Python example in README doesn't work

[ForceBru]

Your first Kompute (Python) says: AttributeError: module 'kp' has no attribute 'Shader'.

Code

The code is the same as in the README - I just imported kp and numpy as np on top. (BTW, it would be nice to be able to copy & paste this code and immediately run it, without adding the imports)

# test_kompute.py
import kp
import numpy as np

def kompute(shader):
    # 1. Create Kompute Manager with default settings (device 0, first queue and no extensions)
    mgr = kp.Manager()

    # 2. Create and initialise Kompute Tensors through manager

    # Default tensor constructor simplifies creation of float values
    tensor_in_a = mgr.tensor([2, 2, 2])
    tensor_in_b = mgr.tensor([1, 2, 3])
    # Explicit type constructor supports uint32, int32, double, float and bool
    tensor_out_a = mgr.tensor_t(np.array([0, 0, 0], dtype=np.uint32))
    tensor_out_b = mgr.tensor_t(np.array([0, 0, 0], dtype=np.uint32))

    params = [tensor_in_a, tensor_in_b, tensor_out_a, tensor_out_b]

    # 3. Create algorithm based on shader (supports buffers & push/spec constants)
    workgroup = (3, 1, 1)
    spec_consts = [2]
    push_consts_a = [2]
    push_consts_b = [3]

    # See documentation shader section for compile_source
    spirv = kp.Shader.compile_source(shader)

    algo = mgr.algorithm(params, spirv, workgroup, spec_consts, push_consts_a)

    # 4. Run operation synchronously using sequence
    (mgr.sequence()
        .record(kp.OpTensorSyncDevice(params))
        .record(kp.OpAlgoDispatch(algo)) # Binds default push consts provided
        .eval() # evaluates the two recorded ops
        .record(kp.OpAlgoDispatch(algo, push_consts_b)) # Overrides push consts
        .eval()) # evaluates only the last recorded op

    # 5. Sync results from the GPU asynchronously
    sq = mgr.sequence()
    sq.eval_async(kp.OpTensorSyncLocal(params))

    # ... Do other work asynchronously whilst GPU finishes

    sq.eval_await()

    # Prints the first output which is: { 4, 8, 12 }
    print(tensor_out_a)
    # Prints the first output which is: { 10, 10, 10 }
    print(tensor_out_b)

if __name__ == "__main__":

    # Define a raw string shader (or use the Kompute tools to compile to SPIRV / C++ header
    # files). This shader shows some of the main components including constants, buffers, etc
    shader = """
        #version 450

        layout (local_size_x = 1) in;

        // The input tensors bind index is relative to index in parameter passed
        layout(set = 0, binding = 0) buffer buf_in_a { float in_a[]; };
        layout(set = 0, binding = 1) buffer buf_in_b { float in_b[]; };
        layout(set = 0, binding = 2) buffer buf_out_a { uint out_a[]; };
        layout(set = 0, binding = 3) buffer buf_out_b { uint out_b[]; };

        // Kompute supports push constants updated on dispatch
        layout(push_constant) uniform PushConstants {
            float val;
        } push_const;

        // Kompute also supports spec constants on initalization
        layout(constant_id = 0) const float const_one = 0;

        void main() {
            uint index = gl_GlobalInvocationID.x;
            out_a[index] += uint( in_a[index] * in_b[index] );
            out_b[index] += uint( const_one * push_const.val );
        }
    """

    kompute(shader)

Running the code

~/test $ python3 test_kompute.py
Traceback (most recent call last):
  File "/Users/forcebru/test/test_kompute.py", line 81, in <module>
    kompute(shader)
  File "/Users/forcebru/test/test_kompute.py", line 26, in kompute
    spirv = kp.Shader.compile_source(shader)
AttributeError: module 'kp' has no attribute 'Shader'
libc++abi.dylib: terminating with uncaught exception of type pybind11::error_already_set: TypeError: 'NoneType' object is not callable
fish: Job 1, 'python3 test_kompute.py' terminated by signal SIGABRT (Abort)
~/test [SIGABRT] $

Side note: terminating with uncaught exception of type pybind11::error_already_set is probably a bug.

A comment in the code says See documentation shader section for compile_source, but I couldn't find this section. Furthermore, the documentation doesn't list Shader under Core Python components. The only documentation entry that mentions the compile_source function is this code, which just says spirv = compile_source(shader) and goes on assuming that the user knows what compile_source is.

I found a function called compile_source in the tests:

kompute/python/test/utils.py

Lines 4 to 7 in c4b3e6e

	def compile_source(source):
	open("tmp_kp_shader.comp", "w").write(source)
	os.system("glslangValidator -V tmp_kp_shader.comp -o tmp_kp_shader.comp.spv")
	return open("tmp_kp_shader.comp.spv", "rb").read()

Is this the function I should be using here?

Versions

• macOS 10.15.7
• NVIDIA GeForce GT 640M
• Vulkan Instance Version: 1.2.189
• Python 3.9.6
• kp 0.8.0

[axsaucedo]

Thank you very much for opening this issue @ForceBru ! This is indeed an error in the documentation, as the code has now been updated to use the util function instead of the internal dependency as per:

kompute/python/test/utils.py

Lines 4 to 7 in c4b3e6e

	def compile_source(source):
	open("tmp_kp_shader.comp", "w").write(source)
	os.system("glslangValidator -V tmp_kp_shader.comp -o tmp_kp_shader.comp.spv")
	return open("tmp_kp_shader.comp.spv", "rb").read()

[axsaucedo]

I've merged now the change - would you be able to test with the updated script?

[ForceBru]

This says:

Traceback (most recent call last):
  File "/Users/forcebru/test/test_kompute.py", line 1, in <module>
    from .utils import compile_source
ImportError: attempted relative import with no known parent package

I just copied the code into /Users/forcebru/test/test_kompute.py, and the utils.py file isn't available there, of course. I manually added the code from python/test/utils.py:

import os

import kp
import numpy as np

def compile_source(source):
    open("tmp_kp_shader.comp", "w").write(source)
    os.system("glslangValidator -V tmp_kp_shader.comp -o tmp_kp_shader.comp.spv")
    return open("tmp_kp_shader.comp.spv", "rb").read()

def kompute(shader):
    # rest of the code

...and got this output:

~/test $ python3 test_kompute.py
tmp_kp_shader.comp
<kp.Tensor object at 0x1279145b0>
<kp.Tensor object at 0x10264f2b0>
~/test $

The comments, however, say:

# Prints the first output which is: { 4, 8, 12 }
print(tensor_out_a)
# Prints the first output which is: { 10, 10, 10 }
print(tensor_out_b)

So, the output is different.

Looks like the correct code should be calling .data, like this:

print(tensor_out_a.data())
print(tensor_out_b.data())

Then the output is:

tmp_kp_shader.comp
[ 4  8 12]
[10 10 10]

---

Why isn't the compile_source function part of the API, though? It seems to be really important, because I can't compile (and thus run) my shaders without it.

[axsaucedo]

Yes that is correct, it's not part of the API, it's only used in the tests. This shouldn't be a blocker as if you do want to perform online compilation of shaders you can use https://github.com/realitix/pyshaderc.

For completeness, this was removed in v0.8.0 as the functionality was very much just for tests and was adding a lot of complexity via the glslang dependency #235

[ForceBru]

Ah, so Kompute doesn't compile shaders, and one should be using either compile_source as a standalone function or the pyshaderc library. Got it, thanks!

[axsaucedo]

That is correct - there's actually some documentation, but may be worth actually extending - I'll add this to have explicitly covered https://kompute.cc/overview/shaders-to-headers.html

[skuep2]

By the way, this is still an issue with the Python Example here:
https://github.com/KomputeProject/kompute/tree/master/examples/python_naive_matmul

Traceback (most recent call last):
  File "/home/developer/Downloads/kompute/benchmark.py", line 56, in <module>
    main()
  File "/home/developer/Downloads/kompute/benchmark.py", line 35, in main
    matmul_op = MatMulOp(mgr)
  File "/home/developer/Downloads/kompute/imp1_naive.py", line 66, in __init__
    self.compiled_shader = kp.Shader.compile_source(self.shader.format(
AttributeError: module 'kp' has no attribute 'Shader'
terminate called after throwing an instance of 'pybind11::error_already_set'
  what():  TypeError: 'NoneType' object is not callable
Aborted (core dumped)

And still, the above problem with the wrong output is also present:

tmp_kp_shader.comp
<kp.Tensor object at 0x7fe60bec0030>
<kp.Tensor object at 0x7fe60bec00b0>

Changing it to print(tensor_out_a.data) outputs

tmp_kp_shader.comp
<bound method PyCapsule.data of <kp.Tensor object at 0x7f4b3612c7f0>>
<bound method PyCapsule.data of <kp.Tensor object at 0x7f4b3612c830>>