12 This document contains the release notes for the LLVM Compiler Infrastructure,
13 release 3.8. Here we describe the status of LLVM, including major improvements
14 from the previous release, improvements in various subprojects of LLVM, and
15 some of the current users of the code. All LLVM releases may be downloaded
16 from the `LLVM releases web site <http://llvm.org/releases/>`_.
18 For more information about LLVM, including information about the latest
19 release, please check out the `main LLVM web site <http://llvm.org/>`_. If you
20 have questions or comments, the `LLVM Developer's Mailing List
21 <http://lists.llvm.org/mailman/listinfo/llvm-dev>`_ is a good place to send
24 Non-comprehensive list of changes in this release
25 =================================================
26 * With this release, the minimum Windows version required for running LLVM is
27 Windows 7. Earlier versions, including Windows Vista and XP are no longer
30 * With this release, the autoconf build system is deprecated. It will be removed
31 in the 3.9 release. Please migrate to using CMake. For more information see:
32 `Building LLVM with CMake <CMake.html>`_
34 * We have documented our C API stability guarantees for both development and
35 release branches, as well as documented how to extend the C API. Please see
36 the `developer documentation <DeveloperPolicy.html#c-api-changes>`_ for more
39 * The C API function ``LLVMLinkModules`` is deprecated. It will be removed in the
40 3.9 release. Please migrate to ``LLVMLinkModules2``. Unlike the old function the
43 * Doesn't take an unused parameter.
44 * Destroys the source instead of only damaging it.
45 * Does not record a message. Use the diagnostic handler instead.
47 * The C API functions ``LLVMParseBitcode``, ``LLVMParseBitcodeInContext``,
48 ``LLVMGetBitcodeModuleInContext`` and ``LLVMGetBitcodeModule`` have been deprecated.
49 They will be removed in 3.9. Please migrate to the versions with a 2 suffix.
50 Unlike the old ones the new ones do not record a diagnostic message. Use
51 the diagnostic handler instead.
53 * The deprecated C APIs ``LLVMGetBitcodeModuleProviderInContext`` and
54 ``LLVMGetBitcodeModuleProvider`` have been removed.
56 * The deprecated C APIs ``LLVMCreateExecutionEngine``, ``LLVMCreateInterpreter``,
57 ``LLVMCreateJITCompiler``, ``LLVMAddModuleProvider`` and ``LLVMRemoveModuleProvider``
60 * With this release, the C API headers have been reorganized to improve build
61 time. Type specific declarations have been moved to Type.h, and error
62 handling routines have been moved to ErrorHandling.h. Both are included in
63 Core.h so nothing should change for projects directly including the headers,
64 but transitive dependencies may be affected.
66 * llvm-ar now supports thin archives.
68 * llvm doesn't produce ``.data.rel.ro.local`` or ``.data.rel`` sections anymore.
70 * Aliases to ``available_externally`` globals are now rejected by the verifier.
72 * The IR Linker has been split into ``IRMover`` that moves bits from one module to
73 another and Linker proper that decides what to link.
75 * Support for dematerializing has been dropped.
77 * ``RegisterScheduler::setDefault`` was removed. Targets that used to call into the
78 command line parser to set the ``DAGScheduler``, and that don't have enough
79 control with ``setSchedulingPreference``, should look into overriding the
80 ``SubTargetHook`` "``getDAGScheduler()``".
82 * ``ilist_iterator<T>`` no longer has implicit conversions to and from ``T*``,
83 since ``ilist_iterator<T>`` may be pointing at the sentinel (which is usually
84 not of type ``T`` at all). To convert from an iterator ``I`` to a pointer,
85 use ``&*I``; to convert from a pointer ``P`` to an iterator, use
86 ``P->getIterator()``. Alternatively, explicit conversions via
87 ``static_cast<T>(U)`` are still available.
89 * ``ilist_node<T>::getNextNode()`` and ``ilist_node<T>::getPrevNode()`` now
90 fail at compile time when the node cannot access its parent list.
91 Previously, when the sentinel was was an ``ilist_half_node<T>``, this API
92 could return the sentinel instead of ``nullptr``. Frustrated callers should
93 be updated to use ``iplist<T>::getNextNode(T*)`` instead. Alternatively, if
94 the node ``N`` is guaranteed not to be the last in the list, it is safe to
95 call ``&*++N->getIterator()`` directly.
97 * The `Kaleidoscope tutorials <tutorial/index.html>`_ have been updated to use
100 * ORC now has a basic set of C bindings.
102 * Optional support for linking clang and the LLVM tools with a single libLLVM
103 shared library. To enable this, pass ``-DLLVM_LINK_LLVM_DYLIB=ON`` to CMake.
104 See `Building LLVM with CMake`_ for more details.
106 * The optimization to move the prologue and epilogue of functions in colder
107 code path (shrink-wrapping) is now enabled by default.
109 * A new target-independent gcc-compatible emulated Thread Local Storage mode
110 is added. When ``-femultated-tls`` flag is used, all accesses to TLS
111 variables are converted to calls to ``__emutls_get_address`` in the runtime
114 * MSVC-compatible exception handling has been completely overhauled. New
115 instructions have been introduced to facilitate this:
116 `New exception handling instructions <ExceptionHandling.html#new-exception-handling-instructions>`_.
117 While we have done our best to test this feature thoroughly, it would
118 not be completely surprising if there were a few lingering issues that
119 early adopters might bump into.
122 Changes to the ARM Backends
123 ---------------------------
125 During this release the AArch64 target has:
127 * Added support for more sanitizers (MSAN, TSAN) and made them compatible with
128 all VMA kernel configurations (currently tested on 39 and 42 bits).
129 * Gained initial LLD support in the new ELF back-end
130 * Extended the Load/Store optimiser and cleaned up some of the bad decisions
132 * Expanded LLDB support, including watchpoints, native building, Renderscript,
133 LLDB-server, debugging 32-bit applications.
134 * Added support for the ``Exynos M1`` chip.
136 During this release the ARM target has:
138 * Gained massive performance improvements on embedded benchmarks due to finally
139 running the stride vectorizer in full form, incrementing the performance gains
140 that we already had in the previous releases with limited stride vectorization.
141 * Expanded LLDB support, including watchpoints, unwind tables
142 * Extended the Load/Store optimiser and cleaned up some of the bad decisions
144 * Simplified code generation for global variable addresses in ELF, resulting in
145 a significant (4% in Chromium) reduction in code size.
146 * Gained some additional code size improvements, though there's still a long road
147 ahead, especially for older cores.
148 * Added some EABI floating point comparison functions to Compiler-RT
149 * Added support for Windows+GNU triple, ``+features`` in ``-mcpu``/``-march`` options.
152 Changes to the MIPS Target
153 --------------------------
155 During this release the MIPS target has:
157 * Significantly extended support for the Integrated Assembler. See below for
159 * Added support for the ``P5600`` processor.
160 * Added support for the ``interrupt`` attribute for MIPS32R2 and later. This
161 attribute will generate a function which can be used as a interrupt handler
162 on bare metal MIPS targets using the static relocation model.
163 * Added support for the ``ERETNC`` instruction found in MIPS32R5 and later.
164 * Added support for OpenCL. See http://portablecl.org/.
166 * Address spaces 1 to 255 are now reserved for software use and conversions
167 between them are no-op casts.
169 * Removed the ``mips16`` value for the ``-mcpu`` option since it is an :abbr:`ASE
170 (Application Specific Extension)` and not a processor. If you were using this,
171 please specify another CPU and use ``-mips16`` to enable MIPS16.
172 * Removed ``copy_u.w`` from 32-bit MSA and ``copy_u.d`` from 64-bit MSA since
173 they have been removed from the MSA specification due to forward compatibility
174 issues. For example, 32-bit MSA code containing ``copy_u.w`` would behave
175 differently on a 64-bit processor supporting MSA. The corresponding intrinsics
176 are still available and may expand to ``copy_s.[wd]`` where this is
177 appropriate for forward compatibility purposes.
178 * Relaxed the ``-mnan`` option to allow ``-mnan=2008`` on MIPS32R2/MIPS64R2 for
179 compatibility with GCC.
180 * Made MIPS64R6 the default CPU for 64-bit Android triples.
182 The MIPS target has also fixed various bugs including the following notable
185 * Fixed reversed operands on ``mthi``/``mtlo`` in the DSP :abbr:`ASE
186 (Application Specific Extension)`.
187 * The code generator no longer uses ``jal`` for calls to absolute immediate
189 * Disabled fast instruction selection on MIPS32R6 and MIPS64R6 since this is not
191 * Corrected addend for ``R_MIPS_HI16`` and ``R_MIPS_PCHI16`` in MCJIT
192 * The code generator no longer crashes when handling subregisters of an 64-bit
193 FPU register with undefined value.
194 * The code generator no longer attempts to use ``$zero`` for operands that do
195 not permit ``$zero``.
196 * Corrected the opcode used for ``ll``/``sc`` when using MIPS32R6/MIPS64R6 and
197 the Integrated Assembler.
198 * Added support for atomic load and atomic store.
199 * Corrected debug info when dynamically re-aligning the stack.
201 We have made a large number of improvements to the integrated assembler for
202 MIPS. In this release, the integrated assembler isn't quite production-ready
203 since there are a few known issues related to bare-metal support, checking
204 immediates on instructions, and the N32/N64 ABI's. However, the current support
205 should be sufficient for many users of the O32 ABI, particularly those targeting
206 MIPS32 on Linux or bare-metal MIPS32.
208 If you would like to try the integrated assembler, please use
211 Changes to the PowerPC Target
212 -----------------------------
214 There are numerous improvements to the PowerPC target in this release:
216 * Shrink wrapping optimization has been enabled for PowerPC Little Endian
218 * Direct move instructions are used when converting scalars to vectors
220 * Thread Sanitizer (TSAN) is now supported for PowerPC
222 * New MI peephole pass to clean up redundant XXPERMDI instructions
224 * Add branch hints to highly biased branch instructions (code reaching
225 unreachable terminators and exceptional control flow constructs)
227 * Promote boolean return values to integer to prevent excessive usage of
230 * Additional vector APIs for vector comparisons and vector merges have been
233 * Many bugs have been identified and fixed
236 Changes to the X86 Target
237 -----------------------------
239 * TLS is enabled for Cygwin as emutls.
241 * Smaller code for materializing 32-bit 1 and -1 constants at ``-Os``.
243 * More efficient code for wide integer compares. (E.g. 64-bit compares
246 * Tail call support for ``thiscall``, ``stdcall``, ``vectorcall``, and
247 ``fastcall`` functions.
249 Changes to the Hexagon Target
250 -----------------------------
252 In addition to general code size and performance improvements, Hexagon target
253 now has basic support for Hexagon V60 architecture and Hexagon Vector
256 Changes to the AVR Target
257 -------------------------
259 Slightly less than half of the AVR backend has been merged in at this point. It is still
260 missing a number large parts which cause it to be unusable, but is well on the
261 road to being completely merged and workable.
263 Changes to the OCaml bindings
264 -----------------------------
266 * The ocaml function link_modules has been replaced with link_modules' which
267 uses LLVMLinkModules2.
270 External Open Source Projects Using LLVM 3.8
271 ============================================
273 An exciting aspect of LLVM is that it is used as an enabling technology for
274 a lot of other language and tools projects. This section lists some of the
275 projects that have already been updated to work with LLVM 3.8.
277 LDC - the LLVM-based D compiler
278 -------------------------------
280 `D <http://dlang.org>`_ is a language with C-like syntax and static typing. It
281 pragmatically combines efficiency, control, and modeling power, with safety and
282 programmer productivity. D supports powerful concepts like Compile-Time Function
283 Execution (CTFE) and Template Meta-Programming, provides an innovative approach
284 to concurrency and offers many classical paradigms.
286 `LDC <http://wiki.dlang.org/LDC>`_ uses the frontend from the reference compiler
287 combined with LLVM as backend to produce efficient native code. LDC targets
288 x86/x86_64 systems like Linux, OS X and Windows and also PowerPC (32/64 bit)
289 and ARM. Ports to other architectures like AArch64 and MIPS64 are underway.
292 Additional Information
293 ======================
295 A wide variety of additional information is available on the `LLVM web page
296 <http://llvm.org/>`_, in particular in the `documentation
297 <http://llvm.org/docs/>`_ section. The web page also contains versions of the
298 API documentation which is up-to-date with the Subversion version of the source
299 code. You can access versions of these documents specific to this release by
300 going into the ``llvm/docs/`` directory in the LLVM tree.
302 If you have any questions or comments about LLVM, please feel free to contact
303 us via the `mailing lists <http://llvm.org/docs/#maillist>`_.