MLIR

Multi-Level IR Compiler Framework

'std' Dialect

This dialect provides documentation for operations within the Standard dialect.

Note: This dialect is a collection of operations for several different concepts, and should be split into multiple more-focused dialects accordingly.

Operations 

std.absf (AbsFOp) 

floating point absolute-value operation

The absf operation computes the absolute value. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats.

Example:

// Scalar absolute value.
%a = absf %b : f64

// SIMD vector element-wise absolute value.
%f = absf %g : vector<4xf32>

// Tensor element-wise absolute value.
%x = absf %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.addcf (AddCFOp) 

complex number addition

The addcf operation takes two complex number operands and returns their sum, a single complex number. All operands and result must be of the same type, a complex number with a floating-point element type.

Example:

%a = addcf %b, %c : complex<f32>

Operands: 

OperandDescription
lhscomplex type with floating-point elements
rhscomplex type with floating-point elements

Results: 

ResultDescription
«unnamed»any type

std.addf (AddFOp) 

floating point addition operation

Syntax:

operation ::= ssa-id `=` `std.addf` ssa-use `,` ssa-use `:` type

The addf operation takes two operands and returns one result, each of these is required to be the same type. This type may be a floating point scalar type, a vector whose element type is a floating point type, or a floating point tensor.

Example:

// Scalar addition.
%a = addf %b, %c : f64

// SIMD vector addition, e.g. for Intel SSE.
%f = addf %g, %h : vector<4xf32>

// Tensor addition.
%x = addf %y, %z : tensor<4x?xbf16>

TODO: In the distant future, this will accept optional attributes for fast math, contraction, rounding mode, and other controls.

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.addi (AddIOp) 

integer addition operation

Syntax:

operation ::= ssa-id `=` `std.addi` ssa-use `,` ssa-use `:` type

The addi operation takes two operands and returns one result, each of these is required to be the same type. This type may be an integer scalar type, a vector whose element type is integer, or a tensor of integers. It has no standard attributes.

Example:

// Scalar addition.
%a = addi %b, %c : i64

// SIMD vector element-wise addition, e.g. for Intel SSE.
%f = addi %g, %h : vector<4xi32>

// Tensor element-wise addition.
%x = addi %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.alloc (AllocOp) 

memory allocation operation

The alloc operation allocates a region of memory, as specified by its memref type.

Example:

%0 = alloc() : memref<8x64xf32, 1>

The optional list of dimension operands are bound to the dynamic dimensions specified in its memref type. In the example below, the ssa value ‘%d’ is bound to the second dimension of the memref (which is dynamic).

%0 = alloc(%d) : memref<8x?xf32, 1>

The optional list of symbol operands are bound to the symbols of the memrefs affine map. In the example below, the ssa value ‘%s’ is bound to the symbol ‘s0’ in the affine map specified in the allocs memref type.

%0 = alloc()[%s] : memref<8x64xf32,
                          affine_map<(d0, d1)[s0] -> ((d0 + s0), d1)>, 1>

This operation returns a single ssa value of memref type, which can be used by subsequent load and store operations.

The optional alignment attribute may be specified to ensure that the region of memory that will be indexed is aligned at the specified byte boundary.

%0 = alloc()[%s] {alignment = 8} :
  memref<8x64xf32, affine_map<(d0, d1)[s0] -> ((d0 + s0), d1)>, 1>

Attributes: 

AttributeMLIR TypeDescription
alignment::mlir::IntegerAttr64-bit signless integer attribute whose minimum value is 0

Operands: 

OperandDescription
valueindex

Results: 

ResultDescription
«unnamed»memref of any type values

std.alloca (AllocaOp) 

stack memory allocation operation

The alloca operation allocates memory on the stack, to be automatically released when control transfers back from the region of its closest surrounding operation with an AutomaticAllocationScope trait. The amount of memory allocated is specified by its memref and additional operands. For example:

%0 = alloca() : memref<8x64xf32>

The optional list of dimension operands are bound to the dynamic dimensions specified in its memref type. In the example below, the SSA value ‘%d’ is bound to the second dimension of the memref (which is dynamic).

%0 = alloca(%d) : memref<8x?xf32>

The optional list of symbol operands are bound to the symbols of the memref’s affine map. In the example below, the SSA value ‘%s’ is bound to the symbol ‘s0’ in the affine map specified in the allocs memref type.

%0 = alloca()[%s] : memref<8x64xf32,
                           affine_map<(d0, d1)[s0] -> ((d0 + s0), d1)>>

This operation returns a single SSA value of memref type, which can be used by subsequent load and store operations. An optional alignment attribute, if specified, guarantees alignment at least to that boundary. If not specified, an alignment on any convenient boundary compatible with the type will be chosen.

Attributes: 

AttributeMLIR TypeDescription
alignment::mlir::IntegerAttr64-bit signless integer attribute whose minimum value is 0

Operands: 

OperandDescription
valueindex

Results: 

ResultDescription
«unnamed»memref of any type values

std.and (AndOp) 

integer binary and

Syntax:

operation ::= ssa-id `=` `std.and` ssa-use `,` ssa-use `:` type

The and operation takes two operands and returns one result, each of these is required to be the same type. This type may be an integer scalar type, a vector whose element type is integer, or a tensor of integers. It has no standard attributes.

Example:

// Scalar integer bitwise and.
%a = and %b, %c : i64

// SIMD vector element-wise bitwise integer and.
%f = and %g, %h : vector<4xi32>

// Tensor element-wise bitwise integer and.
%x = and %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.assume_alignment (AssumeAlignmentOp) 

assertion that gives alignment information to the input memref

Syntax:

operation ::= `std.assume_alignment` $memref `,` $alignment attr-dict `:` type($memref)

The assume_alignment operation takes a memref and an integer of alignment value, and internally annotates the buffer with the given alignment. If the buffer isn’t aligned to the given alignment, the behavior is undefined.

This operation doesn’t affect the semantics of a correct program. It’s for optimization only, and the optimization is best-effort.

Attributes: 

AttributeMLIR TypeDescription
alignment::mlir::IntegerAttr32-bit signless integer attribute whose value is positive

Operands: 

OperandDescription
memrefmemref of any type values

std.atomic_rmw (AtomicRMWOp) 

atomic read-modify-write operation

Syntax:

operation ::= `std.atomic_rmw` $kind $value `,` $memref `[` $indices `]` attr-dict `:` `(` type($value) `,`
              type($memref) `)` `->` type($result)

The atomic_rmw operation provides a way to perform a read-modify-write sequence that is free from data races. The kind enumeration specifies the modification to perform. The value operand represents the new value to be applied during the modification. The memref operand represents the buffer that the read and write will be performed against, as accessed by the specified indices. The arity of the indices is the rank of the memref. The result represents the latest value that was stored.

Example:

%x = atomic_rmw "addf" %value, %I[%i] : (f32, memref<10xf32>) -> f32

Attributes: 

AttributeMLIR TypeDescription
kind::mlir::IntegerAttrallowed 64-bit signless integer cases: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10

Operands: 

OperandDescription
valuesignless integer or floating-point
memrefmemref of signless integer or floating-point values
indicesindex

Results: 

ResultDescription
resultsignless integer or floating-point

std.atomic_yield (AtomicYieldOp) 

yield operation for GenericAtomicRMWOp

Syntax:

operation ::= `std.atomic_yield` $result attr-dict `:` type($result)

“atomic_yield” yields an SSA value from a GenericAtomicRMWOp region.

Operands: 

OperandDescription
resultany type

std.br (BranchOp) 

branch operation

Syntax:

operation ::= `std.br` $dest (`(` $destOperands^ `:` type($destOperands) `)`)? attr-dict

The br operation represents a branch operation in a function. The operation takes variable number of operands and produces no results. The operand number and types for each successor must match the arguments of the block successor.

Example:

^bb2:
  %2 = call @someFn()
  br ^bb3(%2 : tensor<*xf32>)
^bb3(%3: tensor<*xf32>):

Operands: 

OperandDescription
destOperandsany type

Successors: 

SuccessorDescription
destany successor

std.call_indirect (CallIndirectOp) 

indirect call operation

Syntax:

operation ::= `std.call_indirect` $callee `(` $operands `)` attr-dict `:` type($callee)

The call_indirect operation represents an indirect call to a value of function type. Functions are first class types in MLIR, and may be passed as arguments and merged together with block arguments. The operands and result types of the call must match the specified function type.

Function values can be created with the constant operation .

Example:

%31 = call_indirect %15(%0, %1)
        : (tensor<16xf32>, tensor<16xf32>) -> tensor<16xf32>

Operands: 

OperandDescription
calleefunction type
operandsany type

Results: 

ResultDescription
resultsany type

std.call (CallOp) 

call operation

Syntax:

operation ::= `std.call` $callee `(` $operands `)` attr-dict `:` functional-type($operands, results)

The call operation represents a direct call to a function that is within the same symbol scope as the call. The operands and result types of the call must match the specified function type. The callee is encoded as a symbol reference attribute named “callee”.

Example:

%2 = call @my_add(%0, %1) : (f32, f32) -> f32

Attributes: 

AttributeMLIR TypeDescription
callee::mlir::FlatSymbolRefAttrflat symbol reference attribute

Operands: 

OperandDescription
operandsany type

Results: 

ResultDescription
«unnamed»any type

std.ceilf (CeilFOp) 

ceiling of the specified value

Syntax:

operation ::= ssa-id `=` `std.ceilf` ssa-use `:` type

The ceilf operation computes the ceiling of a given value. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar ceiling value.
%a = ceilf %b : f64

// SIMD vector element-wise ceiling value.
%f = ceilf %g : vector<4xf32>

// Tensor element-wise ceiling value.
%x = ceilf %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.cmpf (CmpFOp) 

floating-point comparison operation

Syntax:

operation ::= `std.cmpf` $predicate `,` $lhs `,` $rhs attr-dict `:` type($lhs)

The cmpf operation compares its two operands according to the float comparison rules and the predicate specified by the respective attribute. The predicate defines the type of comparison: (un)orderedness, (in)equality and signed less/greater than (or equal to) as well as predicates that are always true or false. The operands must have the same type, and this type must be a float type, or a vector or tensor thereof. The result is an i1, or a vector/tensor thereof having the same shape as the inputs. Unlike cmpi, the operands are always treated as signed. The u prefix indicates unordered comparison, not unsigned comparison, so “une” means unordered or not equal. For the sake of readability by humans, custom assembly form for the operation uses a string-typed attribute for the predicate. The value of this attribute corresponds to lower-cased name of the predicate constant, e.g., “one” means “ordered not equal”. The string representation of the attribute is merely a syntactic sugar and is converted to an integer attribute by the parser.

Example:

%r1 = cmpf "oeq" %0, %1 : f32
%r2 = cmpf "ult" %0, %1 : tensor<42x42xf64>
%r3 = "std.cmpf"(%0, %1) {predicate: 0} : (f8, f8) -> i1

Attributes: 

AttributeMLIR TypeDescription
predicate::mlir::IntegerAttrallowed 64-bit signless integer cases: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
resultbool-like

std.cmpi (CmpIOp) 

integer comparison operation

Syntax:

operation ::= `std.cmpi` $predicate `,` $lhs `,` $rhs attr-dict `:` type($lhs)

The cmpi operation is a generic comparison for integer-like types. Its two arguments can be integers, vectors or tensors thereof as long as their types match. The operation produces an i1 for the former case, a vector or a tensor of i1 with the same shape as inputs in the other cases.

Its first argument is an attribute that defines which type of comparison is performed. The following comparisons are supported:

  • equal (mnemonic: "eq"; integer value: 0)
  • not equal (mnemonic: "ne"; integer value: 1)
  • signed less than (mnemonic: "slt"; integer value: 2)
  • signed less than or equal (mnemonic: "sle"; integer value: 3)
  • signed greater than (mnemonic: "sgt"; integer value: 4)
  • signed greater than or equal (mnemonic: "sge"; integer value: 5)
  • unsigned less than (mnemonic: "ult"; integer value: 6)
  • unsigned less than or equal (mnemonic: "ule"; integer value: 7)
  • unsigned greater than (mnemonic: "ugt"; integer value: 8)
  • unsigned greater than or equal (mnemonic: "uge"; integer value: 9)

The result is 1 if the comparison is true and 0 otherwise. For vector or tensor operands, the comparison is performed elementwise and the element of the result indicates whether the comparison is true for the operand elements with the same indices as those of the result.

Note: while the custom assembly form uses strings, the actual underlying attribute has integer type (or rather enum class in C++ code) as seen from the generic assembly form. String literals are used to improve readability of the IR by humans.

This operation only applies to integer-like operands, but not floats. The main reason being that comparison operations have diverging sets of attributes: integers require sign specification while floats require various floating point-related particularities, e.g., -ffast-math behavior, IEEE754 compliance, etc ( rationale ). The type of comparison is specified as attribute to avoid introducing ten similar operations, taking into account that they are often implemented using the same operation downstream ( rationale ). The separation between signed and unsigned order comparisons is necessary because of integers being signless. The comparison operation must know how to interpret values with the foremost bit being set: negatives in two’s complement or large positives ( rationale ).

Example:

// Custom form of scalar "signed less than" comparison.
%x = cmpi "slt", %lhs, %rhs : i32

// Generic form of the same operation.
%x = "std.cmpi"(%lhs, %rhs) {predicate = 2 : i64} : (i32, i32) -> i1

// Custom form of vector equality comparison.
%x = cmpi "eq", %lhs, %rhs : vector<4xi64>

// Generic form of the same operation.
%x = "std.cmpi"(%lhs, %rhs) {predicate = 0 : i64}
    : (vector<4xi64>, vector<4xi64>) -> vector<4xi1>

Attributes: 

AttributeMLIR TypeDescription
predicate::mlir::IntegerAttrallowed 64-bit signless integer cases: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
resultbool-like

std.cond_br (CondBranchOp) 

conditional branch operation

Syntax:

operation ::= `std.cond_br` $condition `,`
              $trueDest (`(` $trueDestOperands^ `:` type($trueDestOperands) `)`)? `,`
              $falseDest (`(` $falseDestOperands^ `:` type($falseDestOperands) `)`)?
              attr-dict

The cond_br terminator operation represents a conditional branch on a boolean (1-bit integer) value. If the bit is set, then the first destination is jumped to; if it is false, the second destination is chosen. The count and types of operands must align with the arguments in the corresponding target blocks.

The MLIR conditional branch operation is not allowed to target the entry block for a region. The two destinations of the conditional branch operation are allowed to be the same.

The following example illustrates a function with a conditional branch operation that targets the same block.

Example:

func @select(%a: i32, %b: i32, %flag: i1) -> i32 {
  // Both targets are the same, operands differ
  cond_br %flag, ^bb1(%a : i32), ^bb1(%b : i32)

^bb1(%x : i32) :
  return %x : i32
}

Operands: 

OperandDescription
condition1-bit signless integer
trueDestOperandsany type
falseDestOperandsany type

Successors: 

SuccessorDescription
trueDestany successor
falseDestany successor

std.constant (ConstantOp) 

constant

Syntax:

operation ::= ssa-id `=` `std.constant` attribute-value `:` type

The constant operation produces an SSA value equal to some constant specified by an attribute. This is the way that MLIR uses to form simple integer and floating point constants, as well as more exotic things like references to functions and tensor/vector constants.

Example:

// Integer constant
%1 = constant 42 : i32

// Reference to function @myfn.
%3 = constant @myfn : (tensor<16xf32>, f32) -> tensor<16xf32>

// Equivalent generic forms
%1 = "std.constant"() {value = 42 : i32} : () -> i32
%3 = "std.constant"() {value = @myfn}
   : () -> ((tensor<16xf32>, f32) -> tensor<16xf32>)

MLIR does not allow direct references to functions in SSA operands because the compiler is multithreaded, and disallowing SSA values to directly reference a function simplifies this ( rationale ).

Attributes: 

AttributeMLIR TypeDescription
value::mlir::Attributeany attribute

Results: 

ResultDescription
«unnamed»any type

std.copysign (CopySignOp) 

A copysign operation

Syntax:

operation ::= ssa-id `=` `std.copysign` ssa-use `:` type

The copysign returns a value with the magnitude of the first operand and the sign of the second operand. It takes two operands and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar copysign value.
%a = copysign %b %c : f64

// SIMD vector element-wise copysign value.
%f = copysign %g %h : vector<4xf32>

// Tensor element-wise copysign value.
%x = copysign %y %z : tensor<4x?xf8>

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.cos (CosOp) 

cosine of the specified value

Syntax:

operation ::= ssa-id `=` `std.cos` ssa-use `:` type

The cos operation computes the cosine of a given value. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar cosine value.
%a = cos %b : f64

// SIMD vector element-wise cosine value.
%f = cos %g : vector<4xf32>

// Tensor element-wise cosine value.
%x = cos %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.create_complex (CreateComplexOp) 

creates a complex number

Syntax:

operation ::= `std.create_complex` $real `,` $imaginary attr-dict `:` type($complex)

The create_complex operation creates a complex number from two floating-point operands, the real and the imaginary part.

Example:

%a = create_complex %b, %c : complex<f32>

Operands: 

OperandDescription
realfloating-point
imaginaryfloating-point

Results: 

ResultDescription
complexcomplex type with floating-point elements

std.dealloc (DeallocOp) 

memory deallocation operation

Syntax:

operation ::= `std.dealloc` $memref attr-dict `:` type($memref)

The dealloc operation frees the region of memory referenced by a memref which was originally created by the alloc operation. The dealloc operation should not be called on memrefs which alias an alloc’d memref (e.g. memrefs returned by view operations).

Example:

%0 = alloc() : memref<8x64xf32, (d0, d1) -> (d0, d1), 1>
dealloc %0 : memref<8x64xf32, (d0, d1) -> (d0, d1), 1>

Operands: 

OperandDescription
memrefmemref of any type values

std.dim (DimOp) 

dimension index operation

Syntax:

operation ::= `std.dim` attr-dict $memrefOrTensor `,` $index `:` type($memrefOrTensor)

The dim operation takes a memref/tensor and a dimension operand of type index. It returns the size of the requested dimension of the given memref/tensor. If the dimension index is out of bounds the behavior is undefined.

The specified memref or tensor type is that of the first operand.

Example:

// Always returns 4, can be constant folded:
%c0 = constant 0 : index
%x = = dim %A, %c0 : tensor<4 x ? x f32>

// Returns the dynamic dimension of %A.
%c1 = constant 1 : index
%y = dim %A, %c1 : tensor<4 x ? x f32>

// Equivalent generic form:
%x = "std.dim"(%A, %c0) : (tensor<4 x ? x f32>, index) -> index
%y = "std.dim"(%A, %c1) : (tensor<4 x ? x f32>, index) -> index

Operands: 

OperandDescription
memrefOrTensorany tensor or memref type
indexindex

Results: 

ResultDescription
resultindex

std.divf (DivFOp) 

floating point division operation

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.exp2 (Exp2Op) 

base-2 exponential of the specified value

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.exp (ExpOp) 

base-e exponential of the specified value

Syntax:

operation ::= ssa-id `=` `std.exp` ssa-use `:` type

The exp operation takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar natural exponential.
%a = exp %b : f64

// SIMD vector element-wise natural exponential.
%f = exp %g : vector<4xf32>

// Tensor element-wise natural exponential.
%x = exp %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.extract_element (ExtractElementOp) 

element extract operation

Syntax:

operation ::= `std.extract_element` $aggregate `[` $indices `]` attr-dict `:` type($aggregate)

The extract_element op reads a tensor or vector and returns one element from it specified by an index list. The output of the ‘extract_element’ is a new value with the same type as the elements of the tensor or vector. The arity of indices matches the rank of the accessed value (i.e., if a tensor is of rank 3, then 3 indices are required for the extract. The indices should all be of index type.

Example:

%3 = extract_element %v[%1, %2] : vector<4x4xi32>
%4 = extract_element %t[%1, %2] : tensor<4x4xi32>
%5 = extract_element %ut[%1, %2] : tensor<*xi32>

Operands: 

OperandDescription
aggregatevector of any type values or tensor of any type values
indicesindex

Results: 

ResultDescription
resultany type

std.fpext (FPExtOp) 

cast from floating-point to wider floating-point

Cast a floating-point value to a larger floating-point-typed value. The destination type must to be strictly wider than the source type. Only scalars are currently supported.

Operands: 

OperandDescription
inany type

Results: 

ResultDescription
«unnamed»any type

std.fptosi (FPToSIOp) 

cast from floating-point type to integer type

Cast from a value interpreted as floating-point to the nearest (rounding towards zero) signed integer value.

Operands: 

OperandDescription
inany type

Results: 

ResultDescription
«unnamed»any type

std.fptrunc (FPTruncOp) 

cast from floating-point to narrower floating-point

Truncate a floating-point value to a smaller floating-point-typed value. The destination type must be strictly narrower than the source type. If the value cannot be exactly represented, it is rounded using the default rounding mode. Only scalars are currently supported.

Operands: 

OperandDescription
inany type

Results: 

ResultDescription
«unnamed»any type

std.generic_atomic_rmw (GenericAtomicRMWOp) 

atomic read-modify-write operation with a region

The atomic_rmw operation provides a way to perform a read-modify-write sequence that is free from data races. The memref operand represents the buffer that the read and write will be performed against, as accessed by the specified indices. The arity of the indices is the rank of the memref. The result represents the latest value that was stored. The region contains the code for the modification itself. The entry block has a single argument that represents the value stored in memref[indices] before the write is performed. No side-effecting ops are allowed in the body of GenericAtomicRMWOp.

Example:

%x = generic_atomic_rmw %I[%i] : memref<10xf32> {
  ^bb0(%current_value : f32):
    %c1 = constant 1.0 : f32
    %inc = addf %c1, %current_value : f32
    atomic_yield %inc : f32
}

Operands: 

OperandDescription
memrefmemref of signless integer or floating-point values
indicesindex

Results: 

ResultDescription
resultsignless integer or floating-point

std.im (ImOp) 

extracts the imaginary part of a complex number

Syntax:

operation ::= `std.im` $complex attr-dict `:` type($complex)

The im operation takes a single complex number as its operand and extracts the imaginary part as a floating-point value.

Example:

%a = im %b : complex<f32>

Operands: 

OperandDescription
complexcomplex type with floating-point elements

Results: 

ResultDescription
imaginaryfloating-point

std.index_cast (IndexCastOp) 

cast between index and integer types

Casts between integer scalars and ‘index’ scalars. Index is an integer of platform-specific bit width. If casting to a wider integer, the value is sign-extended. If casting to a narrower integer, the value is truncated.

Operands: 

OperandDescription
inany type

Results: 

ResultDescription
«unnamed»any type

std.load (LoadOp) 

load operation

Syntax:

operation ::= `std.load` $memref `[` $indices `]` attr-dict `:` type($memref)

The load op reads an element from a memref specified by an index list. The output of load is a new value with the same type as the elements of the memref. The arity of indices is the rank of the memref (i.e., if the memref loaded from is of rank 3, then 3 indices are required for the load following the memref identifier).

In an affine.if or affine.for body, the indices of a load are restricted to SSA values bound to surrounding loop induction variables, symbols , results of a constant operation , or the result of an affine.apply operation that can in turn take as arguments all of the aforementioned SSA values or the recursively result of such an affine.apply operation.

Example:

%1 = affine.apply affine_map<(d0, d1) -> (3*d0)> (%i, %j)
%2 = affine.apply affine_map<(d0, d1) -> (d1+1)> (%i, %j)
%12 = load %A[%1, %2] : memref<8x?xi32, #layout, memspace0>

// Example of an indirect load (treated as non-affine)
%3 = affine.apply affine_map<(d0) -> (2*d0 + 1)>(%12)
%13 = load %A[%3, %2] : memref<4x?xi32, #layout, memspace0>

Context: The load and store operations are specifically crafted to fully resolve a reference to an element of a memref, and (in affine affine.if and affine.for operations) the compiler can follow use-def chains (e.g. through affine.apply operations) to precisely analyze references at compile-time using polyhedral techniques. This is possible because of the restrictions on dimensions and symbols in these contexts.

Operands: 

OperandDescription
memrefmemref of any type values
indicesindex

Results: 

ResultDescription
resultany type

std.log10 (Log10Op) 

base-10 logarithm of the specified value

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.log2 (Log2Op) 

base-2 logarithm of the specified value

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.log (LogOp) 

base-e logarithm of the specified value

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.memref_cast (MemRefCastOp) 

memref cast operation

Syntax:

operation ::= ssa-id `=` `std.memref_cast` ssa-use `:` type `to` type

The memref_cast operation converts a memref from one type to an equivalent type with a compatible shape. The source and destination types are compatible if:

a. Both are ranked memref types with the same element type, address space, and rank and:

  1. Both have the same layout or both have compatible strided layouts.
  2. The individual sizes (resp. offset and strides in the case of strided memrefs) may convert constant dimensions to dynamic dimensions and vice-versa.

If the cast converts any dimensions from an unknown to a known size, then it acts as an assertion that fails at runtime if the dynamic dimensions disagree with resultant destination size.

Example:

// Assert that the input dynamic shape matches the destination static shape.
%2 = memref_cast %1 : memref<?x?xf32> to memref<4x4xf32>
// Erase static shape information, replacing it with dynamic information.
%3 = memref_cast %1 : memref<4xf32> to memref<?xf32>

// The same holds true for offsets and strides.

// Assert that the input dynamic shape matches the destination static stride.
%4 = memref_cast %1 : memref<12x4xf32, offset:?, strides: [?, ?]> to
                      memref<12x4xf32, offset:5, strides: [4, 1]>
// Erase static offset and stride information, replacing it with
// dynamic information.
%5 = memref_cast %1 : memref<12x4xf32, offset:5, strides: [4, 1]> to
                      memref<12x4xf32, offset:?, strides: [?, ?]>

b. Either or both memref types are unranked with the same element type, and address space.

Example:

Cast to concrete shape.
    %4 = memref_cast %1 : memref<*xf32> to memref<4x?xf32>

Erase rank information.
    %5 = memref_cast %1 : memref<4x?xf32> to memref<*xf32>

Operands: 

OperandDescription
sourceunranked.memref of any type values or memref of any type values

Results: 

ResultDescription
«unnamed»unranked.memref of any type values or memref of any type values

std.mulf (MulFOp) 

floating point multiplication operation

Syntax:

operation ::= ssa-id `=` `std.mulf` ssa-use `,` ssa-use `:` type

The mulf operation takes two operands and returns one result, each of these is required to be the same type. This type may be a floating point scalar type, a vector whose element type is a floating point type, or a floating point tensor.

Example:

// Scalar multiplication.
%a = mulf %b, %c : f64

// SIMD pointwise vector multiplication, e.g. for Intel SSE.
%f = mulf %g, %h : vector<4xf32>

// Tensor pointwise multiplication.
%x = mulf %y, %z : tensor<4x?xbf16>

TODO: In the distant future, this will accept optional attributes for fast math, contraction, rounding mode, and other controls.

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.muli (MulIOp) 

integer multiplication operation

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.negf (NegFOp) 

floating point negation

Syntax:

operation ::= ssa-id `=` `negf` ssa-use `:` type

The negf operation computes the negation of a given value. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar negation value.
%a = negf %b : f64

// SIMD vector element-wise negation value.
%f = negf %g : vector<4xf32>

// Tensor element-wise negation value.
%x = negf %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.or (OrOp) 

integer binary or

Syntax:

operation ::= ssa-id `=` `or` ssa-use `,` ssa-use `:` type

The or operation takes two operands and returns one result, each of these is required to be the same type. This type may be an integer scalar type, a vector whose element type is integer, or a tensor of integers. It has no standard attributes.

Example:

// Scalar integer bitwise or.
%a = or %b, %c : i64

// SIMD vector element-wise bitwise integer or.
%f = or %g, %h : vector<4xi32>

// Tensor element-wise bitwise integer or.
%x = or %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.prefetch (PrefetchOp) 

prefetch operation

The “prefetch” op prefetches data from a memref location described with subscript indices similar to std.load, and with three attributes: a read/write specifier, a locality hint, and a cache type specifier as shown below:

prefetch %0[%i, %j], read, locality<3>, data : memref<400x400xi32>

The read/write specifier is either ‘read’ or ‘write’, the locality hint ranges from locality<0> (no locality) to locality<3> (extremely local keep in cache). The cache type specifier is either ‘data’ or ‘instr’ and specifies whether the prefetch is performed on data cache or on instruction cache.

Attributes: 

AttributeMLIR TypeDescription
isWrite::mlir::BoolAttrbool attribute
localityHint::mlir::IntegerAttr32-bit signless integer attribute whose minimum value is 0 whose maximum value is 3
isDataCache::mlir::BoolAttrbool attribute

Operands: 

OperandDescription
memrefmemref of any type values
indicesindex

std.rank (RankOp) 

rank operation

Syntax:

operation ::= `std.rank` operands attr-dict `:` type(operands)

The rank operation takes a tensor operand and returns its rank.

Example:

%1 = rank %0 : tensor<*xf32>

Operands: 

OperandDescription
«unnamed»tensor of any type values

Results: 

ResultDescription
«unnamed»index

std.re (ReOp) 

extracts the real part of a complex number

Syntax:

operation ::= `std.re` $complex attr-dict `:` type($complex)

The re operation takes a single complex number as its operand and extracts the real part as a floating-point value.

Example:

%a = re %b : complex<f32>

Operands: 

OperandDescription
complexcomplex type with floating-point elements

Results: 

ResultDescription
realfloating-point

std.remf (RemFOp) 

floating point division remainder operation

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.return (ReturnOp) 

return operation

Syntax:

operation ::= `std.return` attr-dict ($operands^ `:` type($operands))?

The return operation represents a return operation within a function. The operation takes variable number of operands and produces no results. The operand number and types must match the signature of the function that contains the operation.

Example:

func @foo() : (i32, f8) {
  ...
  return %0, %1 : i32, f8
}

Operands: 

OperandDescription
operandsany type

std.rsqrt (RsqrtOp) 

reciprocal of sqrt (1 / sqrt of the specified value)

The rsqrt operation computes the reciprocal of the square root. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.sitofp (SIToFPOp) 

cast from integer type to floating-point

Cast from a value interpreted as signed integer to the corresponding floating-point value. If the value cannot be exactly represented, it is rounded using the default rounding mode. Only scalars are currently supported.

Operands: 

OperandDescription
inany type

Results: 

ResultDescription
«unnamed»any type

std.select (SelectOp) 

select operation

The select operation chooses one value based on a binary condition supplied as its first operand. If the value of the first operand is 1, the second operand is chosen, otherwise the third operand is chosen. The second and the third operand must have the same type.

The operation applies to vectors and tensors elementwise given the shape of all operands is identical. The choice is made for each element individually based on the value at the same position as the element in the condition operand. If an i1 is provided as the condition, the entire vector or tensor is chosen.

The select operation combined with cmpi can be used to implement min and max with signed or unsigned comparison semantics.

Example:

// Custom form of scalar selection.
%x = select %cond, %true, %false : i32

// Generic form of the same operation.
%x = "std.select"(%cond, %true, %false) : (i1, i32, i32) -> i32

// Element-wise vector selection.
%vx = std.select %vcond, %vtrue, %vfalse : vector<42xi1>, vector<42xf32>

// Full vector selection.
%vx = std.select %cond, %vtrue, %vfalse : vector<42xf32>

Operands: 

OperandDescription
conditionbool-like
true_valueany type
false_valueany type

Results: 

ResultDescription
resultany type

std.shift_left (ShiftLeftOp) 

integer left-shift

The shift_left operation shifts an integer value to the left by a variable amount. The low order bits are filled with zeros.

Example:

%1 = constant 5 : i8                       // %1 is 0b00000101
%2 = constant 3 : i8
%3 = shift_left %1, %2 : (i8, i8) -> i8    // %3 is 0b00101000

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.sexti (SignExtendIOp) 

integer sign extension operation

The integer sign extension operation takes an integer input of width M and an integer destination type of width N. The destination bit-width must be larger than the input bit-width (N > M). The top-most (N - M) bits of the output are filled with copies of the most-significant bit of the input.

Example:

%1 = constant 5 : i3            // %1 is 0b101
%2 = sexti %1 : i3 to i6        // %2 is 0b111101
%3 = constant 2 : i3            // %3 is 0b010
%4 = sexti %3 : i3 to i6        // %4 is 0b000010

%5 = sexti %0 : vector<2 x i32> to vector<2 x i64>

Operands: 

OperandDescription
valuesignless-integer-like

Results: 

ResultDescription
«unnamed»signless-integer-like

std.divi_signed (SignedDivIOp) 

signed integer division operation

Syntax:

operation ::= ssa-id `=` `divi_signed` ssa-use `,` ssa-use `:` type

Signed integer division. Rounds towards zero. Treats the leading bit as sign, i.e. 6 / -2 = -3.

Note: the semantics of division by zero or signed division overflow (minimum value divided by -1) is TBD; do NOT assume any specific behavior.

Example:

// Scalar signed integer division.
%a = divis %b, %c : i64

// SIMD vector element-wise division.
%f = divis %g, %h : vector<4xi32>

// Tensor element-wise integer division.
%x = divis %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.remi_signed (SignedRemIOp) 

signed integer division remainder operation

Syntax:

operation ::= ssa-id `=` `std.remi_signed` ssa-use `,` ssa-use `:` type

Signed integer division remainder. Treats the leading bit as sign, i.e. 6 % -2 = 0.

Note: the semantics of division by zero is TBD; do NOT assume any specific behavior.

Example:

// Scalar signed integer division remainder.
%a = remis %b, %c : i64

// SIMD vector element-wise division remainder.
%f = remis %g, %h : vector<4xi32>

// Tensor element-wise integer division remainder.
%x = remis %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.shift_right_signed (SignedShiftRightOp) 

signed integer right-shift

The shift_right_signed operation shifts an integer value to the right by a variable amount. The integer is interpreted as signed. The high order bits in the output are filled with copies of the most-significant bit of the shifted value (which means that the sign of the value is preserved).

Example:

%1 = constant 160 : i8                             // %1 is 0b10100000
%2 = constant 3 : i8
%3 = shift_right_signed %1, %2 : (i8, i8) -> i8    // %3 is 0b11110100
%4 = constant 96 : i8                              // %4 is 0b01100000
%5 = shift_right_signed %4, %2 : (i8, i8) -> i8    // %5 is 0b00001100

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.sin (SinOp) 

sine of the specified value

Syntax:

operation ::= ssa-id `=` `std.sin` ssa-use `:` type

The sin operation computes the sine of a given value. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar sine value.
%a = sin %b : f64

// SIMD vector element-wise sine value.
%f = sin %g : vector<4xf32>

// Tensor element-wise sine value.
%x = sin %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.splat (SplatOp) 

splat or broadcast operation

Syntax:

operation ::= `std.splat` $input attr-dict `:` type($aggregate)

Broadcast the operand to all elements of the result vector or tensor. The operand has to be of either integer or float type. When the result is a tensor, it has to be statically shaped.

Example:

%s = load %A[%i] : memref<128xf32>
%v = splat %s : vector<4xf32>
%t = splat %s : tensor<8x16xi32>

TODO: This operation is easy to extend to broadcast to dynamically shaped tensors in the same way dynamically shaped memrefs are handled.

// Broadcasts %s to a 2-d dynamically shaped tensor, with %m, %n binding
// to the sizes of the two dynamic dimensions.
%m = "foo"() : () -> (index)
%n = "bar"() : () -> (index)
%t = splat %s [%m, %n] : tensor<?x?xi32>

Operands: 

OperandDescription
inputinteger or float type

Results: 

ResultDescription
aggregatevector of any type values or statically shaped tensor of any type values

std.sqrt (SqrtOp) 

sqrt of the specified value

The sqrt operation computes the square root. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar square root value.
%a = sqrt %b : f64
// SIMD vector element-wise square root value.
%f = sqrt %g : vector<4xf32>
// Tensor element-wise square root value.
%x = sqrt %y : tensor<4x?xf32>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.store (StoreOp) 

store operation

Syntax:

operation ::= `std.store` $value `,` $memref `[` $indices `]` attr-dict `:` type($memref)

Store a value to a memref location given by indices. The value stored should have the same type as the elemental type of the memref. The number of arguments provided within brackets need to match the rank of the memref.

In an affine context, the indices of a store are restricted to SSA values bound to surrounding loop induction variables, symbols , results of a constant operation , or the result of an affine.apply operation that can in turn take as arguments all of the aforementioned SSA values or the recursively result of such an affine.apply operation.

Example:

store %100, %A[%1, 1023] : memref<4x?xf32, #layout, memspace0>

Context: The load and store operations are specifically crafted to fully resolve a reference to an element of a memref, and (in polyhedral affine.if and affine.for operations) the compiler can follow use-def chains (e.g. through affine.apply operations) to precisely analyze references at compile-time using polyhedral techniques. This is possible because of the restrictions on dimensions and symbols in these contexts.

Operands: 

OperandDescription
valueany type
memrefmemref of any type values
indicesindex

std.subcf (SubCFOp) 

complex number subtraction

The subcf operation takes two complex number operands and returns their difference, a single complex number. All operands and result must be of the same type, a complex number with a floating-point element type.

Example:

%a = subcf %b, %c : complex<f32>

Operands: 

OperandDescription
lhscomplex type with floating-point elements
rhscomplex type with floating-point elements

Results: 

ResultDescription
«unnamed»any type

std.subf (SubFOp) 

floating point subtraction operation

Operands: 

OperandDescription
lhsfloating-point-like
rhsfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.subi (SubIOp) 

integer subtraction operation

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.subview (SubViewOp) 

memref subview operation

The “subview” operation converts a memref type to another memref type which represents a reduced-size view of the original memref as specified by the operation’s offsets, sizes and strides arguments.

The SubView operation supports the following arguments: *) Memref: the “base” memref on which to create a “view” memref. *) Offsets: memref-rank number of dynamic offsets or static integer attributes into the “base” memref at which to create the “view” memref. *) Sizes: memref-rank number of dynamic sizes or static integer attributes which specify the sizes of the result “view” memref type. *) Strides: memref-rank number of dynamic strides or static integer attributes multiplicatively to the base memref strides in each dimension.

Example 1:

%0 = alloc() : memref<64x4xf32, (d0, d1) -> (d0 * 4 + d1)>

// Create a sub-view of "base" memref '%0' with offset arguments '%c0',
// dynamic sizes for each dimension, and stride arguments '%c1'.
%1 = subview %0[%c0, %c0][%size0, %size1][%c1, %c1]
  : memref<64x4xf32, (d0, d1) -> (d0 * 4 + d1) > to
    memref<?x?xf32, (d0, d1)[s0, s1] -> (d0 * s1 + d1 + s0)>

Example 2:

%0 = alloc() : memref<8x16x4xf32, (d0, d1, d1) -> (d0 * 64 + d1 * 4 + d2)>

// Create a sub-view of "base" memref '%0' with dynamic offsets, sizes,
// and strides.
// Note that dynamic offsets are represented by the linearized dynamic
// offset symbol 's0' in the subview memref layout map, and that the
// dynamic strides operands, after being applied to the base memref
// strides in each dimension, are represented in the view memref layout
// map as symbols 's1', 's2' and 's3'.
%1 = subview %0[%i, %j, %k][%size0, %size1, %size2][%x, %y, %z]
  : memref<8x16x4xf32, (d0, d1, d2) -> (d0 * 64 + d1 * 4 + d2)> to
    memref<?x?x?xf32,
      (d0, d1, d2)[s0, s1, s2, s3] -> (d0 * s1 + d1 * s2 + d2 * s3 + s0)>

Example 3:

%0 = alloc() : memref<8x16x4xf32, (d0, d1, d1) -> (d0 * 64 + d1 * 4 + d2)>

// Subview with constant offsets, sizes and strides.
%1 = subview %0[0, 2, 0][4, 4, 4][64, 4, 1]
  : memref<8x16x4xf32, (d0, d1, d2) -> (d0 * 64 + d1 * 4 + d2)> to
    memref<4x4x4xf32, (d0, d1, d2) -> (d0 * 64 + d1 * 4 + d2 + 8)>

Example 4:

%0 = alloc(%arg0, %arg1) : memref<?x?xf32>

// Subview with constant size, but dynamic offsets and
// strides. The resulting memref has a static shape, but if the
// base memref has an affine map to describe the layout, the result
// memref also uses an affine map to describe the layout. The
// strides of the result memref is computed as follows:
//
// Let #map1 represents the layout of the base memref, and #map2
// represents the layout of the result memref. A #mapsubview can be
// constructed to map an index from the result memref to the base
// memref (note that the description below uses more convenient
// naming for symbols, while in affine maps, symbols are
// represented as unsigned numbers that identify that symbol in the
// given affine map.
//
// #mapsubview = (d0, d1)[o0, o1, t0, t1] -> (d0 * t0 + o0, d1 * t1 + o1)
//
// where, o0, o1, ... are offsets, and t0, t1, ... are strides. Then,
//
// #map2 = #map1.compose(#mapsubview)
//
// If the layout map is represented as
//
// #map1 = (d0, d1)[s0, s1, s2] -> (d0 * s1 + d1 * s2 + s0)
//
// then,
//
// #map2 = (d0, d1)[s0, s1, s2, o0, o1, t0, t1] ->
//              (d0 * s1 * t0 + d1 * s2 * t1 + o0 * s1 + o1 * s2 + s0)
//
// Representing this canonically
//
// #map2 = (d0, d1)[r0, r1, r2] -> (d0 * r1 + d1 * r2 + r0)
//
// where, r0 = o0 * s1 + o1 * s2 + s0, r1 = s1 * t0, r2 = s2 * t1.
%1 = subview %0[%i, %j][4, 4][%x, %y] :
  : memref<?x?xf32, (d0, d1)[s0, s1, s2] -> (d0 * s1 + d1 * s2 + s0)> to
    memref<4x4xf32, (d0, d1)[r0, r1, r2] -> (d0 * r1 + d1 * r2 + r0)>

// Note that the subview op does not guarantee that the result
// memref is "inbounds" w.r.t to base memref. It is upto the client
// to ensure that the subview is accessed in a manner that is
// in-bounds.

}

Attributes: 

AttributeMLIR TypeDescription
static_offsets::mlir::ArrayAttr64-bit integer array attribute
static_sizes::mlir::ArrayAttr64-bit integer array attribute
static_strides::mlir::ArrayAttr64-bit integer array attribute

Operands: 

OperandDescription
sourcememref of any type values
offsetsindex
sizesindex
stridesindex

Results: 

ResultDescription
resultmemref of any type values

std.tanh (TanhOp) 

hyperbolic tangent of the specified value

Syntax:

operation ::= ssa-id `=` `std.tanh` ssa-use `:` type

The tanh operation computes the hyperbolic tangent. It takes one operand and returns one result of the same type. This type may be a float scalar type, a vector whose element type is float, or a tensor of floats. It has no standard attributes.

Example:

// Scalar hyperbolic tangent value.
%a = tanh %b : f64

// SIMD vector element-wise hyperbolic tangent value.
%f = tanh %g : vector<4xf32>

// Tensor element-wise hyperbolic tangent value.
%x = tanh %y : tensor<4x?xf8>

Operands: 

OperandDescription
operandfloating-point-like

Results: 

ResultDescription
«unnamed»any type

std.tensor_cast (TensorCastOp) 

tensor cast operation

Syntax:

operation ::= ssa-id `=` `std.tensor_cast` ssa-use `:` type `to` type

Convert a tensor from one type to an equivalent type without changing any data elements. The source and destination types must both be tensor types with the same element type. If both are ranked, then the rank should be the same and static dimensions should match. The operation is invalid if converting to a mismatching constant dimension.

Example:

// Convert from unknown rank to rank 2 with unknown dimension sizes.
%2 = "std.tensor_cast"(%1) : (tensor<*xf32>) -> tensor<?x?xf32>
%2 = tensor_cast %1 : tensor<*xf32> to tensor<?x?xf32>

// Convert to a type with more known dimensions.
%3 = "std.tensor_cast"(%2) : (tensor<?x?xf32>) -> tensor<4x?xf32>

// Discard static dimension and rank information.
%4 = "std.tensor_cast"(%3) : (tensor<4x?xf32>) -> tensor<?x?xf32>
%5 = "std.tensor_cast"(%4) : (tensor<?x?xf32>) -> tensor<*xf32>

Operands: 

OperandDescription
«unnamed»tensor of any type values

Results: 

ResultDescription
«unnamed»tensor of any type values

std.tensor_from_elements (TensorFromElementsOp) 

tensor from elements operation.

Create a 1D tensor from a range of same-type arguments.

Example:

tensor_from_elements(i_1, ..., i_N) :  tensor<Nxindex>

Operands: 

OperandDescription
elementsany type

Results: 

ResultDescription
resulttensor of any type values

std.tensor_load (TensorLoadOp) 

tensor load operation

Syntax:

operation ::= `std.tensor_load` $memref attr-dict `:` type($memref)

Create a tensor from a memref, making an independent copy of the element data. The result value is a tensor whose shape and element type match the memref operand.

Example:

// Produces a value of tensor<4x?xf32> type.
%12 = tensor_load %10 : memref<4x?xf32, #layout, memspace0>

Operands: 

OperandDescription
memrefmemref of any type values

Results: 

ResultDescription
resulttensor of any type values

std.tensor_store (TensorStoreOp) 

tensor store operation

Syntax:

operation ::= `std.tensor_store` $tensor `,` $memref attr-dict `:` type($memref)

Stores the contents of a tensor into a memref. The first operand is a value of tensor type, the second operand is a value of memref type. The shapes and element types of these must match, and are specified by the memref type.

Example:

%9 = dim %8, 1 : tensor<4x?xf32>
%10 = alloc(%9) : memref<4x?xf32, #layout, memspace0>
tensor_store %8, %10 : memref<4x?xf32, #layout, memspace0>

Operands: 

OperandDescription
tensortensor of any type values
memrefmemref of any type values

std.trunci (TruncateIOp) 

integer truncation operation

The integer truncation operation takes an integer input of width M and an integer destination type of width N. The destination bit-width must be smaller than the input bit-width (N < M). The top-most (N - M) bits of the input are discarded.

Example:

  %1 = constant 21 : i5           // %1 is 0b10101
  %2 = trunci %1 : i5 to i4       // %2 is 0b0101
  %3 = trunci %1 : i5 to i3       // %3 is 0b101

  %5 = trunci %0 : vector<2 x i32> to vector<2 x i16>

Operands: 

OperandDescription
valuesignless-integer-like

Results: 

ResultDescription
«unnamed»signless-integer-like

std.divi_unsigned (UnsignedDivIOp) 

unsigned integer division operation

Syntax:

operation ::= ssa-id `=` `std.divi_unsigned` ssa-use `,` ssa-use `:` type

Unsigned integer division. Rounds towards zero. Treats the leading bit as the most significant, i.e. for i16 given two’s complement representation, 6 / -2 = 6 / (2^16 - 2) = 0.

Note: the semantics of division by zero is TBD; do NOT assume any specific behavior.

Example:

// Scalar unsigned integer division.
%a = diviu %b, %c : i64

// SIMD vector element-wise division.
%f = diviu %g, %h : vector<4xi32>

// Tensor element-wise integer division.
%x = diviu %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.remi_unsigned (UnsignedRemIOp) 

unsigned integer division remainder operation

Syntax:

operation ::= ssa-id `=` `std.remi_unsigned` ssa-use `,` ssa-use `:` type

Unsigned integer division remainder. Treats the leading bit as the most significant, i.e. for i16, 6 % -2 = 6 % (2^16 - 2) = 6.

Note: the semantics of division by zero is TBD; do NOT assume any specific behavior.

Example:

// Scalar unsigned integer division remainder.
%a = remiu %b, %c : i64

// SIMD vector element-wise division remainder.
%f = remiu %g, %h : vector<4xi32>

// Tensor element-wise integer division remainder.
%x = remiu %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.shift_right_unsigned (UnsignedShiftRightOp) 

unsigned integer right-shift

The shift_right_unsigned operation shifts an integer value to the right by a variable amount. The integer is interpreted as unsigned. The high order bits are always filled with zeros.

Example:

%1 = constant 160 : i8                               // %1 is 0b10100000
%2 = constant 3 : i8
%3 = shift_right_unsigned %1, %2 : (i8, i8) -> i8    // %3 is 0b00010100

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.view (ViewOp) 

memref view operation

The “view” operation extracts an N-D contiguous memref with empty layout map with arbitrary element type from a 1-D contiguous memref with empty layout map of i8 element type. The ViewOp supports the following arguments: *) A single dynamic byte-shift operand must be specified which represents a a shift of the base 1-D memref pointer from which to create the resulting contiguous memref view with identity layout. *) A dynamic size operand that must be specified for each dynamic dimension in the resulting view memref type.

The “view” operation gives a structured indexing form to a flat 1-D buffer. Unlike “subview” it can perform a type change. The type change behavior requires the op to have special semantics because, e.g. a byte shift of 3 cannot be represented as an offset on f64. For now, a “view” op:

  1. Only takes a contiguous source memref with 0 offset and empty layout.
  2. Must specify a byte_shift operand (in the future, a special integer attribute may be added to support the folded case).
  3. Returns a contiguous memref with 0 offset and empty layout.

Example:

// Allocate a flat 1D/i8 memref.
%0 = alloc() : memref<2048xi8>

// ViewOp with dynamic offset and static sizes.
%1 = view %0[%offset_1024][] : memref<2048xi8> to memref<64x4xf32>

// ViewOp with dynamic offset and two dynamic size.
%2 = view %0[%offset_1024][%size0, %size1] :
  memref<2048xi8> to memref<?x4x?xf32>

Operands: 

OperandDescription
source1D memref of 8-bit signless integer values
byte_shiftindex
sizesindex

Results: 

ResultDescription
«unnamed»memref of any type values

std.xor (XOrOp) 

integer binary xor

The xor operation takes two operands and returns one result, each of these is required to be the same type. This type may be an integer scalar type, a vector whose element type is integer, or a tensor of integers. It has no standard attributes.

Example:

// Scalar integer bitwise xor.
%a = xor %b, %c : i64

// SIMD vector element-wise bitwise integer xor.
%f = xor %g, %h : vector<4xi32>

// Tensor element-wise bitwise integer xor.
%x = xor %y, %z : tensor<4x?xi8>

Operands: 

OperandDescription
lhssignless-integer-like
rhssignless-integer-like

Results: 

ResultDescription
«unnamed»any type

std.zexti (ZeroExtendIOp) 

integer zero extension operation

The integer zero extension operation takes an integer input of width M and an integer destination type of width N. The destination bit-width must be larger than the input bit-width (N > M). The top-most (N - M) bits of the output are filled with zeros.

Example:

  %1 = constant 5 : i3            // %1 is 0b101
  %2 = zexti %1 : i3 to i6        // %2 is 0b000101
  %3 = constant 2 : i3            // %3 is 0b010
  %4 = zexti %3 : i3 to i6        // %4 is 0b000010

  %5 = zexti %0 : vector<2 x i32> to vector<2 x i64>

Operands: 

OperandDescription
valuesignless-integer-like

Results: 

ResultDescription
«unnamed»signless-integer-like

‘dma_start’ operation 

Syntax:

operation ::= `dma_start` ssa-use`[`ssa-use-list`]` `,`
               ssa-use`[`ssa-use-list`]` `,` ssa-use `,`
               ssa-use`[`ssa-use-list`]` (`,` ssa-use `,` ssa-use)?
              `:` memref-type `,` memref-type `,` memref-type

Starts a non-blocking DMA operation that transfers data from a source memref to a destination memref. The operands include the source and destination memref’s each followed by its indices, size of the data transfer in terms of the number of elements (of the elemental type of the memref), a tag memref with its indices, and optionally two additional arguments corresponding to the stride (in terms of number of elements) and the number of elements to transfer per stride. The tag location is used by a dma_wait operation to check for completion. The indices of the source memref, destination memref, and the tag memref have the same restrictions as any load/store operation in an affine context (whenever DMA operations appear in an affine context). See restrictions on dimensions and symbols in affine contexts. This allows powerful static analysis and transformations in the presence of such DMAs including rescheduling, pipelining / overlap with computation, and checking for matching start/end operations. The source and destination memref need not be of the same dimensionality, but need to have the same elemental type.

For example, a dma_start operation that transfers 32 vector elements from a memref %src at location [%i, %j] to memref %dst at [%k, %l] would be specified as shown below.

Example:

%size = constant 32 : index
%tag = alloc() : memref<1 x i32, affine_map<(d0) -> (d0)>, 4>
%idx = constant 0 : index
dma_start %src[%i, %j], %dst[%k, %l], %size, %tag[%idx] :
     memref<40 x 8 x vector<16xf32>, affine_map<(d0, d1) -> (d0, d1)>, 0>,
     memref<2 x 4 x vector<16xf32>, affine_map<(d0, d1) -> (d0, d1)>, 2>,
     memref<1 x i32>, affine_map<(d0) -> (d0)>, 4>

‘dma_wait’ operation 

Syntax:

operation ::= `dma_wait` ssa-use`[`ssa-use-list`]` `,` ssa-use `:` memref-type

Blocks until the completion of a DMA operation associated with the tag element specified with a tag memref and its indices. The operands include the tag memref followed by its indices and the number of elements associated with the DMA being waited on. The indices of the tag memref have the same restrictions as load/store indices.

Example:

dma_wait %tag[%idx], %size : memref<1 x i32, affine_map<(d0) -> (d0)>, 4>