Implement ASR pass for promoting allocatables to non-allocatables #2734

czgdp1807 · 2023-10-26T16:40:04Z

Closes #2656

certik · 2023-10-27T16:00:55Z

Do you see any speedups?

czgdp1807 · 2023-10-27T16:36:58Z

I will check tomorrow.

czgdp1807 · 2023-10-27T17:01:09Z

Do you see any speedups?

Nopes not in dftatom. And there is one thing that I have to take care of here. So I will do that first.

certik · 2023-10-30T11:46:49Z

Thanks! I have a few questions:

If this PR doesn't give any speedup, is there a way to test it?
It is being added as a default pass, correct? Isn't this rather an optimization pass that should only be used with --fast?
Can one construct an example where this pass speeds things up?
If not, is there any other advantage of this pass?

The code in this PR seems ok, I just want to have good testing of this, or understanding what exactly it does and only call it when it makes sense.

czgdp1807 · 2023-10-30T13:25:25Z

Here's one example where it helps in speeding up the code,

program debug
implicit none

integer :: i, s

s = 0
do i = 1, 200000000
    call prg(10, 10, 10, s)
end do

print *, s

contains

subroutine prg(d1, d2, d3, s)
integer, intent(in) :: d1, d2, d3
integer, intent(out) :: s
integer, allocatable :: d(:, :, :)
allocate(d(d1, d2, d3))
s = s + size(d)
end subroutine


end program

Without the pass

(lf) 18:53:01:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 --skip-pass=promote_allocatable_to_nonallocatable -o a.out
(lf) 18:53:05:~/lfortran_project/lfortran % time ./a.out
-1863462912
./a.out  4.28s user 0.01s system 99% cpu 4.287 total

With the pass

(lf) 18:53:40:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 -o a.out
(lf) 18:53:56:~/lfortran_project/lfortran % time ./a.out
-1863462912
./a.out  3.53s user 0.01s system 99% cpu 3.537 total

That's 1.2x speed up. Because you are avoiding one allocate and one deallocate statement.

certik · 2023-10-30T13:39:06Z

Beautiful! I think this is an optimization pass then. Let's only run it for --fast?

certik · 2023-10-30T13:51:08Z

The speedup seems to disappear when you use --fast:

$ lfortran a.f90 --skip-pass=promote_allocatable_to_nonallocatable --fast
-1863462912
(lf) lfortran/lfortran(alloc) $ time ./a.out
-1863462912
./a.out  4.56s user 0.01s system 99% cpu 4.578 total
(lf) lfortran/lfortran(alloc) $ lfortran a.f90  --fast                         
-1863462912
(lf) lfortran/lfortran(alloc) $ time ./a.out          
-1863462912
./a.out  4.58s user 0.01s system 98% cpu 4.651 total

I can reproduce the speedup without --fast.

Further analysis would be needed. However, since this is an optimization pass, we need to find a benchmark that is spedup with --fast enabled. We need to analyze if this slows things down (like for me above). If it slows things down, this must be an optional pass, never to be run by default in any mode.

In Debug mode (the default) I think this should not be run even if it speeds things up.

It might be that our allocatable and non-allocatable arrays are the same fast, since both are allocated on the heap in the same way. However, we still want this pass, since eventually the non-allocatable arrays should be faster, and they provide further optimization possibilities (to be implemented).

So to move forward:

let's test this pass perhaps with an ASR test.
disable this in Debug mode
benchmark in Release mode --- if this slows the runtime down, then disable it as well in Release.

certik · 2023-10-30T16:13:17Z

P.S. I tested this with --dump-all-passes-fortran and it transforms:

subroutine prg(d1, d2, d3, s)
    integer(4), dimension(:, :, :), allocatable :: d
    integer(4), intent(in) :: d1
    integer(4), intent(in) :: d2
    integer(4), intent(in) :: d3
    integer(4), intent(out) :: s
    allocate(d(d1, d2, d3))
    s = s + size(d)
    deallocate(d) ! Implicit deallocate
    deallocate(d) ! Implicit deallocate
end subroutine prg

Into:

subroutine prg(d1, d2, d3, s)
    integer(4), intent(in) :: d1
    integer(4), intent(in) :: d2
    integer(4), intent(in) :: d3
    integer(4), dimension(d1, d2, d3) :: d
    integer(4), intent(out) :: s
    s = s + size(d)
end subroutine prg

So the pass clearly works. One can see that there are two implicit deallocate statements, which is a bug (unrelated to this PR), just one is enough.

Btw, the new ASR->Fortran output is awesome, it's my favorite way of seeing what the ASR passes do (thanks to @Thirumalai-Shaktivel, @khushi-411, @Shaikh-Ubaid, @Pranavchiku and others).

czgdp1807 · 2023-10-31T07:37:33Z

One more advantage of this pass, it helps in avoiding memory leaks wherever possible like,

program debug
implicit none

integer :: i, s

s = 0
call prg(10, 10, 10, s)

print *, s

contains

subroutine prg(d1, d2, d3, s)
integer, intent(in) :: d1, d2, d3
integer, intent(out) :: s
integer, allocatable :: d(:, :, :)
s = 0
do i = 1, 20
    allocate(d(d1, d2, d3))
    s = s + size(d)
end do
end subroutine

end program

With this pass

(lf) 13:06:04:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 --fast -o a.out   
(lf) 13:06:17:~/lfortran_project/lfortran % leaks --atExit -- ./a.out                                           
a.out(3901) MallocStackLogging: could not tag MSL-related memory as no_footprint, so those pages will be included in process footprint - (null)
a.out(3901) MallocStackLogging: recording malloc and VM allocation stacks using lite mode
20000
Process 3901 is not debuggable. Due to security restrictions, leaks can only show or save contents of readonly memory of restricted processes.

Process:         a.out [3901]
Path:            /Users/USER/*/a.out
Load Address:    0x100f1c000
Identifier:      a.out
Version:         0
Code Type:       ARM64
Platform:        macOS
Parent Process:  leaks [3900]

Date/Time:       2023-10-31 13:06:19.833 +0530
Launch Time:     2023-10-31 13:06:19.797 +0530
OS Version:      macOS 14.0 (23A344)
Report Version:  7
Analysis Tool:   /usr/bin/leaks

Physical footprint:         3265K
Physical footprint (peak):  3265K
Idle exit:                  untracked
----

leaks Report Version: 4.0, multi-line stacks
Process 3901: 191 nodes malloced for 15 KB
Process 3901: 0 leaks for 0 total leaked bytes.

Without this ASR pass

(lf) 12:59:40:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 --skip-pass=promote_allocatable_to_nonallocatable --fast -o a.out
(lf) 12:59:48:~/lfortran_project/lfortran % leaks --atExit -- ./a.out
a.out(3822) MallocStackLogging: could not tag MSL-related memory as no_footprint, so those pages will be included in process footprint - (null)
a.out(3822) MallocStackLogging: recording malloc and VM allocation stacks using lite mode
20000
Process 3822 is not debuggable. Due to security restrictions, leaks can only show or save contents of readonly memory of restricted processes.

Process:         a.out [3822]
Path:            /Users/USER/*/a.out
Load Address:    0x102404000
Identifier:      a.out
Version:         0
Code Type:       ARM64
Platform:        macOS
Parent Process:  leaks [3821]

Date/Time:       2023-10-31 12:59:50.290 +0530
Launch Time:     2023-10-31 12:59:50.252 +0530
OS Version:      macOS 14.0 (23A344)
Report Version:  7
Analysis Tool:   /usr/bin/leaks

Physical footprint:         3313K
Physical footprint (peak):  3313K
Idle exit:                  untracked
----

leaks Report Version: 4.0, multi-line stacks
Process 3822: 210 nodes malloced for 91 KB
Process 3822: 19 leaks for 77824 total leaked bytes.

STACK OF 19 INSTANCES OF 'ROOT LEAK: <malloc in _lfortran_malloc>':
2   a.out                                 0x102407eb8 main + 44
1   liblfortran_runtime.0.20.3.dylib        0x1028a3e5c _lfortran_malloc + 24  lfortran_intrinsics.c:1580
0   libsystem_malloc.dylib                0x180b08a44 _malloc_zone_malloc_instrumented_or_legacy + 136 
====
    19 (76.0K) << TOTAL >>
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124008200> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124009200> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12400c600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12400d600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12400e600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12400f600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124010600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124011600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124012600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124013600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124014600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124015600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124016600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124017600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124018600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x124019600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12401a600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12401b600> [4096]
      1 (4.00K) ROOT LEAK: <malloc in _lfortran_malloc 0x12401c600> [4096]



Binary Images:
       0x102404000 -        0x102407fff +a.out (0) <4681AC7F-0AC3-3A87-823E-F4693369BA5E> /Users/*/a.out
       0x102888000 -        0x10288bffb  libLeaksAtExit.dylib (64561.82.2) <79DF57F9-0882-3396-B7CA-B66DDA910C71> /usr/lib/libLeaksAtExit.dylib
       0x10289c000 -        0x1028a7fff +liblfortran_runtime.0.20.3.dylib (0) <7493D1DE-3E9A-3E86-A92A-C11FF3EDB73D> /Users/*/liblfortran_runtime.0.20.3.dylib
       0x1808f8000 -        0x180946f08  libobjc.A.dylib (906) <1C0364B5-D7E3-3170-BAA6-9C28F400E7CC> /usr/lib/libobjc.A.dylib
       0x180947000 -        0x1809da873  dyld (1.0.0 - 1122.1) <FFD8AB66-C9AB-31DF-AB80-3A3DFF367DDD> /usr/lib/dyld
       0x1809db000 -        0x1809dfff8  libsystem_blocks.dylib (90) <A81CDBDB-D1B1-386E-9029-2F6E0F685E3A> /usr/lib/system/libsystem_blocks.dylib
       0x1809e0000 -        0x180a26fff  libxpc.dylib (2679.0.25) <84971772-939B-3130-8908-B7A38851A516> /usr/lib/system/libxpc.dylib
       0x180a27000 -        0x180a41fff  libsystem_trace.dylib (1481.0.12) <4AF0EA7D-E9F4-30E7-914D-FB292EDB7F37> /usr/lib/system/libsystem_trace.dylib
       0x180a42000 -        0x180adefc7  libcorecrypto.dylib (1608.0.18) <B3AC12C0-8ED6-35A2-86C6-0BFA55BFF333> /usr/lib/system/libcorecrypto.dylib
       0x180adf000 -        0x180b15fff  libsystem_malloc.dylib (474.0.13) <00379936-C90E-3D9D-A569-3F97338F1472> /usr/lib/system/libsystem_malloc.dylib
       0x180b16000 -        0x180b5cfff  libdispatch.dylib (1462.0.4) <8790BA20-19EC-3A36-8975-E34382D9747C> /usr/lib/system/libdispatch.dylib
       0x180b5d000 -        0x180b5ffff  libsystem_featureflags.dylib (85) <0C18273A-0DB8-3522-853B-67A2187EEF07> /usr/lib/system/libsystem_featureflags.dylib
       0x180b60000 -        0x180bdeff3  libsystem_c.dylib (1583.0.14) <1B84A7E4-8958-330C-98B8-27D491DFF69E> /usr/lib/system/libsystem_c.dylib
       0x180bdf000 -        0x180c6cff7  libc++.1.dylib (1600.151) <93FBD681-FC78-33A3-AFFE-568702676142> /usr/lib/libc++.1.dylib
       0x180c6d000 -        0x180c84fff  libc++abi.dylib (1600.151) <43940F08-B65E-3888-8CD3-C52398EB1CA1> /usr/lib/libc++abi.dylib
       0x180c85000 -        0x180cbffef  libsystem_kernel.dylib (10002.1.13) <A7D3C07D-0A1E-3C4C-8FBA-66905E16BF99> /usr/lib/system/libsystem_kernel.dylib
       0x180cc0000 -        0x180cccff3  libsystem_pthread.dylib (519) <E4DEBB6E-421D-33D0-9E17-77AE0E0FE4DC> /usr/lib/system/libsystem_pthread.dylib
       0x180ccd000 -        0x180cf1fff  libdyld.dylib (1122.1) <4BB77515-DBA8-3EDF-9AF7-3C9EAE959EA6> /usr/lib/system/libdyld.dylib
       0x180cf2000 -        0x180cf8ffb  libsystem_platform.dylib (306.0.1) <E7E2F10A-772A-397F-BD19-A7EA5A54E49F> /usr/lib/system/libsystem_platform.dylib
       0x180cf9000 -        0x180d25ffb  libsystem_info.dylib (583.0.1) <8F2A7530-B8D3-3705-B2B7-9347D9495223> /usr/lib/system/libsystem_info.dylib
       0x184069000 -        0x184073ff7  libsystem_darwin.dylib (1583.0.14) <67509907-7BCE-3723-A305-0B042A41FDA5> /usr/lib/system/libsystem_darwin.dylib
       0x1844db000 -        0x1844ebfff  libsystem_notify.dylib (317) <7DA7113E-E87A-3E78-AD4F-77F6EC076E42> /usr/lib/system/libsystem_notify.dylib
       0x1862fa000 -        0x186313ff7  libsystem_networkextension.dylib (1834.0.1.0.1) <9DBF57DB-0008-3584-89A3-1463AB9E4967> /usr/lib/system/libsystem_networkextension.dylib
       0x18638a000 -        0x1863a1fff  libsystem_asl.dylib (398) <2EA1DA8E-A94D-32EA-A64F-DD586FE72599> /usr/lib/system/libsystem_asl.dylib
       0x187d3d000 -        0x187d45ff3  libsystem_symptoms.dylib (1848.0.5) <2D7D536C-1D08-304A-9E67-BFB1E99301FE> /usr/lib/system/libsystem_symptoms.dylib
       0x18ad89000 -        0x18adb2ffb  libsystem_containermanager.dylib (582.0.8) <8E716F51-A02E-3005-9F6F-84163BFAF1AE> /usr/lib/system/libsystem_containermanager.dylib
       0x18bcf9000 -        0x18bcfdfff  libsystem_configuration.dylib (1296.0.1) <5E85DB71-174E-3E61-A3C9-36D5E0CBBF46> /usr/lib/system/libsystem_configuration.dylib
       0x18bcfe000 -        0x18bd03ff3  libsystem_sandbox.dylib (2169.0.12) <3BC895C5-B10E-39F4-B761-E79D5719FB1D> /usr/lib/system/libsystem_sandbox.dylib
       0x18ca17000 -        0x18ca19ffb  libquarantine.dylib (169) <3011123B-C35C-35FC-AB0B-8D775C9109D1> /usr/lib/system/libquarantine.dylib
       0x18d11c000 -        0x18d121ffb  libsystem_coreservices.dylib (152) <0ECFD7D1-0230-3C32-BDD7-4715D94A28E6> /usr/lib/system/libsystem_coreservices.dylib
       0x18d43c000 -        0x18d472ff3  libsystem_m.dylib (3252) <72899DE6-A120-389C-90DC-B94C392EF655> /usr/lib/system/libsystem_m.dylib
       0x18d477000 -        0x18d47effb  libmacho.dylib (1009) <1A7038EC-EE49-35AE-8A3C-C311083795FB> /usr/lib/system/libmacho.dylib
       0x18d49f000 -        0x18d4acff7  libcommonCrypto.dylib (600025) <BFDF8F55-D3DC-3A92-B8A1-8EF165A56F1B> /usr/lib/system/libcommonCrypto.dylib
       0x18d4ad000 -        0x18d4b7fff  libunwind.dylib (1600.112) <FB2051CF-36B6-348F-B8DF-5B2F702F6290> /usr/lib/system/libunwind.dylib
       0x18d4b8000 -        0x18d4bffff  liboah.dylib (312) <5A83FBC3-0944-3D77-8DF2-03E262F67AFA> /usr/lib/liboah.dylib
       0x18d4c0000 -        0x18d4c9ff3  libcopyfile.dylib (196) <65612C42-C5E4-3821-B71D-DDE620FB014C> /usr/lib/system/libcopyfile.dylib
       0x18d4ca000 -        0x18d4cdfff  libcompiler_rt.dylib (103.1) <B29A99B2-7ADE-3371-A774-B690BEC3C406> /usr/lib/system/libcompiler_rt.dylib
       0x18d4ce000 -        0x18d4d2ffb  libsystem_collections.dylib (1583.0.14) <FA0C6F96-A737-3F29-B640-99E313873A21> /usr/lib/system/libsystem_collections.dylib
       0x18d4d3000 -        0x18d4d5ffb  libsystem_secinit.dylib (139) <1C92CD80-54E6-3E70-BB3C-7264A160AD7C> /usr/lib/system/libsystem_secinit.dylib
       0x18d4d6000 -        0x18d4d8ffb  libremovefile.dylib (70) <CE63E895-25E4-350C-BB79-0EFD3AC899E5> /usr/lib/system/libremovefile.dylib
       0x18d4d9000 -        0x18d4d9ffb  libkeymgr.dylib (31) <F7CE9486-FFF5-3CB8-B26F-75811EF4283A> /usr/lib/system/libkeymgr.dylib
       0x18d4da000 -        0x18d4e2fff  libsystem_dnssd.dylib (2200.0.8) <82426BAB-A0A9-3C99-8A86-DABB3A0BA072> /usr/lib/system/libsystem_dnssd.dylib
       0x18d4e3000 -        0x18d4e8fff  libcache.dylib (92) <C683623C-1FF6-3133-9E28-28672FDBA4D3> /usr/lib/system/libcache.dylib
       0x18d4e9000 -        0x18d4eafff  libSystem.B.dylib (1336) <F0A54B2D-8751-35F1-A3CF-F1A02F842211> /usr/lib/libSystem.B.dylib
       0x1a565f000 -        0x1a566aff7  com.apple.MallocStackLogging (1.0 - 64561.91.2) <A193B4F2-B10B-3785-AF87-9E17FD9E2999> /System/Library/PrivateFrameworks/MallocStackLogging.framework/Versions/A/MallocStackLogging
       0x220087000 -        0x22008afff  libsystem_darwindirectory.dylib (86.0.2) <F1E52C1A-7156-3BED-AD5B-1CE3AFFDFCE3> /usr/lib/system/libsystem_darwindirectory.dylib

Speed Up

(lf) 14:08:20:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 --fast -o a.out
(lf) 14:08:28:~/lfortran_project/lfortran % time ./a.out
-1863462912
./a.out  0.00s user 0.00s system 74% cpu 0.007 total
(lf) 14:08:31:~/lfortran_project/lfortran % lfortran /Users/czgdp1807/lfortran_project/debug.f90 --skip-pass=promote_allocatable_to_nonallocatable --fast -o a.out
(lf) 14:08:49:~/lfortran_project/lfortran % time ./a.out
-1863462912
./a.out  2.42s user 1.03s system 93% cpu 3.679 total

czgdp1807 · 2023-10-31T08:48:26Z

I will add the above example as a test. What do you say @certik?

certik · 2023-10-31T11:53:23Z

The code:

do i = 1, 20
    allocate(d(d1, d2, d3))
    s = s + size(d)
end do

should not leak with or without this pass.

certik · 2023-10-31T11:55:22Z

I will add the above example as a test. What do you say @certik?

The test is quite slow (several seconds), so I would not add it, unless it runs "immediately".

See this comment #2734 (comment) what needs to be done to finish the PR.

certik · 2023-10-31T17:46:09Z

TODO:

Remove from Debug
Add a test (that runs quickly), the one above where we get large speedup thanks to this pass.

czgdp1807 · 2023-11-01T07:15:12Z

Add a test (that runs quickly), the one above where we get large speedup thanks to this pass.

Actually if we add the test then we will be able to run it only in Release mode. Now I am not sure what to do here. I am thinking to keep this pass in both FAST and NOFAST mode but only in Release and it will not be there in Debug at all. And in testing the test will be run only in Release mode.

certik · 2023-11-01T12:03:27Z

Actually if we add the test then we will be able to run it only in Release mode.

How so? The test should run in all modes. The ASR pass introduced in this PR will only run in Release mode. Run the test in both Debug and Release.

czgdp1807 · 2023-11-01T12:06:54Z

Well, I mean then I will have to reduce the size of the test. Then we will never now if the test is actually working or not. If I keep the size same then it will take longer in debug mode.

certik · 2023-11-01T12:13:18Z

As we discussed, the test should be small. Indeed, we will not know from just running it that a given performance pass was applied. To know that for sure, you would need to add an ASR check. We can. But as we discussed, it might not be worth it.

czgdp1807 · 2023-11-01T17:22:21Z

@certik This is ready for review.

czgdp1807 · 2023-11-01T17:22:38Z

However please do not merge. I will clean up the git history and merge.

certik · 2023-11-01T20:29:14Z

integration_tests/allocate_16.f90

+integer, intent(out) :: s
+integer, allocatable :: d(:, :, :)
+s = 0
+do i = 1, 200000


In Debug mode this runs in 21ms, which I think is fine.

time ./b.out 200000000 ./b.out 0.01s user 0.00s system 84% cpu 0.021 total

So we can keep it. If needed, we can further lower this number.

Okay. But it segfaults on my mac at this number.

Oh, than that's a bug that we need to fix?

I have to complete it so I am using a smaller number for now. I will fix the segmentation fault later because I have to complete the sync for now.

certik · 2023-11-01T20:40:44Z

src/libasr/pass/pass_manager.h

+                "insert_deallocate",
+                #if !defined(WITH_LFORTRAN_ASSERT)
+                "promote_allocatable_to_nonallocatable"
+                #endif


Why is this ifdef needed here?

For using it only in Release mode. This condition checks for that. If WITH_LFORTRAN_ASSERT is not defined then its Release mode. I think WITH_LFORTRAN_ASSERT is defined only for Debug mode, correct?

No, that's Release/Debug for the LFortran compiler, not the LFortran runtime. Use the compiler_options.fast for that.

certik

Other than that, I think it is fine.

czgdp1807 · 2023-11-07T18:01:09Z

I will complete it tomorrow.

tests/reference/llvm-allocate_02-7f23768.json

tests/tests.toml

certik

I think that this is fine.

…_of_struct_arrays

certik · 2023-11-14T15:02:10Z

Thank you for finishing it!

czgdp1807 force-pushed the alloc branch 2 times, most recently from 53a61b9 to 1520984 Compare October 27, 2023 15:58

czgdp1807 force-pushed the alloc branch from 1520984 to 9037cdf Compare October 30, 2023 06:28

czgdp1807 marked this pull request as ready for review October 30, 2023 07:05

certik marked this pull request as draft October 30, 2023 13:51

czgdp1807 force-pushed the alloc branch from 9037cdf to 9287754 Compare October 31, 2023 08:50

czgdp1807 force-pushed the alloc branch from 9287754 to dcba2a4 Compare November 1, 2023 10:47

czgdp1807 force-pushed the alloc branch from dcba2a4 to d9444a4 Compare November 1, 2023 17:07

czgdp1807 marked this pull request as ready for review November 1, 2023 17:22

czgdp1807 force-pushed the alloc branch from d9444a4 to 335ccc4 Compare November 1, 2023 20:05

certik reviewed Nov 1, 2023

View reviewed changes

certik approved these changes Nov 1, 2023

View reviewed changes

czgdp1807 force-pushed the alloc branch from 335ccc4 to 6f145df Compare November 8, 2023 08:14

certik reviewed Nov 8, 2023

View reviewed changes

tests/reference/llvm-allocate_02-7f23768.json Outdated Show resolved Hide resolved

certik reviewed Nov 8, 2023

View reviewed changes

tests/tests.toml Show resolved Hide resolved

certik marked this pull request as draft November 8, 2023 12:56

czgdp1807 marked this pull request as ready for review November 13, 2023 15:13

certik approved these changes Nov 13, 2023

View reviewed changes

czgdp1807 added 8 commits November 14, 2023 11:46

TEST: Add test for promoting allocatable to non-allocatable

3ea534c

DEV: Add ASR pass for promoting allocatables to non-allocatables

9b030e8

DEV: Register promote_allocatable_to_nonallocatable ASR pass

4b00749

DEV: Implement logic for PointerToDataArray in allocate_array_members…

9126932

…_of_struct_arrays

DEV: Add getFunctionType in ASRUtils

78cfd2d

MISC: Add pass/promote_allocatable_to_nonallocatable.cpp in build script

f554a94

DEV: Register allocate_16 test

6c76694

TEST: Updated reference tests

cae41f9

czgdp1807 force-pushed the alloc branch from 15518b4 to cae41f9 Compare November 14, 2023 06:24

czgdp1807 enabled auto-merge November 14, 2023 06:24

czgdp1807 merged commit 59d7523 into lfortran:main Nov 14, 2023
20 checks passed

czgdp1807 deleted the alloc branch November 14, 2023 06:53

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement ASR pass for promoting allocatables to non-allocatables #2734

Implement ASR pass for promoting allocatables to non-allocatables #2734

czgdp1807 commented Oct 26, 2023

certik commented Oct 27, 2023

czgdp1807 commented Oct 27, 2023

czgdp1807 commented Oct 27, 2023

certik commented Oct 30, 2023 •

edited

czgdp1807 commented Oct 30, 2023

certik commented Oct 30, 2023

certik commented Oct 30, 2023 •

edited

certik commented Oct 30, 2023 •

edited

czgdp1807 commented Oct 31, 2023 •

edited

czgdp1807 commented Oct 31, 2023

certik commented Oct 31, 2023

certik commented Oct 31, 2023

certik commented Oct 31, 2023

czgdp1807 commented Nov 1, 2023

certik commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

certik commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

certik Nov 1, 2023

czgdp1807 Nov 1, 2023

certik Nov 1, 2023

czgdp1807 Nov 8, 2023

certik Nov 1, 2023

czgdp1807 Nov 1, 2023

certik Nov 1, 2023

czgdp1807 Nov 2, 2023

certik left a comment

czgdp1807 commented Nov 7, 2023

certik left a comment

certik commented Nov 14, 2023

Implement ASR pass for promoting allocatables to non-allocatables #2734

Implement ASR pass for promoting allocatables to non-allocatables #2734

Conversation

czgdp1807 commented Oct 26, 2023

certik commented Oct 27, 2023

czgdp1807 commented Oct 27, 2023

czgdp1807 commented Oct 27, 2023

certik commented Oct 30, 2023 • edited

czgdp1807 commented Oct 30, 2023

certik commented Oct 30, 2023

certik commented Oct 30, 2023 • edited

certik commented Oct 30, 2023 • edited

czgdp1807 commented Oct 31, 2023 • edited

czgdp1807 commented Oct 31, 2023

certik commented Oct 31, 2023

certik commented Oct 31, 2023

certik commented Oct 31, 2023

czgdp1807 commented Nov 1, 2023

certik commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

certik commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

czgdp1807 commented Nov 1, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

certik left a comment

Choose a reason for hiding this comment

czgdp1807 commented Nov 7, 2023

certik left a comment

Choose a reason for hiding this comment

certik commented Nov 14, 2023

certik commented Oct 30, 2023 •

edited

certik commented Oct 30, 2023 •

edited

certik commented Oct 30, 2023 •

edited

czgdp1807 commented Oct 31, 2023 •

edited