PATCH: R600/SI: Tablegen refactoring
tom at stellard.net
Fri May 16 10:42:02 PDT 2014
On Fri, May 16, 2014 at 10:11:52AM -0700, Matt Arsenault wrote:
> On May 16, 2014, at 8:17 AM, Tom Stellard <tom at stellard.net> wrote:
> > Hi,
> > The attached patches refactor the SI tablegen definitions to enable to
> > allow for some convenient simplifications in the future.
> > -Tom
> LGTM. For the 4th one, why can’t the pattern be specified on the instruction directly? Is the extra 0 operand to the instruction a problem? I also don’t understand the point of the 6th one. Which instructions have different encodings on different subtargets?
If you include a pattern in the instruction definition, it must contain
all operands of the instruction. Since we aren't using src2 in the
pattern, tablegen will complain.
More information about the llvm-commits