# TritonOps ### `tt.call` (triton::CallOp) _Call operation_ Syntax: ``` operation ::= `tt.call` $callee `(` $operands `)` attr-dict `:` functional-type($operands, results) ``` The `tt.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: ```mlir %2 = tt.call @my_add(%0, %1) : (f32, f32) -> f32 ``` Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `CallOpInterface`, `SymbolUserOpInterface` #### Attributes:
AttributeMLIR TypeDescription
callee::mlir::FlatSymbolRefAttrflat symbol reference attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `operands` | variadic of any type #### Results: | Result | Description | | :----: | ----------- | «unnamed» | variadic of any type ### `tt.func` (triton::FuncOp) _An operation with a name containing a single `SSACFG` region_ Operations within the function cannot implicitly capture values defined outside of the function, i.e. Functions are `IsolatedFromAbove`. All external references must use function arguments or attributes that establish a symbolic connection (e.g. symbols referenced by name via a string attribute like SymbolRefAttr). An external function declaration (used when referring to a function declared in some other module) has no body. While the MLIR textual form provides a nice inline syntax for function arguments, they are internally represented as “block arguments” to the first block in the region. Only dialect attribute names may be specified in the attribute dictionaries for function arguments, results, or the function itself. Example: ```mlir // External function definitions. tt.func @abort() tt.func @scribble(i32, i64, memref) -> f64 // A function that returns its argument twice: tt.func @count(%x: i64) -> (i64, i64) attributes {fruit: "banana"} { return %x, %x: i64, i64 } // A function with an argument attribute tt.func @example_fn_arg(%x: i32 {swift.self = unit}) // A function with a result attribute tt.func @example_fn_result() -> (f64 {dialectName.attrName = 0 : i64}) // A function with an attribute tt.func @example_fn_attr() attributes {dialectName.attrName = false} ``` Traits: `AffineScope`, `AutomaticAllocationScope`, `IsolatedFromAbove`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `CallableOpInterface`, `FunctionOpInterface`, `OpAsmOpInterface`, `Symbol` #### Attributes:
AttributeMLIR TypeDescription
sym_name::mlir::StringAttrstring attribute
function_type::mlir::TypeAttrtype attribute of function type
sym_visibility::mlir::StringAttrstring attribute
arg_attrs::mlir::ArrayAttrArray of dictionary attributes
res_attrs::mlir::ArrayAttrArray of dictionary attributes
### `tt.return` (triton::ReturnOp) _Function return operation_ Syntax: ``` operation ::= `tt.return` attr-dict ($srcs^ `:` type($srcs))? ``` The `tt.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: ```mlir tt.func @foo() : (i32, f8) { ... tt.return %0, %1 : i32, f8 } ``` Traits: `AlwaysSpeculatableImplTrait`, `HasParent`, `ReturnLike`, `TensorSizeTrait`, `Terminator`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)`, `RegionBranchTerminatorOpInterface` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `srcs` | variadic of any type ### `tt.addptr` (triton::AddPtrOp) Syntax: ``` operation ::= `tt.addptr` $ptr `,` $offset attr-dict `:` type($result) `,` type($offset) ``` Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr or ranked tensor of ptr values | `offset` | integer or ranked tensor of integer values #### Results: | Result | Description | | :----: | ----------- | | `result` | ptr or ranked tensor of ptr values ### `tt.advance` (triton::AdvanceOp) _Advance a tensor pointer by offsets_ Syntax: ``` operation ::= `tt.advance` $ptr `,` `[` $offsets `]` attr-dict `:` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr | `offsets` | variadic of 32-bit signless integer #### Results: | Result | Description | | :----: | ----------- | | `result` | ptr ### `tt.assert` (triton::AssertOp) _Device-side assert, as in CUDA for correctness checking_ Syntax: ``` operation ::= `tt.assert` $condition `,` $message `,` $file `,` $func `,` $line attr-dict `:` type($condition) ``` `tt.assert` takes a condition tensor, a message string, a file string, a function string, and a line number. If the condition is false, the message is printed, and the program is aborted. Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
message::mlir::StringAttrstring attribute
file::mlir::StringAttrstring attribute
func::mlir::StringAttrstring attribute
line::mlir::IntegerAttr32-bit signless integer attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `condition` | ranked tensor of floating-point or integer or ptr values ### `tt.atomic_cas` (triton::AtomicCASOp) _Atomic cas_ Syntax: ``` operation ::= `tt.atomic_cas` $sem `,` $scope `,` $ptr `,` $cmp `,` $val attr-dict `:` functional-type(operands, $result) ``` compare $cmp with data $old at location $ptr, if $old == $cmp, store $val to $ptr, else store $old to $ptr, return $old Traits: `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Read on ::mlir::triton::GlobalMemory}`, `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
sem::mlir::triton::MemSemanticAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4{{% markdown %}}Enum cases: * relaxed (`RELAXED`) * acquire (`ACQUIRE`) * release (`RELEASE`) * acq_rel (`ACQUIRE_RELEASE`){{% /markdown %}}
scope::mlir::triton::MemSyncScopeAttr
allowed 32-bit signless integer cases: 1, 2, 3{{% markdown %}}Enum cases: * gpu (`GPU`) * cta (`CTA`) * sys (`SYSTEM`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr or ranked tensor of ptr values | `cmp` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr | `val` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.atomic_rmw` (triton::AtomicRMWOp) _Atomic rmw_ Syntax: ``` operation ::= `tt.atomic_rmw` $atomic_rmw_op `,` $sem `,` $scope `,` $ptr `,` $val (`,` $mask^)? attr-dict `:` functional-type(operands, $result) ``` load data at $ptr, do $rmw_op with $val, and store result to $ptr. return old value at $ptr Traits: `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Read on ::mlir::triton::GlobalMemory}`, `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
atomic_rmw_op::mlir::triton::RMWOpAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10{{% markdown %}}Enum cases: * and (`AND`) * or (`OR`) * xor (`XOR`) * add (`ADD`) * fadd (`FADD`) * max (`MAX`) * min (`MIN`) * umax (`UMAX`) * umin (`UMIN`) * exch (`XCHG`){{% /markdown %}}
sem::mlir::triton::MemSemanticAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4{{% markdown %}}Enum cases: * relaxed (`RELAXED`) * acquire (`ACQUIRE`) * release (`RELEASE`) * acq_rel (`ACQUIRE_RELEASE`){{% /markdown %}}
scope::mlir::triton::MemSyncScopeAttr
allowed 32-bit signless integer cases: 1, 2, 3{{% markdown %}}Enum cases: * gpu (`GPU`) * cta (`CTA`) * sys (`SYSTEM`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr or ranked tensor of ptr values | `val` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr | `mask` | 1-bit signless integer or ranked tensor of 1-bit signless integer values #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.bitcast` (triton::BitcastOp) _Cast between types of the same bitwidth_ Syntax: ``` operation ::= `tt.bitcast` $src attr-dict `:` type($src) `->` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.broadcast` (triton::BroadcastOp) _Broadcast a tensor_ Syntax: ``` operation ::= `tt.broadcast` $src attr-dict `:` type($src) `->` type($result) ``` For a given tensor, broadcast changes one or more dimensions with size 1 to a new size, e.g. tensor<1x32x1xf32> -> tensor<2x32x4xf32>. You cannot change the size of a non-1 dimension. Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultElementType`, `SameOperandsAndResultEncoding`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.cat` (triton::CatOp) _Concatenate 2 tensors_ Syntax: ``` operation ::= `tt.cat` $lhs `,` $rhs attr-dict `:` type($lhs) `->` type($result) ``` Traits: `SameOperandsAndResultElementType`, `SameTypeOperands`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `lhs` | ranked tensor of floating-point or integer or ptr values | `rhs` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.clampf` (triton::ClampFOp) _Clamp operation for floating point types_ Syntax: ``` operation ::= `tt.clampf` $x `,` $min `,` $max `,` `propagateNan` `=` $propagateNan attr-dict `:` type($result) ``` Clamp operation for floating point types. The operation takes three arguments: x, min, and max. It returns a tensor of the same shape as x with its values clamped to the range [min, max]. Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
propagateNan::mlir::triton::PropagateNanAttr
allowed 32-bit signless integer cases: 0, 65535{{% markdown %}}Enum cases: * none (`NONE`) * all (`ALL`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `x` | floating-point or ranked tensor of floating-point values | `min` | floating-point or ranked tensor of floating-point values | `max` | floating-point or ranked tensor of floating-point values #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values ### `tt.dot` (triton::DotOp) _Dot_ Syntax: ``` operation ::= `tt.dot` $a`,` $b`,` $c (`,` `inputPrecision` `=` $inputPrecision^)? attr-dict `:` type($a) `*` type($b) `->` type($d) ``` $d = matrix_multiply($a, $b) + $c. $inputPrecision describes how to exercise the TC when the inputs are f32. It can be one of: tf32, tf32x3, ieee. tf32: use TC with tf32 ops. tf32x3: implement the 3xTF32 trick. For more info see the pass in F32DotTC.cpp ieee: don't use TC, implement dot in software. If the GPU does not have Tensor cores or the inputs are not f32, this flag is ignored. Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
inputPrecision::mlir::triton::InputPrecisionAttr
allowed 32-bit signless integer cases: 0, 1, 2{{% markdown %}}Enum cases: * tf32 (`TF32`) * tf32x3 (`TF32x3`) * ieee (`IEEE`){{% /markdown %}}
maxNumImpreciseAcc::mlir::IntegerAttr32-bit signless integer attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `a` | TensorOrMemDesc instance | `b` | TensorOrMemDesc instance | `c` | ranked tensor of floating-point or integer values #### Results: | Result | Description | | :----: | ----------- | | `d` | ranked tensor of floating-point or integer values ### `tt.elementwise_inline_asm` (triton::ElementwiseInlineAsmOp) _Inline assembly applying an elementwise operation to a group of packed elements._ Syntax: ``` operation ::= `tt.elementwise_inline_asm` $asm_string attr-dict ($args^ `:` type($args))? `->` type($result) ``` Runs an inline asm block to generate one or more tensors. The asm block is given `packed_element` elements at a time. Exactly which elems it receives is unspecified. Traits: `Elementwise`, `SameOperandsAndResultEncoding`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface` #### Attributes:
AttributeMLIR TypeDescription
asm_string::mlir::StringAttrstring attribute
constraints::mlir::StringAttrstring attribute
pure::mlir::BoolAttrbool attribute
packed_element::mlir::IntegerAttr32-bit signless integer attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `args` | variadic of floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | variadic of floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.expand_dims` (triton::ExpandDimsOp) _Expand_dims_ Syntax: ``` operation ::= `tt.expand_dims` $src attr-dict `:` type($src) `->` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultElementType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
axis::mlir::IntegerAttr32-bit signless integer attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.experimental_descriptor_load` (triton::ExperimentalDescriptorLoadOp) _Load from descriptor_ Syntax: ``` operation ::= `tt.experimental_descriptor_load` $desc_ptr `[` $indices `]` oilist( `cacheModifier` `=` $cache | `evictionPolicy` `=` $evict ) attr-dict `:` qualified(type($desc_ptr)) `->` type($result) ``` This operation will be lowered to Nvidia TMA load operation on targets supporting it. `desc_ptr` is a pointer to the TMA descriptor allocated in global memory. The destination tensor type and shape must match the descriptor otherwise the result is undefined. This is an escape hatch and is only there for testing/experimenting. This op will be removed in the future. Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Read on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
cache::mlir::triton::CacheModifierAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4, 5, 6{{% markdown %}}Enum cases: * none (`NONE`) * ca (`CA`) * cg (`CG`) * wb (`WB`) * cs (`CS`) * wt (`WT`){{% /markdown %}}
evict::mlir::triton::EvictionPolicyAttr
allowed 32-bit signless integer cases: 1, 2, 3{{% markdown %}}Enum cases: * evict_normal (`NORMAL`) * evict_first (`EVICT_FIRST`) * evict_last (`EVICT_LAST`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `desc_ptr` | Pointer type (`::mlir::triton::PointerType`) in Triton IR type system | `indices` | variadic of 32-bit signless integer #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.experimental_descriptor_store` (triton::ExperimentalDescriptorStoreOp) _Store value based on descriptor_ Syntax: ``` operation ::= `tt.experimental_descriptor_store` $desc_ptr `[` $indices `]` `,` $src attr-dict `:` qualified(type($desc_ptr)) `,` type($src) ``` This operation will be lowered to Nvidia TMA store operation on targets supporting it. `desc_ptr` is a pointer to the TMA descriptor allocated in global memory. The shape and types of `src` must match the descriptor otherwise the result is undefined. This is an escape hatch and is only there for testing/experimenting. This op will be removed in the future. Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Operands: | Operand | Description | | :-----: | ----------- | | `desc_ptr` | Pointer type (`::mlir::triton::PointerType`) in Triton IR type system | `src` | ranked tensor of floating-point or integer or ptr values | `indices` | variadic of 32-bit signless integer ### `tt.extern_elementwise` (triton::ExternElementwiseOp) Syntax: ``` operation ::= `tt.extern_elementwise` operands attr-dict `:` functional-type(operands, $result) ``` call an external function $symbol implemented in $libpath/$libname with $args return $libpath/$libname:$symbol($args...) Traits: `Elementwise`, `SameOperandsAndResultEncoding`, `SameVariadicOperandSize`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface` #### Attributes:
AttributeMLIR TypeDescription
libname::mlir::StringAttrstring attribute
libpath::mlir::StringAttrstring attribute
symbol::mlir::StringAttrstring attribute
pure::mlir::BoolAttrbool attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `srcs` | variadic of floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.fp_to_fp` (triton::FpToFpOp) _Floating point casting for custom types_ Syntax: ``` operation ::= `tt.fp_to_fp` $src attr-dict (`,` `rounding` `=` $rounding^)? `:` type($src) `->` type($result) ``` Floating point casting for custom types (F8), and non-default rounding modes. F8 <-> FP16, BF16, FP32, FP64 Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
rounding::mlir::triton::RoundingModeAttr
allowed 32-bit signless integer cases: 0, 1{{% markdown %}}Enum cases: * rtz (`RTZ`) * rtne (`RTNE`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of floating-point values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point values ### `tt.get_num_programs` (triton::GetNumProgramsOp) Syntax: ``` operation ::= `tt.get_num_programs` $axis attr-dict `:` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
axis::mlir::triton::ProgramIDDimAttr
allowed 32-bit signless integer cases: 0, 1, 2{{% markdown %}}Enum cases: * x (`X`) * y (`Y`) * z (`Z`){{% /markdown %}}
#### Results: | Result | Description | | :----: | ----------- | | `result` | 32-bit signless integer ### `tt.get_program_id` (triton::GetProgramIdOp) Syntax: ``` operation ::= `tt.get_program_id` $axis attr-dict `:` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
axis::mlir::triton::ProgramIDDimAttr
allowed 32-bit signless integer cases: 0, 1, 2{{% markdown %}}Enum cases: * x (`X`) * y (`Y`) * z (`Z`){{% /markdown %}}
#### Results: | Result | Description | | :----: | ----------- | | `result` | 32-bit signless integer ### `tt.histogram` (triton::HistogramOp) _Return a histgram of the inputs._ Syntax: ``` operation ::= `tt.histogram` $src attr-dict `:` type($src) `->` type($result) ``` Return the histogram of the input tensor. The number of bins is equal to the dimension of the output tensor. Each bins has a width of 1 and bins start at 0. Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of integer values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of integer values ### `tt.int_to_ptr` (triton::IntToPtrOp) _Cast int64 to pointer_ Syntax: ``` operation ::= `tt.int_to_ptr` $src attr-dict `:` type($src) `->` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | 64-bit signless integer or tensor of 64-bit signless integer values #### Results: | Result | Description | | :----: | ----------- | | `result` | ptr or ranked tensor of ptr values ### `tt.join` (triton::JoinOp) _Join two tensors along a new, minor dimension_ Syntax: ``` operation ::= `tt.join` $lhs `,` $rhs attr-dict `:` type($lhs) `->` type($result) ``` For example, if the two input tensors are 4x8xf32, returns a tensor of shape 4x8x2xf32. Because Triton tensors always have a power-of-two number of elements, the two input tensors must have the same shape. Traits: `SameTypeOperands`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `lhs` | ranked tensor of floating-point or integer or ptr values | `rhs` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.load` (triton::LoadOp) _Load from a tensor of pointers or from a tensor pointer_ Syntax: ``` operation ::= `tt.load` $ptr (`,` $mask^)? (`,` $other^)? oilist( `cacheModifier` `=` $cache | `evictionPolicy` `=` $evict ) attr-dict `:` type($ptr) ``` Traits: `AttrSizedOperandSegments`, `SameLoadStoreOperandsAndResultEncoding`, `SameLoadStoreOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `InferTypeOpInterface`, `MemoryEffectOpInterface` #### Attributes:
AttributeMLIR TypeDescription
boundaryCheck::mlir::DenseI32ArrayAttri32 dense array attribute
padding::mlir::triton::PaddingOptionAttr
allowed 32-bit signless integer cases: 1, 2{{% markdown %}}Enum cases: * zero (`PAD_ZERO`) * nan (`PAD_NAN`){{% /markdown %}}
cache::mlir::triton::CacheModifierAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4, 5, 6{{% markdown %}}Enum cases: * none (`NONE`) * ca (`CA`) * cg (`CG`) * wb (`WB`) * cs (`CS`) * wt (`WT`){{% /markdown %}}
evict::mlir::triton::EvictionPolicyAttr
allowed 32-bit signless integer cases: 1, 2, 3{{% markdown %}}Enum cases: * evict_normal (`NORMAL`) * evict_first (`EVICT_FIRST`) * evict_last (`EVICT_LAST`){{% /markdown %}}
isVolatile::mlir::BoolAttrbool attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr or ranked tensor of ptr values or ptr | `mask` | 1-bit signless integer or ranked tensor of 1-bit signless integer values | `other` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.make_range` (triton::MakeRangeOp) _Make range_ Syntax: ``` operation ::= `tt.make_range` attr-dict `:` type($result) ``` Returns an 1D int32 tensor. Values span from $start to $end (exclusive), with step = 1 Traits: `AlwaysSpeculatableImplTrait`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
start::mlir::IntegerAttr32-bit signless integer attribute
end::mlir::IntegerAttr32-bit signless integer attribute
#### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of integer values ### `tt.make_tensor_ptr` (triton::MakeTensorPtrOp) _Make a tensor pointer type with meta information of the parent tensor and the block specified_ Syntax: ``` operation ::= `tt.make_tensor_ptr` $base `,` `[` $shape `]` `,` `[` $strides `]` `,` `[` $offsets `]` attr-dict `:` type($result) ``` `tt.make_tensor_ptr` takes both meta information of the parent tensor and the block tensor, then it returns a pointer to the block tensor, e.g. returns a type of `tt.ptr>`. Traits: `AlwaysSpeculatableImplTrait`, `SameVariadicOperandSize`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
order::mlir::DenseI32ArrayAttri32 dense array attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `base` | ptr | `shape` | variadic of 64-bit signless integer | `strides` | variadic of 64-bit signless integer | `offsets` | variadic of 32-bit signless integer #### Results: | Result | Description | | :----: | ----------- | | `result` | ptr ### `tt.mulhiui` (triton::MulhiUIOp) _Most significant N bits of the 2N-bit product of two integers_ Syntax: ``` operation ::= `tt.mulhiui` $x `,` $y attr-dict `:` type($x) ``` Most significant N bits of the 2N-bit product of two integers. Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `x` | integer or ranked tensor of integer values | `y` | integer or ranked tensor of integer values #### Results: | Result | Description | | :----: | ----------- | | `result` | integer or ranked tensor of integer values ### `tt.precise_divf` (triton::PreciseDivFOp) _Precise div for floating point types_ Syntax: ``` operation ::= `tt.precise_divf` $x `,` $y attr-dict `:` type($x) ``` Precise div for floating point types. Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `x` | floating-point or ranked tensor of floating-point values | `y` | floating-point or ranked tensor of floating-point values #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values ### `tt.precise_sqrt` (triton::PreciseSqrtOp) _Precise sqrt for floating point types_ Syntax: ``` operation ::= `tt.precise_sqrt` $x attr-dict `:` type($x) ``` Precise sqrt for floating point types. Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `x` | floating-point or ranked tensor of floating-point values #### Results: | Result | Description | | :----: | ----------- | | `result` | floating-point or ranked tensor of floating-point values ### `tt.print` (triton::PrintOp) _Device-side print, as in CUDA for debugging_ Syntax: ``` operation ::= `tt.print` $prefix attr-dict (`:` $args^ `:` type($args))? ``` `tt.print` takes a literal string prefix and an arbitrary number of scalar or tensor arguments that should be printed. format are generated automatically from the arguments. Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
prefix::mlir::StringAttrstring attribute
hex::mlir::BoolAttrbool attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `args` | variadic of floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.ptr_to_int` (triton::PtrToIntOp) _Cast pointer to int64_ Syntax: ``` operation ::= `tt.ptr_to_int` $src attr-dict `:` type($src) `->` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `Elementwise`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ptr or ranked tensor of ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | 64-bit signless integer or tensor of 64-bit signless integer values ### `tt.reduce` (triton::ReduceOp) _Reduction using generic combination algorithm_ Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsEncoding`, `SingleBlock`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
axis::mlir::IntegerAttr32-bit signless integer attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `srcs` | variadic of ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | variadic of floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr ### `tt.reduce.return` (triton::ReduceReturnOp) _Terminator for reduce operator_ Syntax: ``` operation ::= `tt.reduce.return` $result attr-dict `:` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `HasParent`, `ReturnLike`, `TensorSizeTrait`, `Terminator`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)`, `RegionBranchTerminatorOpInterface` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `result` | variadic of any type ### `tt.reshape` (triton::ReshapeOp) _Reinterpret a tensor to a different shape. It may change elements order if the attribute is set._ Syntax: ``` operation ::= `tt.reshape` $src attr-dict `:` type($src) `->` type($result) ``` reinterpret a tensor to a different shape. If allow_reorder is set the compiler is free to change the order of elements to generate more efficient code. If efficient_layout is set, this is a hint that the destination layout should be kept for performance reason. The compiler is still free to change it for better performance. Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultElementType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
allow_reorder::mlir::BoolAttrbool attribute
efficient_layout::mlir::UnitAttrunit attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.scan` (triton::ScanOp) _Associative scan using generic combination algorithm_ Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultEncoding`, `SameOperandsAndResultShape`, `SingleBlock`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
axis::mlir::IntegerAttr32-bit signless integer attribute
reverse::mlir::BoolAttrbool attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `srcs` | variadic of ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `result` | variadic of ranked tensor of floating-point or integer or ptr values ### `tt.scan.return` (triton::ScanReturnOp) _Terminator for scan operator_ Syntax: ``` operation ::= `tt.scan.return` $result attr-dict `:` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `HasParent`, `ReturnLike`, `TensorSizeTrait`, `Terminator`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)`, `RegionBranchTerminatorOpInterface` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `result` | variadic of any type ### `tt.splat` (triton::SplatOp) _Splat_ Syntax: ``` operation ::= `tt.splat` $src attr-dict `:` type($src) `->` type($result) ``` Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultElementType`, `SameOperandsAndResultEncoding`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr #### Results: | Result | Description | | :----: | ----------- | | `result` | ranked tensor of floating-point or integer or ptr values ### `tt.split` (triton::SplitOp) _Splits a tensor into two, along its last dimension_ Syntax: ``` operation ::= `tt.split` $src attr-dict `:` type($src) `->` type($outLHS) ``` The input must be a tensor whose last dimension has size 2. Returns two tensors, src[..., 0] and src[..., 1]. For example, if the input shape is 4x8x2xf32, returns two tensors of shape 4x8xf32. Traits: `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Operands: | Operand | Description | | :-----: | ----------- | | `src` | ranked tensor of floating-point or integer or ptr values #### Results: | Result | Description | | :----: | ----------- | | `outLHS` | ranked tensor of floating-point or integer or ptr values | `outRHS` | ranked tensor of floating-point or integer or ptr values ### `tt.store` (triton::StoreOp) _Store by a tensor of pointers or by a tensor pointer_ Syntax: ``` operation ::= `tt.store` $ptr `,` $value (`,` $mask^)? oilist(`cacheModifier` `=` $cache | `evictionPolicy` `=` $evict) attr-dict `:` type($ptr) ``` Traits: `SameLoadStoreOperandsEncoding`, `SameLoadStoreOperandsShape`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `MemoryEffectOpInterface (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{MemoryEffects::Write on ::mlir::triton::GlobalMemory}` #### Attributes:
AttributeMLIR TypeDescription
boundaryCheck::mlir::DenseI32ArrayAttri32 dense array attribute
cache::mlir::triton::CacheModifierAttr
allowed 32-bit signless integer cases: 1, 2, 3, 4, 5, 6{{% markdown %}}Enum cases: * none (`NONE`) * ca (`CA`) * cg (`CG`) * wb (`WB`) * cs (`CS`) * wt (`WT`){{% /markdown %}}
evict::mlir::triton::EvictionPolicyAttr
allowed 32-bit signless integer cases: 1, 2, 3{{% markdown %}}Enum cases: * evict_normal (`NORMAL`) * evict_first (`EVICT_FIRST`) * evict_last (`EVICT_LAST`){{% /markdown %}}
#### Operands: | Operand | Description | | :-----: | ----------- | | `ptr` | ptr or ranked tensor of ptr values or ptr | `value` | floating-point or ranked tensor of floating-point values or integer or ranked tensor of integer values or ptr or ranked tensor of ptr values or ptr | `mask` | 1-bit signless integer or ranked tensor of 1-bit signless integer values ### `tt.trans` (triton::TransOp) _Rearrange the dimensions of a tensor_ Syntax: ``` operation ::= `tt.trans` $src attr-dict `:` type($src) `->` type($result) ``` For example, given a tensor x with shape [1,2,4], transpose(x) with order=[2,0,1] rearranges the tensor to have shape [4,1,2]. Although this op is called "trans", it implements both tl.trans() and tl.permute(). ("permute" might be a better name, but it's called "trans" because originally it only supported 2D tensors.) ## Implementation note on encodings: In the TritonGPU dialect (and probably others), an encoding is chosen for this op's output so it's a nop from the perspective of code generation. For example, suppose tensor x has an encoding such that GPU thread [i,j,k] has a register containing element [i,j,k] of the tensor. Now we transpose x with order [2,1,0], i.e. we reverse the order of its dimensions. In TritonGPU, we will choose a layout for the output of the transpose so that GPU thread [i,j,k] has element [k,j,i] of transpose(x). But this is the same element it had before! All we've done is "rename" the element that thread [i,j,k] has. The "real" transpose -- i.e. moving data between GPU threads -- occurs in convertLayout ops that appear before and/or after the operation. We do this so that you can chain multiple data-movement ops (e.g. transpose+reshape+concat) without going to shared memory after each one. Traits: `AlwaysSpeculatableImplTrait`, `SameOperandsAndResultElementType`, `TensorSizeTrait`, `VerifyTensorLayoutsTrait` Interfaces: `ConditionallySpeculatable`, `InferTypeOpInterface`, `NoMemoryEffect (MemoryEffectOpInterface)` Effects: `MemoryEffects::Effect{}` #### Attributes:
AttributeMLIR TypeDescription
order::mlir::DenseI32ArrayAttri32 dense array attribute
#### Operands: | Operand | Description | | :-----: | ----------- | | `src` | TensorOrMemDesc instance #### Results: | Result | Description | | :----: | ----------- | | `result` | TensorOrMemDesc instance