Mesa (master): r600/sb: handle scratch mem reads on r600

[Dave Airlie]

Module: Mesa
Branch: master
Commit: 22fc5eff803d14e99e9f86db223ceedcef1b9108
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=22fc5eff803d14e99e9f86db223ceedcef1b9108

Author: Dave Airlie <airlied at redhat.com>
Date:   Tue Feb  6 14:17:46 2018 +1000

r600/sb: handle scratch mem reads on r600

On r600 we use the scratch mem with read/read_ind, in that case
sb should track the rw_gpr as a dst instead of a src.

This stops the whole shader being optimised out.

Signed-off-by: Dave Airlie <airlied at redhat.com>
Reviewed-by: Dave Airlie <airlied at redhat.com>

---

 src/gallium/drivers/r600/sb/sb_bc_finalize.cpp |  9 ++++++++-
 src/gallium/drivers/r600/sb/sb_bc_parser.cpp   | 19 +++++++++++++++----
 2 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp b/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
index e737f3d6cd..5c572938a1 100644
--- a/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
+++ b/src/gallium/drivers/r600/sb/sb_bc_finalize.cpp
@@ -778,8 +778,15 @@ void bc_finalizer::finalize_cf(cf_node* c) {
 		int reg = -1;
 		unsigned mask = 0;
 
+
 		for (unsigned chan = 0; chan < 4; ++chan) {
-			value *v = c->src[chan];
+			value *v;
+			if (ctx.hw_class == HW_CLASS_R600 && c->bc.op == CF_OP_MEM_SCRATCH &&
+			    (c->bc.type == 2 || c->bc.type == 3))
+				v = c->dst[chan];
+			else
+				v = c->src[chan];
+
 			if (!v || v->is_undef())
 				continue;
 
diff --git a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
index 2ec842159b..a7b828268b 100644
--- a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
+++ b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
@@ -832,12 +832,23 @@ int bc_parser::prepare_ir() {
 
 			do {
 
-				c->src.resize(4);
+				if (ctx.hw_class == HW_CLASS_R600 && c->bc.op == CF_OP_MEM_SCRATCH &&
+				    (c->bc.type == 2 || c->bc.type == 3)) {
+					c->dst.resize(4);
+					for(int s = 0; s < 4; ++s) {
+						if (c->bc.comp_mask & (1 << s))
+							c->dst[s] =
+								sh->get_gpr_value(true, c->bc.rw_gpr, s, false);
+					}
+				} else {
+					c->src.resize(4);
 
-				for(int s = 0; s < 4; ++s) {
-					if (c->bc.comp_mask & (1 << s))
-						c->src[s] =
+				
+					for(int s = 0; s < 4; ++s) {
+						if (c->bc.comp_mask & (1 << s))
+							c->src[s] =
 								sh->get_gpr_value(true, c->bc.rw_gpr, s, false);
+					}
 				}
 
 				if (((flags & CF_RAT) || (!(flags & CF_STRM))) && (c->bc.type & 1)) { // indexed write