[Mesa-stable] [PATCH 21/44] intel/fs: Rework zero-length URB write handling

[Jason Ekstrand]

Originally we tried to handle this case based on slots_valid.  However,
there are a number of ways that this can go wrong.  For one, we throw
away any trailing slots which either aren't written or are set to
VARYING_SLOT_PAD.  Second, even if PSIZ is a valid slot, we may not
actually write anything there.  Between the lot of these, it was
possible to end up in a case where we tried to do a regular URB write
but ended up with a length of 1 which is invalid.  This commit moves it
to the end and makes it based on a new boolean flag urb_written.

Cc: mesa-stable at lists.freedesktop.org
---
 src/intel/compiler/brw_fs_visitor.cpp | 60 ++++++++++++++++++-----------------
 1 file changed, 31 insertions(+), 29 deletions(-)

diff --git a/src/intel/compiler/brw_fs_visitor.cpp b/src/intel/compiler/brw_fs_visitor.cpp
index fb575ef..f17ef81 100644
--- a/src/intel/compiler/brw_fs_visitor.cpp
+++ b/src/intel/compiler/brw_fs_visitor.cpp
@@ -559,34 +559,6 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
    else
       urb_handle = fs_reg(retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD));
 
-   /* If we don't have any valid slots to write, just do a minimal urb write
-    * send to terminate the shader.  This includes 1 slot of undefined data,
-    * because it's invalid to write 0 data:
-    *
-    * From the Broadwell PRM, Volume 7: 3D Media GPGPU, Shared Functions -
-    * Unified Return Buffer (URB) > URB_SIMD8_Write and URB_SIMD8_Read >
-    * Write Data Payload:
-    *
-    *    "The write data payload can be between 1 and 8 message phases long."
-    */
-   if (vue_map->slots_valid == 0) {
-      /* For GS, just turn EmitVertex() into a no-op.  We don't want it to
-       * end the thread, and emit_gs_thread_end() already emits a SEND with
-       * EOT at the end of the program for us.
-       */
-      if (stage == MESA_SHADER_GEOMETRY)
-         return;
-
-      fs_reg payload = fs_reg(VGRF, alloc.allocate(2), BRW_REGISTER_TYPE_UD);
-      bld.exec_all().MOV(payload, urb_handle);
-
-      fs_inst *inst = bld.emit(SHADER_OPCODE_URB_WRITE_SIMD8, reg_undef, payload);
-      inst->eot = true;
-      inst->mlen = 2;
-      inst->offset = 1;
-      return;
-   }
-
    opcode opcode = SHADER_OPCODE_URB_WRITE_SIMD8;
    int header_size = 1;
    fs_reg per_slot_offsets;
@@ -638,6 +610,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
       last_slot--;
    }
 
+   bool urb_written = false;
    for (slot = 0; slot < vue_map->num_slots; slot++) {
       int varying = vue_map->slot_to_varying[slot];
       switch (varying) {
@@ -723,7 +696,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
        * the last slot or if we need to flush (see BAD_FILE varying case
        * above), emit a URB write send now to flush out the data.
        */
-      if (length == 8 || slot == last_slot)
+      if (length == 8 || (length > 0 && slot == last_slot))
          flush = true;
       if (flush) {
          fs_reg *payload_sources =
@@ -748,8 +721,37 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
          urb_offset = starting_urb_offset + slot + 1;
          length = 0;
          flush = false;
+         urb_written = true;
       }
    }
+
+   /* If we don't have any valid slots to write, just do a minimal urb write
+    * send to terminate the shader.  This includes 1 slot of undefined data,
+    * because it's invalid to write 0 data:
+    *
+    * From the Broadwell PRM, Volume 7: 3D Media GPGPU, Shared Functions -
+    * Unified Return Buffer (URB) > URB_SIMD8_Write and URB_SIMD8_Read >
+    * Write Data Payload:
+    *
+    *    "The write data payload can be between 1 and 8 message phases long."
+    */
+   if (!urb_written) {
+      /* For GS, just turn EmitVertex() into a no-op.  We don't want it to
+       * end the thread, and emit_gs_thread_end() already emits a SEND with
+       * EOT at the end of the program for us.
+       */
+      if (stage == MESA_SHADER_GEOMETRY)
+         return;
+
+      fs_reg payload = fs_reg(VGRF, alloc.allocate(2), BRW_REGISTER_TYPE_UD);
+      bld.exec_all().MOV(payload, urb_handle);
+
+      fs_inst *inst = bld.emit(SHADER_OPCODE_URB_WRITE_SIMD8, reg_undef, payload);
+      inst->eot = true;
+      inst->mlen = 2;
+      inst->offset = 1;
+      return;
+   }
 }
 
 void
-- 
2.5.0.400.gff86faf