diff options
| author | Eric Anholt <[email protected]> | 2018-07-20 14:27:09 -0700 |
|---|---|---|
| committer | Eric Anholt <[email protected]> | 2018-07-23 10:21:43 -0700 |
| commit | 6b73a97f84f86f4c3d3bbbbadf963c20b8e52b57 (patch) | |
| tree | 6c8701d6c84c1478f7e829a08e5bb141777846e5 /src/broadcom/compiler/qpu_schedule.c | |
| parent | 79e0f042bcd6a1956015cf332c2232ade2c2321f (diff) | |
v3d: Implement a small immediates optimization, based on VC4's.
We can do one per instruction, and we have to be careful not to overwrite
raddr_b, but this greatly reduces the pressure on uniform loads
(particularly around ldvpm/stvpm instructions).
total instructions in shared programs: 90768 -> 88220 (-2.81%)
instructions in affected programs: 82711 -> 80163 (-3.08%)
Diffstat (limited to 'src/broadcom/compiler/qpu_schedule.c')
| -rw-r--r-- | src/broadcom/compiler/qpu_schedule.c | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/src/broadcom/compiler/qpu_schedule.c b/src/broadcom/compiler/qpu_schedule.c index af0b9b86b1c..441b6327825 100644 --- a/src/broadcom/compiler/qpu_schedule.c +++ b/src/broadcom/compiler/qpu_schedule.c @@ -670,7 +670,8 @@ qpu_merge_inst(const struct v3d_device_info *devinfo, if (v3d_qpu_uses_mux(b, V3D_QPU_MUX_B)) { if (v3d_qpu_uses_mux(a, V3D_QPU_MUX_B) && - a->raddr_b != b->raddr_b) { + (a->raddr_b != b->raddr_b || + a->sig.small_imm != b->sig.small_imm)) { return false; } merge.raddr_b = b->raddr_b; |