[webml/MPS] No exception is thrown when an argument passed to setOperandValue() is incompatible with the type expected

[Christywl]

Test Env: Chromium Version : nightly build 65.0.3324.0 (revision f9e1a31) OS : Mac OS X 10.13.4

Expected Result:

• Exception should be thrown when operand has the type of INT32 or TENSOR_INT32 and its value isn’t Int32Arrary
• Exception should be thrown when operand has the type of FLOAT32 or TENSOR_FLOAT32 and its value isn’t Float32Arrary
• Exception should be thrown when operand has the type of UINT32 and its value isn’t Uint32Arrary
• Exception should be thrown when operand has the type of TENSOR_QUANT8_ASYMM and its value isn’t Uint8Arrary
• Exception should be thrown when operand is scale and the length of its value is greater than 1.

Actual Result: No exception is thrown when an argument passed to setOperandValue() is incompatible with the type expected

How to reproduce: Set the following incompatible value:

• Add an operand with a FLOAT32 scalar or TENSOR_FLOAT32 tesnsor, but set its value to a different type(not Float32Arrary):

    an Int8Array data
    an Uint8ClampedArray data
    an Int16Array data
    an Uint16Array data
    a Float64Array data
    a DataView data

• Add an operand with a INT32 scalar or TENSOR_INT32 tensor, but set its value to a different type(not Int32Arrary):

  an Int8Array data
  an Uint8ClampedArray data
  an Int16Array data
  an Uint16Array data
  a Float64Array data
  a DataView data

• Add an operand with a UINT32 scalar, but set its value to a different type(not Uint32Arrary):

  an Int8Array data
  an Uint8ClampedArray data
  an Int16Array data
  an Uint16Array data
  a Float64Array data
  a DataView data

• Add an operand with a TENSOR_QUANT8_ASYMM tensor, but set its value to a different type(not Uint8Arrary):

    an Int8Array data
    an Uint8ClampedArray data
    an Int16Array data
    an Uint16Array data
    a Float64Array data
    a DataView data

• Add an operand for a scalar, but the length of its value is greater than 1

For example:

1. set an Int8Array data for a FLOAT32 scalar

   it('raise error when setting an Int8Array data for a FLOAT32 scalar', function() {
   return nn.createModel(options).then((model)=>{
   model.addOperand({type: nn.FLOAT32});
   assert.throws(() => {
     model.setOperandValue(0, new Int8Array([0]));
   });
   });
   });

2. set a Float32Array data which length > 1 for a FLOAT32 scalar

   it('raise error when setting a Float32Array data which length > 1 for a FLOAT32 scalar', function() {
   return nn.createModel(options).then((model)=>{
   model.addOperand({type: nn.FLOAT32});
   assert.throws(() => {
     model.setOperandValue(0, new Float32Array([0, 0]));
   });
   });
   });

[sunlin-link]

Should add more check for the details of operand based on Android NN API

[sunlin-link]

@Christywl Fixed. Please verify

[Christywl]

@sunlin-link , I tried the chromium nightly build 313c1c2, most negative tests passed except for the following case:

• Exception should be thrown when operand is scale and the length of its value is greater than 1.

Steps: Run https://brucedai.github.io/nt/testm/index-local.html?backend=mps, check the tests in #setOperandValue API:

• raise error when setting a Float32Array data which length > 1 for a FLOAT32 scalar
• raise error when setting an Int32Array data which length > 1 for an INT32 scalar
• raise error when setting an Uint32Array data which length > 1 for an UINT32 scalar

[Christywl]

This issue was fixed on nightly build 8e6c6d1.