- To solve the slowness of the MemoryBlock Fill() method when the block has a big size I've created a new class for this MemoryOperations that for now contains only the method Fill() with a part implemented in ASM and with some overloads to semplify its usage from managed code. In future in this class will be added other methods to operate fast on memory for example Cmp() and Copy().
- Adding (not passing) test of the Color struct to the BoxingTest kernel
- Removed BoxingTest kernel inside BCL test: it was a duplicate of the BoxingTest kernel
- Restored correct TestKernelSets
- In the CGS Test kernel the last rectangle is now LimeGreen instead of another type of red (more clear that is working correctly)
- Added to X# the generation of the instruction Shuftps
- Added to X# the generation of the instruction MoveUPS
- Modified Guess Demo to wait for a key press to terminate (it closed suddenly in case the number was guessed)
- The VBE IOMemoryBlock has again size of 1920x1200 (the max usable)
- Added CGS Demo
- Code clean up
- Fixed Single.ToString() on special cases (infinities, NaN and 0) and aumented the range of printable values
- Fixed Double.ToString(): it always printed "Double Overrange" for a bug in opcode ldarga
- Fixed opcode ldarga: the displacement of the argument variable was off of 4 bytes
- Fixed opcodes shr, shr_un and shl when the shift was more that 32 bytes, added to BCL relative tests
- Added BLC tests regarding BitConverter and unsafe code
- Moved the meat of the code of Single.ToString() and Double.ToString() to the class StringHelper together with the analogous methods for numer types
- Re-added _floatsignbit to CosmosAssembler.cs so the neg test should not fail anymore
- Removed all code relative to x87 that I had left under #if false
- Clean up
- Added test for Neg OpCode for float and double
- Conv.U8 wrongly changed the sign of double before converting it to ulong
- Div_Un, Add_Ovf and Add_Ovf_Un had code for floating point but they should not have that for ECMA specs
