r600/sfn: Fix copy-prop with array access

We will have to check whether there is access to an array between the
instructions involved with the copy prop, so for now do not allow it.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21347>

diff --git a/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp b/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
index a265914..6d6dd93 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
+++ b/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
@@ -417,6 +417,9 @@ AluInstr::can_propagate_dest() const
    if (!src_reg->has_flag(Register::ssa))
       return false;
 
+   if (!m_dest->has_flag(Register::ssa))
+      return false;
+
    if (src_reg->pin() == pin_chan)
       return m_dest->pin() == pin_none || m_dest->pin() == pin_free ||
              ((m_dest->pin() == pin_chan || m_dest->pin() == pin_group) &&
@@ -472,10 +475,10 @@ bool AluInstr::can_replace_source(PRegister old_src, PVirtualValue new_src)
    if (!check_readport_validation(old_src, new_src))
       return false;
 
-   /* If the old source is an array element, we assume that there
+   /* If the old or new source is an array element, we assume that there
     * might have been an (untracked) indirect access, so don't replace
     * this source */
-   if (old_src->pin() == pin_array)
+   if (old_src->pin() == pin_array || new_src->pin() == pin_array)
       return false;
 
    auto [addr, dummy, index] = indirect_addr();

diff --git a/src/gallium/drivers/r600/sfn/tests/sfn_test_shaders.cpp b/src/gallium/drivers/r600/sfn/tests/sfn_test_shaders.cpp
index aa43985..a32e0be 100644 (file)
--- a/src/gallium/drivers/r600/sfn/tests/sfn_test_shaders.cpp
+++ b/src/gallium/drivers/r600/sfn/tests/sfn_test_shaders.cpp
@@ -1912,27 +1912,27 @@ ELSE
      ALU MOV A2[S37.x].x : I[0] {W}
      ALU MOV A2[S37.x].y : L[0x3dcccccd] {WL}
 ENDIF
+ALU MOV S1025.x@group{s} : A2[0].x {W}
+ALU MOV S1025.y@group{s} : A2[0].y {WL}
+ALU MOV S1025.z@group{s} : A2[1].x {W}
+ALU MOV S1025.w@group{s} : A2[1].y {WL}
+ALU MOV S1027.x@group{s} : A2[2].x {W}
+ALU MOV S1027.y@group{s} : A2[2].y {WL}
+ALU MOV S1027.z@group{s} : A2[3].x {W}
+ALU MOV S1027.w@group{s} : A2[3].y {WL}
+ALU MOV S1029.x@group{s} : A2[0].z {W}
+ALU MOV S1029.y@group{s} : A2[0].w {WL}
+ALU MOV S1029.z@group{s} : A2[1].z {W}
+ALU MOV S1029.w@group{s} : A2[1].w {WL}
+ALU MOV S1031.x@group{s} : A2[2].z {W}
+ALU MOV S1031.y@group{s} : A2[2].w {WL}
+ALU MOV S1031.z@group{s} : A2[3].z {W}
+ALU MOV S1031.w@group{s} : A2[3].w {WL}
 EXPORT_DONE POS 0 S19.xyzw
-ALU MOV S46.x@group : A2[0].x {W}
-ALU MOV S46.y@group : A2[0].y {W}
-ALU MOV S46.z@group : A2[1].x {W}
-ALU MOV S46.w@group : A2[1].y {WL}
-EXPORT PARAM 0 S46.xyzw
-ALU MOV S47.x@group : A2[2].x {W}
-ALU MOV S47.y@group : A2[2].y {W}
-ALU MOV S47.z@group : A2[3].x {W}
-ALU MOV S47.w@group : A2[3].y {WL}
-EXPORT PARAM 1 S47.xyzw
-ALU MOV S48.x@group : A2[0].z {W}
-ALU MOV S48.y@group : A2[0].w {W}
-ALU MOV S48.z@group : A2[1].z {W}
-ALU MOV S48.w@group : A2[1].w {WL}
-EXPORT PARAM 2 S48.xyzw
-ALU MOV S49.x@group : A2[2].z {W}
-ALU MOV S49.y@group : A2[2].w {W}
-ALU MOV S49.z@group : A2[3].z {W}
-ALU MOV S49.w@group : A2[3].w {WL}
-EXPORT_DONE PARAM 3 S49.xyzw
+EXPORT PARAM 0 S1025.xyzw
+EXPORT PARAM 1 S1027.xyzw
+EXPORT PARAM 2 S1029.xyzw
+EXPORT_DONE PARAM 3 S1031.xyzw
 )";
 
 const char *shader_with_dest_array_opt_scheduled =
@@ -2021,38 +2021,38 @@ ELSE
     ALU MOV A2[AR].y : L[0x3dcccccd] {WL}
   ALU_GROUP_END
 ENDIF
-ALU_GROUP_BEGIN
-  ALU MOV S46.x@group : A2[0].x {W}
-  ALU MOV S46.y@group : A2[0].y {W}
-  ALU MOV S46.z@group : A2[1].x {W}
-  ALU MOV S46.w@group : A2[1].y {W}
-  ALU MOV S47.x@group : A2[2].x {WL}
-ALU_GROUP_END
-ALU_GROUP_BEGIN
-  ALU MOV S48.x@group : A2[0].z {W}
-  ALU MOV S47.y@group : A2[2].y {W}
-  ALU MOV S47.z@group : A2[3].x {W}
-  ALU MOV S47.w@group : A2[3].y {W}
-  ALU MOV S48.y@group : A2[0].w {WL}
-ALU_GROUP_END
-ALU_GROUP_BEGIN
-  ALU MOV S49.x@group : A2[2].z {W}
-  ALU MOV S49.y@group : A2[2].w {W}
-  ALU MOV S48.z@group : A2[1].z {W}
-  ALU MOV S48.w@group : A2[1].w {W}
-  ALU MOV S49.z@group : A2[3].z {WL}
-ALU_GROUP_END
-ALU_GROUP_BEGIN
-  ALU MOV S49.w@group : A2[3].w {WL}
+  ALU_GROUP_BEGIN
+    ALU MOV S1025.x@chgr : A2[0].x {W}
+    ALU MOV S1025.y@chgr : A2[0].y {W}
+    ALU MOV S1025.z@chgr : A2[1].x {W}
+    ALU MOV S1025.w@chgr : A2[1].y {W}
+    ALU MOV S1027.x@group : A2[2].x {WL}
+ ALU_GROUP_END
+ ALU_GROUP_BEGIN
+   ALU MOV S1029.x@chgr : A2[0].z {W}
+   ALU MOV S1027.y@chgr : A2[2].y {W}
+   ALU MOV S1027.z@chgr : A2[3].x {W}
+   ALU MOV S1027.w@chgr : A2[3].y {W}
+   ALU MOV S1029.y@group : A2[0].w {WL}
+ ALU_GROUP_END
+ ALU_GROUP_BEGIN
+  ALU MOV S1031.x@chgr : A2[2].z {W}
+  ALU MOV S1031.y@chgr : A2[2].w {W}
+  ALU MOV S1029.z@chgr : A2[1].z {W}
+  ALU MOV S1029.w@chgr : A2[1].w {W}
+  ALU MOV S1031.z@group : A2[3].z {WL}
+ ALU_GROUP_END
+ ALU_GROUP_BEGIN
+   ALU MOV S1031.w@chgr : A2[3].w {WL}
 ALU_GROUP_END
 BLOCK_END
 BLOCK_START
 EXPORT_DONE POS 0 S19.xyzw
-EXPORT PARAM 0 S46.xyzw
-EXPORT PARAM 1 S47.xyzw
-EXPORT PARAM 2 S48.xyzw
-EXPORT_DONE PARAM 3 S49.xyzw
-BLOCK_END
+EXPORT PARAM 0 S1025.xyzw
+EXPORT PARAM 1 S1027.xyzw
+EXPORT PARAM 2 S1029.xyzw
+EXPORT_DONE PARAM 3 S1031.xyzw
+BLOCK END\n
 )";
 
 const char *shader_with_dest_array2 =