[llvm] 192a34d - [Attributor][OpenMPOpt][FIX] Register simplification callbacks
Johannes Doerfert via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 6 19:29:26 PST 2022
Author: Johannes Doerfert
Date: 2022-03-06T21:28:38-06:00
New Revision: 192a34ddb05c8cd720d3b1c2120416497797f5d3
URL: https://github.com/llvm/llvm-project/commit/192a34ddb05c8cd720d3b1c2120416497797f5d3
DIFF: https://github.com/llvm/llvm-project/commit/192a34ddb05c8cd720d3b1c2120416497797f5d3.diff
LOG: [Attributor][OpenMPOpt][FIX] Register simplification callbacks
Heap-2-stack and heap-2-shared can replace an allocation call with
something else. To avoid us deriving information from the allocator
implementation we register a simplification callback now that will
force us to stop at the call site. We probably should create the
replacement memory eagerly and return that instead though.
Added:
Modified:
llvm/lib/Transforms/IPO/AttributorAttributes.cpp
llvm/lib/Transforms/IPO/OpenMPOpt.cpp
llvm/test/Transforms/Attributor/dereferenceable-1.ll
llvm/test/Transforms/Attributor/heap_to_stack.ll
llvm/test/Transforms/Attributor/noalias.ll
llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
llvm/test/Transforms/OpenMP/replace_globalization.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
index 1910500aa30f7..ad2805b074267 100644
--- a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
+++ b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
@@ -455,6 +455,9 @@ stripAndAccumulateOffsets(Attributor &A, const AbstractAttribute &QueryingAA,
: DepClassTy::NONE);
ConstantRange Range = UseAssumed ? ValueConstantRangeAA.getAssumed()
: ValueConstantRangeAA.getKnown();
+ if (Range.isFullSet())
+ return false;
+
// We can only use the lower part of the range because the upper part can
// be higher than what the value can really be.
if (GetMinOffset)
@@ -5974,6 +5977,16 @@ struct AAHeapToStackFunction final : public AAHeapToStack {
/* CheckPotentiallyDead */ true);
(void)Success;
assert(Success && "Did not expect the call base visit callback to fail!");
+
+ Attributor::SimplifictionCallbackTy SCB =
+ [](const IRPosition &, const AbstractAttribute *,
+ bool &) -> Optional<Value *> { return nullptr; };
+ for (const auto &It : AllocationInfos)
+ A.registerSimplificationCallback(IRPosition::callsite_returned(*It.first),
+ SCB);
+ for (const auto &It : DeallocationInfos)
+ A.registerSimplificationCallback(IRPosition::callsite_returned(*It.first),
+ SCB);
}
const std::string getAsStr() const override {
diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 91b11f26cef85..f3914f36fdfce 100644
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -2966,9 +2966,15 @@ struct AAHeapToSharedFunction : public AAHeapToShared {
auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
auto &RFI = OMPInfoCache.RFIs[OMPRTL___kmpc_alloc_shared];
+ Attributor::SimplifictionCallbackTy SCB =
+ [](const IRPosition &, const AbstractAttribute *,
+ bool &) -> Optional<Value *> { return nullptr; };
for (User *U : RFI.Declaration->users())
- if (CallBase *CB = dyn_cast<CallBase>(U))
+ if (CallBase *CB = dyn_cast<CallBase>(U)) {
MallocCalls.insert(CB);
+ A.registerSimplificationCallback(IRPosition::callsite_returned(*CB),
+ SCB);
+ }
findPotentialRemovedFreeCalls(A);
}
diff --git a/llvm/test/Transforms/Attributor/dereferenceable-1.ll b/llvm/test/Transforms/Attributor/dereferenceable-1.ll
index df88b174c9413..656f30ce2ece7 100644
--- a/llvm/test/Transforms/Attributor/dereferenceable-1.ll
+++ b/llvm/test/Transforms/Attributor/dereferenceable-1.ll
@@ -319,7 +319,7 @@ define void @deref_or_null_and_nonnull(i32* dereferenceable_or_null(100) %0) {
define void @test8(i8* %ptr) #0 {
; IS________OPM: Function Attrs: argmemonly nofree norecurse nosync nounwind writeonly
; IS________OPM-LABEL: define {{[^@]+}}@test8
-; IS________OPM-SAME: (i8* nocapture nofree nonnull writeonly [[PTR:%.*]]) #[[ATTR3:[0-9]+]] {
+; IS________OPM-SAME: (i8* nocapture nofree writeonly [[PTR:%.*]]) #[[ATTR3:[0-9]+]] {
; IS________OPM-NEXT: br label [[TMP1:%.*]]
; IS________OPM: 1:
; IS________OPM-NEXT: [[I_0:%.*]] = phi i32 [ 20, [[TMP0:%.*]] ], [ [[TMP4:%.*]], [[TMP5:%.*]] ]
diff --git a/llvm/test/Transforms/Attributor/heap_to_stack.ll b/llvm/test/Transforms/Attributor/heap_to_stack.ll
index 4ee3c432ac5b1..d646ad4b18e5e 100644
--- a/llvm/test/Transforms/Attributor/heap_to_stack.ll
+++ b/llvm/test/Transforms/Attributor/heap_to_stack.ll
@@ -220,12 +220,21 @@ define void @test3c(i64 %alignment) {
; leave alone a constant-but-invalid alignment
define void @test3d(i8* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test3d
-; CHECK-SAME; (i8* nocapture [[P:%.*]]) {
-; CHECK-NEXT: [[TMP1:%.*]] = tail call noalias i8* @aligned_alloc(i64 noundef 33, i64 noundef 128)
-; CHECK: tail call void @free(i8* noalias nocapture [[TMP1]])
-; CHECK-NEXT: ret void
+; IS________OPM-LABEL: define {{[^@]+}}@test3d
+; IS________OPM-SAME: (i8* nocapture [[P:%.*]]) {
+; IS________OPM-NEXT: [[TMP1:%.*]] = tail call noalias i8* @aligned_alloc(i64 noundef 33, i64 noundef 128)
+; IS________OPM-NEXT: tail call void @nofree_arg_only(i8* nocapture nofree [[TMP1]], i8* nocapture [[P]])
+; IS________OPM-NEXT: tail call void @free(i8* noalias nocapture [[TMP1]])
+; IS________OPM-NEXT: ret void
;
+; IS________NPM-LABEL: define {{[^@]+}}@test3d
+; IS________NPM-SAME: (i8* nocapture [[P:%.*]]) {
+; IS________NPM-NEXT: [[TMP1:%.*]] = tail call noalias i8* @aligned_alloc(i64 noundef 33, i64 noundef 128)
+; IS________NPM-NEXT: tail call void @nofree_arg_only(i8* noalias nocapture nofree [[TMP1]], i8* nocapture [[P]])
+; IS________NPM-NEXT: tail call void @free(i8* noalias nocapture [[TMP1]])
+; IS________NPM-NEXT: ret void
+;
+; CHECK-SAME; (i8* nocapture [[P:%.*]]) {
%1 = tail call noalias i8* @aligned_alloc(i64 33, i64 128)
tail call void @nofree_arg_only(i8* %1, i8* %p)
tail call void @free(i8* %1)
@@ -578,8 +587,9 @@ define i32 @irreducible_cfg(i32 %0) {
; IS________NPM-NEXT: [[TMP14]] = add nsw i32 [[DOT1]], 1
; IS________NPM-NEXT: br label [[TMP8]]
; IS________NPM: 15:
-; IS________NPM-NEXT: [[TMP16:%.*]] = load i32, i32* [[TMP3]], align 4
-; IS________NPM-NEXT: ret i32 [[TMP16]]
+; IS________NPM-NEXT: [[TMP16:%.*]] = bitcast i32* [[TMP3]] to i8*
+; IS________NPM-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP3]], align 4
+; IS________NPM-NEXT: ret i32 [[TMP17]]
;
%2 = call noalias i8* @malloc(i64 4)
%3 = bitcast i8* %2 to i32*
diff --git a/llvm/test/Transforms/Attributor/noalias.ll b/llvm/test/Transforms/Attributor/noalias.ll
index 773e42d7e86c0..4cff1d70f8efd 100644
--- a/llvm/test/Transforms/Attributor/noalias.ll
+++ b/llvm/test/Transforms/Attributor/noalias.ll
@@ -414,11 +414,13 @@ define void @test12_4(){
; NOT_TUNIT_OPM-LABEL: define {{[^@]+}}@test12_4() {
; NOT_TUNIT_OPM-NEXT: [[A:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
; NOT_TUNIT_OPM-NEXT: [[B:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
+; NOT_TUNIT_OPM-NEXT: [[A_0:%.*]] = getelementptr i8, i8* [[A]], i64 0
; NOT_TUNIT_OPM-NEXT: [[A_1:%.*]] = getelementptr i8, i8* [[A]], i64 1
+; NOT_TUNIT_OPM-NEXT: [[B_0:%.*]] = getelementptr i8, i8* [[B]], i64 0
; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* noalias nocapture [[A]], i8* noalias nocapture [[B]])
-; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* nocapture [[A]], i8* nocapture [[A]])
+; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* nocapture [[A]], i8* nocapture [[A_0]])
; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* nocapture [[A]], i8* nocapture [[A_1]])
-; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* nocapture [[A]], i8* nocapture [[B]])
+; NOT_TUNIT_OPM-NEXT: tail call void @two_args(i8* nocapture [[A_0]], i8* nocapture [[B_0]])
; NOT_TUNIT_OPM-NEXT: ret void
;
%A = tail call noalias i8* @malloc(i64 4)
@@ -450,17 +452,12 @@ define void @use_i8_internal(i8* %a) {
}
define void @test13_use_noalias(){
-; IS________OPM-LABEL: define {{[^@]+}}@test13_use_noalias() {
-; IS________OPM-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
-; IS________OPM-NEXT: [[C1:%.*]] = bitcast i8* [[M1]] to i16*
-; IS________OPM-NEXT: [[C2:%.*]] = bitcast i16* [[C1]] to i8*
-; IS________OPM-NEXT: call void @use_i8_internal(i8* noalias nocapture [[C2]])
-; IS________OPM-NEXT: ret void
-;
-; NOT_TUNIT_OPM-LABEL: define {{[^@]+}}@test13_use_noalias() {
-; NOT_TUNIT_OPM-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
-; NOT_TUNIT_OPM-NEXT: call void @use_i8_internal(i8* noalias nocapture [[M1]])
-; NOT_TUNIT_OPM-NEXT: ret void
+; CHECK-LABEL: define {{[^@]+}}@test13_use_noalias() {
+; CHECK-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
+; CHECK-NEXT: [[C1:%.*]] = bitcast i8* [[M1]] to i16*
+; CHECK-NEXT: [[C2:%.*]] = bitcast i16* [[C1]] to i8*
+; CHECK-NEXT: call void @use_i8_internal(i8* noalias nocapture [[C2]])
+; CHECK-NEXT: ret void
;
; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@test13_use_noalias()
; IS__CGSCC_OPM-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 4)
@@ -476,20 +473,14 @@ define void @test13_use_noalias(){
}
define void @test13_use_alias(){
-; IS________OPM-LABEL: define {{[^@]+}}@test13_use_alias() {
-; IS________OPM-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
-; IS________OPM-NEXT: [[C1:%.*]] = bitcast i8* [[M1]] to i16*
-; IS________OPM-NEXT: [[C2A:%.*]] = bitcast i16* [[C1]] to i8*
-; IS________OPM-NEXT: [[C2B:%.*]] = bitcast i16* [[C1]] to i8*
-; IS________OPM-NEXT: call void @use_i8_internal(i8* nocapture [[C2A]])
-; IS________OPM-NEXT: call void @use_i8_internal(i8* nocapture [[C2B]])
-; IS________OPM-NEXT: ret void
-;
-; NOT_TUNIT_OPM-LABEL: define {{[^@]+}}@test13_use_alias() {
-; NOT_TUNIT_OPM-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
-; NOT_TUNIT_OPM-NEXT: call void @use_i8_internal(i8* nocapture [[M1]])
-; NOT_TUNIT_OPM-NEXT: call void @use_i8_internal(i8* nocapture [[M1]])
-; NOT_TUNIT_OPM-NEXT: ret void
+; CHECK-LABEL: define {{[^@]+}}@test13_use_alias() {
+; CHECK-NEXT: [[M1:%.*]] = tail call noalias i8* @malloc(i64 noundef 4)
+; CHECK-NEXT: [[C1:%.*]] = bitcast i8* [[M1]] to i16*
+; CHECK-NEXT: [[C2A:%.*]] = bitcast i16* [[C1]] to i8*
+; CHECK-NEXT: [[C2B:%.*]] = bitcast i16* [[C1]] to i8*
+; CHECK-NEXT: call void @use_i8_internal(i8* nocapture [[C2A]])
+; CHECK-NEXT: call void @use_i8_internal(i8* nocapture [[C2B]])
+; CHECK-NEXT: ret void
;
%m1 = tail call noalias i8* @malloc(i64 4)
%c1 = bitcast i8* %m1 to i16*
diff --git a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
index f650e92a61b5b..54c57a70da7c9 100644
--- a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
@@ -3417,6 +3417,7 @@ define dso_local i32 @conditional_calloc(i32 %x) {
; IS________NPM-NEXT: br label [[IF_END]]
; IS________NPM: if.end:
; IS________NPM-NEXT: [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
+; IS________NPM-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP1]] to i8*
; IS________NPM-NEXT: ret i32 [[TMP2]]
;
; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@conditional_calloc
@@ -3473,10 +3474,12 @@ define dso_local i32 @conditional_calloc_zero(i1 %c) {
; IS________NPM-NEXT: entry:
; IS________NPM-NEXT: [[TMP0:%.*]] = alloca i8, i64 4, align 1
; IS________NPM-NEXT: call void @llvm.memset.p0i8.i64(i8* [[TMP0]], i8 0, i64 4, i1 false)
+; IS________NPM-NEXT: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i32*
; IS________NPM-NEXT: br i1 [[C]], label [[IF_END:%.*]], label [[IF_THEN:%.*]]
; IS________NPM: if.then:
; IS________NPM-NEXT: br label [[IF_END]]
; IS________NPM: if.end:
+; IS________NPM-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP1]] to i8*
; IS________NPM-NEXT: ret i32 0
;
; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@conditional_calloc_zero
diff --git a/llvm/test/Transforms/OpenMP/replace_globalization.ll b/llvm/test/Transforms/OpenMP/replace_globalization.ll
index efec38855c9a8..4ecc97350fcc8 100644
--- a/llvm/test/Transforms/OpenMP/replace_globalization.ll
+++ b/llvm/test/Transforms/OpenMP/replace_globalization.ll
@@ -84,7 +84,15 @@ entry:
ret void
}
-declare i8* @__kmpc_alloc_shared(i64)
+ at offset =global i32 undef
+ at stack = internal addrspace(3) global [1024 x i8] undef
+define private i8* @__kmpc_alloc_shared(i64) {
+ %bc = bitcast [1024 x i8] addrspace(3) * @stack to i8 addrspace(3) *
+ %ac = addrspacecast i8 addrspace(3) * %bc to i8*
+ %l = load i32, i32* @offset
+ %gep = getelementptr i8, i8* %ac, i32 %l
+ ret i8* %gep
+}
declare void @__kmpc_free_shared(i8*, i64)
@@ -122,33 +130,41 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK: @[[S:[a-zA-Z0-9_$"\\.-]+]] = external local_unnamed_addr global i8*
; CHECK: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [113 x i8] c"
; CHECK: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([113 x i8], [113 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
+; CHECK: @[[OFFSET:[a-zA-Z0-9_$"\\.-]+]] = global i32 undef
+; CHECK: @[[STACK:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [1024 x i8] undef
; CHECK: @[[X_SHARED:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [16 x i8] undef, align 4
; CHECK: @[[Y_SHARED:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 4
;.
; CHECK-LABEL: define {{[^@]+}}@foo() {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false, i1 true)
-; CHECK-NEXT: [[X:%.*]] = call align 4 i8* @__kmpc_alloc_shared(i64 4) #[[ATTR5:[0-9]+]]
-; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR4:[0-9]+]]
-; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 [[X]]) #[[ATTR6:[0-9]+]]
-; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[X]], i64 4) #[[ATTR5]]
+; CHECK-NEXT: [[X:%.*]] = call align 4 i8* @__kmpc_alloc_shared(i64 noundef 4) #[[ATTR6:[0-9]+]]
+; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR5:[0-9]+]]
+; CHECK-NEXT: [[X_ON_STACK:%.*]] = bitcast i8* [[X]] to i32*
+; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[X_ON_STACK]] to i8*
+; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 [[TMP0]]) #[[ATTR7:[0-9]+]]
+; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[X]], i64 4) #[[ATTR8:[0-9]+]]
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
; CHECK-NEXT: ret void
;
;
; CHECK-LABEL: define {{[^@]+}}@bar() {
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false, i1 true)
-; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR4]]
+; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR5]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[C]], -1
; CHECK-NEXT: br i1 [[CMP]], label [[MASTER1:%.*]], label [[EXIT:%.*]]
; CHECK: master1:
-; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 addrspacecast (i8 addrspace(3)* getelementptr inbounds ([16 x i8], [16 x i8] addrspace(3)* @x_shared, i32 0, i32 0) to i8*)) #[[ATTR6]]
+; CHECK-NEXT: [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([16 x i8], [16 x i8] addrspace(3)* @x_shared, i32 0, i32 0) to i8*) to [4 x i32]*
+; CHECK-NEXT: [[A0:%.*]] = bitcast [4 x i32]* [[X_ON_STACK]] to i8*
+; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 [[A0]]) #[[ATTR7]]
; CHECK-NEXT: br label [[NEXT:%.*]]
; CHECK: next:
-; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR4]]
+; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR5]]
; CHECK-NEXT: br label [[MASTER2:%.*]]
; CHECK: master2:
-; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @y_shared, i32 0, i32 0) to i8*)) #[[ATTR6]]
+; CHECK-NEXT: [[Y_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @y_shared, i32 0, i32 0) to i8*) to [4 x i32]*
+; CHECK-NEXT: [[B1:%.*]] = bitcast [4 x i32]* [[Y_ON_STACK]] to i8*
+; CHECK-NEXT: call void @use.internalized(i8* nofree align 4 [[B1]]) #[[ATTR7]]
; CHECK-NEXT: br label [[EXIT]]
; CHECK: exit:
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
@@ -157,13 +173,15 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
;
; CHECK-LABEL: define {{[^@]+}}@baz_spmd() {
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 true, i1 true)
-; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR4]]
+; CHECK-NEXT: call void @unknown_no_openmp() #[[ATTR5]]
; CHECK-NEXT: [[C0:%.*]] = icmp eq i32 [[C]], -1
; CHECK-NEXT: br i1 [[C0]], label [[MASTER3:%.*]], label [[EXIT:%.*]]
; CHECK: master3:
-; CHECK-NEXT: [[Z:%.*]] = call align 4 i8* @__kmpc_alloc_shared(i64 24) #[[ATTR5]], !dbg [[DBG9:![0-9]+]]
-; CHECK-NEXT: call void @use.internalized(i8* nofree writeonly align 4 [[Z]]) #[[ATTR6]]
-; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[Z]], i64 24) #[[ATTR5]]
+; CHECK-NEXT: [[Z:%.*]] = call align 4 i8* @__kmpc_alloc_shared(i64 noundef 24) #[[ATTR6]], !dbg [[DBG9:![0-9]+]]
+; CHECK-NEXT: [[Z_ON_STACK:%.*]] = bitcast i8* [[Z]] to [6 x i32]*
+; CHECK-NEXT: [[C1:%.*]] = bitcast [6 x i32]* [[Z_ON_STACK]] to i8*
+; CHECK-NEXT: call void @use.internalized(i8* nofree writeonly align 4 [[C1]]) #[[ATTR7]]
+; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[Z]], i64 24) #[[ATTR8]]
; CHECK-NEXT: br label [[EXIT]]
; CHECK: exit:
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
@@ -184,14 +202,23 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-NEXT: store i8* [[X]], i8** @S, align 8
; CHECK-NEXT: ret void
;
+;
+; CHECK-LABEL: define {{[^@]+}}@__kmpc_alloc_shared
+; CHECK-SAME: (i64 [[TMP0:%.*]]) #[[ATTR1:[0-9]+]] {
+; CHECK-NEXT: [[L:%.*]] = load i32, i32* @offset, align 4
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([1024 x i8], [1024 x i8] addrspace(3)* @stack, i32 0, i32 0) to i8*), i32 [[L]]
+; CHECK-NEXT: ret i8* [[GEP]]
+;
;.
; CHECK: attributes #[[ATTR0]] = { nofree norecurse nosync nounwind willreturn writeonly }
-; CHECK: attributes #[[ATTR1:[0-9]+]] = { nosync nounwind }
-; CHECK: attributes #[[ATTR2:[0-9]+]] = { nounwind readnone speculatable }
-; CHECK: attributes #[[ATTR3:[0-9]+]] = { nofree nosync nounwind readnone speculatable willreturn }
-; CHECK: attributes #[[ATTR4]] = { "llvm.assume"="omp_no_openmp" }
-; CHECK: attributes #[[ATTR5]] = { nounwind }
-; CHECK: attributes #[[ATTR6]] = { nounwind writeonly }
+; CHECK: attributes #[[ATTR1]] = { nofree norecurse nosync nounwind readonly willreturn allocsize(0) }
+; CHECK: attributes #[[ATTR2:[0-9]+]] = { nosync nounwind }
+; CHECK: attributes #[[ATTR3:[0-9]+]] = { nounwind readnone speculatable }
+; CHECK: attributes #[[ATTR4:[0-9]+]] = { nofree nosync nounwind readnone speculatable willreturn }
+; CHECK: attributes #[[ATTR5]] = { "llvm.assume"="omp_no_openmp" }
+; CHECK: attributes #[[ATTR6]] = { nounwind readonly }
+; CHECK: attributes #[[ATTR7]] = { nounwind writeonly }
+; CHECK: attributes #[[ATTR8]] = { nounwind }
;.
; CHECK: [[META0:![0-9]+]] = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 12.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
; CHECK: [[META1:![0-9]+]] = !DIFile(filename: "replace_globalization.c", directory: "/tmp/replace_globalization.c")
More information about the llvm-commits
mailing list