Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Rework GPU runtime system and copies #1991

Merged
merged 121 commits into from
Aug 8, 2023
Merged

Rework GPU runtime system and copies #1991

merged 121 commits into from
Aug 8, 2023

Commits on Jul 21, 2023

  1. Remove permutations from LMADs.

    @athas
    athas committed Jul 21, 2023
    Configuration menu
    Copy the full SHA
    a47a1ad View commit details
    Browse the repository at this point in the history

  2. Fix this.

    @athas
    athas committed Jul 21, 2023
    Configuration menu
    Copy the full SHA
    59ca859 View commit details
    Browse the repository at this point in the history

  3. Detect transpositions properly (sort of).

    @athas
    athas committed Jul 21, 2023
    Configuration menu
    Copy the full SHA
    aaeb3da View commit details
    Browse the repository at this point in the history

  4. Fix unittests.

    @athas
    athas committed Jul 21, 2023
    Configuration menu
    Copy the full SHA
    9418f8e View commit details
    Browse the repository at this point in the history

Commits on Jul 22, 2023

  1. This seems never sound.

    @athas
    athas committed Jul 22, 2023
    Configuration menu
    Copy the full SHA
    876a05e View commit details
    Browse the repository at this point in the history

Commits on Jul 24, 2023

  1. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    bb43fd6 View commit details
    Browse the repository at this point in the history

  2. An ISPC write always has a varying index.

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    d6254cb View commit details
    Browse the repository at this point in the history

  3. Make memory simplification parametric in the RuleBook.

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    100379d View commit details
    Browse the repository at this point in the history

  4. Use inferred memory information when possible.

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    2723a96 View commit details
    Browse the repository at this point in the history

  5. No need for these permutations.

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    d59755b View commit details
    Browse the repository at this point in the history

  6. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Jul 24, 2023
    Configuration menu
    Copy the full SHA
    ac865c5 View commit details
    Browse the repository at this point in the history

Commits on Jul 25, 2023

  1. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    b3a49ef View commit details
    Browse the repository at this point in the history

  2. IxFun/LMAD.rank is useful.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    0e67b72 View commit details
    Browse the repository at this point in the history

  3. New IxFun concept: embedding. 

    Replaces rebasing, I hope.
    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    67bfc54 View commit details
    Browse the repository at this point in the history

  4. Half-baked testing.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    b519de6 View commit details
    Browse the repository at this point in the history

  5. Remove rebasing.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    63c1dd3 View commit details
    Browse the repository at this point in the history

  6. More readable this way.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    ee0c3c7 View commit details
    Browse the repository at this point in the history

  7. embed=>expand, and also simplify it a bit.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    1fb3ba8 View commit details
    Browse the repository at this point in the history

  8. I cannot believe this actually works.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    e24d0e6 View commit details
    Browse the repository at this point in the history

  9. Better detection of transpositions. 

    We are actually still too conservative.  A fully dynamic approach is
needed.
    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    7ccfb67 View commit details
    Browse the repository at this point in the history

  10. Just do the same as in the thread case.

    @athas
    athas committed Jul 25, 2023
    Configuration menu
    Copy the full SHA
    ab177f2 View commit details
    Browse the repository at this point in the history

Commits on Jul 26, 2023

  1. We will need this cool function.

    @athas
    athas committed Jul 26, 2023
    Configuration menu
    Copy the full SHA
    5e174c3 View commit details
    Browse the repository at this point in the history

Commits on Jul 27, 2023

  1. Begin rework of copying in codegen. 

    Adds LMADCopy to Imp and implements generic code generation.  Very
slow, but at least functional for the C backend.
    @athas
    athas committed Jul 27, 2023
    Configuration menu
    Copy the full SHA
    f71f4cc View commit details
    Browse the repository at this point in the history

  2. Add mechanism for optimised copy functions.

    @athas
    athas committed Jul 27, 2023
    Configuration menu
    Copy the full SHA
    4672e99 View commit details
    Browse the repository at this point in the history

  3. Correct contiguousness checking.

    @athas
    athas committed Jul 27, 2023
    Configuration menu
    Copy the full SHA
    36f2497 View commit details
    Browse the repository at this point in the history

  4. Start work on GPU copies.

    @athas
    athas committed Jul 27, 2023
    Configuration menu
    Copy the full SHA
    35acf6f View commit details
    Browse the repository at this point in the history

Commits on Jul 28, 2023

  1. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    6cf7b92 View commit details
    Browse the repository at this point in the history

  2. Better handling of cost centre names.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    0db4d43 View commit details
    Browse the repository at this point in the history

  3. Remove unused things.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    2caca00 View commit details
    Browse the repository at this point in the history

  4. Implement optimised transpose for OpenCL.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    4e652a1 View commit details
    Browse the repository at this point in the history

  5. Use distinct CUDA runtime system.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    ccfad48 View commit details
    Browse the repository at this point in the history

  6. Enable large kernel.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    ba632e4 View commit details
    Browse the repository at this point in the history

  7. Placeholder; will get back to this.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    2e47ab9 View commit details
    Browse the repository at this point in the history

  8. Remove ImpCode.copy.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    774dfc6 View commit details
    Browse the repository at this point in the history

  9. Oops.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    f32f2e4 View commit details
    Browse the repository at this point in the history

  10. Also must be flipped.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    5789f24 View commit details
    Browse the repository at this point in the history

  11. Multiply with element size.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    f34d378 View commit details
    Browse the repository at this point in the history

  12. Oops, also this one.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    b7dc0d1 View commit details
    Browse the repository at this point in the history

  13. Almost-general LMAD copying on GPU.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    b7beb50 View commit details
    Browse the repository at this point in the history

  14. Check for empty first.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    0569334 View commit details
    Browse the repository at this point in the history

  15. Fix typo.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    315d93d View commit details
    Browse the repository at this point in the history

  16. Consistent naming.

    @athas
    athas committed Jul 28, 2023
    Configuration menu
    Copy the full SHA
    8c6002d View commit details
    Browse the repository at this point in the history

Commits on Jul 29, 2023

  1. Consistent logging of copies.

    @athas
    athas committed Jul 29, 2023
    Configuration menu
    Copy the full SHA
    9f40ae2 View commit details
    Browse the repository at this point in the history

  2. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Jul 29, 2023
    Configuration menu
    Copy the full SHA
    1ffec5c View commit details
    Browse the repository at this point in the history

  3. Record in cabal file.

    @athas
    athas committed Jul 29, 2023
    Configuration menu
    Copy the full SHA
    c55031e View commit details
    Browse the repository at this point in the history

  4. Correct this.

    @athas
    athas committed Jul 29, 2023
    Configuration menu
    Copy the full SHA
    c6bd83c View commit details
    Browse the repository at this point in the history

Commits on Jul 30, 2023

  1. Consistent transpose logging.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    068ac1b View commit details
    Browse the repository at this point in the history

  2. Minor fixes; still does not catch enough cases.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    595a97b View commit details
    Browse the repository at this point in the history

  3. More appropriate name.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    6cf01c2 View commit details
    Browse the repository at this point in the history

  4. Shorten lines.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    fa416df View commit details
    Browse the repository at this point in the history

  5. Add remaining kernels.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    9580e60 View commit details
    Browse the repository at this point in the history

  6. Start work on generic GPU layer.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    d378c4d View commit details
    Browse the repository at this point in the history

  7. Write transposition generically.

    @athas
    athas committed Jul 30, 2023
    Configuration menu
    Copy the full SHA
    097dc31 View commit details
    Browse the repository at this point in the history

Commits on Jul 31, 2023

  1. Also free builtin kernels.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    575447f View commit details
    Browse the repository at this point in the history

  2. Put portability layer in library.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    52ff34b View commit details
    Browse the repository at this point in the history

  3. CUDA-like local memory management in OpenCL backend.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    d0236ab View commit details
    Browse the repository at this point in the history

  4. Use initialisers instead of assignments.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    844a2ef View commit details
    Browse the repository at this point in the history

  5. Use kernel argument for local memory.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    ef3c877 View commit details
    Browse the repository at this point in the history

  6. Initial work on CUDA support.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    f9572f3 View commit details
    Browse the repository at this point in the history

  7. Fiddle more with CUDA backend.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    d4118c7 View commit details
    Browse the repository at this point in the history

  8. Pass these constants.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    866b6d9 View commit details
    Browse the repository at this point in the history

  9. Also gpu.h.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    c1988c7 View commit details
    Browse the repository at this point in the history

  10. Should be void.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    5438c47 View commit details
    Browse the repository at this point in the history

  11. Actually GPU.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    b97ec6f View commit details
    Browse the repository at this point in the history

  12. Actually this.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    673e7f5 View commit details
    Browse the repository at this point in the history

  13. Actually initialise builtin kernels.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    373ba68 View commit details
    Browse the repository at this point in the history

  14. Shim this for CUDA.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    bd203e5 View commit details
    Browse the repository at this point in the history

  15. Hide this in macro.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    30cc4a3 View commit details
    Browse the repository at this point in the history

  16. Need cast.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    af8a023 View commit details
    Browse the repository at this point in the history

  17. Oops.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    2f1795c View commit details
    Browse the repository at this point in the history

  18. Oops

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    383cefe View commit details
    Browse the repository at this point in the history

  19. Try to restrict transpose group count.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    76ba93e View commit details
    Browse the repository at this point in the history

  20. We also need this wrapper.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    2c33905 View commit details
    Browse the repository at this point in the history

  21. Only limit inner dimensions.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    dfcb049 View commit details
    Browse the repository at this point in the history

  22. Handle enormous kernels.

    @athas
    athas committed Jul 31, 2023
    Configuration menu
    Copy the full SHA
    db66b8d View commit details
    Browse the repository at this point in the history

Commits on Aug 1, 2023

  1. Proper alignment.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    069e17b View commit details
    Browse the repository at this point in the history

  2. Use gpu_ functions for loading kernels.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    11d2a1c View commit details
    Browse the repository at this point in the history

  3. Also use gpu_ loading in CUDA backend.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    59dcb2c View commit details
    Browse the repository at this point in the history

  4. Fix indentation.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    34b2544 View commit details
    Browse the repository at this point in the history

  5. Fix CUDA shared memory, I think.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    c52e19b View commit details
    Browse the repository at this point in the history

  6. I guess it is always aligned in CUDA.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    0d690a2 View commit details
    Browse the repository at this point in the history

  7. CUDA does not have this.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    60a5538 View commit details
    Browse the repository at this point in the history

  8. Oops.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    b1382ac View commit details
    Browse the repository at this point in the history

  9. Move some shared logic into GPU codegen module.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    3c13758 View commit details
    Browse the repository at this point in the history

  10. Centralise even more GPU stuff.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    ec5556a View commit details
    Browse the repository at this point in the history

  11. More centralisation.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    1075b59 View commit details
    Browse the repository at this point in the history

  12. Only used in one spot.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    352c128 View commit details
    Browse the repository at this point in the history

  13. Generic and consistent.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    7afe5c5 View commit details
    Browse the repository at this point in the history

  14. Use OpenCL terminology.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    73b70b1 View commit details
    Browse the repository at this point in the history

  15. More OpenCL terms.

    @athas
    athas committed Aug 1, 2023
    Configuration menu
    Copy the full SHA
    bf9e950 View commit details
    Browse the repository at this point in the history

Commits on Aug 2, 2023

  1. Unify GPU memory allocation.

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    256cfa6 View commit details
    Browse the repository at this point in the history

  2. Unify GPU allocation.

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    2b21310 View commit details
    Browse the repository at this point in the history

  3. Unify GPU scalar read/write.

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    d21a835 View commit details
    Browse the repository at this point in the history

  4. Move this around to be more similar to CUDA.

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    9eef2d1 View commit details
    Browse the repository at this point in the history

  5. Make ISPC backend slightly operational again.

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    6962a9f View commit details
    Browse the repository at this point in the history

  6. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    57ea298 View commit details
    Browse the repository at this point in the history

  7. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Aug 2, 2023
    Configuration menu
    Copy the full SHA
    f448f53 View commit details
    Browse the repository at this point in the history

Commits on Aug 3, 2023

  1. Fix ISPC codegen.

    @athas
    athas committed Aug 3, 2023
    Configuration menu
    Copy the full SHA
    15c0eec View commit details
    Browse the repository at this point in the history

  2. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Aug 3, 2023
    Configuration menu
    Copy the full SHA
    6fa39b5 View commit details
    Browse the repository at this point in the history

  3. Restore basic Python support.

    @athas
    athas committed Aug 3, 2023
    Configuration menu
    Copy the full SHA
    a381d4a View commit details
    Browse the repository at this point in the history

Commits on Aug 4, 2023

  1. Centralise GPU local memory handling.

    @athas
    athas committed Aug 4, 2023
    Configuration menu
    Copy the full SHA
    e5ce2c0 View commit details
    Browse the repository at this point in the history

  2. Implement LMAD copy for sequential Python.

    @athas
    athas committed Aug 4, 2023
    Configuration menu
    Copy the full SHA
    d31df9d View commit details
    Browse the repository at this point in the history

  3. Fix.

    @athas
    athas committed Aug 4, 2023
    Configuration menu
    Copy the full SHA
    93a152f View commit details
    Browse the repository at this point in the history

  4. Fix address calculation.

    @athas
    athas committed Aug 4, 2023
    Configuration menu
    Copy the full SHA
    0539183 View commit details
    Browse the repository at this point in the history

  5. Fix PyOpenCL local memory.

    @athas
    athas committed Aug 4, 2023
    Configuration menu
    Copy the full SHA
    a721bdf View commit details
    Browse the repository at this point in the history

Commits on Aug 7, 2023

  1. Merge branch 'master' into lmad-no-perm

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    ce96ef5 View commit details
    Browse the repository at this point in the history

  2. Use efficient copies in PyOpenCL backend as well.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    1dbddae View commit details
    Browse the repository at this point in the history

  3. Implement efficient host<->gpu copying.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    ae1aef3 View commit details
    Browse the repository at this point in the history

  4. Unify GPU copying.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    231fea3 View commit details
    Browse the repository at this point in the history

  5. No need to export all the guts.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    c14cbbd View commit details
    Browse the repository at this point in the history

  6. Move more stuff out of boilerplate module.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    08d5ee7 View commit details
    Browse the repository at this point in the history

  7. This is unused.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    ebe4148 View commit details
    Browse the repository at this point in the history

  8. Simplify further.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    05841e6 View commit details
    Browse the repository at this point in the history

  9. No need to keep these separate.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    adc3e86 View commit details
    Browse the repository at this point in the history

  10. Eliminate need for boilerplate modules.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    644a49e View commit details
    Browse the repository at this point in the history

  11. Clean up these types.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    28fc9ca View commit details
    Browse the repository at this point in the history

  12. Proper error detection in these API functions.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    1f02851 View commit details
    Browse the repository at this point in the history

  13. Properly ignore these.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    4ae1316 View commit details
    Browse the repository at this point in the history

  14. Oops, these were the CUDA functions.

    @athas
    athas committed Aug 7, 2023
    Configuration menu
    Copy the full SHA
    50ccf11 View commit details
    Browse the repository at this point in the history

Commits on Aug 8, 2023

  1. Manifest simplification rule.

    @athas
    athas committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    ea30bb3 View commit details
    Browse the repository at this point in the history

  2. Fewer allocations needed.

    @athas
    athas committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    5fc50c0 View commit details
    Browse the repository at this point in the history