diff options
| author | Marek Olšák <[email protected]> | 2019-07-03 22:56:58 -0400 |
|---|---|---|
| committer | Marek Olšák <[email protected]> | 2019-07-09 17:24:16 -0400 |
| commit | 270a8ab6481f584be791ee1a827c12a536070d52 (patch) | |
| tree | 1b08b936f4cabcdb1353d6c895b504057d8a3516 | |
| parent | ab1f36a1d3949542c73f8643f4d68a84fd453f83 (diff) | |
radeonsi/gfx10: launch 2 compute waves per CU before going onto the next CU
Acked-by: Pierre-Eric Pelloux-Prayer <[email protected]>
Acked-by: Dave Airlie <[email protected]>
| -rw-r--r-- | src/gallium/drivers/radeonsi/si_compute.c | 11 |
1 files changed, 9 insertions, 2 deletions
diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c index 6c3509b5226..0989181aba4 100644 --- a/src/gallium/drivers/radeonsi/si_compute.c +++ b/src/gallium/drivers/radeonsi/si_compute.c @@ -810,12 +810,19 @@ static void si_emit_dispatch_packets(struct si_context *sctx, struct si_screen *sscreen = sctx->screen; struct radeon_cmdbuf *cs = sctx->gfx_cs; bool render_cond_bit = sctx->render_cond && !sctx->render_cond_force_off; + unsigned threads_per_threadgroup = + info->block[0] * info->block[1] * info->block[2]; unsigned waves_per_threadgroup = - DIV_ROUND_UP(info->block[0] * info->block[1] * info->block[2], 64); + DIV_ROUND_UP(threads_per_threadgroup, 64); + unsigned threadgroups_per_cu = 1; + + if (sctx->chip_class >= GFX10 && waves_per_threadgroup == 1) + threadgroups_per_cu = 2; radeon_set_sh_reg(cs, R_00B854_COMPUTE_RESOURCE_LIMITS, si_get_compute_resource_limits(sscreen, waves_per_threadgroup, - sctx->cs_max_waves_per_sh, 1)); + sctx->cs_max_waves_per_sh, + threadgroups_per_cu)); unsigned dispatch_initiator = S_00B800_COMPUTE_SHADER_EN(1) | |