feat: Introduce and implement IntrinsicImpureSubroutine
#3650
Conversation
Pranavchiku
added
asr
src/libasr/ASR.asdl
Outdated
| @@ -57,6 +57,7 @@ stmt | |||
| | Stop(expr? code) | |||
| | Assert(expr test, expr? msg) | |||
| | SubroutineCall(symbol name, symbol? original_name, call_arg* args, expr? dt) | |||
| | IntrinsicElementalSubroutine(int intrinsic_id, expr* args, int overload_id) | |||
There was a problem hiding this comment.
In ASR, IntrinsicElemental must be side-effects-free and deterministic. The random number is neither side-effects-free nor deterministic. The only Fortran intrinsic that would fit into IntrinsicElementalSubroutine is mvbits. However, mvbits can also be implemented as a function in ASR:
CALL MVBITS(FROM, FROMPOS, LEN, TO, TOPOS)
Gets represented as:
TO = MVBITS(FROM, FROMPOS, LEN, TO, TOPOS)
Now all arguments to MVBITS function are intent(in), and it can be implemented as IntrinsicElementalFunction.
Let's implement mvbits using IntrinsicElementalFunction and ensure everything works. Once we convince ourselves that there is no issue, then we know we don't need IntrinsicElementalSubroutine.
Regarding a random number, that has to be IntrinsicImpureSubroutine, per #1658
Pranavchiku
force-pushed
the
intrinsic-subroutine
branch
from
1df8bef
to
84993fc
Compare
Pranavchiku
changed the title
IntrinsicElementalSubroutineIntrinsicImpureSubroutine
|
This PR is ready to review. |
|
Question: does this work when we pass an array to |
|
Yes it does. |
|
Ok, it works for an array: https://github.com/lfortran/lfortran/pull/3650/files#diff-e452e55b5ec5613929ce4d9cde98f54efe0d931da016ea716fff121841b69b79R84, specifically, it works for a 1D array. Can you now please add tests for a 2D and 3D array and fix it up to work for them too (if it doesn't)? The PR looks good, so I merged it. The 2D and 3D array extension can be done in a subsequent PR, I think the design does not close the door to it. |
|
It does work for array of every dimension. |
|
https://github.com/lfortran/lfortran/pull/3650/files#diff-c00bf96f361ecbfafda30ce2ddfedf0bd47d741ee3e432f0e395d6fbdc816929R588 takes care by creating do loops recursively. program main
real :: x(5, 6, 7, 8)
call random_number(x)
print *, x
end program% lfortran a.f90
7.82636926e-06 1.31537795e-01 7.55605340e-01 4.58650142e-01 5.32767236e-01 2.18959183e-01 4.70446162e-02 6.78864717e-01
6.79296434e-01 9.34692919e-01 3.83502066e-01 5.19416392e-01 8.30965340e-01 3.45721096e-02 5.34616336e-02 5.29700220e-01
6.71149373e-01 7.69818621e-03 3.83415639e-01 6.68422356e-02 4.17485982e-01 6.86772704e-01 5.88976622e-01 9.30436492e-01
8.46166909e-01 5.26928782e-01 9.19648930e-02 6.53918982e-01 4.15999353e-01 7.01190591e-01 9.10320818e-01 7.62198031e-01
2.62452990e-01 4.74645123e-02 7.36081898e-01 3.28234226e-01 6.32638574e-01 7.56410480e-01 9.91037369e-01 3.65338683e-01
2.47038886e-01 9.82550263e-01 7.22660422e-01 7.53355861e-01 6.51518583e-01 7.26858824e-02 6.31634712e-01 8.84707153e-01
2.72709966e-01 4.36411411e-01 7.66494751e-01 4.77731764e-01 2.37774432e-01 2.74906844e-01 3.59264970e-01 1.66507199e-01
......... |
|
Very good! Can you add a test for this? |
Closes #2591
ASR Definition