1.. _amdgpu-dwarf-extensions-for-heterogeneous-debugging:
2
3********************************************
4DWARF Extensions For Heterogeneous Debugging
5********************************************
6
7.. contents::
8   :local:
9
10.. warning::
11
12   This document describes **provisional extensions** to DWARF Version 5
13   [:ref:`DWARF <amdgpu-dwarf-DWARF>`] to support heterogeneous debugging. It is
14   not currently fully implemented and is subject to change.
15
16.. _amdgpu-dwarf-introduction:
17
18Introduction
19============
20
21AMD [:ref:`AMD <amdgpu-dwarf-AMD>`] has been working on supporting heterogeneous
22computing through the AMD Radeon Open Compute Platform (ROCm) [:ref:`AMD-ROCm
23<amdgpu-dwarf-AMD-ROCm>`]. A heterogeneous computing program can be written in a
24high level language such as C++ or Fortran with OpenMP pragmas, OpenCL, or HIP
25(a portable C++ programming environment for heterogeneous computing [:ref:`HIP
26<amdgpu-dwarf-HIP>`]). A heterogeneous compiler and runtime allows a program to
27execute on multiple devices within the same native process. Devices could
28include CPUs, GPUs, DSPs, FPGAs, or other special purpose accelerators.
29Currently HIP programs execute on systems with CPUs and GPUs.
30
31ROCm is fully open sourced and includes contributions to open source projects
32such as LLVM for compilation [:ref:`LLVM <amdgpu-dwarf-LLVM>`] and GDB for
33debugging [:ref:`GDB <amdgpu-dwarf-GDB>`], as well as collaboration with other
34third party projects such as the GCC compiler [:ref:`GCC <amdgpu-dwarf-GCC>`]
35and the Perforce TotalView HPC debugger [:ref:`Perforce-TotalView
36<amdgpu-dwarf-Perforce-TotalView>`].
37
38To support debugging heterogeneous programs several features that are not
39provided by current DWARF Version 5 [:ref:`DWARF <amdgpu-dwarf-DWARF>`] have
40been identified. This document contains a collection of extensions to address
41providing those features.
42
43The :ref:`amdgpu-dwarf-motivation` section describes the issues that are being
44addressed for heterogeneous computing. That is followed by the
45:ref:`amdgpu-dwarf-changes-relative-to-dwarf-version-5` section containing the
46textual changes for the extensions relative to the DWARF Version 5 standard.
47Then there is an :ref:`amdgpu-dwarf-examples` section that links to the AMD GPU
48specific usage of the extensions that includes an example. Finally, there is a
49:ref:`amdgpu-dwarf-references` section. There are a number of notes included
50that raise open questions, or provide alternative approaches considered. The
51extensions seek to be general in nature and backwards compatible with DWARF
52Version 5. The goal is to be applicable to meeting the needs of any
53heterogeneous system and not be vendor or architecture specific.
54
55A fundamental aspect of the extensions is that it allows DWARF expression
56location descriptions as stack elements. The extensions are based on DWARF
57Version 5 and maintains compatibility with DWARF Version 5. After attempting
58several alternatives, the current thinking is that such extensions to DWARF
59Version 5 are the simplest and cleanest ways to support debugging optimized GPU
60code. It also appears to be generally useful and may be able to address other
61reported DWARF issues, as well as being helpful in providing better optimization
62support for non-GPU code.
63
64General feedback on these extensions is sought, together with suggestions on how
65to clarify, simplify, or organize them. If their is general interest then some
66or all of these extensions could be submitted as future DWARF proposals.
67
68We are in the process of modifying LLVM and GDB to support these extensions
69which is providing experience and insights. We plan to upstream the changes to
70those projects for any final form of the extensions.
71
72The author very much appreciates the input provided so far by many others which
73has been incorporated into this current version.
74
75.. _amdgpu-dwarf-motivation:
76
77Motivation
78==========
79
80This document presents a set of backwards compatible extensions to DWARF Version
815 [:ref:`DWARF <amdgpu-dwarf-DWARF>`] to support heterogeneous debugging.
82
83The remainder of this section provides motivation for each extension in
84terms of heterogeneous debugging on commercially available AMD GPU hardware
85(AMDGPU). The goal is to add support to the AMD [:ref:`AMD <amdgpu-dwarf-AMD>`]
86open source Radeon Open Compute Platform (ROCm) [:ref:`AMD-ROCm
87<amdgpu-dwarf-AMD-ROCm>`] which is an implementation of the industry standard
88for heterogeneous computing devices defined by the Heterogeneous System
89Architecture (HSA) Foundation [:ref:`HSA <amdgpu-dwarf-HSA>`]. ROCm includes the
90LLVM compiler [:ref:`LLVM <amdgpu-dwarf-LLVM>`] with upstreamed support for
91AMDGPU [:ref:`AMDGPU-LLVM <amdgpu-dwarf-AMDGPU-LLVM>`]. The goal is to also add
92the GDB debugger [:ref:`GDB <amdgpu-dwarf-GDB>`] with upstreamed support for
93AMDGPU [:ref:`AMD-ROCgdb <amdgpu-dwarf-AMD-ROCgdb>`]. In addition, the goal is
94to work with third parties to enable support for AMDGPU debugging in the GCC
95compiler [:ref:`GCC <amdgpu-dwarf-GCC>`] and the Perforce TotalView HPC debugger
96[:ref:`Perforce-TotalView <amdgpu-dwarf-Perforce-TotalView>`].
97
98However, the extensions are intended to be vendor and architecture neutral. They
99are believed to apply to other heterogeneous hardware devices including GPUs,
100DSPs, FPGAs, and other specialized hardware. These collectively include similar
101characteristics and requirements as AMDGPU devices. Some of the extension can
102also apply to traditional CPU hardware that supports large vector registers.
103Compilers can map source languages and extensions that describe large scale
104parallel execution onto the lanes of the vector registers. This is common in
105programming languages used in ML and HPC. The extensions also include improved
106support for optimized code on any architecture. Some of the generalizations may
107also benefit other issues that have been raised.
108
109The extensions have evolved through collaboration with many individuals and
110active prototyping within the GDB debugger and LLVM compiler. Input has also
111been very much appreciated from the developers working on the Perforce TotalView
112HPC Debugger and GCC compiler.
113
114The AMDGPU has several features that require additional DWARF functionality in
115order to support optimized code.
116
117AMDGPU optimized code may spill vector registers to non-global address space
118memory, and this spilling may be done only for lanes that are active on entry
119to the subprogram. To support this, a location description that can be created
120as a masked select is required. See ``DW_OP_LLVM_select_bit_piece``.
121
122Since the active lane mask may be held in a register, a way to get the value
123of a register on entry to a subprogram is required. To support this an
124operation that returns the caller value of a register as specified by the Call
125Frame Information (CFI) is required. See ``DW_OP_LLVM_call_frame_entry_reg``
126and :ref:`amdgpu-dwarf-call-frame-information`.
127
128Current DWARF uses an empty expression to indicate an undefined location
129description. Since the masked select composite location description operation
130takes more than one location description, it is necessary to have an explicit
131way to specify an undefined location description. Otherwise it is not possible
132to specify that a particular one of the input location descriptions is
133undefined. See ``DW_OP_LLVM_undefined``.
134
135CFI describes restoring callee saved registers that are spilled. Currently CFI
136only allows a location description that is a register, memory address, or
137implicit location description. AMDGPU optimized code may spill scalar
138registers into portions of vector registers. This requires extending CFI to
139allow any location description. See
140:ref:`amdgpu-dwarf-call-frame-information`.
141
142The vector registers of the AMDGPU are represented as their full wavefront
143size, meaning the wavefront size times the dword size. This reflects the
144actual hardware and allows the compiler to generate DWARF for languages that
145map a thread to the complete wavefront. It also allows more efficient DWARF to
146be generated to describe the CFI as only a single expression is required for
147the whole vector register, rather than a separate expression for each lane's
148dword of the vector register. It also allows the compiler to produce DWARF
149that indexes the vector register if it spills scalar registers into portions
150of a vector register.
151
152Since DWARF stack value entries have a base type and AMDGPU registers are a
153vector of dwords, the ability to specify that a base type is a vector is
154required. See ``DW_AT_LLVM_vector_size``.
155
156If the source language is mapped onto the AMDGPU wavefronts in a SIMT manner,
157then the variable DWARF location expressions must compute the location for a
158single lane of the wavefront. Therefore, a DWARF operation is required to denote
159the current lane, much like ``DW_OP_push_object_address`` denotes the current
160object. The ``DW_OP_*piece`` operations only allow literal indices. Therefore, a
161way to use a computed offset of an arbitrary location description (such as a
162vector register) is required. See ``DW_OP_LLVM_push_lane``,
163``DW_OP_LLVM_offset``, ``DW_OP_LLVM_offset_uconst``, and
164``DW_OP_LLVM_bit_offset``.
165
166If the source language is mapped onto the AMDGPU wavefronts in a SIMT manner
167the compiler can use the AMDGPU execution mask register to control which lanes
168are active. To describe the conceptual location of non-active lanes a DWARF
169expression is needed that can compute a per lane PC. For efficiency, this is
170done for the wavefront as a whole. This expression benefits by having a masked
171select composite location description operation. This requires an attribute
172for source location of each lane. The AMDGPU may update the execution mask for
173whole wavefront operations and so needs an attribute that computes the current
174active lane mask. See ``DW_OP_LLVM_select_bit_piece``, ``DW_OP_LLVM_extend``,
175``DW_AT_LLVM_lane_pc``, and ``DW_AT_LLVM_active_lane``.
176
177AMDGPU needs to be able to describe addresses that are in different kinds of
178memory. Optimized code may need to describe a variable that resides in pieces
179that are in different kinds of storage which may include parts of registers,
180memory that is in a mixture of memory kinds, implicit values, or be undefined.
181DWARF has the concept of segment addresses. However, the segment cannot be
182specified within a DWARF expression, which is only able to specify the offset
183portion of a segment address. The segment index is only provided by the entity
184that specifies the DWARF expression. Therefore, the segment index is a
185property that can only be put on complete objects, such as a variable. That
186makes it only suitable for describing an entity (such as variable or
187subprogram code) that is in a single kind of memory. Therefore, AMDGPU uses
188the DWARF concept of address spaces. For example, a variable may be allocated
189in a register that is partially spilled to the call stack which is in the
190private address space, and partially spilled to the local address space.
191
192DWARF uses the concept of an address in many expression operations but does not
193define how it relates to address spaces. For example,
194``DW_OP_push_object_address`` pushes the address of an object. Other contexts
195implicitly push an address on the stack before evaluating an expression. For
196example, the ``DW_AT_use_location`` attribute of the
197``DW_TAG_ptr_to_member_type``. The expression that uses the address needs to
198do so in a general way and not need to be dependent on the address space of
199the address. For example, a pointer to member value may want to be applied to
200an object that may reside in any address space.
201
202The number of registers and the cost of memory operations is much higher for
203AMDGPU than a typical CPU. The compiler attempts to optimize whole variables
204and arrays into registers. Currently DWARF only allows
205``DW_OP_push_object_address`` and related operations to work with a global
206memory location. To support AMDGPU optimized code it is required to generalize
207DWARF to allow any location description to be used. This allows registers, or
208composite location descriptions that may be a mixture of memory, registers, or
209even implicit values.
210
211DWARF Version 5 does not allow location descriptions to be entries on the
212DWARF stack. They can only be the final result of the evaluation of a DWARF
213expression. However, by allowing a location description to be a first-class
214entry on the DWARF stack it becomes possible to compose expressions containing
215both values and location descriptions naturally. It allows objects to be
216located in any kind of memory address space, in registers, be implicit values,
217be undefined, or a composite of any of these. By extending DWARF carefully,
218all existing DWARF expressions can retain their current semantic meaning.
219DWARF has implicit conversions that convert from a value that represents an
220address in the default address space to a memory location description. This
221can be extended to allow a default address space memory location description
222to be implicitly converted back to its address value. This allows all DWARF
223Version 5 expressions to retain their same meaning, while adding the ability
224to explicitly create memory location descriptions in non-default address
225spaces and generalizing the power of composite location descriptions to any
226kind of location description. See :ref:`amdgpu-dwarf-operation-expressions`.
227
228To allow composition of composite location descriptions, an explicit operation
229that indicates the end of the definition of a composite location description
230is required. This can be implied if the end of a DWARF expression is reached,
231allowing current DWARF expressions to remain legal. See
232``DW_OP_LLVM_piece_end``.
233
234The ``DW_OP_plus`` and ``DW_OP_minus`` can be defined to operate on a memory
235location description in the default target architecture specific address space
236and a generic type value to produce an updated memory location description. This
237allows them to continue to be used to offset an address. To generalize
238offsetting to any location description, including location descriptions that
239describe when bytes are in registers, are implicit, or a composite of these, the
240``DW_OP_LLVM_offset``, ``DW_OP_LLVM_offset_uconst``, and
241``DW_OP_LLVM_bit_offset`` offset operations are added. Unlike ``DW_OP_plus``,
242``DW_OP_plus_uconst``, and ``DW_OP_minus`` arithmetic operations, these do not
243define that integer overflow causes wrap-around. The offset operations can
244operate on location storage of any size. For example, implicit location storage
245could be any number of bits in size. It is simpler to define offsets that exceed
246the size of the location storage as being an evaluation error, than having to
247force an implementation to support potentially infinite precision offsets to
248allow it to correctly track a series of positive and negative offsets that may
249transiently overflow or underflow, but end up in range. This is simple for the
250arithmetic operations as they are defined in terms of two's compliment
251arithmetic on a base type of a fixed size.
252
253Having the offset operations allows ``DW_OP_push_object_address`` to push a
254location description that may be in a register, or be an implicit value, and the
255DWARF expression of ``DW_TAG_ptr_to_member_type`` can contain them to offset
256within it. ``DW_OP_LLVM_bit_offset`` generalizes DWARF to work with bit fields
257which is not possible in DWARF Version 5.
258
259The DWARF ``DW_OP_xderef*`` operations allow a value to be converted into an
260address of a specified address space which is then read. But it provides no
261way to create a memory location description for an address in the non-default
262address space. For example, AMDGPU variables can be allocated in the local
263address space at a fixed address. It is required to have an operation to
264create an address in a specific address space that can be used to define the
265location description of the variable. Defining this operation to produce a
266location description allows the size of addresses in an address space to be
267larger than the generic type. See ``DW_OP_LLVM_form_aspace_address``.
268
269If the ``DW_OP_LLVM_form_aspace_address`` operation had to produce a value
270that can be implicitly converted to a memory location description, then it
271would be limited to the size of the generic type which matches the size of the
272default address space. Its value would be undefined and likely not match any
273value in the actual program. By making the result a location description, it
274allows a consumer great freedom in how it implements it. The implicit
275conversion back to a value can be limited only to the default address space to
276maintain compatibility with DWARF Version 5. For other address spaces the
277producer can use the new operations that explicitly specify the address space.
278
279``DW_OP_breg*`` treats the register as containing an address in the default
280address space. It is required to be able to specify the address space of the
281register value. See ``DW_OP_LLVM_aspace_bregx``.
282
283Similarly, ``DW_OP_implicit_pointer`` treats its implicit pointer value as
284being in the default address space. It is required to be able to specify the
285address space of the pointer value. See
286``DW_OP_LLVM_aspace_implicit_pointer``.
287
288Almost all uses of addresses in DWARF are limited to defining location
289descriptions, or to be dereferenced to read memory. The exception is
290``DW_CFA_val_offset`` which uses the address to set the value of a register.
291By defining the CFA DWARF expression as being a memory location description,
292it can maintain what address space it is, and that can be used to convert the
293offset address back to an address in that address space. See
294:ref:`amdgpu-dwarf-call-frame-information`.
295
296This approach allows all existing DWARF to have the identical semantics. It
297allows the compiler to explicitly specify the address space it is using. For
298example, a compiler could choose to access private memory in a swizzled manner
299when mapping a source language to a wavefront in a SIMT manner, or to access
300it in an unswizzled manner if mapping the same language with the wavefront
301being the thread. It also allows the compiler to mix the address space it uses
302to access private memory. For example, for SIMT it can still spill entire
303vector registers in an unswizzled manner, while using a swizzled private
304memory for SIMT variable access. This approach allows memory location
305descriptions for different address spaces to be combined using the regular
306``DW_OP_*piece`` operations.
307
308Location descriptions are an abstraction of storage, they give freedom to the
309consumer on how to implement them. They allow the address space to encode lane
310information so they can be used to read memory with only the memory
311description and no extra arguments. The same set of operations can operate on
312locations independent of their kind of storage. The ``DW_OP_deref*`` therefore
313can be used on any storage kind. ``DW_OP_xderef*`` is unnecessary, except to
314become a more compact way to convert a non-default address space address
315followed by dereferencing it.
316
317In DWARF Version 5 a location description is defined as a single location
318description or a location list. A location list is defined as either
319effectively an undefined location description or as one or more single
320location descriptions to describe an object with multiple places. The
321``DW_OP_push_object_address`` and ``DW_OP_call*`` operations can put a
322location description on the stack. Furthermore, debugger information entry
323attributes such as ``DW_AT_data_member_location``, ``DW_AT_use_location``, and
324``DW_AT_vtable_elem_location`` are defined as pushing a location description
325on the expression stack before evaluating the expression. However, DWARF
326Version 5 only allows the stack to contain values and so only a single memory
327address can be on the stack which makes these incapable of handling location
328descriptions with multiple places, or places other than memory. Since these
329extensions allow the stack to contain location descriptions, the operations are
330generalized to support location descriptions that can have multiple places.
331This is backwards compatible with DWARF Version 5 and allows objects with
332multiple places to be supported. For example, the expression that describes
333how to access the field of an object can be evaluated with a location
334description that has multiple places and will result in a location description
335with multiple places as expected. With this change, the separate DWARF Version
3365 sections that described DWARF expressions and location lists have been
337unified into a single section that describes DWARF expressions in general.
338This unification seems to be a natural consequence and a necessity of allowing
339location descriptions to be part of the evaluation stack.
340
341For those familiar with the definition of location descriptions in DWARF Version
3425, the definitions in these extensions are presented differently, but does
343in fact define the same concept with the same fundamental semantics. However,
344it does so in a way that allows the concept to extend to support address
345spaces, bit addressing, the ability for composite location descriptions to be
346composed of any kind of location description, and the ability to support
347objects located at multiple places. Collectively these changes expand the set
348of processors that can be supported and improves support for optimized code.
349
350Several approaches were considered, and the one presented appears to be the
351cleanest and offers the greatest improvement of DWARF's ability to support
352optimized code. Examining the GDB debugger and LLVM compiler, it appears only
353to require modest changes as they both already have to support general use of
354location descriptions. It is anticipated that will also be the case for other
355debuggers and compilers.
356
357As an experiment, GDB was modified to evaluate DWARF Version 5 expressions
358with location descriptions as stack entries and implicit conversions. All GDB
359tests have passed, except one that turned out to be an invalid test by DWARF
360Version 5 rules. The code in GDB actually became simpler as all evaluation was
361on the stack and there was no longer a need to maintain a separate structure
362for the location description result. This gives confidence of the backwards
363compatibility.
364
365Since the AMDGPU supports languages such as OpenCL [:ref:`OpenCL
366<amdgpu-dwarf-OpenCL>`], there is a need to define source language address
367classes so they can be used in a consistent way by consumers. It would also be
368desirable to add support for using them in defining language types rather than
369the current target architecture specific address spaces. See
370:ref:`amdgpu-dwarf-segment_addresses`.
371
372A ``DW_AT_LLVM_augmentation`` attribute is added to a compilation unit
373debugger information entry to indicate that there is additional target
374architecture specific information in the debugging information entries of that
375compilation unit. This allows a consumer to know what extensions are present
376in the debugger information entries as is possible with the augmentation
377string of other sections. The format that should be used for the augmentation
378string in the lookup by name table and CFI Common Information Entry is also
379recommended to allow a consumer to parse the string when it contains
380information from multiple vendors.
381
382The AMDGPU supports programming languages that include online compilation
383where the source text may be created at runtime. Therefore, a way to embed the
384source text in the debug information is required. For example, the OpenCL
385language runtime supports online compilation. See
386:ref:`amdgpu-dwarf-line-number-information`.
387
388Support to allow MD5 checksums to be optionally present in the line table is
389added. This allows linking together compilation units where some have MD5
390checksums and some do not. In DWARF Version 5 the file timestamp and file size
391can be optional, but if the MD5 checksum is present it must be valid for all
392files. See :ref:`amdgpu-dwarf-line-number-information`.
393
394Support is added for the HIP programming language [:ref:`HIP
395<amdgpu-dwarf-HIP>`] which is supported by the AMDGPU. See
396:ref:`amdgpu-dwarf-language-names`.
397
398The following sections provide the definitions for the additional operations,
399as well as clarifying how existing expression operations, CFI operations, and
400attributes behave with respect to generalized location descriptions that
401support address spaces and location descriptions that support multiple places.
402It has been defined such that it is backwards compatible with DWARF Version 5.
403The definitions are intended to fully define well-formed DWARF in a consistent
404style based on the DWARF Version 5 specification. Non-normative text is shown
405in *italics*.
406
407The names for the new operations, attributes, and constants include "\
408``LLVM``\ " and are encoded with vendor specific codes so these extensions can
409be implemented as an LLVM vendor extension to DWARF Version 5. If accepted these
410names would not include the "\ ``LLVM``\ " and would not use encodings in the
411vendor range.
412
413The extensions are described in
414:ref:`amdgpu-dwarf-changes-relative-to-dwarf-version-5` and are
415organized to follow the section ordering of DWARF Version 5. It includes notes
416to indicate the corresponding DWARF Version 5 sections to which they pertain.
417Other notes describe additional changes that may be worth considering, and to
418raise questions.
419
420.. _amdgpu-dwarf-changes-relative-to-dwarf-version-5:
421
422Changes Relative to DWARF Version 5
423===================================
424
425General Description
426-------------------
427
428Attribute Types
429~~~~~~~~~~~~~~~
430
431.. note::
432
433  This augments DWARF Version 5 section 2.2 and Table 2.2.
434
435The following table provides the additional attributes. See
436:ref:`amdgpu-dwarf-debugging-information-entry-attributes`.
437
438.. table:: Attribute names
439   :name: amdgpu-dwarf-attribute-names-table
440
441   =========================== ====================================
442   Attribute                   Usage
443   =========================== ====================================
444   ``DW_AT_LLVM_active_lane``  SIMD or SIMT active lanes
445   ``DW_AT_LLVM_augmentation`` Compilation unit augmentation string
446   ``DW_AT_LLVM_lane_pc``      SIMD or SIMT lane program location
447   ``DW_AT_LLVM_lanes``        SIMD or SIMT thread lane count
448   ``DW_AT_LLVM_vector_size``  Base type vector size
449   =========================== ====================================
450
451.. _amdgpu-dwarf-expressions:
452
453DWARF Expressions
454~~~~~~~~~~~~~~~~~
455
456.. note::
457
458  This section, and its nested sections, replaces DWARF Version 5 section 2.5
459  and section 2.6. The new DWARF expression operation extensions are defined as
460  well as clarifying the extensions to already existing DWARF Version 5
461  operations. It is based on the text of the existing DWARF Version 5 standard.
462
463DWARF expressions describe how to compute a value or specify a location.
464
465*The evaluation of a DWARF expression can provide the location of an object, the
466value of an array bound, the length of a dynamic string, the desired value
467itself, and so on.*
468
469If the evaluation of a DWARF expression does not encounter an error, then it can
470either result in a value (see :ref:`amdgpu-dwarf-expression-value`) or a
471location description (see :ref:`amdgpu-dwarf-location-description`). When a
472DWARF expression is evaluated, it may be specified whether a value or location
473description is required as the result kind.
474
475If a result kind is specified, and the result of the evaluation does not match
476the specified result kind, then the implicit conversions described in
477:ref:`amdgpu-dwarf-memory-location-description-operations` are performed if
478valid. Otherwise, the DWARF expression is ill-formed.
479
480If the evaluation of a DWARF expression encounters an evaluation error, then the
481result is an evaluation error.
482
483.. note::
484
485  Decided to define the concept of an evaluation error. An alternative is to
486  introduce an undefined value base type in a similar way to location
487  descriptions having an undefined location description. Then operations that
488  encounter an evaluation error can return the undefined location description or
489  value with an undefined base type.
490
491  All operations that act on values would return an undefined entity if given an
492  undefined value. The expression would then always evaluate to completion, and
493  can be tested to determine if it is an undefined entity.
494
495  However, this would add considerable additional complexity and does not match
496  that GDB throws an exception when these evaluation errors occur.
497
498If a DWARF expression is ill-formed, then the result is undefined.
499
500The following sections detail the rules for when a DWARF expression is
501ill-formed or results in an evaluation error.
502
503A DWARF expression can either be encoded as a operation expression (see
504:ref:`amdgpu-dwarf-operation-expressions`), or as a location list expression
505(see :ref:`amdgpu-dwarf-location-list-expressions`).
506
507.. _amdgpu-dwarf-expression-evaluation-context:
508
509DWARF Expression Evaluation Context
510+++++++++++++++++++++++++++++++++++
511
512A DWARF expression is evaluated in a context that can include a number of
513context elements.  If multiple context elements are specified then they must be
514self consistent or the result of the evaluation is undefined. The context
515elements that can be specified are:
516
517*A current result kind*
518
519  The kind of result required by the DWARF expression evaluation. If specified
520  it can be a location description or a value.
521
522*A current thread*
523
524  The target architecture thread identifier of the source program thread of
525  execution for which a user presented expression is currently being evaluated.
526
527  It is required for operations that are related to target architecture threads.
528
529  *For example, the* ``DW_OP_form_tls_address`` *operation and*
530  ``DW_OP_LLVM_form_aspace_address`` *operation when given an address space that
531  is thread specific.*
532
533*A current lane*
534
535  The target architecture lane identifier of the source program thread of
536  execution for which a user presented expression is currently being evaluated.
537  This applies to languages that are implemented using a SIMD or SIMT execution
538  model.
539
540  It is required for operations that are related to target architecture lanes.
541
542  *For example, the* ``DW_OP_LLVM_push_lane`` *operation and*
543  ``DW_OP_LLVM_form_aspace_address`` *operation when given an address space that
544  is lane specific.*
545
546  If specified, it must be consistent with any specified current thread and
547  current target architecture. It is consistent with a thread if it identifies a
548  lane of the thread. It is consistent with a target architecture if it is a
549  valid lane identifier of the target architecture. Otherwise the result is
550  undefined.
551
552*A current call frame*
553
554  The target architecture call frame identifier. It identifies a call frame that
555  corresponds to an active invocation of a subprogram in the current thread. It
556  is identified by its address on the call stack. The address is referred to as
557  the Canonical Frame Address (CFA). The call frame information is used to
558  determine the CFA for the call frames of the current thread's call stack (see
559  :ref:`amdgpu-dwarf-call-frame-information`).
560
561  It is required for operations that specify target architecture registers to
562  support virtual unwinding of the call stack.
563
564  *For example, the* ``DW_OP_*reg*`` *operations.*
565
566  If specified, it must be an active call frame in the current thread. If the
567  current lane is specified, then that lane must have been active on entry to
568  the call frame (see the ``DW_AT_LLVM_lane_pc`` attribute). Otherwise the
569  result is undefined.
570
571  If it is the currently executing call frame, then it is termed the top call
572  frame.
573
574*A current program location*
575
576  The target architecture program location corresponding to the current call
577  frame of the current thread.
578
579  The program location of the top call frame is the target architecture program
580  counter for the current thread. The call frame information is used to obtain
581  the value of the return address register to determine the program location of
582  the other call frames (see :ref:`amdgpu-dwarf-call-frame-information`).
583
584  It is required for the evaluation of location list expressions to select
585  amongst multiple program location ranges. It is required for operations that
586  specify target architecture registers to support virtual unwinding of the call
587  stack (see :ref:`amdgpu-dwarf-call-frame-information`).
588
589  If specified:
590
591  * If the current lane is not specified:
592
593    * If the current call frame is the top call frame, it must be the current
594      target architecture program location.
595
596    * If the current call frame F is not the top call frame, it must be the
597      program location associated with the call site in the current caller frame
598      F that invoked the callee frame.
599
600  * If the current lane is specified and the architecture program location LPC
601    computed by the ``DW_AT_LLVM_lane_pc`` attribute for the current lane is not
602    the undefined location description (indicating the lane was not active on
603    entry to the call frame), it must be LPC.
604
605  * Otherwise the result is undefined.
606
607*A current compilation unit*
608
609  The compilation unit debug information entry that contains the DWARF expression
610  being evaluated.
611
612  It is required for operations that reference debug information associated with
613  the same compilation unit, including indicating if such references use the
614  32-bit or 64-bit DWARF format. It can also provide the default address space
615  address size if no current target architecture is specified.
616
617  *For example, the* ``DW_OP_constx`` *and* ``DW_OP_addrx`` *operations.*
618
619  *Note that this compilation unit may not be the same as the compilation unit
620  determined from the loaded code object corresponding to the current program
621  location. For example, the evaluation of the expression E associated with a
622  ``DW_AT_location`` attribute of the debug information entry operand of the
623  ``DW_OP_call*`` operations is evaluated with the compilation unit that
624  contains E and not the one that contains the ``DW_OP_call*`` operation
625  expression.*
626
627*A current target architecture*
628
629  The target architecture.
630
631  It is required for operations that specify target architecture specific
632  entities.
633
634  *For example, target architecture specific entities include DWARF register
635  identifiers, DWARF lane identifiers, DWARF address space identifiers, the
636  default address space, and the address space address sizes.*
637
638  If specified:
639
640  * If the current thread is specified, then the current target architecture
641    must be the same as the target architecture of the current thread.
642
643  * If the current compilation unit is specified, then the current target
644    architecture default address space address size must be the same as he
645    ``address_size`` field in the header of the current compilation unit and any
646    associated entry in the ``.debug_aranges`` section.
647
648  * If the current program location is specified, then the current target
649    architecture must be the same as the target architecture of any line number
650    information entry (see :ref:`amdgpu-dwarf-line-number-information`)
651    corresponding to the current program location.
652
653  * If the current program location is specified, then the current target
654    architecture default address space address size must be the same as he
655    ``address_size`` field in the header of any entry corresponding to the
656    current program location in the ``.debug_addr``, ``.debug_line``,
657    ``.debug_rnglists``, ``.debug_rnglists.dwo``, ``.debug_loclists``, and
658    ``.debug_loclists.dwo`` sections.
659
660  * Otherwise the result is undefined.
661
662*A current object*
663
664  The location description of a program object.
665
666  It is required for the ``DW_OP_push_object_address`` operation.
667
668  *For example, the* ``DW_AT_data_location`` *attribute on type debug
669  information entries specifies the the program object corresponding to a
670  runtime descriptor as the current object when it evaluates its associated
671  expression.*
672
673  The result is undefined if the location descriptor is invalid (see
674  :ref:`amdgpu-dwarf-location-description`).
675
676*An initial stack*
677
678  This is a list of values or location descriptions that will be pushed on the
679  operation expression evaluation stack in the order provided before evaluation
680  of an operation expression starts.
681
682  Some debugger information entries have attributes that evaluate their DWARF
683  expression value with initial stack entries. In all other cases the initial
684  stack is empty.
685
686  The result is undefined if any location descriptors are invalid (see
687  :ref:`amdgpu-dwarf-location-description`).
688
689If the evaluation requires a context element that is not specified, then the
690result of the evaluation is an error.
691
692*A DWARF expression for the location description may be able to be evaluated
693without a thread, lane, call frame, program location, or architecture context.
694For example, the location of a global variable may be able to be evaluated
695without such context. If the expression evaluates with an error then it may
696indicate the variable has been optimized and so requires more context.*
697
698*The DWARF expression for call frame information (see
699:ref:`amdgpu-dwarf-call-frame-information`) operations are restricted to those
700that do not require the compilation unit context to be specified.*
701
702The DWARF is ill-formed if all the ``address_size`` fields in the headers of all
703the entries in the ``.debug_info``, ``.debug_addr``, ``.debug_line``,
704``.debug_rnglists``, ``.debug_rnglists.dwo``, ``.debug_loclists``, and
705``.debug_loclists.dwo`` sections corresponding to any given program location do
706not match.
707
708.. _amdgpu-dwarf-expression-value:
709
710DWARF Expression Value
711++++++++++++++++++++++
712
713A value has a type and a literal value. It can represent a literal value of any
714supported base type of the target architecture. The base type specifies the size
715and encoding of the literal value.
716
717.. note::
718
719  It may be desirable to add an implicit pointer base type encoding. It would be
720  used for the type of the value that is produced when the ``DW_OP_deref*``
721  operation retrieves the full contents of an implicit pointer location storage
722  created by the ``DW_OP_implicit_pointer`` or
723  ``DW_OP_LLVM_aspace_implicit_pointer`` operations. The literal value would
724  record the debugging information entry and byte displacement specified by the
725  associated ``DW_OP_implicit_pointer`` or
726  ``DW_OP_LLVM_aspace_implicit_pointer`` operations.
727
728There is a distinguished base type termed the generic type, which is an integral
729type that has the size of an address in the target architecture default address
730space and unspecified signedness.
731
732*The generic type is the same as the unspecified type used for stack operations
733defined in DWARF Version 4 and before.*
734
735An integral type is a base type that has an encoding of ``DW_ATE_signed``,
736``DW_ATE_signed_char``, ``DW_ATE_unsigned``, ``DW_ATE_unsigned_char``,
737``DW_ATE_boolean``, or any target architecture defined integral encoding in the
738inclusive range ``DW_ATE_lo_user`` to ``DW_ATE_hi_user``.
739
740.. note::
741
742  It is unclear if ``DW_ATE_address`` is an integral type. GDB does not seem to
743  consider it as integral.
744
745.. _amdgpu-dwarf-location-description:
746
747DWARF Location Description
748++++++++++++++++++++++++++
749
750*Debugging information must provide consumers a way to find the location of
751program variables, determine the bounds of dynamic arrays and strings, and
752possibly to find the base address of a subprogram’s call frame or the return
753address of a subprogram. Furthermore, to meet the needs of recent computer
754architectures and optimization techniques, debugging information must be able to
755describe the location of an object whose location changes over the object’s
756lifetime, and may reside at multiple locations simultaneously during parts of an
757object's lifetime.*
758
759Information about the location of program objects is provided by location
760descriptions.
761
762Location descriptions can consist of one or more single location descriptions.
763
764A single location description specifies the location storage that holds a
765program object and a position within the location storage where the program
766object starts. The position within the location storage is expressed as a bit
767offset relative to the start of the location storage.
768
769A location storage is a linear stream of bits that can hold values. Each
770location storage has a size in bits and can be accessed using a zero-based bit
771offset. The ordering of bits within a location storage uses the bit numbering
772and direction conventions that are appropriate to the current language on the
773target architecture.
774
775There are five kinds of location storage:
776
777*memory location storage*
778  Corresponds to the target architecture memory address spaces.
779
780*register location storage*
781  Corresponds to the target architecture registers.
782
783*implicit location storage*
784  Corresponds to fixed values that can only be read.
785
786*undefined location storage*
787  Indicates no value is available and therefore cannot be read or written.
788
789*composite location storage*
790  Allows a mixture of these where some bits come from one location storage and
791  some from another location storage, or from disjoint parts of the same
792  location storage.
793
794.. note::
795
796  It may be better to add an implicit pointer location storage kind used by the
797  ``DW_OP_implicit_pointer`` and ``DW_OP_LLVM_aspace_implicit_pointer``
798  operations. It would specify the debugger information entry and byte offset
799  provided by the operations.
800
801*Location descriptions are a language independent representation of addressing
802rules. They are created using DWARF operation expressions of arbitrary
803complexity. They can be the result of evaluating a debugger information entry
804attribute that specifies an operation expression. In this usage they can
805describe the location of an object as long as its lifetime is either static or
806the same as the lexical block (see DWARF Version 5 section 3.5) that owns it,
807and it does not move during its lifetime. They can be the result of evaluating a
808debugger information entry attribute that specifies a location list expression.
809In this usage they can describe the location of an object that has a limited
810lifetime, changes its location during its lifetime, or has multiple locations
811over part or all of its lifetime.*
812
813If a location description has more than one single location description, the
814DWARF expression is ill-formed if the object value held in each single location
815description's position within the associated location storage is not the same
816value, except for the parts of the value that are uninitialized.
817
818*A location description that has more than one single location description can
819only be created by a location list expression that has overlapping program
820location ranges, or certain expression operations that act on a location
821description that has more than one single location description. There are no
822operation expression operations that can directly create a location description
823with more than one single location description.*
824
825*A location description with more than one single location description can be
826used to describe objects that reside in more than one piece of storage at the
827same time. An object may have more than one location as a result of
828optimization. For example, a value that is only read may be promoted from memory
829to a register for some region of code, but later code may revert to reading the
830value from memory as the register may be used for other purposes. For the code
831region where the value is in a register, any change to the object value must be
832made in both the register and the memory so both regions of code will read the
833updated value.*
834
835*A consumer of a location description with more than one single location
836description can read the object's value from any of the single location
837descriptions (since they all refer to location storage that has the same value),
838but must write any changed value to all the single location descriptions.*
839
840The evaluation of an expression may require context elements to create a
841location description. If such a location description is accessed, the storage it
842denotes is that associated with the context element values specified when the
843location description was created, which may differ from the context at the time
844it is accessed.
845
846*For example, creating a register location description requires the thread
847context: the location storage is for the specified register of that thread.
848Creating a memory location description for an address space may required a
849thread and a lane context: the location storage is the memory associated with
850that thread and lane.*
851
852If any of the context elements required to create a location description change,
853the location description becomes invalid and accessing it is undefined.
854
855*Examples of context that can invalidate a location description are:*
856
857* *The thread context is required and execution causes the thread to terminate.*
858* *The call frame context is required and further execution causes the call
859  frame to return to the calling frame.*
860* *The program location is required and further execution of the thread occurs.
861  That could change the location list entry or call frame information entry that
862  applies.*
863* *An operation uses call frame information:*
864
865  * *Any of the frames used in the virtual call frame unwinding return.*
866  * *The top call frame is used, the program location is used to select the call
867    frame information entry, and further execution of the thread occurs.*
868
869*A DWARF expression can be used to compute a location description for an object.
870A subsequent DWARF expression evaluation can be given the object location
871description as the object context or initial stack context to compute a
872component of the object. The final result is undefined if the object location
873description becomes invalid between the two expression evaluations.*
874
875A change of a thread's program location may not make a location description
876invalid, yet may still render it as no longer meaningful. Accessing such a
877location description, or using it as the object context or initial stack context
878of an expression evaluation, may produce an undefined result.
879
880*For example, a location description may specify a register that no longer holds
881the intended program object after a program location change. One way to avoid
882such problems is to recompute location descriptions associated with threads when
883their program locations change.*
884
885.. _amdgpu-dwarf-operation-expressions:
886
887DWARF Operation Expressions
888+++++++++++++++++++++++++++
889
890An operation expression is comprised of a stream of operations, each consisting
891of an opcode followed by zero or more operands. The number of operands is
892implied by the opcode.
893
894Operations represent a postfix operation on a simple stack machine. Each stack
895entry can hold either a value or a location description. Operations can act on
896entries on the stack, including adding entries and removing entries. If the kind
897of a stack entry does not match the kind required by the operation and is not
898implicitly convertible to the required kind (see
899:ref:`amdgpu-dwarf-memory-location-description-operations`), then the DWARF
900operation expression is ill-formed.
901
902Evaluation of an operation expression starts with an empty stack on which the
903entries from the initial stack provided by the context are pushed in the order
904provided. Then the operations are evaluated, starting with the first operation
905of the stream. Evaluation continues until either an operation has an evaluation
906error, or until one past the last operation of the stream is reached.
907
908The result of the evaluation is:
909
910* If an operation has an evaluation error, or an operation evaluates an
911  expression that has an evaluation error, then the result is an evaluation
912  error.
913
914* If the current result kind specifies a location description, then:
915
916  * If the stack is empty, the result is a location description with one
917    undefined location description.
918
919    *This rule is for backwards compatibility with DWARF Version 5 which has no
920    explicit operation to create an undefined location description, and uses an
921    empty operation expression for this purpose.*
922
923  * If the top stack entry is a location description, or can be converted
924    to one (see :ref:`amdgpu-dwarf-memory-location-description-operations`),
925    then the result is that, possibly converted, location description. Any other
926    entries on the stack are discarded.
927
928  * Otherwise the DWARF expression is ill-formed.
929
930    .. note::
931
932      Could define this case as returning an implicit location description as
933      if the ``DW_OP_implicit`` operation is performed.
934
935* If the current result kind specifies a value, then:
936
937  * If the top stack entry is a value, or can be converted to one (see
938    :ref:`amdgpu-dwarf-memory-location-description-operations`), then the result
939    is that, possibly converted, value. Any other entries on the stack are
940    discarded.
941
942  * Otherwise the DWARF expression is ill-formed.
943
944* If the current result kind is not specified, then:
945
946  * If the stack is empty, the result is a location description with one
947    undefined location description.
948
949    *This rule is for backwards compatibility with DWARF Version 5 which has no
950    explicit operation to create an undefined location description, and uses an
951    empty operation expression for this purpose.*
952
953    .. note::
954
955      This rule is consistent with the rule above for when a location
956      description is requested. However, GDB appears to report this as an error
957      and no GDB tests appear to cause an empty stack for this case.
958
959  * Otherwise, the top stack entry is returned. Any other entries on the stack
960    are discarded.
961
962An operation expression is encoded as a byte block with some form of prefix that
963specifies the byte count. It can be used:
964
965* as the value of a debugging information entry attribute that is encoded using
966  class ``exprloc`` (see DWARF Version 5 section 7.5.5),
967
968* as the operand to certain operation expression operations,
969
970* as the operand to certain call frame information operations (see
971  :ref:`amdgpu-dwarf-call-frame-information`),
972
973* and in location list entries (see
974  :ref:`amdgpu-dwarf-location-list-expressions`).
975
976.. _amdgpu-dwarf-stack-operations:
977
978Stack Operations
979################
980
981The following operations manipulate the DWARF stack. Operations that index the
982stack assume that the top of the stack (most recently added entry) has index 0.
983They allow the stack entries to be either a value or location description.
984
985If any stack entry accessed by a stack operation is an incomplete composite
986location description (see
987:ref:`amdgpu-dwarf-composite-location-description-operations`), then the DWARF
988expression is ill-formed.
989
990.. note::
991
992  These operations now support stack entries that are values and location
993  descriptions.
994
995.. note::
996
997  If it is desired to also make them work with incomplete composite location
998  descriptions, then would need to define that the composite location storage
999  specified by the incomplete composite location description is also replicated
1000  when a copy is pushed. This ensures that each copy of the incomplete composite
1001  location description can update the composite location storage they specify
1002  independently.
1003
10041.  ``DW_OP_dup``
1005
1006    ``DW_OP_dup`` duplicates the stack entry at the top of the stack.
1007
10082.  ``DW_OP_drop``
1009
1010    ``DW_OP_drop`` pops the stack entry at the top of the stack and discards it.
1011
10123.  ``DW_OP_pick``
1013
1014    ``DW_OP_pick`` has a single unsigned 1-byte operand that represents an index
1015    I. A copy of the stack entry with index I is pushed onto the stack.
1016
10174.  ``DW_OP_over``
1018
1019    ``DW_OP_over`` pushes a copy of the entry with index 1.
1020
1021    *This is equivalent to a ``DW_OP_pick 1`` operation.*
1022
10235.  ``DW_OP_swap``
1024
1025    ``DW_OP_swap`` swaps the top two stack entries. The entry at the top of the
1026    stack becomes the second stack entry, and the second stack entry becomes the
1027    top of the stack.
1028
10296.  ``DW_OP_rot``
1030
1031    ``DW_OP_rot`` rotates the first three stack entries. The entry at the top of
1032    the stack becomes the third stack entry, the second entry becomes the top of
1033    the stack, and the third entry becomes the second entry.
1034
1035.. _amdgpu-dwarf-control-flow-operations:
1036
1037Control Flow Operations
1038#######################
1039
1040The following operations provide simple control of the flow of a DWARF operation
1041expression.
1042
10431.  ``DW_OP_nop``
1044
1045    ``DW_OP_nop`` is a place holder. It has no effect on the DWARF stack
1046    entries.
1047
10482.  ``DW_OP_le``, ``DW_OP_ge``, ``DW_OP_eq``, ``DW_OP_lt``, ``DW_OP_gt``,
1049    ``DW_OP_ne``
1050
1051    .. note::
1052
1053      The same as in DWARF Version 5 section 2.5.1.5.
1054
10553.  ``DW_OP_skip``
1056
1057    ``DW_OP_skip`` is an unconditional branch. Its single operand is a 2-byte
1058    signed integer constant. The 2-byte constant is the number of bytes of the
1059    DWARF expression to skip forward or backward from the current operation,
1060    beginning after the 2-byte constant.
1061
1062    If the updated position is at one past the end of the last operation, then
1063    the operation expression evaluation is complete.
1064
1065    Otherwise, the DWARF expression is ill-formed if the updated operation
1066    position is not in the range of the first to last operation inclusive, or
1067    not at the start of an operation.
1068
10694.  ``DW_OP_bra``
1070
1071    ``DW_OP_bra`` is a conditional branch. Its single operand is a 2-byte signed
1072    integer constant. This operation pops the top of stack. If the value popped
1073    is not the constant 0, the 2-byte constant operand is the number of bytes of
1074    the DWARF operation expression to skip forward or backward from the current
1075    operation, beginning after the 2-byte constant.
1076
1077    If the updated position is at one past the end of the last operation, then
1078    the operation expression evaluation is complete.
1079
1080    Otherwise, the DWARF expression is ill-formed if the updated operation
1081    position is not in the range of the first to last operation inclusive, or
1082    not at the start of an operation.
1083
10845.  ``DW_OP_call2, DW_OP_call4, DW_OP_call_ref``
1085
1086    ``DW_OP_call2``, ``DW_OP_call4``, and ``DW_OP_call_ref`` perform DWARF
1087    procedure calls during evaluation of a DWARF expression.
1088
1089    ``DW_OP_call2`` and ``DW_OP_call4``, have one operand that is, respectively,
1090    a 2-byte or 4-byte unsigned offset DR that represents the byte offset of a
1091    debugging information entry D relative to the beginning of the current
1092    compilation unit.
1093
1094    ``DW_OP_call_ref`` has one operand that is a 4-byte unsigned value in the
1095    32-bit DWARF format, or an 8-byte unsigned value in the 64-bit DWARF format,
1096    that represents the byte offset DR of a debugging information entry D
1097    relative to the beginning of the ``.debug_info`` section that contains the
1098    current compilation unit. D may not be in the current compilation unit.
1099
1100    .. note:
1101
1102      DWARF Version 5 states that DR can be an offset in a ``.debug_info``
1103      section other than the one that contains the current compilation unit. It
1104      states that relocation of references from one executable or shared object
1105      file to another must be performed by the consumer. But given that DR is
1106      defined as an offset in a ``.debug_info`` section this seems impossible.
1107      If DR was defined as an implementation defined value, then the consumer
1108      could choose to interpret the value in an implementation defined manner to
1109      reference a debug information in another executable or shared object.
1110
1111      In ELF the ``.debug_info`` section is in a non-\ ``PT_LOAD`` segment so
1112      standard dynamic relocations cannot be used. But even if they were loaded
1113      segments and dynamic relocations were used, DR would need to be the
1114      address of D, not an offset in a ``.debug_info`` section. That would also
1115      need DR to be the size of a global address. So it would not be possible to
1116      use the 32-bit DWARF format in a 64-bit global address space. In addition,
1117      the consumer would need to determine what executable or shared object the
1118      relocated address was in so it could determine the containing compilation
1119      unit.
1120
1121      GDB only interprets DR as an offset in the ``.debug_info`` section that
1122      contains the current compilation unit.
1123
1124      This comment also applies to ``DW_OP_implicit_pointer`` and
1125      ``DW_OP_LLVM_aspace_implicit_pointer``.
1126
1127    *Operand interpretation of* ``DW_OP_call2``\ *,* ``DW_OP_call4``\ *, and*
1128    ``DW_OP_call_ref`` *is exactly like that for* ``DW_FORM_ref2``\ *,
1129    ``DW_FORM_ref4``\ *, and* ``DW_FORM_ref_addr``\ *, respectively.*
1130
1131    The call operation is evaluated by:
1132
1133    * If D has a ``DW_AT_location`` attribute that is encoded as a ``exprloc``
1134      that specifies an operation expression E, then execution of the current
1135      operation expression continues from the first operation of E. Execution
1136      continues until one past the last operation of E is reached, at which
1137      point execution continues with the operation following the call operation.
1138      The operations of E are evaluated with the same current context, except
1139      current compilation unit is the one that contains D and the stack is the
1140      same as that being used by the call operation. After the call operation
1141      has been evaluated, the stack is therefore as it is left by the evaluation
1142      of the operations of E. Since E is evaluated on the same stack as the call
1143      operation, E can use, and/or remove entries already on the stack, and can
1144      add new entries to the stack.
1145
1146      *Values on the stack at the time of the call may be used as parameters by
1147      the called expression and values left on the stack by the called expression
1148      may be used as return values by prior agreement between the calling and
1149      called expressions.*
1150
1151    * If D has a ``DW_AT_location`` attribute that is encoded as a ``loclist`` or
1152      ``loclistsptr``, then the specified location list expression E is
1153      evaluated. The evaluation of E uses the current context, except the result
1154      kind is a location description, the compilation unit is the one that
1155      contains D, and the initial stack is empty. The location description
1156      result is pushed on the stack.
1157
1158      .. note::
1159
1160        This rule avoids having to define how to execute a matched location list
1161        entry operation expression on the same stack as the call when there are
1162        multiple matches. But it allows the call to obtain the location
1163        description for a variable or formal parameter which may use a location
1164        list expression.
1165
1166        An alternative is to treat the case when D has a ``DW_AT_location``
1167        attribute that is encoded as a ``loclist`` or ``loclistsptr``, and the
1168        specified location list expression E' matches a single location list
1169        entry with operation expression E, the same as the ``exprloc`` case and
1170        evaluate on the same stack.
1171
1172        But this is not attractive as if the attribute is for a variable that
1173        happens to end with a non-singleton stack, it will not simply put a
1174        location description on the stack. Presumably the intent of using
1175        ``DW_OP_call*`` on a variable or formal parameter debugger information
1176        entry is to push just one location description on the stack. That
1177        location description may have more than one single location description.
1178
1179        The previous rule for ``exprloc`` also has the same problem as normally
1180        a variable or formal parameter location expression may leave multiple
1181        entries on the stack and only return the top entry.
1182
1183        GDB implements ``DW_OP_call*`` by always executing E on the same stack.
1184        If the location list has multiple matching entries, it simply picks the
1185        first one and ignores the rest. This seems fundamentally at odds with
1186        the desire to supporting multiple places for variables.
1187
1188        So, it feels like ``DW_OP_call*`` should both support pushing a location
1189        description on the stack for a variable or formal parameter, and also
1190        support being able to execute an operation expression on the same stack.
1191        Being able to specify a different operation expression for different
1192        program locations seems a desirable feature to retain.
1193
1194        A solution to that is to have a distinct ``DW_AT_LLVM_proc`` attribute
1195        for the ``DW_TAG_dwarf_procedure`` debugging information entry. Then the
1196        ``DW_AT_location`` attribute expression is always executed separately
1197        and pushes a location description (that may have multiple single
1198        location descriptions), and the ``DW_AT_LLVM_proc`` attribute expression
1199        is always executed on the same stack and can leave anything on the
1200        stack.
1201
1202        The ``DW_AT_LLVM_proc`` attribute could have the new classes
1203        ``exprproc``, ``loclistproc``, and ``loclistsptrproc`` to indicate that
1204        the expression is executed on the same stack. ``exprproc`` is the same
1205        encoding as ``exprloc``. ``loclistproc`` and ``loclistsptrproc`` are the
1206        same encoding as their non-\ ``proc`` counterparts, except the DWARF is
1207        ill-formed if the location list does not match exactly one location list
1208        entry and a default entry is required. These forms indicate explicitly
1209        that the matched single operation expression must be executed on the
1210        same stack. This is better than ad hoc special rules for ``loclistproc``
1211        and ``loclistsptrproc`` which are currently clearly defined to always
1212        return a location description. The producer then explicitly indicates
1213        the intent through the attribute classes.
1214
1215        Such a change would be a breaking change for how GDB implements
1216        ``DW_OP_call*``. However, are the breaking cases actually occurring in
1217        practice? GDB could implement the current approach for DWARF Version 5,
1218        and the new semantics for DWARF Version 6 which has been done for some
1219        other features.
1220
1221        Another option is to limit the execution to be on the same stack only to
1222        the evaluation of an expression E that is the value of a
1223        ``DW_AT_location`` attribute of a ``DW_TAG_dwarf_procedure`` debugging
1224        information entry. The DWARF would be ill-formed if E is a location list
1225        expression that does not match exactly one location list entry. In all
1226        other cases the evaluation of an expression E that is the value of a
1227        ``DW_AT_location`` attribute would evaluate E with the current context,
1228        except the result kind is a location description, the compilation unit
1229        is the one that contains D, and the initial stack is empty. The location
1230        description result is pushed on the stack.
1231
1232    * If D has a ``DW_AT_const_value`` attribute with a value V, then it is as
1233      if a ``DW_OP_implicit_value V`` operation was executed.
1234
1235      *This allows a call operation to be used to compute the location
1236      description for any variable or formal parameter regardless of whether the
1237      producer has optimized it to a constant. This is consistent with the
1238      ``DW_OP_implicit_pointer`` operation.*
1239
1240      .. note::
1241
1242        Alternatively, could deprecate using ``DW_AT_const_value`` for
1243        ``DW_TAG_variable`` and ``DW_TAG_formal_parameter`` debugger information
1244        entries that are constants and instead use ``DW_AT_location`` with an
1245        operation expression that results in a location description with one
1246        implicit location description. Then this rule would not be required.
1247
1248    * Otherwise, there is no effect and no changes are made to the stack.
1249
1250      .. note::
1251
1252        In DWARF Version 5, if D does not have a ``DW_AT_location`` then
1253        ``DW_OP_call*`` is defined to have no effect. It is unclear that this is
1254        the right definition as a producer should be able to rely on using
1255        ``DW_OP_call*`` to get a location description for any non-\
1256        ``DW_TAG_dwarf_procedure`` debugging information entries. Also, the
1257        producer should not be creating DWARF with ``DW_OP_call*`` to a
1258        ``DW_TAG_dwarf_procedure`` that does not have a ``DW_AT_location``
1259        attribute. So, should this case be defined as an ill-formed DWARF
1260        expression?
1261
1262    *The* ``DW_TAG_dwarf_procedure`` *debugging information entry can be used to
1263    define DWARF procedures that can be called.*
1264
1265.. _amdgpu-dwarf-value-operations:
1266
1267Value Operations
1268################
1269
1270This section describes the operations that push values on the stack.
1271
1272Each value stack entry has a type and a literal value and can represent a
1273literal value of any supported base type of the target architecture. The base
1274type specifies the size and encoding of the literal value.
1275
1276Instead of a base type, value stack entries can have a distinguished generic
1277type, which is an integral type that has the size of an address in the target
1278architecture default address space and unspecified signedness.
1279
1280*The generic type is the same as the unspecified type used for stack operations
1281defined in DWARF Version 4 and before.*
1282
1283An integral type is a base type that has an encoding of ``DW_ATE_signed``,
1284``DW_ATE_signed_char``, ``DW_ATE_unsigned``, ``DW_ATE_unsigned_char``,
1285``DW_ATE_boolean``, or any target architecture defined integral encoding in the
1286inclusive range ``DW_ATE_lo_user`` to ``DW_ATE_hi_user``.
1287
1288.. note::
1289
1290  Unclear if ``DW_ATE_address`` is an integral type. GDB does not seem to
1291  consider it as integral.
1292
1293.. _amdgpu-dwarf-literal-operations:
1294
1295Literal Operations
1296^^^^^^^^^^^^^^^^^^
1297
1298The following operations all push a literal value onto the DWARF stack.
1299
1300Operations other than ``DW_OP_const_type`` push a value V with the generic type.
1301If V is larger than the generic type, then V is truncated to the generic type
1302size and the low-order bits used.
1303
13041.  ``DW_OP_lit0``, ``DW_OP_lit1``, ..., ``DW_OP_lit31``
1305
1306    ``DW_OP_lit<N>`` operations encode an unsigned literal value N from 0
1307    through 31, inclusive. They push the value N with the generic type.
1308
13092.  ``DW_OP_const1u``, ``DW_OP_const2u``, ``DW_OP_const4u``, ``DW_OP_const8u``
1310
1311    ``DW_OP_const<N>u`` operations have a single operand that is a 1, 2, 4, or
1312    8-byte unsigned integer constant U, respectively. They push the value U with
1313    the generic type.
1314
13153.  ``DW_OP_const1s``, ``DW_OP_const2s``, ``DW_OP_const4s``, ``DW_OP_const8s``
1316
1317    ``DW_OP_const<N>s`` operations have a single operand that is a 1, 2, 4, or
1318    8-byte signed integer constant S, respectively. They push the value S with
1319    the generic type.
1320
13214.  ``DW_OP_constu``
1322
1323    ``DW_OP_constu`` has a single unsigned LEB128 integer operand N. It pushes
1324    the value N with the generic type.
1325
13265.  ``DW_OP_consts``
1327
1328    ``DW_OP_consts`` has a single signed LEB128 integer operand N. It pushes the
1329    value N with the generic type.
1330
13316.  ``DW_OP_constx``
1332
1333    ``DW_OP_constx`` has a single unsigned LEB128 integer operand that
1334    represents a zero-based index into the ``.debug_addr`` section relative to
1335    the value of the ``DW_AT_addr_base`` attribute of the associated compilation
1336    unit. The value N in the ``.debug_addr`` section has the size of the generic
1337    type. It pushes the value N with the generic type.
1338
1339    *The* ``DW_OP_constx`` *operation is provided for constants that require
1340    link-time relocation but should not be interpreted by the consumer as a
1341    relocatable address (for example, offsets to thread-local storage).*
1342
13439.  ``DW_OP_const_type``
1344
1345    ``DW_OP_const_type`` has three operands. The first is an unsigned LEB128
1346    integer DR that represents the byte offset of a debugging information entry
1347    D relative to the beginning of the current compilation unit, that provides
1348    the type T of the constant value. The second is a 1-byte unsigned integral
1349    constant S. The third is a block of bytes B, with a length equal to S.
1350
1351    TS is the bit size of the type T. The least significant TS bits of B are
1352    interpreted as a value V of the type D. It pushes the value V with the type
1353    D.
1354
1355    The DWARF is ill-formed if D is not a ``DW_TAG_base_type`` debugging
1356    information entry in the current compilation unit, or if TS divided by 8
1357    (the byte size) and rounded up to a whole number is not equal to S.
1358
1359    *While the size of the byte block B can be inferred from the type D
1360    definition, it is encoded explicitly into the operation so that the
1361    operation can be parsed easily without reference to the* ``.debug_info``
1362    *section.*
1363
136410. ``DW_OP_LLVM_push_lane`` *New*
1365
1366    ``DW_OP_LLVM_push_lane`` pushes the target architecture lane identifier of
1367    the current lane as a value with the generic type.
1368
1369    *For languages that are implemented using a SIMD or SIMT execution model,
1370    this is the lane number that corresponds to the source language thread of
1371    execution upon which the user is focused.*
1372
1373.. _amdgpu-dwarf-arithmetic-logical-operations:
1374
1375Arithmetic and Logical Operations
1376^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1377
1378.. note::
1379
1380  This section is the same as DWARF Version 5 section 2.5.1.4.
1381
1382.. _amdgpu-dwarf-type-conversions-operations:
1383
1384Type Conversion Operations
1385^^^^^^^^^^^^^^^^^^^^^^^^^^
1386
1387.. note::
1388
1389  This section is the same as DWARF Version 5 section 2.5.1.6.
1390
1391.. _amdgpu-dwarf-general-operations:
1392
1393Special Value Operations
1394^^^^^^^^^^^^^^^^^^^^^^^^
1395
1396There are these special value operations currently defined:
1397
13981.  ``DW_OP_regval_type``
1399
1400    ``DW_OP_regval_type`` has two operands. The first is an unsigned LEB128
1401    integer that represents a register number R. The second is an unsigned
1402    LEB128 integer DR that represents the byte offset of a debugging information
1403    entry D relative to the beginning of the current compilation unit, that
1404    provides the type T of the register value.
1405
1406    The operation is equivalent to performing ``DW_OP_regx R; DW_OP_deref_type
1407    DR``.
1408
1409    .. note::
1410
1411      Should DWARF allow the type T to be a larger size than the size of the
1412      register R? Restricting a larger bit size avoids any issue of conversion
1413      as the, possibly truncated, bit contents of the register is simply
1414      interpreted as a value of T. If a conversion is wanted it can be done
1415      explicitly using a ``DW_OP_convert`` operation.
1416
1417      GDB has a per register hook that allows a target specific conversion on a
1418      register by register basis. It defaults to truncation of bigger registers.
1419      Removing use of the target hook does not cause any test failures in common
1420      architectures. If the compiler for a target architecture did want some
1421      form of conversion, including a larger result type, it could always
1422      explicitly used the ``DW_OP_convert`` operation.
1423
1424      If T is a larger type than the register size, then the default GDB
1425      register hook reads bytes from the next register (or reads out of bounds
1426      for the last register!). Removing use of the target hook does not cause
1427      any test failures in common architectures (except an illegal hand written
1428      assembly test). If a target architecture requires this behavior, these
1429      extensions allow a composite location description to be used to combine
1430      multiple registers.
1431
14322.  ``DW_OP_deref``
1433
1434    S is the bit size of the generic type divided by 8 (the byte size) and
1435    rounded up to a whole number. DR is the offset of a hypothetical debug
1436    information entry D in the current compilation unit for a base type of the
1437    generic type.
1438
1439    The operation is equivalent to performing ``DW_OP_deref_type S, DR``.
1440
14413.  ``DW_OP_deref_size``
1442
1443    ``DW_OP_deref_size`` has a single 1-byte unsigned integral constant that
1444    represents a byte result size S.
1445
1446    TS is the smaller of the generic type bit size and S scaled by 8 (the byte
1447    size). If TS is smaller than the generic type bit size then T is an unsigned
1448    integral type of bit size TS, otherwise T is the generic type. DR is the
1449    offset of a hypothetical debug information entry D in the current
1450    compilation unit for a base type T.
1451
1452    .. note::
1453
1454      Truncating the value when S is larger than the generic type matches what
1455      GDB does. This allows the generic type size to not be an integral byte
1456      size. It does allow S to be arbitrarily large. Should S be restricted to
1457      the size of the generic type rounded up to a multiple of 8?
1458
1459    The operation is equivalent to performing ``DW_OP_deref_type S, DR``, except
1460    if T is not the generic type, the value V pushed is zero-extended to the
1461    generic type bit size and its type changed to the generic type.
1462
14634.  ``DW_OP_deref_type``
1464
1465    ``DW_OP_deref_type`` has two operands. The first is a 1-byte unsigned
1466    integral constant S. The second is an unsigned LEB128 integer DR that
1467    represents the byte offset of a debugging information entry D relative to
1468    the beginning of the current compilation unit, that provides the type T of
1469    the result value.
1470
1471    TS is the bit size of the type T.
1472
1473    *While the size of the pushed value V can be inferred from the type T, it is
1474    encoded explicitly as the operand S so that the operation can be parsed
1475    easily without reference to the* ``.debug_info`` *section.*
1476
1477    .. note::
1478
1479      It is unclear why the operand S is needed. Unlike ``DW_OP_const_type``,
1480      the size is not needed for parsing. Any evaluation needs to get the base
1481      type T to push with the value to know its encoding and bit size.
1482
1483    It pops one stack entry that must be a location description L.
1484
1485    A value V of TS bits is retrieved from the location storage LS specified by
1486    one of the single location descriptions SL of L.
1487
1488    *If L, or the location description of any composite location description
1489    part that is a subcomponent of L, has more than one single location
1490    description, then any one of them can be selected as they are required to
1491    all have the same value. For any single location description SL, bits are
1492    retrieved from the associated storage location starting at the bit offset
1493    specified by SL. For a composite location description, the retrieved bits
1494    are the concatenation of the N bits from each composite location part PL,
1495    where N is limited to the size of PL.*
1496
1497    V is pushed on the stack with the type T.
1498
1499    .. note::
1500
1501      This definition makes it an evaluation error if L is a register location
1502      description that has less than TS bits remaining in the register storage.
1503      Particularly since these extensions extend location descriptions to have
1504      a bit offset, it would be odd to define this as performing sign extension
1505      based on the type, or be target architecture dependent, as the number of
1506      remaining bits could be any number. This matches the GDB implementation
1507      for ``DW_OP_deref_type``.
1508
1509      These extensions define ``DW_OP_*breg*`` in terms of
1510      ``DW_OP_regval_type``. ``DW_OP_regval_type`` is defined in terms of
1511      ``DW_OP_regx``, which uses a 0 bit offset, and ``DW_OP_deref_type``.
1512      Therefore, it requires the register size to be greater or equal to the
1513      address size of the address space. This matches the GDB implementation for
1514      ``DW_OP_*breg*``.
1515
1516    The DWARF is ill-formed if D is not in the current compilation unit, D is
1517    not a ``DW_TAG_base_type`` debugging information entry, or if TS divided by
1518    8 (the byte size) and rounded up to a whole number is not equal to S.
1519
1520    .. note::
1521
1522      This definition allows the base type to be a bit size since there seems no
1523      reason to restrict it.
1524
1525    It is an evaluation error if any bit of the value is retrieved from the
1526    undefined location storage or the offset of any bit exceeds the size of the
1527    location storage LS specified by any single location description SL of L.
1528
1529    See :ref:`amdgpu-dwarf-implicit-location-descriptions` for special rules
1530    concerning implicit location descriptions created by the
1531    ``DW_OP_implicit_pointer`` and ``DW_OP_LLVM_implicit_aspace_pointer``
1532    operations.
1533
15345.  ``DW_OP_xderef`` *Deprecated*
1535
1536    ``DW_OP_xderef`` pops two stack entries. The first must be an integral type
1537    value that represents an address A. The second must be an integral type
1538    value that represents a target architecture specific address space
1539    identifier AS.
1540
1541    The operation is equivalent to performing ``DW_OP_swap;
1542    DW_OP_LLVM_form_aspace_address; DW_OP_deref``. The value V retrieved is left
1543    on the stack with the generic type.
1544
1545    *This operation is deprecated as the* ``DW_OP_LLVM_form_aspace_address``
1546    *operation can be used and provides greater expressiveness.*
1547
15486.  ``DW_OP_xderef_size`` *Deprecated*
1549
1550    ``DW_OP_xderef_size`` has a single 1-byte unsigned integral constant that
1551    represents a byte result size S.
1552
1553    It pops two stack entries. The first must be an integral type value that
1554    represents an address A. The second must be an integral type value that
1555    represents a target architecture specific address space identifier AS.
1556
1557    The operation is equivalent to performing ``DW_OP_swap;
1558    DW_OP_LLVM_form_aspace_address; DW_OP_deref_size S``. The zero-extended
1559    value V retrieved is left on the stack with the generic type.
1560
1561    *This operation is deprecated as the* ``DW_OP_LLVM_form_aspace_address``
1562    *operation can be used and provides greater expressiveness.*
1563
15647.  ``DW_OP_xderef_type`` *Deprecated*
1565
1566    ``DW_OP_xderef_type`` has two operands. The first is a 1-byte unsigned
1567    integral constant S. The second operand is an unsigned LEB128 integer DR
1568    that represents the byte offset of a debugging information entry D relative
1569    to the beginning of the current compilation unit, that provides the type T
1570    of the result value.
1571
1572    It pops two stack entries. The first must be an integral type value that
1573    represents an address A. The second must be an integral type value that
1574    represents a target architecture specific address space identifier AS.
1575
1576    The operation is equivalent to performing ``DW_OP_swap;
1577    DW_OP_LLVM_form_aspace_address; DW_OP_deref_type S R``. The value V
1578    retrieved is left on the stack with the type D.
1579
1580    *This operation is deprecated as the* ``DW_OP_LLVM_form_aspace_address``
1581    *operation can be used and provides greater expressiveness.*
1582
15838.  ``DW_OP_entry_value`` *Deprecated*
1584
1585    ``DW_OP_entry_value`` pushes the value of an expression that is evaluated in
1586    the context of the calling frame.
1587
1588    *It may be used to determine the value of arguments on entry to the current
1589    call frame provided they are not clobbered.*
1590
1591    It has two operands. The first is an unsigned LEB128 integer S. The second
1592    is a block of bytes, with a length equal S, interpreted as a DWARF
1593    operation expression E.
1594
1595    E is evaluated with the current context, except the result kind is
1596    unspecified, the call frame is the one that called the current frame, the
1597    program location is the call site in the calling frame, the object is
1598    unspecified, and the initial stack is empty. The calling frame information
1599    is obtained by virtually unwinding the current call frame using the call
1600    frame information (see :ref:`amdgpu-dwarf-call-frame-information`).
1601
1602    If the result of E is a location description L (see
1603    :ref:`amdgpu-dwarf-register-location-descriptions`), and the last operation
1604    executed by E is a ``DW_OP_reg*`` for register R with a target architecture
1605    specific base type of T, then the contents of the register are retrieved as
1606    if a ``DW_OP_deref_type DR`` operation was performed where DR is the offset
1607    of a hypothetical debug information entry in the current compilation unit
1608    for T. The resulting value V s pushed on the stack.
1609
1610    *Using* ``DW_OP_reg*`` *provides a more compact form for the case where the
1611    value was in a register on entry to the subprogram.*
1612
1613    .. note:
1614
1615      It is unclear how this provides a more compact expression, as
1616      ``DW_OP_regval_type`` could be used which is marginally larger.
1617
1618    If the result of E is a value V, then V is pushed on the stack.
1619
1620    Otherwise, the DWARF expression is ill-formed.
1621
1622    *The* ``DW_OP_entry_value`` *operation is deprecated as its main usage is
1623    provided by other means. DWARF Version 5 added the*
1624    ``DW_TAG_call_site_parameter`` *debugger information entry for call sites
1625    that has* ``DW_AT_call_value``\ *,* ``DW_AT_call_data_location``\ *, and*
1626    ``DW_AT_call_data_value`` *attributes that provide DWARF expressions to
1627    compute actual parameter values at the time of the call, and requires the
1628    producer to ensure the expressions are valid to evaluate even when virtually
1629    unwound. The* ``DW_OP_LLVM_call_frame_entry_reg`` *operation provides access
1630    to registers in the virtually unwound calling frame.*
1631
1632    .. note::
1633
1634      GDB only implements ``DW_OP_entry_value`` when E is exactly
1635      ``DW_OP_reg*`` or ``DW_OP_breg*; DW_OP_deref*``.
1636
1637.. _amdgpu-dwarf-location-description-operations:
1638
1639Location Description Operations
1640###############################
1641
1642This section describes the operations that push location descriptions on the
1643stack.
1644
1645General Location Description Operations
1646^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1647
16481.  ``DW_OP_LLVM_offset`` *New*
1649
1650    ``DW_OP_LLVM_offset`` pops two stack entries. The first must be an integral
1651    type value that represents a byte displacement B. The second must be a
1652    location description L.
1653
1654    It adds the value of B scaled by 8 (the byte size) to the bit offset of each
1655    single location description SL of L, and pushes the updated L.
1656
1657    It is an evaluation error if the updated bit offset of any SL is less than 0
1658    or greater than or equal to the size of the location storage specified by
1659    SL.
1660
16612.  ``DW_OP_LLVM_offset_uconst`` *New*
1662
1663    ``DW_OP_LLVM_offset_uconst`` has a single unsigned LEB128 integer operand
1664    that represents a byte displacement B.
1665
1666    The operation is equivalent to performing ``DW_OP_constu B;
1667    DW_OP_LLVM_offset``.
1668
1669    *This operation is supplied specifically to be able to encode more field
1670    displacements in two bytes than can be done with* ``DW_OP_lit*;
1671    DW_OP_LLVM_offset``\ *.*
1672
1673    .. note::
1674
1675      Should this be named ``DW_OP_LLVM_offset_uconst`` to match
1676      ``DW_OP_plus_uconst``, or ``DW_OP_LLVM_offset_constu`` to match
1677      ``DW_OP_constu``?
1678
16793.  ``DW_OP_LLVM_bit_offset`` *New*
1680
1681    ``DW_OP_LLVM_bit_offset`` pops two stack entries. The first must be an
1682    integral type value that represents a bit displacement B. The second must be
1683    a location description L.
1684
1685    It adds the value of B to the bit offset of each single location description
1686    SL of L, and pushes the updated L.
1687
1688    It is an evaluation error if the updated bit offset of any SL is less than 0
1689    or greater than or equal to the size of the location storage specified by
1690    SL.
1691
16924.  ``DW_OP_push_object_address``
1693
1694    ``DW_OP_push_object_address`` pushes the location description L of the
1695    current object.
1696
1697    *This object may correspond to an independent variable that is part of a
1698    user presented expression that is being evaluated. The object location
1699    description may be determined from the variable's own debugging information
1700    entry or it may be a component of an array, structure, or class whose
1701    address has been dynamically determined by an earlier step during user
1702    expression evaluation.*
1703
1704    *This operation provides explicit functionality (especially for arrays
1705    involving descriptions) that is analogous to the implicit push of the base
1706    location description of a structure prior to evaluation of a
1707    ``DW_AT_data_member_location`` to access a data member of a structure.*
1708
1709    .. note::
1710
1711      This operation could be removed and the object location description
1712      specified as the initial stack as for ``DW_AT_data_member_location``.
1713
1714      The only attribute that specifies a current object is
1715      ``DW_AT_data_location`` so the non-normative text seems to overstate how
1716      this is being used. Or are there other attributes that need to state they
1717      pass an object?
1718
17195.  ``DW_OP_LLVM_call_frame_entry_reg`` *New*
1720
1721    ``DW_OP_LLVM_call_frame_entry_reg`` has a single unsigned LEB128 integer
1722    operand that represents a target architecture register number R.
1723
1724    It pushes a location description L that holds the value of register R on
1725    entry to the current subprogram as defined by the call frame information
1726    (see :ref:`amdgpu-dwarf-call-frame-information`).
1727
1728    *If there is no call frame information defined, then the default rules for
1729    the target architecture are used. If the register rule is* undefined\ *, then
1730    the undefined location description is pushed. If the register rule is* same
1731    value\ *, then a register location description for R is pushed.*
1732
1733.. _amdgpu-dwarf-undefined-location-description-operations:
1734
1735Undefined Location Description Operations
1736^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1737
1738*The undefined location storage represents a piece or all of an object that is
1739present in the source but not in the object code (perhaps due to optimization).
1740Neither reading nor writing to the undefined location storage is meaningful.*
1741
1742An undefined location description specifies the undefined location storage.
1743There is no concept of the size of the undefined location storage, nor of a bit
1744offset for an undefined location description. The ``DW_OP_LLVM_*offset``
1745operations leave an undefined location description unchanged. The
1746``DW_OP_*piece`` operations can explicitly or implicitly specify an undefined
1747location description, allowing any size and offset to be specified, and results
1748in a part with all undefined bits.
1749
17501.  ``DW_OP_LLVM_undefined`` *New*
1751
1752    ``DW_OP_LLVM_undefined`` pushes a location description L that comprises one
1753    undefined location description SL.
1754
1755.. _amdgpu-dwarf-memory-location-description-operations:
1756
1757Memory Location Description Operations
1758^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1759
1760Each of the target architecture specific address spaces has a corresponding
1761memory location storage that denotes the linear addressable memory of that
1762address space. The size of each memory location storage corresponds to the range
1763of the addresses in the corresponding address space.
1764
1765*It is target architecture defined how address space location storage maps to
1766target architecture physical memory. For example, they may be independent
1767memory, or more than one location storage may alias the same physical memory
1768possibly at different offsets and with different interleaving. The mapping may
1769also be dictated by the source language address classes.*
1770
1771A memory location description specifies a memory location storage. The bit
1772offset corresponds to a bit position within a byte of the memory. Bits accessed
1773using a memory location description, access the corresponding target
1774architecture memory starting at the bit position within the byte specified by
1775the bit offset.
1776
1777A memory location description that has a bit offset that is a multiple of 8 (the
1778byte size) is defined to be a byte address memory location description. It has a
1779memory byte address A that is equal to the bit offset divided by 8.
1780
1781A memory location description that does not have a bit offset that is a multiple
1782of 8 (the byte size) is defined to be a bit field memory location description.
1783It has a bit position B equal to the bit offset modulo 8, and a memory byte
1784address A equal to the bit offset minus B that is then divided by 8.
1785
1786The address space AS of a memory location description is defined to be the
1787address space that corresponds to the memory location storage associated with
1788the memory location description.
1789
1790A location description that is comprised of one byte address memory location
1791description SL is defined to be a memory byte address location description. It
1792has a byte address equal to A and an address space equal to AS of the
1793corresponding SL.
1794
1795``DW_ASPACE_none`` is defined as the target architecture default address space.
1796
1797If a stack entry is required to be a location description, but it is a value V
1798with the generic type, then it is implicitly converted to a location description
1799L with one memory location description SL. SL specifies the memory location
1800storage that corresponds to the target architecture default address space with a
1801bit offset equal to V scaled by 8 (the byte size).
1802
1803.. note::
1804
1805  If it is wanted to allow any integral type value to be implicitly converted to
1806  a memory location description in the target architecture default address
1807  space:
1808
1809    If a stack entry is required to be a location description, but is a value V
1810    with an integral type, then it is implicitly converted to a location
1811    description L with a one memory location description SL. If the type size of
1812    V is less than the generic type size, then the value V is zero extended to
1813    the size of the generic type. The least significant generic type size bits
1814    are treated as a twos-complement unsigned value to be used as an address A.
1815    SL specifies memory location storage corresponding to the target
1816    architecture default address space with a bit offset equal to A scaled by 8
1817    (the byte size).
1818
1819  The implicit conversion could also be defined as target architecture specific.
1820  For example, GDB checks if V is an integral type. If it is not it gives an
1821  error. Otherwise, GDB zero-extends V to 64 bits. If the GDB target defines a
1822  hook function, then it is called. The target specific hook function can modify
1823  the 64-bit value, possibly sign extending based on the original value type.
1824  Finally, GDB treats the 64-bit value V as a memory location address.
1825
1826If a stack entry is required to be a location description, but it is an implicit
1827pointer value IPV with the target architecture default address space, then it is
1828implicitly converted to a location description with one single location
1829description specified by IPV. See
1830:ref:`amdgpu-dwarf-implicit-location-descriptions`.
1831
1832.. note::
1833
1834  Is this rule required for DWARF Version 5 backwards compatibility? If not, it
1835  can be eliminated, and the producer can use
1836  ``DW_OP_LLVM_form_aspace_address``.
1837
1838If a stack entry is required to be a value, but it is a location description L
1839with one memory location description SL in the target architecture default
1840address space with a bit offset B that is a multiple of 8, then it is implicitly
1841converted to a value equal to B divided by 8 (the byte size) with the generic
1842type.
1843
18441.  ``DW_OP_addr``
1845
1846    ``DW_OP_addr`` has a single byte constant value operand, which has the size
1847    of the generic type, that represents an address A.
1848
1849    It pushes a location description L with one memory location description SL
1850    on the stack. SL specifies the memory location storage corresponding to the
1851    target architecture default address space with a bit offset equal to A
1852    scaled by 8 (the byte size).
1853
1854    *If the DWARF is part of a code object, then A may need to be relocated. For
1855    example, in the ELF code object format, A must be adjusted by the difference
1856    between the ELF segment virtual address and the virtual address at which the
1857    segment is loaded.*
1858
18592.  ``DW_OP_addrx``
1860
1861    ``DW_OP_addrx`` has a single unsigned LEB128 integer operand that represents
1862    a zero-based index into the ``.debug_addr`` section relative to the value of
1863    the ``DW_AT_addr_base`` attribute of the associated compilation unit. The
1864    address value A in the ``.debug_addr`` section has the size of the generic
1865    type.
1866
1867    It pushes a location description L with one memory location description SL
1868    on the stack. SL specifies the memory location storage corresponding to the
1869    target architecture default address space with a bit offset equal to A
1870    scaled by 8 (the byte size).
1871
1872    *If the DWARF is part of a code object, then A may need to be relocated. For
1873    example, in the ELF code object format, A must be adjusted by the difference
1874    between the ELF segment virtual address and the virtual address at which the
1875    segment is loaded.*
1876
18773.  ``DW_OP_LLVM_form_aspace_address`` *New*
1878
1879    ``DW_OP_LLVM_form_aspace_address`` pops top two stack entries. The first
1880    must be an integral type value that represents a target architecture
1881    specific address space identifier AS. The second must be an integral type
1882    value that represents an address A.
1883
1884    The address size S is defined as the address bit size of the target
1885    architecture specific address space that corresponds to AS.
1886
1887    A is adjusted to S bits by zero extending if necessary, and then treating the
1888    least significant S bits as a twos-complement unsigned value A'.
1889
1890    It pushes a location description L with one memory location description SL
1891    on the stack. SL specifies the memory location storage LS that corresponds
1892    to AS with a bit offset equal to A' scaled by 8 (the byte size).
1893
1894    If AS is an address space that is specific to context elements, then LS
1895    corresponds to the location storage associated with the current context.
1896
1897    *For example, if AS is for per thread storage then LS is the location
1898    storage for the current thread. For languages that are implemented using a
1899    SIMD or SIMT execution model, then if AS is for per lane storage then LS is
1900    the location storage for the current lane of the current thread. Therefore,
1901    if L is accessed by an operation, the location storage selected when the
1902    location description was created is accessed, and not the location storage
1903    associated with the current context of the access operation.*
1904
1905    The DWARF expression is ill-formed if AS is not one of the values defined by
1906    the target architecture specific ``DW_ASPACE_*`` values.
1907
1908    See :ref:`amdgpu-dwarf-implicit-location-descriptions` for special rules
1909    concerning implicit pointer values produced by dereferencing implicit
1910    location descriptions created by the ``DW_OP_implicit_pointer`` and
1911    ``DW_OP_LLVM_implicit_aspace_pointer`` operations.
1912
19134.  ``DW_OP_form_tls_address``
1914
1915    ``DW_OP_form_tls_address`` pops one stack entry that must be an integral
1916    type value and treats it as a thread-local storage address TA.
1917
1918    It pushes a location description L with one memory location description SL
1919    on the stack. SL is the target architecture specific memory location
1920    description that corresponds to the thread-local storage address TA.
1921
1922    The meaning of the thread-local storage address TA is defined by the
1923    run-time environment. If the run-time environment supports multiple
1924    thread-local storage blocks for a single thread, then the block
1925    corresponding to the executable or shared library containing this DWARF
1926    expression is used.
1927
1928    *Some implementations of C, C++, Fortran, and other languages support a
1929    thread-local storage class. Variables with this storage class have distinct
1930    values and addresses in distinct threads, much as automatic variables have
1931    distinct values and addresses in each subprogram invocation. Typically,
1932    there is a single block of storage containing all thread-local variables
1933    declared in the main executable, and a separate block for the variables
1934    declared in each shared library. Each thread-local variable can then be
1935    accessed in its block using an identifier. This identifier is typically a
1936    byte offset into the block and pushed onto the DWARF stack by one of the*
1937    ``DW_OP_const*`` *operations prior to the* ``DW_OP_form_tls_address``
1938    *operation. Computing the address of the appropriate block can be complex
1939    (in some cases, the compiler emits a function call to do it), and difficult
1940    to describe using ordinary DWARF location descriptions. Instead of forcing
1941    complex thread-local storage calculations into the DWARF expressions, the*
1942    ``DW_OP_form_tls_address`` *allows the consumer to perform the computation
1943    based on the target architecture specific run-time environment.*
1944
19455.  ``DW_OP_call_frame_cfa``
1946
1947    ``DW_OP_call_frame_cfa`` pushes the location description L of the Canonical
1948    Frame Address (CFA) of the current subprogram, obtained from the call frame
1949    information on the stack. See :ref:`amdgpu-dwarf-call-frame-information`.
1950
1951    *Although the value of the* ``DW_AT_frame_base`` *attribute of the debugger
1952    information entry corresponding to the current subprogram can be computed
1953    using a location list expression, in some cases this would require an
1954    extensive location list because the values of the registers used in
1955    computing the CFA change during a subprogram execution. If the call frame
1956    information is present, then it already encodes such changes, and it is
1957    space efficient to reference that using the* ``DW_OP_call_frame_cfa``
1958    *operation.*
1959
19606.  ``DW_OP_fbreg``
1961
1962    ``DW_OP_fbreg`` has a single signed LEB128 integer operand that represents a
1963    byte displacement B.
1964
1965    The location description L for the *frame base* of the current subprogram is
1966    obtained from the ``DW_AT_frame_base`` attribute of the debugger information
1967    entry corresponding to the current subprogram as described in
1968    :ref:`amdgpu-dwarf-debugging-information-entry-attributes`.
1969
1970    The location description L is updated as if the ``DW_OP_LLVM_offset_uconst
1971    B`` operation was applied. The updated L is pushed on the stack.
1972
19737.  ``DW_OP_breg0``, ``DW_OP_breg1``, ..., ``DW_OP_breg31``
1974
1975    The ``DW_OP_breg<N>`` operations encode the numbers of up to 32 registers,
1976    numbered from 0 through 31, inclusive. The register number R corresponds to
1977    the N in the operation name.
1978
1979    They have a single signed LEB128 integer operand that represents a byte
1980    displacement B.
1981
1982    The address space identifier AS is defined as the one corresponding to the
1983    target architecture specific default address space.
1984
1985    The address size S is defined as the address bit size of the target
1986    architecture specific address space corresponding to AS.
1987
1988    The contents of the register specified by R are retrieved as if a
1989    ``DW_OP_regval_type R, DR`` operation was performed where DR is the offset
1990    of a hypothetical debug information entry in the current compilation unit
1991    for an unsigned integral base type of size S bits. B is added and the least
1992    significant S bits are treated as an unsigned value to be used as an address
1993    A.
1994
1995    They push a location description L comprising one memory location
1996    description LS on the stack. LS specifies the memory location storage that
1997    corresponds to AS with a bit offset equal to A scaled by 8 (the byte size).
1998
19998.  ``DW_OP_bregx``
2000
2001    ``DW_OP_bregx`` has two operands. The first is an unsigned LEB128 integer
2002    that represents a register number R. The second is a signed LEB128
2003    integer that represents a byte displacement B.
2004
2005    The action is the same as for ``DW_OP_breg<N>``, except that R is used as
2006    the register number and B is used as the byte displacement.
2007
20089.  ``DW_OP_LLVM_aspace_bregx`` *New*
2009
2010    ``DW_OP_LLVM_aspace_bregx`` has two operands. The first is an unsigned
2011    LEB128 integer that represents a register number R. The second is a signed
2012    LEB128 integer that represents a byte displacement B. It pops one stack
2013    entry that is required to be an integral type value that represents a target
2014    architecture specific address space identifier AS.
2015
2016    The action is the same as for ``DW_OP_breg<N>``, except that R is used as
2017    the register number, B is used as the byte displacement, and AS is used as
2018    the address space identifier.
2019
2020    The DWARF expression is ill-formed if AS is not one of the values defined by
2021    the target architecture specific ``DW_ASPACE_*`` values.
2022
2023    .. note::
2024
2025      Could also consider adding ``DW_OP_aspace_breg0, DW_OP_aspace_breg1, ...,
2026      DW_OP_aspace_bref31`` which would save encoding size.
2027
2028.. _amdgpu-dwarf-register-location-descriptions:
2029
2030Register Location Description Operations
2031^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2032
2033There is a register location storage that corresponds to each of the target
2034architecture registers. The size of each register location storage corresponds
2035to the size of the corresponding target architecture register.
2036
2037A register location description specifies a register location storage. The bit
2038offset corresponds to a bit position within the register. Bits accessed using a
2039register location description access the corresponding target architecture
2040register starting at the specified bit offset.
2041
20421.  ``DW_OP_reg0``, ``DW_OP_reg1``, ..., ``DW_OP_reg31``
2043
2044    ``DW_OP_reg<N>`` operations encode the numbers of up to 32 registers,
2045    numbered from 0 through 31, inclusive. The target architecture register
2046    number R corresponds to the N in the operation name.
2047
2048    The operation is equivalent to performing ``DW_OP_regx R``.
2049
20502.  ``DW_OP_regx``
2051
2052    ``DW_OP_regx`` has a single unsigned LEB128 integer operand that represents
2053    a target architecture register number R.
2054
2055    If the current call frame is the top call frame, it pushes a location
2056    description L that specifies one register location description SL on the
2057    stack. SL specifies the register location storage that corresponds to R with
2058    a bit offset of 0 for the current thread.
2059
2060    If the current call frame is not the top call frame, call frame information
2061    (see :ref:`amdgpu-dwarf-call-frame-information`) is used to determine the
2062    location description that holds the register for the current call frame and
2063    current program location of the current thread. The resulting location
2064    description L is pushed.
2065
2066    *Note that if call frame information is used, the resulting location
2067    description may be register, memory, or undefined.*
2068
2069    *An implementation may evaluate the call frame information immediately, or
2070    may defer evaluation until L is accessed by an operation. If evaluation is
2071    deferred, R and the current context can be recorded in L. When accessed, the
2072    recorded context is used to evaluate the call frame information, not the
2073    current context of the access operation.*
2074
2075*These operations obtain a register location. To fetch the contents of a
2076register, it is necessary to use* ``DW_OP_regval_type``\ *, use one of the*
2077``DW_OP_breg*`` *register-based addressing operations, or use* ``DW_OP_deref*``
2078*on a register location description.*
2079
2080.. _amdgpu-dwarf-implicit-location-descriptions:
2081
2082Implicit Location Description Operations
2083^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2084
2085Implicit location storage represents a piece or all of an object which has no
2086actual location in the program but whose contents are nonetheless known, either
2087as a constant or can be computed from other locations and values in the program.
2088
2089An implicit location description specifies an implicit location storage. The bit
2090offset corresponds to a bit position within the implicit location storage. Bits
2091accessed using an implicit location description, access the corresponding
2092implicit storage value starting at the bit offset.
2093
20941.  ``DW_OP_implicit_value``
2095
2096    ``DW_OP_implicit_value`` has two operands. The first is an unsigned LEB128
2097    integer that represents a byte size S. The second is a block of bytes with a
2098    length equal to S treated as a literal value V.
2099
2100    An implicit location storage LS is created with the literal value V and a
2101    size of S.
2102
2103    It pushes location description L with one implicit location description SL
2104    on the stack. SL specifies LS with a bit offset of 0.
2105
21062.  ``DW_OP_stack_value``
2107
2108    ``DW_OP_stack_value`` pops one stack entry that must be a value V.
2109
2110    An implicit location storage LS is created with the literal value V and a
2111    size equal to V's base type size.
2112
2113    It pushes a location description L with one implicit location description SL
2114    on the stack. SL specifies LS with a bit offset of 0.
2115
2116    *The* ``DW_OP_stack_value`` *operation specifies that the object does not
2117    exist in memory, but its value is nonetheless known. In this form, the
2118    location description specifies the actual value of the object, rather than
2119    specifying the memory or register storage that holds the value.*
2120
2121    See :ref:`amdgpu-dwarf-implicit-location-descriptions` for special rules
2122    concerning implicit pointer values produced by dereferencing implicit
2123    location descriptions created by the ``DW_OP_implicit_pointer`` and
2124    ``DW_OP_LLVM_implicit_aspace_pointer`` operations.
2125
2126    .. note::
2127
2128      Since location descriptions are allowed on the stack, the
2129      ``DW_OP_stack_value`` operation no longer terminates the DWARF operation
2130      expression execution as in DWARF Version 5.
2131
21323.  ``DW_OP_implicit_pointer``
2133
2134    *An optimizing compiler may eliminate a pointer, while still retaining the
2135    value that the pointer addressed.* ``DW_OP_implicit_pointer`` *allows a
2136    producer to describe this value.*
2137
2138    ``DW_OP_implicit_pointer`` *specifies an object is a pointer to the target
2139    architecture default address space that cannot be represented as a real
2140    pointer, even though the value it would point to can be described. In this
2141    form, the location description specifies a debugging information entry that
2142    represents the actual location description of the object to which the
2143    pointer would point. Thus, a consumer of the debug information would be able
2144    to access the dereferenced pointer, even when it cannot access the pointer
2145    itself.*
2146
2147    ``DW_OP_implicit_pointer`` has two operands. The first operand is a 4-byte
2148    unsigned value in the 32-bit DWARF format, or an 8-byte unsigned value in
2149    the 64-bit DWARF format, that represents the byte offset DR of a debugging
2150    information entry D relative to the beginning of the ``.debug_info`` section
2151    that contains the current compilation unit. The second operand is a signed
2152    LEB128 integer that represents a byte displacement B.
2153
2154    *Note that D may not be in the current compilation unit.*
2155
2156    *The first operand interpretation is exactly like that for*
2157    ``DW_FORM_ref_addr``\ *.*
2158
2159    The address space identifier AS is defined as the one corresponding to the
2160    target architecture specific default address space.
2161
2162    The address size S is defined as the address bit size of the target
2163    architecture specific address space corresponding to AS.
2164
2165    An implicit location storage LS is created with the debugging information
2166    entry D, address space AS, and size of S.
2167
2168    It pushes a location description L that comprises one implicit location
2169    description SL on the stack. SL specifies LS with a bit offset of 0.
2170
2171    It is an evaluation error if a ``DW_OP_deref*`` operation pops a location
2172    description L', and retrieves S bits, such that any retrieved bits come from
2173    an implicit location storage that is the same as LS, unless both the
2174    following conditions are met:
2175
2176    1.  All retrieved bits come from an implicit location description that
2177        refers to an implicit location storage that is the same as LS.
2178
2179        *Note that all bits do not have to come from the same implicit location
2180        description, as L' may involve composite location descriptors.*
2181
2182    2.  The bits come from consecutive ascending offsets within their respective
2183        implicit location storage.
2184
2185    *These rules are equivalent to retrieving the complete contents of LS.*
2186
2187    If both the above conditions are met, then the value V pushed by the
2188    ``DW_OP_deref*`` operation is an implicit pointer value IPV with a target
2189    architecture specific address space of AS, a debugging information entry of
2190    D, and a base type of T. If AS is the target architecture default address
2191    space, then T is the generic type. Otherwise, T is a target architecture
2192    specific integral type with a bit size equal to S.
2193
2194    If IPV is either implicitly converted to a location description (only done
2195    if AS is the target architecture default address space) or used by
2196    ``DW_OP_LLVM_form_aspace_address`` (only done if the address space popped by
2197    ``DW_OP_LLVM_form_aspace_address`` is AS), then the resulting location
2198    description RL is:
2199
2200    * If D has a ``DW_AT_location`` attribute, the DWARF expression E from the
2201      ``DW_AT_location`` attribute is evaluated with the current context, except
2202      that the result kind is a location description, the compilation unit is
2203      the one that contains D, the object is unspecified, and the initial stack
2204      is empty. RL is the expression result.
2205
2206      *Note that E is evaluated with the context of the expression accessing
2207      IPV, and not the context of the expression that contained the*
2208      ``DW_OP_implicit_pointer`` *or* ``DW_OP_LLVM_aspace_implicit_pointer``
2209      *operation that created L.*
2210
2211    * If D has a ``DW_AT_const_value`` attribute, then an implicit location
2212      storage RLS is created from the ``DW_AT_const_value`` attribute's value
2213      with a size matching the size of the ``DW_AT_const_value`` attribute's
2214      value. RL comprises one implicit location description SRL. SRL specifies
2215      RLS with a bit offset of 0.
2216
2217      .. note::
2218
2219        If using ``DW_AT_const_value`` for variables and formal parameters is
2220        deprecated and instead ``DW_AT_location`` is used with an implicit
2221        location description, then this rule would not be required.
2222
2223    * Otherwise, it is an evaluation error.
2224
2225    The bit offset of RL is updated as if the ``DW_OP_LLVM_offset_uconst B``
2226    operation was applied.
2227
2228    If a ``DW_OP_stack_value`` operation pops a value that is the same as IPV,
2229    then it pushes a location description that is the same as L.
2230
2231    It is an evaluation error if LS or IPV is accessed in any other manner.
2232
2233    *The restrictions on how an implicit pointer location description created
2234    by* ``DW_OP_implicit_pointer`` *and* ``DW_OP_LLVM_aspace_implicit_pointer``
2235    *can be used are to simplify the DWARF consumer. Similarly, for an implicit
2236    pointer value created by* ``DW_OP_deref*`` *and* ``DW_OP_stack_value``\ .*
2237
22384.  ``DW_OP_LLVM_aspace_implicit_pointer`` *New*
2239
2240    ``DW_OP_LLVM_aspace_implicit_pointer`` has two operands that are the same as
2241    for ``DW_OP_implicit_pointer``.
2242
2243    It pops one stack entry that must be an integral type value that represents
2244    a target architecture specific address space identifier AS.
2245
2246    The location description L that is pushed on the stack is the same as for
2247    ``DW_OP_implicit_pointer``, except that the address space identifier used is
2248    AS.
2249
2250    The DWARF expression is ill-formed if AS is not one of the values defined by
2251    the target architecture specific ``DW_ASPACE_*`` values.
2252
2253    .. note::
2254
2255      This definition of ``DW_OP_LLVM_aspace_implicit_pointer`` may change when
2256      full support for address classes is added as required for languages such
2257      as OpenCL/SyCL.
2258
2259*Typically a* ``DW_OP_implicit_pointer`` *or*
2260``DW_OP_LLVM_aspace_implicit_pointer`` *operation is used in a DWARF expression
2261E*\ :sub:`1` *of a* ``DW_TAG_variable`` *or* ``DW_TAG_formal_parameter``
2262*debugging information entry D*\ :sub:`1`\ *'s* ``DW_AT_location`` *attribute.
2263The debugging information entry referenced by the* ``DW_OP_implicit_pointer``
2264*or* ``DW_OP_LLVM_aspace_implicit_pointer`` *operations is typically itself a*
2265``DW_TAG_variable`` *or* ``DW_TAG_formal_parameter`` *debugging information
2266entry D*\ :sub:`2` *whose* ``DW_AT_location`` *attribute gives a second DWARF
2267expression E*\ :sub:`2`\ *.*
2268
2269*D*\ :sub:`1` *and E*\ :sub:`1` *are describing the location of a pointer type
2270object. D*\ :sub:`2` *and E*\ :sub:`2` *are describing the location of the
2271object pointed to by that pointer object.*
2272
2273*However, D*\ :sub:`2` *may be any debugging information entry that contains a*
2274``DW_AT_location`` *or* ``DW_AT_const_value`` *attribute (for example,*
2275``DW_TAG_dwarf_procedure``\ *). By using E*\ :sub:`2`\ *, a consumer can
2276reconstruct the value of the object when asked to dereference the pointer
2277described by E*\ :sub:`1` *which contains the* ``DW_OP_implicit_pointer`` or
2278``DW_OP_LLVM_aspace_implicit_pointer`` *operation.*
2279
2280.. _amdgpu-dwarf-composite-location-description-operations:
2281
2282Composite Location Description Operations
2283^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2284
2285A composite location storage represents an object or value which may be
2286contained in part of another location storage or contained in parts of more
2287than one location storage.
2288
2289Each part has a part location description L and a part bit size S. L can have
2290one or more single location descriptions SL. If there are more than one SL then
2291that indicates that part is located in more than one place. The bits of each
2292place of the part comprise S contiguous bits from the location storage LS
2293specified by SL starting at the bit offset specified by SL. All the bits must
2294be within the size of LS or the DWARF expression is ill-formed.
2295
2296A composite location storage can have zero or more parts. The parts are
2297contiguous such that the zero-based location storage bit index will range over
2298each part with no gaps between them. Therefore, the size of a composite location
2299storage is the sum of the size of its parts. The DWARF expression is ill-formed
2300if the size of the contiguous location storage is larger than the size of the
2301memory location storage corresponding to the largest target architecture
2302specific address space.
2303
2304A composite location description specifies a composite location storage. The bit
2305offset corresponds to a bit position within the composite location storage.
2306
2307There are operations that create a composite location storage.
2308
2309There are other operations that allow a composite location storage to be
2310incrementally created. Each part is created by a separate operation. There may
2311be one or more operations to create the final composite location storage. A
2312series of such operations describes the parts of the composite location storage
2313that are in the order that the associated part operations are executed.
2314
2315To support incremental creation, a composite location storage can be in an
2316incomplete state. When an incremental operation operates on an incomplete
2317composite location storage, it adds a new part, otherwise it creates a new
2318composite location storage. The ``DW_OP_LLVM_piece_end`` operation explicitly
2319makes an incomplete composite location storage complete.
2320
2321A composite location description that specifies a composite location storage
2322that is incomplete is termed an incomplete composite location description. A
2323composite location description that specifies a composite location storage that
2324is complete is termed a complete composite location description.
2325
2326If the top stack entry is a location description that has one incomplete
2327composite location description SL after the execution of an operation expression
2328has completed, SL is converted to a complete composite location description.
2329
2330*Note that this conversion does not happen after the completion of an operation
2331expression that is evaluated on the same stack by the* ``DW_OP_call*``
2332*operations. Such executions are not a separate evaluation of an operation
2333expression, but rather the continued evaluation of the same operation expression
2334that contains the* ``DW_OP_call*`` *operation.*
2335
2336If a stack entry is required to be a location description L, but L has an
2337incomplete composite location description, then the DWARF expression is
2338ill-formed. The exception is for the operations involved in incrementally
2339creating a composite location description as described below.
2340
2341*Note that a DWARF operation expression may arbitrarily compose composite
2342location descriptions from any other location description, including those that
2343have multiple single location descriptions, and those that have composite
2344location descriptions.*
2345
2346*The incremental composite location description operations are defined to be
2347compatible with the definitions in DWARF Version 5.*
2348
23491.  ``DW_OP_piece``
2350
2351    ``DW_OP_piece`` has a single unsigned LEB128 integer that represents a byte
2352    size S.
2353
2354    The action is based on the context:
2355
2356    * If the stack is empty, then a location description L comprised of one
2357      incomplete composite location description SL is pushed on the stack.
2358
2359      An incomplete composite location storage LS is created with a single part
2360      P. P specifies a location description PL and has a bit size of S scaled by
2361      8 (the byte size). PL is comprised of one undefined location description
2362      PSL.
2363
2364      SL specifies LS with a bit offset of 0.
2365
2366    * Otherwise, if the top stack entry is a location description L comprised of
2367      one incomplete composite location description SL, then the incomplete
2368      composite location storage LS that SL specifies is updated to append a new
2369      part P. P specifies a location description PL and has a bit size of S
2370      scaled by 8 (the byte size). PL is comprised of one undefined location
2371      description PSL. L is left on the stack.
2372
2373    * Otherwise, if the top stack entry is a location description or can be
2374      converted to one, then it is popped and treated as a part location
2375      description PL. Then:
2376
2377      * If the top stack entry (after popping PL) is a location description L
2378        comprised of one incomplete composite location description SL, then the
2379        incomplete composite location storage LS that SL specifies is updated to
2380        append a new part P. P specifies the location description PL and has a
2381        bit size of S scaled by 8 (the byte size). L is left on the stack.
2382
2383      * Otherwise, a location description L comprised of one incomplete
2384        composite location description SL is pushed on the stack.
2385
2386        An incomplete composite location storage LS is created with a single
2387        part P. P specifies the location description PL and has a bit size of S
2388        scaled by 8 (the byte size).
2389
2390        SL specifies LS with a bit offset of 0.
2391
2392    * Otherwise, the DWARF expression is ill-formed
2393
2394    *Many compilers store a single variable in sets of registers or store a
2395    variable partially in memory and partially in registers.* ``DW_OP_piece``
2396    *provides a way of describing where a part of a variable is located.*
2397
2398    *If a non-0 byte displacement is required, the* ``DW_OP_LLVM_offset``
2399    *operation can be used to update the location description before using it as
2400    the part location description of a* ``DW_OP_piece`` *operation.*
2401
2402    *The evaluation rules for the* ``DW_OP_piece`` *operation allow it to be
2403    compatible with the DWARF Version 5 definition.*
2404
2405    .. note::
2406
2407      Since these extensions allow location descriptions to be entries on the
2408      stack, a simpler operation to create composite location descriptions. For
2409      example, just one operation that specifies how many parts, and pops pairs
2410      of stack entries for the part size and location description. Not only
2411      would this be a simpler operation and avoid the complexities of incomplete
2412      composite location descriptions, but it may also have a smaller encoding
2413      in practice. However, the desire for compatibility with DWARF Version 5 is
2414      likely a stronger consideration.
2415
24162.  ``DW_OP_bit_piece``
2417
2418    ``DW_OP_bit_piece`` has two operands. The first is an unsigned LEB128
2419    integer that represents the part bit size S. The second is an unsigned
2420    LEB128 integer that represents a bit displacement B.
2421
2422    The action is the same as for ``DW_OP_piece``, except that any part created
2423    has the bit size S, and the location description PL of any created part is
2424    updated as if the ``DW_OP_constu B; DW_OP_LLVM_bit_offset`` operations were
2425    applied.
2426
2427    ``DW_OP_bit_piece`` *is used instead of* ``DW_OP_piece`` *when the piece to
2428    be assembled is not byte-sized or is not at the start of the part location
2429    description.*
2430
2431    *If a computed bit displacement is required, the* ``DW_OP_LLVM_bit_offset``
2432    *operation can be used to update the location description before using it as
2433    the part location description of a* ``DW_OP_bit_piece`` *operation.*
2434
2435    .. note::
2436
2437      The bit offset operand is not needed as ``DW_OP_LLVM_bit_offset`` can be
2438      used on the part's location description.
2439
24403.  ``DW_OP_LLVM_piece_end`` *New*
2441
2442    If the top stack entry is not a location description L comprised of one
2443    incomplete composite location description SL, then the DWARF expression is
2444    ill-formed.
2445
2446    Otherwise, the incomplete composite location storage LS specified by SL is
2447    updated to be a complete composite location description with the same parts.
2448
24494.  ``DW_OP_LLVM_extend`` *New*
2450
2451    ``DW_OP_LLVM_extend`` has two operands. The first is an unsigned LEB128
2452    integer that represents the element bit size S. The second is an unsigned
2453    LEB128 integer that represents a count C.
2454
2455    It pops one stack entry that must be a location description and is treated
2456    as the part location description PL.
2457
2458    A location description L comprised of one complete composite location
2459    description SL is pushed on the stack.
2460
2461    A complete composite location storage LS is created with C identical parts
2462    P. Each P specifies PL and has a bit size of S.
2463
2464    SL specifies LS with a bit offset of 0.
2465
2466    The DWARF expression is ill-formed if the element bit size or count are 0.
2467
24685.  ``DW_OP_LLVM_select_bit_piece`` *New*
2469
2470    ``DW_OP_LLVM_select_bit_piece`` has two operands. The first is an unsigned
2471    LEB128 integer that represents the element bit size S. The second is an
2472    unsigned LEB128 integer that represents a count C.
2473
2474    It pops three stack entries. The first must be an integral type value that
2475    represents a bit mask value M. The second must be a location description
2476    that represents the one-location description L1. The third must be a
2477    location description that represents the zero-location description L0.
2478
2479    A complete composite location storage LS is created with C parts P\ :sub:`N`
2480    ordered in ascending N from 0 to C-1 inclusive. Each P\ :sub:`N` specifies
2481    location description PL\ :sub:`N` and has a bit size of S.
2482
2483    PL\ :sub:`N` is as if the ``DW_OP_LLVM_bit_offset N*S`` operation was
2484    applied to PLX\ :sub:`N`\ .
2485
2486    PLX\ :sub:`N` is the same as L0 if the N\ :sup:`th` least significant bit of
2487    M is a zero, otherwise it is the same as L1.
2488
2489    A location description L comprised of one complete composite location
2490    description SL is pushed on the stack. SL specifies LS with a bit offset of
2491    0.
2492
2493    The DWARF expression is ill-formed if S or C are 0, or if the bit size of M
2494    is less than C.
2495
2496.. _amdgpu-dwarf-location-list-expressions:
2497
2498DWARF Location List Expressions
2499+++++++++++++++++++++++++++++++
2500
2501*To meet the needs of recent computer architectures and optimization techniques,
2502debugging information must be able to describe the location of an object whose
2503location changes over the object’s lifetime, and may reside at multiple
2504locations during parts of an object's lifetime. Location list expressions are
2505used in place of operation expressions whenever the object whose location is
2506being described has these requirements.*
2507
2508A location list expression consists of a series of location list entries. Each
2509location list entry is one of the following kinds:
2510
2511*Bounded location description*
2512
2513  This kind of location list entry provides an operation expression that
2514  evaluates to the location description of an object that is valid over a
2515  lifetime bounded by a starting and ending address. The starting address is the
2516  lowest address of the address range over which the location is valid. The
2517  ending address is the address of the first location past the highest address
2518  of the address range.
2519
2520  The location list entry matches when the current program location is within
2521  the given range.
2522
2523  There are several kinds of bounded location description entries which differ
2524  in the way that they specify the starting and ending addresses.
2525
2526*Default location description*
2527
2528  This kind of location list entry provides an operation expression that
2529  evaluates to the location description of an object that is valid when no
2530  bounded location description entry applies.
2531
2532  The location list entry matches when the current program location is not
2533  within the range of any bounded location description entry.
2534
2535*Base address*
2536
2537  This kind of location list entry provides an address to be used as the base
2538  address for beginning and ending address offsets given in certain kinds of
2539  bounded location description entries. The applicable base address of a bounded
2540  location description entry is the address specified by the closest preceding
2541  base address entry in the same location list. If there is no preceding base
2542  address entry, then the applicable base address defaults to the base address
2543  of the compilation unit (see DWARF Version 5 section 3.1.1).
2544
2545  In the case of a compilation unit where all of the machine code is contained
2546  in a single contiguous section, no base address entry is needed.
2547
2548*End-of-list*
2549
2550  This kind of location list entry marks the end of the location list
2551  expression.
2552
2553The address ranges defined by the bounded location description entries of a
2554location list expression may overlap. When they do, they describe a situation in
2555which an object exists simultaneously in more than one place.
2556
2557If all of the address ranges in a given location list expression do not
2558collectively cover the entire range over which the object in question is
2559defined, and there is no following default location description entry, it is
2560assumed that the object is not available for the portion of the range that is
2561not covered.
2562
2563The result of the evaluation of a DWARF location list expression is:
2564
2565* If the current program location is not specified, then it is an evaluation
2566  error.
2567
2568  .. note::
2569
2570    If the location list only has a single default entry, should that be
2571    considered a match if there is no program location? If there are non-default
2572    entries then it seems it has to be an evaluation error when there is no
2573    program location as that indicates the location depends on the program
2574    location which is not known.
2575
2576* If there are no matching location list entries, then the result is a location
2577  description that comprises one undefined location description.
2578
2579* Otherwise, the operation expression E of each matching location list entry is
2580  evaluated with the current context, except that the result kind is a location
2581  description, the object is unspecified, and the initial stack is empty. The
2582  location list entry result is the location description returned by the
2583  evaluation of E.
2584
2585  The result is a location description that is comprised of the union of the
2586  single location descriptions of the location description result of each
2587  matching location list entry.
2588
2589A location list expression can only be used as the value of a debugger
2590information entry attribute that is encoded using class ``loclist`` or
2591``loclistsptr`` (see DWARF Version 5 section 7.5.5). The value of the attribute
2592provides an index into a separate object file section called ``.debug_loclists``
2593or ``.debug_loclists.dwo`` (for split DWARF object files) that contains the
2594location list entries.
2595
2596A ``DW_OP_call*`` and ``DW_OP_implicit_pointer`` operation can be used to
2597specify a debugger information entry attribute that has a location list
2598expression. Several debugger information entry attributes allow DWARF
2599expressions that are evaluated with an initial stack that includes a location
2600description that may originate from the evaluation of a location list
2601expression.
2602
2603*This location list representation, the* ``loclist`` *and* ``loclistsptr``
2604*class, and the related* ``DW_AT_loclists_base`` *attribute are new in DWARF
2605Version 5. Together they eliminate most, or all of the code object relocations
2606previously needed for location list expressions.*
2607
2608.. note::
2609
2610  The rest of this section is the same as DWARF Version 5 section 2.6.2.
2611
2612.. _amdgpu-dwarf-segment_addresses:
2613
2614Segmented Addresses
2615~~~~~~~~~~~~~~~~~~~
2616
2617.. note::
2618
2619  This augments DWARF Version 5 section 2.12.
2620
2621DWARF address classes are used for source languages that have the concept of
2622memory spaces. They are used in the ``DW_AT_address_class`` attribute for
2623pointer type, reference type, subprogram, and subprogram type debugger
2624information entries.
2625
2626Each DWARF address class is conceptually a separate source language memory space
2627with its own lifetime and aliasing rules. DWARF address classes are used to
2628specify the source language memory spaces that pointer type and reference type
2629values refer, and to specify the source language memory space in which variables
2630are allocated.
2631
2632The set of currently defined source language DWARF address classes, together
2633with source language mappings, is given in
2634:ref:`amdgpu-dwarf-address-class-table`.
2635
2636Vendor defined source language address classes may be defined using codes in the
2637range ``DW_ADDR_LLVM_lo_user`` to ``DW_ADDR_LLVM_hi_user``.
2638
2639.. table:: Address class
2640   :name: amdgpu-dwarf-address-class-table
2641
2642   ========================= ============ ========= ========= =========
2643   Address Class Name        Meaning      C/C++     OpenCL    CUDA/HIP
2644   ========================= ============ ========= ========= =========
2645   ``DW_ADDR_none``          generic      *default* generic   *default*
2646   ``DW_ADDR_LLVM_global``   global                 global
2647   ``DW_ADDR_LLVM_constant`` constant               constant  constant
2648   ``DW_ADDR_LLVM_group``    thread-group           local     shared
2649   ``DW_ADDR_LLVM_private``  thread                 private
2650   ``DW_ADDR_LLVM_lo_user``
2651   ``DW_ADDR_LLVM_hi_user``
2652   ========================= ============ ========= ========= =========
2653
2654DWARF address spaces correspond to target architecture specific linear
2655addressable memory areas. They are used in DWARF expression location
2656descriptions to describe in which target architecture specific memory area data
2657resides.
2658
2659*Target architecture specific DWARF address spaces may correspond to hardware
2660supported facilities such as memory utilizing base address registers, scratchpad
2661memory, and memory with special interleaving. The size of addresses in these
2662address spaces may vary. Their access and allocation may be hardware managed
2663with each thread or group of threads having access to independent storage. For
2664these reasons they may have properties that do not allow them to be viewed as
2665part of the unified global virtual address space accessible by all threads.*
2666
2667*It is target architecture specific whether multiple DWARF address spaces are
2668supported and how source language DWARF address classes map to target
2669architecture specific DWARF address spaces. A target architecture may map
2670multiple source language DWARF address classes to the same target architecture
2671specific DWARF address class. Optimization may determine that variable lifetime
2672and access pattern allows them to be allocated in faster scratchpad memory
2673represented by a different DWARF address space.*
2674
2675Although DWARF address space identifiers are target architecture specific,
2676``DW_ASPACE_none`` is a common address space supported by all target
2677architectures.
2678
2679DWARF address space identifiers are used by:
2680
2681* The DWARF expression operations: ``DW_OP_LLVM_aspace_bregx``,
2682  ``DW_OP_LLVM_form_aspace_address``, ``DW_OP_LLVM_implicit_aspace_pointer``,
2683  and ``DW_OP_xderef*``.
2684
2685* The CFI instructions: ``DW_CFA_def_aspace_cfa`` and
2686  ``DW_CFA_def_aspace_cfa_sf``.
2687
2688.. note::
2689
2690  With the definition of DWARF address classes and DWARF address spaces in these
2691  extensions, DWARF Version 5 table 2.7 needs to be updated. It seems it is an
2692  example of DWARF address spaces and not DWARF address classes.
2693
2694.. note::
2695
2696  With the expanded support for DWARF address spaces in these extensions, it may
2697  be worth examining if DWARF segments can be eliminated and DWARF address
2698  spaces used instead.
2699
2700  That may involve extending DWARF address spaces to also be used to specify
2701  code locations. In target architectures that use different memory areas for
2702  code and data this would seem a natural use for DWARF address spaces. This
2703  would allow DWARF expression location descriptions to be used to describe the
2704  location of subprograms and entry points that are used in expressions
2705  involving subprogram pointer type values.
2706
2707  Currently, DWARF expressions assume data and code resides in the same default
2708  DWARF address space, and only the address ranges in DWARF location list
2709  entries and in the ``.debug_aranges`` section for accelerated access for
2710  addresses allow DWARF segments to be used to distinguish.
2711
2712.. note::
2713
2714  Currently, DWARF defines address class values as being target architecture
2715  specific. It is unclear how language specific memory spaces are intended to be
2716  represented in DWARF using these.
2717
2718  For example, OpenCL defines memory spaces (called address spaces in OpenCL)
2719  for ``global``, ``local``, ``constant``, and ``private``. These are part of
2720  the type system and are modifiers to pointer types. In addition, OpenCL
2721  defines ``generic`` pointers that can reference either the ``global``,
2722  ``local``, or ``private`` memory spaces. To support the OpenCL language the
2723  debugger would want to support casting pointers between the ``generic`` and
2724  other memory spaces, querying what memory space a ``generic`` pointer value is
2725  currently referencing, and possibly using pointer casting to form an address
2726  for a specific memory space out of an integral value.
2727
2728  The method to use to dereference a pointer type or reference type value is
2729  defined in DWARF expressions using ``DW_OP_xderef*`` which uses a target
2730  architecture specific address space.
2731
2732  DWARF defines the ``DW_AT_address_class`` attribute on pointer type and
2733  reference type debugger information entries. It specifies the method to use to
2734  dereference them. Why is the value of this not the same as the address space
2735  value used in ``DW_OP_xderef*``? In both cases it is target architecture
2736  specific and the architecture presumably will use the same set of methods to
2737  dereference pointers in both cases.
2738
2739  Since ``DW_AT_address_class`` uses a target architecture specific value, it
2740  cannot in general capture the source language memory space type modifier
2741  concept. On some architectures all source language memory space modifiers may
2742  actually use the same method for dereferencing pointers.
2743
2744  One possibility is for DWARF to add an ``DW_TAG_LLVM_address_class_type``
2745  debugger information entry type modifier that can be applied to a pointer type
2746  and reference type. The ``DW_AT_address_class`` attribute could be re-defined
2747  to not be target architecture specific and instead define generalized language
2748  values (as presented above for DWARF address classes in the table
2749  :ref:`amdgpu-dwarf-address-class-table`) that will support OpenCL and other
2750  languages using memory spaces. The ``DW_AT_address_class`` attribute could be
2751  defined to not be applied to pointer types or reference types, but instead
2752  only to the new ``DW_TAG_LLVM_address_class_type`` type modifier debugger
2753  information entry.
2754
2755  If a pointer type or reference type is not modified by
2756  ``DW_TAG_LLVM_address_class_type`` or if ``DW_TAG_LLVM_address_class_type``
2757  has no ``DW_AT_address_class`` attribute, then the pointer type or reference
2758  type would be defined to use the ``DW_ADDR_none`` address class as currently.
2759  Since modifiers can be chained, it would need to be defined if multiple
2760  ``DW_TAG_LLVM_address_class_type`` modifiers were legal, and if so if the
2761  outermost one is the one that takes precedence.
2762
2763  A target architecture implementation that supports multiple address spaces
2764  would need to map ``DW_ADDR_none`` appropriately to support CUDA-like
2765  languages that have no address classes in the type system but do support
2766  variable allocation in address classes. Such variable allocation would result
2767  in the variable's location description needing an address space.
2768
2769  The approach presented in :ref:`amdgpu-dwarf-address-class-table` is to define
2770  the default ``DW_ADDR_none`` to be the generic address class and not the
2771  global address class. This matches how CLANG and LLVM have added support for
2772  CUDA-like languages on top of existing C++ language support. This allows all
2773  addresses to be generic by default which matches CUDA-like languages.
2774
2775  An alternative approach is to define ``DW_ADDR_none`` as being the global
2776  address class and then change ``DW_ADDR_LLVM_global`` to
2777  ``DW_ADDR_LLVM_generic``. This would match the reality that languages that do
2778  not support multiple memory spaces only have one default global memory space.
2779  Generally, in these languages if they expose that the target architecture
2780  supports multiple address spaces, the default one is still the global memory
2781  space. Then a language that does support multiple memory spaces has to
2782  explicitly indicate which pointers have the added ability to reference more
2783  than the global memory space. However, compilers generating DWARF for
2784  CUDA-like languages would then have to define every CUDA-like language pointer
2785  type or reference type using ``DW_TAG_LLVM_address_class_type`` with a
2786  ``DW_AT_address_class`` attribute of ``DW_ADDR_LLVM_generic`` to match the
2787  language semantics.
2788
2789  A new ``DW_AT_LLVM_address_space`` attribute could be defined that can be
2790  applied to pointer type, reference type, subprogram, and subprogram type to
2791  describe how objects having the given type are dereferenced or called (the
2792  role that ``DW_AT_address_class`` currently provides). The values of
2793  ``DW_AT_address_space`` would be target architecture specific and the same as
2794  used in ``DW_OP_xderef*``.
2795
2796.. note::
2797
2798  Some additional changes will be made to support languages such as OpenCL/SyCL
2799  that allow address class pointer casting and queries.
2800
2801  This requires the compiler to provide the mapping from address space to
2802  address class which may be runtime and not target architecture dependent. Some
2803  implementations may have a one-to-one mapping from source language address
2804  class to target architecture address space, and some may have a many-to-one
2805  mapping which requires knowledge of the address class when determining if
2806  pointer address class casts are allowed.
2807
2808  The changes will likely add an attribute that has an expression provided by
2809  the compiler to map from address class to address space. The
2810  ``DW_OP_implicit_pointer`` and ``DW_OP_LLVM_aspace_implicit_pointer``
2811  operations may be changed as the current IPV definition may not provide enough
2812  information when used to cast between address classes. Other attributes and
2813  operations may be needed. The legal casts between address classes may need to
2814  be defined on a per language address class basis.
2815
2816.. _amdgpu-dwarf-debugging-information-entry-attributes:
2817
2818Debugging Information Entry Attributes
2819~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2820
2821.. note::
2822
2823  This section provides changes to existing debugger information entry
2824  attributes and defines attributes added by these extensions. These would be
2825  incorporated into the appropriate DWARF Version 5 chapter 2 sections.
2826
28271.  ``DW_AT_location``
2828
2829    Any debugging information entry describing a data object (which includes
2830    variables and parameters) or common blocks may have a ``DW_AT_location``
2831    attribute, whose value is a DWARF expression E.
2832
2833    The result of the attribute is obtained by evaluating E with a context that
2834    has a result kind of a location description, an unspecified object, the
2835    compilation unit that contains E, an empty initial stack, and other context
2836    elements corresponding to the source language thread of execution upon which
2837    the user is focused, if any. The result of the evaluation is the location
2838    description of the base of the data object.
2839
2840    See :ref:`amdgpu-dwarf-control-flow-operations` for special evaluation rules
2841    used by the ``DW_OP_call*`` operations.
2842
2843    .. note::
2844
2845      Delete the description of how the ``DW_OP_call*`` operations evaluate a
2846      ``DW_AT_location`` attribute as that is now described in the operations.
2847
2848    .. note::
2849
2850      See the discussion about the ``DW_AT_location`` attribute in the
2851      ``DW_OP_call*`` operation. Having each attribute only have a single
2852      purpose and single execution semantics seems desirable. It makes it easier
2853      for the consumer that no longer have to track the context. It makes it
2854      easier for the producer as it can rely on a single semantics for each
2855      attribute.
2856
2857      For that reason, limiting the ``DW_AT_location`` attribute to only
2858      supporting evaluating the location description of an object, and using a
2859      different attribute and encoding class for the evaluation of DWARF
2860      expression *procedures* on the same operation expression stack seems
2861      desirable.
2862
28632.  ``DW_AT_const_value``
2864
2865    .. note::
2866
2867      Could deprecate using the ``DW_AT_const_value`` attribute for
2868      ``DW_TAG_variable`` or ``DW_TAG_formal_parameter`` debugger information
2869      entries that have been optimized to a constant. Instead,
2870      ``DW_AT_location`` could be used with a DWARF expression that produces an
2871      implicit location description now that any location description can be
2872      used within a DWARF expression. This allows the ``DW_OP_call*`` operations
2873      to be used to push the location description of any variable regardless of
2874      how it is optimized.
2875
28763.  ``DW_AT_frame_base``
2877
2878    A ``DW_TAG_subprogram`` or ``DW_TAG_entry_point`` debugger information entry
2879    may have a ``DW_AT_frame_base`` attribute, whose value is a DWARF expression
2880    E.
2881
2882    The result of the attribute is obtained by evaluating E with a context that
2883    has a result kind of a location description, an unspecified object, the
2884    compilation unit that contains E, an empty initial stack, and other context
2885    elements corresponding to the source language thread of execution upon which
2886    the user is focused, if any.
2887
2888    The DWARF is ill-formed if E contains an ``DW_OP_fbreg`` operation, or the
2889    resulting location description L is not comprised of one single location
2890    description SL.
2891
2892    If SL a register location description for register R, then L is replaced
2893    with the result of evaluating a ``DW_OP_bregx R, 0`` operation. This
2894    computes the frame base memory location description in the target
2895    architecture default address space.
2896
2897    *This allows the more compact* ``DW_OPreg*`` *to be used instead of*
2898    ``DW_OP_breg* 0``\ *.*
2899
2900    .. note::
2901
2902      This rule could be removed and require the producer to create the required
2903      location description directly using ``DW_OP_call_frame_cfa``,
2904      ``DW_OP_breg*``, or ``DW_OP_LLVM_aspace_bregx``. This would also then
2905      allow a target to implement the call frames within a large register.
2906
2907    Otherwise, the DWARF is ill-formed if SL is not a memory location
2908    description in any of the target architecture specific address spaces.
2909
2910    The resulting L is the *frame base* for the subprogram or entry point.
2911
2912    *Typically, E will use the* ``DW_OP_call_frame_cfa`` *operation or be a
2913    stack pointer register plus or minus some offset.*
2914
29154.  ``DW_AT_data_member_location``
2916
2917    For a ``DW_AT_data_member_location`` attribute there are two cases:
2918
2919    1.  If the attribute is an integer constant B, it provides the offset in
2920        bytes from the beginning of the containing entity.
2921
2922        The result of the attribute is obtained by evaluating a
2923        ``DW_OP_LLVM_offset B`` operation with an initial stack comprising the
2924        location description of the beginning of the containing entity.  The
2925        result of the evaluation is the location description of the base of the
2926        member entry.
2927
2928        *If the beginning of the containing entity is not byte aligned, then the
2929        beginning of the member entry has the same bit displacement within a
2930        byte.*
2931
2932    2.  Otherwise, the attribute must be a DWARF expression E which is evaluated
2933        with a context that has a result kind of a location description, an
2934        unspecified object, the compilation unit that contains E, an initial
2935        stack comprising the location description of the beginning of the
2936        containing entity, and other context elements corresponding to the
2937        source language thread of execution upon which the user is focused, if
2938        any. The result of the evaluation is the location description of the
2939        base of the member entry.
2940
2941    .. note::
2942
2943      The beginning of the containing entity can now be any location
2944      description, including those with more than one single location
2945      description, and those with single location descriptions that are of any
2946      kind and have any bit offset.
2947
29485.  ``DW_AT_use_location``
2949
2950    The ``DW_TAG_ptr_to_member_type`` debugging information entry has a
2951    ``DW_AT_use_location`` attribute whose value is a DWARF expression E. It is
2952    used to compute the location description of the member of the class to which
2953    the pointer to member entry points.
2954
2955    *The method used to find the location description of a given member of a
2956    class, structure, or union is common to any instance of that class,
2957    structure, or union and to any instance of the pointer to member type. The
2958    method is thus associated with the pointer to member type, rather than with
2959    each object that has a pointer to member type.*
2960
2961    The ``DW_AT_use_location`` DWARF expression is used in conjunction with the
2962    location description for a particular object of the given pointer to member
2963    type and for a particular structure or class instance.
2964
2965    The result of the attribute is obtained by evaluating E with a context that
2966    has a result kind of a location description, an unspecified object, the
2967    compilation unit that contains E, an initial stack comprising two entries,
2968    and other context elements corresponding to the source language thread of
2969    execution upon which the user is focused, if any. The first stack entry is
2970    the value of the pointer to member object itself. The second stack entry is
2971    the location description of the base of the entire class, structure, or
2972    union instance containing the member whose location is being calculated. The
2973    result of the evaluation is the location description of the member of the
2974    class to which the pointer to member entry points.
2975
29766.  ``DW_AT_data_location``
2977
2978    The ``DW_AT_data_location`` attribute may be used with any type that
2979    provides one or more levels of hidden indirection and/or run-time parameters
2980    in its representation. Its value is a DWARF operation expression E which
2981    computes the location description of the data for an object. When this
2982    attribute is omitted, the location description of the data is the same as
2983    the location description of the object.
2984
2985    The result of the attribute is obtained by evaluating E with a context that
2986    has a result kind of a location description, an object that is the location
2987    description of the data descriptor, the compilation unit that contains E, an
2988    empty initial stack, and other context elements corresponding to the source
2989    language thread of execution upon which the user is focused, if any. The
2990    result of the evaluation is the location description of the base of the
2991    member entry.
2992
2993    *E will typically involve an operation expression that begins with a*
2994    ``DW_OP_push_object_address`` *operation which loads the location
2995    description of the object which can then serve as a description in
2996    subsequent calculation.*
2997
2998    .. note::
2999
3000      Since ``DW_AT_data_member_location``, ``DW_AT_use_location``, and
3001      ``DW_AT_vtable_elem_location`` allow both operation expressions and
3002      location list expressions, why does ``DW_AT_data_location`` not allow
3003      both? In all cases they apply to data objects so less likely that
3004      optimization would cause different operation expressions for different
3005      program location ranges. But if supporting for some then should be for
3006      all.
3007
3008      It seems odd this attribute is not the same as
3009      ``DW_AT_data_member_location`` in having an initial stack with the
3010      location description of the object since the expression has to need it.
3011
30127.  ``DW_AT_vtable_elem_location``
3013
3014    An entry for a virtual function also has a ``DW_AT_vtable_elem_location``
3015    attribute whose value is a DWARF expression E.
3016
3017    The result of the attribute is obtained by evaluating E with a context that
3018    has a result kind of a location description, an unspecified object, the
3019    compilation unit that contains E, an initial stack comprising the location
3020    description of the object of the enclosing type, and other context elements
3021    corresponding to the source language thread of execution upon which the user
3022    is focused, if any. The result of the evaluation is the location description
3023    of the slot for the function within the virtual function table for the
3024    enclosing class.
3025
30268.  ``DW_AT_static_link``
3027
3028    If a ``DW_TAG_subprogram`` or ``DW_TAG_entry_point`` debugger information
3029    entry is lexically nested, it may have a ``DW_AT_static_link`` attribute,
3030    whose value is a DWARF expression E.
3031
3032    The result of the attribute is obtained by evaluating E with a context that
3033    has a result kind of a location description, an unspecified object, the
3034    compilation unit that contains E, an empty initial stack, and other context
3035    elements corresponding to the source language thread of execution upon which
3036    the user is focused, if any. The result of the evaluation is the location
3037    description L of the *canonical frame address* (see
3038    :ref:`amdgpu-dwarf-call-frame-information`) of the relevant call frame of
3039    the subprogram instance that immediately lexically encloses the current call
3040    frame's subprogram or entry point.
3041
3042    The DWARF is ill-formed if L is is not comprised of one memory location
3043    description for one of the target architecture specific address spaces.
3044
30459.  ``DW_AT_return_addr``
3046
3047    A ``DW_TAG_subprogram``, ``DW_TAG_inlined_subroutine``, or
3048    ``DW_TAG_entry_point`` debugger information entry may have a
3049    ``DW_AT_return_addr`` attribute, whose value is a DWARF expression E.
3050
3051    The result of the attribute is obtained by evaluating E with a context that
3052    has a result kind of a location description, an unspecified object, the
3053    compilation unit that contains E, an empty initial stack, and other context
3054    elements corresponding to the source language thread of execution upon which
3055    the user is focused, if any. The result of the evaluation is the location
3056    description L of the place where the return address for the current call
3057    frame's subprogram or entry point is stored.
3058
3059    The DWARF is ill-formed if L is not comprised of one memory location
3060    description for one of the target architecture specific address spaces.
3061
3062    .. note::
3063
3064      It is unclear why ``DW_TAG_inlined_subroutine`` has a
3065      ``DW_AT_return_addr`` attribute but not a ``DW_AT_frame_base`` or
3066      ``DW_AT_static_link`` attribute. Seems it would either have all of them or
3067      none. Since inlined subprograms do not have a call frame it seems they
3068      would have none of these attributes.
3069
307010. ``DW_AT_call_value``, ``DW_AT_call_data_location``, and
3071    ``DW_AT_call_data_value``
3072
3073    A ``DW_TAG_call_site_parameter`` debugger information entry may have a
3074    ``DW_AT_call_value`` attribute, whose value is a DWARF operation expression
3075    E\ :sub:`1`\ .
3076
3077    The result of the ``DW_AT_call_value`` attribute is obtained by evaluating
3078    E\ :sub:`1` with a context that has a result kind of a value, an unspecified
3079    object, the compilation unit that contains E, an empty initial stack, and
3080    other context elements corresponding to the source language thread of
3081    execution upon which the user is focused, if any. The resulting value V\
3082    :sub:`1` is the value of the parameter at the time of the call made by the
3083    call site.
3084
3085    For parameters passed by reference, where the code passes a pointer to a
3086    location which contains the parameter, or for reference type parameters, the
3087    ``DW_TAG_call_site_parameter`` debugger information entry may also have a
3088    ``DW_AT_call_data_location`` attribute whose value is a DWARF operation
3089    expression E\ :sub:`2`\ , and a ``DW_AT_call_data_value`` attribute whose
3090    value is a DWARF operation expression E\ :sub:`3`\ .
3091
3092    The value of the ``DW_AT_call_data_location`` attribute is obtained by
3093    evaluating E\ :sub:`2` with a context that has a result kind of a location
3094    description, an unspecified object, the compilation unit that contains E, an
3095    empty initial stack, and other context elements corresponding to the source
3096    language thread of execution upon which the user is focused, if any. The
3097    resulting location description L\ :sub:`2` is the location where the
3098    referenced parameter lives during the call made by the call site. If E\
3099    :sub:`2` would just be a ``DW_OP_push_object_address``, then the
3100    ``DW_AT_call_data_location`` attribute may be omitted.
3101
3102    The value of the ``DW_AT_call_data_value`` attribute is obtained by
3103    evaluating E\ :sub:`3` with a context that has a result kind of a value, an
3104    unspecified object, the compilation unit that contains E, an empty initial
3105    stack, and other context elements corresponding to the source language
3106    thread of execution upon which the user is focused, if any. The resulting
3107    value V\ :sub:`3` is the value in L\ :sub:`2` at the time of the call made
3108    by the call site.
3109
3110    The result of these attributes is undefined if the current call frame is
3111    not for the subprogram containing the ``DW_TAG_call_site_parameter``
3112    debugger information entry or the current program location is not for the
3113    call site containing the ``DW_TAG_call_site_parameter`` debugger information
3114    entry in the current call frame.
3115
3116    *The consumer may have to virtually unwind to the call site (see*
3117    :ref:`amdgpu-dwarf-call-frame-information`\ *) in order to evaluate these
3118    attributes. This will ensure the source language thread of execution upon
3119    which the user is focused corresponds to the call site needed to evaluate
3120    the expression.*
3121
3122    If it is not possible to avoid the expressions of these attributes from
3123    accessing registers or memory locations that might be clobbered by the
3124    subprogram being called by the call site, then the associated attribute
3125    should not be provided.
3126
3127    *The reason for the restriction is that the parameter may need to be
3128    accessed during the execution of the callee. The consumer may virtually
3129    unwind from the called subprogram back to the caller and then evaluate the
3130    attribute expressions. The call frame information (see*
3131    :ref:`amdgpu-dwarf-call-frame-information`\ *) will not be able to restore
3132    registers that have been clobbered, and clobbered memory will no longer have
3133    the value at the time of the call.*
3134
313511. ``DW_AT_LLVM_lanes`` *New*
3136
3137    For languages that are implemented using a SIMD or SIMT execution model, a
3138    ``DW_TAG_subprogram``, ``DW_TAG_inlined_subroutine``, or
3139    ``DW_TAG_entry_point`` debugger information entry may have a
3140    ``DW_AT_LLVM_lanes`` attribute whose value is an integer constant that is
3141    the number of lanes per thread. This is the static number of lanes per
3142    thread. It is not the dynamic number of lanes with which the thread was
3143    initiated, for example, due to smaller or partial work-groups.
3144
3145    If not present, the default value of 1 is used.
3146
3147    The DWARF is ill-formed if the value is 0.
3148
314912. ``DW_AT_LLVM_lane_pc`` *New*
3150
3151    For languages that are implemented using a SIMD or SIMT execution model, a
3152    ``DW_TAG_subprogram``, ``DW_TAG_inlined_subroutine``, or
3153    ``DW_TAG_entry_point`` debugging information entry may have a
3154    ``DW_AT_LLVM_lane_pc`` attribute whose value is a DWARF expression E.
3155
3156    The result of the attribute is obtained by evaluating E with a context that
3157    has a result kind of a location description, an unspecified object, the
3158    compilation unit that contains E, an empty initial stack, and other context
3159    elements corresponding to the source language thread of execution upon which
3160    the user is focused, if any.
3161
3162    The resulting location description L is for a thread lane count sized vector
3163    of generic type elements. The thread lane count is the value of the
3164    ``DW_AT_LLVM_lanes`` attribute. Each element holds the conceptual program
3165    location of the corresponding lane, where the least significant element
3166    corresponds to the first target architecture specific lane identifier and so
3167    forth. If the lane was not active when the current subprogram was called,
3168    its element is an undefined location description.
3169
3170    ``DW_AT_LLVM_lane_pc`` *allows the compiler to indicate conceptually where
3171    each lane of a SIMT thread is positioned even when it is in divergent
3172    control flow that is not active.*
3173
3174    *Typically, the result is a location description with one composite location
3175    description with each part being a location description with either one
3176    undefined location description or one memory location description.*
3177
3178    If not present, the thread is not being used in a SIMT manner, and the
3179    thread's current program location is used.
3180
318113. ``DW_AT_LLVM_active_lane`` *New*
3182
3183    For languages that are implemented using a SIMD or SIMT execution model, a
3184    ``DW_TAG_subprogram``, ``DW_TAG_inlined_subroutine``, or
3185    ``DW_TAG_entry_point`` debugger information entry may have a
3186    ``DW_AT_LLVM_active_lane`` attribute whose value is a DWARF expression E.
3187
3188    The result of the attribute is obtained by evaluating E with a context that
3189    has a result kind of a value, an unspecified object, the compilation unit
3190    that contains E, an empty initial stack, and other context elements
3191    corresponding to the source language thread of execution upon which the user
3192    is focused, if any.
3193
3194    The DWARF is ill-formed if the resulting value V is not an integral value.
3195
3196    The resulting V is a bit mask of active lanes for the current program
3197    location. The N\ :sup:`th` least significant bit of the mask corresponds to
3198    the N\ :sup:`th` lane. If the bit is 1 the lane is active, otherwise it is
3199    inactive.
3200
3201    *Some targets may update the target architecture execution mask for regions
3202    of code that must execute with different sets of lanes than the current
3203    active lanes. For example, some code must execute with all lanes made
3204    temporarily active.* ``DW_AT_LLVM_active_lane`` *allows the compiler to
3205    provide the means to determine the source language active lanes.*
3206
3207    If not present and ``DW_AT_LLVM_lanes`` is greater than 1, then the target
3208    architecture execution mask is used.
3209
321014. ``DW_AT_LLVM_vector_size`` *New*
3211
3212    A ``DW_TAG_base_type`` debugger information entry for a base type T may have
3213    a ``DW_AT_LLVM_vector_size`` attribute whose value is an integer constant
3214    that is the vector type size N.
3215
3216    The representation of a vector base type is as N contiguous elements, each
3217    one having the representation of a base type T' that is the same as T
3218    without the ``DW_AT_LLVM_vector_size`` attribute.
3219
3220    If a ``DW_TAG_base_type`` debugger information entry does not have a
3221    ``DW_AT_LLVM_vector_size`` attribute, then the base type is not a vector
3222    type.
3223
3224    The DWARF is ill-formed if N is not greater than 0.
3225
3226    .. note::
3227
3228      LLVM has mention of a non-upstreamed debugger information entry that is
3229      intended to support vector types. However, that was not for a base type so
3230      would not be suitable as the type of a stack value entry. But perhaps that
3231      could be replaced by using this attribute.
3232
323315. ``DW_AT_LLVM_augmentation`` *New*
3234
3235    A ``DW_TAG_compile_unit`` debugger information entry for a compilation unit
3236    may have a ``DW_AT_LLVM_augmentation`` attribute, whose value is an
3237    augmentation string.
3238
3239    *The augmentation string allows producers to indicate that there is
3240    additional vendor or target specific information in the debugging
3241    information entries. For example, this might be information about the
3242    version of vendor specific extensions that are being used.*
3243
3244    If not present, or if the string is empty, then the compilation unit has no
3245    augmentation string.
3246
3247    The format for the augmentation string is:
3248
3249      | ``[``\ *vendor*\ ``:v``\ *X*\ ``.``\ *Y*\ [\ ``:``\ *options*\ ]\ ``]``\ *
3250
3251    Where *vendor* is the producer, ``vX.Y`` specifies the major X and minor Y
3252    version number of the extensions used, and *options* is an optional string
3253    providing additional information about the extensions. The version number
3254    must conform to semantic versioning [:ref:`SEMVER <amdgpu-dwarf-SEMVER>`].
3255    The *options* string must not contain the "\ ``]``\ " character.
3256
3257    For example:
3258
3259      ::
3260
3261        [abc:v0.0][def:v1.2:feature-a=on,feature-b=3]
3262
3263Program Scope Entities
3264----------------------
3265
3266.. _amdgpu-dwarf-language-names:
3267
3268Unit Entities
3269~~~~~~~~~~~~~
3270
3271.. note::
3272
3273  This augments DWARF Version 5 section 3.1.1 and Table 3.1.
3274
3275Additional language codes defined for use with the ``DW_AT_language`` attribute
3276are defined in :ref:`amdgpu-dwarf-language-names-table`.
3277
3278.. table:: Language Names
3279   :name: amdgpu-dwarf-language-names-table
3280
3281   ==================== =============================
3282   Language Name        Meaning
3283   ==================== =============================
3284   ``DW_LANG_LLVM_HIP`` HIP Language.
3285   ==================== =============================
3286
3287The HIP language [:ref:`HIP <amdgpu-dwarf-HIP>`] can be supported by extending
3288the C++ language.
3289
3290Other Debugger Information
3291--------------------------
3292
3293Accelerated Access
3294~~~~~~~~~~~~~~~~~~
3295
3296.. _amdgpu-dwarf-lookup-by-name:
3297
3298Lookup By Name
3299++++++++++++++
3300
3301Contents of the Name Index
3302##########################
3303
3304.. note::
3305
3306  The following provides changes to DWARF Version 5 section 6.1.1.1.
3307
3308  The rule for debugger information entries included in the name index in the
3309  optional ``.debug_names`` section is extended to also include named
3310  ``DW_TAG_variable`` debugging information entries with a ``DW_AT_location``
3311  attribute that includes a ``DW_OP_LLVM_form_aspace_address`` operation.
3312
3313The name index must contain an entry for each debugging information entry that
3314defines a named subprogram, label, variable, type, or namespace, subject to the
3315following rules:
3316
3317* ``DW_TAG_variable`` debugging information entries with a ``DW_AT_location``
3318  attribute that includes a ``DW_OP_addr``, ``DW_OP_LLVM_form_aspace_address``,
3319  or ``DW_OP_form_tls_address`` operation are included; otherwise, they are
3320  excluded.
3321
3322Data Representation of the Name Index
3323#####################################
3324
3325Section Header
3326^^^^^^^^^^^^^^
3327
3328.. note::
3329
3330  The following provides an addition to DWARF Version 5 section 6.1.1.4.1 item
3331  14 ``augmentation_string``.
3332
3333A null-terminated UTF-8 vendor specific augmentation string, which provides
3334additional information about the contents of this index. If provided, the
3335recommended format for augmentation string is:
3336
3337  | ``[``\ *vendor*\ ``:v``\ *X*\ ``.``\ *Y*\ [\ ``:``\ *options*\ ]\ ``]``\ *
3338
3339Where *vendor* is the producer, ``vX.Y`` specifies the major X and minor Y
3340version number of the extensions used in the DWARF of the compilation unit, and
3341*options* is an optional string providing additional information about the
3342extensions. The version number must conform to semantic versioning [:ref:`SEMVER
3343<amdgpu-dwarf-SEMVER>`]. The *options* string must not contain the "\ ``]``\ "
3344character.
3345
3346For example:
3347
3348  ::
3349
3350    [abc:v0.0][def:v1.2:feature-a=on,feature-b=3]
3351
3352.. note::
3353
3354  This is different to the definition in DWARF Version 5 but is consistent with
3355  the other augmentation strings and allows multiple vendor extensions to be
3356  supported.
3357
3358.. _amdgpu-dwarf-line-number-information:
3359
3360Line Number Information
3361~~~~~~~~~~~~~~~~~~~~~~~
3362
3363The Line Number Program Header
3364++++++++++++++++++++++++++++++
3365
3366Standard Content Descriptions
3367#############################
3368
3369.. note::
3370
3371  This augments DWARF Version 5 section 6.2.4.1.
3372
3373.. _amdgpu-dwarf-line-number-information-dw-lnct-llvm-source:
3374
33751.  ``DW_LNCT_LLVM_source``
3376
3377    The component is a null-terminated UTF-8 source text string with "\ ``\n``\
3378    " line endings. This content code is paired with the same forms as
3379    ``DW_LNCT_path``. It can be used for file name entries.
3380
3381    The value is an empty null-terminated string if no source is available. If
3382    the source is available but is an empty file then the value is a
3383    null-terminated single "\ ``\n``\ ".
3384
3385    *When the source field is present, consumers can use the embedded source
3386    instead of attempting to discover the source on disk using the file path
3387    provided by the* ``DW_LNCT_path`` *field. When the source field is absent,
3388    consumers can access the file to get the source text.*
3389
3390    *This is particularly useful for programming languages that support runtime
3391    compilation and runtime generation of source text. In these cases, the
3392    source text does not reside in any permanent file. For example, the OpenCL
3393    language [:ref:`OpenCL <amdgpu-dwarf-OpenCL>`] supports online compilation.*
3394
33952.  ``DW_LNCT_LLVM_is_MD5``
3396
3397    ``DW_LNCT_LLVM_is_MD5`` indicates if the ``DW_LNCT_MD5`` content kind, if
3398    present, is valid: when 0 it is not valid and when 1 it is valid. If
3399    ``DW_LNCT_LLVM_is_MD5`` content kind is not present, and ``DW_LNCT_MD5``
3400    content kind is present, then the MD5 checksum is valid.
3401
3402    ``DW_LNCT_LLVM_is_MD5`` is always paired with the ``DW_FORM_udata`` form.
3403
3404    *This allows a compilation unit to have a mixture of files with and without
3405    MD5 checksums. This can happen when multiple relocatable files are linked
3406    together.*
3407
3408.. _amdgpu-dwarf-call-frame-information:
3409
3410Call Frame Information
3411~~~~~~~~~~~~~~~~~~~~~~
3412
3413.. note::
3414
3415  This section provides changes to existing call frame information and defines
3416  instructions added by these extensions. Additional support is added for
3417  address spaces. Register unwind DWARF expressions are generalized to allow any
3418  location description, including those with composite and implicit location
3419  descriptions.
3420
3421  These changes would be incorporated into the DWARF Version 5 section 6.1.
3422
3423.. _amdgpu-dwarf-structure_of-call-frame-information:
3424
3425Structure of Call Frame Information
3426+++++++++++++++++++++++++++++++++++
3427
3428The register rules are:
3429
3430*undefined*
3431  A register that has this rule has no recoverable value in the previous frame.
3432  The previous value of this register is the undefined location description (see
3433  :ref:`amdgpu-dwarf-undefined-location-description-operations`).
3434
3435  *By convention, the register is not preserved by a callee.*
3436
3437*same value*
3438  This register has not been modified from the previous caller frame.
3439
3440  If the current frame is the top frame, then the previous value of this
3441  register is the location description L that specifies one register location
3442  description SL. SL specifies the register location storage that corresponds to
3443  the register with a bit offset of 0 for the current thread.
3444
3445  If the current frame is not the top frame, then the previous value of this
3446  register is the location description obtained using the call frame information
3447  for the callee frame and callee program location invoked by the current caller
3448  frame for the same register.
3449
3450  *By convention, the register is preserved by the callee, but the callee has
3451  not modified it.*
3452
3453*offset(N)*
3454  N is a signed byte offset. The previous value of this register is saved at the
3455  location description computed as if the DWARF operation expression
3456  ``DW_OP_LLVM_offset N`` is evaluated with the current context, except the
3457  result kind is a location description, the compilation unit is unspecified,
3458  the object is unspecified, and an initial stack comprising the location
3459  description of the current CFA (see
3460  :ref:`amdgpu-dwarf-operation-expressions`).
3461
3462*val_offset(N)*
3463  N is a signed byte offset. The previous value of this register is the memory
3464  byte address of the location description computed as if the DWARF operation
3465  expression ``DW_OP_LLVM_offset N`` is evaluated with the current context,
3466  except the result kind is a location description, the compilation unit is
3467  unspecified, the object is unspecified, and an initial stack comprising the
3468  location description of the current CFA (see
3469  :ref:`amdgpu-dwarf-operation-expressions`).
3470
3471  The DWARF is ill-formed if the CFA location description is not a memory byte
3472  address location description, or if the register size does not match the size
3473  of an address in the address space of the current CFA location description.
3474
3475  *Since the CFA location description is required to be a memory byte address
3476  location description, the value of val_offset(N) will also be a memory byte
3477  address location description since it is offsetting the CFA location
3478  description by N bytes. Furthermore, the value of val_offset(N) will be a
3479  memory byte address in the same address space as the CFA location
3480  description.*
3481
3482  .. note::
3483
3484    Should DWARF allow the address size to be a different size to the size of
3485    the register? Requiring them to be the same bit size avoids any issue of
3486    conversion as the bit contents of the register is simply interpreted as a
3487    value of the address.
3488
3489    GDB has a per register hook that allows a target specific conversion on a
3490    register by register basis. It defaults to truncation of bigger registers,
3491    and to actually reading bytes from the next register (or reads out of bounds
3492    for the last register) for smaller registers. There are no GDB tests that
3493    read a register out of bounds (except an illegal hand written assembly
3494    test).
3495
3496*register(R)*
3497  This register has been stored in another register numbered R.
3498
3499  The previous value of this register is the location description obtained using
3500  the call frame information for the current frame and current program location
3501  for register R.
3502
3503  The DWARF is ill-formed if the size of this register does not match the size
3504  of register R or if there is a cyclic dependency in the call frame
3505  information.
3506
3507  .. note::
3508
3509    Should this also allow R to be larger than this register? If so is the value
3510    stored in the low order bits and it is undefined what is stored in the
3511    extra upper bits?
3512
3513*expression(E)*
3514  The previous value of this register is located at the location description
3515  produced by evaluating the DWARF operation expression E (see
3516  :ref:`amdgpu-dwarf-operation-expressions`).
3517
3518  E is evaluated with the current context, except the result kind is a location
3519  description, the compilation unit is unspecified, the object is unspecified,
3520  and an initial stack comprising the location description of the current CFA
3521  (see :ref:`amdgpu-dwarf-operation-expressions`).
3522
3523*val_expression(E)*
3524  The previous value of this register is the value produced by evaluating the
3525  DWARF operation expression E (see :ref:`amdgpu-dwarf-operation-expressions`).
3526
3527  E is evaluated with the current context, except the result kind is a value,
3528  the compilation unit is unspecified, the object is unspecified, and an initial
3529  stack comprising the location description of the current CFA (see
3530  :ref:`amdgpu-dwarf-operation-expressions`).
3531
3532  The DWARF is ill-formed if the resulting value type size does not match the
3533  register size.
3534
3535  .. note::
3536
3537    This has limited usefulness as the DWARF expression E can only produce
3538    values up to the size of the generic type. This is due to not allowing any
3539    operations that specify a type in a CFI operation expression. This makes it
3540    unusable for registers that are larger than the generic type. However,
3541    *expression(E)* can be used to create an implicit location description of
3542    any size.
3543
3544*architectural*
3545  The rule is defined externally to this specification by the augmenter.
3546
3547A Common Information Entry (CIE) holds information that is shared among many
3548Frame Description Entries (FDE). There is at least one CIE in every non-empty
3549``.debug_frame`` section. A CIE contains the following fields, in order:
3550
35511.  ``length`` (initial length)
3552
3553    A constant that gives the number of bytes of the CIE structure, not
3554    including the length field itself. The size of the length field plus the
3555    value of length must be an integral multiple of the address size specified
3556    in the ``address_size`` field.
3557
35582.  ``CIE_id`` (4 or 8 bytes, see
3559    :ref:`amdgpu-dwarf-32-bit-and-64-bit-dwarf-formats`)
3560
3561    A constant that is used to distinguish CIEs from FDEs.
3562
3563    In the 32-bit DWARF format, the value of the CIE id in the CIE header is
3564    0xffffffff; in the 64-bit DWARF format, the value is 0xffffffffffffffff.
3565
35663.  ``version`` (ubyte)
3567
3568    A version number. This number is specific to the call frame information and
3569    is independent of the DWARF version number.
3570
3571    The value of the CIE version number is 4.
3572
3573    .. note::
3574
3575      Would this be increased to 5 to reflect the changes in these extensions?
3576
35774.  ``augmentation`` (sequence of UTF-8 characters)
3578
3579    A null-terminated UTF-8 string that identifies the augmentation to this CIE
3580    or to the FDEs that use it. If a reader encounters an augmentation string
3581    that is unexpected, then only the following fields can be read:
3582
3583    * CIE: length, CIE_id, version, augmentation
3584    * FDE: length, CIE_pointer, initial_location, address_range
3585
3586    If there is no augmentation, this value is a zero byte.
3587
3588    *The augmentation string allows users to indicate that there is additional
3589    vendor and target architecture specific information in the CIE or FDE which
3590    is needed to virtually unwind a stack frame. For example, this might be
3591    information about dynamically allocated data which needs to be freed on exit
3592    from the routine.*
3593
3594    *Because the* ``.debug_frame`` *section is useful independently of any*
3595    ``.debug_info`` *section, the augmentation string always uses UTF-8
3596    encoding.*
3597
3598    The recommended format for the augmentation string is:
3599
3600      | ``[``\ *vendor*\ ``:v``\ *X*\ ``.``\ *Y*\ [\ ``:``\ *options*\ ]\ ``]``\ *
3601
3602    Where *vendor* is the producer, ``vX.Y`` specifies the major X and minor Y
3603    version number of the extensions used, and *options* is an optional string
3604    providing additional information about the extensions. The version number
3605    must conform to semantic versioning [:ref:`SEMVER <amdgpu-dwarf-SEMVER>`].
3606    The *options* string must not contain the "\ ``]``\ " character.
3607
3608    For example:
3609
3610      ::
3611
3612        [abc:v0.0][def:v1.2:feature-a=on,feature-b=3]
3613
36145.  ``address_size`` (ubyte)
3615
3616    The size of a target address in this CIE and any FDEs that use it, in bytes.
3617    If a compilation unit exists for this frame, its address size must match the
3618    address size here.
3619
36206.  ``segment_selector_size`` (ubyte)
3621
3622    The size of a segment selector in this CIE and any FDEs that use it, in
3623    bytes.
3624
36257.  ``code_alignment_factor`` (unsigned LEB128)
3626
3627    A constant that is factored out of all advance location instructions (see
3628    :ref:`amdgpu-dwarf-row-creation-instructions`). The resulting value is
3629    ``(operand * code_alignment_factor)``.
3630
36318.  ``data_alignment_factor`` (signed LEB128)
3632
3633    A constant that is factored out of certain offset instructions (see
3634    :ref:`amdgpu-dwarf-cfa-definition-instructions` and
3635    :ref:`amdgpu-dwarf-register-rule-instructions`). The resulting value is
3636    ``(operand * data_alignment_factor)``.
3637
36389.  ``return_address_register`` (unsigned LEB128)
3639
3640    An unsigned LEB128 constant that indicates which column in the rule table
3641    represents the return address of the subprogram. Note that this column might
3642    not correspond to an actual machine register.
3643
3644    The value of the return address register is used to determine the program
3645    location of the caller frame. The program location of the top frame is the
3646    target architecture program counter value of the current thread.
3647
364810. ``initial_instructions`` (array of ubyte)
3649
3650    A sequence of rules that are interpreted to create the initial setting of
3651    each column in the table.
3652
3653    The default rule for all columns before interpretation of the initial
3654    instructions is the undefined rule. However, an ABI authoring body or a
3655    compilation system authoring body may specify an alternate default value for
3656    any or all columns.
3657
365811. ``padding`` (array of ubyte)
3659
3660    Enough ``DW_CFA_nop`` instructions to make the size of this entry match the
3661    length value above.
3662
3663An FDE contains the following fields, in order:
3664
36651.  ``length`` (initial length)
3666
3667    A constant that gives the number of bytes of the header and instruction
3668    stream for this subprogram, not including the length field itself. The size
3669    of the length field plus the value of length must be an integral multiple of
3670    the address size.
3671
36722.  ``CIE_pointer`` (4 or 8 bytes, see
3673    :ref:`amdgpu-dwarf-32-bit-and-64-bit-dwarf-formats`)
3674
3675    A constant offset into the ``.debug_frame`` section that denotes the CIE
3676    that is associated with this FDE.
3677
36783.  ``initial_location`` (segment selector and target address)
3679
3680    The address of the first location associated with this table entry. If the
3681    segment_selector_size field of this FDE’s CIE is non-zero, the initial
3682    location is preceded by a segment selector of the given length.
3683
36844.  ``address_range`` (target address)
3685
3686    The number of bytes of program instructions described by this entry.
3687
36885.  ``instructions`` (array of ubyte)
3689
3690    A sequence of table defining instructions that are described in
3691    :ref:`amdgpu-dwarf-call-frame-instructions`.
3692
36936.  ``padding`` (array of ubyte)
3694
3695    Enough ``DW_CFA_nop`` instructions to make the size of this entry match the
3696    length value above.
3697
3698.. _amdgpu-dwarf-call-frame-instructions:
3699
3700Call Frame Instructions
3701+++++++++++++++++++++++
3702
3703Some call frame instructions have operands that are encoded as DWARF operation
3704expressions E (see :ref:`amdgpu-dwarf-operation-expressions`). The DWARF
3705operations that can be used in E have the following restrictions:
3706
3707* ``DW_OP_addrx``, ``DW_OP_call2``, ``DW_OP_call4``, ``DW_OP_call_ref``,
3708  ``DW_OP_const_type``, ``DW_OP_constx``, ``DW_OP_convert``,
3709  ``DW_OP_deref_type``, ``DW_OP_fbreg``, ``DW_OP_implicit_pointer``,
3710  ``DW_OP_regval_type``, ``DW_OP_reinterpret``, and ``DW_OP_xderef_type``
3711  operations are not allowed because the call frame information must not depend
3712  on other debug sections.
3713
3714* ``DW_OP_push_object_address`` is not allowed because there is no object
3715  context to provide a value to push.
3716
3717* ``DW_OP_LLVM_push_lane`` is not allowed because the call frame instructions
3718  describe the actions for the whole thread, not the lanes independently.
3719
3720* ``DW_OP_call_frame_cfa`` and ``DW_OP_entry_value`` are not allowed because
3721  their use would be circular.
3722
3723* ``DW_OP_LLVM_call_frame_entry_reg`` is not allowed if evaluating E causes a
3724  circular dependency between ``DW_OP_LLVM_call_frame_entry_reg`` operations.
3725
3726  *For example, if a register R1 has a* ``DW_CFA_def_cfa_expression``
3727  *instruction that evaluates a* ``DW_OP_LLVM_call_frame_entry_reg`` *operation
3728  that specifies register R2, and register R2 has a*
3729  ``DW_CFA_def_cfa_expression`` *instruction that that evaluates a*
3730  ``DW_OP_LLVM_call_frame_entry_reg`` *operation that specifies register R1.*
3731
3732*Call frame instructions to which these restrictions apply include*
3733``DW_CFA_def_cfa_expression``\ *,* ``DW_CFA_expression``\ *, and*
3734``DW_CFA_val_expression``\ *.*
3735
3736.. _amdgpu-dwarf-row-creation-instructions:
3737
3738Row Creation Instructions
3739#########################
3740
3741.. note::
3742
3743  These instructions are the same as in DWARF Version 5 section 6.4.2.1.
3744
3745.. _amdgpu-dwarf-cfa-definition-instructions:
3746
3747CFA Definition Instructions
3748###########################
3749
37501.  ``DW_CFA_def_cfa``
3751
3752    The ``DW_CFA_def_cfa`` instruction takes two unsigned LEB128 operands
3753    representing a register number R and a (non-factored) byte displacement B.
3754    AS is set to the target architecture default address space identifier. The
3755    required action is to define the current CFA rule to be the result of
3756    evaluating the DWARF operation expression ``DW_OP_constu AS;
3757    DW_OP_aspace_bregx R, B`` as a location description.
3758
37592.  ``DW_CFA_def_cfa_sf``
3760
3761    The ``DW_CFA_def_cfa_sf`` instruction takes two operands: an unsigned LEB128
3762    value representing a register number R and a signed LEB128 factored byte
3763    displacement B. AS is set to the target architecture default address space
3764    identifier. The required action is to define the current CFA rule to be the
3765    result of evaluating the DWARF operation expression ``DW_OP_constu AS;
3766    DW_OP_aspace_bregx R, B*data_alignment_factor`` as a location description.
3767
3768    *The action is the same as* ``DW_CFA_def_cfa``\ *, except that the second
3769    operand is signed and factored.*
3770
37713.  ``DW_CFA_def_aspace_cfa`` *New*
3772
3773    The ``DW_CFA_def_aspace_cfa`` instruction takes three unsigned LEB128
3774    operands representing a register number R, a (non-factored) byte
3775    displacement B, and a target architecture specific address space identifier
3776    AS. The required action is to define the current CFA rule to be the result
3777    of evaluating the DWARF operation expression ``DW_OP_constu AS;
3778    DW_OP_aspace_bregx R, B`` as a location description.
3779
3780    If AS is not one of the values defined by the target architecture specific
3781    ``DW_ASPACE_*`` values then the DWARF expression is ill-formed.
3782
37834.  ``DW_CFA_def_aspace_cfa_sf`` *New*
3784
3785    The ``DW_CFA_def_cfa_sf`` instruction takes three operands: an unsigned
3786    LEB128 value representing a register number R, a signed LEB128 factored byte
3787    displacement B, and an unsigned LEB128 value representing a target
3788    architecture specific address space identifier AS. The required action is to
3789    define the current CFA rule to be the result of evaluating the DWARF
3790    operation expression ``DW_OP_constu AS; DW_OP_aspace_bregx R,
3791    B*data_alignment_factor`` as a location description.
3792
3793    If AS is not one of the values defined by the target architecture specific
3794    ``DW_ASPACE_*`` values, then the DWARF expression is ill-formed.
3795
3796    *The action is the same as* ``DW_CFA_aspace_def_cfa``\ *, except that the
3797    second operand is signed and factored.*
3798
37995.  ``DW_CFA_def_cfa_register``
3800
3801    The ``DW_CFA_def_cfa_register`` instruction takes a single unsigned LEB128
3802    operand representing a register number R. The required action is to define
3803    the current CFA rule to be the result of evaluating the DWARF operation
3804    expression ``DW_OP_constu AS; DW_OP_aspace_bregx R, B`` as a location
3805    description. B and AS are the old CFA byte displacement and address space
3806    respectively.
3807
3808    If the subprogram has no current CFA rule, or the rule was defined by a
3809    ``DW_CFA_def_cfa_expression`` instruction, then the DWARF is ill-formed.
3810
38116.  ``DW_CFA_def_cfa_offset``
3812
3813    The ``DW_CFA_def_cfa_offset`` instruction takes a single unsigned LEB128
3814    operand representing a (non-factored) byte displacement B. The required
3815    action is to define the current CFA rule to be the result of evaluating the
3816    DWARF operation expression ``DW_OP_constu AS; DW_OP_aspace_bregx R, B`` as a
3817    location description. R and AS are the old CFA register number and address
3818    space respectively.
3819
3820    If the subprogram has no current CFA rule, or the rule was defined by a
3821    ``DW_CFA_def_cfa_expression`` instruction, then the DWARF is ill-formed.
3822
38237.  ``DW_CFA_def_cfa_offset_sf``
3824
3825    The ``DW_CFA_def_cfa_offset_sf`` instruction takes a signed LEB128 operand
3826    representing a factored byte displacement B. The required action is to
3827    define the current CFA rule to be the result of evaluating the DWARF
3828    operation expression ``DW_OP_constu AS; DW_OP_aspace_bregx R,
3829    B*data_alignment_factor`` as a location description. R and AS are the old
3830    CFA register number and address space respectively.
3831
3832    If the subprogram has no current CFA rule, or the rule was defined by a
3833    ``DW_CFA_def_cfa_expression`` instruction, then the DWARF is ill-formed.
3834
3835    *The action is the same as* ``DW_CFA_def_cfa_offset``\ *, except that the
3836    operand is signed and factored.*
3837
38388.  ``DW_CFA_def_cfa_expression``
3839
3840    The ``DW_CFA_def_cfa_expression`` instruction takes a single operand encoded
3841    as a ``DW_FORM_exprloc`` value representing a DWARF operation expression E.
3842    The required action is to define the current CFA rule to be the result of
3843    evaluating E with the current context, except the result kind is a location
3844    description, the compilation unit is unspecified, the object is unspecified,
3845    and an empty initial stack.
3846
3847    *See* :ref:`amdgpu-dwarf-call-frame-instructions` *regarding restrictions on
3848    the DWARF expression operations that can be used in E.*
3849
3850    The DWARF is ill-formed if the result of evaluating E is not a memory byte
3851    address location description.
3852
3853.. _amdgpu-dwarf-register-rule-instructions:
3854
3855Register Rule Instructions
3856##########################
3857
38581.  ``DW_CFA_undefined``
3859
3860    The ``DW_CFA_undefined`` instruction takes a single unsigned LEB128 operand
3861    that represents a register number R. The required action is to set the rule
3862    for the register specified by R to ``undefined``.
3863
38642.  ``DW_CFA_same_value``
3865
3866    The ``DW_CFA_same_value`` instruction takes a single unsigned LEB128 operand
3867    that represents a register number R. The required action is to set the rule
3868    for the register specified by R to ``same value``.
3869
38703.  ``DW_CFA_offset``
3871
3872    The ``DW_CFA_offset`` instruction takes two operands: a register number R
3873    (encoded with the opcode) and an unsigned LEB128 constant representing a
3874    factored displacement B. The required action is to change the rule for the
3875    register specified by R to be an *offset(B\*data_alignment_factor)* rule.
3876
3877    .. note::
3878
3879      Seems this should be named ``DW_CFA_offset_uf`` since the offset is
3880      unsigned factored.
3881
38824.  ``DW_CFA_offset_extended``
3883
3884    The ``DW_CFA_offset_extended`` instruction takes two unsigned LEB128
3885    operands representing a register number R and a factored displacement B.
3886    This instruction is identical to ``DW_CFA_offset``, except for the encoding
3887    and size of the register operand.
3888
3889    .. note::
3890
3891      Seems this should be named ``DW_CFA_offset_extended_uf`` since the
3892      displacement is unsigned factored.
3893
38945.  ``DW_CFA_offset_extended_sf``
3895
3896    The ``DW_CFA_offset_extended_sf`` instruction takes two operands: an
3897    unsigned LEB128 value representing a register number R and a signed LEB128
3898    factored displacement B. This instruction is identical to
3899    ``DW_CFA_offset_extended``, except that B is signed.
3900
39016.  ``DW_CFA_val_offset``
3902
3903    The ``DW_CFA_val_offset`` instruction takes two unsigned LEB128 operands
3904    representing a register number R and a factored displacement B. The required
3905    action is to change the rule for the register indicated by R to be a
3906    *val_offset(B\*data_alignment_factor)* rule.
3907
3908    .. note::
3909
3910      Seems this should be named ``DW_CFA_val_offset_uf`` since the displacement
3911      is unsigned factored.
3912
3913    .. note::
3914
3915      An alternative is to define ``DW_CFA_val_offset`` to implicitly use the
3916      target architecture default address space, and add another operation that
3917      specifies the address space.
3918
39197.  ``DW_CFA_val_offset_sf``
3920
3921    The ``DW_CFA_val_offset_sf`` instruction takes two operands: an unsigned
3922    LEB128 value representing a register number R and a signed LEB128 factored
3923    displacement B. This instruction is identical to ``DW_CFA_val_offset``,
3924    except that B is signed.
3925
39268.  ``DW_CFA_register``
3927
3928    The ``DW_CFA_register`` instruction takes two unsigned LEB128 operands
3929    representing register numbers R1 and R2 respectively. The required action is
3930    to set the rule for the register specified by R1 to be a *register(R2)* rule.
3931
39329.  ``DW_CFA_expression``
3933
3934    The ``DW_CFA_expression`` instruction takes two operands: an unsigned LEB128
3935    value representing a register number R, and a ``DW_FORM_block`` value
3936    representing a DWARF operation expression E. The required action is to
3937    change the rule for the register specified by R to be an *expression(E)*
3938    rule.
3939
3940    *That is, E computes the location description where the register value can
3941    be retrieved.*
3942
3943    *See* :ref:`amdgpu-dwarf-call-frame-instructions` *regarding restrictions on
3944    the DWARF expression operations that can be used in E.*
3945
394610. ``DW_CFA_val_expression``
3947
3948    The ``DW_CFA_val_expression`` instruction takes two operands: an unsigned
3949    LEB128 value representing a register number R, and a ``DW_FORM_block`` value
3950    representing a DWARF operation expression E. The required action is to
3951    change the rule for the register specified by R to be a *val_expression(E)*
3952    rule.
3953
3954    *That is, E computes the value of register R.*
3955
3956    *See* :ref:`amdgpu-dwarf-call-frame-instructions` *regarding restrictions on
3957    the DWARF expression operations that can be used in E.*
3958
3959    If the result of evaluating E is not a value with a base type size that
3960    matches the register size, then the DWARF is ill-formed.
3961
396211. ``DW_CFA_restore``
3963
3964    The ``DW_CFA_restore`` instruction takes a single operand (encoded with the
3965    opcode) that represents a register number R. The required action is to
3966    change the rule for the register specified by R to the rule assigned it by
3967    the ``initial_instructions`` in the CIE.
3968
396912. ``DW_CFA_restore_extended``
3970
3971    The ``DW_CFA_restore_extended`` instruction takes a single unsigned LEB128
3972    operand that represents a register number R. This instruction is identical
3973    to ``DW_CFA_restore``, except for the encoding and size of the register
3974    operand.
3975
3976Row State Instructions
3977######################
3978
3979.. note::
3980
3981  These instructions are the same as in DWARF Version 5 section 6.4.2.4.
3982
3983Padding Instruction
3984###################
3985
3986.. note::
3987
3988  These instructions are the same as in DWARF Version 5 section 6.4.2.5.
3989
3990Call Frame Instruction Usage
3991++++++++++++++++++++++++++++
3992
3993.. note::
3994
3995  The same as in DWARF Version 5 section 6.4.3.
3996
3997.. _amdgpu-dwarf-call-frame-calling-address:
3998
3999Call Frame Calling Address
4000++++++++++++++++++++++++++
4001
4002.. note::
4003
4004  The same as in DWARF Version 5 section 6.4.4.
4005
4006Data Representation
4007-------------------
4008
4009.. _amdgpu-dwarf-32-bit-and-64-bit-dwarf-formats:
4010
401132-Bit and 64-Bit DWARF Formats
4012~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4013
4014.. note::
4015
4016  This augments DWARF Version 5 section 7.4.
4017
40181.  Within the body of the ``.debug_info`` section, certain forms of attribute
4019    value depend on the choice of DWARF format as follows. For the 32-bit DWARF
4020    format, the value is a 4-byte unsigned integer; for the 64-bit DWARF format,
4021    the value is an 8-byte unsigned integer.
4022
4023    .. table:: ``.debug_info`` section attribute form roles
4024      :name: amdgpu-dwarf-debug-info-section-attribute-form-roles-table
4025
4026      ================================== ===================================
4027      Form                               Role
4028      ================================== ===================================
4029      DW_FORM_line_strp                  offset in ``.debug_line_str``
4030      DW_FORM_ref_addr                   offset in ``.debug_info``
4031      DW_FORM_sec_offset                 offset in a section other than
4032                                         ``.debug_info`` or ``.debug_str``
4033      DW_FORM_strp                       offset in ``.debug_str``
4034      DW_FORM_strp_sup                   offset in ``.debug_str`` section of
4035                                         supplementary object file
4036      DW_OP_call_ref                     offset in ``.debug_info``
4037      DW_OP_implicit_pointer             offset in ``.debug_info``
4038      DW_OP_LLVM_aspace_implicit_pointer offset in ``.debug_info``
4039      ================================== ===================================
4040
4041Format of Debugging Information
4042~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4043
4044Attribute Encodings
4045+++++++++++++++++++
4046
4047.. note::
4048
4049  This augments DWARF Version 5 section 7.5.4 and Table 7.5.
4050
4051The following table gives the encoding of the additional debugging information
4052entry attributes.
4053
4054.. table:: Attribute encodings
4055   :name: amdgpu-dwarf-attribute-encodings-table
4056
4057   ================================== ====== ===================================
4058   Attribute Name                     Value  Classes
4059   ================================== ====== ===================================
4060   DW_AT_LLVM_active_lane             0x3e08 exprloc, loclist
4061   DW_AT_LLVM_augmentation            0x3e09 string
4062   DW_AT_LLVM_lanes                   0x3e0a constant
4063   DW_AT_LLVM_lane_pc                 0x3e0b exprloc, loclist
4064   DW_AT_LLVM_vector_size             0x3e0c constant
4065   ================================== ====== ===================================
4066
4067DWARF Expressions
4068~~~~~~~~~~~~~~~~~
4069
4070.. note::
4071
4072  Rename DWARF Version 5 section 7.7 to reflect the unification of location
4073  descriptions into DWARF expressions.
4074
4075Operation Expressions
4076+++++++++++++++++++++
4077
4078.. note::
4079
4080  Rename DWARF Version 5 section 7.7.1 and delete section 7.7.2 to reflect the
4081  unification of location descriptions into DWARF expressions.
4082
4083  This augments DWARF Version 5 section 7.7.1 and Table 7.9.
4084
4085The following table gives the encoding of the additional DWARF expression
4086operations.
4087
4088.. table:: DWARF Operation Encodings
4089   :name: amdgpu-dwarf-operation-encodings-table
4090
4091   ================================== ===== ======== ===============================
4092   Operation                          Code  Number   Notes
4093                                            of
4094                                            Operands
4095   ================================== ===== ======== ===============================
4096   DW_OP_LLVM_form_aspace_address     0xe1     0
4097   DW_OP_LLVM_push_lane               0xe2     0
4098   DW_OP_LLVM_offset                  0xe3     0
4099   DW_OP_LLVM_offset_uconst           0xe4     1     ULEB128 byte displacement
4100   DW_OP_LLVM_bit_offset              0xe5     0
4101   DW_OP_LLVM_call_frame_entry_reg    0xe6     1     ULEB128 register number
4102   DW_OP_LLVM_undefined               0xe7     0
4103   DW_OP_LLVM_aspace_bregx            0xe8     2     ULEB128 register number,
4104                                                     ULEB128 byte displacement
4105   DW_OP_LLVM_aspace_implicit_pointer 0xe9     2     4-byte or 8-byte offset of DIE,
4106                                                     SLEB128 byte displacement
4107   DW_OP_LLVM_piece_end               0xea     0
4108   DW_OP_LLVM_extend                  0xeb     2     ULEB128 bit size,
4109                                                     ULEB128 count
4110   DW_OP_LLVM_select_bit_piece        0xec     2     ULEB128 bit size,
4111                                                     ULEB128 count
4112   ================================== ===== ======== ===============================
4113
4114Location List Expressions
4115+++++++++++++++++++++++++
4116
4117.. note::
4118
4119  Rename DWARF Version 5 section 7.7.3 to reflect that location lists are a kind
4120  of DWARF expression.
4121
4122Source Languages
4123~~~~~~~~~~~~~~~~
4124
4125.. note::
4126
4127  This augments DWARF Version 5 section 7.12 and Table 7.17.
4128
4129The following table gives the encoding of the additional DWARF languages.
4130
4131.. table:: Language encodings
4132   :name: amdgpu-dwarf-language-encodings-table
4133
4134   ==================== ====== ===================
4135   Language Name        Value  Default Lower Bound
4136   ==================== ====== ===================
4137   ``DW_LANG_LLVM_HIP`` 0x8100 0
4138   ==================== ====== ===================
4139
4140Address Class and Address Space Encodings
4141~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4142
4143.. note::
4144
4145  This replaces DWARF Version 5 section 7.13.
4146
4147The encodings of the constants used for the currently defined address classes
4148are given in :ref:`amdgpu-dwarf-address-class-encodings-table`.
4149
4150.. table:: Address class encodings
4151   :name: amdgpu-dwarf-address-class-encodings-table
4152
4153   ========================== ======
4154   Address Class Name         Value
4155   ========================== ======
4156   ``DW_ADDR_none``           0x0000
4157   ``DW_ADDR_LLVM_global``    0x0001
4158   ``DW_ADDR_LLVM_constant``  0x0002
4159   ``DW_ADDR_LLVM_group``     0x0003
4160   ``DW_ADDR_LLVM_private``   0x0004
4161   ``DW_ADDR_LLVM_lo_user``   0x8000
4162   ``DW_ADDR_LLVM_hi_user``   0xffff
4163   ========================== ======
4164
4165Line Number Information
4166~~~~~~~~~~~~~~~~~~~~~~~
4167
4168.. note::
4169
4170  This augments DWARF Version 5 section 7.22 and Table 7.27.
4171
4172The following table gives the encoding of the additional line number header
4173entry formats.
4174
4175.. table:: Line number header entry format encodings
4176  :name: amdgpu-dwarf-line-number-header-entry-format-encodings-table
4177
4178  ====================================  ====================
4179  Line number header entry format name  Value
4180  ====================================  ====================
4181  ``DW_LNCT_LLVM_source``               0x2001
4182  ``DW_LNCT_LLVM_is_MD5``               0x2002
4183  ====================================  ====================
4184
4185Call Frame Information
4186~~~~~~~~~~~~~~~~~~~~~~
4187
4188.. note::
4189
4190  This augments DWARF Version 5 section 7.24 and Table 7.29.
4191
4192The following table gives the encoding of the additional call frame information
4193instructions.
4194
4195.. table:: Call frame instruction encodings
4196   :name: amdgpu-dwarf-call-frame-instruction-encodings-table
4197
4198   ======================== ====== ====== ================ ================ ================
4199   Instruction              High 2 Low 6  Operand 1        Operand 2        Operand 3
4200                            Bits   Bits
4201   ======================== ====== ====== ================ ================ ================
4202   DW_CFA_def_aspace_cfa    0      0x30   ULEB128 register ULEB128 offset   ULEB128 address space
4203   DW_CFA_def_aspace_cfa_sf 0      0x31   ULEB128 register SLEB128 offset   ULEB128 address space
4204   ======================== ====== ====== ================ ================ ================
4205
4206Attributes by Tag Value (Informative)
4207-------------------------------------
4208
4209.. note::
4210
4211  This augments DWARF Version 5 Appendix A and Table A.1.
4212
4213The following table provides the additional attributes that are applicable to
4214debugger information entries.
4215
4216.. table:: Attributes by tag value
4217   :name: amdgpu-dwarf-attributes-by-tag-value-table
4218
4219   ============================= =============================
4220   Tag Name                      Applicable Attributes
4221   ============================= =============================
4222   ``DW_TAG_base_type``          * ``DW_AT_LLVM_vector_size``
4223   ``DW_TAG_compile_unit``       * ``DW_AT_LLVM_augmentation``
4224   ``DW_TAG_entry_point``        * ``DW_AT_LLVM_active_lane``
4225                                 * ``DW_AT_LLVM_lane_pc``
4226                                 * ``DW_AT_LLVM_lanes``
4227   ``DW_TAG_inlined_subroutine`` * ``DW_AT_LLVM_active_lane``
4228                                 * ``DW_AT_LLVM_lane_pc``
4229                                 * ``DW_AT_LLVM_lanes``
4230   ``DW_TAG_subprogram``         * ``DW_AT_LLVM_active_lane``
4231                                 * ``DW_AT_LLVM_lane_pc``
4232                                 * ``DW_AT_LLVM_lanes``
4233   ============================= =============================
4234
4235.. _amdgpu-dwarf-examples:
4236
4237Examples
4238========
4239
4240The AMD GPU specific usage of the features in these extensions, including
4241examples, is available at *User Guide for AMDGPU Backend* section
4242:ref:`amdgpu-dwarf-debug-information`.
4243
4244.. note::
4245
4246  Change examples to use ``DW_OP_LLVM_offset`` instead of ``DW_OP_add`` when
4247  acting on a location description.
4248
4249  Need to provide examples of new features.
4250
4251.. _amdgpu-dwarf-references:
4252
4253References
4254==========
4255
4256    .. _amdgpu-dwarf-AMD:
4257
42581.  [AMD] `Advanced Micro Devices <https://www.amd.com/>`__
4259
4260    .. _amdgpu-dwarf-AMD-ROCm:
4261
42622.  [AMD-ROCm] `AMD ROCm Platform <https://rocm-documentation.readthedocs.io>`__
4263
4264    .. _amdgpu-dwarf-AMD-ROCgdb:
4265
42663.  [AMD-ROCgdb] `AMD ROCm Debugger (ROCgdb) <https://github.com/ROCm-Developer-Tools/ROCgdb>`__
4267
4268    .. _amdgpu-dwarf-AMDGPU-LLVM:
4269
42704.  [AMDGPU-LLVM] `User Guide for AMDGPU LLVM Backend <https://llvm.org/docs/AMDGPUUsage.html>`__
4271
4272    .. _amdgpu-dwarf-CUDA:
4273
42745.  [CUDA] `Nvidia CUDA Language <https://docs.nvidia.com/cuda/cuda-c-programming-guide/>`__
4275
4276    .. _amdgpu-dwarf-DWARF:
4277
42786.  [DWARF] `DWARF Debugging Information Format <http://dwarfstd.org/>`__
4279
4280    .. _amdgpu-dwarf-ELF:
4281
42827.  [ELF] `Executable and Linkable Format (ELF) <http://www.sco.com/developers/gabi/>`__
4283
4284    .. _amdgpu-dwarf-GCC:
4285
42868.  [GCC] `GCC: The GNU Compiler Collection <https://www.gnu.org/software/gcc/>`__
4287
4288    .. _amdgpu-dwarf-GDB:
4289
42909.  [GDB] `GDB: The GNU Project Debugger <https://www.gnu.org/software/gdb/>`__
4291
4292    .. _amdgpu-dwarf-HIP:
4293
429410. [HIP] `HIP Programming Guide <https://rocm-documentation.readthedocs.io/en/latest/Programming_Guides/Programming-Guides.html#hip-programing-guide>`__
4295
4296    .. _amdgpu-dwarf-HSA:
4297
429811. [HSA] `Heterogeneous System Architecture (HSA) Foundation <http://www.hsafoundation.com/>`__
4299
4300    .. _amdgpu-dwarf-LLVM:
4301
430212. [LLVM] `The LLVM Compiler Infrastructure <https://llvm.org/>`__
4303
4304    .. _amdgpu-dwarf-OpenCL:
4305
430613. [OpenCL] `The OpenCL Specification Version 2.0 <http://www.khronos.org/registry/cl/specs/opencl-2.0.pdf>`__
4307
4308    .. _amdgpu-dwarf-Perforce-TotalView:
4309
431014. [Perforce-TotalView] `Perforce TotalView HPC Debugging Software <https://totalview.io/products/totalview>`__
4311
4312    .. _amdgpu-dwarf-SEMVER:
4313
431415. [SEMVER] `Semantic Versioning <https://semver.org/>`__
4315