Syntax

Slicing

NumPy-style slicing with start:end:step syntax for accessing tensor elements:

src = Buffer(<4, 8>, int8, __INPUT_MEMORY__);

slice_1 = src[1, :4];      // Row 1, columns 0-3
slice_2 = src[1:3, :];     // Rows 1-2, all columns
slice_3 = src[::2, ::2];   // Every 2nd element

All parameters are optional. Omitted values default to 0 for start, dimension size for end, and 1 for step.

For Loops

Python-like range() with a required carry clause for loop-carried variables:

for i in range(10) carry () {
    Print(i);
};

for i in range(0, 10, 2) carry () {
    Print(i);  // 0, 2, 4, 6, 8
};

a = 1;
b = 1;
for i in range(6) carry (a, b) {
    c = a + b;
    a = b;
    b = c;
};

Loop Unrolling

Apply the @unroll decorator to duplicate the loop body at compile time:

@unroll
for i in range(4) carry () {
    // Body duplicated 4 times
};

All statements end with ; including loop and conditional blocks.

Conditionals

if (x == 1) carry () {
    // then branch
} else {
    // else branch
};

Operators

Comparison

==

Equal

!=

Not equal

<<

Less than

>>

Greater than

<=

Less or equal

>=

Greater or equal

&&

Logical AND

<< and >> are comparison operators, not bit shifts.

Arithmetic

+

Add

-

Subtract

*

Multiply

/

Divide

%

Modulo