82 Commits

Author SHA1 Message Date
Zoltan Majo
5751808a6c 8148754: C2 loop unrolling fails due to unexpected graph shape
Check if graph shape is appropriate for optimization, bail out optimization if not.

Reviewed-by: kvn, twisti, shade, dnsimon
2016-03-21 09:51:20 +01:00
Michael Berg
3ade110066 8149421: Vectorized Post Loops
Add vectorised post loop for counted loops with vectors.

Reviewed-by: kvn
2016-02-12 16:12:15 -08:00
Tobias Hartmann
ca56200977 6675699: need comprehensive fix for unconstrained ConvI2L with narrowed type
Emit CastII to make narrow ConvI2L dependent on the corresponding range check.

Reviewed-by: kvn, roland
2016-01-18 08:40:25 +01:00
Roland Westrelin
631c143469 8145322: Code generated from unsafe loops can be slightly improved
Improve code generated from checkIndex and unsafe loops

Reviewed-by: kvn, thartmann
2016-01-11 16:33:27 +01:00
Nils Eliasson
2f9e32256d Merge 2015-11-13 13:31:48 +01:00
Christian Thalinger
6896030b96 8140424: don't prefix developer and notproduct flag variables with CONST_ in product builds
Reviewed-by: goetz, stefank
2015-11-11 16:32:17 -10:00
Vladimir Kozlov
6ed8c23581 8141624: Limit calculation of pre loop during super word optimization is wrong
Reviewed-by: vlivanov, kvn
2015-11-10 21:36:35 -08:00
Michael C Berg
e86e38619e 8140779: Code generation fixes for avx512
Assembler refactoring. Abstract away instruction attributes.

Reviewed-by: kvn, roland, iveresov
2015-11-09 11:26:41 -08:00
Jesper Wilhelmsson
1a5787d42e Merge 2015-11-05 19:31:57 +01:00
Jan Civlin
0485593fbc 8139340: SuperWord enhancement to support vector conditional move (CMovVD) on Intel AVX cpu
Emit vector conditional moves

Reviewed-by: kvn
2015-10-26 19:33:31 -07:00
Nils Eliasson
5a5faf94bf 8137167: JEP165: Compiler Control: Implementation task
Compiler Control JEP

Reviewed-by: roland, twisti, zmajo, simonis
2015-10-20 18:07:28 +02:00
Goetz Lindenmaier
83b3b21342 8139040: Fix initializations before ShouldNotReachHere() etc. and enable -Wuninitialized on linux
Reviewed-by: stuefe, coleenp, roland
2015-10-22 13:07:10 -04:00
Dean Long
344a9becee Merge 2015-10-17 19:40:30 -04:00
Jesper Wilhelmsson
7078962b9c Merge 2015-10-15 13:28:22 +02:00
Jesper Wilhelmsson
e75e2e04f7 Merge 2015-10-07 01:03:24 +02:00
Felix Yang
12cba200a3 8138583: aarch64: add support for vectorizing fabs/fneg
Reviewed-by: aph, roland
2015-09-30 04:35:39 -04:00
Jan Civlin
115afda88e 8136725: Provide utility for creation a counted loop reserve copy (clone)
Make it easier to revert to the original loop should that be needed

Reviewed-by: kvn
2015-10-09 12:17:58 -07:00
David Lindholm
1e71f67736 8080775: Better argument formatting for assert() and friends
Reviewed-by: kbarrett, pliden
2015-09-29 11:02:08 +02:00
Roland Westrelin
f8abd0e843 8136820: Generate better code for some Unsafe addressing patterns
Reshape address computation to move invariant part out of loops

Reviewed-by: kvn
2015-09-17 16:53:42 +02:00
Tobias Hartmann
f298249547 8134739: compiler/loopopts/superword/TestVectorizationWithInvariant crashes in loop opts
Bail out of superword optimization if loop was removed (i.e., if zero-trip Opaque1Node was removed).

Reviewed-by: kvn, roland
2015-09-17 08:08:47 +02:00
Michael Berg
e75f5a5cde 8135028: support for vectorizing double precision sqrt
Reviewed-by: kvn, twisti
2015-09-09 10:34:17 -07:00
Goetz Lindenmaier
ad601cca0c 8131676: Fix warning 'negative int converted to unsigned' after 8085932
Also fix 64-bit constant added in 8076276.

Reviewed-by: kvn
2015-07-16 14:18:13 +02:00
Jan Civlin
2a4f18ccbd 8085932: Fixing bugs in detecting memory alignments in SuperWord
Reviewed-by: kvn
2015-07-14 09:55:22 -07:00
Vladimir Kozlov
b5284a93ce 8129920: Vectorized loop unrolling
Optimize loop opts for vectorizible loops.

Reviewed-by: kvn, roland
2015-07-10 11:59:09 -07:00
Vladimir Kozlov
d71bb8c72b 8129893: 8129094 fix is incomplete
Move old and add new primitive type check.

Reviewed-by: iveresov
2015-06-30 12:12:18 -07:00
Vladimir Kozlov
d62598136b 8129094: assert(is_java_primitive(bt)) failed: only primitive type vectors
Ignored non-primitive type memory operation regardless its (or its inputs) control.

Reviewed-by: roland, mcberg
2015-06-18 11:09:29 -07:00
Michael Berg
7c7b91845f 8080325: SuperWord loop unrolling analysis
Determine loop unroll factor based on supported vectors sizes.

Reviewed-by: roland, kvn
2015-06-16 16:10:36 -07:00
Roland Westrelin
a3c77df5f1 8077504: Unsafe load can loose control dependency and cause crash
Node::depends_only_on_test() should return false for Unsafe loads

Reviewed-by: kvn, adinn
2015-05-21 13:54:07 +02:00
Tobias Hartmann
e113dfc652 8079343: Crash in PhaseIdealLoop with "assert(!had_error) failed: bad dominance"
C2 should not try to vectorize loops with loop variant vector base address.

Reviewed-by: kvn
2015-05-11 07:44:46 +02:00
Tobias Hartmann
43a5abc876 8078497: C2's superword optimization causes unaligned memory accesses
Prevent vectorization of memory operations with different invariant offsets if unaligned memory accesses are not allowed.

Reviewed-by: kvn
2015-05-08 12:19:17 +02:00
Jan Civlin
c7d33de202 8076284: Improve vectorization of parallel streams
Improve vectorization of java/util/stream/Streams$RangeIntSpliterator::forEachRemaining() method and enable loop vectorization in a given method on demand.

Reviewed-by: kvn
2015-05-05 12:33:57 -07:00
Michael C Berg
a880f3d139 8078563: Restrict reduction optimization
Reviewed-by: kvn, twisti
2015-04-29 15:34:04 -07:00
Vladimir Kozlov
81bb7521bd 8076523: assert(((ABS(iv_adjustment_in_bytes) % elt_size) == 0)) fails in superword.cpp
Check that offset % mem_oper_size == 0 when alignment is verified during vectorization.

Reviewed-by: iveresov
2015-04-02 17:16:39 -07:00
Michael C Berg
9e55e44c85 8074981: Integer/FP scalar reduction optimization
Add scalar reduction optimization to C2 to take advantage of vector instructions in modern x86 CPUs.

Reviewed-by: kvn, twisti
2015-04-01 18:07:50 -07:00
Vladimir Kozlov
7764490363 8052081: Optimize generated by C2 code for Intel's Atom processor
Allow to execute vectorization and crc32 optimization on Atom. Enable UseFPUForSpilling by default on x86.

Reviewed-by: roland
2014-08-05 15:02:10 -07:00
Tobias Hartmann
70a55eac34 8054033: Remove unused references to Compile*
Removed unused references to Compile* missed by JDK-8034812.

Reviewed-by: kvn, roland
2014-08-05 09:58:52 +02:00
Tobias Hartmann
2a0815a55e 8034812: remove IDX_INIT macro hack in Node class
The IDX_INIT macro used by Node::Node(...) to retrieve the Compile object is removed and replaced by a call to Compile::current(). The Node constructor, new operator and all calls to it are adapted accordingly.

Reviewed-by: kvn, jrose, iveresov, goetz
2014-06-02 08:07:29 +02:00
Vladimir Kozlov
26de6296e0 8041351: Crash in src/share/vm/opto/loopnode.cpp:3215 - assert(!had_error) failed: bad dominance
Add missing is_mem() check when we collect load nodes in SuperWord::co_locate_pack().

Reviewed-by: iveresov
2014-04-22 17:45:56 -07:00
Morris Meyer
6db303a301 8001532: C2 node files refactoring
Split C2 node files into cast, convert, countbits, intrinsic, move, narrowptr and opaquenode classes

Reviewed-by: kvn, morris
2014-04-01 09:05:20 -07:00
Niclas Adlertz
94a86285a7 8010500: [parfait] Possible null pointer dereference at hotspot/src/share/vm/opto/loopnode.hpp
Added NULL check for loopnode() in get_pre_loop_end()

Reviewed-by: kvn, roland
2014-02-26 07:46:46 +01:00
Mikael Vidstedt
a0da47fd66 8029233: Update copyright year to match last edit in jdk8 hotspot repository for 2013
Copyright year updated for files modified during 2013

Reviewed-by: twisti, iveresov
2013-12-24 11:48:39 -08:00
Vladimir Kozlov
656949d014 8004867: VM crashing with assert "share/vm/opto/node.hpp:357 - assert(i < _max) failed: oob"
Added few checks and early bailout from Superword optimization to avoid such cases in a future.

Reviewed-by: roland, twisti
2013-02-18 16:47:15 -08:00
Vladimir Kozlov
fc84b11227 8001635: assert(in_bb(n)) failed: must be
Added missed check that Load node is in processed loop block.

Reviewed-by: twisti
2012-10-26 11:48:04 -07:00
Vladimir Kozlov
a9c2b6a900 8001183: incorrect results of char vectors right shift operaiton
Do vector right shift operation for small int types only after loads

Reviewed-by: jrose, dlong
2012-10-23 13:06:37 -07:00
Vladimir Ivanov
6674423523 7199654: Remove LoadUI2LNode
Removed LoadUI2L node from Ideal nodes, use match rule in .ad files instead.

Reviewed-by: kvn
2012-10-09 12:40:05 -07:00
Vladimir Kozlov
e8332c76e9 7201026: add vector for shift count
Add generation of vectors for scalar shift count.

Reviewed-by: roland, twisti, dlong
2012-10-02 12:25:13 -07:00
Bharadwaj Yadavalli
d5d2e78faa 7193318: C2: remove number of inputs requirement from Node's new operator
Deleted placement new operator of Node - node(size_t, Compile *, int).

Reviewed-by: kvn, twisti
2012-09-27 09:38:42 -07:00
Vladimir Kozlov
973fdc81a1 7200264: 7192963 changes disabled shift vectors
Replaced is_vector_use() call with explicit check for vector shift's count.

Reviewed-by: twisti, roland, dlong, vlivanov
2012-09-25 15:48:17 -07:00
Vladimir Kozlov
25f18c20d9 7199010: incorrect vector alignment
Fixed vectors alignment when several arrays are accessed in one loop.

Reviewed-by: roland, twisti
2012-09-19 16:50:26 -07:00
Vladimir Kozlov
811c047ec1 7196199: java/text/Bidi/Bug6665028.java failed: Bidi run count incorrect
Save whole XMM/YMM registers in safepoint interrupt handler.

Reviewed-by: roland, twisti
2012-09-17 19:39:07 -07:00