[Mlir-commits] [mlir] [uArch][XeGPU] Add XeGPU uArch definition. (PR #153706)

[Rolf Morel]

================
@@ -23,179 +23,167 @@
 #include <tuple>
 
 #include "mlir/IR/Types.h"
+#include "llvm/ADT/SmallVector.h"
 
 namespace mlir {
 namespace xegpu {
 namespace uArch {
-// Architecture HW component hierarchy to present thread, core, socket ...
-struct uArchHierarchyComponent {
-  std::string name = ""; // optional name of the hierarchy component
-  // no. of lower hierarchy component it contains, e.g., for PVC XeCore it
-  // contains 8 threads, so no_of_component=8
-  uint32_t no_of_component;
-  // Constructor
-  uArchHierarchyComponent(const std::string &name, uint32_t no_of_component)
-      : name(name), no_of_component(no_of_component) {}
-};
 
 // An enum class to represent the scope of an instruction
-enum class InstructionScopeEnum { WorkItem, Subgroup, Workgroup, Cluster };
-
-// A struct to represent basic information about an instruction
-// This struct is used to represent the information about an instruction in the
-// uArch The information includes:
-// - the name of the instruction,
-// - the description of the instruction
-// - the scope of the instruction,
-//
-// The information is represented as strings
-// For example, the information about an instruction can be represented as:
-// Instruction instr = {"dpas", "Dot Product Accumulate Systolic  (DPAS) is a
-// matrix multiply-add operation", "subgroup"};
+enum class InstructionScope { Lane, Subgroup, Workgroup, Cluster };
+enum class InstructionKind {
+  DPAS, // Dot Product Accumulate Systolic (DPAS) is a matrix
+        // multiply-add operation
+  // @TODO: Add more instructions as needed
+};
 
+llvm::StringRef toString(InstructionKind name) {
----------------
rolfmorel wrote:

Nit: would this make more sense as a method?

(Similarly, parseInstructionKind _could_ be a static method.)

Matter of taste though. If there's precedence in the codebase, please ignore this comment.

https://github.com/llvm/llvm-project/pull/153706