Added load/storePolyWord/NativeWord functions to RunCall.

Use TBZ/TBNZ for tag test and branches.

Move code to check for RTS exceptions into the RTS caller for the X86. This…

Combine adjacent pushes, combine stack resets and use post-indexing to include…

Use ldp and stp for adjacent loads and stores where possible.

Begin peep-hole optimisation. Eliminate unnecessary branches and straighten…

Handle allocations of > 4096 which was omitted in earlier change.

Move allocation code and branch tables into pre-assembler "instructions".

Convert ForeignCall to use the pre-assembler layer. Remove ArmSequences and…

Remove temporary functions that converted the pre-assembly code to the assembly…

Return small numbers of multiple results in registers.

Fix another case where the optimiser added a new tuple but didn't set the…

Clean up the conflict test for arguments of long format arbitrary precision…

Support for multiple results from function calls.

Don't attempt to merge cache states for the moment. It seems to be broken.

Fix the optimiser so that it sets the correct argument type for newly…

Use hints to improve the chances that we will be able to retain cache states…

Add a dummy instruction if we're caching to see if this is the problem.

Fix incorrect test.

Use an indexed load in 32-in-64 if the offset is zero and this is the last…

Merge the cache state on multiple paths.

Include the cache registers in the save mask for allocations so that if there…

Change the assertion into a test and don't optimise if the result of the load…

Cache the combination of loading a value from the stack and getting the…

Refine the checking code.

Add basic caching of stack entries and object ids.

Add temporary debugging code to check that the cache register and the value…

When picking a register try to avoid those that are caching values.

Separate allocation of new registers at a "destination" from getting the…

Remove extra code to make sure conflicts are symmetrical.

Improve register hinting by propagating "real hints".

Move conflict state code into the register allocator.

Construct the result list in the conversion function rather than mapping it.

Allocate the registers at the same time as converting the code. This changes…

AllocateRegisters now produces concrete icode and ICodeToArm64Code now handles…

Turn the icode type into a polytype to allow for a version with concrete…

Convert to generating pre-code. Use load/store pairs for exception handling.

Completed conversion of instructions.

More instructions including loops.

Added some extra instructions at the assembly level and completed the pre…

Convert more instructions.

Begin the process of generating pre-assembler code rather than assembly code…

Add pre-assembly functor and signature. Split icode loads and stores for…

Add further operations at the pre-assembler level.

Added extra instructions including floating point.

Sort out ordering of instruction sequences.

Must clear the argument registers from the cache after a CompareByteVector or…

Volatile refs are byte memory so don't need to be initialised in the allocation.

Don't try to print a value with stringOfWord: it may not be a valid address.

Fix initialisation of multiple items.

Optimise various constant operations so that "ref" is now minimal.

Check for exceptions raised in the RTS in the RTS call wrapper code and remove…

Save active registers in the long case of an arbitrary precision operation…

Fix store allocation profiling.

String length words ARE tagged when loaded using LoadStoreUntaggedUnsigned so…

Fix typos in immutable load function names.

X0 is used for the exception packet so has to be reserved in raises and handles.

Must not use X30 for the result of AllocateMemoryFixed/Variable because it…

Fix typo in version string.

Remove unused exception identifiers.

Vector indexing should use load-from-immutable.

Remove a piece of debugging code that includes PolyML.makestring. This seems…

Fix typo that caused Int.quot to fail. Use sign-extending indexed loads/stores…

Remove fall-back to old code-generator.

Implemented floating point operations.

Implement loads and stores to C memory.

Implement mutex creation, locking and unlocking.

Add a "containers" argument to FunctionCall to ensure that containers are not…

When extending the stack update any registers with addresses in the stack.

Implemented block moves and byte comparisons.

Implement LargeWord div and mod.

Fixed a couple of mistakes in the last commit.

Add back=up file for Int missed in last commit.

When optimising short byte equality and short moves use the immutable version…

Fix addition when combining an index with a constant offset.

Split stores to update from stores to initialise.

Implement store of untagged values to set string length. Also implement stores…

Add storeByte/WordInitialise to store without release. Update version to 5.9.1…

Fix fixed int remainder.

Implemented division and remainder.

Implement multiplication.

Implement indexed case and exception handlers.

Implement conversion between LargeWord and Word and Int.

Allow register spilling.

Word.toInt needs to check that the value is small enough to convert to a…

Implement clear-mutable and allocate-byte-memory.

Cannot use the same register for the source and destination in BitFieldInsert.

Need to allocate a register for every argument in BeginFunction even if it…

Separate shifts (ubfm and sbfm) from moving bits into a register.

Merge branch 'master' into ARM64Development

Implemented allocation of arrays and vectors.

Implement shifts for Word and LargeWord.

Implement arbitrary precision addition and subtraction.

Separate boxing/unboxing of large-word values from floating point and include…

Implement LoadUntagged used to get the length of a string.

Implemented MemoryCellLength and MemoryCellFlags.

Implement logical operations for word and LargeWord and comparisons for…

Implement BeginLoop and JumpLoop.

Add a container type to the argument types. This should allow code to return a…

Removed stack closure detection since it isn't used and greatly simplified the…