freedreno/a6xx: Fix SP_HS_UNKNOWN_A831 value and document it
It appears that storage for varyings in a wave has an upper
limit of wavesize * max_a831 where max_a831 is 64.
Exceeding the limit seam to force gpu to reduce primitives
processed per wave, at least calculations make sense with
such interpretation.
With blob SP_HS_UNKNOWN_A831 never exceeds 64 and setting
it to 65 in freedreno leads to a hang.
On A630 tests (patch_size=3 + gl_Position + array of vec4)
have shown such relation:
| Num of vec4 | A831 | PC_HS_INPUT_SIZE |
|-------------|------|------------------|
| 1 | 0x10 | 0xc |
| 2 | 0x14 | 0xf |
| 3 | 0x18 | 0x12 |
| 4 | 0x1c | 0x15 |
| 5 | 0x20 | 0x18 |
| 6 | 0x24 | 0x1b |
| 7 | 0x28 | 0x1e |
| 8 | 0x2c | 0x21 |
| 9 | 0x30 | 0x24 |
| 10 | 0x34 | 0x27 |
| 11 | 0x38 | 0x2a |
| 12 | 0x3c | 0x2d |
| 13 | 0x3f | 0x30 |
| 14 | 0x40 | 0x33 |
| 15 | 0x3d | 0x36 |
| 16 | 0x3d | 0x39 |
| 17 | 0x40 | 0x3c |
| 18 | 0x3f | 0x3f |
| 19 | 0x3e | 0x42 |
| 20 | 0x3d | 0x45 |
| 21 | 0x3f | 0x48 |
| 22 | 0x3d | 0x4b |
| 23 | 0x40 | 0x4e |
| 24 | 0x3d | 0x51 |
| 25 | 0x3f | 0x54 |
| 26 | 0x3c | 0x57 |
| 27 | 0x3e | 0x5a |
| 28 | 0x40 | 0x5d |
| 29 | 0x3c | 0x60 |
| 30 | 0x3e | 0x63 |
| 31 | 0x40 | 0x66 |
|-------------|------|------------------|
Brief tests with high patch sizes also confirm that formula
matches blob behaviour.
A831 is not a limit for storage available for one thread, so
naming it as SP_HS_WAVE_INPUT_SIZE would make more sense.
Fixes:
47e2c195 "freedreno/a6xx: Program state for tessellation stages"
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7917>
projects / platform / upstream / mesa.git / commit