NEWS.md 8.99 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
* sparc: Support computed goto
38
* arm: Support computed goto
39
* Bugfixes
Matthias Braun's avatar
Matthias Braun committed
40

Christoph Mallon's avatar
Christoph Mallon committed
41
libFirm 1.22.0 (2015-12-31)
42
43
---------------------------
* Improved PIC support, linux/elf is now supported
Christoph Mallon's avatar
Christoph Mallon committed
44
* Many additional local optimization rules (instcombine)
45
* Inline assembly support for sparc/amd64
yb9976's avatar
yb9976 committed
46
* Improved constant bit analysis
47
* Improved load/store optimization, featuring compound value optimizations
yb9976's avatar
yb9976 committed
48
* Improved reassociation pass
Christoph Mallon's avatar
Christoph Mallon committed
49
* Improved amd64 (aka x86_64) backend, it is now able to bootstrap libfirm
50
51
52
53
54
* 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
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
* 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)
77
* Support `__attribute__((visibility("hidden")))` and `__attribute__((visible("protected")))`
Christoph Mallon's avatar
Christoph Mallon committed
78
* Optionally use CMake as build system
79
* Support `make install` with the usual variables (INSTALLPREFIX, ...)
80
81
* Bugfixes

Matthias Braun's avatar
Matthias Braun committed
82
83
84
85
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
86
* Improved local optimization rules (esp. conversions)
Matthias Braun's avatar
Matthias Braun committed
87
88
89
* Make compiler more deterministic
* Bugfixes

90
91
libFirm 1.20.0 (2011-12-07)
---------------------------
92

93
94
95
96
97
98
99
100
101
* 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
102
* Bugfixes (we did a lot of csmith testing)
103

104
105
106
107
108
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
109
libFirm 1.19.0 (2011-03-15)
110
111
---------------------------

Matthias Braun's avatar
Matthias Braun committed
112
113
114
115
116
117
* 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
118
libFirm 1.18.1 (2010-05-05)
119
120
---------------------------

Matthias Braun's avatar
Matthias Braun committed
121
* Fix bug where stackframe was not always setup for -fno-omit-frame-pointer
Matthias Braun's avatar
Matthias Braun committed
122
* bugfixes in Asm handling
Matthias Braun's avatar
Matthias Braun committed
123

124
125
126
libFirm 1.18.0 (2010-04-15)
---------------------------

Matthias Braun's avatar
Matthias Braun committed
127
* Includes "Preference Guided Register Assignment" algorithm
Matthias Braun's avatar
Matthias Braun committed
128
* Experimental Value Range Propagation
129
* Loop Inversion and experimental Loop Unrolling code
Matthias Braun's avatar
Matthias Braun committed
130
131
132
133
134
135
* 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

136
137
138
libFirm 1.17.0 (2009-05-15)
---------------------------

Matthias Braun's avatar
Matthias Braun committed
139
* bugfixes
yb9976's avatar
yb9976 committed
140
* advanced load/store optimization which hoists loads out of loops
Matthias Braun's avatar
Matthias Braun committed
141
142
143
144
* 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
145
* Support and optimizations for trampolines
Matthias Braun's avatar
Matthias Braun committed
146
147
* fix PIC support

148
149
150
libFirm 1.16.0 (2009-01-28)
---------------------------

Christoph Mallon's avatar
Christoph Mallon committed
151
* bugfixes
152
* support for builtin nodes
Christoph Mallon's avatar
Christoph Mallon committed
153

154
155
libFirm 1.15.0 (2008-12-01)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
156
157
* bugfixes

158
159
160
libFirm 1.14.0 (2008-11-22)
---------------------------

yb9976's avatar
yb9976 committed
161
* Implementation of Clicks Combined Analysis/Optimizations
Matthias Braun's avatar
Matthias Braun committed
162
163
164
165
166
167
168
169
* 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
170
* improve peephole optimizations for x86
Matthias Braun's avatar
Matthias Braun committed
171
172
* bugfixes

173
174
libFirm 1.13.0 (2008-07-31)
---------------------------
Michael Beck's avatar
Michael Beck committed
175

Michael Beck's avatar
updated    
Michael Beck committed
176
* VanDrunen's GVN-PRE fixed
Michael Beck's avatar
Michael Beck committed
177
* operator strength reduce fixed and improved
yb9976's avatar
typos    
yb9976 committed
178
* fixed 64bit code generation for some rare compare cases
Michael Beck's avatar
updated    
Michael Beck committed
179
* better tailrecursion optimization: handles x * func() and x + func()
yb9976's avatar
typos    
yb9976 committed
180
* improved inliner: better heuristics for inlining, can now inline recursive calls
Michael Beck's avatar
updated    
Michael Beck committed
181
* improved spiller
Michael Beck's avatar
updated    
Michael Beck committed
182
183
* lowering of CopyB nodes
* better memory disambiguator
Michael Beck's avatar
updated    
Michael Beck committed
184
* float->64bit conversion fixed for x87
185
* removed old verbosity level based debugging: all modules use the new debug facility
Michael Beck's avatar
updated    
Michael Beck committed
186
* Improved Confirm based optimization and conditional evaluation (using Confirm nodes)
Michael Beck's avatar
updated    
Michael Beck committed
187
* BugFixes: tail recursion, load/store optimization, lowering of structure return, conditional
Michael Beck's avatar
updated    
Michael Beck committed
188
  evaluation, removal of unused methods
Michael Beck's avatar
updated    
Michael Beck committed
189
* reduced numer of indirections for backend operation
Michael Beck's avatar
updated    
Michael Beck committed
190
* ia32 Backend: supports more CPU architectures
Michael Beck's avatar
updated    
Michael Beck committed
191
* ARM Backend: fixed frame access
Matthias Braun's avatar
Matthias Braun committed
192
* support for special segments (like constructors, destructors)
Michael Beck's avatar
Michael Beck committed
193

194
195
libFirm 1.12.1 (2008-02-18)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
196
197
198
199

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

200
201
libFirm 1.12.0 (2008-02-14)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
202

Matthias Braun's avatar
Matthias Braun committed
203
* dependency on libcore and libobstack dropped
Matthias Braun's avatar
Matthias Braun committed
204
205
206
207
* there's an alternative easier to use way to construct compound initializers
* bugfixes
* improved support for exceptions
* speed improvements
yb9976's avatar
yb9976 committed
208
* optimization of known libc functions
Matthias Braun's avatar
Matthias Braun committed
209

210
211
libFirm 1.11.0 (2008-11-05)
---------------------------
Matthias Braun's avatar
Matthias Braun committed
212
213
214
215
216
217

* 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
218
219
* More local optimizations
* New backend peephole optimizations
Matthias Braun's avatar
Matthias Braun committed
220
* Explicit status flag modeling (only for x86 for now)
yb9976's avatar
yb9976 committed
221
* Improvements of Load/Store optimization and alias analysis
222
* All C benchmarks from Spec CINT2000 work now (with our edg frontend)