Addressed comments on placement of default values in the specification

docs/spec.md

@@ -197,7 +197,9 @@ values of type `tensor<T>`).
   `f = (i - zero_point) * scale`, where `scale` and `zero_point` are called
   **quantization parameters**. The `storage_min` and `storage_max`, being
   optional, have default values which should satisfy the associated constraints
-  as laid out in this document.
+  as laid out in this document. The `storage_min` and `storage_max` are optional
+  in the grammar, but have default values of `min_value(storage_type)` and
+  `max_value(storage_type)` respectively.
 
 Quantization can be **per-tensor**, meaning, having one `scale` and/or
 `zero_point` for the entire tensor or can be **per-axis**, meaning, having
@@ -213,33 +215,25 @@ Quantized types satisfy the following constraints:
 * (C1) `num_bits(storage_type) < num_bits(expressed_type)`.
 * (C2) `type(storage_min) = storage_type`.
 * (C3) `type(storage_max) = storage_type`.
-* (C4) If `storage_min` is empty and `d = num_bits(storage_type)`:
-  * `storage_min = is_signed(storage_type) ? -2^(d-1) : 0`.
-* (C5) If `storage_max` is empty:
-  * `storage_max = is_signed(storage_type) ? 2^(d-1) : 2^d - 1`.
-* (C6) If `storage_min` is not empty:
-  * If `is_signed(storage_type)`, `storage_min >= -2^(d-1)`.
-  * If `is_unsigned(storage_type)`, `storage_min >= 0`.
-* (C7) If `storage_max` is not empty:
-  * If `is_signed(storage_type)`, `storage_max <= 2^(d-1)-1`
-  * If `is_unsigned(storage_type)`, `storage_max <= 2^d - 1`.
-* (C8) `storage_max - storage_min > 0`.
-* (C9) For all `i`, `type(scales[i]) = expressed_type`.
-* (C10) For all `i`, `scales[i] > 0`.
-* (C11) For all `i`, `storage_min <= zero_points[i] <= storage_max`.
-* (C12) For all `i`, `type(zero_points[i]) = storage_type`.
-* (C13) `size(scales) = size(zero_points)`.
-* (C14) If `quantization_dimension` is empty, then `size(scales) = 1`.
-* (C15) If `quantization_dimension` is not empty, then
+* (C4) `storage_min >= min_value(storage_type)`.
+* (C5) `storage_max <= max_value(storage_type)`.
+* (C6) `storage_max - storage_min > 0`.
+* (C7) For all `i`, `type(scales[i]) = expressed_type`.
+* (C8) For all `i`, `scales[i] > 0`.
+* (C9) For all `i`, `storage_min <= zero_points[i] <= storage_max`.
+* (C10) For all `i`, `type(zero_points[i]) = storage_type`.
+* (C11) `size(scales) = size(zero_points)`.
+* (C12) If `quantization_dimension` is empty, then `size(scales) = 1`.
+* (C13) If `quantization_dimension` is not empty, then
   `0 <= quantization_dimension`.
 
 Furthermore, tensors of quantized types satisfy the following constraints:
 
 * For per-tensor quantization:
   * No additional constraints.
 * For per-axis quantization:
-  * (C16) `quantization_dimension < size(shape)`.
-  * (C17) `size(scales) = shape[quantization_dimension]`.
+  * (C14) `quantization_dimension < size(shape)`.
+  * (C15) `size(scales) = shape[quantization_dimension]`.
 
 ```ebnf
 FunctionType ::= '(' [ValueType {',' ValueType}] ')' '->' '(' [ValueType {',' ValueType}] ')'