-lower-quant-ops ¶

Lower quant.dcast and quant.qcast ops

Lower quantization (quant.qcast) and dequantization (quant.dcast) ops into other core dialects.

The lowering process generates storage type casts in the form of quant.scast ops to act as an interface between the original quantized types of operands and results and their corresponding storage types used in the generated arithmetic computations.

-normalize-quant-types ¶

Normalize generic quantized types to specific quantized types

This pass converts generic quantized types in the quant dialect to more specific types when possible.

The following conversions are performed:

1. Sub-channel to per-axis: If the shape of the scales tensor of sub-channel quantized type has all but one non-one value, it is converted to a per-axis quantized type.

   For example:

   • !quant.uniform<i8:f32:{0:1}, {{2.0}, {3.0}}> -> !quant.uniform<i8:f32:0, {2.0, 3.0}>
   • tensor<?x?x!quant.uniform<i8:f32:{0:1,1:4}, {{2.0}, {3.0}}>> -> tensor<?x?x!quant.uniform<i8:f32:0, {2.0, 3.0}>>

2. Sub-channel to per-tensor: If a sub-channel quantized type has only one scale or zero-point, it is converted to a per-tensor quantized type.

   For example:

   • !quant.uniform<i8:f32:{}, {{2.0}}> -> !quant.uniform<i8:f32, 2.0>
   • tensor<?x?x!quant.uniform<i8:f32:{0:1, 0:4}, {{2.0}}>> -> tensor<?x?x!quant.uniform<i8:f32, 2.0>>

The rationale for these conversions is that the decompositions / handling of more precise quantized types tends to be more efficient than treating everything as subchannel type.

-strip-func-quant-types ¶

Strip quantized types from function headers

Identify occurrences of function arguments using a quantized type and replace them with a new value of the corresponding storage (signless integer) type. For each converted argument, a quant.scast op is introduced at the head of the function’s entry block converting the new integer argument into the original quantized value.