[Mlir-commits] [mlir] Sub-channel quantized type implementation (PR #120172)

[Kevin Gleason]

================
@@ -31,6 +31,39 @@ def LowerQuantOps : Pass<"lower-quant-ops", "func::FuncOp"> {
   ];
 }
 
+def NormalizeQuantTypes : Pass<"normalize-quant-types"> {
+  let summary = "Normalize generic quantized types to specific quantized types";
+  let description = [{
+    This pass converts generic quantized types in the `quant` dialect to more
+    specific types when possible.
----------------
GleasonK wrote:

Could add a note that the rationale for doing this is that the decompositions / handling of more precise quantized types tends to be more efficient than treating everything as subchannel.

https://github.com/llvm/llvm-project/pull/120172