Hi Robin,
I once had different comments for those but either I never pushed them
or they got buried in the process of refactoring. The explanatory
comment explaining vlmax is also in "nowhere land" below autovec_use_vlmax_p.
(it says vsetvli instead of vsetvl as well...) It would be useful
to move it to above the function comments you touch.
I would like to move this comment to insn_expander::emit_insn body
before set avl in another patch which add VLS avl_type.
+/* Emit RVV insn which vl is the number of units of the vector mode.
+ This function can only be used before LRA pass or for VLS_AVL_IMM modes. */
Emit an RVV insn with a vector length that equals the number of units of
the vector mode. For VLA modes this corresponds to VLMAX.
Unless the vector length can be encoded in the vsetivl[i] instruction this
function must only be used as long as we can create pseudo registers.
This is because it will set a pseudo register to VLMAX using vsetvl and
use this as definition for the vector length.
Besides, we could add a const_vlmax_p () || can_create_pseudo_p assert here?
+/* Like emit_vlmax_insn but can be only used after LRA pass that can't create
+ pseudo register. */
Like emit_vlmax_insn but must only be used when we cannot create pseudo
registers anymore. This function, however, takes a predefined vector
length from the value in VL.
+/* Emit RVV insn which vl is the VL argument. */
+emit_nonvlmax_insn (unsigned icode, unsigned insn_flags, rtx *ops, rtx vl)
I think I renamed this to emit_len_insn or something before but Juzhe didn't
like it ;)
How about something like:
Emit an RVV insn with a predefined vector length. Contrary to emit_vlmax_insn
the instruction's vector length is not deduced from its mode but taken from
the value in VL.
Thank you very much, I used all of them. Here the V2 patch:
https://gcc.gnu.org/pipermail/gcc-patches/2023-September/631114.html
--
Best,
Lehua (RiVAI)
lehua.d...@rivai.ai
