llvm.org GIT mirror llvm / cd197f3
ReleaseNotes: tidy up for the release git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_70@341640 91177308-0d34-0410-b5e6-96231b3b80d8 Hans Wennborg 1 year, 1 month ago
1 changed file(s) with 47 addition(s) and 91 deletion(s). Raw diff Collapse all Expand all
33
44 .. contents::
55 :local:
6
7 .. warning::
8 These are in-progress notes for the upcoming LLVM 7 release.
9 Release notes for previous releases can be found on
10 `the Download Page `_.
116
127
138 Introduction
1712 release 7.0.0. Here we describe the status of LLVM, including major improvements
1813 from the previous release, improvements in various subprojects of LLVM, and
1914 some of the current users of the code. All LLVM releases may be downloaded
20 from the `LLVM releases web site ://llvm.org/releases/>`_.
15 from the `LLVM releases web site s://llvm.org/releases/>`_.
2116
2217 For more information about LLVM, including information about the latest
23 release, please check out the `main LLVM web site ://llvm.org/>`_. If you
18 release, please check out the `main LLVM web site s://llvm.org/>`_. If you
2419 have questions or comments, the `LLVM Developer's Mailing List
25 ://lists.llvm.org/mailman/listinfo/llvm-dev>`_ is a good place to send
20 s://lists.llvm.org/mailman/listinfo/llvm-dev>`_ is a good place to send
2621 them.
27
28 Note that if you are reading this file from a Subversion checkout or the main
29 LLVM web page, this document applies to the *next* release, not the current
30 one. To see the release notes for a specific release, please see the `releases
31 page `_.
3222
3323 Non-comprehensive list of changes in this release
3424 =================================================
35 .. NOTE
36 For small 1-3 sentence descriptions, just add an entry at the end of
37 this list. If your description won't fit comfortably in one bullet
38 point (e.g. maybe you would like to give an example of the
39 functionality, or simply have a lot to talk about), see the `NOTE` below
40 for adding a new subsection.
4125
4226 * The Windows installer no longer includes a Visual Studio integration.
4327 Instead, a new
44 `LLVM Compiler Toolchain Visual Studio extension `
28 `LLVM Compiler Toolchain Visual Studio extension `_
4529 is available on the Visual Studio Marketplace. The new integration includes
4630 support for Visual Studio 2017.
4731
4832 * Libraries have been renamed from 7.0 to 7. This change also impacts
4933 downstream libraries like lldb.
5034
51 * The LoopInstSimplify pass (-loop-instsimplify) has been removed.
35 * The LoopInstSimplify pass (``-loop-instsimplify``) has been removed.
5236
5337 * Symbols starting with ``?`` are no longer mangled by LLVM when using the
5438 Windows ``x`` or ``w`` IR mangling schemes.
6347 information available in LLVM to statically predict the performance of
6448 machine code for a specific CPU.
6549
66 * The optimization flag to merge constants (-fmerge-all-constants) is no longer
67 applied by default.
68
6950 * Optimization of floating-point casts is improved. This may cause surprising
70 results for code that is relying on the undefined behavior of overflowing
51 results for code that is relying on the undefined behavior of overflowing
7152 casts. The optimization can be disabled by specifying a function attribute:
72 "strict-float-cast-overflow"="false". This attribute may be created by the
53 ``"strict-float-cast-overflow"="false"``. This attribute may be created by the
7354 clang option ``-fno-strict-float-cast-overflow``.
74 Code sanitizers can be used to detect affected patterns. The option for
75 detecting this problem alone is "-fsanitize=float-cast-overflow":
55 Code sanitizers can be used to detect affected patterns. The clang option for
56 detecting this problem alone is ``-fsanitize=float-cast-overflow``:
7657
7758 .. code-block:: c
7859
8566
8667 .. code-block:: bash
8768
88 clang -O1 ftrunc.c -fsanitize=float-cast-overflow ; ./a.out
69 clang -O1 ftrunc.c -fsanitize=float-cast-overflow ; ./a.out
8970 ftrunc.c:5:15: runtime error: 4.29497e+09 is outside the range of representable values of type 'int'
9071 junk in the ftrunc: 0.000000
9172
10384 git grep -l 'DEBUG' | xargs perl -pi -e 's/\bDEBUG\s?\(/LLVM_DEBUG(/g'
10485 git diff -U0 master | ../clang/tools/clang-format/clang-format-diff.py -i -p1 -style LLVM
10586
106 * Early support for UBsan, X-Ray instrumentation and libFuzzer (x86 and x86_64) for OpenBSD. Support for MSan
107 (x86_64), X-Ray instrumentation and libFuzzer (x86 and x86_64) for FreeBSD.
87 * Early support for UBsan, X-Ray instrumentation and libFuzzer (x86 and x86_64)
88 for OpenBSD. Support for MSan (x86_64), X-Ray instrumentation and libFuzzer
89 (x86 and x86_64) for FreeBSD.
10890
10991 * ``SmallVector`` shrank from ``sizeof(void*) * 4 + sizeof(T)`` to
11092 ``sizeof(void*) + sizeof(unsigned) * 2``, smaller than ``std::vector`` on
11193 64-bit platforms. The maximum capacity is now restricted to ``UINT32_MAX``.
11294 Since SmallVector doesn't have the exception-safety pessimizations some
113 implementations saddle std::vector with and is better at using ``realloc``,
114 it's now a better choice even on the heap (although when TinyPtrVector works,
115 it's even smaller).
95 implementations saddle ``std::vector`` with and is better at using ``realloc``,
96 it's now a better choice even on the heap (although when ``TinyPtrVector`` works,
97 that's even smaller).
11698
11799 * Preliminary/experimental support for DWARF v5 debugging information,
118 including the new .debug_names accelerator table. DWARF emitted at ``-O0``
100 including the new ``.debug_names`` accelerator table. DWARF emitted at ``-O0``
119101 should be fully DWARF v5 compliant. Type units and split DWARF are known
120102 not to be compliant, and higher optimization levels will still emit some
121103 information in v4 format.
128110 but it can now handle leftover C declarations in preprocessor output, if
129111 given output from a preprocessor run externally.)
130112
131 * CodeView debug info can now be emitted MinGW configurations, if requested.
132
133 * The :program:`opt` tool now supports the :option:`-load-pass-plugin` for
113 * CodeView debug info can now be emitted for MinGW configurations, if requested.
114
115 * The :program:`opt` tool now supports the ``-load-pass-plugin`` option for
134116 loading pass plugins for the new PassManager.
135117
136 * Note..
137
138 .. NOTE
139 If you would like to document a larger change, then you can add a
140 subsection about it right here. You can copy the following boilerplate
141 and un-indent it (the indentation causes it to be inside this comment).
142
143 Special New Feature
144 -------------------
145
146 Makes programs 10x faster by doing Special New Thing.
147118
148119 Changes to the LLVM IR
149120 ----------------------
150121
151 * The signatures for the builtins @llvm.memcpy, @llvm.memmove, and @llvm.memset
152 have changed. Alignment is no longer an argument, and are instead conveyed as
153 parameter attributes.
154
155 * invariant.group.barrier has been renamed to launder.invariant.group.
156
157 * invariant.group metadata can now refer only empty metadata nodes.
122 * The signatures for the builtins ``@llvm.memcpy``, ``@llvm.memmove``, and
123 ``@llvm.memset`` have changed. Alignment is no longer an argument, and are
124 instead conveyed as parameter attributes.
125
126 * ``invariant.group.barrier`` has been renamed to ``launder.invariant.group``.
127
128 * ``invariant.group`` metadata can now refer only to empty metadata nodes.
158129
159130 Changes to the AArch64 Target
160131 -----------------------------
200171 * Introduced definitions of ``[d]rem``, ``[d]remu``,
201172 and microMIPSR6 ``ll/sc`` instructions.
202173
203 * Shrink-wrapping is now supported and enabled by default (except for -O0).
174 * Shrink-wrapping is now supported and enabled by default (except for ``-O0``).
204175
205176 * Extended size reduction pass by the LWP and SWP instructions.
206177
213184
214185 * Improved the selection of multiple instructions.
215186
216 * Load/store lb, sb, ld, sd, lld, ... instructions
187 * Load/store ``lb``, ``sb``, ``ld``, ``sd``, ``lld``, ... instructions
217188 now support 32/64-bit offsets.
218189
219190 * Added support for ``y``, ``M``, and ``L`` inline assembler operand codes.
221192 * Extended list of relocations supported by the ``.reloc`` directive
222193
223194 * Fixed using a wrong register class for creating an emergency
224 spill slot for mips3 / n64 abi.
195 spill slot for mips3 / n64 ABI.
225196
226197 * MIPS relocation types were generated for microMIPS code.
227198
228199 * Corrected definitions of multiple instructions (``lwp``, ``swp``, ``ctc2``,
229200 ``cfc2``, ``sync``, ``synci``, ``cvt.d.w``, ...).
230201
231 * Fixed atomic operations at O0 level.
202 * Fixed atomic operations at ``-O0`` level.
232203
233204 * Fixed local dynamic TLS with Sym64
234205
239210
240211 * Replaced the list scheduler for post register allocation with the machine scheduler.
241212
242 * Added support for coldcc calling convention.
213 * Added support for ``coldcc`` calling convention.
243214
244215 * Added support for ``symbol@high`` and ``symbol@higha`` symbol modifiers.
245216
246 * Added support for quad-precision floating point type (``__float128``) under the llvm option `-enable-ppc-quad-precision`.
217 * Added support for quad-precision floating point type (``__float128``) under the llvm option ``-enable-ppc-quad-precision``.
247218
248219 * Added dump function to ``LatencyPriorityQueue``.
249220
292263 environments - in MSVC environments, long doubles are the same size as
293264 normal doubles.)
294265
295 Changes to the AMDGPU Target
296 -----------------------------
297
298 During this release ...
299
300 Changes to the AVR Target
301 -----------------------------
302
303 During this release ...
304
305266 Changes to the OCaml bindings
306267 -----------------------------
307268
308 * Remove ``add_bb_vectorize``.
269 * Removed ``add_bb_vectorize``.
309270
310271
311272 Changes to the C API
312273 --------------------
313274
314 * Remove ``LLVMAddBBVectorizePass``. The implementation was removed and the C
275 * Removed ``LLVMAddBBVectorizePass``. The implementation was removed and the C
315276 interface was made a deprecated no-op in LLVM 5. Use
316277 ``LLVMAddSLPVectorizePass`` instead to get the supported SLP vectorizer.
317278
318279 Changes to the DAG infrastructure
319280 ---------------------------------
320 * ADDC/ADDE/SUBC/SUBE are now deprecated and will default to expand. Backends
321 that wish to continue to use these opcodes should explicitely request so
281 * ``ADDC``/``ADDE``/``SUBC``/``SUBE`` are now deprecated and will default to expand. Backends
282 that wish to continue to use these opcodes should explicitely request to do so
322283 using ``setOperationAction`` in their ``TargetLowering``. New backends
323 should use UADDO/ADDCARRY/USUBO/SUBCARRY instead of the deprecated opcodes.
324
325 * The SETCCE opcode has now been removed in favor of SETCCCARRY.
326
327 * TableGen now supports multi-alternative pattern fragments via the PatFrags
328 class. PatFrag is now derived from PatFrags, which may require minor
329 changes to backends that directly access PatFrag members.
330
331 External Open Source Projects Using LLVM 7
332 ==========================================
333
334 * A project...
284 should use ``UADDO``/``ADDCARRY``/``USUBO``/``SUBCARRY`` instead of the deprecated opcodes.
285
286 * The ``SETCCE`` opcode has now been removed in favor of ``SETCCCARRY``.
287
288 * TableGen now supports multi-alternative pattern fragments via the ``PatFrags``
289 class. ``PatFrag`` is now derived from ``PatFrags``, which may require minor
290 changes to backends that directly access ``PatFrag`` members.
335291
336292
337293 Additional Information
338294 ======================
339295
340296 A wide variety of additional information is available on the `LLVM web page
341 `_, in particular in the `documentation
342 ://llvm.org/docs/>`_ section. The web page also contains versions of the
297 s://llvm.org/>`_, in particular in the `documentation
298 `_ section. The web page also contains versions of the
343299 API documentation which is up-to-date with the Subversion version of the source
344300 code. You can access versions of these documents specific to this release by
345301 going into the ``llvm/docs/`` directory in the LLVM tree.
346302
347303 If you have any questions or comments about LLVM, please feel free to contact
348 us via the `mailing lists ://llvm.org/docs/#maillist>`_.
304 us via the `mailing lists s://llvm.org/docs/#mailing-lists>`_.