On 7/9/24 8:14 PM, Xi Ruoyao wrote:
On Tue, 2024-07-09 at 16:10 -0700, Vineet Gupta wrote:
On 7/3/24 21:35, Xi Ruoyao wrote:
On Sun, 2024-06-30 at 17:47 -0700, Vineet Gupta wrote:
- Don't hardcode SI in patterns, try to keep X to avoid potential
sign extension pitfalls. Implementation wise requires skipping
:MODE specifier in match_operand which is flagged as missing mode
warning.
I'm unsure about this. GCC Internal says:
‘isfinitem2’
Return 1 if operand 1 is a finite floating point number and 0 otherwise.
m is a scalar floating point mode. Operand 0 has mode SImode, and
operand 1 has mode m.
Likewise for isnormalm2. BTW isinfm2 is missing in the page.
So per the doc SImode is required. At least the doc should be updated
to say "operand 0 has an integer mode" or something if doing so is
intentionally allowed.
While the man page does say these return int, for which SI mode whould
suffice, I feel that keeping it X is more flexible.
So we should change the gcc documentation - so you want to send a patch ?
I do want to change it (it'll make implementing this for LoongArch
easier as well) but I don't know what we precisely allow here. "Any
integer mode" or something?
I'd lean towards any integer mode. I would expect that most targets
would ultimately use their native word_mode.
Jeff
