In D150862#4359123, @StephenFan wrote:

I think we still resolved part of this problem instead of all of it. The reason is for

define void @foo(ptr %dest) {
  %p1 = getelementptr i8, ptr %dest, i32 2048
  store i8 1, ptr %p1
  %p2 = getelementptr i8, ptr %dest, i32 2049
  store i8 1, ptr %p2
  %p3 = getelementptr i8, ptr %dest, i32 2050
  store i8 1, ptr %p3
  %p4 = getelementptr i8, ptr %dest, i32 2051
  store i8 1, ptr %p4
  ret void
}

the extra addi a1, a0, 1 is produced by 2048 can not be represented in Int<12>. And this patch just resolved this kind of case. Consider that if the offset is 4194305, the addi still exists since the instruction selector selects it to

lui	a1, 1024
addiw	a1, a1, 1

If you want to solve all of these kinds of cases, you need to repeat all the logic that selecting a large constant. So maybe handle it after instruction selection is a better way?

• Rebase
• Fix assertion failures due to dangling value handle references
• Add a test from @StephenFan, which tries to find an offset divisible by 4096 to use lui + add.

+ Rebase
+ Fix assertion failures

In D150665#4351984, @craig.topper wrote:

In D150665#4351914, @dtcxzyw wrote:

In D150665#4350578, @craig.topper wrote:

I feel like this should be handled where the chain was formed or in DAG combine. The whole reason the chain exists is to provide a common base. If the common base isn't right we should fix that.

I cannot find code about forming chains in RISCVISelLowering.cpp and RISCVISelDAGToDAG.cpp. Should I implement some virtual member functions of class TargetLowering?

I think the splitting is done by CodeGenPrepare::splitLargeGEPOffsets()

In D150665#4350578, @craig.topper wrote:

I feel like this should be handled where the chain was formed or in DAG combine. The whole reason the chain exists is to provide a common base. If the common base isn't right we should fix that.

Rebase and address comments.

• Rebase
• Find the best (base, offset) pair to emit compressed load/store instructions with C/Zca/Zcf/Zcd.
• Add additional tests with large offsets

fix comment and rebase

Resolve conflicts before committing

Update tests

Update tests

Update tests

In D147789#4251449, @nikic wrote:

Can you base this on nowrap flags instead of re-deriving them?

Add tests

Fix test errors.