shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
156,083 Commits 1,712 Branches 298 Tags
Commit Graph

77 Commits

Author SHA1 Message Date
Matt Arsenault
3c71dabd88 Fix typo 
llvm-svn: 188915
 2013-08-21 18:54:47 +00:00
Arnold Schwaighofer
e1f3ab69d1 SLPVectorizer: Fix invalid iterator errors 
Update iterator when the SLP vectorizer changes the instructions in the basic
block by restarting the traversal of the basic block.

Patch by Yi Jiang!

Fixes PR 16899.

llvm-svn: 188832
 2013-08-20 21:21:45 +00:00
Joerg Sonnenberger
8e3050db51 PR 16899: Do not modify the basic block using the iterator, but keep the 
next value. This avoids crashes due to invalidation.

Patch by Joey Gouly.

llvm-svn: 188605
 2013-08-17 11:04:47 +00:00
Matt Arsenault
5cae894a13 Fix spelling 
llvm-svn: 188506
 2013-08-15 23:11:03 +00:00
Nadav Rotem
e23147bbd4 Fix PR16797 - Support PHINodes with multiple inputs from the same basic block. 
Do not generate new vector values for the same entries because we know that the incoming values
from the same block must be identical.

llvm-svn: 188185
 2013-08-12 17:46:44 +00:00
Nadav Rotem
5defea90e6 SLPVectorizer: Fix PR16777. PHInodes may use multiple extracted values that come from different blocks. 
Thanks Alexey Samsonov.

llvm-svn: 187663
 2013-08-02 18:40:24 +00:00
Nadav Rotem
d9c74cc6d3 SLPVectorier: update the debug location for the new instructions. 
llvm-svn: 187363
 2013-07-29 18:18:46 +00:00
Nadav Rotem
750e42cba3 Don't vectorize when the attribute NoImplicitFloat is used. 
llvm-svn: 187340
 2013-07-29 05:13:00 +00:00
Nadav Rotem
3e50c68956 Update the comment 
llvm-svn: 187316
 2013-07-27 23:28:47 +00:00
Nadav Rotem
cfd40da9b1 SLP Vectorier: Don't vectorize really short chains because they are already handled by the SelectionDAG store-vectorizer, which does a better job in deciding when to vectorize. 
llvm-svn: 187267
 2013-07-26 23:07:55 +00:00
Nadav Rotem
9ce0f779bc SLP Vectorizer: Disable the vectorization of non power of two chains, such as <3 x float>, because we dont have a good cost model for these types. 
llvm-svn: 187265
 2013-07-26 22:53:11 +00:00
Nadav Rotem
cf0dcdc71c When we vectorize across multiple basic blocks we may vectorize PHINodes that create a cycle. We already break the cycle on phi-nodes, but arithmetic operations are still uplicated. This patch adds code that checks if the operation that we are vectorizing was vectorized during the visit of the operands and uses this value if it can. 
llvm-svn: 186883
 2013-07-22 22:18:07 +00:00
Nadav Rotem
d7ff88a8d9 Delete unused helper functions. 
llvm-svn: 186808
 2013-07-22 05:19:22 +00:00
Nadav Rotem
f6bb6a464c Revert a part of r186420. Don't forbid multiple store chains that merge. 
llvm-svn: 186786
 2013-07-21 06:12:57 +00:00
Nadav Rotem
e210839f5b fix an 80-col line. 
llvm-svn: 186733
 2013-07-19 23:14:01 +00:00
Nadav Rotem
c069c25518 Use LLVMs ADTs that improve the compile time of this pass. 
llvm-svn: 186732
 2013-07-19 23:12:19 +00:00
Nadav Rotem
5c9a193a65 SLPVectorizer: Improve the compile time of isConsecutive by reordering the conditions that check GEPs and eliminate two of the calls to accumulateConstantOffset. 
llvm-svn: 186731
 2013-07-19 23:11:15 +00:00
Nadav Rotem
bb3398f000 Handle constants without going through SCEV. 
llvm-svn: 186593
 2013-07-18 18:34:21 +00:00
Nadav Rotem
de2815a5f7 SLPVectorizer: Speedup isConsecutive by manually checking GEPs with multiple indices. 
This brings the compile time of the SLP-Vectorizer to about 2.5% of OPT for my testcase.

llvm-svn: 186592
 2013-07-18 18:20:45 +00:00
Nadav Rotem
7d7036b8c6 SLPVectorizer: Speedup isConsecutive (that checks if two addresses are consecutive in memory) by checking for additional patterns that don't need to go through SCEV. 
llvm-svn: 186563
 2013-07-18 04:33:20 +00:00
Nadav Rotem
43639e8492 Fix a comment. 
llvm-svn: 186541
 2013-07-17 22:41:16 +00:00
Nadav Rotem
3072baeb9c Add a micro optimization to catch cases where the PtrA equals PtrB. 
llvm-svn: 186531
 2013-07-17 19:52:25 +00:00
Nadav Rotem
2202317fce SLPVectorizer: Accelerate the isConsecutive check by replacing the subtraction of the two values with a simple SCEV expression that adds the offset to one of the pointers that we compare. 
llvm-svn: 186479
 2013-07-17 00:48:31 +00:00
Nadav Rotem
d2e8c4cdea flip the scev minus direction to simplify the code. 
llvm-svn: 186466
 2013-07-16 22:57:06 +00:00
Nadav Rotem
8f924f3891 SLPVectorizer: Improve the compile time of isConsecutive by adding a simple constant-gep check before using SCEV. 
This check does not always work because not all of the GEPs use a constant offset, but it happens often enough to reduce the number of times we use SCEV.

llvm-svn: 186465
 2013-07-16 22:51:07 +00:00
Nadav Rotem
26bf9a0c75 SLPVectorizer: Reduce the compile time of the consecutive store lookup. 
Process groups of stores in chunks of 16.

llvm-svn: 186420
 2013-07-16 15:25:17 +00:00
Nadav Rotem
1c1d6c1666 PR16628: Fix a bug in the code that merges compares. 
Compares return i1 but they compare different types.

llvm-svn: 186359
 2013-07-15 22:52:48 +00:00
Nadav Rotem
d9f3f4548e SLPVectorizer: change the order in which we search for vectorization candidates. Do stores first and PHIs second. 
llvm-svn: 186277
 2013-07-14 06:15:46 +00:00
Nadav Rotem
89c41bf06a SLPVectorizer: Sink and enable CSE for ExtractElements. 
llvm-svn: 186145
 2013-07-12 06:09:24 +00:00
Nadav Rotem
fa3c2db211 SLPVectorize: Replace the code that checks for vectorization candidates in successor blocks with code that scans PHINodes. 
Before we could vectorize PHINodes scanning successors was a good way of finding candidates. Now we can vectorize the phinodes which is simpler.

llvm-svn: 186139
 2013-07-12 00:04:18 +00:00
Nadav Rotem
db06b139fd Remove an argument that we dont use anymore. 
llvm-svn: 186116
 2013-07-11 20:56:13 +00:00
Nadav Rotem
08efb262a9 Fix a warning. 
llvm-svn: 186064
 2013-07-11 05:39:02 +00:00
Nadav Rotem
b8dd66f655 SLPVectorizer: refactor the code that places extracts. Place the code that decides where to put extracts in the build-tree phase. This allows us to take the cost of the extracts into account. 
llvm-svn: 186058
 2013-07-11 04:54:05 +00:00
Nadav Rotem
d7b574e5b3 Fix PR16571, which is a bug in the code that checks that all of the types in the bundle are uniform. 
llvm-svn: 185970
 2013-07-09 21:38:08 +00:00
Nadav Rotem
861bef7dd0 Set the default insert point to the first instruction, and not to end() 
llvm-svn: 185953
 2013-07-09 17:55:36 +00:00
Nadav Rotem
c9c57518ab This patch changes the saved IRBuilder insert point from BasicBlock::iterator to AssertingVH. 
Commit 185883 fixes a bug in the IRBuilder that should fix the ASan bot. AssertingVH can help in exposing some RAUW problems.

Thanks Ben and Alexey!

llvm-svn: 185886
 2013-07-08 23:31:13 +00:00
Nadav Rotem
2ee35771a8 Clear the builder insert point between tree-vectorization phases. 
llvm-svn: 185777
 2013-07-07 14:57:18 +00:00
Nadav Rotem
2041b742d4 SLPVectorizer: Implement DCE as part of vectorization. 
This is a complete re-write if the bottom-up vectorization class.
Before this commit we scanned the instruction tree 3 times. First in search of merge points for the trees. Second, for estimating the cost. And finally for vectorization.
There was a lot of code duplication and adding the DCE exposed bugs. The new design is simpler and DCE was a part of the design.
In this implementation we build the tree once. After that we estimate the cost by scanning the different entries in the constructed tree (in any order). The vectorization phase also works on the built tree.

llvm-svn: 185774
 2013-07-07 06:57:07 +00:00
Craig Topper
af0dea1347 Use SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifying the vector size. 
llvm-svn: 185606
 2013-07-04 01:31:24 +00:00
Nadav Rotem
0a25727f31 We preserve the CFG and some of the analysis passes. 
llvm-svn: 185251
 2013-06-29 05:38:15 +00:00
Nadav Rotem
e00343446c Update docs. 
llvm-svn: 185250
 2013-06-29 05:37:19 +00:00
Nadav Rotem
060be733a5 SLP Vectorizer: Add support for trees with external users. 
To support this we have to insert 'extractelement' instructions to pick the right lane.
We had this functionality before but I removed it when we moved to the multi-block design because it was too complicated.

llvm-svn: 185230
 2013-06-28 22:07:09 +00:00
Nadav Rotem
8edefb3665 No need to use a Set when a vector would do. 
llvm-svn: 185047
 2013-06-27 00:14:13 +00:00
Nadav Rotem
93f880fb77 SLP: When searching for vectorization opportunities scan the blocks in post-order because we grow chains upwards. 
llvm-svn: 185041
 2013-06-26 23:44:45 +00:00
Nadav Rotem
7f0d6d7975 SLP: Dont erase instructions during vectorization because it prevents the outerloops from iterating over the instructions. 
llvm-svn: 185040
 2013-06-26 23:43:23 +00:00
Nadav Rotem
4c5b2d1de6 Erase all of the instructions that we RAUWed 
llvm-svn: 184969
 2013-06-26 17:16:09 +00:00
Nadav Rotem
f4ca3994b8 Do not add cse-ed instructions into the visited map because we dont want to consider them as a candidate for replacement of instructions to be visited. 
llvm-svn: 184966
 2013-06-26 16:54:53 +00:00
Nadav Rotem
0794acc1da SLPVectorizer: support slp-vectorization of PHINodes between basic blocks 
llvm-svn: 184888
 2013-06-25 23:04:09 +00:00
Nadav Rotem
3de032a3b6 Fix a typo in the code that collected the costs recursively. 
llvm-svn: 184827
 2013-06-25 05:30:56 +00:00
Nadav Rotem
9c7c997a7e Rename the variable to fix a warning. Thanks Andy Gibbs. 
llvm-svn: 184749
 2013-06-24 15:59:47 +00:00