[Mlir-commits] [mlir] Quantile Type and Low FP Support (PR #190321)

[Javed Absar]

================
@@ -547,6 +548,111 @@ class CalibratedQuantizedType
   double getMax() const;
 };
 
+/*Syntax:
+
+    ```
+    quantile-type ::= `!quant.quantile` `<` type `:` type `,` `{` float-list `}`
+   `,` `<` int `,` int `>`? `>`
+    ```
+
+    A quantile type represents a quantile-based floating point encoding, where
+    discrete storage values are totally defined by the floating-point values
+   entries in a quantile lookup table of F8/F16/F32.
----------------
javedabsar1 wrote:

You have a good description and overview in the PR/commit message e.g.
```....
quantile<ui4:f16, {nf4_table}>
quantile<ui4:f16, {custom_table}>
```
..but in here description is bit scant. The 'quantile' type is a bit complex representation  (type-interface) quantile on top of quant, so it will be helpful if you expand on the explanations here. 
e.g. take your example '!quant.quantile<ui4:f16, {-1.0,-0.696,0.0,0.079,1.0}, <-8,7>>'
then you could explain the entires, min,max etc a bit more.

https://github.com/llvm/llvm-project/pull/190321