Skip to content

Commit 10c11e4

Browse files
kawashima-fjsmithp35
kawashima-fj
authored and
smithp35
committed
This option allows selecting the TLS size in the local exec TLS model,
which is the default TLS model for non-PIC objects. This allows large/ many thread local variables or a compact/fast code in an executable. Specification is same as that of GCC. For example, the code model option precedes the TLS size option. TLS access models other than local-exec are not changed. It means supoort of the large code model is only in the local exec TLS model. Patch By KAWASHIMA Takahiro (kawashima-fj <[email protected]>) Reviewers: dmgreen, mstorsjo, t.p.northover, peter.smith, ostannard Reviewd By: peter.smith Committed by: peter.smith Differential Revision: https://reviews.llvm.org/D71688
1 parent 96b8e1a commit 10c11e4

File tree

13 files changed

+279
-71
lines changed

13 files changed

+279
-71
lines changed

‎clang/include/clang/Basic/CodeGenOptions.def

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -333,6 +333,9 @@ ENUM_CODEGENOPT(VecLib, VectorLibrary, 2, NoLibrary)
333333
/// The default TLS model to use.
334334
ENUM_CODEGENOPT(DefaultTLSModel, TLSModel, 2, GeneralDynamicTLSModel)
335335

336+
/// Bit size of immediate TLS offsets (0 == use the default).
337+
VALUE_CODEGENOPT(TLSSize, 8, 0)
338+
336339
/// Number of path components to strip when emitting checks. (0 == full
337340
/// filename)
338341
VALUE_CODEGENOPT(EmitCheckPathComponentsToStrip, 32, 0)

‎clang/include/clang/Driver/Options.td

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2178,6 +2178,9 @@ def mwatchsimulator_version_min_EQ : Joined<["-"], "mwatchsimulator-version-min=
21782178
def march_EQ : Joined<["-"], "march=">, Group<m_Group>, Flags<[CoreOption]>;
21792179
def masm_EQ : Joined<["-"], "masm=">, Group<m_Group>, Flags<[DriverOption]>;
21802180
def mcmodel_EQ : Joined<["-"], "mcmodel=">, Group<m_Group>;
2181+
def mtls_size_EQ : Joined<["-"], "mtls-size=">, Group<m_Group>, Flags<[DriverOption, CC1Option]>,
2182+
HelpText<"Specify bit size of immediate TLS offsets (AArch64 ELF only): "
2183+
"12 (for 4KB) | 24 (for 16MB, default) | 32 (for 4GB) | 48 (for 256TB, needs -mcmodel=large)">;
21812184
def mimplicit_it_EQ : Joined<["-"], "mimplicit-it=">, Group<m_Group>;
21822185
def mdefault_build_attributes : Joined<["-"], "mdefault-build-attributes">, Group<m_Group>;
21832186
def mno_default_build_attributes : Joined<["-"], "mno-default-build-attributes">, Group<m_Group>;

‎clang/lib/CodeGen/BackendUtil.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -474,6 +474,7 @@ static void initTargetOptions(llvm::TargetOptions &Options,
474474
Options.FunctionSections = CodeGenOpts.FunctionSections;
475475
Options.DataSections = CodeGenOpts.DataSections;
476476
Options.UniqueSectionNames = CodeGenOpts.UniqueSectionNames;
477+
Options.TLSSize = CodeGenOpts.TLSSize;
477478
Options.EmulatedTLS = CodeGenOpts.EmulatedTLS;
478479
Options.ExplicitEmulatedTLS = CodeGenOpts.ExplicitEmulatedTLS;
479480
Options.DebuggerTuning = CodeGenOpts.getDebuggerTuning();

‎clang/lib/Driver/ToolChains/Clang.cpp

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4531,6 +4531,19 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
45314531
CmdArgs.push_back(A->getValue());
45324532
}
45334533

4534+
if (Arg *A = Args.getLastArg(options::OPT_mtls_size_EQ)) {
4535+
StringRef Value = A->getValue();
4536+
unsigned TLSSize = 0;
4537+
Value.getAsInteger(10, TLSSize);
4538+
if (!Triple.isAArch64() || !Triple.isOSBinFormatELF())
4539+
D.Diag(diag::err_drv_unsupported_opt_for_target)
4540+
<< A->getOption().getName() << TripleStr;
4541+
if (TLSSize != 12 && TLSSize != 24 && TLSSize != 32 && TLSSize != 48)
4542+
D.Diag(diag::err_drv_invalid_int_value)
4543+
<< A->getOption().getName() << Value;
4544+
Args.AddLastArg(CmdArgs, options::OPT_mtls_size_EQ);
4545+
}
4546+
45344547
// Add the target cpu
45354548
std::string CPU = getCPUName(Args, Triple, /*FromAs*/ false);
45364549
if (!CPU.empty()) {

‎clang/lib/Frontend/CompilerInvocation.cpp

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1268,6 +1268,8 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
12681268
}
12691269
}
12701270

1271+
Opts.TLSSize = getLastArgIntValue(Args, OPT_mtls_size_EQ, 0, Diags);
1272+
12711273
if (Arg *A = Args.getLastArg(OPT_fdenormal_fp_math_EQ)) {
12721274
StringRef Val = A->getValue();
12731275
Opts.FPDenormalMode = llvm::parseDenormalFPAttribute(Val);

‎clang/test/Driver/tls-size.c

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,26 @@
1+
// Options for AArch64 ELF
2+
// RUN: %clang -### -target aarch64-linux-gnu -mtls-size=12 %s 2>&1 \
3+
// RUN: | FileCheck -check-prefix=CHECK-12 %s
4+
// RUN: %clang -### -target aarch64-linux-gnu -mtls-size=24 %s 2>&1 \
5+
// RUN: | FileCheck -check-prefix=CHECK-24 %s
6+
// RUN: %clang -### -target aarch64-linux-gnu -mtls-size=32 %s 2>&1 \
7+
// RUN: | FileCheck -check-prefix=CHECK-32 %s
8+
// RUN: %clang -### -target aarch64-linux-gnu -mtls-size=48 %s 2>&1 \
9+
// RUN: | FileCheck -check-prefix=CHECK-48 %s
10+
11+
// Unsupported target
12+
// RUN: not %clang -target aarch64-unknown-windows-msvc -mtls-size=24 %s 2>&1 \
13+
// RUN: | FileCheck -check-prefix=UNSUPPORTED-TARGET %s
14+
// RUN: not %clang -target x86_64-linux-gnu -mtls-size=24 %s 2>&1 \
15+
// RUN: | FileCheck -check-prefix=UNSUPPORTED-TARGET %s
16+
17+
// Invalid option value
18+
// RUN: not %clang -target aarch64-linux-gnu -mtls-size=0 %s 2>&1 \
19+
// RUN: | FileCheck -check-prefix=INVALID-VALUE %s
20+
21+
// CHECK-12: "-cc1" {{.*}}"-mtls-size=12"
22+
// CHECK-24: "-cc1" {{.*}}"-mtls-size=24"
23+
// CHECK-32: "-cc1" {{.*}}"-mtls-size=32"
24+
// CHECK-48: "-cc1" {{.*}}"-mtls-size=48"
25+
// UNSUPPORTED-TARGET: error: unsupported option
26+
// INVALID-VALUE: error: invalid integral value

‎llvm/include/llvm/CodeGen/CommandFlags.inc

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -238,6 +238,10 @@ static cl::opt<bool>
238238
cl::desc("Emit functions into separate sections"),
239239
cl::init(false));
240240

241+
static cl::opt<unsigned> TLSSize("tls-size",
242+
cl::desc("Bit size of immediate TLS offsets"),
243+
cl::init(0));
244+
241245
static cl::opt<bool> EmulatedTLS("emulated-tls",
242246
cl::desc("Use emulated TLS model"),
243247
cl::init(false));
@@ -305,6 +309,7 @@ static TargetOptions InitTargetOptionsFromCodeGenFlags() {
305309
Options.DataSections = DataSections;
306310
Options.FunctionSections = FunctionSections;
307311
Options.UniqueSectionNames = UniqueSectionNames;
312+
Options.TLSSize = TLSSize;
308313
Options.EmulatedTLS = EmulatedTLS;
309314
Options.ExplicitEmulatedTLS = EmulatedTLS.getNumOccurrences() > 0;
310315
Options.ExceptionModel = ExceptionModel;

‎llvm/include/llvm/Target/TargetOptions.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -231,6 +231,9 @@ namespace llvm {
231231
/// noreturn calls, even if TrapUnreachable is true.
232232
unsigned NoTrapAfterNoreturn : 1;
233233

234+
/// Bit size of immediate TLS offsets (0 == use the default).
235+
unsigned TLSSize : 8;
236+
234237
/// EmulatedTLS - This flag enables emulated TLS model, using emutls
235238
/// function in the runtime library..
236239
unsigned EmulatedTLS : 1;

‎llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

Lines changed: 104 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -4591,6 +4591,97 @@ AArch64TargetLowering::LowerDarwinGlobalTLSAddress(SDValue Op,
45914591
return DAG.getCopyFromReg(Chain, DL, AArch64::X0, PtrVT, Chain.getValue(1));
45924592
}
45934593

4594+
/// Convert a thread-local variable reference into a sequence of instructions to
4595+
/// compute the variable's address for the local exec TLS model of ELF targets.
4596+
/// The sequence depends on the maximum TLS area size.
4597+
SDValue AArch64TargetLowering::LowerELFTLSLocalExec(const GlobalValue *GV,
4598+
SDValue ThreadBase,
4599+
const SDLoc &DL,
4600+
SelectionDAG &DAG) const {
4601+
EVT PtrVT = getPointerTy(DAG.getDataLayout());
4602+
SDValue TPOff, Addr;
4603+
4604+
switch (DAG.getTarget().Options.TLSSize) {
4605+
default:
4606+
llvm_unreachable("Unexpected TLS size");
4607+
4608+
case 12: {
4609+
// mrs x0, TPIDR_EL0
4610+
// add x0, x0, :tprel_lo12:a
4611+
SDValue Var = DAG.getTargetGlobalAddress(
4612+
GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_PAGEOFF);
4613+
return SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, ThreadBase,
4614+
Var,
4615+
DAG.getTargetConstant(0, DL, MVT::i32)),
4616+
0);
4617+
}
4618+
4619+
case 24: {
4620+
// mrs x0, TPIDR_EL0
4621+
// add x0, x0, :tprel_hi12:a
4622+
// add x0, x0, :tprel_lo12_nc:a
4623+
SDValue HiVar = DAG.getTargetGlobalAddress(
4624+
GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_HI12);
4625+
SDValue LoVar = DAG.getTargetGlobalAddress(
4626+
GV, DL, PtrVT, 0,
4627+
AArch64II::MO_TLS | AArch64II::MO_PAGEOFF | AArch64II::MO_NC);
4628+
Addr = SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, ThreadBase,
4629+
HiVar,
4630+
DAG.getTargetConstant(0, DL, MVT::i32)),
4631+
0);
4632+
return SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, Addr,
4633+
LoVar,
4634+
DAG.getTargetConstant(0, DL, MVT::i32)),
4635+
0);
4636+
}
4637+
4638+
case 32: {
4639+
// mrs x1, TPIDR_EL0
4640+
// movz x0, #:tprel_g1:a
4641+
// movk x0, #:tprel_g0_nc:a
4642+
// add x0, x1, x0
4643+
SDValue HiVar = DAG.getTargetGlobalAddress(
4644+
GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_G1);
4645+
SDValue LoVar = DAG.getTargetGlobalAddress(
4646+
GV, DL, PtrVT, 0,
4647+
AArch64II::MO_TLS | AArch64II::MO_G0 | AArch64II::MO_NC);
4648+
TPOff = SDValue(DAG.getMachineNode(AArch64::MOVZXi, DL, PtrVT, HiVar,
4649+
DAG.getTargetConstant(16, DL, MVT::i32)),
4650+
0);
4651+
TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKXi, DL, PtrVT, TPOff, LoVar,
4652+
DAG.getTargetConstant(0, DL, MVT::i32)),
4653+
0);
4654+
return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadBase, TPOff);
4655+
}
4656+
4657+
case 48: {
4658+
// mrs x1, TPIDR_EL0
4659+
// movz x0, #:tprel_g2:a
4660+
// movk x0, #:tprel_g1_nc:a
4661+
// movk x0, #:tprel_g0_nc:a
4662+
// add x0, x1, x0
4663+
SDValue HiVar = DAG.getTargetGlobalAddress(
4664+
GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_G2);
4665+
SDValue MiVar = DAG.getTargetGlobalAddress(
4666+
GV, DL, PtrVT, 0,
4667+
AArch64II::MO_TLS | AArch64II::MO_G1 | AArch64II::MO_NC);
4668+
SDValue LoVar = DAG.getTargetGlobalAddress(
4669+
GV, DL, PtrVT, 0,
4670+
AArch64II::MO_TLS | AArch64II::MO_G0 | AArch64II::MO_NC);
4671+
TPOff = SDValue(DAG.getMachineNode(AArch64::MOVZXi, DL, PtrVT, HiVar,
4672+
DAG.getTargetConstant(32, DL, MVT::i32)),
4673+
0);
4674+
TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKXi, DL, PtrVT, TPOff, MiVar,
4675+
DAG.getTargetConstant(16, DL, MVT::i32)),
4676+
0);
4677+
TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKXi, DL, PtrVT, TPOff, LoVar,
4678+
DAG.getTargetConstant(0, DL, MVT::i32)),
4679+
0);
4680+
return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadBase, TPOff);
4681+
}
4682+
}
4683+
}
4684+
45944685
/// When accessing thread-local variables under either the general-dynamic or
45954686
/// local-dynamic system, we make a "TLS-descriptor" call. The variable will
45964687
/// have a descriptor, accessible via a PC-relative ADRP, and whose first entry
@@ -4628,15 +4719,7 @@ SDValue
46284719
AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
46294720
SelectionDAG &DAG) const {
46304721
assert(Subtarget->isTargetELF() && "This function expects an ELF target");
4631-
if (getTargetMachine().getCodeModel() == CodeModel::Large)
4632-
report_fatal_error("ELF TLS only supported in small memory model");
4633-
// Different choices can be made for the maximum size of the TLS area for a
4634-
// module. For the small address model, the default TLS size is 16MiB and the
4635-
// maximum TLS size is 4GiB.
4636-
// FIXME: add -mtls-size command line option and make it control the 16MiB
4637-
// vs. 4GiB code sequence generation.
4638-
// FIXME: add tiny codemodel support. We currently generate the same code as
4639-
// small, which may be larger than needed.
4722+
46404723
const GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
46414724

46424725
TLSModel::Model Model = getTargetMachine().getTLSModel(GA->getGlobal());
@@ -4646,6 +4729,17 @@ AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
46464729
Model = TLSModel::GeneralDynamic;
46474730
}
46484731

4732+
if (getTargetMachine().getCodeModel() == CodeModel::Large &&
4733+
Model != TLSModel::LocalExec)
4734+
report_fatal_error("ELF TLS only supported in small memory model or "
4735+
"in local exec TLS model");
4736+
// Different choices can be made for the maximum size of the TLS area for a
4737+
// module. For the small address model, the default TLS size is 16MiB and the
4738+
// maximum TLS size is 4GiB.
4739+
// FIXME: add tiny and large code model support for TLS access models other
4740+
// than local exec. We currently generate the same code as small for tiny,
4741+
// which may be larger than needed.
4742+
46494743
SDValue TPOff;
46504744
EVT PtrVT = getPointerTy(DAG.getDataLayout());
46514745
SDLoc DL(Op);
@@ -4654,23 +4748,7 @@ AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
46544748
SDValue ThreadBase = DAG.getNode(AArch64ISD::THREAD_POINTER, DL, PtrVT);
46554749

46564750
if (Model == TLSModel::LocalExec) {
4657-
SDValue HiVar = DAG.getTargetGlobalAddress(
4658-
GV, DL, PtrVT, 0, AArch64II::MO_TLS | AArch64II::MO_HI12);
4659-
SDValue LoVar = DAG.getTargetGlobalAddress(
4660-
GV, DL, PtrVT, 0,
4661-
AArch64II::MO_TLS | AArch64II::MO_PAGEOFF | AArch64II::MO_NC);
4662-
4663-
SDValue TPWithOff_lo =
4664-
SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, ThreadBase,
4665-
HiVar,
4666-
DAG.getTargetConstant(0, DL, MVT::i32)),
4667-
0);
4668-
SDValue TPWithOff =
4669-
SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, TPWithOff_lo,
4670-
LoVar,
4671-
DAG.getTargetConstant(0, DL, MVT::i32)),
4672-
0);
4673-
return TPWithOff;
4751+
return LowerELFTLSLocalExec(GV, ThreadBase, DL, DAG);
46744752
} else if (Model == TLSModel::InitialExec) {
46754753
TPOff = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_TLS);
46764754
TPOff = DAG.getNode(AArch64ISD::LOADgot, DL, PtrVT, TPOff);

‎llvm/lib/Target/AArch64/AArch64ISelLowering.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -696,6 +696,8 @@ class AArch64TargetLowering : public TargetLowering {
696696
SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
697697
SDValue LowerDarwinGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
698698
SDValue LowerELFGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
699+
SDValue LowerELFTLSLocalExec(const GlobalValue *GV, SDValue ThreadBase,
700+
const SDLoc &DL, SelectionDAG &DAG) const;
699701
SDValue LowerELFTLSDescCallSeq(SDValue SymAddr, const SDLoc &DL,
700702
SelectionDAG &DAG) const;
701703
SDValue LowerWindowsGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;

0 commit comments

Comments
 (0)