[PATCH] D120000: [1/2] TLS loads opimization (hoist)

[Xiang Zhang via Phabricator via llvm-commits]

xiangzhangllvm created this revision.
xiangzhangllvm added reviewers: LuoYuanke, pengfei, andrew.w.kaylor, clin1.
Herald added subscribers: dexonsmith, hiraditya, mgorny.
xiangzhangllvm requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

When we access a TLS variable in PIC mode, it usually get the TLS address by calling a lib function, some like
callq   __tls_get_addr at PLT
This call was not show in IR or MIR, usually tag by a target-special flag (like pic) and generated in Assembly Printing.
So it is usually call it every time when TLS variable is accessed. Many of them are duplicated, especially in loops.

This patch is try to optimize it. It identifies/eliminate Redundant TLS Loads by hoist the TLS access when the related option is set. 
The Clang option will add in patch [2/2]

For example:

  static __thread int x;
  int g();
  int f(int c) {
    int *px = &x;
    while (c--)
      *px += g();
    return *px;
  }

will generated Redundant TLS Loads by compiling it with
 Clang++ -fPIC -ftls-model=global-dynamic -O2 -S

  .LBB0_2:                                # %while.body
                                          # =>This Inner Loop Header: Depth=1
          callq   _Z1gv at PLT
          movl    %eax, %ebp
          leaq    _ZL1x at TLSLD(%rip), %rdi
          callq   __tls_get_addr at PLT
          addl    _ZL1x at DTPOFF(%rax), %ebp
          movl    %ebp, _ZL1x at DTPOFF(%rax)
          addl    $-1, %ebx
          jne     .LBB0_2
          jmp     .LBB0_3
  .LBB0_4:                                # %entry.while.end_crit_edge
          leaq    _ZL1x at TLSLD(%rip), %rdi
          callq   __tls_get_addr at PLT
          movl    _ZL1x at DTPOFF(%rax), %ebp

The Redundant TLS Loads will hurt the performance, especially in loops.
 So we try to eliminate/move them if required by customers, let it be:

  # %bb.0:                                # %entry
           ...
           movl    %edi, %ebx
           leaq    _ZL1x at TLSLD(%rip), %rdi
           callq   __tls_get_addr at PLT
           leaq    _ZL1x at DTPOFF(%rax), %r14
           testl   %ebx, %ebx
           je      .LBB0_1
   .LBB0_2:                                # %while.body
                                           # =>This Inner Loop Header: Depth=1
           callq   _Z1gv at PLT
           addl    (%r14), %eax
           movl    %eax, (%r14)
           addl    $-1, %ebx
           jne     .LBB0_2
           jmp     .LBB0_3


https://reviews.llvm.org/D120000

Files:
  llvm/include/llvm/CodeGen/CodeGenPassBuilder.h
  llvm/include/llvm/CodeGen/MachinePassRegistry.def
  llvm/include/llvm/IR/Module.h
  llvm/include/llvm/InitializePasses.h
  llvm/include/llvm/LinkAllPasses.h
  llvm/include/llvm/Transforms/Scalar.h
  llvm/include/llvm/Transforms/Scalar/TLSVariableControl.h
  llvm/lib/CodeGen/TargetPassConfig.cpp
  llvm/lib/IR/Module.cpp
  llvm/lib/Passes/PassBuilder.cpp
  llvm/lib/Passes/PassRegistry.def
  llvm/lib/Transforms/Scalar/CMakeLists.txt
  llvm/lib/Transforms/Scalar/Scalar.cpp
  llvm/lib/Transforms/Scalar/TLSVariableControl.cpp
  llvm/test/CodeGen/X86/tls-loads-control.ll
  llvm/test/CodeGen/X86/tls-loads-control2.ll
  llvm/test/CodeGen/X86/tls-loads-control3.ll
  llvm/tools/llc/llc.cpp

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D120000.409470.patch
Type: text/x-patch
Size: 49178 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220217/5f47be44/attachment-0001.bin>