Allow unordered loads to be considered invariant in CodeGen

The actual code change is fairly straight forward, but exercising it isn't. First, it turned out we weren't adding the appropriate flags in SelectionDAG. Second, it turned out that we've got some optimization gaps, so obvious test cases don't work.

My first attempt (in atomic-unordered.ll) points out a deficiency in our peephole-opt folding logic which I plan to fix separately. Instead, I'm exercising this through MachineLICM.

Differential Revision: https://reviews.llvm.org/D59375



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356494 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/MachineInstr.cpp

@@ -1312,9 +1312,11 @@ bool MachineInstr::isDereferenceableInvariantLoad(AliasAnalysis *AA) const {
   const MachineFrameInfo &MFI = getParent()->getParent()->getFrameInfo();
 
   for (MachineMemOperand *MMO : memoperands()) {
-    if (MMO->isVolatile()) return false;
-    // TODO: Figure out whether isAtomic is really necessary (see D57601).
-    if (MMO->isAtomic()) return false;
+    if (!MMO->isUnordered())
+      // If the memory operand has ordering side effects, we can't move the
+      // instruction.  Such an instruction is technically an invariant load,
+      // but the caller code would need updated to expect that.
+      return false;
     if (MMO->isStore()) return false;
     if (MMO->isInvariant() && MMO->isDereferenceable())
       continue;

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -4564,6 +4564,11 @@ void SelectionDAGBuilder::visitAtomicLoad(const LoadInst &I) {
   auto Flags = MachineMemOperand::MOLoad;
   if (I.isVolatile())
     Flags |= MachineMemOperand::MOVolatile;
+  if (I.getMetadata(LLVMContext::MD_invariant_load) != nullptr)
+    Flags |= MachineMemOperand::MOInvariant;
+  if (isDereferenceablePointer(I.getPointerOperand(), DAG.getDataLayout()))
+    Flags |= MachineMemOperand::MODereferenceable;
+
   Flags |= TLI.getMMOFlags(I);
 
   MachineMemOperand *MMO =

test/CodeGen/X86/atomic-unordered.ll

@@ -2546,7 +2546,7 @@ define i64 @fold_constant_fence(i64 %arg) {
   ret i64 %ret
 }
 
-define i64 @fold_invariant_clobber(i64* %p, i64 %arg) {
+define i64 @fold_invariant_clobber(i64* dereferenceable(8) %p, i64 %arg) {
 ; CHECK-O0-LABEL: fold_invariant_clobber:
 ; CHECK-O0:       # %bb.0:
 ; CHECK-O0-NEXT:    movq (%rdi), %rax
@@ -2567,7 +2567,7 @@ define i64 @fold_invariant_clobber(i64* %p, i64 %arg) {
 }
 
 
-define i64 @fold_invariant_fence(i64* %p, i64 %arg) {
+define i64 @fold_invariant_fence(i64* dereferenceable(8) %p, i64 %arg) {
 ; CHECK-O0-LABEL: fold_invariant_fence:
 ; CHECK-O0:       # %bb.0:
 ; CHECK-O0-NEXT:    movq (%rdi), %rdi

test/CodeGen/X86/hoist-invariant-load.ll

@@ -73,27 +73,35 @@ define void @test_unordered(i8* %x) uwtable ssp {
 ; CHECK:       ## %bb.0: ## %entry
 ; CHECK-NEXT:    pushq %rbp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    pushq %r14
+; CHECK-NEXT:    pushq %r15
 ; CHECK-NEXT:    .cfi_def_cfa_offset 24
-; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    pushq %r14
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
-; CHECK-NEXT:    .cfi_offset %rbx, -32
-; CHECK-NEXT:    .cfi_offset %r14, -24
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset %rbx, -40
+; CHECK-NEXT:    .cfi_offset %r14, -32
+; CHECK-NEXT:    .cfi_offset %r15, -24
 ; CHECK-NEXT:    .cfi_offset %rbp, -16
 ; CHECK-NEXT:    movq %rdi, %rbx
 ; CHECK-NEXT:    movl $10000, %ebp ## imm = 0x2710
-; CHECK-NEXT:    movq _objc_msgSend@{{.*}}(%rip), %r14
+; CHECK-NEXT:    movq {{.*}}(%rip), %r14
+; CHECK-NEXT:    movq _objc_msgSend@{{.*}}(%rip), %r15
 ; CHECK-NEXT:    .p2align 4, 0x90
 ; CHECK-NEXT:  LBB1_1: ## %for.body
 ; CHECK-NEXT:    ## =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    movq {{.*}}(%rip), %rsi
 ; CHECK-NEXT:    movq %rbx, %rdi
-; CHECK-NEXT:    callq *%r14
+; CHECK-NEXT:    movq %r14, %rsi
+; CHECK-NEXT:    callq *%r15
 ; CHECK-NEXT:    decl %ebp
 ; CHECK-NEXT:    jne LBB1_1
 ; CHECK-NEXT:  ## %bb.2: ## %for.end
+; CHECK-NEXT:    addq $8, %rsp
 ; CHECK-NEXT:    popq %rbx
 ; CHECK-NEXT:    popq %r14
+; CHECK-NEXT:    popq %r15
 ; CHECK-NEXT:    popq %rbp
 ; CHECK-NEXT:    retq
 entry: