Removal of Pointless References in Function Args

[mondus]

Fixes #825

Data types fixed

• unsigned int
• int
• bool
• char
• unsigned char
• size_type (unsigned int)
• float
• double
• signed char (we have some use of this)
• EnvironmentManager::size_type
• uint64_t
• int64_t
• uint16_t
• int16_t
• T
• T2 (from use in macro properties)
• cudaStream_t
• size_t

Each items has been searched for using const type & as well as const type& as linting allows either.

Where const references appear in declarations they have been made primitive (pass by value) type argument and const has been removed as it is meaningless. Within definitions const primitive types are allowed as this signals an intention that the argument should not be changed in the function body.

[mondus]

There are actually quite a few places inside curve and message iterator functions where this might save a few registers.

[Robadob]

There are actually quite a few places inside curve and message iterator functions where this might save a few registers.

Much of the device API stuff is marked inline/forceinline. So if it is actually being inlined, the function calls are presumably irrelevant.

[mondus]

Template types have also been modified to be passed by value. This may be detrimental if we decide to support complex types in the future although (as Rob has pointed out) most of the device functions are inlined so copies wouldn't need to be made for function calls. All C++ tests pass Python tests not run due to #939.

[Robadob]

Windows/Release/SEATBELTS=ON tests pass. I need to look through the code changes before I put a review in though.

[Robadob]

Looked through all the changes, added a bunch of suggestions where I think you made mistakes. However, given the length of changes my eyes may have glazed over and missed some.

Doesn't appear you changed any non-const references, which is where I was worried issues may occur.

Additionally you should probably do:
cudaStream_t doesn't need to be a reference, it's merely a typedef for a ptr
size_t too (CUDAScatter.cuh:243)

[ptheywood]

Tests all pased for a debug build under linux (C++ and python). I won't do a full review right now, but will take a look when Rob's suggestions have been resolved to see if any bits still jump out.

---

It's possible to run the python tests as long as you do it one folder at a time, which somehow avoids the gc related issues (usually)

[mondus]

Looked through all the changes, added a bunch of suggestions where I think you made mistakes. However, given the length of changes my eyes may have glazed over and missed some.

Doesn't appear you changed any non-const references, which is where I was worried issues may occur.

Additionally you should probably do: cudaStream_t doesn't need to be a reference, it's merely a typedef for a ptr size_t too (CUDAScatter.cuh:243)

Additional types added. Tests passing with SEATBELTS.