LLVM 4 Merge #23

jcarlson23 · 2017-05-06T20:46:20Z

I tested this with fi_tests under PTABen and while there were 4 failures (I think, it was just handful for sure) I don't think (although I'm not entirely sure) these were unexpected (non-default cases referred to Questions on PTABen test cases ). I ran a few other tests and had generally positive results as well so, I didn't run them all (On my Linux VM this went pretty smooth but on my Mac I'm still struggling a bit although I did find some of my issues, perhaps the key ones).

This has the BridgedGEPIterator solution implemented along with the previous PR's more mundane fixes, again the NodeRef inclusion in the GraphTraits was the most notable as I recall, to update SVF with LLVM 4.

… like so far it's line and file information, i.e diagnostic so..

…ysisUtil looks more like I might miss some diagnostic info

…vs X* in the node references, but let me take a look at this before rushing right in (I can do that this afternoon)

…arification of pointers or double pointers in WPASolver

…aking hey

…ngRef, inserting a few NodeRef, etc, and now compiling a fair amount of the project, but still more to go...

…e some new virtual methods in 4.0 aren't implemented...

…back to Yulei (with notes to help him...)

…ing the LLVMGetGlobalContext rather than getGlobalContext

…est should be pretty much ready to go

…re - just for a clean environment, tired of my wonkiness

yuleisui · 2017-05-07T01:15:13Z

Hi Jared,

Thanks! I will take a look at it today.

yuleisui · 2017-05-07T01:55:39Z

Your PR works very well on small test cases including fi_tests and fs_tests. I will start testing it using large benchmarks cpu2000/2006 by comparing the statistics between the existing repo and new one after PR (e.g., points-to size). If everything works, we can merge it asap.

yuleisui

Option -ander and -fspta work perfectly for all the micro-benchmarks.
I have also tested on CPU2000. Your PR produces identical points-to sets.
Very good work!!

Unfortunately, option wpa -ander -locMM *.bc causes crashes for some micro-benchmarks in fi_tests.
Could you please fix it as below?

Thanks

yuleisui · 2017-05-07T03:03:40Z

lib/MemoryModel/LocMemModel.cpp

@@ -60,7 +60,7 @@ bool LocSymTableInfo::computeGepOffset(const llvm::User *V, LocationSet& ls) {
        if (index <= baseIndex) {
            /// variant offset
            // Handling pointer types
-            if (const PointerType* pty = dyn_cast<PointerType>(*gi)) {
+	  if (const PointerType* pty = dyn_cast<PointerType>(gi.getIndexedType())) {


For LocMemModel, we can also get rid of the gep_type_iterator and gi.getIndexedType(), and stick with your BridgedGepIterator for now:).

Please use *gi intead of gi.getIndexedType() in LocMemModel. Please also replace gep_type_iterator at lines 50 and 57 with the BridgedGepIterator.

yuleisui · 2017-05-07T03:21:51Z

lib/MemoryModel/LocMemModel.cpp

                const Type* et = at->getElementType();
                Size_t sz = getTypeSizeInBytes(et);
                ls.offset += idx * sz;
            }
            // Handling struct here
-            else if (const StructType *ST = dyn_cast<StructType>(*gi)) {


Shall we use gi.getStructTypeOrNull? I am not very sure whether these two are equivalent or not. Please keep the original one as you did in MemModel.cpp

LocMemModel.cpp updated!

…ting PR with this commit

yuleisui

Thanks, Jared!

Finally, we almost get there:) I will merge your PR once the following two are fixed.

I will revise the wiki accordingly.

yuleisui · 2017-05-07T12:11:28Z

lib/MemoryModel/LocMemModel.cpp

@@ -94,19 +95,19 @@ bool LocSymTableInfo::computeGepOffset(const llvm::User *V, LocationSet& ls) {
            // Handling pointer types
            // These GEP instructions are simply making address computations from the base pointer address
            // e.g. idx1 = (char*) &MyVar + 4,  at this case gep only one offset index (idx)
-            if (const PointerType* pty = dyn_cast<PointerType>(*gi)) {
+            if (const PointerType* pty = dyn_cast<PointerType>(gi.getIndexedType())) {


Please change this back.

yuleisui · 2017-05-07T12:11:43Z

lib/MemoryModel/LocMemModel.cpp

                const Type* et = pty->getElementType();
                Size_t sz = getTypeSizeInBytes(et);
                ls.offset += idx * sz;
            }
            // Calculate the size of the array element
-            else if(const ArrayType* at = dyn_cast<ArrayType>(*gi)) {


yuleisui · 2017-05-07T12:22:24Z

lib/MemoryModel/LocMemModel.cpp

                const Type* et = pty->getElementType();
                Size_t sz = getTypeSizeInBytes(et);
                ls.offset += idx * sz;
            }
            // Calculate the size of the array element
-            else if(const ArrayType* at = dyn_cast<ArrayType>(*gi)) {
+            else if(const ArrayType* at = dyn_cast<ArrayType>(gi.getIndexedType())) {


Should be this line.

Sorry, my bad on missing this one, trying to update the commits in between breakfast ;-P

…gedIterator

yuleisui · 2017-05-07T13:26:42Z

Jared,

Please link your account to your commits, so we can acknowledge your contribution:

9ef3981

Thanks

jcarlson23 · 2017-05-07T20:01:33Z

Thanks @rockysui! I think I just took care of that (added the additional email). Thanks again.

fix LCD

jcarlson23 and others added 22 commits March 28, 2017 17:30

Starting to move this into 4.0; likely with some errors, but it looks…

54415ad

… like so far it's line and file information, i.e diagnostic so..

A few compilation fixes, the dataflow looks reasonable while the anal…

73f69af

…ysisUtil looks more like I might miss some diagnostic info

I think I almost have the specializations right for the PTACallgraph

ea5c7d4

Ok, fixed the PTACallGraph graph trait issue - at least for compilation

9abbd56

Fixed the SVFG Noderef issue, but looks like I need to clear the X** …

fdab9cd

…vs X* in the node references, but let me take a look at this before rushing right in (I can do that this afternoon)

Had to clean up where I was, should be moving forward again, needs cl…

ef3bb89

…arification of pointers or double pointers in WPASolver

Have an iterator conflict but if I can get past that looks like I'm m…

b62082f

…aking hey

Fixed that iterator mismatch by changing the GNODE type in WPASolver.h

cab36c2

Fixed const char to StringRef in the getPassName() of BreakConstantExpr

af491dc

A number of small fixes, mostly replacing const char* with llvm::Stri…

5fbd66b

…ngRef, inserting a few NodeRef, etc, and now compiling a fair amount of the project, but still more to go...

Couple trivial fixes, more StringRef and a ref to pointer

de31c27

Starting to compile actual products but linkage errors likely indicat…

7b7e39d

…e some new virtual methods in 4.0 aren't implemented...

1% more to go...

837d2d2

Ok, down to one little header issue and then write it up and pass it …

099e767

…back to Yulei (with notes to help him...)

Fixed that header issue, looks like the next few are pretty simple

e5c9dd3

Fixed the wpa and saber tools - just needed a few args changed and us…

37111d4

…ing the LLVMGetGlobalContext rather than getGlobalContext

Ready to issue a PR, one issue for others to decide I think but the r…

29ae917

…est should be pretty much ready to go

first offline attempt at fixing the cmake list

fdc9e2c

Adding in Will's fix, I want to test this out a bit

ef62cdc

Merged in the Bridged GEP iterator; time to fire up a VM and test the…

e857e0f

…re - just for a clean environment, tired of my wonkiness

ierge branch 'master' into llvm-4

66f62ab

removing a file and verifying compilation after the merge

517d3d3

yuleisui reviewed May 7, 2017

View reviewed changes

jcarlson23 added 3 commits May 7, 2017 07:35

Updating LocMemModel.cpp per Yulei's comments

6cd7aef

Missed one getIndexedType in LocMemModel for updated PR

acca31f

quick typo fix, verified compilation, no testing beyond trivial, upda…

31cffc2

…ting PR with this commit

yuleisui reviewed May 7, 2017

View reviewed changes

Couple quick changes from getIndexedType() back to *gi using the brid…

9ef3981

…gedIterator

yuleisui merged commit 437b2d5 into SVF-tools:master May 7, 2017

yuleisui added a commit that referenced this pull request Aug 30, 2019

Merge pull request #23 from kisslune/master

1a4ca4c

fix LCD

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

LLVM 4 Merge #23

LLVM 4 Merge #23

jcarlson23 commented May 6, 2017

yuleisui commented May 7, 2017

yuleisui commented May 7, 2017

yuleisui left a comment

yuleisui May 7, 2017

yuleisui May 7, 2017

jcarlson23 May 7, 2017

yuleisui left a comment •

edited

Loading

yuleisui May 7, 2017

yuleisui May 7, 2017

yuleisui May 7, 2017

jcarlson23 May 7, 2017

yuleisui commented May 7, 2017

jcarlson23 commented May 7, 2017

LLVM 4 Merge #23

LLVM 4 Merge #23

Conversation

jcarlson23 commented May 6, 2017

yuleisui commented May 7, 2017

yuleisui commented May 7, 2017

yuleisui left a comment

Choose a reason for hiding this comment

yuleisui May 7, 2017

Choose a reason for hiding this comment

yuleisui May 7, 2017

Choose a reason for hiding this comment

jcarlson23 May 7, 2017

Choose a reason for hiding this comment

yuleisui left a comment • edited Loading

Choose a reason for hiding this comment

yuleisui May 7, 2017

Choose a reason for hiding this comment

yuleisui May 7, 2017

Choose a reason for hiding this comment

yuleisui May 7, 2017

Choose a reason for hiding this comment

jcarlson23 May 7, 2017

Choose a reason for hiding this comment

yuleisui commented May 7, 2017

jcarlson23 commented May 7, 2017

yuleisui left a comment •

edited

Loading