Add CUDA Examples/Improve CUDA robustness#202

Open

SousaTrashBin wants to merge 99 commits into

add_some_cuda_examples

Commits on Jul 5, 2026

add gpu decorator (@gpu)

SousaTrashBin
authored and
cursoragent
committed
add some gpu helper functions such as Tensor_length

SousaTrashBin
authored and
cursoragent
committed
add some gpu tests (for now uses CPU but should work nonetheless)

SousaTrashBin
authored and
cursoragent
committed
remove gpu tests, still need to verify if syntax will stay the same or not

SousaTrashBin
authored and
cursoragent
committed
add gpu subset code validation (including recursion checking)

SousaTrashBin
authored and
cursoragent
committed
add verification to types and ops (for now only supports builtins)

SousaTrashBin
authored and
cursoragent
committed
add gpu subset ast representation as well as a way to convert between normal/core ast and gpu ast (not sure if this is the smartest way of doing this)

SousaTrashBin
authored and
cursoragent
committed
not sure why this is needed (reminder to ask)

SousaTrashBin
authored and
cursoragent
committed
add some gpu subset testing (still need to study a bit of llvm syntax before finishing this)

SousaTrashBin
authored and
cursoragent
committed
eventually this will be the file that hosts the conversion between the GPU AST and llvm instructions

SousaTrashBin
authored and
cursoragent
committed
still need to add the generation of the kernel itself

SousaTrashBin
authored and
cursoragent
committed
fix, was using a raw str but should instead use the Name dataclass

SousaTrashBin
authored and
cursoragent
committed
forgot to setup auto ruff

SousaTrashBin
authored and
cursoragent
committed
add llvm decorator

SousaTrashBin
authored and
cursoragent
committed
remove unused files

SousaTrashBin
authored and
cursoragent
committed
add `CPULLVMPipeline` implementation, integrate function compilation and LLVM IR generation

SousaTrashBin
authored and
cursoragent
committed
add gpu decorator (@gpu)

SousaTrashBin
authored and
cursoragent
committed
add some gpu tests (for now uses CPU but should work nonetheless)

SousaTrashBin
authored and
cursoragent
committed
remove gpu tests, still need to verify if syntax will stay the same or not

SousaTrashBin
authored and
cursoragent
committed
add gpu subset code validation (including recursion checking)

SousaTrashBin
authored and
cursoragent
committed
add verification to types and ops (for now only supports builtins)

SousaTrashBin
authored and
cursoragent
committed
add gpu subset ast representation as well as a way to convert between normal/core ast and gpu ast (not sure if this is the smartest way of doing this)

SousaTrashBin
authored and
cursoragent
committed
not sure why this is needed (reminder to ask)

SousaTrashBin
authored and
cursoragent
committed
add some gpu subset testing (still need to study a bit of llvm syntax before finishing this)

SousaTrashBin
authored and
cursoragent
committed
eventually this will be the file that hosts the conversion between the GPU AST and llvm instructions

SousaTrashBin
authored and
cursoragent
committed
still need to add the generation of the kernel itself

SousaTrashBin
authored and
cursoragent
committed
fix, was using a raw str but should instead use the Name dataclass

SousaTrashBin
authored and
cursoragent
committed
forgot to setup auto ruff

SousaTrashBin
authored and
cursoragent
committed
remove unused files

SousaTrashBin
authored and
cursoragent
committed
fix(llvm): only compile @llvm targets; align tests with master

alcides
authored and
cursoragent
committed
feat(llvm): add new llvm backend pipeline class `MultiBackendPipeline` which can be used by multiple backends (assuming they have a LLVM Execution/IrGenerator/Lowerer)

SousaTrashBin
authored and
cursoragent
committed
feat(temp): add way to skip elaboration/type checking just to test operations related with Vectors, should be removed once elaboration is fixed

SousaTrashBin
authored and
cursoragent
committed
add gpu decorator (@gpu)

SousaTrashBin
authored and
cursoragent
committed
add some gpu tests (for now uses CPU but should work nonetheless)

SousaTrashBin
authored and
cursoragent
committed
remove gpu tests, still need to verify if syntax will stay the same or not

SousaTrashBin
authored and
cursoragent
committed
add gpu subset code validation (including recursion checking)

SousaTrashBin
authored and
cursoragent
committed
add verification to types and ops (for now only supports builtins)

SousaTrashBin
authored and
cursoragent
committed
add gpu subset ast representation as well as a way to convert between normal/core ast and gpu ast (not sure if this is the smartest way of doing this)

SousaTrashBin
authored and
cursoragent
committed
not sure why this is needed (reminder to ask)

SousaTrashBin
authored and
cursoragent
committed
add some gpu subset testing (still need to study a bit of llvm syntax before finishing this)

SousaTrashBin
authored and
cursoragent
committed
eventually this will be the file that hosts the conversion between the GPU AST and llvm instructions

SousaTrashBin
authored and
cursoragent
committed
still need to add the generation of the kernel itself

SousaTrashBin
authored and
cursoragent
committed
fix, was using a raw str but should instead use the Name dataclass

SousaTrashBin
authored and
cursoragent
committed
forgot to setup auto ruff

SousaTrashBin
authored and
cursoragent
committed
fix validate method signature; now receives a list of valid function names that it can call which should have all @llvm functions (possibly could have built-ins there as well), also added is_top_le…

SousaTrashBin
authored and
cursoragent
committed
remove unused files

SousaTrashBin
authored and
cursoragent
committed
update some LLVM AST types and terms data, also add some str representations (especially helpful for debugging)

SousaTrashBin
authored and
cursoragent
committed
update last test to represent another step of the verification pipeline (Full Application)

SousaTrashBin
authored and
cursoragent
committed
add another verification step (Full Application);

SousaTrashBin
authored and
cursoragent
committed
fix some type signature bugs

SousaTrashBin
authored and
cursoragent
committed
done some changes to ensure lowering works even with anf partial application (needs further testing, but probably will be easier to create e2e tests)

SousaTrashBin
authored and
cursoragent
committed
ruff linting

SousaTrashBin
authored and
cursoragent
committed
extend `llvm ast` with vector operations, and add new LLVM term representations

SousaTrashBin
authored and
cursoragent
committed
add `CPULLVMPipeline` implementation, integrate function compilation and LLVM IR generation

SousaTrashBin
authored and
cursoragent
committed
add `LLVMCast` term, improve vector operation handling in IR generation and execution

SousaTrashBin
authored and
cursoragent
committed
extend LLVM lowering with vector operation handling and built-in function support, refactor validation steps and type inference methods

SousaTrashBin
authored and
cursoragent
committed
refactor validation and lowering in LLVM backend

SousaTrashBin
authored and
cursoragent
committed
add `to_ir` methods for LLVM types and `accept` methods for LLVM terms to support IR generation and visitor pattern

SousaTrashBin
authored and
cursoragent
committed
refactor according to mypy and ruff

SousaTrashBin
authored and
cursoragent
committed
fix rebase-introduced duplicate LLVM definitions

alcides
authored and
cursoragent
committed
feat(cuda): add some cuda examples, histogram (which is not still working properly) and wordcount based on lusiadas, add montecarlo pi aprox and update busy to better understand if it's working or not

SousaTrashBin
authored and
cursoragent
committed
feat: add cupy-cuda12 as an optional dependency for the gpu/cuda kernel execution

SousaTrashBin
authored and
cursoragent
committed
feat: improve type conversion/casting

SousaTrashBin
authored and
cursoragent
committed
feat: improve/simplify cuda conversion

SousaTrashBin
authored and
cursoragent
committed
feat: add optimization pass to llvm_ir execution

SousaTrashBin
authored and
cursoragent
committed
feat: simplify cuda executor

SousaTrashBin
authored and
cursoragent
committed
feat: add some extra decorator metadata (maybe some are not needed such as cache, need to verify)

SousaTrashBin
authored and
cursoragent
committed
feat: expand decorator metadata to include target, opt_level, and log_ir validations

SousaTrashBin
authored and
cursoragent
committed
feat: fix some minor issues relative to the llvm pipeline

SousaTrashBin
authored and
cursoragent
committed
feat: implement find_calls method for LLVM terms and simplify CUDA conversion logic to fix a kernel duplication bug

SousaTrashBin
authored and
cursoragent
committed
feat(cpu): migrate LLVM decorators to CPU decorators and enhance backend execution/fallback logic

SousaTrashBin
authored and
cursoragent
committed
fix(cuda): add verification to LLVM module before target machine creation

SousaTrashBin
authored and
cursoragent
committed
fix(cuda): specify speed level

SousaTrashBin
authored and
cursoragent
committed
feat(cpu): add `Vector_size` implementation, add header-based size management on vectors

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add some CPU and GPU simple tests that execute on the CI/CD pipeline

SousaTrashBin
authored and
cursoragent
committed
fix: change func name

SousaTrashBin
authored and
cursoragent
committed
fix: remove ';' from imports

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): improve backend execution robustness and error logging

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): fix infinite recursion fallback for CPU/GPU execution

SousaTrashBin
authored and
cursoragent
committed
fix(cpu): prevent redundant block creation in function conversion

SousaTrashBin
authored and
cursoragent
committed
fix(llvm): some fixes according to ruff and mypy

SousaTrashBin
authored and
cursoragent
committed
feat(cpu): add some cpu test examples

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add `LLVMVector` class to handle pointer-based vector operations (still didn't test if this works on gpu)

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add back the gpu busy example

SousaTrashBin
authored and
cursoragent
committed
fix: use full parameter list for partial application resolution in lowerer

authored and
cursoragent
committed
fix: post-rebase fixes for mypy, import syntax, and missing LLVM AST nodes

authored and
cursoragent
committed
fix: register @llvm as backward-compatible alias for @cpu decorator

authored and
cursoragent
committed
fix: add Vector.size to Vector library for evaluator fallback

authored and
cursoragent
committed
feat(llvm): squash cuda/cpu backend development commits

SousaTrashBin
authored and
cursoragent
committed
fix: remove debug print, upgrade fallback logs to warning, add missing LLVMVectorSize alias

authored and
cursoragent
committed
chore: remove dead CPULLVMPipeline (superseded by MultiBackendPipeline)

authored and
cursoragent
committed
refactor(llvm): clean up function names, make `invoke` function cleaner by extracting multiple implementation steps to simpler individual functions

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add debug LLVM IR/PTX file writing

SousaTrashBin
authored and
cursoragent
committed
feat(core): add `Vector` type support and remove GPU/Tensor-related code

SousaTrashBin
authored and
cursoragent
committed
refactor(core): remove unused `Vector` type definition (now it's a built-in)

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): improve type validation and resolution

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add raw vector runtime helpers

SousaTrashBin
authored and
cursoragent
committed
refactor(llvm): simplify backend metadata

SousaTrashBin
authored and
cursoragent
committed
feat(llvm): add planned CUDA vector execution

SousaTrashBin
authored and
cursoragent
committed