I wrote a very simple shader program with a task and mesh shader.
I defined the size of the task work group by:
layout(local_size_x = 32) in;
I launch the task shader with N number of work groups by calling:
glDrawMeshTasksNV(0, N);
I pass the global invocation of the task shader by defining:
taskNV out Task
{
uint instanced_id;
} OUT;
with adding this line in the ‘main’ in the mesh shader:
OUT.instanced_id = gl_GlobalInvocationID.x;
and reading this value in the mesh shader:
int instance = int(IN.instanced_id);
In the mesh shader I render a square according to this value in, a different location for each of the value of instance.
What I get is only N squares, and not (local_size_x - 1) * N.
It seams that the task shader executes only N times, the local thread Id equals 0 and is not changing.
I would expect that is will executes (local_size_x - 1) * N.
To check myself, I called the task shader with
glDrawMeshTasksNV(0, (local_size_x - 1) * N))
, with the task work group set to 1, and I get (local_size_x - 1) * N) squares as expected.
I read the GLSL mesh shader specs, ask BING, and it seems that my assumptions are correct.
I also changed the driver as one of the BING recommendation.
Am I missing something?
Many thanks