Generation of 32bit value for <4 x i8> looks wrong

[dneto0]

When accumulating the bits, this:

       IntValue = (IntValue << Idx) | Val;      

Should probably be:

       IntValue = (IntValue << 8 * Idx) | Val;      

[dneto0]

This example shows the problem, and more:

kernel void foo(global uchar4* A) {
 *A = (uchar4)(7,8,9,10);
}

Generates code like:

     %uint_7 = OpConstant %uint 7
     %uint_8 = OpConstant %uint 8
     %uint_9 = OpConstant %uint 9
    %uint_10 = OpConstant %uint 10
         %17 = OpConstantComposite %uint %uint_7 %uint_8 %uint_9 %uint_10
     ...
         %24 = OpFunction %void None %8
         %25 = OpLabel
         %26 = OpAccessChain %_ptr_StorageBuffer_uint %23 %uint_0_0 %uint_0_0
               OpStore %26 %17

That OpConstantComposite is just wrong.