[Mlir-commits] [mlir] [mlir][AMDGPU] Add PermlaneOp (PR #154345)

[Tim Gymnich]

================
@@ -656,6 +656,60 @@ def AMDGPU_SwizzleBitModeOp : AMDGPU_Op<"swizzle_bitmode",
   }];
 }
 
+def AMDGPU_PermlanePerm : I32EnumAttr<"PermlanePerm",
+    "The possible permutations for a permlane operation",
+    [
+      I32EnumAttrCase<"swap_16",  0>,
+      I32EnumAttrCase<"swap_32",  1>,
+    ]> {
+  let genSpecializedAttr = 0;
+  let cppNamespace = "::mlir::amdgpu";
+}
+
+def AMDGPU_PermlanePermAttr : EnumAttr<AMDGPU_Dialect, AMDGPU_PermlanePerm,
+  "permlane_perm">;
+
+def AMDGPU_PermlaneOp : AMDGPU_Op<"permlane", [Pure, AllTypesMatch<["result", "src"]>]>,
+Arguments<(ins AnyIntegerOrFloatOr1DVector:$src, 
+               AMDGPU_PermlanePermAttr:$kind,
+               DefaultValuedAttr<BoolAttr, "false">:$fetch_inactive,
+               DefaultValuedAttr<BoolAttr, "false">:$bound_ctrl)> {
+  let summary = "AMDGPU permlane op";
+  let description = [{
+    High-level wrapper on `rocdl.permlane.*` variants.
+
+    Supports arbitrary int/float/vector types, which will be repacked to i32 and
+    one or more `rocdl.permlane.*` ops during lowering.
+    The following lane permutations are supported:
+    - Swap the data between odd and even rows of 16 lanes (`swap_16`)
+    - Swap the data between the first 32 lanes and the last 32 lanes (`swap_32`)
+
+    Format example:
+    ```
+    %0 = amdgpu.permlane %src swap_16 : f16
+    %1 = amdgpu.permlane %src swap_32 { fetch_inactive = true, bound_ctrl = true } : f16
+    ```
+
+    Operands:
+    * `$src`: Vector register to permute across lanes
+    * `$kind`: The kind of permutation operation.
+    * `$fetch_inactive`: Optional. Used to dertermine behavior of invalid lanes (disabled thread or out-of-range).
----------------
tgymnich wrote:

The documentation for these flags was taken from the ISA manual. I guess the distinction between invalid and out-of-bounds does not really apply to permlane. So I replaced invalid and out-of-bounds with disabled lane.

https://github.com/llvm/llvm-project/pull/154345