[BUG] LLVM ERROR: out of memory

[chadat23]

Bug description

I had some very basic tutorial sorts of things working, then I tried to get a little fancier, using tensors, and received the below message:

(nnfs) chad@ComputerBot:~/mojo/nnfs$ mojo ch02.mojo
LLVM ERROR: out of memory
Allocation failed
[14384:14384:20240920,235455.687881:ERROR elf_dynamic_array_reader.h:64] tag not found
[14384:14384:20240920,235455.688169:ERROR elf_dynamic_array_reader.h:64] tag not found
[14384:14384:20240920,235455.688755:ERROR elf_dynamic_array_reader.h:64] tag not found
[14384:14384:20240920,235455.688852:ERROR elf_dynamic_array_reader.h:64] tag not found
Please submit a bug report to https://github.com/modularml/mojo/issues and include the crash backtrace along with all the relevant source codes.
Stack dump:
0. Program arguments: mojo ch02.mojo
Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var LLVM_SYMBOLIZER_PATH to point to it):
0 mojo 0x000055d0c093847b
1 mojo 0x000055d0c09363d9
2 mojo 0x000055d0c0938b1d
3 libc.so.6 0x00007fa4280b9050
4 libc.so.6 0x00007fa428107e3c
5 libc.so.6 0x00007fa4280b8fb2 gsignal + 18
6 libc.so.6 0x00007fa4280a3472 abort + 211
7 mojo 0x000055d0c0905df9
8 mojo 0x000055d0c0905e45
9 libAsyncRTRuntimeGlobals.so 0x00007fa4285653c3
10 libAsyncRTRuntimeGlobals.so 0x00007fa428574164
11 libKGENCompilerRTShared.so 0x00007fa424ede389 KGEN_CompilerRT_AlignedAlloc + 137
12 libKGENCompilerRTShared.so 0x00007fa3cf1b7348 KGEN_CompilerRT_AlignedAlloc + 18446744072269697096
mojo crashed!
Please file a bug report.
Aborted (core dumped)

It happened every time I tried to run the code. I don't know what to make of it but it says to submit a bug report so...!

Steps to reproduce

Here's the code:

from tensor import Tensor

def main():
var inputs = SIMD[DType.float64, 4](1, 2, 3, 2.5)
var weights = Tensor[DType.float64](3, 4)
weights = ((0.2, 0.8, -0.5, 1), (0.5, -0.91, 0.26, -0.5), (-0.26, -0.27, 0.17, 0.87))
var biases = SIMD[DType.float64, 4](2, 3, 0.5)

var outputs = SIMD[DType.float64, 4](0, 0, 0)
for i in range(weights.dim(0)):
    for j in range(weights.dim(1)):
        outputs[i] += inputs[i] * weights[i][j]
    outputs[i] += biases[i]

print(outputs)

System information

OS: Debian 12
mojo 24.5.0 (e8aacb95)
modular version: "bash: modular: command not found", but as I said, I did previously compile a few things with mojo after installing things via the curl command from the Getting Started page earlier today

[Moosems]

How much memory does your system have?

[arthurevans]

Are you following a specific tutorial here? It looks like you're adapting https://github.com/kumaraayush67/NNFS from Python?

I don't see a Tensor initializer that will take a tuple of tuples like this. (In which case the compiler should really give a more helpful error than "out of memory.").

But you could initialize a Tensor with a variadic list of values. That would look like this:

`var weights = Tensor[DType.float64](shape, 0.2, 0.8, -0.5, 1, 0.5, -0.91, 0.26, -0.5, -0.26, -0.27, 0.17, 0.87)`

If you copied this Mojo code from somewhere, please let us know so we can get it fixed.

[chadat23]

Only 6 gigs of ram (I wasn't planning on running meaningful modelings on this computer, just learn some language basics)

Motivated by the second comment, I ended up with the below code which runs.

from tensor import Tensor, TensorShape

def main():
var inputs = SIMD[DType.float64, 4](1, 2, 3, 2.5)
var weights = Tensor[DType.float64](
TensorShape(3, 4),
0.2, 0.8, -0.5, 1,
0.5, -0.91, 0.26, -0.5,
-0.26, -0.27, 0.17, 0.87
)
var biases = SIMD[DType.float64, 4](2, 3, 0.5)

var outputs = SIMD[DType.float64, 4](0, 0, 0)
for i in range(weights.dim(0)):
    for j in range(weights.dim(1)):
        outputs[i] += inputs[i] * weights[i][j]
    outputs[i] += biases[i]

print(outputs)