvm_tirx.ir.FuncType
GenericConst	tvm_tirx.ir.GenericConst
MissingType	tvm_tirx.ir.MissingType
Op	tvm_tirx.ir.Op
PrimType	tvm_tirx.ir.PrimType
Range	tvm_tirx.ir.Range
StringImm	tvm_tirx.ir.StringImm
StringType	tvm_tirx.ir.StringType
TensorRegion	tvm_tirx.ir.TensorRegion
TensorRegionType	tvm_tirx.ir.TensorRegionType
Type	tvm_tirx.ir.Type
Var	tvm_tirx.ir.Var
AlreadyEmitted	tvm_tirx.script.ir_builder.AlreadyEmitted
IRBuilder	tvm_tirx.script.ir_builder.IRBuilder
IRModuleFrame	tvm_tirx.script.ir_builder.IRModuleFrame
Missing	tvm_tirx.ir.MissingType
NoneType	tvm_tirx.ir.NoneType
Any	tvm_tirx.ir.AnyType
Opaque	tvm_tirx.ir.OpaqueType
Prim	tvm_tirx.ir.PrimType
String	tvm_tirx.ir.StringType
Ptr	tvm_tirx.ir.PtrType
Func	tvm_tirx.ir.FuncType
FloatImm	tvm_tirx.ir.FloatImm
IntImm	tvm_tirx.ir.IntImm
TensorLoad	tvm_tirx.ir.TensorLoad
MemoryScope	tvm_tirx.runtime.MemoryScope
Axis	tvm_tirx.tirx.layout.Axis
ComposeLayout	tvm_tirx.tirx.ComposeLayout
Iter	tvm_tirx.tirx.layout.Iter
Layout	tvm_tirx.tirx.Layout
TileLayout	tvm_tirx.tirx.TileLayout
ThreadScope	tvm_tirx.tirx.ThreadScope
TensorMap	tvm_tirx.tirx.TensorMapType
Tensor	tvm_tirx.tirx.TensorType
CallEffectKind	tvm_tirx.ir.CallEffectKind
Broadcast	tvm_tirx.ir.prim.Broadcast
Ramp	tvm_tirx.ir.prim.Ramp
Shuffle	tvm_tirx.ir.prim.Shuffle
EQ	tvm_tirx.ir.prim.EQ
GE	tvm_tirx.ir.prim.GE
GT	tvm_tirx.ir.prim.GT
LE	tvm_tirx.ir.prim.LE
LT	tvm_tirx.ir.prim.LT
NE	tvm_tirx.ir.prim.NE
Add	tvm_tirx.ir.prim.Add
And	tvm_tirx.ir.prim.And
BitwiseAnd	tvm_tirx.ir.prim.BitwiseAnd
BitwiseNot	tvm_tirx.ir.prim.BitwiseNot
BitwiseOr	tvm_tirx.ir.prim.BitwiseOr
BitwiseXor	tvm_tirx.ir.prim.BitwiseXor
Cast	tvm_tirx.ir.prim.Cast
Div	tvm_tirx.ir.prim.Div
FloorDiv	tvm_tirx.ir.prim.FloorDiv
FloorMod	tvm_tirx.ir.prim.FloorMod
LShift	tvm_tirx.ir.prim.LShift
Max	tvm_tirx.ir.prim.Max
Min	tvm_tirx.ir.prim.Min
Mod	tvm_tirx.ir.prim.Mod
Mul	tvm_tirx.ir.prim.Mul
Not	tvm_tirx.ir.prim.Not
Or	tvm_tirx.ir.prim.Or
RShift	tvm_tirx.ir.prim.RShift
Select	tvm_tirx.ir.prim.Select
Sub	tvm_tirx.ir.prim.Sub
CallFFIKernelAttr	tvm_tirx.tirx.CallFFIKernelAttr
broadcast	tvm_tirx.ir.prim.Broadcast
ramp	tvm_tirx.ir.prim.Ramp

Functions

Tuple(*fields)	Construct a tuple type for a TIRx function or binding annotation.
resolve_global_info_(content)	Implements tvm_tirx.script.ir_builder.parser_protocol.resolve_global_info_().
alloc_tensor([placement, attrs, ...])	Allocate the tensor described by one explicit Tensor type argument.
decl_tensor(data, *, ty_args[, attrs, loc, ty])	Declare a pointer-backed tensor with one explicit Tensor type argument.
alloc_shared(shape[, dtype, data, scope, ...])	Emit a tensor allocation and return its bound variable in a receipt.
alloc_local(shape[, dtype, data, scope, ...])	Emit a tensor allocation and return its bound variable in a receipt.
tmem(shape[, dtype, data, scope, align, ...])	Emit a tensor allocation and return its bound variable in a receipt.
alloc_scalar([dtype, scope, annotations])	Allocate a zero-dimensional tensor (scalar), with optional allocation annotations.
decl_scalar(dtype, data, scope[, layout])	Declare a zero-dimensional tensor (scalar) from a pointer.
shared_scalar([dtype, annotations])	Allocate a zero-dimensional tensor in shared memory.
local_scalar([dtype, init, name, annotations])	Allocate scalar local storage, optionally naming and initializing it.
assign(dst, value)	Store a value into scalar storage and return its emission receipt.
static_assert(x[, message])
int8x2	A primitive scalar or vector data type, including the void sentinel.
int16x2	A primitive scalar or vector data type, including the void sentinel.
int32x2	A primitive scalar or vector data type, including the void sentinel.
int64x2	A primitive scalar or vector data type, including the void sentinel.
int8x4	A primitive scalar or vector data type, including the void sentinel.
int16x4	A primitive scalar or vector data type, including the void sentinel.
int32x4	A primitive scalar or vector data type, including the void sentinel.
int64x4	A primitive scalar or vector data type, including the void sentinel.
int8x8	A primitive scalar or vector data type, including the void sentinel.
int16x8	A primitive scalar or vector data type, including the void sentinel.
int32x8	A primitive scalar or vector data type, including the void sentinel.
int64x8	A primitive scalar or vector data type, including the void sentinel.
int8x16	A primitive scalar or vector data type, including the void sentinel.
int16x16	A primitive scalar or vector data type, including the void sentinel.
int32x16	A primitive scalar or vector data type, including the void sentinel.
int64x16	A primitive scalar or vector data type, including the void sentinel.
int8x32	A primitive scalar or vector data type, including the void sentinel.
int16x32	A primitive scalar or vector data type, including the void sentinel.
int32x32	A primitive scalar or vector data type, including the void sentinel.
int64x32	A primitive scalar or vector data type, including the void sentinel.
int8x64	A primitive scalar or vector data type, including the void sentinel.
int16x64	A primitive scalar or vector data type, including the void sentinel.
int32x64	A primitive scalar or vector data type, including the void sentinel.
int64x64	A primitive scalar or vector data type, including the void sentinel.
uint8x2	A primitive scalar or vector data type, including the void sentinel.
uint16x2	A primitive scalar or vector data type, including the void sentinel.
uint32x2	A primitive scalar or vector data type, including the void sentinel.
uint64x2	A primitive scalar or vector data type, including the void sentinel.
uint8x4	A primitive scalar or vector data type, including the void sentinel.
uint16x4	A primitive scalar or vector data type, including the void sentinel.
uint32x4	A primitive scalar or vector data type, including the void sentinel.
uint64x4	A primitive scalar or vector data type, including the void sentinel.
uint8x8	A primitive scalar or vector data type, including the void sentinel.
uint16x8	A primitive scalar or vector data type, including the void sentinel.
uint32x8	A primitive scalar or vector data type, including the void sentinel.
uint64x8	A primitive scalar or vector data type, including the void sentinel.
uint8x16	A primitive scalar or vector data type, including the void sentinel.
uint16x16	A primitive scalar or vector data type, including the void sentinel.
uint32x16	A primitive scalar or vector data type, including the void sentinel.
uint64x16	A primitive scalar or vector data type, including the void sentinel.
uint8x32	A primitive scalar or vector data type, including the void sentinel.
uint16x32	A primitive scalar or vector data type, including the void sentinel.
uint32x32	A primitive scalar or vector data type, including the void sentinel.
uint64x32	A primitive scalar or vector data type, including the void sentinel.
uint8x64	A primitive scalar or vector data type, including the void sentinel.
uint16x64	A primitive scalar or vector data type, including the void sentinel.
uint32x64	A primitive scalar or vector data type, including the void sentinel.
uint64x64	A primitive scalar or vector data type, including the void sentinel.
float16x2	A primitive scalar or vector data type, including the void sentinel.
float32x2	A primitive scalar or vector data type, including the void sentinel.
float64x2	A primitive scalar or vector data type, including the void sentinel.
float16x4	A primitive scalar or vector data type, including the void sentinel.
float32x4	A primitive scalar or vector data type, including the void sentinel.
float64x4	A primitive scalar or vector data type, including the void sentinel.
float16x8	A primitive scalar or vector data type, including the void sentinel.
float32x8	A primitive scalar or vector data type, including the void sentinel.
float64x8	A primitive scalar or vector data type, including the void sentinel.
float16x16	A primitive scalar or vector data type, including the void sentinel.
float32x16	A primitive scalar or vector data type, including the void sentinel.
float64x16	A primitive scalar or vector data type, including the void sentinel.
float16x32	A primitive scalar or vector data type, including the void sentinel.
float32x32	A primitive scalar or vector data type, including the void sentinel.
float64x32	A primitive scalar or vector data type, including the void sentinel.
float16x64	A primitive scalar or vector data type, including the void sentinel.
float32x64	A primitive scalar or vector data type, including the void sentinel.
float64x64	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x2	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x4	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x8	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x16	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x32	A primitive scalar or vector data type, including the void sentinel.
float8_e3m4x64	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x2	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x4	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x8	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x16	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x32	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3x64	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuz	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx2	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx4	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx8	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx16	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx32	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3b11fnuzx64	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fn	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx2	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx4	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx8	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx16	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx32	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnx64	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuz	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx2	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx4	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx8	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx16	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx32	A primitive scalar or vector data type, including the void sentinel.
float8_e4m3fnuzx64	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x2	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x4	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x8	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x16	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x32	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2x64	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuz	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx2	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx4	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx8	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx16	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx32	A primitive scalar or vector data type, including the void sentinel.
float8_e5m2fnuzx64	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnu	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux2	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux4	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux8	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux16	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux32	A primitive scalar or vector data type, including the void sentinel.
float8_e8m0fnux64	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fn	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx2	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx4	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx8	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx16	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx32	A primitive scalar or vector data type, including the void sentinel.
float6_e2m3fnx64	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fn	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx2	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx4	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx8	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx16	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx32	A primitive scalar or vector data type, including the void sentinel.
float6_e3m2fnx64	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fn	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx2	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx4	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx8	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx16	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx32	A primitive scalar or vector data type, including the void sentinel.
float4_e2m1fnx64	A primitive scalar or vector data type, including the void sentinel.
f16	A primitive scalar or vector data type, including the void sentinel.
f32	A primitive scalar or vector data type, including the void sentinel.
f64	A primitive scalar or vector data type, including the void sentinel.
bf16	A primitive scalar or vector data type, including the void sentinel.
i8	A primitive scalar or vector data type, including the void sentinel.
i16	A primitive scalar or vector data type, including the void sentinel.
i32	A primitive scalar or vector data type, including the void sentinel.
i64	A primitive scalar or vector data type, including the void sentinel.
u8	A primitive scalar or vector data type, including the void sentinel.
u16	A primitive scalar or vector data type, including the void sentinel.
u32	A primitive scalar or vector data type, including the void sentinel.
u64	A primitive scalar or vector data type, including the void sentinel.
handle([dtype, storage_scope])	Create a TIR var that represents a pointer.
void	A primitive scalar or vector data type, including the void sentinel.
ptr(dtype[, storage_scope])	The pointer declaration function.
index_map(mapping, *[, inverse_index_map, ...])	Create a TIR Index mapping
target(target_config[, host])	Create a target
meta_class(cls)	Decorator for utility classes used inside @T.function.
bool	A primitive scalar or vector data type, including the void sentinel.
call_kernel(kernel, launch_args, *args, **kwargs)	Call an external kernel.
cast(value[, dtype, ty, loc])	Cast an expression to the requested data type.
Let(expr, where)	Create a Let expression binding
min(a, b[, loc])	Compute the minimum value of two expressions.
max(a, b[, loc])	Compute the maximum value of two expressions.
register_script_namespace(name, namespace)	Bind an explicit construction namespace to T and its builder aliases.
expose_script_op(name)	Select a completed Op and publish it to declared API destinations.
expose_script_type(type_key)	Select a completed Type and publish it to declared API destinations.
abs(x[, loc])	Get absolute value of the input element-wise.
acos(x, *[, ty, loc])	Take acos of input x.
acosh(x, *[, ty, loc])	Take acos of input x.
address_of(obj[, loc, ty])	Returns the address of a tensor element or addressable variable.
atomic_add(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.atomic_add.
call_spirv_pure_glsl450(*args[, attrs, ...])	Construct a call to tirx.call_spirv_pure_glsl450.
device_scope(*args[, attrs, ty_args, loc, ty])	Internal host/device splitting boundary.
prefetch(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.prefetch.
static_handle(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.static_handle.
asin(x, *[, ty, loc])	Take asin of input x.
asinh(x, *[, ty, loc])	Take asinh of input x.
atan(x, *[, ty, loc])	Take atan of input x.
atan2(x1, x2, *[, ty, loc])	Take arctan2(x1, x2).
atanh(x, *[, ty, loc])	Take atanh of input x.
bitwise_and(x, y[, loc])	Take bitwise and of two values
bitwise_not(x[, loc])	Take bitwise not of input value
bitwise_or(x, y[, loc])	Take bitwise or of two values
bitwise_xor(x, y[, loc])	Take bitwise xor of two values
ceil(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.ceil.
clz(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.clz.
copysign(x1, x2, *[, ty, loc])	Change the sign of x1 to that of x2, element-wise.
cos(x, *[, ty, loc])	Take cos of input x.
cosh(x, *[, ty, loc])	Take cosh of input x.
erf(x, *[, ty, loc])	Take gauss error function of the input x.
exp(x, *[, ty, loc])	Take exponential of input x.
exp2(x, *[, ty, loc])	Calculate 2**x
exp10(x, *[, ty, loc])	Calculate 10**x
gpu_thread_filter(var, pred, *[, loc, ty])	Thread-set filter escape hatch.
gpu_active_thread_selector(var, pred[, loc, ty])	Analysis-only active-thread selector.
floor(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.floor.
ceildiv(lhs, rhs[, loc])	Generic ceildiv operator.
floordiv(a, b[, loc])	Compute the floordiv of two expressions.
floormod(a, b[, loc])	Compute the floormod of two expressions.
fmod(x, y, *[, ty, loc])	Return the remainder of x divided by y with the same sign as x.
fma(x, y, z, *[, ty, loc])	Take fused multiply-add of input x, y, z.
hypot(x1, x2, *[, ty, loc])	Equivalent to sqrt(x1**2 + x2**2), element-wise.
if_then_else(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.if_then_else.
infinity(*, ty[, loc])	infinity value of ty
isfinite(x[, loc])	Check if input value is finite.
isinf(x[, loc])	Check if input value is infinite.
isnan(x[, loc, ty])	Check if input value is Nan.
isnullptr(x[, loc, ty])	Check if input value is nullptr.
ldexp(x1, x2, *[, ty, loc])	Returns x1 * (2 ** x2).
likely(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.likely.
log(x, *[, ty, loc])	Take log of input x.
log1p(x, *[, ty, loc])	Take log(x + 1) with respect to input x.
log2(x, *[, ty, loc])	Take log2 of input x.
log10(x, *[, ty, loc])	Take log10 of input x.
max_value(*, ty[, loc])	maximum value of ty
min_value(*, ty[, loc])	minimum value of ty
nearbyint(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.nearbyint.
nextafter(x1, x2, *[, ty, loc])	Return the next floating-point value after x1 towards x2.
popcount(x, *[, ty, loc])	Count the number of set bits in input x.
pow(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.pow.
round(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.round.
rsqrt(x, *[, ty, loc])	Take reciprocal of square root of input x.
shift_left(x, y[, loc])	Return the result of x left shifted by y bits.
shift_right(x, y[, loc])	Return the result of x right shifted by y bits.
sigmoid(x, *[, ty, loc])	Quick function to get sigmoid
sin(x, *[, ty, loc])	Take sin of input x.
sinh(x, *[, ty, loc])	Take sinh of input x.
sqrt(x, *[, ty, loc])	Take square root of input x.
tan(x, *[, ty, loc])	Take tan of input x.
tanh(x, *[, ty, loc])	Take hyperbolic tanh of input x.
gpu_thread_return(*[, ty, loc])	Return from the current GPU thread without a function value.
trunc(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.trunc.
truncdiv(a, b[, loc])	Compute the truncdiv of two expressions.
truncmod(a, b[, loc])	Compute the truncmod of two expressions.
ptr_byte_offset(data, byte_offset, *, ty[, loc])	Return data + byte_offset with pointer type ty.
throw_last_error(*[, ty, loc])	Throw TVMGetLastError()
stack_alloca(dtype_str, num, *[, ty, loc])	Return new on stack dtype[num]
stack_make_shape(*args[, ty, loc])	Allocate a shape tuple on stack, return the handle
stack_make_dltensor(data, shape, strides, ...)	Allocate a Tensor(DLTensor) on stack, return the handle
call_packed(*args[, loc, ty])	Build expression by call an external packed function.
call_ffi_kernel(*args[, launch])	Call a kernel with a CUDA LaunchConfig, or read a printed low-level call.
call_cpacked(*args[, loc, ty])	Build expression by call an external packed function.
call_packed_lowered(*args[, loc, ty])	Lowered version of call packed.
call_cpacked_lowered(*args[, loc, ty])	Lowered version of call c-packed.
tensor_data_ptr(tensor, *[, ty, loc])	Project a tensor variable's physical pointer.
handle_add_byte_offset(handle, offset, *[, ...])	Add offset to handle
abi_field_set(arr, index, field, value, *[, ...])	Set value in struct field in array
abi_field_get(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.abi_field_get.
gpu_thread_invariant(cond, *[, ty, loc])	Mark condition as thread invariant.
gpu_thread_allreduce(combine, identity, ...)	Perform an all-reduce inside a thread block.
gpu_load_matrix_sync(fragment, m, n, k, ...)	TVM intrinsic for tensor core load operators
gpu_mma_sync(fragment_d, index_d, ...[, ty, loc])	TVM intrinsic for tensor core mma_sync operators
gpu_fill_fragment(fragment, m, n, k, index, ...)	TVM intrinsic for tensor core fill_fragment operators
gpu_store_matrix_sync(fragment, m, n, k, ...)	TVM intrinsic for tensor core store operators
gpu_storage_sync(storage_scope, *[, ty, loc])	Synchronize accesses in the specified storage scope.
cpu_parallel_barrier(*[, ty, loc])	Synchronize all workers in the current CPU parallel launch.
kernel_replace_point(*[, ty, loc])	Mark where a transform should replace generated kernel initialization.
gpu_warp_shuffle(mask, value, warp_id, ...)	Exchange value between threads inside a warp.
gpu_warp_shuffle_up(mask, value, offset, ...)	Copy value from a lane with lower (by offset) index relative to caller.
gpu_warp_shuffle_down(mask, value, offset, ...)	Copy value from a lane with higher (by offset) index relative to caller.
gpu_warp_shuffle_xor(mask, value, lane_mask, ...)	Copy value from a lane with index computed by src_lane_idx ^ lane_mask.
gpu_warp_activemask(*[, ty, loc])	Return a 32-bit mask indicates currently active threads in a calling warp.
assume([cond, ty, loc])	Provide a true statement that can be used for simplifications
assume_aligned(tensor, alignment_bytes, *[, ...])	Assume the tensor's base address is aligned to alignment_bytes.
assume_always(predicate, *[, ty, loc])	Guarantee a predicate throughout its containing scope, including mutations.
undef(*[, ty, loc])	Returns an initialized but arbitrary value
alloc_workspace(device_type, device_id, ...)	Backend function to allocate temporal workspace
free_workspace(device_type, device_id, ptr, *)	Backend function to free temporal workspace.
vscale(*[, ty, loc])	Get the target's vscale value.
ignore_loop_partition(predicate, *[, ty, loc])	Annotate a predicate not be considered as target condition of loop partition.
reinterpret(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.reinterpret.
call_extern(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.call_extern.
call_intrin(func_name, *args, ty[, attrs, loc])	Call an intrinsic with an explicit result type.
call_llvm_intrin(name, *args, ty[, loc])	Build expression by calling a llvm intrinsic function
call_llvm_pure_intrin(name, *args, ty[, loc])	Build expression by calling a pure llvm intrinsic function
call_pure_extern(*args[, attrs, ty_args, ...])	Construct a call to tirx.call_pure_extern.
vector_low(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.vector_low.
vector_high(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.vector_high.
vector_combine(*args[, attrs, ty_args, loc, ty])	Construct a call to tirx.vector_combine.
get_active_lane_mask(*args[, attrs, ...])	Construct a call to tirx.get_active_lane_mask.
masked_load(tensor, index, *args[, ty, loc])	Load vector lanes selected by a predicate mask.
masked_store(tensor, value, index, *args[, ...])	Store vector lanes selected by a predicate mask.
gpu_dp4a(vec1, vec2[, acc])	Dot product of two int8x4 vectors and add an optional accumulator
fabs(*args[, attrs, ty_args, loc, ty])	Construct a call to prim.fabs.
logical_and(*values)	Construct scalar or vector conjunction from eager operands.
logical_or(*values)	Construct scalar or vector disjunction from eager operands.
logical_not(value)	Negate a host or IR value.
select(condition, true_value, false_value)	Construct a scalar conditional or select a host value.
if_then_else_(condition, true_value, false_value)	Implements tvm_tirx.script.ir_builder.parser_protocol.if_then_else_().
and_(*values)	Implements tvm_tirx.script.ir_builder.parser_protocol.and_().
or_(*values)	Implements tvm_tirx.script.ir_builder.parser_protocol.or_().
not_(value)	Implements tvm_tirx.script.ir_builder.parser_protocol.not_().
lt_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.lt_().
le_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.le_().
gt_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.gt_().
ge_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.ge_().
eq_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.eq_().
ne_(lhs, rhs, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.ne_().
add_to_parent(stmt)	Add a statement to the parent frame.
assert_(condition[, message, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.assert_().
break_(*[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.break_().
continue_(*[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.continue_().
else_(*[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.else_().
evaluate(value)	Emit an expression, or pass through a call or statement emission receipt.
grid(*extents[, dtype])	The grid For statement.
if_(condition, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.if_().
region(op, args[, body_params, attrs])	Construct a result-free region with operation-defined body parameters.
return_([value, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.return_().
then_(*[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.then_().
unpack_(value)	Implements tvm_tirx.script.ir_builder.parser_protocol.unpack_().
while_(condition, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.while_().
function([function])	Parse a Python function into a function of the selected IR language variant.
function.builder(**options)	Build a TIRx function with Python body semantics.
function_(*[, private, persistent, decl, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.function_().
arg_(name, annotation, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.arg_().
func_name_(name)	Implements tvm_tirx.script.ir_builder.parser_protocol.func_name_().
func_attr(attrs)	The Function annotation statement.
check_well_formed_(function)	Validate a completed TIRx function.
call_global_var_(function, args)	Implements tvm_tirx.script.ir_builder.parser_protocol.call_global_var_().
bind_([value, ty, name, loc, value_loc, ...])	Implements tvm_tirx.script.ir_builder.parser_protocol.bind_().
decl_mutable_cell_([value, ty, name, loc, ...])	Implements tvm_tirx.script.ir_builder.parser_protocol.decl_mutable_cell_().
set_mutable_cell_(target, value, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.set_mutable_cell_().
setitem_(target, key, value, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.setitem_().
setattr_(target, name, value, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.setattr_().
tensor_store(dest, indices, value)	Emit a tensor store and return a receipt for the stored statement.
for_(iterable, *[, names, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.for_().
range_(*args[, annotations])	Implements tvm_tirx.script.ir_builder.parser_protocol.range_().
serial(start[, stop, annotations, step, ...])	The serial For statement.
parallel(start[, stop, annotations, step, dtype])	The parallel For statement.
vectorized(start[, stop, annotations, step, ...])	The vectorized For statement.
unroll(start[, stop, annotations, step, dtype])	The unrolled For statement.
device_entry(*launch_values[, launch, ...])	Enter a device kernel with independent CUDA launch and compile settings.
device_context(device_type, device_id, *[, ...])	Supply lexical device context for allocation and packed-call lowering.
compute_scope(name, *[, attrs, body_params])	Outline the body as a named CPU compute helper.
parallel_launch(*[, attrs, body_params])	Launch a CPU worker team around parallel loops and team barriers.
launch_thread(thread_tag, extent, *[, ...])	Launch a hardware or virtual thread with a fresh lexical variable.
func_ret(annotation, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.func_ret_type_().
emit(value, *[, loc])	Implements tvm_tirx.script.ir_builder.parser_protocol.emit_().
frame_scope(frames)	Enter multiple IRBuilder frames without deep nesting.
tensor_indices(tensor, index)	Translate logical flat or multidimensional indices for a concrete tensor.
inline([function])	Decorate a helper that constructs IR in its caller's active frames.
macro([function])	Decorate a helper that constructs IR in its caller's active frames.
jit([func, private, check_well_formed, ...])	Decorator: capture the kernel and defer parsing until .specialize().

Alias	Canonical API
make_node	tvm_tirx.ir.make_node
at_	tvm_tirx.script.ir_builder.at_
resolve_global_info_args	tvm_tirx.script.ir_builder.resolve_global_info_args
with_at_group_	tvm_tirx.script.ir_builder.with_at_group_
Lambda	tvm_tirx.script.ir_builder.Lambda
constexpr	tvm_tirx.script.ir_builder.constexpr
dtype	tvm_tirx.script.ir_builder.dtype
dynamic	tvm_tirx.script.ir_builder.dynamic
meta_var	tvm_tirx.script.ir_builder.meta_var
module_attrs	tvm_tirx.script.ir_builder.module_attrs
module_get_attr	tvm_tirx.script.ir_builder.module_get_attr
module_global_infos	tvm_tirx.script.ir_builder.module_global_infos
module_set_attr	tvm_tirx.script.ir_builder.module_set_attr
decl_function	tvm_tirx.script.ir_builder.decl_function
ir_module	tvm_tirx.script.parser.ir_module
module_member_	tvm_tirx.script.ir_builder.module_member_
smem	tvm_tirx.tirx.script.alloc_shared
int8	tvm_tirx.tirx.script.i8
int16	tvm_tirx.tirx.script.i16
int32	tvm_tirx.tirx.script.i32
int64	tvm_tirx.tirx.script.i64
uint8	tvm_tirx.tirx.script.u8
uint16	tvm_tirx.tirx.script.u16
uint32	tvm_tirx.tirx.script.u32
uint64	tvm_tirx.tirx.script.u64
float16	tvm_tirx.tirx.script.f16
float32	tvm_tirx.tirx.script.f32
float64	tvm_tirx.tirx.script.f64
bfloat16	tvm_tirx.tirx.script.bf16
boolean	tvm_tirx.tirx.script.bool
llvm_lookup_intrinsic_id	tvm_tirx.codegen.llvm.lookup_intrinsic_id
func_ret_type_	tvm_tirx.tirx.script.func_ret
emit_	tvm_tirx.tirx.script.emit
pyfunc	tvm_tirx.script.parser.pyfunc

Namespaces

• tvm_tirx.tirx.script.ir_builder
• tvm_tirx.tirx.script.jit
• tvm_tirx.tirx.script.nvshmem
• tvm_tirx.tirx.script.ptx
• tvm_tirx.tirx.script.ptx_legacy