[llvm] [Matrix] Use tiled loops automatically for large kernels. (PR #179325)

[Nathan Corbyn via llvm-commits]

================
@@ -68,9 +68,10 @@ static cl::opt<unsigned> TileSize(
     "fuse-matrix-tile-size", cl::init(4), cl::Hidden,
     cl::desc(
         "Tile size for matrix instruction fusion using square-shaped tiles."));
-static cl::opt<bool> TileUseLoops("fuse-matrix-use-loops", cl::init(false),
-                                  cl::Hidden,
-                                  cl::desc("Generate loop nest for tiling."));
+static cl::opt<unsigned>
+    TileLoopsThreshold("fuse-matrix-loops-threshold", cl::init(200), cl::Hidden,
+                       cl::desc("Generate loop nests for tiling when expected "
+                                "number of operations exceeds threshold."));
----------------
cofibrant wrote:

Yeah, fair enough. I'm happy either way. (Though, in fairness, I think the impact of 0 to disable would just be `bool UseLoops = TileLoopsThreshold && ...` and a note that 0 disables in the `cl::desc`.)

https://github.com/llvm/llvm-project/pull/179325