Mismatch between specification and tests for quantized types

[jtristan]

What happened?

The tests in stablehlo/stablehlo/tests/interpret and stablehlo/stablehlo/testdata use a different syntax than what is described in the specification. Here is a grammar that describes the actual syntax of quantized types in those tests:

QuantizedTensorType ::= 'tensor' '<' Shape QuantizedTensorElementType '>'
QuantizedTensorElementType ::= '!quant.uniform' '<'
                  QuantizationStorageType
                  ['<' QuantizationStorageMin ':' QuantizationStorageMax '>']
                  ':' QuantizationExpressedType
                  [':' QuantizationDimension]
                  ',' QuantizationParameters '>'
QuantizationStorageType ::= IntegerType
QuantizationStorageMin ::= **IntegerLiteral**
QuantizationStorageMax ::= **IntegerLiteral**
QuantizationExpressedType ::= FloatType
QuantizationDimension ::= **IntegerLiteral**
QuantizationParameters ::= QuantizationParameter
                         | '{' QuantizationParameter {',' QuantizationParameter} '}'
QuantizationParameter ::= QuantizationScale **[ ':' QuantizationZeroPoint ]**
QuantizationScale ::= **FloatLiteral**
QuantizationZeroPoint ::= **IntegerLiteral**

where ** and ** emphasize where the differences with the specification are

Steps to reproduce your issue

No response

Version information

No response

[sdasgup3]

Dear @jtristan
I am assuming that you refer to the absence of the non terminals highlighted above. Please note that QuantizationStorageMin, QuantizationStorageMax are optional and If not present inferred from the storage type.
QuantizationDimension is relevant only for per-axis quantization and absent for per-tensor.
QuantizationDimension is optional if the zero-point is zero.

Please let me know if I am missing any discrepancies.

[jtristan]

No, the discrepancies are that the spec defines the grammar of QuantizationStorageMin, QuantizationStorageMax, QuantizationDimension , QuantizationScale , and QuantizationZeroPoint respectively as IntegerConstant , IntegerConstant , IntegerConstant , FloatConstant , IntegerConstant but that it should be IntegerLiteral, IntegerLiteral, IntegerLiteral, FloatLiteral, IntegerLiteral instead. Finally, the spec doesn't capture that QuantizationZeroPoint is optional.

[sdasgup3]

Thanks for pointing those out. Will fix is asap.