maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[rs4gc] avoid insert base computation instructions for deopt uses 

If we have a value live over a call which is used for deopt at the call, we know that the value must be a base pointer. We can avoid potentially inserting IR to materialize a base for this value.

In it's current form, this is mostly a compile time optimization.   Building the base pointer graph (and then optimizing it away again) is a relatively expensive operation.  We also sometimes end up with better codegen in practice - due to failures in optimizing away the inserted base pointer propogation - but those are optimization bugs we're fixing concurrently.

The alternative to this would be to extend the base pointer inference with the ability to generally reuse multiple-base input instructions (phis and selects).  That's somewhat invasive and complicated, so we're defering it a bit longer.

Differential Revision: https://reviews.llvm.org/D97885
This commit is contained in:
Philip Reames 2021-03-05 09:55:36 -08:00
parent 2b50ce1524
commit 99f93dd3a5
3 changed files with 37 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
llvm/lib/Transforms/Scalar/RewriteStatepointsForGC.cpp
View File

@ -1236,7 +1236,20 @@ static void findBasePointers(DominatorTree &DT, DefiningValueMapTy &DVCache,
CallBase *Call,
PartiallyConstructedSafepointRecord &result) {
MapVector<Value *, Value *> PointerToBase;
findBasePointers(result.LiveSet, PointerToBase, &DT, DVCache);
StatepointLiveSetTy PotentiallyDerivedPointers = result.LiveSet;
// We assume that all pointers passed to deopt are base pointers; as an
// optimization, we can use this to avoid seperately materializing the base
// pointer graph. This is only relevant since we're very conservative about
// generating new conflict nodes during base pointer insertion. If we were
// smarter there, this would be irrelevant.
if (auto Opt = Call->getOperandBundle(LLVMContext::OB_deopt))
for (Value *V : Opt->Inputs) {
if (!PotentiallyDerivedPointers.count(V))
continue;
PotentiallyDerivedPointers.remove(V);
PointerToBase[V] = V;
}
findBasePointers(PotentiallyDerivedPointers, PointerToBase, &DT, DVCache);
if (PrintBasePointers) {
errs() << "Base Pairs (w/o Relocation):\n";

20
llvm/test/Transforms/RewriteStatepointsForGC/base-pointers.ll
View File

@ -152,4 +152,24 @@ next: ; preds = %merge
ret i64 addrspace(1)* %bdv
}
; We know from the deopt use that %bdv must be a base value, and as
; result can avoid materializing the extra copy of the BDV phi node.
; (Even without a general forward analysis)
define i64 addrspace(1)* @test6(i1 %cnd, i64 addrspace(1)* %obj, i64 addrspace(1)* %obj2) gc "statepoint-example" {
; CHECK-LABEL: @test6
entry:
br label %merge
merge: ; preds = %merge, %entry
; CHECK-LABEL: merge:
; CHECK-NEXT: phi
; CHECK-NEXT: br i1
%bdv = phi i64 addrspace(1)* [ %obj, %entry ], [ %obj2, %merge ]
br i1 %cnd, label %merge, label %next
next: ; preds = %merge
call void @foo() [ "deopt"(i64 addrspace(1)* %bdv) ]
ret i64 addrspace(1)* %bdv
}
declare void @foo()

10
llvm/test/Transforms/RewriteStatepointsForGC/scalar-base-vector.ll
View File

@ -147,24 +147,20 @@ define void @test6() gc "statepoint-example" {
; CHECK-NEXT: bb:
; CHECK-NEXT: br label [[HEADER:%.*]]
; CHECK: header:
; CHECK-NEXT: [[TMP_BASE:%.*]] = phi i8 addrspace(1)* [ [[TMP6_BASE:%.*]], [[LATCH:%.*]] ], [ null, [[BB:%.*]] ], !is_base_value !0
; CHECK-NEXT: [[TMP:%.*]] = phi i8 addrspace(1)* [ [[TMP6:%.*]], [[LATCH]] ], [ undef, [[BB]] ]
; CHECK-NEXT: [[TMP:%.*]] = phi i8 addrspace(1)* [ [[TMP6:%.*]], [[LATCH:%.*]] ], [ undef, [[BB:%.*]] ]
; CHECK-NEXT: br label [[BB10:%.*]]
; CHECK: bb10:
; CHECK-NEXT: [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 2882400000, i32 0, void ()* @spam, i32 0, i32 0, i32 0, i32 0) [ "deopt"(i8 addrspace(1)* [[TMP]]), "gc-live"(i8 addrspace(1)* [[TMP]], i8 addrspace(1)* [[TMP_BASE]]) ]
; CHECK-NEXT: [[TMP_RELOCATED:%.*]] = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token [[STATEPOINT_TOKEN]], i32 1, i32 0)
; CHECK-NEXT: [[TMP_BASE_RELOCATED:%.*]] = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token [[STATEPOINT_TOKEN]], i32 1, i32 1)
; CHECK-NEXT: [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 2882400000, i32 0, void ()* @spam, i32 0, i32 0, i32 0, i32 0) [ "deopt"(i8 addrspace(1)* [[TMP]]), "gc-live"(i8 addrspace(1)* [[TMP]]) ]
; CHECK-NEXT: [[TMP_RELOCATED:%.*]] = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT: br label [[BB25:%.*]]
; CHECK: bb25:
; CHECK-NEXT: [[STATEPOINT_TOKEN1:%.*]] = call token (i64, i32, <2 x i8 addrspace(1)*> ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_v2p1i8f(i64 2882400000, i32 0, <2 x i8 addrspace(1)*> ()* @baz, i32 0, i32 0, i32 0, i32 0)
; CHECK-NEXT: [[TMP262:%.*]] = call <2 x i8 addrspace(1)*> @llvm.experimental.gc.result.v2p1i8(token [[STATEPOINT_TOKEN1]])
; CHECK-NEXT: [[BASE_EE:%.*]] = extractelement <2 x i8 addrspace(1)*> [[TMP262]], i32 0, !is_base_value !0
; CHECK-NEXT: [[TMP27:%.*]] = extractelement <2 x i8 addrspace(1)*> [[TMP262]], i32 0
; CHECK-NEXT: br i1 undef, label [[BB7:%.*]], label [[LATCH]]
; CHECK: bb7:
; CHECK-NEXT: br label [[LATCH]]
; CHECK: latch:
; CHECK-NEXT: [[TMP6_BASE]] = phi i8 addrspace(1)* [ [[BASE_EE]], [[BB25]] ], [ [[BASE_EE]], [[BB7]] ], !is_base_value !0
; CHECK-NEXT: [[TMP6]] = phi i8 addrspace(1)* [ [[TMP27]], [[BB25]] ], [ [[TMP27]], [[BB7]] ]
; CHECK-NEXT: br label [[HEADER]]
;