[all-commits] [llvm/llvm-project] adf797: [ARM] Describe defs/uses of VLLDM and VLSTM

Momchil Velikov via All-commits all-commits at lists.llvm.org
Tue Jun 23 08:29:39 PDT 2020


  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: adf7973fd35d899a1500d2f379b4e6874b4de20f
      https://github.com/llvm/llvm-project/commit/adf7973fd35d899a1500d2f379b4e6874b4de20f
  Author: Momchil Velikov <momchil.velikov at arm.com>
  Date:   2020-06-23 (Tue, 23 Jun 2020)

  Changed paths:
    M llvm/lib/Target/ARM/ARMExpandPseudoInsts.cpp
    M llvm/lib/Target/ARM/ARMInstrVFP.td
    A llvm/test/CodeGen/ARM/cmse-vlldm-no-reorder.ll
    A llvm/test/CodeGen/ARM/cmse-vlldm-no-reorder.mir
    M llvm/test/CodeGen/ARM/vlldm-vlstm-uops.mir

  Log Message:
  -----------
  [ARM] Describe defs/uses of VLLDM and VLSTM

The VLLDM and VLSTM instructions are incompletely specified.  They
(potentially) write (or read, respectively) registers Q0-Q7, VPR, and
FPSCR, but the compiler is unaware of it.

In the new test case `cmse-vlldm-no-reorder.ll` case the compiler
missed an anti-dependency and reordered a `VLLDM` ahead of the
instruction, which stashed the return value from the non-secure call,
effectively clobbering said value.

This test case does not fail with upstream LLVM, because of scheduling
differences and I couldn't find a test case for the VLSTM either.

Differential Revision: https://reviews.llvm.org/D81586




More information about the All-commits mailing list