1 //===-- llvm/Target/TargetOptions.h - Target Options ------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file defines command line option flags that are shared across various
10 // targets.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_TARGET_TARGETOPTIONS_H
15 #define LLVM_TARGET_TARGETOPTIONS_H
16 
17 #include "llvm/ADT/FloatingPointMode.h"
18 #include "llvm/MC/MCTargetOptions.h"
19 
20 #include <memory>
21 
22 namespace llvm {
23   struct fltSemantics;
24   class MachineFunction;
25   class MemoryBuffer;
26 
27   namespace FloatABI {
28     enum ABIType {
29       Default, // Target-specific (either soft or hard depending on triple, etc).
30       Soft,    // Soft float.
31       Hard     // Hard float.
32     };
33   }
34 
35   namespace FPOpFusion {
36     enum FPOpFusionMode {
37       Fast,     // Enable fusion of FP ops wherever it's profitable.
38       Standard, // Only allow fusion of 'blessed' ops (currently just fmuladd).
39       Strict    // Never fuse FP-ops.
40     };
41   }
42 
43   namespace JumpTable {
44     enum JumpTableType {
45       Single,          // Use a single table for all indirect jumptable calls.
46       Arity,           // Use one table per number of function parameters.
47       Simplified,      // Use one table per function type, with types projected
48                        // into 4 types: pointer to non-function, struct,
49                        // primitive, and function pointer.
50       Full             // Use one table per unique function type
51     };
52   }
53 
54   namespace ThreadModel {
55     enum Model {
56       POSIX,  // POSIX Threads
57       Single  // Single Threaded Environment
58     };
59   }
60 
61   enum class BasicBlockSection {
62     All,    // Use Basic Block Sections for all basic blocks.  A section
63             // for every basic block can significantly bloat object file sizes.
64     List,   // Get list of functions & BBs from a file. Selectively enables
65             // basic block sections for a subset of basic blocks which can be
66             // used to control object size bloats from creating sections.
67     Labels, // Do not use Basic Block Sections but label basic blocks.  This
68             // is useful when associating profile counts from virtual addresses
69             // to basic blocks.
70     Preset, // Similar to list but the blocks are identified by passes which
71             // seek to use Basic Block Sections, e.g. MachineFunctionSplitter.
72             // This option cannot be set via the command line.
73     None    // Do not use Basic Block Sections.
74   };
75 
76   enum class StackProtectorGuards {
77     None,
78     TLS,
79     Global
80   };
81 
82   enum class EABI {
83     Unknown,
84     Default, // Default means not specified
85     EABI4,   // Target-specific (either 4, 5 or gnu depending on triple).
86     EABI5,
87     GNU
88   };
89 
90   /// Identify a debugger for "tuning" the debug info.
91   ///
92   /// The "debugger tuning" concept allows us to present a more intuitive
93   /// interface that unpacks into different sets of defaults for the various
94   /// individual feature-flag settings, that suit the preferences of the
95   /// various debuggers.  However, it's worth remembering that debuggers are
96   /// not the only consumers of debug info, and some variations in DWARF might
97   /// better be treated as target/platform issues. Fundamentally,
98   /// o if the feature is useful (or not) to a particular debugger, regardless
99   ///   of the target, that's a tuning decision;
100   /// o if the feature is useful (or not) on a particular platform, regardless
101   ///   of the debugger, that's a target decision.
102   /// It's not impossible to see both factors in some specific case.
103   ///
104   /// The "tuning" should be used to set defaults for individual feature flags
105   /// in DwarfDebug; if a given feature has a more specific command-line option,
106   /// that option should take precedence over the tuning.
107   enum class DebuggerKind {
108     Default,  // No specific tuning requested.
109     GDB,      // Tune debug info for gdb.
110     LLDB,     // Tune debug info for lldb.
111     SCE       // Tune debug info for SCE targets (e.g. PS4).
112   };
113 
114   /// Enable abort calls when global instruction selection fails to lower/select
115   /// an instruction.
116   enum class GlobalISelAbortMode {
117     Disable,        // Disable the abort.
118     Enable,         // Enable the abort.
119     DisableWithDiag // Disable the abort but emit a diagnostic on failure.
120   };
121 
122   class TargetOptions {
123   public:
TargetOptions()124     TargetOptions()
125         : UnsafeFPMath(false), NoInfsFPMath(false), NoNaNsFPMath(false),
126           NoTrappingFPMath(true), NoSignedZerosFPMath(false),
127           HonorSignDependentRoundingFPMathOption(false), NoZerosInBSS(false),
128           GuaranteedTailCallOpt(false), StackSymbolOrdering(true),
129           EnableFastISel(false), EnableGlobalISel(false), UseInitArray(false),
130           DisableIntegratedAS(false), RelaxELFRelocations(false),
131           FunctionSections(false), DataSections(false),
132           IgnoreXCOFFVisibility(false), UniqueSectionNames(true),
133           UniqueBasicBlockSectionNames(false), TrapUnreachable(false),
134           NoTrapAfterNoreturn(false), TLSSize(0), EmulatedTLS(false),
135           ExplicitEmulatedTLS(false), EnableIPRA(false),
136           EmitStackSizeSection(false), EnableMachineOutliner(false),
137           EnableMachineFunctionSplitter(false), SupportsDefaultOutlining(false),
138           EmitAddrsig(false), EmitCallSiteInfo(false),
139           SupportsDebugEntryValues(false), EnableDebugEntryValues(false),
140           ValueTrackingVariableLocations(false), ForceDwarfFrameSection(false),
141           XRayOmitFunctionIndex(false),
142           FPDenormalMode(DenormalMode::IEEE, DenormalMode::IEEE) {}
143 
144     /// DisableFramePointerElim - This returns true if frame pointer elimination
145     /// optimization should be disabled for the given machine function.
146     bool DisableFramePointerElim(const MachineFunction &MF) const;
147 
148     /// UnsafeFPMath - This flag is enabled when the
149     /// -enable-unsafe-fp-math flag is specified on the command line.  When
150     /// this flag is off (the default), the code generator is not allowed to
151     /// produce results that are "less precise" than IEEE allows.  This includes
152     /// use of X86 instructions like FSIN and FCOS instead of libcalls.
153     unsigned UnsafeFPMath : 1;
154 
155     /// NoInfsFPMath - This flag is enabled when the
156     /// -enable-no-infs-fp-math flag is specified on the command line. When
157     /// this flag is off (the default), the code generator is not allowed to
158     /// assume the FP arithmetic arguments and results are never +-Infs.
159     unsigned NoInfsFPMath : 1;
160 
161     /// NoNaNsFPMath - This flag is enabled when the
162     /// -enable-no-nans-fp-math flag is specified on the command line. When
163     /// this flag is off (the default), the code generator is not allowed to
164     /// assume the FP arithmetic arguments and results are never NaNs.
165     unsigned NoNaNsFPMath : 1;
166 
167     /// NoTrappingFPMath - This flag is enabled when the
168     /// -enable-no-trapping-fp-math is specified on the command line. This
169     /// specifies that there are no trap handlers to handle exceptions.
170     unsigned NoTrappingFPMath : 1;
171 
172     /// NoSignedZerosFPMath - This flag is enabled when the
173     /// -enable-no-signed-zeros-fp-math is specified on the command line. This
174     /// specifies that optimizations are allowed to treat the sign of a zero
175     /// argument or result as insignificant.
176     unsigned NoSignedZerosFPMath : 1;
177 
178     /// HonorSignDependentRoundingFPMath - This returns true when the
179     /// -enable-sign-dependent-rounding-fp-math is specified.  If this returns
180     /// false (the default), the code generator is allowed to assume that the
181     /// rounding behavior is the default (round-to-zero for all floating point
182     /// to integer conversions, and round-to-nearest for all other arithmetic
183     /// truncations).  If this is enabled (set to true), the code generator must
184     /// assume that the rounding mode may dynamically change.
185     unsigned HonorSignDependentRoundingFPMathOption : 1;
186     bool HonorSignDependentRoundingFPMath() const;
187 
188     /// NoZerosInBSS - By default some codegens place zero-initialized data to
189     /// .bss section. This flag disables such behaviour (necessary, e.g. for
190     /// crt*.o compiling).
191     unsigned NoZerosInBSS : 1;
192 
193     /// GuaranteedTailCallOpt - This flag is enabled when -tailcallopt is
194     /// specified on the commandline. When the flag is on, participating targets
195     /// will perform tail call optimization on all calls which use the fastcc
196     /// calling convention and which satisfy certain target-independent
197     /// criteria (being at the end of a function, having the same return type
198     /// as their parent function, etc.), using an alternate ABI if necessary.
199     unsigned GuaranteedTailCallOpt : 1;
200 
201     /// StackAlignmentOverride - Override default stack alignment for target.
202     unsigned StackAlignmentOverride = 0;
203 
204     /// StackSymbolOrdering - When true, this will allow CodeGen to order
205     /// the local stack symbols (for code size, code locality, or any other
206     /// heuristics). When false, the local symbols are left in whatever order
207     /// they were generated. Default is true.
208     unsigned StackSymbolOrdering : 1;
209 
210     /// EnableFastISel - This flag enables fast-path instruction selection
211     /// which trades away generated code quality in favor of reducing
212     /// compile time.
213     unsigned EnableFastISel : 1;
214 
215     /// EnableGlobalISel - This flag enables global instruction selection.
216     unsigned EnableGlobalISel : 1;
217 
218     /// EnableGlobalISelAbort - Control abort behaviour when global instruction
219     /// selection fails to lower/select an instruction.
220     GlobalISelAbortMode GlobalISelAbort = GlobalISelAbortMode::Enable;
221 
222     /// UseInitArray - Use .init_array instead of .ctors for static
223     /// constructors.
224     unsigned UseInitArray : 1;
225 
226     /// Disable the integrated assembler.
227     unsigned DisableIntegratedAS : 1;
228 
229     /// Compress DWARF debug sections.
230     DebugCompressionType CompressDebugSections = DebugCompressionType::None;
231 
232     unsigned RelaxELFRelocations : 1;
233 
234     /// Emit functions into separate sections.
235     unsigned FunctionSections : 1;
236 
237     /// Emit data into separate sections.
238     unsigned DataSections : 1;
239 
240     /// Do not emit visibility attribute for xcoff.
241     unsigned IgnoreXCOFFVisibility : 1;
242 
243     unsigned UniqueSectionNames : 1;
244 
245     /// Use unique names for basic block sections.
246     unsigned UniqueBasicBlockSectionNames : 1;
247 
248     /// Emit target-specific trap instruction for 'unreachable' IR instructions.
249     unsigned TrapUnreachable : 1;
250 
251     /// Do not emit a trap instruction for 'unreachable' IR instructions behind
252     /// noreturn calls, even if TrapUnreachable is true.
253     unsigned NoTrapAfterNoreturn : 1;
254 
255     /// Bit size of immediate TLS offsets (0 == use the default).
256     unsigned TLSSize : 8;
257 
258     /// EmulatedTLS - This flag enables emulated TLS model, using emutls
259     /// function in the runtime library..
260     unsigned EmulatedTLS : 1;
261 
262     /// Whether -emulated-tls or -no-emulated-tls is set.
263     unsigned ExplicitEmulatedTLS : 1;
264 
265     /// This flag enables InterProcedural Register Allocation (IPRA).
266     unsigned EnableIPRA : 1;
267 
268     /// Emit section containing metadata on function stack sizes.
269     unsigned EmitStackSizeSection : 1;
270 
271     /// Enables the MachineOutliner pass.
272     unsigned EnableMachineOutliner : 1;
273 
274     /// Enables the MachineFunctionSplitter pass.
275     unsigned EnableMachineFunctionSplitter : 1;
276 
277     /// Set if the target supports default outlining behaviour.
278     unsigned SupportsDefaultOutlining : 1;
279 
280     /// Emit address-significance table.
281     unsigned EmitAddrsig : 1;
282 
283     /// Emit basic blocks into separate sections.
284     BasicBlockSection BBSections = BasicBlockSection::None;
285 
286     /// Memory Buffer that contains information on sampled basic blocks and used
287     /// to selectively generate basic block sections.
288     std::shared_ptr<MemoryBuffer> BBSectionsFuncListBuf;
289 
290     /// The flag enables call site info production. It is used only for debug
291     /// info, and it is restricted only to optimized code. This can be used for
292     /// something else, so that should be controlled in the frontend.
293     unsigned EmitCallSiteInfo : 1;
294     /// Set if the target supports the debug entry values by default.
295     unsigned SupportsDebugEntryValues : 1;
296     /// When set to true, the EnableDebugEntryValues option forces production
297     /// of debug entry values even if the target does not officially support
298     /// it. Useful for testing purposes only. This flag should never be checked
299     /// directly, always use \ref ShouldEmitDebugEntryValues instead.
300      unsigned EnableDebugEntryValues : 1;
301     /// NOTE: There are targets that still do not support the debug entry values
302     /// production.
303     bool ShouldEmitDebugEntryValues() const;
304 
305     // When set to true, use experimental new debug variable location tracking,
306     // which seeks to follow the values of variables rather than their location,
307     // post isel.
308     unsigned ValueTrackingVariableLocations : 1;
309 
310     /// Emit DWARF debug frame section.
311     unsigned ForceDwarfFrameSection : 1;
312 
313     /// Emit XRay Function Index section
314     unsigned XRayOmitFunctionIndex : 1;
315 
316     /// Stack protector guard offset to use.
317     unsigned StackProtectorGuardOffset : 32;
318 
319     /// Stack protector guard mode to use, e.g. tls, global.
320     StackProtectorGuards StackProtectorGuard =
321                                          StackProtectorGuards::None;
322 
323     /// Stack protector guard reg to use, e.g. usually fs or gs in X86.
324     std::string StackProtectorGuardReg = "None";
325 
326     /// FloatABIType - This setting is set by -float-abi=xxx option is specfied
327     /// on the command line. This setting may either be Default, Soft, or Hard.
328     /// Default selects the target's default behavior. Soft selects the ABI for
329     /// software floating point, but does not indicate that FP hardware may not
330     /// be used. Such a combination is unfortunately popular (e.g.
331     /// arm-apple-darwin). Hard presumes that the normal FP ABI is used.
332     FloatABI::ABIType FloatABIType = FloatABI::Default;
333 
334     /// AllowFPOpFusion - This flag is set by the -fuse-fp-ops=xxx option.
335     /// This controls the creation of fused FP ops that store intermediate
336     /// results in higher precision than IEEE allows (E.g. FMAs).
337     ///
338     /// Fast mode - allows formation of fused FP ops whenever they're
339     /// profitable.
340     /// Standard mode - allow fusion only for 'blessed' FP ops. At present the
341     /// only blessed op is the fmuladd intrinsic. In the future more blessed ops
342     /// may be added.
343     /// Strict mode - allow fusion only if/when it can be proven that the excess
344     /// precision won't effect the result.
345     ///
346     /// Note: This option only controls formation of fused ops by the
347     /// optimizers.  Fused operations that are explicitly specified (e.g. FMA
348     /// via the llvm.fma.* intrinsic) will always be honored, regardless of
349     /// the value of this option.
350     FPOpFusion::FPOpFusionMode AllowFPOpFusion = FPOpFusion::Standard;
351 
352     /// ThreadModel - This flag specifies the type of threading model to assume
353     /// for things like atomics
354     ThreadModel::Model ThreadModel = ThreadModel::POSIX;
355 
356     /// EABIVersion - This flag specifies the EABI version
357     EABI EABIVersion = EABI::Default;
358 
359     /// Which debugger to tune for.
360     DebuggerKind DebuggerTuning = DebuggerKind::Default;
361 
362   private:
363     /// Flushing mode to assume in default FP environment.
364     DenormalMode FPDenormalMode;
365 
366     /// Flushing mode to assume in default FP environment, for float/vector of
367     /// float.
368     DenormalMode FP32DenormalMode;
369 
370   public:
setFPDenormalMode(DenormalMode Mode)371     void setFPDenormalMode(DenormalMode Mode) {
372       FPDenormalMode = Mode;
373     }
374 
setFP32DenormalMode(DenormalMode Mode)375     void setFP32DenormalMode(DenormalMode Mode) {
376       FP32DenormalMode = Mode;
377     }
378 
getRawFPDenormalMode()379     DenormalMode getRawFPDenormalMode() const {
380       return FPDenormalMode;
381     }
382 
getRawFP32DenormalMode()383     DenormalMode getRawFP32DenormalMode() const {
384       return FP32DenormalMode;
385     }
386 
387     DenormalMode getDenormalMode(const fltSemantics &FPType) const;
388 
389     /// What exception model to use
390     ExceptionHandling ExceptionModel = ExceptionHandling::None;
391 
392     /// Machine level options.
393     MCTargetOptions MCOptions;
394   };
395 
396 } // End llvm namespace
397 
398 #endif
399