On 11/12/22 14:29, Philipp Tomsich wrote:
When if-conversion encounters sequences using immediates, the
sequences can't trivially map back onto vt.maskc/vt.maskcn (even if
benefitial) due to vt.maskc and vt.maskcn not having immediate forms.
This adds a splitter to rewrite opportunities for XVentanaCondOps that
operate on an immediate by first putting the immediate into a register
to enable the non-immediate vt.maskc/vt.maskcn instructions to operate
on the value.
Consider code, such as
long func2 (long a, long c)
{
if (c)
a = 2;
else
a = 5;
return a;
}
which will be converted to
func2:
seqz a0,a2
neg a0,a0
andi a0,a0,3
addi a0,a0,2
ret
Following this change, we generate
li a0,3
vt.maskcn a0,a0,a2
addi a0,a0,2
ret
This commit also introduces a simple unit test for if-conversion with
immediate (literal) values as the sources for simple sets in the THEN
and ELSE blocks. The test checks that Ventana's conditional mask
instruction (vt.maskc<n>) is emitted as part of the resultant branchless
instruction sequence.
gcc/ChangeLog:
* config/riscv/xventanacondops.md: Support immediates for
vt.maskc/vt.maskcn through a splitter.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/xventanacondops-ifconv-imm.c: New test.
OK once we've cleared the non-technical hurdles to committing vendor
specific extensions.
Jeff
