[pypy-commit] pypy vecopt: added integer/float types to zjit test (int 8, 16, 32, 64, float 32, 64)

2015-05-12 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77298:40afd88ea5d8 Date: 2015-05-12 16:21 +0200 http://bitbucket.org/pypy/pypy/changeset/40afd88ea5d8/ Log:added integer/float types to zjit test (int 8,16,32,64, float 32,64) extended test_add to use different types

[pypy-commit] pypy vecopt: vector add now passing for double floating points

2015-05-12 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77297:786c0adbf389 Date: 2015-05-12 09:51 +0200 http://bitbucket.org/pypy/pypy/changeset/786c0adbf389/ Log:vector add now passing for double floating points diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy vecopt: typed expand to float/int

2015-05-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77340:f82a4d3d1831 Date: 2015-05-15 18:40 +0200 http://bitbucket.org/pypy/pypy/changeset/f82a4d3d1831/ Log:typed expand to float/int added extract/insert integer x86 ops added missing add/sub/mul/div float ops new

[pypy-commit] pypy vecopt: split box_(un)pack into float_(un)pack and int_(un)pack, adjusted some tests that unrolled to often (scheduler splits packed instructions if size is too big)

2015-05-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77339:2861d6325f12 Date: 2015-05-15 11:07 +0200 http://bitbucket.org/pypy/pypy/changeset/2861d6325f12/ Log:split box_(un)pack into float_(un)pack and int_(un)pack, adjusted some tests that unrolled to often (scheduler

[pypy-commit] pypy vecopt: finished the unpacking for float32/64 and int32/64. added x86 packed mul operations (don't know if we ever can use them for int64)

2015-05-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77392:199b27a762f8 Date: 2015-05-19 16:14 +0200 http://bitbucket.org/pypy/pypy/changeset/199b27a762f8/ Log:finished the unpacking for float32/64 and int32/64. added x86 packed mul operations (don't know if we ever can

[pypy-commit] pypy vecopt: removed annotation errors that occur when building the jit

2015-05-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77393:18c3705b8f89 Date: 2015-05-19 16:59 +0200 http://bitbucket.org/pypy/pypy/changeset/18c3705b8f89/ Log:removed annotation errors that occur when building the jit removed commented (old) code from assembler do not

[pypy-commit] pypy vecopt: removed a bug where packtype was modified but not copied before that

2015-05-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77390:0e3498ee6eb4 Date: 2015-05-19 10:11 +0200 http://bitbucket.org/pypy/pypy/changeset/0e3498ee6eb4/ Log:removed a bug where packtype was modified but not copied before that diff --git a/pypy/module/micronumpy/compile.py

[pypy-commit] pypy vecopt: cvtpd2dq packs ints to the lower quadword. that is why it did not work, starting to rethink the conversion function

2015-05-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77391:43ebe5044bd1 Date: 2015-05-19 11:33 +0200 http://bitbucket.org/pypy/pypy/changeset/43ebe5044bd1/ Log:cvtpd2dq packs ints to the lower quadword. that is why it did not work, starting to rethink the conversion

[pypy-commit] pypy vecopt: excluded getfield gc/raw from the load tracking. there is no index on this operation. index is in the descriptor

2015-05-20 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77427:1efff313cfd7 Date: 2015-05-20 08:51 +0200 http://bitbucket.org/pypy/pypy/changeset/1efff313cfd7/ Log:excluded getfield gc/raw from the load tracking. there is no index on this operation. index is in the descriptor

[pypy-commit] pypy vecopt: helping the rtyper

2015-05-20 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77426:20eb63b24b80 Date: 2015-05-20 08:05 +0200 http://bitbucket.org/pypy/pypy/changeset/20eb63b24b80/ Log:helping the rtyper diff --git a/rpython/jit/backend/llsupport/descr.py b/rpython/jit/backend/llsupport/descr.py ---

[pypy-commit] pypy vecopt: added debug output in the vecopt method and log it in the logfile

2015-05-20 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77429:bc8e3589d68a Date: 2015-05-20 14:39 +0200 http://bitbucket.org/pypy/pypy/changeset/bc8e3589d68a/ Log:added debug output in the vecopt method and log it in the logfile diff --git a/rpython/jit/metainterp/jitprof.py

[pypy-commit] pypy vecopt: turned off vectorize opt for all jit drivers and enabled vectorize opt in micronumpy loop jit drivers

2015-05-20 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77428:46704e37a322 Date: 2015-05-20 10:39 +0200 http://bitbucket.org/pypy/pypy/changeset/46704e37a322/ Log:turned off vectorize opt for all jit drivers and enabled vectorize opt in micronumpy loop jit drivers resolved a

[pypy-commit] pypy vecopt: problem with float abs (expanded variable but did not remove second use), andps constant pointed to neg constant

2015-06-05 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77876:062f2c00c3f0 Date: 2015-06-05 08:56 +0200 http://bitbucket.org/pypy/pypy/changeset/062f2c00c3f0/ Log:problem with float abs (expanded variable but did not remove second use), andps constant pointed to neg constant

[pypy-commit] pypy vecopt: in the case #pack 2 the accum is not split

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77986:f7b571afa033 Date: 2015-06-09 11:53 +0200 http://bitbucket.org/pypy/pypy/changeset/f7b571afa033/ Log:in the case #pack 2 the accum is not split diff --git a/rpython/jit/metainterp/optimizeopt/dependency.py

[pypy-commit] pypy vecopt: accumulation box is now used instead of the scalar box, need to adapt guard exit to produce correct result

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77975:74976aacf58b Date: 2015-06-09 10:01 +0200 http://bitbucket.org/pypy/pypy/changeset/74976aacf58b/ Log:accumulation box is now used instead of the scalar box, need to adapt guard exit to produce correct result diff

[pypy-commit] pypy vecopt: need to call method to get the result :)

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77978:7aba38455439 Date: 2015-06-09 10:08 +0200 http://bitbucket.org/pypy/pypy/changeset/7aba38455439/ Log:need to call method to get the result :) diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: updated tests to provide arguments for some changed methods. float vector box signed is always false

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77980:716facf22bdc Date: 2015-06-09 10:54 +0200 http://bitbucket.org/pypy/pypy/changeset/716facf22bdc/ Log:updated tests to provide arguments for some changed methods. float vector box signed is always false diff --git

[pypy-commit] pypy vecopt: added test to ensure sum is not prohibited by cost calculation

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77979:450cfd82b800 Date: 2015-06-09 10:17 +0200 http://bitbucket.org/pypy/pypy/changeset/450cfd82b800/ Log:added test to ensure sum is not prohibited by cost calculation diff --git

[pypy-commit] pypy vecopt: generating vector box for accumulation before the label and renaming occurances

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77974:66758cffe3af Date: 2015-06-09 09:34 +0200 http://bitbucket.org/pypy/pypy/changeset/66758cffe3af/ Log:generating vector box for accumulation before the label and renaming occurances diff --git

[pypy-commit] pypy vecopt: scheduler emits packs that are interdependent (only within pack and marked accum)

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77970:c853e8ea2f01 Date: 2015-06-09 08:34 +0200 http://bitbucket.org/pypy/pypy/changeset/c853e8ea2f01/ Log:scheduler emits packs that are interdependent (only within pack and marked accum) diff --git

[pypy-commit] pypy vecopt: extract instructions sse4 had some wrong parameters in assembler

2015-06-05 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77921:3c733c6463df Date: 2015-06-05 19:40 +0200 http://bitbucket.org/pypy/pypy/changeset/3c733c6463df/ Log:extract instructions sse4 had some wrong parameters in assembler added test case for that diff --git

[pypy-commit] pypy vecopt: int expansion for int16 and int8 added, int32/16 test added first already passes

2015-06-05 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77902:eb3cc9cf75f4 Date: 2015-06-05 12:59 +0200 http://bitbucket.org/pypy/pypy/changeset/eb3cc9cf75f4/ Log:int expansion for int16 and int8 added, int32/16 test added first already passes diff --git

[pypy-commit] pypy vecopt: updating the input type. when items are packed their count changes, thus the input and output type must be updated

2015-06-05 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77903:9f502085659b Date: 2015-06-05 13:14 +0200 http://bitbucket.org/pypy/pypy/changeset/9f502085659b/ Log:updating the input type. when items are packed their count changes, thus the input and output type must be

[pypy-commit] pypy vecopt: tested insert x86 opcodes

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77946:815c8b4a7c5b Date: 2015-06-08 08:39 +0200 http://bitbucket.org/pypy/pypy/changeset/815c8b4a7c5b/ Log:tested insert x86 opcodes diff --git a/rpython/jit/backend/x86/rx86.py b/rpython/jit/backend/x86/rx86.py ---

[pypy-commit] pypy vecopt: removed manual test since it is tested in auto test now

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77949:1fc0d9cd2612 Date: 2015-06-08 10:28 +0200 http://bitbucket.org/pypy/pypy/changeset/1fc0d9cd2612/ Log:removed manual test since it is tested in auto test now fixed some other tests in the x86 backend diff --git

[pypy-commit] pypy vecopt: removed allocation of comparison objects to track guard comparisons. This is handled by the guard strengthing operation

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77957:3bc0e1fb8db8 Date: 2015-06-08 15:12 +0200 http://bitbucket.org/pypy/pypy/changeset/3bc0e1fb8db8/ Log:removed allocation of comparison objects to track guard comparisons. This is handled by the guard strengthing

[pypy-commit] pypy vecopt: moved out guard strengthening (on arith level) and scheduling from vectorize.py and schedule.py

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77958:cec809035d0c Date: 2015-06-08 15:25 +0200 http://bitbucket.org/pypy/pypy/changeset/cec809035d0c/ Log:moved out guard strengthening (on arith level) and scheduling from vectorize.py and schedule.py diff --git

[pypy-commit] pypy vecopt: can_be_packed did not consider the case if origin pack is None, some small other refactorings

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77960:5d3c3f5b1a5a Date: 2015-06-08 15:56 +0200 http://bitbucket.org/pypy/pypy/changeset/5d3c3f5b1a5a/ Log:can_be_packed did not consider the case if origin pack is None, some small other refactorings diff --git

[pypy-commit] pypy vecopt: finished refactor the structure

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77959:fab36fa4cf6d Date: 2015-06-08 15:39 +0200 http://bitbucket.org/pypy/pypy/changeset/fab36fa4cf6d/ Log:finished refactor the structure diff --git a/rpython/jit/metainterp/optimizeopt/guard.py

[pypy-commit] pypy vecopt: started to find reduce/accumulation functions that are vectorizable (e.g. sum)

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77954:90b97695ef67 Date: 2015-06-08 14:15 +0200 http://bitbucket.org/pypy/pypy/changeset/90b97695ef67/ Log:started to find reduce/accumulation functions that are vectorizable (e.g. sum) diff --git

[pypy-commit] pypy vecopt: reverted the 8 immediate suffix (solved differently for tests)

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77953:6e0e98c3d70a Date: 2015-06-08 10:50 +0200 http://bitbucket.org/pypy/pypy/changeset/6e0e98c3d70a/ Log:reverted the 8 immediate suffix (solved differently for tests) diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: finding reduceables works, scheduling needs to be adapted next

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77956:ac80f41576c2 Date: 2015-06-08 15:03 +0200 http://bitbucket.org/pypy/pypy/changeset/ac80f41576c2/ Log:finding reduceables works, scheduling needs to be adapted next diff --git

[pypy-commit] pypy vecopt: added accumulation pair

2015-06-08 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77955:a3b3e446b864 Date: 2015-06-08 14:38 +0200 http://bitbucket.org/pypy/pypy/changeset/a3b3e446b864/ Log:added accumulation pair diff --git a/rpython/jit/metainterp/optimizeopt/test/test_vectorize.py

[pypy-commit] pypy vecopt: remember expansion (reduces register pressure if constant/variable is use more often). but not for heterogeneous expanded vectors, I don't think this happens frequently

2015-06-03 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77799:95db7332c363 Date: 2015-06-03 08:53 +0200 http://bitbucket.org/pypy/pypy/changeset/95db7332c363/ Log:remember expansion (reduces register pressure if constant/variable is use more often). but not for heterogeneous

[pypy-commit] pypy vmprof-address: sorting the code ranges again

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vmprof-address Changeset: r78009:795973683fee Date: 2015-06-10 12:06 +0200 http://bitbucket.org/pypy/pypy/changeset/795973683fee/ Log:sorting the code ranges again diff --git a/rpython/tool/jitlogparser/parser.py

[pypy-commit] pypy vecopt: loading all code regions instead of the merged ones from World

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78008:09753b7b85af Date: 2015-06-10 12:05 +0200 http://bitbucket.org/pypy/pypy/changeset/09753b7b85af/ Log:loading all code regions instead of the merged ones from World diff --git a/rpython/tool/jitlogparser/parser.py

[pypy-commit] pypy vmprof-address: manually loading address from log to get all addresses and be able to display all assembler dumps

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vmprof-address Changeset: r78007:c1dc41980f3b Date: 2015-06-10 12:01 +0200 http://bitbucket.org/pypy/pypy/changeset/c1dc41980f3b/ Log:manually loading address from log to get all addresses and be able to display all assembler dumps diff

[pypy-commit] pypy vecopt: added a new field to the resume guard descr to handle accumulation variables at guard exit

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78012:6497803fbba6 Date: 2015-06-10 14:10 +0200 http://bitbucket.org/pypy/pypy/changeset/6497803fbba6/ Log:added a new field to the resume guard descr to handle accumulation variables at guard exit implemented the

[pypy-commit] pypy vmprof-address: changes to view the bootstrap address of vmprof

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vmprof-address Changeset: r78006:2ff1050d072a Date: 2015-06-10 11:41 +0200 http://bitbucket.org/pypy/pypy/changeset/2ff1050d072a/ Log:changes to view the bootstrap address of vmprof diff --git a/rpython/jit/backend/tool/viewcode.py

[pypy-commit] pypy vecopt: vector boxes need to be considered on guard exit

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78005:37e903d27286 Date: 2015-06-10 11:39 +0200 http://bitbucket.org/pypy/pypy/changeset/37e903d27286/ Log:vector boxes need to be considered on guard exit diff --git a/rpython/jit/backend/llsupport/assembler.py

[pypy-commit] jitviewer vmprof-address: adding vmprof address to the right side of a loop

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vmprof-address Changeset: r272:0d1b70823758 Date: 2015-06-10 10:50 +0200 http://bitbucket.org/pypy/jitviewer/changeset/0d1b70823758/ Log:adding vmprof address to the right side of a loop extended the search to be able to lookup hex

[pypy-commit] jitviewer vmprof-address: merged defaut

2015-06-10 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vmprof-address Changeset: r273:294cee8b9fda Date: 2015-06-10 10:51 +0200 http://bitbucket.org/pypy/jitviewer/changeset/294cee8b9fda/ Log:merged defaut diff --git a/.hgtags b/.hgtags --- a/.hgtags +++ b/.hgtags @@ -4,3 +4,4 @@

[pypy-commit] pypy vecopt: silenced rpython complaints

2015-06-04 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77873:e5d7788d191b Date: 2015-06-04 18:47 +0200 http://bitbucket.org/pypy/pypy/changeset/e5d7788d191b/ Log:silenced rpython complaints diff --git a/rpython/jit/backend/x86/assembler.py b/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy vecopt: adapted test_zjit to changes in micro numpy

2015-06-04 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77865:a0cf34db417d Date: 2015-06-04 18:10 +0200 http://bitbucket.org/pypy/pypy/changeset/a0cf34db417d/ Log:adapted test_zjit to changes in micro numpy added a stub for int expansion diff --git

[pypy-commit] pypy vecopt: merged release-2.6.x

2015-06-04 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77863:d3969a3c7bb3 Date: 2015-06-04 14:43 +0200 http://bitbucket.org/pypy/pypy/changeset/d3969a3c7bb3/ Log:merged release-2.6.x diff too long, truncating to 2000 out of 4 lines diff --git a/.hgtags b/.hgtags --- a/.hgtags

[pypy-commit] pypy vecopt: added missing vec counters to jitprof test

2015-06-04 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77864:b41e4cb3e026 Date: 2015-06-04 14:53 +0200 http://bitbucket.org/pypy/pypy/changeset/b41e4cb3e026/ Log:added missing vec counters to jitprof test diff --git a/rpython/jit/backend/x86/regalloc.py

[pypy-commit] pypy vecopt: int expand 64 bit working

2015-06-04 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77868:812b3caa438c Date: 2015-06-04 18:37 +0200 http://bitbucket.org/pypy/pypy/changeset/812b3caa438c/ Log:int expand 64 bit working diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: preventing accumulator flush within the trace. the plan is only to support in in guard exits

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77991:be2247303111 Date: 2015-06-09 14:18 +0200 http://bitbucket.org/pypy/pypy/changeset/be2247303111/ Log:preventing accumulator flush within the trace. the plan is only to support in in guard exits diff --git

[pypy-commit] pypy vecopt: works now for the int16 case

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77990:ec09c339cc68 Date: 2015-06-09 13:55 +0200 http://bitbucket.org/pypy/pypy/changeset/ec09c339cc68/ Log:works now for the int16 case diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: moving the type determination from the scheduling transformation to the find pack routines,

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77989:2982e1a68dcf Date: 2015-06-09 13:36 +0200 http://bitbucket.org/pypy/pypy/changeset/2982e1a68dcf/ Log:moving the type determination from the scheduling transformation to the find pack routines, this helps to better

[pypy-commit] pypy vecopt: preventing the transformation of cumsum (some problems that followed from last changes)

2015-06-09 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77993:652fca81f3d1 Date: 2015-06-09 15:41 +0200 http://bitbucket.org/pypy/pypy/changeset/652fca81f3d1/ Log:preventing the transformation of cumsum (some problems that followed from last changes) diff --git

[pypy-commit] pypy vecopt: started to bail out for some operations that cannot be implemented efficiently in SSE4.1

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78266:c9e6e55b3900 Date: 2015-06-23 16:48 +0200 http://bitbucket.org/pypy/pypy/changeset/c9e6e55b3900/ Log:started to bail out for some operations that cannot be implemented efficiently in SSE4.1 diff --git

[pypy-commit] pypy vecopt: moved around the timing to be able to translate the project with it

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78260:5af70e37351c Date: 2015-06-23 14:31 +0200 http://bitbucket.org/pypy/pypy/changeset/5af70e37351c/ Log:moved around the timing to be able to translate the project with it diff --git a/rpython/jit/metainterp/compile.py

[pypy-commit] pypy vecopt: some changes after being tested with test_zjit

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78262:fefc4fafe4f6 Date: 2015-06-23 14:47 +0200 http://bitbucket.org/pypy/pypy/changeset/fefc4fafe4f6/ Log:some changes after being tested with test_zjit diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: added test case for conversion int64 - int8

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78265:77645ab90fd3 Date: 2015-06-23 15:34 +0200 http://bitbucket.org/pypy/pypy/changeset/77645ab90fd3/ Log:added test case for conversion int64 - int8 diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: missing format character

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78263:58b4e8859909 Date: 2015-06-23 14:52 +0200 http://bitbucket.org/pypy/pypy/changeset/58b4e8859909/ Log:missing format character diff --git a/rpython/jit/metainterp/warmspot.py b/rpython/jit/metainterp/warmspot.py ---

[pypy-commit] pypy vecopt: missing variable definition

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78274:7e3a332fec89 Date: 2015-06-23 19:30 +0200 http://bitbucket.org/pypy/pypy/changeset/7e3a332fec89/ Log:missing variable definition diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: moved timing functions into the execute assembler, commented out the guard as vec opt for this translation test

2015-06-25 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78307:f5d62531d777 Date: 2015-06-25 12:10 +0200 http://bitbucket.org/pypy/pypy/changeset/f5d62531d777/ Log:moved timing functions into the execute assembler, commented out the guard as vec opt for this translation test

[pypy-commit] pypy vecopt: don't follow wrong dependency chains, excluded fail args

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78318:8afb499c0584 Date: 2015-06-26 09:56 +0200 http://bitbucket.org/pypy/pypy/changeset/8afb499c0584/ Log:don't follow wrong dependency chains, excluded fail args only store is not allowed to compute operations if the

[pypy-commit] pypy vecopt: setting to safe default

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78279:1bd1ae3b3117 Date: 2015-06-24 08:57 +0200 http://bitbucket.org/pypy/pypy/changeset/1bd1ae3b3117/ Log:setting to safe default diff --git a/rpython/jit/metainterp/warmspot.py b/rpython/jit/metainterp/warmspot.py ---

[pypy-commit] pypy vecopt: reformated the log output to time the vecopt traces

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78289:c061d25dca84 Date: 2015-06-24 12:08 +0200 http://bitbucket.org/pypy/pypy/changeset/c061d25dca84/ Log:reformated the log output to time the vecopt traces diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: added comment in doc

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78288:16da7f0d90bd Date: 2015-06-24 12:04 +0200 http://bitbucket.org/pypy/pypy/changeset/16da7f0d90bd/ Log:added comment in doc diff --git a/rpython/doc/jit/vectorization.rst b/rpython/doc/jit/vectorization.rst ---

[pypy-commit] pypy vecopt: well, should be int_mul not int_add...

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78290:2d5ff3f421e1 Date: 2015-06-24 12:17 +0200 http://bitbucket.org/pypy/pypy/changeset/2d5ff3f421e1/ Log:well, should be int_mul not int_add... diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: added a new resop class for casting operations, added a test to ensure they are created correctly

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78286:e812d5febce9 Date: 2015-06-24 10:36 +0200 http://bitbucket.org/pypy/pypy/changeset/e812d5febce9/ Log:added a new resop class for casting operations, added a test to ensure they are created correctly diff --git

[pypy-commit] pypy vecopt: preventing int signext from 32 - 32

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78287:3b569b13ba22 Date: 2015-06-24 12:00 +0200 http://bitbucket.org/pypy/pypy/changeset/3b569b13ba22/ Log:preventing int signext from 32 - 32 preventing packed int mul for 64 bit cannot be done with an sse opcode (see

[pypy-commit] pypy vecopt: doc additions, reenabled the int8 expand test (passes now)

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78292:e70ae41089d7 Date: 2015-06-24 13:58 +0200 http://bitbucket.org/pypy/pypy/changeset/e70ae41089d7/ Log:doc additions, reenabled the int8 expand test (passes now) diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: modified tests for missing packed int64 mul (was incorrect before) and prevented int64-int16 conversion

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78291:e8681afe010b Date: 2015-06-24 13:39 +0200 http://bitbucket.org/pypy/pypy/changeset/e8681afe010b/ Log:modified tests for missing packed int64 mul (was incorrect before) and prevented int64-int16 conversion diff

[pypy-commit] pypy vecopt: logging condition changed for timing in optimize_trace

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78294:26d736866fa2 Date: 2015-06-24 15:07 +0200 http://bitbucket.org/pypy/pypy/changeset/26d736866fa2/ Log:logging condition changed for timing in optimize_trace diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: rpython translation issues

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78295:5008c5321939 Date: 2015-06-24 15:19 +0200 http://bitbucket.org/pypy/pypy/changeset/5008c5321939/ Log:rpython translation issues diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: adding guards as vector instructions. i'm not yet sure how this will work out, but could help to generate better loops for reductions

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78299:269e30fb6042 Date: 2015-06-24 17:31 +0200 http://bitbucket.org/pypy/pypy/changeset/269e30fb6042/ Log:adding guards as vector instructions. i'm not yet sure how this will work out, but could help to generate better

[pypy-commit] pypy vecopt: add a new doc file to describe vecopt, describe what the current opt. is capable of and added some limitations

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78250:77f58b1dcf7f Date: 2015-06-23 09:33 +0200 http://bitbucket.org/pypy/pypy/changeset/77f58b1dcf7f/ Log:add a new doc file to describe vecopt, describe what the current opt. is capable of and added some limitations

[pypy-commit] pypy vecopt: rpython issue

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78172:f5454aafa6d6 Date: 2015-06-18 11:51 +0200 http://bitbucket.org/pypy/pypy/changeset/f5454aafa6d6/ Log:rpython issue diff --git a/rpython/jit/backend/x86/assembler.py b/rpython/jit/backend/x86/assembler.py ---

[pypy-commit] pypy regalloc: added logic to put every variable into the register depending on how far away the next usage is (less distance is more likely to end up in a register)

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: regalloc Changeset: r78174:be630ecb8639 Date: 2015-06-18 13:04 +0200 http://bitbucket.org/pypy/pypy/changeset/be630ecb8639/ Log:added logic to put every variable into the register depending on how far away the next usage is (less distance

[pypy-commit] pypy regalloc: storing the variable use/def information in the tuple of longevity as third parameter

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: regalloc Changeset: r78164:73506cd059de Date: 2015-06-18 09:58 +0200 http://bitbucket.org/pypy/pypy/changeset/73506cd059de/ Log:storing the variable use/def information in the tuple of longevity as third parameter diff --git

[pypy-commit] pypy regalloc: added more information to the output of regalloc.py

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: regalloc Changeset: r78242:ce341006d844 Date: 2015-06-22 12:08 +0200 http://bitbucket.org/pypy/pypy/changeset/ce341006d844/ Log:added more information to the output of regalloc.py diff --git a/rpython/jit/backend/tool/regalloc.py

[pypy-commit] pypy vecopt: remembering the position of the guard exit to resume the regallocator at a guard exit

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78169:6466422700f1 Date: 2015-06-18 11:41 +0200 http://bitbucket.org/pypy/pypy/changeset/6466422700f1/ Log:remembering the position of the guard exit to resume the regallocator at a guard exit diff --git

[pypy-commit] pypy vecopt: typo in arm assembler.py

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78170:e5cf1e51db67 Date: 2015-06-18 11:43 +0200 http://bitbucket.org/pypy/pypy/changeset/e5cf1e51db67/ Log:typo in arm assembler.py diff --git a/rpython/jit/backend/arm/assembler.py b/rpython/jit/backend/arm/assembler.py ---

[pypy-commit] pypy regalloc: exchanged the heuristic for the register allocator. it now takes the live range whose next use is furthest in the future

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: regalloc Changeset: r78168:0dcfffa6613e Date: 2015-06-18 11:00 +0200 http://bitbucket.org/pypy/pypy/changeset/0dcfffa6613e/ Log:exchanged the heuristic for the register allocator. it now takes the live range whose next use is furthest in

[pypy-commit] pypy vecopt: missing bracket

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78256:7cdaa1548c93 Date: 2015-06-23 12:26 +0200 http://bitbucket.org/pypy/pypy/changeset/7cdaa1548c93/ Log:missing bracket diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: added some temporary timing code for the trace execution

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78255:052ab2d7af35 Date: 2015-06-23 12:10 +0200 http://bitbucket.org/pypy/pypy/changeset/052ab2d7af35/ Log:added some temporary timing code for the trace execution diff --git a/rpython/jit/metainterp/compile.py

[pypy-commit] pypy vecopt: timing the vecopt and printing it to the logfiles

2015-06-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78254:96394198b3ec Date: 2015-06-23 10:47 +0200 http://bitbucket.org/pypy/pypy/changeset/96394198b3ec/ Log:timing the vecopt and printing it to the logfiles diff --git a/rpython/jit/metainterp/optimizeopt/guard.py

[pypy-commit] pypy vecopt: tyring to make things more easier, complexity gets hard to manage if extending the accumulation. trying to prevent the splitting entering the scheduling (work in progress)

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78322:7419dfb817a7 Date: 2015-06-26 14:48 +0200 http://bitbucket.org/pypy/pypy/changeset/7419dfb817a7/ Log:tyring to make things more easier, complexity gets hard to manage if extending the accumulation. trying to

[pypy-commit] pypy vecopt: finishing up changes

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78324:313bcd2938c3 Date: 2015-06-26 16:34 +0200 http://bitbucket.org/pypy/pypy/changeset/313bcd2938c3/ Log:finishing up changes diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: rewritten scheduling tests to add type (could not be inferred easily), cost model passing again

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78326:d1a942296dd8 Date: 2015-06-26 17:46 +0200 http://bitbucket.org/pypy/pypy/changeset/d1a942296dd8/ Log:rewritten scheduling tests to add type (could not be inferred easily), cost model passing again diff --git

[pypy-commit] pypy vecopt: test_vectorize passing again

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78325:2df6133da026 Date: 2015-06-26 16:56 +0200 http://bitbucket.org/pypy/pypy/changeset/2df6133da026/ Log:test_vectorize passing again diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: fixed guard_false for packed arguments, the last test (set_slice) not working

2015-06-27 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78332:ccca4d45dd30 Date: 2015-06-27 19:24 +0200 http://bitbucket.org/pypy/pypy/changeset/ccca4d45dd30/ Log:fixed guard_false for packed arguments, the last test (set_slice) not working diff --git

[pypy-commit] pypy vecopt: added guard_true/false for vector register as first argument

2015-06-27 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78330:9150ce6cdf52 Date: 2015-06-27 18:10 +0200 http://bitbucket.org/pypy/pypy/changeset/9150ce6cdf52/ Log:added guard_true/false for vector register as first argument diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: all any (but not any casting from float) execute correctly

2015-06-27 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78331:12de4acb3b37 Date: 2015-06-27 18:54 +0200 http://bitbucket.org/pypy/pypy/changeset/12de4acb3b37/ Log:all any (but not any casting from float) execute correctly diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: docu comment

2015-06-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78300:01fbd63c766c Date: 2015-06-24 17:35 +0200 http://bitbucket.org/pypy/pypy/changeset/01fbd63c766c/ Log:docu comment diff --git a/rpython/doc/jit/vectorization.rst b/rpython/doc/jit/vectorization.rst ---

[pypy-commit] pypy vecopt: added llgraph impl for reduce (+,-,*)n

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78029:fa11cb002cc1 Date: 2015-06-11 11:35 +0200 http://bitbucket.org/pypy/pypy/changeset/fa11cb002cc1/ Log:added llgraph impl for reduce (+,-,*)n diff --git a/rpython/jit/backend/llgraph/runner.py

[pypy-commit] pypy vecopt: added uint vector addition test

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78027:685a9ce976c8 Date: 2015-06-11 11:16 +0200 http://bitbucket.org/pypy/pypy/changeset/685a9ce976c8/ Log:added uint vector addition test diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: removed set_trace call

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78028:2c6a43445680 Date: 2015-06-11 11:27 +0200 http://bitbucket.org/pypy/pypy/changeset/2c6a43445680/ Log:removed set_trace call diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: added tests and parameterized one

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78026:645362226e3e Date: 2015-06-11 10:55 +0200 http://bitbucket.org/pypy/pypy/changeset/645362226e3e/ Log:added tests and parameterized one diff --git a/pypy/module/micronumpy/compile.py b/pypy/module/micronumpy/compile.py

[pypy-commit] pypy vecopt: added test case to ensure correctness of transformation and find all missing assembler implementations

2015-06-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78108:04d9da20e80b Date: 2015-06-15 12:04 +0200 http://bitbucket.org/pypy/pypy/changeset/04d9da20e80b/ Log:added test case to ensure correctness of transformation and find all missing assembler implementations diff

[pypy-commit] pypy vecopt: jit.promote in the numpy internal iterator generates slightly different traces, adjusted tests with the new constant

2015-06-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78107:3038636d0684 Date: 2015-06-15 10:49 +0200 http://bitbucket.org/pypy/pypy/changeset/3038636d0684/ Log:jit.promote in the numpy internal iterator generates slightly different traces, adjusted tests with the new

[pypy-commit] pypy vecopt: added test case to sum on llgraph

2015-06-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78099:2e80beb1f2b4 Date: 2015-06-15 09:20 +0200 http://bitbucket.org/pypy/pypy/changeset/2e80beb1f2b4/ Log:added test case to sum on llgraph diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: error in the test case added array instead of summed scalar

2015-06-15 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78106:1d9b7639a828 Date: 2015-06-15 09:40 +0200 http://bitbucket.org/pypy/pypy/changeset/1d9b7639a828/ Log:error in the test case added array instead of summed scalar diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: added a vectorize_user parameter to try the vectorization on user code (array[i], numpy.array[i]), removed some old checks for test_axissum

2015-05-27 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77618:c5bf438f0e12 Date: 2015-05-27 11:13 +0200 http://bitbucket.org/pypy/pypy/changeset/c5bf438f0e12/ Log:added a vectorize_user parameter to try the vectorization on user code (array[i], numpy.array[i]), removed some

[pypy-commit] pypy vecopt: bug fix. did not copy descriptors while unrolling instructions, simplifcations and use copy_all_attr of the descr instead of doing it manually

2015-05-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77679:27be296beda0 Date: 2015-05-29 12:29 +0200 http://bitbucket.org/pypy/pypy/changeset/27be296beda0/ Log:bug fix. did not copy descriptors while unrolling instructions, simplifcations and use copy_all_attr of the

[pypy-commit] pypy vecopt: generalized extend

2015-05-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77688:d89d708eeea2 Date: 2015-05-29 14:36 +0200 http://bitbucket.org/pypy/pypy/changeset/d89d708eeea2/ Log:generalized extend dont overwrite resume at loop header descr in unrolling (this will lead to stronger

<    1   2   3   4   5   6   7   8   9   10   >