//===- XeGPUAttrs.td - XeGPU dialect attributes definition --*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef MLIR_DIALECT_XEGPU_IR_XEGPUATTRS_TD
#define MLIR_DIALECT_XEGPU_IR_XEGPUATTRS_TD
include "mlir/Dialect/XeGPU/IR/XeGPUDialect.td"
include "mlir/IR/AttrTypeBase.td"
include "mlir/IR/EnumAttr.td"
class XeGPUAttr<string name, string attrMnemonic, list<Trait> traits = [],
string baseCppClass = "::mlir::Attribute">
: AttrDef<XeGPU_Dialect, name, traits, baseCppClass> {
let mnemonic = attrMnemonic;
}
class XeGPU_TensorDescAttr<string name, string attrMnemonic, list<Trait> traits = [],
string baseCppClass = "::mlir::Attribute">
: XeGPUAttr<name, attrMnemonic, traits, baseCppClass> {
let assemblyFormat = "`<` struct(params) `>`";
}
def XeGPU_BlockTensorDescAttr: XeGPU_TensorDescAttr<"BlockTensorDesc", "block_tdesc_attr"> {
let summary = [{a composite attribute for `TensorDescType`}];
let description = [{`BlockTensorDesc` (or `block_tdesc_attr`) is a composite
attribute defined for `TensorDescType` for describing following
properties of a `TensorDesc`.
1. `memory_space`: It describes where the data block described by the
TensorDesc is located, `Global` device memory or `Shared` local memory.
It is default to `Global`.
2. `array_length`: It describes how many horizontally consecutive blocks
will be loaded by a hardware load instruction. If the TensorDesc shape
is 8x16, with array_length = 2. The loaded block shape will be acctually
8x32. Its default value is 1.
3. `boundary_check`: It is used to indicates the hardware whether to do
out-of-boundary check. The default value is true.
}];
let parameters = (ins
OptionalParameter<"MemorySpaceAttr">: $memory_space,
OptionalParameter<"IntegerAttr", "1">: $array_length,
OptionalParameter<"BoolAttr", "true">: $boundary_check
);
let builders = [
AttrBuilder<(ins
CArg<"xegpu::MemorySpace", "xegpu::MemorySpace::Global">:$memory_space,
CArg<"int", "1">:$array_length,
CArg<"bool", "true">: $boundary_check
)>
];
}
def XeGPU_ScatterTensorDescAttr: XeGPU_TensorDescAttr<"ScatterTensorDesc", "scatter_tdesc_attr"> {
let summary = [{a composite attribute for `TensorDescType`}];
let description = [{`ScatterTensorDesc` (or `scatter_tdesc_attr`) is a composite
attribute defined for `TensorDescType` for describing following
properties of a `TensorDesc`.
1. `memory_space`: It describes where the data block described by the
TensorDesc is located, `Global` device memory or `Shared` local memory.
It is default to `Global`.
2. `chunk_size`: indicates number of continious elements accessed for each
offset, default is 1. It is used with `scattered` attr only.
}];
let parameters = (ins
OptionalParameter<"MemorySpaceAttr">: $memory_space,
OptionalParameter<"IntegerAttr", "1">: $chunk_size
);
let builders = [
AttrBuilder<(ins
CArg<"xegpu::MemorySpace", "xegpu::MemorySpace::Global">:$memory_space,
CArg<"int", "1">: $chunk_size
)>
];
}
//===----------------------------------------------------------------------===//
// XeGPU Memory Scope Enums.
//===----------------------------------------------------------------------===//
def XeGPU_MemorySpaceGlobal: I32EnumAttrCase<"Global", 0, "global">;
def XeGPU_MemorySpaceShared: I32EnumAttrCase<"SLM", 3, "slm">;
def XeGPU_MemorySpace: I32EnumAttr<"MemorySpace",
"The address space of the memory the tensor descritor is created for",
[XeGPU_MemorySpaceGlobal, XeGPU_MemorySpaceShared]> {
let genSpecializedAttr = 0;
let cppNamespace = "::mlir::xegpu";
}
def XeGPU_MemorySpaceAttr:
EnumAttr<XeGPU_Dialect, XeGPU_MemorySpace, "memory_space"> {
let summary = [{Describe the location of data described by a `TensorDesc`:
Global device memory (`Global`) or Shared local memory (`SLM`).}];
let assemblyFormat = "$value";
}
//===----------------------------------------------------------------------===//
// XeGPU Cache Enums.
//===----------------------------------------------------------------------===//
def XeGPU_CachePolicyCached: I32EnumAttrCase<"CACHED", 0, "cached">; // valid for read and write
def XeGPU_CachePolicyUncached: I32EnumAttrCase<"UNCACHED", 1, "uncached">; // valid for read and write
def XeGPU_CachePolicyStreaming: I32EnumAttrCase<"STREAMING", 2, "streaming">; // valid for read only
def XeGPU_CachePolicyInvalid: I32EnumAttrCase<"READ_INVALIDATE", 3, "read_invalidate">; // valid for read only
def XeGPU_CachePolicyWriteBack: I32EnumAttrCase<"WRITE_BACK", 4, "write_back">; // valid for write only
def XeGPU_CachePolicyWriteThrough: I32EnumAttrCase<"WRITE_THROUGH", 5, "write_through">; // valid for write only
def XeGPU_CachePolicyEnums : I32EnumAttr<"CachePolicy", "Cache policy",
[XeGPU_CachePolicyCached, XeGPU_CachePolicyUncached,
XeGPU_CachePolicyStreaming, XeGPU_CachePolicyInvalid,
XeGPU_CachePolicyWriteBack, XeGPU_CachePolicyWriteThrough]> {
let genSpecializedAttr = 0;
let cppNamespace = "::mlir::xegpu";
}
def XeGPU_CacheHintAttr
: EnumAttr<XeGPU_Dialect, XeGPU_CachePolicyEnums, "cache_hint"> {
let summary = [{Describe the cache settings for prefetch/load/store operators}];
let assemblyFormat = "`<` $value `>`";
}
def XeGPU_FenceScopeWorkgroup: I32EnumAttrCase<"Workgroup", 0, "workgroup">;
def XeGPU_FenceScopeGPU: I32EnumAttrCase<"GPU", 1, "gpu">;
def XeGPU_FenceScope: I32EnumAttr<"FenceScope",
"The enumeration for the scope of fence operation.",
[XeGPU_FenceScopeWorkgroup, XeGPU_FenceScopeGPU]> {
let genSpecializedAttr = 0;
let cppNamespace = "::mlir::xegpu";
}
def XeGPU_FenceScopeAttr:
EnumAttr<XeGPU_Dialect, XeGPU_FenceScope, "fence_scope"> {
let summary = [{Describes the scope of fence.
"workgroup" means that the scope is within each work group.
"gpu" means the scope is across work groups within the gpu.}];
let assemblyFormat = "$value";
}
def XeGPU_SGMapAttr : XeGPUAttr<"SGMap", "sg_map"> {
let summary = [{
Describes the mapping between work item (WI) and the 2D tensor specified by the tensor descriptor.
}];
let description = [{
To distribute the XeGPU operation to work items, the tensor_desc must be specified with the sg_map
attribute at the tensor description creation time.
Within the `sg_map`, `wi_layout` specifies the layout of work items,
describing the mapping of work items to the tensor.
wi_layout[0] x wi_layout[1] must be equal to the total number of work items within a subgroup.
`wi_data` specifies the minimum number of data elements assigned to each work item for a single distribution.
E.g., #xegpu.sg_map<wi_layout = [1, 16], wi_data = [1, 1]>
In this example, the subgroup has 16 work items in wi_layout=[1, 16],
each accessing 1 element as specified by wi_data=[1, 1].
`wi_data[0] * wi_data[1]` can be greater than 1, meaning that each work item operates on multiple elements,
which is eventually lowered to "SIMT-flavor" vector, like SPIR-V vector or llvm vector, or packed to a storage data type.
The multiple elements indicated by `wi_data` can only be from one dimension and must be contiguous in the memory along either dimension.
}];
let parameters = (ins
ArrayRefParameter<"uint32_t">:$wi_layout,
ArrayRefParameter<"uint32_t">:$wi_data);
let builders = [
AttrBuilder<(ins)>
];
let hasCustomAssemblyFormat = 1;
let genVerifyDecl = 1;
}
#endif // MLIR_DIALECT_XEGPU_IR_XEGPUATTRS_TD