[pypy-commit] pypy vecopt: first try to merge iterators in call2

2015-05-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77689:b47f0d81442a Date: 2015-05-29 15:55 +0200 http://bitbucket.org/pypy/pypy/changeset/b47f0d81442a/ Log:first try to merge iterators in call2 diff --git a/pypy/module/micronumpy/iterators.py

[pypy-commit] pypy vecopt: distinct between input/output argument in vector type conversion

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77736:cb7dddccc7f0 Date: 2015-06-01 09:14 +0200 http://bitbucket.org/pypy/pypy/changeset/cb7dddccc7f0/ Log:distinct between input/output argument in vector type conversion call2 uses a list to track iterator and their

[pypy-commit] pypy vecopt: reverted call2, list + index left the access in the trace

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77737:c191d623bf18 Date: 2015-06-01 09:27 +0200 http://bitbucket.org/pypy/pypy/changeset/c191d623bf18/ Log:reverted call2, list + index left the access in the trace diff --git a/pypy/module/micronumpy/iterators.py

[pypy-commit] pypy vecopt: started to add cost model, all enabled zjit tests passing now

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77738:c322539e68de Date: 2015-06-01 09:41 +0200 http://bitbucket.org/pypy/pypy/changeset/c322539e68de/ Log:started to add cost model, all enabled zjit tests passing now diff --git

[pypy-commit] pypy vecopt: rpython compliant, started to impl cost model

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77739:443876d529b4 Date: 2015-06-01 10:15 +0200 http://bitbucket.org/pypy/pypy/changeset/443876d529b4/ Log:rpython compliant, started to impl cost model diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy vecopt: activated all but 3 zjit tests (pow, take missing), all others pass

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77741:09b0ee52aaf2 Date: 2015-06-01 15:47 +0200 http://bitbucket.org/pypy/pypy/changeset/09b0ee52aaf2/ Log:activated all but 3 zjit tests (pow, take missing), all others pass added a jit param vec_cost to still be able

[pypy-commit] pypy vecopt: removed unused dict, lead to rpy error (no values included)

2015-06-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77742:9492770f0741 Date: 2015-06-01 15:52 +0200 http://bitbucket.org/pypy/pypy/changeset/9492770f0741/ Log:removed unused dict, lead to rpy error (no values included) diff --git

[pypy-commit] pypy vecopt: factored out a rename object for trace operations

2015-05-28 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77639:de37694c94d1 Date: 2015-05-28 10:16 +0200 http://bitbucket.org/pypy/pypy/changeset/de37694c94d1/ Log:factored out a rename object for trace operations same as for guard strengthing is not not used anymore (should

[pypy-commit] pypy vecopt: retinkering the dependency construction, statements with sideeffects need stronger dependencies

2015-05-28 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77640:f83b729acb89 Date: 2015-05-28 13:04 +0200 http://bitbucket.org/pypy/pypy/changeset/f83b729acb89/ Log:retinkering the dependency construction, statements with sideeffects need stronger dependencies improved the

[pypy-commit] pypy vecopt: correctly emitting normal guards not subject of the strengh opt

2015-05-28 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77655:f9c2fe944393 Date: 2015-05-28 15:50 +0200 http://bitbucket.org/pypy/pypy/changeset/f9c2fe944393/ Log:correctly emitting normal guards not subject of the strengh opt copying failargs, descr and frame info on

[pypy-commit] pypy vecopt: uncommented more tests to see how well current version works

2015-05-28 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77658:d004834cb66f Date: 2015-05-28 18:06 +0200 http://bitbucket.org/pypy/pypy/changeset/d004834cb66f/ Log:uncommented more tests to see how well current version works diff --git a/pypy/module/micronumpy/test/test_zjit.py

[pypy-commit] pypy vecopt: excluded OP_GC from vectorization. Those ops can load complex objects which are not subject of this optimization

2015-06-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77756:f80a25298cf3 Date: 2015-06-02 11:57 +0200 http://bitbucket.org/pypy/pypy/changeset/f80a25298cf3/ Log:excluded OP_GC from vectorization. Those ops can load complex objects which are not subject of this optimization

[pypy-commit] pypy vecopt: wrong assumption: it can happen that a call that is exception guarded is not immediatley after the call

2015-06-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77758:cd53c1c662a5 Date: 2015-06-02 14:00 +0200 http://bitbucket.org/pypy/pypy/changeset/cd53c1c662a5/ Log:wrong assumption: it can happen that a call that is exception guarded is not immediatley after the call diff

[pypy-commit] pypy vecopt: vector boxes in label and jump args are not correctly dispatched as xmm registers

2015-06-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77768:8ef2e618034c Date: 2015-06-02 16:07 +0200 http://bitbucket.org/pypy/pypy/changeset/8ef2e618034c/ Log:vector boxes in label and jump args are not correctly dispatched as xmm registers diff --git

[pypy-commit] pypy vecopt: expanding variables/constants are not inlined before the instruction but gathered in another list to be prepended to the label operation

2015-06-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77760:692b8f945d3a Date: 2015-06-02 15:19 +0200 http://bitbucket.org/pypy/pypy/changeset/692b8f945d3a/ Log:expanding variables/constants are not inlined before the instruction but gathered in another list to be

[pypy-commit] pypy vecopt: removed primitive type mixin

2015-06-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77759:20d16b623806 Date: 2015-06-02 14:43 +0200 http://bitbucket.org/pypy/pypy/changeset/20d16b623806/ Log:removed primitive type mixin added two schedule tests for constant/variable expansion, need to move the

[pypy-commit] pypy vecopt: correctly emit reduction operation in a guard exit that compiles a bridge (was missing before)

2015-07-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78376:ac77811327eb Date: 2015-07-01 12:35 +0200 http://bitbucket.org/pypy/pypy/changeset/ac77811327eb/ Log:correctly emit reduction operation in a guard exit that compiles a bridge (was missing before) added prod(...)

[pypy-commit] pypy vecopt: resolving issues with * accumulation, there where some assembler routines i did not implement correctly (but where not invoked beforehand)

2015-07-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78378:88fff9cde657 Date: 2015-07-01 13:48 +0200 http://bitbucket.org/pypy/pypy/changeset/88fff9cde657/ Log:resolving issues with * accumulation, there where some assembler routines i did not implement correctly (but

[pypy-commit] pypy vecopt: restricted input size for vectorization to 75 operations max (dont see any chance for it to succeed with more than 20-30 operations)

2015-07-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78383:a01f8e4831c4 Date: 2015-07-01 18:05 +0200 http://bitbucket.org/pypy/pypy/changeset/a01f8e4831c4/ Log:restricted input size for vectorization to 75 operations max (dont see any chance for it to succeed with more

[pypy-commit] pypy vecopt: updated docs

2015-07-01 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78382:c878b10c3713 Date: 2015-07-01 16:05 +0200 http://bitbucket.org/pypy/pypy/changeset/c878b10c3713/ Log:updated docs diff --git a/rpython/doc/jit/vectorization.rst b/rpython/doc/jit/vectorization.rst ---

[pypy-commit] pypy vecopt: preventing signext from int16, int8 to greater or vice versa, the first direction was ignored up to now

2015-07-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78398:8022a49a8008 Date: 2015-07-02 09:07 +0200 http://bitbucket.org/pypy/pypy/changeset/8022a49a8008/ Log:preventing signext from int16,int8 to greater or vice versa, the first direction was ignored up to now diff

[pypy-commit] pypy vecopt: product assembler now uses SHUFPD instead of SHUFPS and moves two floats instead of one to the accumulator

2015-07-02 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78405:03f3796197b8 Date: 2015-07-02 10:48 +0200 http://bitbucket.org/pypy/pypy/changeset/03f3796197b8/ Log:product assembler now uses SHUFPD instead of SHUFPS and moves two floats instead of one to the accumulator

[pypy-commit] pypy vecopt: pass zjit, splitting is not necessary anymore, because the packing phase determines if it is able to combine pairs or not

2015-06-28 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78340:3ad6a660ea4c Date: 2015-06-28 17:25 +0200 http://bitbucket.org/pypy/pypy/changeset/3ad6a660ea4c/ Log:pass zjit, splitting is not necessary anymore, because the packing phase determines if it is able to combine

[pypy-commit] pypy vecopt: signext (due to its missing x86 opcodes) is weigthed as not a profitable operation

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78025:eaef430ca49d Date: 2015-06-11 10:08 +0200 http://bitbucket.org/pypy/pypy/changeset/eaef430ca49d/ Log:signext (due to its missing x86 opcodes) is weigthed as not a profitable operation diff --git

[pypy-commit] pypy vecopt: some test changes, removed an error when the output type is none

2015-06-11 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78030:c291ce811739 Date: 2015-06-11 17:15 +0200 http://bitbucket.org/pypy/pypy/changeset/c291ce811739/ Log:some test changes, removed an error when the output type is none diff --git

[pypy-commit] pypy vecopt: removed rpython problems

2015-05-25 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77545:118fa7ba7526 Date: 2015-05-25 18:00 +0200 http://bitbucket.org/pypy/pypy/changeset/118fa7ba7526/ Log:removed rpython problems diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt: removed the const arg for each vecop (but not load)

2015-05-25 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77543:bccd719ea178 Date: 2015-05-25 17:01 +0200 http://bitbucket.org/pypy/pypy/changeset/bccd719ea178/ Log:removed the const arg for each vecop (but not load) rewrote the unpacking/packing and gave it a new structure

[pypy-commit] pypy vecopt: added float_neg and float_abs implementations to x86

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77560:8b6acbbc80c6 Date: 2015-05-26 09:35 +0200 http://bitbucket.org/pypy/pypy/changeset/8b6acbbc80c6/ Log:added float_neg and float_abs implementations to x86 added a new single float neg/abs constants to the heap

[pypy-commit] pypy vecopt: removed count from float_expand asm args, added float expand for single float (parts of it in previous commit)

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77561:2223e2dde667 Date: 2015-05-26 09:39 +0200 http://bitbucket.org/pypy/pypy/changeset/2223e2dde667/ Log:removed count from float_expand asm args, added float expand for single float (parts of it in previous commit)

[pypy-commit] pypy vecopt: constant expansion in trace now adds the constants to the heap memory (instead of crashing)

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77562:ab0e626b4c0c Date: 2015-05-26 10:29 +0200 http://bitbucket.org/pypy/pypy/changeset/ab0e626b4c0c/ Log:constant expansion in trace now adds the constants to the heap memory (instead of crashing) diff --git

[pypy-commit] pypy vecopt: added test case for float_truediv (was not included in transformation), float_neg, float_abs

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77559:39f5a5e73321 Date: 2015-05-26 08:37 +0200 http://bitbucket.org/pypy/pypy/changeset/39f5a5e73321/ Log:added test case for float_truediv (was not included in transformation), float_neg, float_abs added float_abs,

[pypy-commit] pypy vecopt: rpy test passes for recent changes

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77563:1115d37e0455 Date: 2015-05-26 10:42 +0200 http://bitbucket.org/pypy/pypy/changeset/1115d37e0455/ Log:rpy test passes for recent changes added new test file for scheduling specific tests diff --git

[pypy-commit] pypy vecopt: resolved problem with guard strengthening (boolinverse needed if guard_false)

2015-05-22 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77496:3931485d86f0 Date: 2015-05-22 17:11 +0200 http://bitbucket.org/pypy/pypy/changeset/3931485d86f0/ Log:resolved problem with guard strengthening (boolinverse needed if guard_false) guard implication supported (might

[pypy-commit] pypy vecopt: refactored the dependency construction for guards

2015-05-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77452:2cdfa2593741 Date: 2015-05-21 09:56 +0200 http://bitbucket.org/pypy/pypy/changeset/2cdfa2593741/ Log:refactored the dependency construction for guards guard relaxation is now simpler and faster saving backwards

[pypy-commit] pypy vecopt: rewritten the guard strengthening. it is now independent from vecopt (still contained in the same file though).

2015-05-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77454:1f73ac83382c Date: 2015-05-21 18:26 +0200 http://bitbucket.org/pypy/pypy/changeset/1f73ac83382c/ Log:rewritten the guard strengthening. it is now independent from vecopt (still contained in the same file though).

[pypy-commit] pypy vecopt: removed attribute error when printing unrolled/vectorized loop

2015-05-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77453:45b8d9752af2 Date: 2015-05-21 10:15 +0200 http://bitbucket.org/pypy/pypy/changeset/45b8d9752af2/ Log:removed attribute error when printing unrolled/vectorized loop diff --git

[pypy-commit] pypy vecopt: enhanced test env to check schedule operation

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77583:9eec1d3c661e Date: 2015-05-26 14:17 +0200 http://bitbucket.org/pypy/pypy/changeset/9eec1d3c661e/ Log:enhanced test env to check schedule operation added same_shape to AbstractValue (needed for comparison of two

[pypy-commit] pypy vecopt: added missing result for pack instruction (expand box to vector)

2015-05-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r77584:0aebdb7b396b Date: 2015-05-26 16:56 +0200 http://bitbucket.org/pypy/pypy/changeset/0aebdb7b396b/ Log:added missing result for pack instruction (expand box to vector) added some instructions for float/double

[pypy-commit] pypy vecopt: setting the orig label arguments in the test case

2015-08-03 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78754:0359a9053df7 Date: 2015-08-03 17:26 +0200 http://bitbucket.org/pypy/pypy/changeset/0359a9053df7/ Log:setting the orig label arguments in the test case diff --git

[pypy-commit] pypy vecopt: failargs can be none

2015-08-03 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78753:6ef92e33b506 Date: 2015-08-03 16:49 +0200 http://bitbucket.org/pypy/pypy/changeset/6ef92e33b506/ Log:failargs can be none diff --git a/rpython/jit/metainterp/optimizeopt/vectorize.py

[pypy-commit] pypy vecopt-merge: changes to get translation going, vec_params is a str parameter like enabled_opts

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79033:b5a77dd8c0bf Date: 2015-08-18 12:22 +0200 http://bitbucket.org/pypy/pypy/changeset/b5a77dd8c0bf/ Log:changes to get translation going, vec_params is a str parameter like enabled_opts diff --git

[pypy-commit] pypy vecopt-merge: added some comments to the state switching

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79005:299b91377f5f Date: 2015-08-17 10:40 +0200 http://bitbucket.org/pypy/pypy/changeset/299b91377f5f/ Log:added some comments to the state switching diff --git a/pypy/module/micronumpy/loop.py

[pypy-commit] pypy vecopt-merge: added iterator sharing to call1 (only one possibility which makes things easier)

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79006:a610dc93d3a2 Date: 2015-08-17 10:51 +0200 http://bitbucket.org/pypy/pypy/changeset/a610dc93d3a2/ Log:added iterator sharing to call1 (only one possibility which makes things easier) diff --git

[pypy-commit] pypy vecopt-merge: moved post/pre unpacking from vectorize.py into schedule.py, this removes unnecessary object allocations for each scheduling step, makes things more readable

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79007:318a57aaddb8 Date: 2015-08-17 11:18 +0200 http://bitbucket.org/pypy/pypy/changeset/318a57aaddb8/ Log:moved post/pre unpacking from vectorize.py into schedule.py, this removes unnecessary object allocations

[pypy-commit] pypy vecopt-merge-opt: closing, work was continued in vecopt-merge

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge-opt Changeset: r79008:b414e83c5a5f Date: 2015-08-17 11:19 +0200 http://bitbucket.org/pypy/pypy/changeset/b414e83c5a5f/ Log:closing, work was continued in vecopt-merge ___ pypy-commit

[pypy-commit] pypy vecopt-merge: added a new jit parameter vec_params which encode the previously known vectorize_user, vec_cost and add a maximum trace length and a ratio in the fast path to skip vec

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79029:fc65e2cfbacc Date: 2015-08-18 10:41 +0200 http://bitbucket.org/pypy/pypy/changeset/fc65e2cfbacc/ Log:added a new jit parameter vec_params which encode the previously known vectorize_user, vec_cost and add a

[pypy-commit] pypy vecopt-merge: removed recently added STR_PARAMETERS as it does not allow specialization

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79037:328fce203b85 Date: 2015-08-18 13:47 +0200 http://bitbucket.org/pypy/pypy/changeset/328fce203b85/ Log:removed recently added STR_PARAMETERS as it does not allow specialization diff --git

[pypy-commit] pypy vecopt-merge: removed last usage of STR_PARAMS

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79038:13d3acac4d6a Date: 2015-08-18 14:00 +0200 http://bitbucket.org/pypy/pypy/changeset/13d3acac4d6a/ Log:removed last usage of STR_PARAMS diff --git a/rpython/rlib/jit.py b/rpython/rlib/jit.py --- a/rpython/rlib/jit.py

[pypy-commit] pypy vecopt-merge: vec_params in interp_jit.py must be a separate case to wrap it in a w_str

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79036:483562ca6a0f Date: 2015-08-18 13:14 +0200 http://bitbucket.org/pypy/pypy/changeset/483562ca6a0f/ Log:vec_params in interp_jit.py must be a separate case to wrap it in a w_str diff --git

[pypy-commit] pypy vecopt-merge: set_param second parameter needs to be specialized, thus cannot be variable

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79035:f874a9f93f05 Date: 2015-08-18 12:48 +0200 http://bitbucket.org/pypy/pypy/changeset/f874a9f93f05/ Log:set_param second parameter needs to be specialized, thus cannot be variable diff --git

[pypy-commit] pypy vecopt-merge: typo name should be name1, added test case that creates cycle in dependency

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79039:61d63941604d Date: 2015-08-18 14:40 +0200 http://bitbucket.org/pypy/pypy/changeset/61d63941604d/ Log:typo name should be name1, added test case that creates cycle in dependency diff --git

[pypy-commit] pypy vecopt-merge: further cut down the search space to move guards to an earlier position, at the same time it seems that I have made it weaker. some guards are not moved anymore (need

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79052:8fa916fbce20 Date: 2015-08-19 13:19 +0200 http://bitbucket.org/pypy/pypy/changeset/8fa916fbce20/ Log:further cut down the search space to move guards to an earlier position, at the same time it seems that I

[pypy-commit] pypy vecopt-merge: adapted tests to ignore the jump. the jump operation is not tracked anymore as a dependency, but it is always emitted at the latest position in the trace

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79054:fc7129745344 Date: 2015-08-19 14:03 +0200 http://bitbucket.org/pypy/pypy/changeset/fc7129745344/ Log:adapted tests to ignore the jump. the jump operation is not tracked anymore as a dependency, but it is

[pypy-commit] pypy vecopt-merge: since jump is not scheduled in emit_into anymore it needs correct treatment

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79053:6750bbcedc0e Date: 2015-08-19 13:28 +0200 http://bitbucket.org/pypy/pypy/changeset/6750bbcedc0e/ Log:since jump is not scheduled in emit_into anymore it needs correct treatment diff --git

[pypy-commit] pypy vecopt-merge: why not split the parameters up into several different integer ones, why not do it this way in the first place?

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79056:7594920913e7 Date: 2015-08-19 14:54 +0200 http://bitbucket.org/pypy/pypy/changeset/7594920913e7/ Log:why not split the parameters up into several different integer ones, why not do it this way in the first

[pypy-commit] pypy vecopt-merge: typo

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79055:ad32f76090e6 Date: 2015-08-19 14:08 +0200 http://bitbucket.org/pypy/pypy/changeset/ad32f76090e6/ Log:typo diff --git a/rpython/rlib/jit.py b/rpython/rlib/jit.py --- a/rpython/rlib/jit.py +++ b/rpython/rlib/jit.py

[pypy-commit] pypy vecopt-merge: documentation additions (command line flags), added description of the ABC optimization, note on limitations

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79117:3c01987639c7 Date: 2015-08-21 15:43 +0200 http://bitbucket.org/pypy/pypy/changeset/3c01987639c7/ Log:documentation additions (command line flags), added description of the ABC optimization, note on

[pypy-commit] pypy vecopt-merge: docu

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79120:e7a3de7993aa Date: 2015-08-21 16:05 +0200 http://bitbucket.org/pypy/pypy/changeset/e7a3de7993aa/ Log:docu diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt-merge: translation issues

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79122:8cc48941c8b9 Date: 2015-08-21 17:19 +0200 http://bitbucket.org/pypy/pypy/changeset/8cc48941c8b9/ Log:translation issues diff --git a/rpython/jit/metainterp/compile.py b/rpython/jit/metainterp/compile.py ---

[pypy-commit] pypy vecopt-merge: reverted the repr_of_arg, added more debug info to update_accumlocation_loc

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79123:dc014b620f41 Date: 2015-08-21 18:52 +0200 http://bitbucket.org/pypy/pypy/changeset/dc014b620f41/ Log:reverted the repr_of_arg, added more debug info to update_accumlocation_loc diff --git

[pypy-commit] pypy vecopt-merge: fixed issues with test cases for llgraph

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79124:38a19b56ef23 Date: 2015-08-21 19:25 +0200 http://bitbucket.org/pypy/pypy/changeset/38a19b56ef23/ Log:fixed issues with test cases for llgraph diff --git a/rpython/jit/backend/llgraph/runner.py

[pypy-commit] pypy vecopt-merge: updated command line vec parameter

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79125:754c30aac049 Date: 2015-08-21 19:27 +0200 http://bitbucket.org/pypy/pypy/changeset/754c30aac049/ Log:updated command line vec parameter diff --git a/pypy/module/pypyjit/test_pypy_c/test_micronumpy.py

[pypy-commit] pypy vecopt-merge: translation issue, assigning to wrong class

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79169:2e20dba31d03 Date: 2015-08-23 20:36 +0200 http://bitbucket.org/pypy/pypy/changeset/2e20dba31d03/ Log:translation issue, assigning to wrong class diff --git a/rpython/jit/metainterp/optimizeopt/guard.py

[pypy-commit] pypy vecopt-merge: added docu to LoopVersion, translation issue with tuple/None unification

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79170:70512ea6c862 Date: 2015-08-23 20:50 +0200 http://bitbucket.org/pypy/pypy/changeset/70512ea6c862/ Log:added docu to LoopVersion, translation issue with tuple/None unification diff --git

[pypy-commit] pypy vecopt-merge: need the 4th element in the tuple, the reference to descr.version is removed!

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79174:c0ed1ea7fc3b Date: 2015-08-23 21:44 +0200 http://bitbucket.org/pypy/pypy/changeset/c0ed1ea7fc3b/ Log:need the 4th element in the tuple, the reference to descr.version is removed! diff --git

[pypy-commit] pypy vecopt-merge: saving of accum state more explicit

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79156:caedfb3ee99b Date: 2015-08-23 15:14 +0200 http://bitbucket.org/pypy/pypy/changeset/caedfb3ee99b/ Log:saving of accum state more explicit diff --git a/rpython/jit/backend/llsupport/assembler.py

[pypy-commit] pypy vecopt-merge: updated failure for accum to be printed at runtime

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79150:b06be6fc7944 Date: 2015-08-23 08:58 +0200 http://bitbucket.org/pypy/pypy/changeset/b06be6fc7944/ Log:updated failure for accum to be printed at runtime diff --git a/rpython/jit/backend/x86/regalloc.py

[pypy-commit] pypy vecopt-merge: used not_implemented function already present in regalloc.py

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79151:12bbf27bd64c Date: 2015-08-23 12:29 +0200 http://bitbucket.org/pypy/pypy/changeset/12bbf27bd64c/ Log:used not_implemented function already present in regalloc.py diff --git a/rpython/jit/backend/x86/regalloc.py

[pypy-commit] pypy vecopt-merge: a all new stitch bridge that considers register mapping. works for accumulation values as well

2015-08-23 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79159:0a4078644343 Date: 2015-08-23 17:24 +0200 http://bitbucket.org/pypy/pypy/changeset/0a4078644343/ Log:a all new stitch bridge that considers register mapping. works for accumulation values as well diff --git

[pypy-commit] pypy vecopt-merge: memory error does not happen anymore on vecopt-merge

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79204:9e571b42712d Date: 2015-08-24 15:27 +0200 http://bitbucket.org/pypy/pypy/changeset/9e571b42712d/ Log:memory error does not happen anymore on vecopt-merge diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy vecopt-merge: translation issue, insert index must be proven = 0

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79208:b891a59cd1a7 Date: 2015-08-24 17:42 +0200 http://bitbucket.org/pypy/pypy/changeset/b891a59cd1a7/ Log:translation issue, insert index must be proven = 0 diff --git a/rpython/jit/metainterp/history.py

[pypy-commit] pypy vecopt-merge: do not store the version on the descr anymore, but on an external data structure just for this purpose

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79207:89a3655d9226 Date: 2015-08-24 17:34 +0200 http://bitbucket.org/pypy/pypy/changeset/89a3655d9226/ Log:do not store the version on the descr anymore, but on an external data structure just for this purpose

[pypy-commit] pypy vecopt-merge: commented print

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79209:ce56ac915e9a Date: 2015-08-24 18:23 +0200 http://bitbucket.org/pypy/pypy/changeset/ce56ac915e9a/ Log:commented print diff --git a/rpython/jit/metainterp/compile.py b/rpython/jit/metainterp/compile.py ---

[pypy-commit] pypy vecopt-merge: updated llgraph to work with the new model (again :)

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79211:3742fae37b90 Date: 2015-08-24 19:21 +0200 http://bitbucket.org/pypy/pypy/changeset/3742fae37b90/ Log:updated llgraph to work with the new model (again :) diff --git a/rpython/jit/backend/llgraph/runner.py

[pypy-commit] pypy vecopt-merge: removed commented prints

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79210:eab397a1b567 Date: 2015-08-24 19:19 +0200 http://bitbucket.org/pypy/pypy/changeset/eab397a1b567/ Log:removed commented prints diff --git a/rpython/jit/metainterp/compile.py b/rpython/jit/metainterp/compile.py ---

[pypy-commit] pypy vecopt-merge: accum problem with saving data on fail descr (work in progress)

2015-08-20 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79087:a9ccb0f0842b Date: 2015-08-20 13:37 +0200 http://bitbucket.org/pypy/pypy/changeset/a9ccb0f0842b/ Log:accum problem with saving data on fail descr (work in progress) diff --git a/rpython/jit/backend/x86/regalloc.py

[pypy-commit] pypy vecopt-merge: llgraph is now working with accum data

2015-08-21 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79091:10582848a63f Date: 2015-08-20 14:18 +0200 http://bitbucket.org/pypy/pypy/changeset/10582848a63f/ Log:llgraph is now working with accum data diff --git a/rpython/jit/backend/llgraph/runner.py

[pypy-commit] pypy vecopt-merge: added two tests ensuring that paths are constructed correctly, Path was not copied

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79040:4c2ccf278516 Date: 2015-08-18 17:08 +0200 http://bitbucket.org/pypy/pypy/changeset/4c2ccf278516/ Log:added two tests ensuring that paths are constructed correctly, Path was not copied ensured that paths are

[pypy-commit] pypy vecopt-merge: testing the path iteration, there is still a problem in analyse_index_calculation

2015-08-19 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79050:a376c01d579f Date: 2015-08-19 11:55 +0200 http://bitbucket.org/pypy/pypy/changeset/a376c01d579f/ Log:testing the path iteration, there is still a problem in analyse_index_calculation diff --git

[pypy-commit] pypy vecopt-merge: path iteration on a dependecy graph can enable blacklisting nodes that have already been visited.this cuts down the search space. this is only valid if a property must

2015-08-18 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79042:4886c173125a Date: 2015-08-18 17:49 +0200 http://bitbucket.org/pypy/pypy/changeset/4886c173125a/ Log:path iteration on a dependecy graph can enable blacklisting nodes that have already been visited.this cuts

[pypy-commit] pypy vecopt-merge-iterator-sharing: activated get/setarrayitem_gc if their descriptor uses primitive data (int/float)

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge-iterator-sharing Changeset: r79000:40eb3eef0bac Date: 2015-08-17 09:55 +0200 http://bitbucket.org/pypy/pypy/changeset/40eb3eef0bac/ Log:activated get/setarrayitem_gc if their descriptor uses primitive data (int/float) added

[pypy-commit] pypy vecopt-merge: merged iterator sharing into the vecopt-merge

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge Changeset: r79002:06ec92fa38c0 Date: 2015-08-17 10:30 +0200 http://bitbucket.org/pypy/pypy/changeset/06ec92fa38c0/ Log:merged iterator sharing into the vecopt-merge diff --git a/pypy/module/micronumpy/iterators.py

[pypy-commit] pypy vecopt-merge-iterator-sharing: iterator sharing simplifications, removed the state list, indices for the iterator and moved the sharing logic in it's own function

2015-08-17 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt-merge-iterator-sharing Changeset: r79001:355ee4d5a425 Date: 2015-08-17 10:29 +0200 http://bitbucket.org/pypy/pypy/changeset/355ee4d5a425/ Log:iterator sharing simplifications, removed the state list, indices for the iterator and

[pypy-commit] pypy memoryerror: reverted some changes to get to the root of the problem

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror Changeset: r79179:a2ed66c15857 Date: 2015-08-24 10:27 +0200 http://bitbucket.org/pypy/pypy/changeset/a2ed66c15857/ Log:reverted some changes to get to the root of the problem diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy memoryerror3: furhter stripping down the commit

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79194:10602d3c86f4 Date: 2015-08-24 12:45 +0200 http://bitbucket.org/pypy/pypy/changeset/10602d3c86f4/ Log:furhter stripping down the commit diff --git a/rpython/jit/metainterp/compile.py

[pypy-commit] pypy memoryerror3: changes to translate this version

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79186:8b191cded36c Date: 2015-08-24 11:36 +0200 http://bitbucket.org/pypy/pypy/changeset/8b191cded36c/ Log:changes to translate this version diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy memoryerror3: translation issues

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79188:e119ce67b48b Date: 2015-08-24 11:56 +0200 http://bitbucket.org/pypy/pypy/changeset/e119ce67b48b/ Log:translation issues diff --git a/rpython/jit/metainterp/compile.py b/rpython/jit/metainterp/compile.py ---

[pypy-commit] pypy memoryerror3: finished the refactoring. the root loop has now it's own version, which makes generate_pending_loop_versions easier to understand. a fail descr now always points to th

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79187:743dc64b5f8a Date: 2015-08-21 14:10 +0200 http://bitbucket.org/pypy/pypy/changeset/743dc64b5f8a/ Log:finished the refactoring. the root loop has now it's own version, which makes

[pypy-commit] pypy memoryerror: some more changes

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror Changeset: r79184:e967b7cba547 Date: 2015-08-24 10:51 +0200 http://bitbucket.org/pypy/pypy/changeset/e967b7cba547/ Log:some more changes diff --git a/rpython/jit/metainterp/compile.py b/rpython/jit/metainterp/compile.py ---

[pypy-commit] pypy memoryerror2: start from the other way arround and apply changes until it does not work anymore

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror2 Changeset: r79185:0e7fdd3779b2 Date: 2015-08-24 11:05 +0200 http://bitbucket.org/pypy/pypy/changeset/0e7fdd3779b2/ Log:start from the other way arround and apply changes until it does not work anymore diff --git

[pypy-commit] pypy memoryerror3: memory error does not occur anymore

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79199:fbcc105f757d Date: 2015-08-24 13:41 +0200 http://bitbucket.org/pypy/pypy/changeset/fbcc105f757d/ Log:memory error does not occur anymore diff --git a/rpython/jit/backend/x86/assembler.py

[pypy-commit] pypy memoryerror3: still working

2015-08-24 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: memoryerror3 Changeset: r79200:f4a6026863c1 Date: 2015-08-24 13:51 +0200 http://bitbucket.org/pypy/pypy/changeset/f4a6026863c1/ Log:still working diff --git a/rpython/jit/metainterp/history.py b/rpython/jit/metainterp/history.py ---

[pypy-commit] pypy vecopt: rpython translation issues, adding reshape operation to zjit tests

2015-06-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78342:b785c8f713af Date: 2015-06-29 10:49 +0200 http://bitbucket.org/pypy/pypy/changeset/b785c8f713af/ Log:rpython translation issues, adding reshape operation to zjit tests diff --git a/pypy/module/micronumpy/compile.py

[pypy-commit] pypy vecopt: rewritten splitting of packs, added asserts to ensure the impl assumptions are correct. some tests broke (it is not yet finished)

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78321:2e935c4aa59d Date: 2015-06-26 12:29 +0200 http://bitbucket.org/pypy/pypy/changeset/2e935c4aa59d/ Log:rewritten splitting of packs, added asserts to ensure the impl assumptions are correct. some tests broke (it is

[pypy-commit] pypy vecopt: splitting must be done differently

2015-06-26 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78320:835955fe1216 Date: 2015-06-26 10:22 +0200 http://bitbucket.org/pypy/pypy/changeset/835955fe1216/ Log:splitting must be done differently diff --git a/rpython/jit/metainterp/optimizeopt/schedule.py

[pypy-commit] pypy vecopt: that was a nasty problem. entering the vecopt trace through the preamble only worked for non accum/expanded traces, otherwise the arguments would not match

2015-07-31 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78728:948a227eab7f Date: 2015-07-31 16:01 +0200 http://bitbucket.org/pypy/pypy/changeset/948a227eab7f/ Log:that was a nasty problem. entering the vecopt trace through the preamble only worked for non accum/expanded

[pypy-commit] pypy vecopt: once again I did not copy the descr...

2015-07-31 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78729:95b6c048c720 Date: 2015-07-31 16:31 +0200 http://bitbucket.org/pypy/pypy/changeset/95b6c048c720/ Log:once again I did not copy the descr... diff --git a/rpython/jit/metainterp/optimizeopt/guard.py

[pypy-commit] pypy vecopt: pblendw_xxi - only tests byte for the i argument

2015-07-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78700:1544d293e4b7 Date: 2015-07-29 08:56 +0200 http://bitbucket.org/pypy/pypy/changeset/1544d293e4b7/ Log:pblendw_xxi - only tests byte for the i argument last two tests in test_vectorize diff --git

[pypy-commit] pypy vecopt: moved vec_guard_false out of the assembler into the vector_ext file,

2015-07-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: vecopt Changeset: r78701:9c7a794d8c78 Date: 2015-07-29 09:31 +0200 http://bitbucket.org/pypy/pypy/changeset/9c7a794d8c78/ Log:moved vec_guard_false out of the assembler into the vector_ext file, adapted test_micronumpy to use int_is_ture

[pypy-commit] pypy default: #2101 ported fix from vecopt to default

2015-07-29 Thread plan_rich
Author: Richard Plangger r...@pasra.at Branch: Changeset: r78704:bdfcbfcfddb7 Date: 2015-07-29 12:12 +0200 http://bitbucket.org/pypy/pypy/changeset/bdfcbfcfddb7/ Log:#2101 ported fix from vecopt to default diff --git a/rpython/jit/codewriter/jtransform.py

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