[PATCH] D61884: [RISCV] Support stack offset exceed 32-bit for RV64

[Shiva Chen via Phabricator via llvm-commits]

shiva0217 marked 2 inline comments as done.
shiva0217 added inline comments.


================
Comment at: lib/Target/RISCV/RISCVFrameLowering.cpp:88
+    } else {
+      TII->movImm64(MBB, MBBI, DL, ScratchReg, Val, Flag);
+    }
----------------
lenary wrote:
> I think you can probably get rid of `movImm32`, and only use movImm64 (the logic in `generateInstSeq` covers 32-bit operands), which I think will address the TODO in `movImm32`.
> 
> Then my only worry would be how we error when we need a larger-than-32-bit offset on RISC-V 32, which would probably cause errors somewhere way before this code anyway.
Yes, there's some test case improvement by replacing movImm32 with the generateInstSeq one, thanks!

I would add an error message for RV32 with a larger-than-32-bit offset. We could update the message when we support a larger-than-32-bit offset in RV32.


================
Comment at: test/CodeGen/RISCV/rv64-large-stack.ll:42
+entry:
+  %w = alloca [100000000 x { fp128, fp128 }], align 16
+  %0 = bitcast [100000000 x { fp128, fp128 }]* %w to i8*
----------------
lenary wrote:
> I think this test case can be significantly simplified. You should just need a single alloca, and then maybe a call that uses a pointer into the alloca?
Yes, it becomes significantly cleaner, thanks!


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61884/new/

https://reviews.llvm.org/D61884