Sanjoy Das 5b5782c20e [RewriteStatepointsForGC] Fix a bug on creating gc_relocate for pointer to vector of pointers
Summary:
In RewriteStatepointsForGC pass, we create a gc_relocate intrinsic for
each relocated pointer, and the gc_relocate has the same type with the
pointer. During the creation of gc_relocate intrinsic, llvm requires to
mangle its type. However, llvm does not support mangling of all possible
types. RewriteStatepointsForGC will hit an assertion failure when it
tries to create a gc_relocate for pointer to vector of pointers because
mangling for vector of pointers is not supported.

This patch changes the way RewriteStatepointsForGC pass creates
gc_relocate. For each relocated pointer, we erase the type of pointers
and create an unified gc_relocate of type i8 addrspace(1)*. Then a
bitcast is inserted to convert the gc_relocate to the correct type. In
this way, gc_relocate does not need to deal with different types of
pointers and the unsupported type mangling is no longer a problem. This
change would also ease further merge when LLVM erases types of pointers
and introduces an unified pointer type.

Some minor changes are also introduced to gc_relocate related part in
InstCombineCalls, CodeGenPrepare, and Verifier accordingly.

Patch by Chen Li!

Reviewers: reames, AndyAyers, sanjoy

Reviewed By: sanjoy

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9592

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237009 91177308-0d34-0410-b5e6-96231b3b80d8
2015-05-11 18:49:34 +00:00

101 lines
4.6 KiB
LLVM

; RUN: opt %s -rewrite-statepoints-for-gc -S 2>&1 | FileCheck %s
declare i64 addrspace(1)* @generate_obj()
declare void @use_obj(i64 addrspace(1)*)
; The rewriting needs to make %obj loop variant by inserting a phi
; of the original value and it's relocation.
define void @def_use_safepoint() gc "statepoint-example" {
; CHECK-LABEL: def_use_safepoint
entry:
%obj = call i64 addrspace(1)* @generate_obj()
br label %loop
loop:
; CHECK: phi i64 addrspace(1)*
; CHECK-DAG: [ %obj.relocated.casted, %loop ]
; CHECK-DAG: [ %obj, %entry ]
call void @use_obj(i64 addrspace(1)* %obj)
%safepoint_token = call i32 (void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(void ()* @do_safepoint, i32 0, i32 0, i32 0, i32 5, i32 0, i32 -1, i32 0, i32 0, i32 0)
br label %loop
}
declare void @do_safepoint()
declare void @parse_point(i64 addrspace(1)*)
define i64 addrspace(1)* @test1(i32 %caller, i8 addrspace(1)* %a, i8 addrspace(1)* %b, i32 %unknown) gc "statepoint-example" {
; CHECK-LABEL: test1
entry:
br i1 undef, label %left, label %right
left:
%a.cast = bitcast i8 addrspace(1)* %a to i64 addrspace(1)*
; CHECK: left:
; CHECK-NEXT: %a.cast = bitcast i8 addrspace(1)* %a to i64 addrspace(1)*
; CHECK-NEXT: [[CAST_L:%.*]] = bitcast i8 addrspace(1)* %a to i64 addrspace(1)*
; Our safepoint placement pass calls removeUnreachableBlocks, which does a bunch
; of simplifications to branch instructions. This bug is visible only when
; there are multiple branches into the same block from the same predecessor, and
; the following ceremony is to make that artefact survive a call to
; removeUnreachableBlocks. As an example, "br i1 undef, label %merge, label %merge"
; will get simplified to "br label %merge" by removeUnreachableBlocks.
switch i32 %unknown, label %right [ i32 0, label %merge
i32 1, label %merge
i32 5, label %merge
i32 3, label %right ]
right:
%b.cast = bitcast i8 addrspace(1)* %b to i64 addrspace(1)*
br label %merge
; CHECK: right:
; CHECK-NEXT: %b.cast = bitcast i8 addrspace(1)* %b to i64 addrspace(1)*
; CHECK-NEXT: [[CAST_R:%.*]] = bitcast i8 addrspace(1)* %b to i64 addrspace(1)*
merge:
; CHECK: merge:
; CHECK-NEXT: %base_phi = phi i64 addrspace(1)* [ [[CAST_L]], %left ], [ [[CAST_L]], %left ], [ [[CAST_L]], %left ], [ [[CAST_R]], %right ], !is_base_value !0
%value = phi i64 addrspace(1)* [ %a.cast, %left], [ %a.cast, %left], [ %a.cast, %left], [ %b.cast, %right]
%safepoint_token = call i32 (void (i64 addrspace(1)*)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidp1i64f(void (i64 addrspace(1)*)* @parse_point, i32 1, i32 0, i64 addrspace(1)* %value, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0)
ret i64 addrspace(1)* %value
}
;; The purpose of this test is to ensure that when two live values share a
;; base defining value with inherent conflicts, we end up with a *single*
;; base phi/select per such node. This is testing an optimization, not a
;; fundemental correctness criteria
define void @test2(i1 %cnd, i64 addrspace(1)* %base_obj, i64 addrspace(1)* %base_arg2) gc "statepoint-example" {
; CHECK-LABEL: @test2
entry:
%obj = getelementptr i64, i64 addrspace(1)* %base_obj, i32 1
br label %loop
loop: ; preds = %loop, %entry
; CHECK-LABEL: loop
; CHECK: %base_phi = phi i64 addrspace(1)*
; CHECK-DAG: [ %base_obj, %entry ]
; Given the two selects are equivelent, so are their base phis - ideally,
; we'd have commoned these, but that's a missed optimization, not correctness.
; CHECK-DAG: [ [[DISCARD:%base_select.*.relocated.casted]], %loop ]
; CHECK-NOT: base_phi2
; CHECK: next = select
; CHECK: base_select
; CHECK: extra2 = select
; CHECK: base_select
; CHECK: statepoint
;; Both 'next' and 'extra2' are live across the backedge safepoint...
%current = phi i64 addrspace(1)* [ %obj, %entry ], [ %next, %loop ]
%extra = phi i64 addrspace(1)* [ %obj, %entry ], [ %extra2, %loop ]
%nexta = getelementptr i64, i64 addrspace(1)* %current, i32 1
%next = select i1 %cnd, i64 addrspace(1)* %nexta, i64 addrspace(1)* %base_arg2
%extra2 = select i1 %cnd, i64 addrspace(1)* %nexta, i64 addrspace(1)* %base_arg2
%safepoint_token = call i32 (void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(void ()* @foo, i32 0, i32 0, i32 0, i32 5, i32 0, i32 -1, i32 0, i32 0, i32 0)
br label %loop
}
declare void @foo()
declare i32 @llvm.experimental.gc.statepoint.p0f_isVoidf(void ()*, i32, i32, ...)
declare i32 @llvm.experimental.gc.statepoint.p0f_isVoidp1i64f(void (i64 addrspace(1)*)*, i32, i32, ...)