Mesa (master): i965/fs: Take the shared mathbox into account in instruction scheduling.
Eric Anholt
anholt at kemper.freedesktop.org
Thu Jan 20 01:36:53 UTC 2011
Module: Mesa
Branch: master
Commit: b41d323c90da47d720c015a18aa90e9efbb689aa
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=b41d323c90da47d720c015a18aa90e9efbb689aa
Author: Eric Anholt <eric at anholt.net>
Date: Tue Jan 18 21:26:53 2011 -0800
i965/fs: Take the shared mathbox into account in instruction scheduling.
I don't have evidence for this amounting to any improvement,
but it does codify a bit more what we understand so far about
the pipeline.
---
.../dri/i965/brw_fs_schedule_instructions.cpp | 15 +++++++++++++++
1 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp
index 00aa99d..c8f0b27 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp
@@ -438,6 +438,21 @@ instruction_scheduler::schedule_instructions(fs_inst *next_block_header)
instructions.push_tail(child);
}
}
+
+ /* Shared resource: the mathbox. There's one per EU (on later
+ * generations, it's even more limited pre-gen6), so if we send
+ * something off to it then the next math isn't going to make
+ * progress until the first is done.
+ */
+ if (chosen->inst->is_math()) {
+ foreach_iter(exec_list_iterator, iter, instructions) {
+ schedule_node *n = (schedule_node *)iter.get();
+
+ if (n->inst->is_math())
+ n->unblocked_time = MAX2(n->unblocked_time,
+ time + chosen->latency);
+ }
+ }
}
assert(instructions_to_schedule == 0);
More information about the mesa-commit
mailing list