lyken
/
nac3
forked from M-Labs/nac3
1
0
Fork 0
Code Pull Requests Activity

core: Implement numpy.matmul for 2D-2D ndarrays

This commit is contained in:
David Mak 2024-04-19 19:00:07 +08:00
parent 5dfcc63978
commit 847615fc2f
5 changed files with 434 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
nac3core/src/codegen/expr.rs
View File

@ -384,7 +384,7 @@ impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
rhs: BasicValueEnum<'ctx>,
) -> BasicValueEnum<'ctx> {
let (BasicValueEnum::FloatValue(lhs), BasicValueEnum::FloatValue(rhs)) = (lhs, rhs) else {
unreachable!()
unreachable!("Expected (FloatValue, FloatValue), got ({}, {})", lhs.get_type(), rhs.get_type())
};
match op {
Operator::Add => self.builder.build_float_add(lhs, rhs, "fadd").map(Into::into).unwrap(),
@ -589,8 +589,9 @@ impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
// even if this assumption is violated, it does not matter as exception unwinding is
// slow anyway...
let cond = call_expect(self, cond, i1_true, Some("expect"));
let current_fun = self.builder.get_insert_block().unwrap().get_parent().unwrap();
let then_block = self.ctx.append_basic_block(current_fun, "succ");
let current_bb = self.builder.get_insert_block().unwrap();
let current_fun = current_bb.get_parent().unwrap();
let then_block = self.ctx.insert_basic_block_after(current_bb, "succ");
let exn_block = self.ctx.append_basic_block(current_fun, "fail");
self.builder.build_conditional_branch(cond, then_block, exn_block).unwrap();
self.builder.position_at_end(exn_block);
@ -1148,27 +1149,45 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
let left_val = NDArrayValue::from_ptr_val(
left_val.into_pointer_value(),
llvm_usize,
None
None,
);
let res = numpy::ndarray_elementwise_binop_impl(
generator,
ctx,
ndarray_dtype1,
if is_aug_assign { Some(left_val) } else { None },
(left_val.as_ptr_value().into(), false),
(right_val, false),
|generator, ctx, (lhs, rhs)| {
gen_binop_expr_with_values(
generator,
ctx,
(&Some(ndarray_dtype1), lhs),
op,
(&Some(ndarray_dtype2), rhs),
ctx.current_loc,
is_aug_assign,
)?.unwrap().to_basic_value_enum(ctx, generator, ndarray_dtype1)
},
)?;
let right_val = NDArrayValue::from_ptr_val(
right_val.into_pointer_value(),
llvm_usize,
None,
);
let res = if *op == Operator::MatMult {
// MatMult is the only binop which is not an elementwise op
numpy::ndarray_matmul_2d(
generator,
ctx,
ndarray_dtype1,
if is_aug_assign { Some(left_val) } else { None },
left_val,
right_val,
)?
} else {
numpy::ndarray_elementwise_binop_impl(
generator,
ctx,
ndarray_dtype1,
if is_aug_assign { Some(left_val) } else { None },
(left_val.as_ptr_value().into(), false),
(right_val.as_ptr_value().into(), false),
|generator, ctx, (lhs, rhs)| {
gen_binop_expr_with_values(
generator,
ctx,
(&Some(ndarray_dtype1), lhs),
op,
(&Some(ndarray_dtype2), rhs),
ctx.current_loc,
is_aug_assign,
)?.unwrap().to_basic_value_enum(ctx, generator, ndarray_dtype1)
},
)?
};
Ok(Some(res.as_ptr_value().into()))
} else {

332
nac3core/src/codegen/numpy.rs
View File

@ -1,9 +1,5 @@
use inkwell::{
IntPredicate,
types::BasicType,
values::{BasicValueEnum, IntValue, PointerValue}
};
use nac3parser::ast::StrRef;
use inkwell::{IntPredicate, OptimizationLevel, types::BasicType, values::{BasicValueEnum, IntValue, PointerValue}};
use nac3parser::ast::{Operator, StrRef};
use crate::{
codegen::{
classes::{
@ -14,17 +10,20 @@ use crate::{
TypedArrayLikeAccessor,
TypedArrayLikeAdapter,
UntypedArrayLikeAccessor,
UntypedArrayLikeMutator,
},
CodeGenContext,
CodeGenerator,
expr::gen_binop_expr_with_values,
irrt::{
call_ndarray_calc_broadcast,
call_ndarray_calc_broadcast_index,
call_ndarray_calc_nd_indices,
call_ndarray_calc_size,
},
llvm_intrinsics::call_memcpy_generic,
stmt::gen_for_callback_incrementing,
llvm_intrinsics,
llvm_intrinsics::{call_memcpy_generic},
stmt::{gen_for_callback_incrementing, gen_if_else_expr_callback},
},
symbol_resolver::ValueEnum,
toplevel::{
@ -86,6 +85,8 @@ fn create_ndarray_dyn_shape<'ctx, 'a, G, V, LenFn, DataFn>(
ctx.current_loc,
);
// TODO: Disallow dim_sz > u32_MAX
Ok(())
},
llvm_usize.const_int(1, false),
@ -171,6 +172,8 @@ fn create_ndarray_const_shape<'ctx, G: CodeGenerator + ?Sized>(
[None, None, None],
ctx.current_loc,
);
// TODO: Disallow dim_sz > u32_MAX
}
let ndarray = generator.gen_var_alloc(
@ -824,6 +827,319 @@ pub fn ndarray_elementwise_binop_impl<'ctx, G, ValueFn>(
Ok(ndarray)
}
/// LLVM-typed implementation for computing matrix multiplication between two 2D `ndarray`s.
///
/// * `elem_ty` - The element type of the `NDArray`.
/// * `res` - The `ndarray` instance to write results into, or [`None`] if the result should be
/// written to a new `ndarray`.
pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
elem_ty: Type,
res: Option<NDArrayValue<'ctx>>,
lhs: NDArrayValue<'ctx>,
rhs: NDArrayValue<'ctx>,
) -> Result<NDArrayValue<'ctx>, String> {
let llvm_i32 = ctx.ctx.i32_type();
let llvm_usize = generator.get_size_type(ctx.ctx);
if cfg!(debug_assertions) {
let lhs_ndims = lhs.load_ndims(ctx);
let rhs_ndims = rhs.load_ndims(ctx);
// lhs.ndims == 2
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
lhs_ndims,
llvm_usize.const_int(2, false),
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
// rhs.ndims == 2
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
rhs_ndims,
llvm_usize.const_int(2, false),
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
if let Some(res) = res {
let res_ndims = res.load_ndims(ctx);
let res_dim0 = unsafe {
res.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
};
let res_dim1 = unsafe {
res.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
};
let lhs_dim0 = unsafe {
lhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
};
let rhs_dim1 = unsafe {
rhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
};
// res.ndims == 2
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
res_ndims,
llvm_usize.const_int(2, false),
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
// res.dims[0] == lhs.dims[0]
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
lhs_dim0,
res_dim0,
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
// res.dims[1] == rhs.dims[0]
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
rhs_dim1,
res_dim1,
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
}
}
if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
let lhs_dim1 = unsafe {
lhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
};
let rhs_dim0 = unsafe {
rhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
};
// lhs.dims[1] == rhs.dims[0]
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
lhs_dim1,
rhs_dim0,
"",
).unwrap(),
"0:ValueError",
"",
[None, None, None],
ctx.current_loc,
);
}
let lhs = if res.is_some_and(|res| res.as_ptr_value() == lhs.as_ptr_value()) {
ndarray_copy_impl(generator, ctx, elem_ty, lhs)?
} else {
lhs
};
let ndarray = res.unwrap_or_else(|| {
create_ndarray_dyn_shape(
generator,
ctx,
elem_ty,
&(lhs, rhs),
|_, _, _| {
Ok(llvm_usize.const_int(2, false))
},
|generator, ctx, (lhs, rhs), idx| {
gen_if_else_expr_callback(
generator,
ctx,
|_, ctx| {
Ok(ctx.builder.build_int_compare(
IntPredicate::EQ,
idx,
llvm_usize.const_zero(),
"",
).unwrap())
},
|generator, ctx| {
Ok(Some(unsafe {
lhs.dim_sizes().get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_zero(),
None,
)
}))
},
|generator, ctx| {
Ok(Some(unsafe {
rhs.dim_sizes().get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(1, false),
None,
)
}))
},
).map(|v| v.map(BasicValueEnum::into_int_value).unwrap())
},
).unwrap()
});
let llvm_ndarray_ty = ctx.get_llvm_type(generator, elem_ty);
ndarray_fill_indexed(
generator,
ctx,
ndarray,
|generator, ctx, idx| {
llvm_intrinsics::call_expect(
ctx,
idx.size(ctx, generator).get_type().const_int(2, false),
idx.size(ctx, generator),
None,
);
let common_dim = {
let lhs_idx1 = unsafe {
lhs.dim_sizes().get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(1, false),
None,
)
};
let rhs_idx0 = unsafe {
rhs.dim_sizes().get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_zero(),
None,
)
};
let idx = llvm_intrinsics::call_expect(ctx, rhs_idx0, lhs_idx1, None);
ctx.builder.build_int_truncate(idx, llvm_i32, "").unwrap()
};
let idx0 = unsafe {
let idx0 = idx.get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_zero(),
None,
);
ctx.builder.build_int_truncate(idx0, llvm_i32, "").unwrap()
};
let idx1 = unsafe {
let idx1 = idx.get_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(1, false),
None,
);
ctx.builder.build_int_truncate(idx1, llvm_i32, "").unwrap()
};
let result_addr = generator.gen_var_alloc(ctx, llvm_ndarray_ty, None)?;
let result_identity = ndarray_zero_value(generator, ctx, elem_ty);
ctx.builder.build_store(result_addr, result_identity).unwrap();
gen_for_callback_incrementing(
generator,
ctx,
llvm_i32.const_zero(),
(common_dim, false),
|generator, ctx, i| {
let i = ctx.builder.build_int_truncate(i, llvm_i32, "").unwrap();
let ab_idx = generator.gen_array_var_alloc(
ctx,
llvm_i32.into(),
llvm_usize.const_int(2, false),
None,
)?;
let a = unsafe {
ab_idx.set_unchecked(ctx, generator, &llvm_usize.const_zero(), idx0.into());
ab_idx.set_unchecked(ctx, generator, &llvm_usize.const_int(1, false), i.into());
lhs.data().get_unchecked(ctx, generator, &ab_idx, None)
};
let b = unsafe {
ab_idx.set_unchecked(ctx, generator, &llvm_usize.const_zero(), i.into());
ab_idx.set_unchecked(ctx, generator, &llvm_usize.const_int(1, false), idx1.into());
rhs.data().get_unchecked(ctx, generator, &ab_idx, None)
};
let a_mul_b = gen_binop_expr_with_values(
generator,
ctx,
(&Some(elem_ty), a),
&Operator::Mult,
(&Some(elem_ty), b),
ctx.current_loc,
false,
)?.unwrap().to_basic_value_enum(ctx, generator, elem_ty)?;
let result = ctx.builder.build_load(result_addr, "").unwrap();
let result = gen_binop_expr_with_values(
generator,
ctx,
(&Some(elem_ty), result),
&Operator::Add,
(&Some(elem_ty), a_mul_b),
ctx.current_loc,
false,
)?.unwrap().to_basic_value_enum(ctx, generator, elem_ty)?;
ctx.builder.build_store(result_addr, result).unwrap();
Ok(())
},
llvm_usize.const_int(1, false),
)?;
let result = ctx.builder.build_load(result_addr, "").unwrap();
Ok(result)
}
)?;
Ok(ndarray)
}
/// Generates LLVM IR for `ndarray.empty`.
pub fn gen_ndarray_empty<'ctx>(
context: &mut CodeGenContext<'ctx, '_>,

19
nac3core/src/codegen/stmt.rs
View File

@ -495,14 +495,14 @@ pub fn gen_for_callback<'ctx, 'a, G, I, InitFn, CondFn, BodyFn, UpdateFn>(
BodyFn: FnOnce(&mut G, &mut CodeGenContext<'ctx, 'a>, I) -> Result<(), String>,
UpdateFn: FnOnce(&mut G, &mut CodeGenContext<'ctx, 'a>, I) -> Result<(), String>,
{
let current = ctx.builder.get_insert_block().and_then(BasicBlock::get_parent).unwrap();
let init_bb = ctx.ctx.append_basic_block(current, "for.init");
let current_bb = ctx.builder.get_insert_block().unwrap();
let init_bb = ctx.ctx.insert_basic_block_after(current_bb, "for.init");
// The BB containing the loop condition check
let cond_bb = ctx.ctx.append_basic_block(current, "for.cond");
let body_bb = ctx.ctx.append_basic_block(current, "for.body");
let cond_bb = ctx.ctx.insert_basic_block_after(init_bb, "for.cond");
let body_bb = ctx.ctx.insert_basic_block_after(cond_bb, "for.body");
// The BB containing the increment expression
let update_bb = ctx.ctx.append_basic_block(current, "for.update");
let cont_bb = ctx.ctx.append_basic_block(current, "for.end");
let update_bb = ctx.ctx.insert_basic_block_after(body_bb, "for.update");
let cont_bb = ctx.ctx.insert_basic_block_after(update_bb, "for.end");
// store loop bb information and restore it later
let loop_bb = ctx.loop_target.replace((update_bb, cont_bb));
@ -719,12 +719,10 @@ pub fn gen_if_else_expr_callback<'ctx, 'a, G, CondFn, ThenFn, ElseFn, R>(
R: BasicValue<'ctx>,
{
let current_bb = ctx.builder.get_insert_block().unwrap();
let current_fn = current_bb.get_parent().unwrap();
let end_bb = ctx.ctx.append_basic_block(current_fn, "if.end");
let then_bb = ctx.ctx.insert_basic_block_after(current_bb, "if.then");
let else_bb = ctx.ctx.insert_basic_block_after(current_bb, "if.else");
let else_bb = ctx.ctx.insert_basic_block_after(then_bb, "if.else");
let end_bb = ctx.ctx.insert_basic_block_after(else_bb, "if.end");
let cond = cond_fn(generator, ctx)?;
assert_eq!(cond.get_type().get_bit_width(), ctx.ctx.bool_type().get_bit_width());
@ -742,6 +740,7 @@ pub fn gen_if_else_expr_callback<'ctx, 'a, G, CondFn, ThenFn, ElseFn, R>(
ctx.builder.build_unconditional_branch(end_bb).unwrap();
}
ctx.builder.position_at_end(end_bb);
let phi = match (then_val, else_val) {
(Some(tv), Some(ev)) => {
let tv_ty = tv.as_basic_value_enum().get_type();

45
nac3core/src/typecheck/magic_methods.rs
View File

@ -291,6 +291,17 @@ pub fn impl_mod(
impl_binop(unifier, store, ty, other_ty, ret_ty, &[Operator::Mod]);
}
/// [Operator::MatMult]
pub fn impl_matmul(
unifier: &mut Unifier,
store: &PrimitiveStore,
ty: Type,
other_ty: &[Type],
ret_ty: Option<Type>,
) {
impl_binop(unifier, store, ty, other_ty, ret_ty, &[Operator::MatMult])
}
/// `UAdd`, `USub`
pub fn impl_sign(unifier: &mut Unifier, _store: &PrimitiveStore, ty: Type, ret_ty: Option<Type>) {
impl_unaryop(unifier, ty, ret_ty, &[Unaryop::UAdd, Unaryop::USub]);
@ -431,7 +442,38 @@ pub fn typeof_binop(
}
}
Operator::MatMult => typeof_ndarray_broadcast(unifier, primitives, lhs, rhs)?,
Operator::MatMult => {
let (_, lhs_ndims) = unpack_ndarray_var_tys(unifier, lhs);
let lhs_ndims = match &*unifier.get_ty_immutable(lhs_ndims) {
TypeEnum::TLiteral { values, .. } => {
assert_eq!(values.len(), 1);
u64::try_from(values[0].clone()).unwrap()
}
_ => unreachable!(),
};
let (_, rhs_ndims) = unpack_ndarray_var_tys(unifier, rhs);
let rhs_ndims = match &*unifier.get_ty_immutable(rhs_ndims) {
TypeEnum::TLiteral { values, .. } => {
assert_eq!(values.len(), 1);
u64::try_from(values[0].clone()).unwrap()
}
_ => unreachable!(),
};
match (lhs_ndims, rhs_ndims) {
(2, 2) => typeof_ndarray_broadcast(unifier, primitives, lhs, rhs)?,
(lhs, rhs) if lhs == 0 || rhs == 0 => {
return Err(format!(
"Input operand {} does not have enough dimensions (has {lhs}, requires {rhs})",
(rhs == 0) as u8
))
}
(lhs, rhs) => {
return Err(format!("ndarray.__matmul__ on {lhs}D and {rhs}D operands not supported"))
}
}
}
Operator::Div => {
if is_left_ndarray || is_right_ndarray {
typeof_ndarray_broadcast(unifier, primitives, lhs, rhs)?
@ -610,6 +652,7 @@ pub fn set_primitives_magic_methods(store: &PrimitiveStore, unifier: &mut Unifie
impl_div(unifier, store, ndarray_t, &[ndarray_t, ndarray_dtype_t], None);
impl_floordiv(unifier, store, ndarray_t, &[ndarray_unsized_t, ndarray_unsized_dtype_t], None);
impl_mod(unifier, store, ndarray_t, &[ndarray_unsized_t, ndarray_unsized_dtype_t], None);
impl_matmul(unifier, store, ndarray_t, &[ndarray_t], Some(ndarray_t));
impl_sign(unifier, store, ndarray_t, Some(ndarray_t));
impl_invert(unifier, store, ndarray_t, Some(ndarray_t));
impl_eq(unifier, store, ndarray_t, &[ndarray_unsized_t, ndarray_unsized_dtype_t], None);

15
nac3standalone/demo/src/ndarray.py
View File

@ -429,6 +429,19 @@ def test_ndarray_ipow_broadcast_scalar():
output_ndarray_float_2(x)
def test_ndarray_matmul():
x = np_identity(2)
y = x @ np_ones([2, 2])
output_ndarray_float_2(x)
output_ndarray_float_2(y)
def test_ndarray_imatmul():
x = np_identity(2)
x @= np_ones([2, 2])
output_ndarray_float_2(x)
def test_ndarray_pos():
x_int32 = np_full([2, 2], -2)
y_int32 = +x_int32
@ -696,6 +709,8 @@ def run() -> int32:
test_ndarray_ipow()
test_ndarray_ipow_broadcast()
test_ndarray_ipow_broadcast_scalar()
test_ndarray_matmul()
test_ndarray_imatmul()
test_ndarray_pos()
test_ndarray_neg()
test_ndarray_inv()