On 12/09/2014 07:39 AM, Vadim Girlin wrote:
On 12/09/2014 05:18 AM, Dave Airlie wrote:
On 8 December 2014 at 20:41, Vadim Girlin <vadimgir...@gmail.com> wrote:
On 12/06/2014 07:13 AM, Vadim Girlin wrote:
On 12/04/2014 01:43 AM, Dave Airlie wrote:
Hi Vadim,
I've been looking with Glenn's help into a bug in sb for a couple of
weeks now triggered by a change in how GLSL generates switch
statements.
I understand you probably aren't too interested in r600g but I believe
I'm hitting a design level problem and I would like some advice.
So it appears that GLSL can create loops that don't repeat for switch
statements, and it appears SB wasn't ready to handle such a thing.
Hi, Dave,
I suspect we should rather get rid of such loops somehow, i.e. convert
to something else, the loop that never repeats is not really a loop
anyway. AFAICS "continue" is not supported in switch statements
according to GLSL specs, so the loops generated for switch will
never be
repeated. Am I missing something? Even if repeating is possible
somehow,
at least we can get rid of the loops that are not repeated.
I think loops are less efficient than other control flow
instructions on
r600g hw (at least because they increase stack usage), and possibly on
other hw too.
In fact it seems sb basically gets rid of it already in IR, it just
doesn't know how to translate resulting control flow to ISA, because so
far it only supports specific control flow structure for if-then-else
that was previously preserved during optimizations. I think it may be
not very hard to implement support for that in finalizer, I'll look
into
it.
In fact handling that control flow in finalizer is not as easy as I
hoped,
probably impossible, at least if we want to make it efficient. I forgot
about the limitations of R600 ISA.
OTOH it seems I've managed to fix the issues with loops, the patch is
attached (it's meant to be used instead of 7b0067d2). There are no
piglit
regressions on evergreen, but I didn't test any real apps.
This does seem to fix the problems in piglit, and looks close to what
I was attempting but written by someone who knows what they are doing :-)
What is the sb_sched.cpp change for at the end for?
It fixes those scheduler/regalloc errors for switch tests.
Unfortunately, now I've installed some benchmarks for testing and AFAICS
this patch breaks at least lightsmark 2008, so it seems the condition
removed by the patch was there for a reason.
I'll probably try to come up with better fix.
New patch is attached, the only difference is in the sb_sched.cpp (it
disables copy coalescing for some "unsafe" cases, so it may leave more
MOVs than previously, but I don't think there will be any noticeable
effect on performance).
So far I don't see any problems with it, but I don't have many GL apps
on the test machine. At least lightsmark and unigine demos work for me.
Vadim
Vadim
Dave.
>From d2d16fa39c7b4e871d67e05bad92a540d7e5ea68 Mon Sep 17 00:00:00 2001
From: Vadim Girlin <vadimgir...@gmail.com>
Date: Wed, 10 Dec 2014 14:41:10 +0300
Subject: [PATCH] r600g/sb: fix issues with loops created for switch
---
src/gallium/drivers/r600/sb/sb_bc_finalize.cpp | 2 ++
src/gallium/drivers/r600/sb/sb_bc_parser.cpp | 2 ++
src/gallium/drivers/r600/sb/sb_if_conversion.cpp | 4 ++--
src/gallium/drivers/r600/sb/sb_ir.h | 9 +++++++--
src/gallium/drivers/r600/sb/sb_sched.cpp | 3 +++
5 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp b/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
index f0849ca..3f362c4 100644
--- a/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
+++ b/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
@@ -110,6 +110,8 @@ int bc_finalizer::run() {
void bc_finalizer::finalize_loop(region_node* r) {
+ update_nstack(r);
+
cf_node *loop_start = sh.create_cf(CF_OP_LOOP_START_DX10);
cf_node *loop_end = sh.create_cf(CF_OP_LOOP_END);
diff --git a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
index d787e5b..403f938 100644
--- a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
+++ b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
@@ -758,6 +758,8 @@ int bc_parser::prepare_loop(cf_node* c) {
c->insert_before(reg);
rep->move(c, end->next);
+ reg->src_loop = true;
+
loop_stack.push(reg);
return 0;
}
diff --git a/src/gallium/drivers/r600/sb/sb_if_conversion.cpp b/src/gallium/drivers/r600/sb/sb_if_conversion.cpp
index 93edace..3f2b1b1 100644
--- a/src/gallium/drivers/r600/sb/sb_if_conversion.cpp
+++ b/src/gallium/drivers/r600/sb/sb_if_conversion.cpp
@@ -115,13 +115,13 @@ void if_conversion::convert_kill_instructions(region_node *r,
bool if_conversion::check_and_convert(region_node *r) {
depart_node *nd1 = static_cast<depart_node*>(r->first);
- if (!nd1->is_depart())
+ if (!nd1->is_depart() || nd1->target != r)
return false;
if_node *nif = static_cast<if_node*>(nd1->first);
if (!nif->is_if())
return false;
depart_node *nd2 = static_cast<depart_node*>(nif->first);
- if (!nd2->is_depart())
+ if (!nd2->is_depart() || nd2->target != r)
return false;
value* &em = nif->cond;
diff --git a/src/gallium/drivers/r600/sb/sb_ir.h b/src/gallium/drivers/r600/sb/sb_ir.h
index 85c3d06..711c2eb 100644
--- a/src/gallium/drivers/r600/sb/sb_ir.h
+++ b/src/gallium/drivers/r600/sb/sb_ir.h
@@ -1089,7 +1089,8 @@ typedef std::vector<repeat_node*> repeat_vec;
class region_node : public container_node {
protected:
region_node(unsigned id) : container_node(NT_REGION, NST_LIST), region_id(id),
- loop_phi(), phi(), vars_defined(), departs(), repeats() {}
+ loop_phi(), phi(), vars_defined(), departs(), repeats(), src_loop()
+ {}
public:
unsigned region_id;
@@ -1101,12 +1102,16 @@ public:
depart_vec departs;
repeat_vec repeats;
+ // true if region was created for loop in the parser, sometimes repeat_node
+ // may be optimized away so we need to remember this information
+ bool src_loop;
+
virtual bool accept(vpass &p, bool enter);
unsigned dep_count() { return departs.size(); }
unsigned rep_count() { return repeats.size() + 1; }
- bool is_loop() { return !repeats.empty(); }
+ bool is_loop() { return src_loop || !repeats.empty(); }
container_node* get_entry_code_location() {
node *p = first;
diff --git a/src/gallium/drivers/r600/sb/sb_sched.cpp b/src/gallium/drivers/r600/sb/sb_sched.cpp
index 1413916..4fbdc4f 100644
--- a/src/gallium/drivers/r600/sb/sb_sched.cpp
+++ b/src/gallium/drivers/r600/sb/sb_sched.cpp
@@ -1527,6 +1527,9 @@ bool post_scheduler::check_copy(node *n) {
if (!s->is_prealloc()) {
recolor_local(s);
+
+ if (!s->chunk || s->chunk != d->chunk)
+ return false;
}
if (s->gpr == d->gpr) {
--
2.1.0
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
