Commit Graph

96 Commits

Author SHA1 Message Date
Chris Lattner
fe54b2a495 allow partially materialized modules to be written out, which just strips out
the functions which haven't been read.

llvm-svn: 36999
2007-05-11 23:51:59 +00:00
Chris Lattner
062b15b7cc Make a preemptive bitcode format change to support PR1146. This lets us do
pr1146 in llvm 2.1 without ugly code to emulate old behavior.  This should
be merged into the 2.0 release branch.

llvm-svn: 36928
2007-05-08 05:38:01 +00:00
Chris Lattner
6056636de8 fix off-by-one that caused the llvm2cpp test to fail
llvm-svn: 36880
2007-05-06 19:27:46 +00:00
Chris Lattner
dc7a11940e Flush the file after writing bitcode so that clients who don't close their
ofstreams will be ok.

llvm-svn: 36878
2007-05-06 19:19:23 +00:00
Chris Lattner
af0a11cd65 Fix a nasty problem where we would miss enumeration of some types. This fixes
issues with CE_CAST etc.

llvm-svn: 36864
2007-05-06 08:35:19 +00:00
Chris Lattner
0c204417b2 Allow structs with zero fields.
llvm-svn: 36862
2007-05-06 08:21:50 +00:00
Chris Lattner
8b0bda0104 add a missing check
llvm-svn: 36859
2007-05-06 07:33:01 +00:00
Jeff Cohen
64156becaa Keep header file free of 'possible loss of data' warnings.
llvm-svn: 36832
2007-05-06 03:23:14 +00:00
Jeff Cohen
ce844b2aaa Unbreak VC++.
llvm-svn: 36831
2007-05-06 03:12:47 +00:00
Chris Lattner
ad8aadac37 add abbrevs for binops and casts. This shrinks a testcase from 725132->682500
bytes.

llvm-svn: 36829
2007-05-06 02:38:57 +00:00
Chris Lattner
bf23240f44 add a new CreateBitcodeWriterPass method, which creates a bitcode writer as
a pass

llvm-svn: 36828
2007-05-06 02:30:12 +00:00
Chris Lattner
7eb319ab65 implement reading/writing of inlineasm objects
llvm-svn: 36827
2007-05-06 01:58:20 +00:00
Chris Lattner
0505a9f632 add some abbrevs for ret and unreachable, shrinking kc++ from 2717360->2705388
llvm-svn: 36823
2007-05-06 01:28:01 +00:00
Chris Lattner
50460e7c48 enumerate the operands of a constant before we enumerate the constant itself
This avoids fwd references in the reader.

llvm-svn: 36822
2007-05-06 01:00:28 +00:00
Chris Lattner
6d0c5eb739 add a denser encoding for null terminated strings, add a 6-bit abbrev as
well.  This shrinks kc++ from 2724088 to 2717360 bytes.

llvm-svn: 36821
2007-05-06 00:53:07 +00:00
Chris Lattner
63002bb235 add an abbreviation for the string constants opzn, shrinking the constnats
block from:

  Block ID #11 (CONSTANTS_BLOCK):
      Num Instances: 1722
         Total Size: 3.85976e+06b/482470B/120617W
          % of file: 16.7609
       Average Size: 2241.44b/280.18B/70.045W
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 1/0.00058072
    Tot/Avg Records: 26423/15.3444
      % Abbrev Recs: 69.1746

to:

 Block ID #11 (CONSTANTS_BLOCK):
      Num Instances: 1724
         Total Size: 2.62406e+06b/328008B/82001.9W
          % of file: 12.041
       Average Size: 1522.08b/190.26B/47.5649W
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 2/0.00116009
    Tot/Avg Records: 26280/15.2436
      % Abbrev Recs: 68.9992

This shrinks kc++ from 2815788 to 2724088 bytes, which means the bitcode
file is now smaller than the bytecode file.

llvm-svn: 36820
2007-05-06 00:42:18 +00:00
Chris Lattner
34b256e1a6 implement the 'string constant' optimization. This shrinks kc.bit from
2878544 to 2815788

llvm-svn: 36818
2007-05-06 00:35:24 +00:00
Chris Lattner
9df45744b0 further reduce the redundancy of types in the instruction encoding. This
shrinks function bodies in kc++ from 891913B to 884073B

llvm-svn: 36817
2007-05-06 00:21:25 +00:00
Chris Lattner
32d91f39ca stop encoding type/value pairs when the type is implied by the value.
This shrinks the function block of kc++ from 1055K to 906K

llvm-svn: 36816
2007-05-06 00:00:00 +00:00
Chris Lattner
ed5887b739 add support for BLOCKINFO records at the module level. This fixes the reader
issues reid noticed last night.

llvm-svn: 36785
2007-05-05 18:57:30 +00:00
Chris Lattner
b865b049eb add an abbrev for loads. This shrinks the function block about 50K, from:
Block ID #12 (FUNCTION_BLOCK):
      Num Instances: 2344
         Total Size: 8.8434e+06b/1.10542e+06B/276356W
          % of file: 35.6726
       Average Size: 3772.78b/471.598B/117.899W
  Tot/Avg SubBlocks: 4065/1.73422
    Tot/Avg Abbrevs: 0/0
    Tot/Avg Records: 128487/54.8153
      % Abbrev Recs: 0

to:

 Block ID #12 (FUNCTION_BLOCK):
      Num Instances: 2344
         Total Size: 8.44518e+06b/1.05565e+06B/263912W
          % of file: 34.6203
       Average Size: 3602.89b/450.362B/112.59W
  Tot/Avg SubBlocks: 4065/1.73422
    Tot/Avg Abbrevs: 0/0
    Tot/Avg Records: 128487/54.8153
      % Abbrev Recs: 22.2077

llvm-svn: 36779
2007-05-05 07:44:49 +00:00
Chris Lattner
db60118ac2 add abbrevs for the constants tables. This shrinks it from 4.49755e6 bits
to 3.85972e6 bits in kc++

llvm-svn: 36778
2007-05-05 07:36:14 +00:00
Chris Lattner
e5194cc663 Add abbreviations to the TYPE_BLOCK for pointers, functions, structs and arrays.
This shrinks the type_block of kc++ from 139901 bits to 99389 bits (0.55% to 0.39%
of the file), a 40% reduction.

This shrink the record from:

  Block ID #10 (TYPE_BLOCK):
      Num Instances: 1
         Total Size: 139901b/17487.6B/4371.91W
          % of file: 0.549306
        Num Abbrevs: 0
        Num Records: 3203
      % Abbrev Recs: 0

to:

  Block ID #10 (TYPE_BLOCK):
      Num Instances: 1
         Total Size: 99389b/12423.6B/3105.91W
          % of file: 0.390862
        Num Abbrevs: 4
        Num Records: 3203
      % Abbrev Recs: 99.6566

With a common histogram of:

       Code Histogram:
                1613    POINTER
                1100    FUNCTION
                255     STRUCT
                224     ARRAY
                5       INTEGER
                2       OPAQUE
                1       LABEL
                1       DOUBLE
                1       VOID
                1       NUMENTRY

llvm-svn: 36776
2007-05-05 06:30:12 +00:00
Chris Lattner
967ab6fe6b add a char6 abbrev for bbnames and value names. This represents each character
with 6 bits where possible.  This shrinks kc++ from 3324164B to 3183584B.  The
old VST was:

  Block ID #14 (VALUE_SYMTAB):
         Total Size: 1.26713e+07b/1.58391e+06B/395978W
       Average Size: 5403.53b/675.442B/168.86W
          % of file: 47.6484

The new one is:

  Block ID #14 (VALUE_SYMTAB):
         Total Size: 1.15467e+07b/1.44334e+06B/360834W
       Average Size: 4923.96b/615.495B/153.874W
          % of file: 45.3368

This is 11% smaller than the VST in the bytecode format.

llvm-svn: 36771
2007-05-05 01:26:50 +00:00
Chris Lattner
277c22f0a9 add an abbreviation for the type symtab, this shrinks the TST from 175197 bits
to 103165 bits:

Old:
  Block ID #13 (TYPE_SYMTAB):
      Num Instances: 1
         Total Size: 175197b/21899.6B/5474.91W
       Average Size: 175197b/21899.6B/5474.91W
          % of file: 0.657023
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 0/0
    Tot/Avg Records: 255/255
      % Abbrev Recs: 0

New:
  Block ID #13 (TYPE_SYMTAB):
      Num Instances: 1
         Total Size: 103165b/12895.6B/3223.91W
       Average Size: 103165b/12895.6B/3223.91W
          % of file: 0.387937
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 1/1
    Tot/Avg Records: 255/255
      % Abbrev Recs: 100

llvm-svn: 36769
2007-05-05 00:47:19 +00:00
Chris Lattner
eca2bb1334 Implement support for globally associating abbrevs with block IDs, which
relieves us from having to emit the abbrevs into each instance of the block.
This shrinks kc.bit from 3368K to 3333K, but will be a more significant win
once instructions are abbreviated.

The VST went from:

  Block ID #14 (VALUE_SYMTAB):
      Num Instances: 2345
         Total Size: 1.29508e+07b/1.61885e+06B/404713W
       Average Size: 5522.73b/690.342B/172.585W
          % of file: 48.0645
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 7035/3
    Tot/Avg Records: 120924/51.5667
      % Abbrev Recs: 100

to:

  Block ID #14 (VALUE_SYMTAB):
      Num Instances: 2345
         Total Size: 1.26713e+07b/1.58391e+06B/395978W
       Average Size: 5403.53b/675.442B/168.86W
          % of file: 47.5198
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 0/0
    Tot/Avg Records: 120924/51.5667
      % Abbrev Recs: 100

because we didn't emit the same 3 abbrevs 2345 times :)

llvm-svn: 36767
2007-05-05 00:17:00 +00:00
Chris Lattner
28e448c9b5 allow the 8-bit abbrev to be used for either bb or other values
llvm-svn: 36761
2007-05-04 21:31:13 +00:00
Chris Lattner
b94c8f4ed8 shave another 150K off of kc++, by using a 7-bit encoding for BB names
where we can.  This shrinks kc++'s down to 3368K, with a VST record of:

  Block ID #14 (VALUE_SYMTAB):
      Num Instances: 2345
         Total Size: 1.29508e+07b/1.61885e+06B/404713W
       Average Size: 5522.73b/690.342B/172.585W
          % of file: 48.0645
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 7035/3
    Tot/Avg Records: 120924/51.5667
      % Abbrev Recs: 100

Isn't it nice to be able to optimizer bc size without touching the reader? :)

llvm-svn: 36759
2007-05-04 20:58:35 +00:00
Chris Lattner
2eaf2a2e40 where possible, encode symtab names with 7 bits per char instead of 8. This
shaves 110K off kc++ to 3514K.  Before:

  Block ID #14 (VALUE_SYMTAB):
      Num Instances: 2345
         Total Size: 1.50425e+07b/1.88031e+06B/470077W
       Average Size: 6414.69b/801.837B/200.459W
          % of file: 51.8057
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 2345/1
    Tot/Avg Records: 120924/51.5667
      % Abbrev Recs: 85.1791

after:

  Block ID #14 (VALUE_SYMTAB):
      Num Instances: 2345
         Total Size: 1.41229e+07b/1.76536e+06B/441341W
       Average Size: 6022.56b/752.82B/188.205W
          % of file: 50.2295
  Tot/Avg SubBlocks: 0/0
    Tot/Avg Abbrevs: 4690/2
    Tot/Avg Records: 120924/51.5667
      % Abbrev Recs: 85.1791

llvm-svn: 36758
2007-05-04 20:52:02 +00:00
Chris Lattner
94c8c289d0 Encode all value symtab strings as arrays of 8-bit fixed width integers,
instead of the default inefficient encoding.  This shrinks kc++ from 4134K
to 3629K

llvm-svn: 36755
2007-05-04 20:34:50 +00:00
Chris Lattner
52cbac8a1a eliminate internal length fields from record. Records already know their
total length

llvm-svn: 36752
2007-05-04 19:11:41 +00:00
Chris Lattner
f0e3a89fed stub out creation of BLOCKINFO_BLOCK
llvm-svn: 36749
2007-05-04 18:26:27 +00:00
Chris Lattner
ddc5d5ce50 optimize constant layout. This fixes encoding of 181.mcf (by ensuring
integer structure idx's are emitted before constant expr geps) and shrinks
files slightly.  For example kc++ shrinks from 4326188 to 4240128 bytes.

llvm-svn: 36742
2007-05-04 05:21:47 +00:00
Chris Lattner
f59434ceca simple optimization for the type table
llvm-svn: 36741
2007-05-04 05:05:48 +00:00
Chris Lattner
f403c9a42c storeinst ctor takes isvolatile before alignment. With this, 176.gcc roundtrips
llvm-svn: 36738
2007-05-04 03:57:30 +00:00
Chris Lattner
c0f186afab fix encoding of invoke instructions. kc++ now round trips
llvm-svn: 36737
2007-05-04 03:52:24 +00:00
Chris Lattner
0fa3401f43 fix a misplaced error
llvm-svn: 36736
2007-05-04 03:50:29 +00:00
Chris Lattner
2f3c85827e encode and read param attrs along with function type. WE can now roundtrip Olden/voronoi loslessly
llvm-svn: 36735
2007-05-04 03:41:34 +00:00
Chris Lattner
ccb3d450ba add support for reading the param attrs block
llvm-svn: 36731
2007-05-04 03:30:17 +00:00
Chris Lattner
2eeffc8c5a fix a bug encoding param attrs
llvm-svn: 36730
2007-05-04 03:14:09 +00:00
Chris Lattner
36f37e3f99 remove dead code
llvm-svn: 36728
2007-05-04 03:02:54 +00:00
Chris Lattner
990f7492d6 add writer support for param attrs
llvm-svn: 36724
2007-05-04 02:59:04 +00:00
Reid Spencer
8148449b40 Allow this to compile with gcc 4.0.X
llvm-svn: 36723
2007-05-04 01:43:33 +00:00
Chris Lattner
13e18fdf09 pass by reference, not by value
llvm-svn: 36721
2007-05-04 00:45:24 +00:00
Chris Lattner
54a0a58b82 trivial scaffolding for param attrs
llvm-svn: 36720
2007-05-04 00:44:52 +00:00
Chris Lattner
21928b0039 enumerate parameter attr lists.
llvm-svn: 36709
2007-05-03 22:46:43 +00:00
Chris Lattner
76e0063529 Encoding calling conv info in call/invoke instrs, tree add now round trips completely
llvm-svn: 36707
2007-05-03 22:34:03 +00:00
Chris Lattner
96aa97284a the type field for a store is the type of the pointer, not the value.
With this fix I can round trip treeaadd, only losing calling conv info.

llvm-svn: 36706
2007-05-03 22:21:59 +00:00
Chris Lattner
83281f01ab don't bother encoding # varargs
llvm-svn: 36705
2007-05-03 22:19:34 +00:00
Chris Lattner
888ca49f41 fix encoding of BB names in the symtab
llvm-svn: 36704
2007-05-03 22:18:21 +00:00