From 20bbc175a4eea5604357bae6690efa1bb1f37feb Mon Sep 17 00:00:00 2001 From: Jason Ekstrand Date: Sat, 4 May 2019 23:51:23 -0500 Subject: [PATCH] intel/fs/copy-prop: Purge unused ACPs If the destination of an ACP entry exists only within this block, then there's no need to keep it for dataflow analysis. We can delete it from the out_acp table and avoid growing the bitsets any bigger than we absolutely have to. This reduces the maximum number of global ACP entries in the vs-isnan-dvec with software fp64 on Kaby Lake from 8630 to 3942 and takes the execution time of the piglit vs-isnan-dvec test from about 1:16.2 on an unoptimized debug build (what we run in CI) with NIR_VALIDATE=0 to about 56.4 seconds. Reviewed-by: Kenneth Graunke Reviewed-by: Matt Turner --- src/intel/compiler/brw_fs_copy_propagation.cpp | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/src/intel/compiler/brw_fs_copy_propagation.cpp b/src/intel/compiler/brw_fs_copy_propagation.cpp index f3c59d8..ac8cf21 100644 --- a/src/intel/compiler/brw_fs_copy_propagation.cpp +++ b/src/intel/compiler/brw_fs_copy_propagation.cpp @@ -886,6 +886,25 @@ fs_visitor::opt_copy_propagation() foreach_block (block, cfg) { progress = opt_copy_propagation_local(copy_prop_ctx, block, out_acp[block->num]) || progress; + + /* If the destination of an ACP entry exists only within this block, + * then there's no need to keep it for dataflow analysis. We can delete + * it from the out_acp table and avoid growing the bitsets any bigger + * than we absolutely have to. + * + * Because nothing in opt_copy_propagation_local touches the block + * start/end IPs and opt_copy_propagation_local is incapable of + * extending the live range of an ACP destination beyond the block, + * it's safe to use the liveness information in this way. + */ + for (unsigned a = 0; a < ACP_HASH_SIZE; a++) { + foreach_in_list_safe(acp_entry, entry, &out_acp[block->num][a]) { + assert(entry->dst.file == VGRF); + if (block->start_ip <= virtual_grf_start[entry->dst.nr] && + virtual_grf_end[entry->dst.nr] <= block->end_ip) + entry->remove(); + } + } } /* Do dataflow analysis for those available copies. */ -- 2.7.4