#549 Trivial missing optimisation on Z80

[Alan Cox]

This shows up a fair bit in my code, although longs in general are badly handled

static uint32_t foo;

uint16_t blkoff(void)
{
return foo & 511;
}

Generates a 32bit load into HL/BC and then masks H nicely. It should have generated a 16bit load

Even casting the value before doing the & 511 doesn't get rid of the bogus loads.

[Philipp Klaus Krause]

Ticket moved from /p/sdcc/bugs/2698/

Can't be converted:

• _category: Z80

[Philipp Klaus Krause]

This optimization would be something to look into soon after the 3.7.0 release.

Philipp

[Philipp Klaus Krause]

I've put this onto the TODO list for 3.8.0 in the wiki. I'll also have a look at Coremark for Z80; Coremark is mostly a 32-bit integer benchmark, so it might cover a few similar issues.

Philipp

[Philipp Klaus Krause]

I can't reproduce this in current SDCC, and get:

ld  hl, (_foo)
ld  a, h
and a, #0x01
ld  h, a

Which looks good to me. I tried with 32-bit and 64-bit foo. Do you still see unnecessary loads into registers for this code?
If yes, which SDCC version and options are used?

Philipp

[Philipp Klaus Krause]

Closing, since it can't be reproduced in current SDCC:

Philipp