[PATCH] D79785: [ARM] Register pressure with -mthumb forces register reload before each call

Mon Jun 22 03:11:05 PDT 2020

prathamesh added a comment.

In D79785#2100272 <https://reviews.llvm.org/D79785#2100272>, @ostannard wrote:

> > only if the register holding function's address got spilled
>
> I don't see any checks for this in the code. Why doesn't this trigger on every tLDRpci -> tBLXr pair, even if the register did not get spilled?

Oops sorry about the comment - "register holding function's address gets spilled", I got mixed up on that.
The point of adding foldMemoryOperand was to fold tLDRpci, tBLXr pair to tBL instead of reloading the function's address and calling it indirectly.

For above case, where tBLXr isn't folded, it's because tLDRpci is paired with COPY, instead of tBLXr:

  %16:tgpr = tLDRpci %const.0, 14, $noreg :: (load 4 from constant-pool)
  %23:tgpr = COPY %16:tgpr
  tBLXr 14, $noreg, %23:tgpr, <regmask $lr $d8 $d9 $d10 $d11 $d12 $d13 $d14 $d15 $q4 $q5 $q6 $q7 $r4 $r5 $r6 $r7 $r8 $r9 $r10 $r11 $s16 $s17 $s18 $s19 $s20 $s21 $s22 $s23 $s24 $s25 $s26 $s27 and 35 more...>, implicit-def dead $lr, implicit $sp, implicit $r0, implicit $r1, implicit $r2, implicit $r3, implicit-def $sp

Dumping LoadMI and MI in Thumb1InstrInfo::foldMemoryOperandImpl hook show first two insns above respectively.

Thanks,
Prathamesh

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D79785/new/

https://reviews.llvm.org/D79785