sparse-llvm outstanding fixes

sparse-llvm outstanding fixes

[Dibyendu Majumdar]

Hi,

This is just a summary of the fixes that are not yet in any patch series:

+ Incorrect generation of GEP offsets due to offset being unsigned.
+ op_fpcast implementation
+ op_setval implementation
+ Fix for variadic function call based upon type information in pseudo values
+ Fix for local variables and simple assignments, including support
for floating type values.

Regards
Dibyendu

Re: sparse-llvm outstanding fixes

[Dibyendu Majumdar]

On 13 March 2017 at 12:08, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> Hi,
>
> This is just a summary of the fixes that are not yet in any patch series:
>
> + Incorrect generation of GEP offsets due to offset being unsigned.
> + op_fpcast implementation
> + op_setval implementation
> + Fix for variadic function call based upon type information in pseudo values
> + Fix for local variables and simple assignments, including support
> for floating type values.
>

Also function ptr call syntax of the type (*f)() results in wrong
linearized output. There is no fix for this yet.

Regards

Re: sparse-llvm outstanding fixes

[Dibyendu Majumdar]

Hi,

I have been reviewing and have selectively merged changes from the
latest v5 LLVM fix series. There are still some issues that are
outstanding:

On 13 March 2017 at 19:04, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
>> This is just a summary of the fixes that are not yet in any patch series:
>>
>> + Incorrect generation of GEP offsets due to offset being unsigned.
>> + Support for local variables is missing (local variables get declared as global)
> + function ptr call syntax of the type (*f)() results in wrong linearized output.

+ Initialization of local objects must zero out the object before
executing instructions to initialize.
+ The simplification of instructions that are of type floating should
be disabled to ensure correct output.
+ The simplification phase generates incorrect code when a variable is
initialized in a branch and the usage is such that variable is
initialized prior to use.
+ The handling of floats is not yet complete in sparse-llvm

I am not able to test the published version of sparse-llvm as it is
missing some of the fixes  are not present.

Thanks and Regards
Dibyendu