; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes='rewrite-statepoints-for-gc,verify<safepoint-ir>' -S | FileCheck %s

declare void @use_obj(ptr addrspace(1)) "gc-leaf-function"
declare void @do_safepoint()


; Profitable test
define i32 @test_remat(ptr addrspace(1) %base, i1 %cond) gc "statepoint-example" {
; CHECK-LABEL: @test_remat(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[DERIVED:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE:%.*]], i32 16
; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT:    br i1 [[COND:%.*]], label [[HERE:%.*]], label [[THERE:%.*]]
; CHECK:       here:
; CHECK-NEXT:    [[STATEPOINT_TOKEN1:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED2:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN1]], i32 0, i32 0)
; CHECK-NEXT:    br label [[MERGE:%.*]]
; CHECK:       there:
; CHECK-NEXT:    [[STATEPOINT_TOKEN3:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED4:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN3]], i32 0, i32 0)
; CHECK-NEXT:    br label [[MERGE]]
; CHECK:       merge:
; CHECK-NEXT:    [[DOT0:%.*]] = phi ptr addrspace(1) [ [[BASE_RELOCATED2]], [[HERE]] ], [ [[BASE_RELOCATED4]], [[THERE]] ]
; CHECK-NEXT:    [[DERIVED_REMAT:%.*]] = getelementptr i32, ptr addrspace(1) [[DOT0]], i32 16
; CHECK-NEXT:    [[RET:%.*]] = load i32, ptr addrspace(1) [[DERIVED_REMAT]], align 4
; CHECK-NEXT:    ret i32 [[RET]]
;
entry:
  %derived = getelementptr i32, ptr addrspace(1) %base, i32 16
  call void @do_safepoint() [ "deopt"() ]
  br i1 %cond, label %here, label %there

here:
  call void @do_safepoint() [ "deopt"() ]
  br label %merge

there:
  call void @do_safepoint() [ "deopt"() ]
  br label %merge

merge:
  %ret = load i32, ptr addrspace(1) %derived
  ret i32 %ret
}

; Unprofitable test
define i32 @test_many_uses(ptr addrspace(1) %base, i1 %cond) gc "statepoint-example" {
; CHECK-LABEL: @test_many_uses(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[DERIVED:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE:%.*]], i32 16
; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT:    [[DERIVED_REMAT:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED]], i32 16
; CHECK-NEXT:    br i1 [[COND:%.*]], label [[HERE:%.*]], label [[THERE:%.*]]
; CHECK:       here:
; CHECK-NEXT:    [[STATEPOINT_TOKEN3:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED4:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN3]], i32 0, i32 0)
; CHECK-NEXT:    [[DERIVED_REMAT1:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED4]], i32 16
; CHECK-NEXT:    call void @use_obj(ptr addrspace(1) [[DERIVED_REMAT1]])
; CHECK-NEXT:    br label [[MERGE:%.*]]
; CHECK:       there:
; CHECK-NEXT:    call void @use_obj(ptr addrspace(1) [[DERIVED_REMAT]])
; CHECK-NEXT:    [[STATEPOINT_TOKEN5:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED6:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN5]], i32 0, i32 0)
; CHECK-NEXT:    [[DERIVED_REMAT2:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED6]], i32 16
; CHECK-NEXT:    br label [[MERGE]]
; CHECK:       merge:
; CHECK-NEXT:    [[DOT0:%.*]] = phi ptr addrspace(1) [ [[DERIVED_REMAT1]], [[HERE]] ], [ [[DERIVED_REMAT2]], [[THERE]] ]
; CHECK-NEXT:    call void @use_obj(ptr addrspace(1) [[DOT0]])
; CHECK-NEXT:    [[RET:%.*]] = load i32, ptr addrspace(1) [[DOT0]], align 4
; CHECK-NEXT:    ret i32 [[RET]]
;
entry:
  %derived = getelementptr i32, ptr addrspace(1) %base, i32 16
  call void @do_safepoint() [ "deopt" () ]
  br i1 %cond, label %here, label %there

here:
  call void @do_safepoint() [ "deopt"() ]
  call void @use_obj(ptr addrspace(1) %derived)
  br label %merge

there:
  call void @use_obj(ptr addrspace(1) %derived)
  call void @do_safepoint() [ "deopt"() ]
  br label %merge

merge:
  call void @use_obj(ptr addrspace(1) %derived)
  %ret = load i32, ptr addrspace(1) %derived
  ret i32 %ret
}

; Remat before phi - not implemented
define i32 @test_phi(ptr addrspace(1) %base, i1 %cond) gc "statepoint-example" {
; CHECK-LABEL: @test_phi(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[DERIVED1:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE:%.*]], i32 16
; CHECK-NEXT:    [[DERIVED2:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE]], i32 32
; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT:    [[DERIVED2_REMAT:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED]], i32 32
; CHECK-NEXT:    [[DERIVED1_REMAT:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED]], i32 16
; CHECK-NEXT:    br i1 [[COND:%.*]], label [[HERE:%.*]], label [[THERE:%.*]]
; CHECK:       here:
; CHECK-NEXT:    br label [[MERGE:%.*]]
; CHECK:       there:
; CHECK-NEXT:    br label [[MERGE]]
; CHECK:       merge:
; CHECK-NEXT:    [[PHI1:%.*]] = phi ptr addrspace(1) [ [[DERIVED1_REMAT]], [[HERE]] ], [ [[DERIVED2_REMAT]], [[THERE]] ]
; CHECK-NEXT:    [[RET:%.*]] = load i32, ptr addrspace(1) [[PHI1]], align 4
; CHECK-NEXT:    ret i32 [[RET]]
;
entry:
  %derived1 = getelementptr i32, ptr addrspace(1) %base, i32 16
  %derived2 = getelementptr i32, ptr addrspace(1) %base, i32 32
  call void @do_safepoint() [ "deopt"() ]
  br i1 %cond, label %here, label %there

here:
  br label %merge

there:
  br label %merge

merge:
  %phi1 = phi ptr addrspace(1) [ %derived1, %here ], [ %derived2, %there ]
  %ret = load i32, ptr addrspace(1) %phi1
  ret i32 %ret
}

; Several uses per block
; TODO: We could rematerialize once per block for several uses if there is not statepoint between
define i32 @test_same_block_with_statepoint(ptr addrspace(1) %base, i1 %cond) gc "statepoint-example" {
; CHECK-LABEL: @test_same_block_with_statepoint(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[DERIVED:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE:%.*]], i32 16
; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT:    br i1 [[COND:%.*]], label [[HERE:%.*]], label [[THERE:%.*]]
; CHECK:       here:
; CHECK-NEXT:    [[DERIVED_REMAT3:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED]], i32 16
; CHECK-NEXT:    call void @use_obj(ptr addrspace(1) [[DERIVED_REMAT3]])
; CHECK-NEXT:    [[STATEPOINT_TOKEN4:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED5:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN4]], i32 0, i32 0)
; CHECK-NEXT:    [[DERIVED_REMAT2:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED5]], i32 16
; CHECK-NEXT:    call void @use_obj(ptr addrspace(1) [[DERIVED_REMAT2]])
; CHECK-NEXT:    [[DERIVED_REMAT1:%.*]] = getelementptr i32, ptr addrspace(1) [[BASE_RELOCATED5]], i32 16
; CHECK-NEXT:    [[DUMMY:%.*]] = load i32, ptr addrspace(1) [[DERIVED_REMAT1]], align 4
; CHECK-NEXT:    [[STATEPOINT_TOKEN6:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED5]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED7:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN6]], i32 0, i32 0)
; CHECK-NEXT:    br label [[MERGE:%.*]]
; CHECK:       there:
; CHECK-NEXT:    [[STATEPOINT_TOKEN8:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED9:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN8]], i32 0, i32 0)
; CHECK-NEXT:    br label [[MERGE]]
; CHECK:       merge:
; CHECK-NEXT:    [[DOT0:%.*]] = phi ptr addrspace(1) [ [[BASE_RELOCATED7]], [[HERE]] ], [ [[BASE_RELOCATED9]], [[THERE]] ]
; CHECK-NEXT:    [[DERIVED_REMAT:%.*]] = getelementptr i32, ptr addrspace(1) [[DOT0]], i32 16
; CHECK-NEXT:    [[RET:%.*]] = load i32, ptr addrspace(1) [[DERIVED_REMAT]], align 4
; CHECK-NEXT:    ret i32 [[RET]]
;
entry:
  %derived = getelementptr i32, ptr addrspace(1) %base, i32 16
  call void @do_safepoint() [ "deopt" () ]
  br i1 %cond, label %here, label %there

here:
  call void @use_obj(ptr addrspace(1) %derived)
  call void @do_safepoint() [ "deopt"() ]
  call void @use_obj(ptr addrspace(1) %derived)
  %dummy = load i32, ptr addrspace(1) %derived
  call void @do_safepoint() [ "deopt" () ]
  br label %merge

there:
  call void @do_safepoint() [ "deopt"() ]
  br label %merge

merge:
  %ret = load i32, ptr addrspace(1) %derived
  ret i32 %ret
}

; Test long chain with sub-chain rematerialized
; TODO: If we rematerialized longer chain first (%v4), then shorter on (%v0) would become dead
define void @test_chain(ptr addrspace(1) %base, i1 %cond1) gc "statepoint-example" {
; CHECK-LABEL: @test_chain(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[V0:%.*]] = getelementptr i8, ptr addrspace(1) [[BASE:%.*]], i64 16
; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(ptr addrspace(1) [[BASE]]), "gc-live"(ptr addrspace(1) [[BASE]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN]], i32 0, i32 0)
; CHECK-NEXT:    br i1 [[COND1:%.*]], label [[BLOCK3:%.*]], label [[COMMON_RET:%.*]]
; CHECK:       block3:
; CHECK-NEXT:    [[V0_REMAT:%.*]] = getelementptr i8, ptr addrspace(1) [[BASE_RELOCATED]], i64 16
; CHECK-NEXT:    [[V4:%.*]] = getelementptr i8, ptr addrspace(1) [[V0_REMAT]], i64 70
; CHECK-NEXT:    [[STATEPOINT_TOKEN1:%.*]] = call token (i64, i32, ptr, i32, i32, ...) @llvm.experimental.gc.statepoint.p0(i64 2882400000, i32 0, ptr elementtype(void ()) @do_safepoint, i32 0, i32 0, i32 0, i32 0) [ "deopt"(ptr addrspace(1) [[BASE_RELOCATED]]), "gc-live"(ptr addrspace(1) [[BASE_RELOCATED]]) ]
; CHECK-NEXT:    [[BASE_RELOCATED2:%.*]] = call coldcc ptr addrspace(1) @llvm.experimental.gc.relocate.p1(token [[STATEPOINT_TOKEN1]], i32 0, i32 0)
; CHECK-NEXT:    [[V0_REMAT_REMAT:%.*]] = getelementptr i8, ptr addrspace(1) [[BASE_RELOCATED2]], i64 16
; CHECK-NEXT:    [[V4_REMAT:%.*]] = getelementptr i8, ptr addrspace(1) [[V0_REMAT_REMAT]], i64 70
; CHECK-NEXT:    [[V5:%.*]] = load atomic i8, ptr addrspace(1) [[V4_REMAT]] unordered, align 2
; CHECK-NEXT:    br label [[COMMON_RET]]
; CHECK:       common.ret:
; CHECK-NEXT:    ret void
;
entry:
  %v0 = getelementptr i8, ptr addrspace(1) %base, i64 16
  call void @do_safepoint() [ "deopt"(ptr addrspace(1) %base) ]
  br i1 %cond1, label %block3, label %common.ret

block3:
  %v4 = getelementptr i8, ptr addrspace(1) %v0, i64 70
  call void @do_safepoint() [ "deopt"(ptr addrspace(1) %base) ]
  %v5 = load atomic i8, ptr addrspace(1) %v4 unordered, align 2
  br label %common.ret

common.ret:
  ret void
}