Commit Graph

486 Commits

Author SHA1 Message Date
David Mak 875d534de4 ast: Use `{filename}:{row}:{col}` for location output 2023-12-08 15:48:54 +08:00
David Mak 5c5620692f core: Add np_{round,floor,ceil}
These functions are NumPy variants of round/floor/ceil, which returns
floats instead of ints.
2023-11-23 13:45:07 +08:00
David Mak 0af1e37e99 core: Prefix all NumPy/SciPy functions with np_/sp_spec 2023-11-23 13:35:23 +08:00
David Mak 10538b5296 core: Update insta snapshots 2023-11-09 13:00:27 +08:00
David Mak d322c91697 core: Change bitshift operators to accept int32/uint32 for RHS operand 2023-11-09 12:16:20 +08:00
David Mak 3231eb0d78 core: Add compile-time error and runtime assertion for negative shifts 2023-11-09 12:16:20 +08:00
David Mak 08a5050f9a core: Implement non-trivial builtin functions using IRRT 2023-11-06 12:57:23 +08:00
David Mak 0a84f7ac31 Add CodeGenerator::gen_block and refactor to use it 2023-11-04 13:42:44 +08:00
David Mak fd787ca3f5 core: Remove trunc
The behavior of trunc is already implemented by casts and is therefore
redundant.
2023-11-04 13:35:53 +08:00
David Mak 4dbe07a0c0 core: Revert breaking changes to round-family functions
These functions should return ints as the math.* functions do instead of
following the convention of numpy.* functions.
2023-11-04 13:35:53 +08:00
David Mak 2e055e8ab1 core: Replace rint implementation with LLVM intrinsic 2023-11-04 13:35:53 +08:00
David Mak c6b9aefe00 core: Fix int32-to-uint64 conversion
This conversion should be sign-extended.
2023-11-03 16:24:26 +08:00
David Mak 8ad09748d0 core: Fix conversion from float to unsigned types
These conversions also need to wraparound.
2023-11-03 16:24:26 +08:00
David Mak 7a5a2db842 core: Fix handling of float-to-int32 casts
Out-of-bound conversions should be wrapped around.
2023-11-03 16:24:26 +08:00
David Mak 92d6f0a5d3 core: Implement bitwise not for unsigned ints and fix implementation 2023-11-03 16:24:26 +08:00
David Mak ff1fed112c core: Rework gamma/gammaln to match SciPy behavior
Matches behavior for infinities and NaNs.
2023-11-01 18:03:29 +08:00
David Mak 36a6a7b8cd core: Replace TopLevelDef comments with documentation 2023-11-01 18:03:29 +08:00
David Mak 2b635a0b97 core: Implement numpy and scipy functions 2023-11-01 18:03:29 +08:00
David Mak 60ad100fbb core: Implement and expose {isinf,isnan} 2023-11-01 18:03:29 +08:00
David Mak 7cf7634985 core: Add create_fn_by_* functions
Used for abstracting the creation of function from different sources.
2023-11-01 18:03:29 +08:00
David Mak 068f0d9faf core: Do not cast floor/ceil result to int
NumPy explicitly states that the return type of the floor/ceil is float.
2023-11-01 18:03:29 +08:00
David Mak 95810d4229 core: Remove {ceil64,floor64,round,round64}
These are not present in NumPy or Artiq.
2023-11-01 18:03:29 +08:00
David Mak e95586f61e core: Fix IR generation of `for` loop containing break/continue
Fix cases where the body BB would have two terminators because of a
preceding continue/break statement already emitting a terminator.
2023-11-01 13:21:27 +08:00
David Mak 50230e61f3 core: Simplify loop condition check for list comprehension 2023-10-06 12:24:03 +08:00
David Mak 0205161e35 core: Simplify list creation for comprehension 2023-10-06 12:22:38 +08:00
David Mak a2fce49b26 core: Allocate exceptions at the beginning of function
Only one instance of exception is necessary, as exceptions will always
be initialized before being thrown.
2023-10-06 12:13:20 +08:00
David Mak 60a503a791 core: Allocate more stack variables at the beginning of function
All allocas for temporary objects are now placed in the beginning of the
function. Allocas for on-temporary objects are not modified because
these variables may appear in a loop and thus must be uniquely
allocated by different allocas.
2023-10-06 11:42:47 +08:00
David Mak 0c49b30a90 core: Restore debug info before function call is invoked
Previously, the IR which sets up the call to the target function will
have its debug location pointing at the last argument of the function
call instead of the function call itself.
2023-10-06 11:35:23 +08:00
David Mak c7de22287e core: Fix restoration of stack address
All allocas for temporary objects are now placed in the beginning of the
function. Allocas for on-temporary objects are not modified because
these variables may appear in a loop and thus must be uniquely
represented.
2023-10-06 11:34:23 +08:00
David Mak 1a54aaa1c0 core: Restore debug location when generating allocas
Debug location is lost when moving the builder cursor.
2023-10-06 11:11:50 +08:00
David Mak 480a4bc0ad core: Implement comparison operators for unsigned types 2023-10-05 17:13:10 +08:00
David Mak 2a775d822e core: Demote dead code into a stdout warning 2023-10-04 18:03:25 +08:00
David Mak 48c6498d1f core: Fix restoration of loop target in try statement
old_loop_target is only assigned if ctx.loop_target is overwritten,
meaning that if ctx.loop_target is never overwritten, ctx.loop_target
will always be overwritten to None.

We fix this by only restoring from old_loop_target if we previously
assigned to old_loop_target.
2023-09-28 20:00:02 +08:00
David Mak 2a38d5160e meta: Respect opt flags when performing whole-module optimization 2023-09-28 19:58:54 +08:00
David Mak cb39f61e79 core: Fix passing structure arguments to extern functions
All parameters with a structure type in extern functions are marked as
`byref` instead of `byval`, as most ABIs require the first several
arguments to be passed in registers before spilling into the stack.

`byval` breaks this contract by explicitly requiring all arguments to be
 passed in the stack, breaking interop with libraries written in other
 languages.
2023-09-28 15:02:35 +08:00
David Mak 176f250bdb core: Fix missing conversion to i1 for IfExp 2023-09-28 10:06:40 +08:00
David Mak acdb1de6fe meta: Improve documentation for various modified classes 2023-09-25 15:42:07 +08:00
David Mak 31dcd2dde9 core: Use i8 for boolean variable allocation
In LLVM, i1 represents a 1-byte integer with a single valid bit; The
rest of the 7 upper bits are undefined. This causes problems when
using these variables in memory operations (e.g. memcpy/memmove as
needed by List slicing and assignment).

We fix this by treating all local boolean variables as i8 so that they
are well-defined for memory operations. Function ABIs will continue to
use i1, as memory operations cannot be directly performed on function
arguments or return types, instead they are always converted back into
local boolean variables (which are i8s anyways).

Fixes #315.
2023-09-25 15:42:07 +08:00
David Mak fc93fc2f0e core: Move bitcode verification error message into panic message 2023-09-22 17:16:29 +08:00
David Mak dd42022633 core: Minor refactor allocate_list 2023-09-22 17:16:29 +08:00
David Mak 6dfc43c8b0 core: Add name to build_gep_and_load 2023-09-22 17:16:29 +08:00
David Mak ab2360d7a0 core: Remove emit_llvm from CodeGenLLVMOptions
We instead output an LLVM bitcode file when the option is specified on
the command-line.
2023-09-22 17:16:29 +08:00
David Mak ee1ee4ab3b core: Replace deprecated _ExtInt with _BitInt 2023-09-22 17:16:29 +08:00
David Mak 3e430b9b40 core: Fix missing changes for codegen tests
Apparently the changes were dropped after rebasing.
2023-09-22 17:16:21 +08:00
David Mak 048fcb0a69 core: Switch to LLVM New Pass Manager 2023-09-18 11:35:15 +08:00
David Mak 676d07657a core: Add target field to CodeGenLLVMOptions
For specifying the target machine options when optimizing and linking.

This field is currently unused but will be required in a future
commit.
2023-09-18 09:46:24 +08:00
David Mak 2482a1ef9b core: Add CodeGenTargetMachineOptions
Needed in a future commit.
2023-09-18 09:41:49 +08:00
David Mak 3993a5cf3f core: Add LLVM options to WorkerRegistry 2023-09-12 10:57:05 +08:00
David Mak 39724de598 core: Add CodeGenLLVMOptions
For specifying LLVM options during code generation.
2023-09-12 10:57:04 +08:00
David Mak 4481d48709 core: Use C-style for loop logic for iterables
Index increment is now performed at the end of the loop body.
2023-09-06 20:09:38 +08:00