Returns true iff both instructions perform identical operation. For the purposes of Loop transformation, we only consider operations of type tX = tX <

Returns nonzero if types are just different precisions of the same underlying type

Returns nonzero if types are just different precisions of the same underlying type

Attempt to transform a single loop into a for-statement

Check if a set of definitions can be split and does the split. Returns nonzero if a split took place

Build a loopinfo array of all the loops in this shader phase

Do flow control analysis on the instructions and build the define-use and use-define chains

Coalesce the split temps back based on their original temp register. Keep uint/int/float operations separate

The number of interfaces that are defined in this array.

The number of types that implement this interface

Returns true if we can do direct assignment between types (mostly for mediump<->highp floats etc)

Returns the result swizzle operand for an instruction, or NULL if all src operands have swizzles

Returns true if all the usages of this definitions are instructions that deal with floating point data

Returns nonzero if psInst is a sample instruction and the sampler has medium or low precision

Returns true if the given instruction is a non-vectorized int or uint comparison instruction that reads from at least one temp and writes to a temp

TODO: share with Runtime/GfxDevice/InstancingUtilities.h

these are for now metal only (but can be trivially added for other backends if needed) they are useful mostly for diagnostics as interim values are ac

In DX bytecode, all const arrays are vec4's, and all arrays are stuffed to one large array. Luckily, each chunk is always accessed with suboperand plu

Read D3D11_SHADER_TYPE_DESC