Home My Page Projects Code Snippets Project Openings SML/NJ
Summary Activity Forums Tracker Lists Tasks Docs Surveys News SCM Files

SCM Repository

[smlnj] Diff of /sml/trunk/HISTORY
ViewVC logotype

Diff of /sml/trunk/HISTORY

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1096, Tue Feb 26 16:59:02 2002 UTC revision 1145, Fri Mar 15 02:30:53 2002 UTC
# Line 14  Line 14 
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Matthias Blume  Name: Matthias Blume
17    Date: 2002/03/14 21:30:00 EST
18    Tag: blume-20020314-controls
19    Description:
20    
21    Controls:
22    ---------
23    
24    1. Factored out the recently-added Controls : CONTROLS stuff and put
25       it into its own library $/controls-lib.cm.  The source tree for
26       this is under src/smlnj-lib/Controls.
27    
28    2. Changed the names of types and functions in this interface, so they
29       make a bit more "sense":
30    
31          module -> registry
32          'a registry -> 'a group
33    
34    3. The interface now deals in ref cells only.  The getter/setter interface
35       is (mostly) gone.
36    
37    4. Added a function that lets one register an already-existing ref cell.
38    
39    5. Made the corresponding modifications to the rest of the code so that
40       everything compiles again.
41    
42    6. Changed the implementation of Controls.MLRISC back to something closer
43       to the original.  In particular, this module (and therefore MLRISC)
44       does not depend on Controls.  There now is some link-time code in
45       int-sys.sml that registers the MLRISC controls with the Controls
46       module.
47    
48    CM:
49    ---
50    
51      * One can now specify the lambda-split aggressiveness in init.cmi.
52    
53    ----------------------------------------------------------------------
54    Name: Allen Leung
55    Date: 2002/03/13 17:30:00 EST
56    Tag: leunga-20020313-x86-fp-unary
57    Description:
58    
59    Bug fix for:
60    
61    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
62    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
63    > - fun f(x,(y,z)) = Real.~ y;
64    > [autoloading]
65    > [autoloading done]
66    >       fchsl   (%eax), 184(%esp)
67    > Error: MLRisc bug: X86MCEmitter.emitInstr
68    >
69    > uncaught exception Error
70    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
71    
72    The problem was that the code generator did not generate any fp registers
73    in this case, and the ra didn't know that it needed to run the X86FP phase to
74    translate the pseudo fp instruction.   This only happened with unary fp
75    operators in certain situations.
76    
77    ----------------------------------------------------------------------
78    Name: Matthias Blume
79    Date: 2002/03/13 14:00:00 EST
80    Tag: blume-20020313-overload-etc
81    Description:
82    
83    1. Added _overload as a synonym for overload for backward compatibility.
84       (Control.overloadKW must be true for either version to be accepted.)
85    
86    2. Fixed bug in install script that caused more things to be installed
87       than what was requested in config/targets.
88    
89    3. Made CM aware of the (_)overload construct so that autoloading
90       works.
91    
92    ----------------------------------------------------------------------
93    Name: Matthias Blume
94    Date: 2002/03/12 22:03:00 EST
95    Tag: blume-20020312-url
96    Description:
97    
98    Forgot to update BOOT and srcarchiveurl.
99    
100    ----------------------------------------------------------------------
101    Name: Matthias Blume
102    Date: 2002/03/12 17:30:00 EST
103    Tag: blume-20020312-version110392
104    Description:
105    
106    Yet another version number bump (because of small changes to the
107    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
108    
109    Changes:
110    
111      The new pid generation scheme described a few weeks ago was overly
112      complicated.  I implemented a new mechanism that is simpler and
113      provides a bit more "stability":  Once CM has seen a compilation
114      unit, it keeps its identity constant (as long as you do not delete
115      those crucial CM/GUID/* files).  This means that when you change
116      an interface, compile, then go back to the old interface, and
117      compile again, you arrive at the original pid.
118    
119      There now also is a mechanism that instructs CM to use the plain
120      environment hash as a module's pid (effectively making its GUID
121      the empty string).  For this, "noguid" must be specified as an
122      option to the .sml file in question within its .cm file.
123      This is most useful for code that is being generated by tools such
124      as ml-nlffigen (because during development programmers tend to
125      erase the tool's entire output directory tree including CM's cached
126      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
127      revert to the old, broken behavior of SML/NJ, but in specific cases
128      where there is no danger of interface confusion, its use is ok
129      (I think).
130    
131      ml-nlffigen by default generates "noguid" annotations.  They can be
132      turned off by specifying -guid in its command line.
133    
134    ----------------------------------------------------------------------
135    Name: Lal George
136    Date: 2002/03/12 12 14:42:36 EST
137    Tag: george-20020312-frequency-computation
138    Description:
139    
140    Integrated jump chaining and static block frequency into the
141    compiler. More details and numbers later.
142    
143    ----------------------------------------------------------------------
144    Name: Lal George
145    Date: 2002/03/11 11 22:38:53 EST
146    Tag: george-20020311-jump-chain-elim
147    Description:
148    
149    Tested the jump chain elimination on all architectures (except the
150    hppa).  This is on by default right now and is profitable for the
151    alpha and x86, however, it may not be profitable for the sparc and ppc
152    when compiling the compiler.
153    
154    The gc test will typically jump to a label at the end of the cluster,
155    where there is another jump to an external cluster containing the actual
156    code to invoke gc. This is to allow factoring of common gc invocation
157    sequences. That is to say, we generate:
158    
159            f:
160               testgc
161               ja   L1      % jump if above to L1
162    
163            L1:
164               jmp L2
165    
166    
167    After jump chain elimination the 'ja L1' instructions is converted to
168    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
169    up being implemented in their long form (if L2 is far away) using:
170    
171            jbe     L3      % jump if below or equal to L3
172            jmp     L2
173         L3:
174            ...
175    
176    
177    For large compilation units L2  may be far away.
178    
179    
180    ----------------------------------------------------------------------
181    Name: Matthias Blume
182    Date: 2002/03/11 13:30:00 EST
183    Tag: blume-20020311-mltreeeval
184    Description:
185    
186    A functor parameter was missing.
187    
188    ----------------------------------------------------------------------
189    Name: Allen Leung
190    Date: 2002/03/11 10:30:00 EST
191    Tag: leunga-20020311-runtime-string0
192    Description:
193    
194       The representation of the empty string now points to a
195    legal null terminated C string instead of unit.  It is now possible
196    to convert an ML string into C string with InlineT.CharVector.getData.
197    This compiles into one single machine instruction.
198    
199    ----------------------------------------------------------------------
200    Name: Allen Leung
201    Date: 2002/03/10 23:55:00 EST
202    Tag: leunga-20020310-x86-call
203    Description:
204    
205       Added machine generation for CALL instruction (relative displacement mode)
206    
207    ----------------------------------------------------------------------
208    Name: Matthias Blume
209    Date: 2002/03/08 16:05:00
210    Tag: blume-20020308-entrypoints
211    Description:
212    
213    Version number bumped to 110.39.1.  NEW BOOTFILES!
214    
215    Entrypoints: non-zero offset into a code object where execution should begin.
216    
217    - Added the notion of an entrypoint to CodeObj.
218    - Added reading/writing of entrypoint info to Binfile.
219    - Made runtime system bootloader aware of entrypoints.
220    - Use the address of the label of the first function given to mlriscGen
221      as the entrypoint.  This address is currently always 0, but it will
222      not be 0 once we turn on block placement.
223    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
224      with entry points) from mlriscGen.
225    
226    ----------------------------------------------------------------------
227    Name: Allen Leung
228    Date: 2002/03/07 20:45:00 EST
229    Tag: leunga-20020307-x86-cmov
230    Description:
231    
232       Bug fixes for CMOVcc on x86.
233    
234       1. Added machine code generation for CMOVcc
235       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
236       3. CMOVcc cannot have an immediate operand as argument.
237    
238    ----------------------------------------------------------------------
239    Name: Matthias Blume
240    Date: 2002/03/07 16:15:00 EST
241    Tag: blume-20020307-controls
242    Description:
243    
244    This is a very large but mostly boring patch which makes (almost)
245    every tuneable compiler knob (i.e., pretty much everything under
246    Control.* plus a few other things) configurable via both the command
247    line and environment variables in the style CM did its configuration
248    until now.
249    
250    Try starting sml with '-h' (or, if you are brave, '-H')
251    
252    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
253    implements the underlying generic mechanism.
254    
255    The interface to some of the existing such facilities has changed somewhat.
256    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
257    (The getFoo interface is still there for backward-compatibility, but its
258    use is deprecated.)
259    
260    The ml-build script passes -Cxxx=yyy command-line arguments through so
261    that one can now twiddle the compiler settings when using this "batch"
262    compiler.
263    
264    TODO items:
265    
266    We should go through and throw out all controls that are no longer
267    connected to anything.  Moreover, we should go through and provide
268    meaningful (and correct!) documentation strings for those controls
269    that still are connected.
270    
271    Currently, multiple calls to Controls.new are accepted (only the first
272    has any effect).  Eventually we should make sure that every control
273    is being made (via Controls.new) exactly once.  Future access can then
274    be done using Controls.acc.
275    
276    Finally, it would probably be a good idea to use the getter-setter
277    interface to controls rather than ref cells.  For the time being, both
278    styles are provided by the Controls module, but getter-setter pairs are
279    better if thread-safety is of any concern because they can be wrapped.
280    
281    *****************************************
282    
283    One bug fix: The function blockPlacement in three of the MLRISC
284    backpatch files used to be hard-wired to one of two possibilities at
285    link time (according to the value of the placementFlag).  But (I
286    think) it should rather sense the flag every time.
287    
288    *****************************************
289    
290    Other assorted changes (by other people who did not supply a HISTORY entry):
291    
292    1. the cross-module inliner now works much better (Monnier)
293    2. representation of weights, frequencies, and probabilities in MLRISC
294       changed in preparation of using those for weighted block placement
295       (Reppy, George)
296    
297    ----------------------------------------------------------------------
298    Name: Lal George
299    Date: 2002/03/07 14:44:24 EST 2002
300    Tag: george-20020307-weighted-block-placement
301    
302    Tested the weighted block placement optimization on all architectures
303    (except the hppa) using AMPL to generate the block and edge frequencies.
304    Changes were required in the machine properties to correctly
305    categorize trap instructions. There is an MLRISC flag
306    "weighted-block-placement" that can be used to enable weighted block
307    placement, but this will be ineffective without block/edge
308    frequencies (coming soon).
309    
310    
311    ----------------------------------------------------------------------
312    Name: Lal George
313    Date: 2002/03/05 17:24:48 EST
314    Tag: george-20020305-linkage-cluster
315    
316    In order to support the block placement optimization, a new cluster
317    is generated as the very first cluster (called the linkage cluster).
318    It contains a single jump to the 'real' entry point for the compilation
319    unit. Block placement has no effect on the linkage cluster itself, but
320    all the other clusters  have full freedom in the manner in which they
321    reorder blocks or functions.
322    
323    On the x86 the typical linkage code that is generated is:
324       ----------------------
325            .align 2
326       L0:
327            addl    $L1-L0, 72(%esp)
328            jmp     L1
329    
330    
331            .align  2
332       L1:
333       ----------------------
334    
335    72(%esp) is the memory location for the stdlink register. This
336    must contain the address of the CPS function being called. In the
337    above example, it contains the address of  L0; before
338    calling L1 (the real entry point for the compilation unit), it
339    must contain the address for L1, and hence
340    
341            addl $L1-L0, 72(%esp)
342    
343    I have tested this on all architectures except the hppa.The increase
344    in code size is of course negligible
345    
346    ----------------------------------------------------------------------
347    Name: Allen Leung
348    Date: 2002/03/03 13:20:00 EST
349    Tag: leunga-20020303-mlrisc-tools
350    
351      Added #[ ... ] expressions to mlrisc tools
352    
353    ----------------------------------------------------------------------
354    Name: Matthias Blume
355    Date: 2002/02/27 12:29:00 EST
356    Tag: blume-20020227-cdebug
357    Description:
358    
359    - made types in structure C and C_Debug to be equal
360    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
361    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
362    
363    ----------------------------------------------------------------------
364    Name: Matthias Blume
365  Date: 2002/02/26 12:00:00 EST  Date: 2002/02/26 12:00:00 EST
366  Tag: blume-20020226-ffi  Tag: blume-20020226-ffi
367  Description:  Description:
# Line 124  Line 472 
472         TOTAL                                   2375.26u  57.21s  48.00g         TOTAL                                   2375.26u  57.21s  48.00g
473    
474  As you can see the old cps spill module suffers from some serious  As you can see the old cps spill module suffers from some serious
475  performance problem but since I cannot decipher the old code fully,  performance problem.  But since I cannot decipher the old code fully,
476  innstead of patching the problems up, I'm reimplementing it  instead of patching the problems up, I'm reimplementing it
477  with a different algorithm.  The new code is more modular,  with a different algorithm.  The new code is more modular,
478  smaller when compiled, and substantially faster  smaller when compiled, and substantially faster
479  (O(n log n) time and O(n) space).  Timing of the new spill module:  (O(n log n) time and O(n) space).  Timing of the new spill module:

Legend:
Removed from v.1096  
changed lines
  Added in v.1145

root@smlnj-gforge.cs.uchicago.edu
ViewVC Help
Powered by ViewVC 1.0.0