Skip to content

GitLab

ir3/a6xx: do not double threadsize if exceeding branchstack limit and make possible to specify branchstack up to 64

Merged Danylo Piliaiev requested to merge Danil/mesa:ir3/a6xx/branch-stack-size into master

On a6xx there is such dependency between branchstack bitfield and the amount of nested ifs, which could be seen with blob:

IFs   BRANCHSTACK
0	0
1	1
2	2
3	2
4	3
5	3
6	4
...
59	30
60	31
61	31
62	32
63	32
64	32

Tested by running shader with switch-case varying the number of cases. Shader started to produce wrong result starting from 65 cases.
The shader was:

layout(local_size_x=128, local_size_y=1, local_size_z=1) in;

layout(binding = 0) buffer block {
            uint values[];
};

void main()
{
    uint x = 0;
    switch (gl_GlobalInvocationID.x) {
        case 0: { x = NUM_CASES; break; }
        case 1: { x = NUM_CASES - 1; break; }
        ...
        case NUM_CASES: { x = 1; break; }
    }
    
    values[gl_GlobalInvocationID.x] = x;
}

When branchstack exceeds the limit - blob does nothing about it. However, the solution would be not doubling the wave size (blob keeps it doubled).

CC: @cwabbott0

Edited by Danylo Piliaiev