NEWS.md 8.93 KB
Newer Older
yb9976's avatar
yb9976 committed
1
libFirm 1.22.1 (2016-01-07)
Matthias Braun's avatar
Matthias Braun committed
2
---------------------------
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
* make: Fix cmake/make build
* ia32: New just in time compilation mode which compiles into a memory buffer
* amd64: Support PIC with PLT for ELF
* ia32: Add `ia32-get_ip={pop,thunk}`
* ia32: Generate `mov $~1, %r; rol x, %r` for `~(1 << x)`
* ia32: Generate `mov $~0x80000000, %r; ror x, %r` for `~(0x80000000 >> x)`
* be: Stub support for the asm constraint modifier `%`
* amd64, ia32: Support the asm constraint `e`
* amd64, ia32: Support the asm modifier `c`
* ia32: Support JIT compilation
* be: Improve permutation moving for copy coalescing
* ir: Improve handling of negative overflow in float to int tarval conversion
* amd64: Improve matching of immediates during instruction selection
* amd64: Add peephole optimization `mov $0, %r` -> `xorl %r, %r`
* amd64: Add peephole optimization `lea c(%r), %r` -> `add $c, %r`
* amd64: Add peephole optimization `lea (%r1, %r2), %r1` -> `add %r2, %r1` and the commutated case
* ia32: Add peephole optimization `lea c(, %i, 2), %d` -> `lea c(%i, %i), %d`
* ia32: Add peephole optimization `lea (%b), %d` -> `mov %b, %d`
* ia32: Add peephole optimization `testl $0x0000XX00, %eRx` -> `testb $0xXX, %Rh`
* ia32: Generate slightly better code to load the floating-point constants `-0.0` and `-1.0`
* amd64: Reduce number of stack adjustments
* api: Set the length of an array type solely when creating the array type
* api: Set whether a function type is variadic and its calling convention and additional properties solely when creating the function type
* api: Automatically infer the mode when creating Add, And, Div, Eor, Minus, Mod, Mul, Mulh, Mux, Not, Or, Shl, Shr, Shrs and Sub nodes
* api: Remove the notion of atomic entities, use the initializer access functions instead
* api: Remove visibility from types
* api: Remove the type flag `tf_variable_size`, test the array size for 0 instead
* api: Remove `plist`, use `pdeq` instead
* api: Remove `get_{class,segment,struct,union}_{ident,name}()`, use `get_compound_{ident,name}()` instead
* ir: Improve IR graph verifier
* arm: Improve address mode use in instruction selection
* ir: Slightly improve preservation of debug info during transformations
* amd64: Improve use of `lea` in instruction selection
36
* sparc: Support the asm constaint `K`
37
* Bugfixes
Matthias Braun's avatar
Matthias Braun committed
38

Christoph Mallon's avatar
Christoph Mallon committed
39
libFirm 1.22.0 (2015-12-31)
40
41
---------------------------
* Improved PIC support, linux/elf is now supported
Christoph Mallon's avatar
Christoph Mallon committed
42
* Many additional local optimization rules (instcombine)
43
* Inline assembly support for sparc/amd64
yb9976's avatar
yb9976 committed
44
* Improved constant bit analysis
45
* Improved load/store optimization, featuring compound value optimizations
yb9976's avatar
yb9976 committed
46
* Improved reassociation pass
Christoph Mallon's avatar
Christoph Mallon committed
47
* Improved amd64 (aka x86_64) backend, it is now able to bootstrap libfirm
48
49
50
51
52
* Improved arm backend (but still experimental)
* Improved inliner (can inline compound types)
* Compiletime performance improvements
* Cleanups and API simplifications
* Switch to C99 and corresponding code cleanup and simplification
Christoph Mallon's avatar
Christoph Mallon committed
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
* More details in dumped IR graphs
* Streamlined IR
* Improved IR graph verifiers
* Further unify backend infrastructure
* Support for arrays as parameter and return types
* Updated documentation
* Perform precise 80bit spilling for x87 floating point
* Improved permutation decomposition for copy coalescing
* Added bitwise 'don't care' analysis and use it for optimizing occult constants
* Support builtin 'saturating increment' (amd64, ia32, sparc)
* Improved generated code for division by constant
* Support builtin 'compare and swap' (amd64, ia32, sparc)
* Support alias entities
* Improved scalar conversion optimization
* Added store combine optimization
* Overhauled representation and handling of infinite loops
* Improved use of flag results generated as side effect of instructions (ia32)
* Improved calculation of execution frequency estimation for large graphs
* Improved instruction selection (amd64, arm, ia32, sparc)
* Improved x87 floating point code generation
* Support early clobbers in inline assembly
* Support x87 constraints in inline assembly (ia32)
75
* Support `__attribute__((visibility("hidden")))` and `__attribute__((visible("protected")))`
Christoph Mallon's avatar
Christoph Mallon committed
76
* Optionally use CMake as build system
77
* Support `make install` with the usual variables (INSTALLPREFIX, ...)
78
79
* Bugfixes

Matthias Braun's avatar
Matthias Braun committed
80
81
82
83
libFirm 1.21.0 (2012-11-16)
---------------------------
* Improvements of x86 backend (esp. x87 floatingpoint code)
* Improvements to sparc backend (better use of delay slots)
yb9976's avatar
yb9976 committed
84
* Improved local optimization rules (esp. conversions)
Matthias Braun's avatar
Matthias Braun committed
85
86
87
* Make compiler more deterministic
* Bugfixes

88
89
libFirm 1.20.0 (2011-12-07)
---------------------------
90

91
92
93
94
95
96
97
98
99
* Further improvemens to sparc backend (SPEC2000 works with softfloat)
* Tuning of x86 backend
* Software floatingpoint lowerer
* Fixed firm profiling
* New pass management allowing to specify pre-/postconditions for passes
* Remove dependency on liblpp, add support for gurobi ILP solver
* Experimental dwarf debugging support
* Code cleanups, refactoring
* Restructured API documentation
yb9976's avatar
yb9976 committed
100
* Bugfixes (we did a lot of csmith testing)
101

102
103
104
105
106
libFirm 1.19.1 (2011-05-17)
---------------------------

* Fix some set_XXX functions not being exported in the shared library

Matthias Braun's avatar
Matthias Braun committed
107
libFirm 1.19.0 (2011-03-15)
108
109
---------------------------

Matthias Braun's avatar
Matthias Braun committed
110
111
112
113
114
115
* Includes "SSA-Based Register Allocation with PBQP"
* Improved Sparc backend
* New (optimistic) fixpoint based value-range propagation/bit analysis
* Code cleanup and refactoring
* Bugfixes

Matthias Braun's avatar
Matthias Braun committed
116
libFirm 1.18.1 (2010-05-05)
117
118
---------------------------

Matthias Braun's avatar
Matthias Braun committed
119
* Fix bug where stackframe was not always setup for -fno-omit-frame-pointer
Matthias Braun's avatar
Matthias Braun committed
120
* bugfixes in Asm handling
Matthias Braun's avatar
Matthias Braun committed
121

122
123
124
libFirm 1.18.0 (2010-04-15)
---------------------------

Matthias Braun's avatar
Matthias Braun committed
125
* Includes "Preference Guided Register Assignment" algorithm
Matthias Braun's avatar
Matthias Braun committed
126
* Experimental Value Range Propagation
127
* Loop Inversion and experimental Loop Unrolling code
Matthias Braun's avatar
Matthias Braun committed
128
129
130
131
132
133
* Simplified construction interface. Most node constructors don't need graph/block arguments anymore.
* Reworked type interface. Type names are optional now. Support for additional linkage types that among others support C++ 'linkonce' semantics now.
* Small changes in constructors and node getters/setters (mostly adding 'const' to some getters)
* code cleanup, smaller improvements in API specification
* bugfixes

134
135
136
libFirm 1.17.0 (2009-05-15)
---------------------------

Matthias Braun's avatar
Matthias Braun committed
137
* bugfixes
yb9976's avatar
yb9976 committed
138
* advanced load/store optimization which hoists loads out of loops
Matthias Braun's avatar
Matthias Braun committed
139
140
141
142
* Internal restruturing: Alot of node structures are automatically generated
   from a specification file now.
* Add support for multiple calling conventions
* New experimental support for reading and writing programgraphs to disk
yb9976's avatar
yb9976 committed
143
* Support and optimizations for trampolines
Matthias Braun's avatar
Matthias Braun committed
144
145
* fix PIC support

146
147
148
libFirm 1.16.0 (2009-01-28)
---------------------------

Christoph Mallon's avatar
Christoph Mallon committed
149
* bugfixes
150
* support for builtin nodes
Christoph Mallon's avatar
Christoph Mallon committed
151

152
153
libFirm 1.15.0 (2008-12-01)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
154
155
* bugfixes

156
157
158
libFirm 1.14.0 (2008-11-22)
---------------------------

yb9976's avatar
yb9976 committed
159
* Implementation of Clicks Combined Analysis/Optimizations
Matthias Braun's avatar
Matthias Braun committed
160
161
162
163
164
165
166
167
* New switch lowering code
* support for global asm statements
* improved asm support
* PIC support for Mac OS X
* New register pressure minimizing scheduler
* Improvements to spill algorithm
* fix endless loop problems
* further improve inlining heuristics
yb9976's avatar
yb9976 committed
168
* improve peephole optimizations for x86
Matthias Braun's avatar
Matthias Braun committed
169
170
* bugfixes

171
172
libFirm 1.13.0 (2008-07-31)
---------------------------
Michael Beck's avatar
Michael Beck committed
173

Michael Beck's avatar
updated    
Michael Beck committed
174
* VanDrunen's GVN-PRE fixed
Michael Beck's avatar
Michael Beck committed
175
* operator strength reduce fixed and improved
yb9976's avatar
typos    
yb9976 committed
176
* fixed 64bit code generation for some rare compare cases
Michael Beck's avatar
updated    
Michael Beck committed
177
* better tailrecursion optimization: handles x * func() and x + func()
yb9976's avatar
typos    
yb9976 committed
178
* improved inliner: better heuristics for inlining, can now inline recursive calls
Michael Beck's avatar
updated    
Michael Beck committed
179
* improved spiller
Michael Beck's avatar
updated    
Michael Beck committed
180
181
* lowering of CopyB nodes
* better memory disambiguator
Michael Beck's avatar
updated    
Michael Beck committed
182
* float->64bit conversion fixed for x87
183
* removed old verbosity level based debugging: all modules use the new debug facility
Michael Beck's avatar
updated    
Michael Beck committed
184
* Improved Confirm based optimization and conditional evaluation (using Confirm nodes)
Michael Beck's avatar
updated    
Michael Beck committed
185
* BugFixes: tail recursion, load/store optimization, lowering of structure return, conditional
Michael Beck's avatar
updated    
Michael Beck committed
186
  evaluation, removal of unused methods
Michael Beck's avatar
updated    
Michael Beck committed
187
* reduced numer of indirections for backend operation
Michael Beck's avatar
updated    
Michael Beck committed
188
* ia32 Backend: supports more CPU architectures
Michael Beck's avatar
updated    
Michael Beck committed
189
* ARM Backend: fixed frame access
Matthias Braun's avatar
Matthias Braun committed
190
* support for special segments (like constructors, destructors)
Michael Beck's avatar
Michael Beck committed
191

192
193
libFirm 1.12.1 (2008-02-18)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
194
195
196
197

* bugfixes for new style initializers with bitfield types
* make lowerer look at constant initializers too

198
199
libFirm 1.12.0 (2008-02-14)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
200

Matthias Braun's avatar
Matthias Braun committed
201
* dependency on libcore and libobstack dropped
Matthias Braun's avatar
Matthias Braun committed
202
203
204
205
* there's an alternative easier to use way to construct compound initializers
* bugfixes
* improved support for exceptions
* speed improvements
yb9976's avatar
yb9976 committed
206
* optimization of known libc functions
Matthias Braun's avatar
Matthias Braun committed
207

208
209
libFirm 1.11.0 (2008-11-05)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
210
211
212
213
214
215

* Lots of bugfixes
* Compilation speed improved
* Completely improved and rewritten handling of x86 address mode
* Optimized Mul -> Lea,Shift,Add transformation
* 64bit operations fixed and improved
yb9976's avatar
yb9976 committed
216
217
* More local optimizations
* New backend peephole optimizations
Matthias Braun's avatar
Matthias Braun committed
218
* Explicit status flag modeling (only for x86 for now)
yb9976's avatar
yb9976 committed
219
* Improvements of Load/Store optimization and alias analysis
220
* All C benchmarks from Spec CINT2000 work now (with our edg frontend)