Is "rev" provided through PropGCC?

DavidZemon · 2015-04-18 16:05

The Propeller has a couple pretty cool instructions with no equivalent in C. Specifically, I'm looking for the rev instruction. Is that provided through PropGCC? If not, could it be? I know I can use it via inline assembly, but that's quite verbose for a single instruction.

ersmith · 2015-04-18 17:07

SwimDude0614 wrote: »

The Propeller has a couple pretty cool instructions with no equivalent in C. Specifically, I'm looking for the rev instruction. Is that provided through PropGCC? If not, could it be? I know I can use it via inline assembly, but that's quite verbose for a single instruction.

x = __builtin_propeller_rev(y, n);

Pretty much any Propeller instruction which doesn't have an "obvious" way to write it in C (the lock, wait, etc. instructions, for example) will have a __builtin_propeller_xxx function, where xxx is the PASM opcode (all lower case).

The <propeller.h> include file has nice macro wrappers for many of these, but that's optional.

Eric

Ariba · 2015-04-18 17:09

Have you tried this?

   __builtin_propeller_rev(myVar,0);

I have never used this one, but I think all the Assembler instructions have an equivalent __builtin_ function.

Andy

Edit. Okay Eric was a bit faster - but only a bit..

DavidZemon · 2015-04-18 17:36

ersmith wrote: »
x = __builtin_propeller_rev(y, n);
Pretty much any Propeller instruction which doesn't have an "obvious" way to write it in C (the lock, wait, etc. instructions, for example) will have a __builtin_propeller_xxx function, where xxx is the PASM opcode (all lower case).

The <propeller.h> include file has nice macro wrappers for many of these, but that's optional.

Eric

I thought I saw something like that but couldn't find it. Thanks!

Andy, thanks to you too :P

DavidZemon · 2015-04-19 08:13

Is there a __builtin_propeller_andn? It seems to be missing and x &= ~(y) doesn't compile to andn.

--Edit--
Or maybe it does, but not when used with CMM?

Dave Hein · 2015-04-19 08:24

No, there is no andn builtin. Here's the complete list.

propeller/propeller.c: * void __builtin_propeller_clkset(unsigned mode)
propeller/propeller.c: * int __builtin_propeller_cogid(void)
propeller/propeller.c: * int __builtin_propeller_coginit(unsigned mode)
propeller/propeller.c: * void __builtin_propeller_cogstop(int id)
propeller/propeller.c: * unsigned __builtin_propeller_rev(unsigned x, unsigned n)
propeller/propeller.c: * unsigned __builtin_propeller_waitcnt(unsigned c, unsigned d)
propeller/propeller.c: * void __builtin_propeller_waitpeq(unsigned state, unsigned mask)
propeller/propeller.c: * void __builtin_propeller_waitpne(unsigned state, unsigned mask)
propeller/propeller.c: * void __builtin_propeller_waitvid(unsigned colors, unsigned pixels)
propeller/propeller.c: * int __builtin_propeller_locknew(void)
propeller/propeller.c: * void __builtin_propeller_lockret(int x)
propeller/propeller.c: * int __builtin_propeller_lockset(int x)
propeller/propeller.c: * void __builtin_propeller_lockclr(int x)

ersmith · 2015-04-19 15:28

SwimDude0614 wrote: »

Is there a __builtin_propeller_andn? It seems to be missing and x &= ~(y) doesn't compile to andn.

I'm not sure about PropGCC 1.0, but later versions (like alpha v1_9_0) compile:

unsigned test(unsigned x, unsigned y)
{
    return x & ~y;
}

to

	.text
	.global	_test
_test
	andn	r0, r1
	lret

in both LMM and CMM mode.

I have seen cases where andn is not used inside loops, though -- unfortunately the loop optimization code sometimes hoists the "~y" part out of the loop, and ends up using more registers as a result.

Is "rev" provided through PropGCC?

Comments