specials: memory: add read_capable option

[maass-hamburg]

add read_capable option, so we can
better detect when no read is done.

Signed-off-by: Fin Maaß [email protected]

[sbourdeauducq]

What's the context for this?

[maass-hamburg]

What's the context for this?

@sbourdeauducq
first of all it reduces unused code in the generated verilog file

the titanium and topaz from efinix internally have EFX_RAM10 (simple dual port, one port for reading and one port for writing, can be on different clocks) and EFX_DPRAM10 (true dual port, both can read and write at the same time).
Only the EFX_RAM10 supports byte enable/we_granularity, the EFX_DPRAM10 does not and the sythesis tool will automatically use the right type, but will complain and fail if a memory will use a EFX_DPRAM10, that uses byte enable/we_granularity. Because of that in litex FullMemoryWE is used on all memory, that have we_granularity, but it is only necessary for memory, that will use EFX_DPRAM10 and not EFX_RAM10, so i want to detect that and only do the spiting of the memory, where we can't use the EFX_RAM10 with byte enable/we_granularity. For that I need to be able to check if the read part of a port is used. So in the end only one block with 512 x 16 is used, when the we_granularity is 8, instead of two 1024 x 8, when the needed Memory Depth is 512 or less.

[maass-hamburg]

under this condition

((global_granularity % 8 == 0) or (global_granularity % 10 == 0)) and len([port for port in orig.ports if port.dat_r is not None]) <= 1 and len([port for port in orig.ports if port.dat_w is not None]) <= 1

in FullMemoryWE, for efinix Titanium/Topaz fpgas the memory don't need to be split.

• granularity must be dividable by 8 or 10
• only one or no read port
• only one or no write port

[maass-hamburg]

@sbourdeauducq ping