i965/gen6/gs: Implement GS_OPCODE_URB_WRITE_ALLOCATE.

Gen6 geometry shaders need to allocate URB handles for each new vertex they
emit after the first (the URB handle for the first vertex is obtained via the
FF_SYNC message).

This opcode adds the URB allocation mechanism to regular URB writes.

Acked-by: Kenneth Graunke <[email protected]>
Reviewed-by: Jordan Justen <[email protected]>

5 files changed, 42 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 27a513ac58d..0b5735cc02e 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -941,6 +941,14 @@ enum opcode {
    GS_OPCODE_URB_WRITE,
 
    /**
+    * Write geometry shader output data to the URB and request a new URB
+    * handle (gen6).
+    *
+    * This opcode doesn't do an implied move from R0 to the first MRF.
+    */
+   GS_OPCODE_URB_WRITE_ALLOCATE,
+
+   /**
     * Terminate the geometry shader thread by doing an empty URB write.
     *
     * This opcode doesn't do an implied move from R0 to the first MRF.  This
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 4eea74e7e10..b2352468450 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -508,6 +508,8 @@ brw_instruction_name(enum opcode op)
 
    case GS_OPCODE_URB_WRITE:
       return "gs_urb_write";
+   case GS_OPCODE_URB_WRITE_ALLOCATE:
+      return "gs_urb_write_allocate";
    case GS_OPCODE_THREAD_END:
       return "gs_thread_end";
    case GS_OPCODE_SET_WRITE_OFFSET:
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
index fa992766a43..4210ee00a54 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
@@ -274,6 +274,7 @@ vec4_visitor::implied_mrf_writes(vec4_instruction *inst)
    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
       return 3;
    case GS_OPCODE_URB_WRITE:
+   case GS_OPCODE_URB_WRITE_ALLOCATE:
    case GS_OPCODE_THREAD_END:
       return 0;
    case GS_OPCODE_FF_SYNC:
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index d2d2a8f5fac..cb863c80595 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -646,6 +646,7 @@ private:
 
    void generate_vs_urb_write(vec4_instruction *inst);
    void generate_gs_urb_write(vec4_instruction *inst);
+   void generate_gs_urb_write_allocate(vec4_instruction *inst);
    void generate_gs_thread_end(vec4_instruction *inst);
    void generate_gs_set_write_offset(struct brw_reg dst,
                                      struct brw_reg src0,
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
index c1eda12a7f3..f98c472430c 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
@@ -467,6 +467,32 @@ vec4_generator::generate_gs_urb_write(vec4_instruction *inst)
 }
 
 void
+vec4_generator::generate_gs_urb_write_allocate(vec4_instruction *inst)
+{
+   struct brw_reg src = brw_message_reg(inst->base_mrf);
+
+   /* We pass the temporary passed in src0 as the writeback register */
+   brw_urb_WRITE(p,
+                 inst->get_src(this->prog_data, 0), /* dest */
+                 inst->base_mrf, /* starting mrf reg nr */
+                 src,
+                 BRW_URB_WRITE_ALLOCATE_COMPLETE,
+                 inst->mlen,
+                 1, /* response len */
+                 inst->offset,  /* urb destination offset */
+                 BRW_URB_SWIZZLE_INTERLEAVE);
+
+   /* Now put allocated urb handle in dst.0 */
+   brw_push_insn_state(p);
+   brw_set_default_access_mode(p, BRW_ALIGN_1);
+   brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+   brw_MOV(p, get_element_ud(inst->get_dst(), 0),
+           get_element_ud(inst->get_src(this->prog_data, 0), 0));
+   brw_set_default_access_mode(p, BRW_ALIGN_16);
+   brw_pop_insn_state(p);
+}
+
+void
 vec4_generator::generate_gs_thread_end(vec4_instruction *inst)
 {
    struct brw_reg src = brw_message_reg(inst->base_mrf);
@@ -1309,6 +1335,10 @@ vec4_generator::generate_code(const cfg_t *cfg)
          generate_gs_urb_write(inst);
          break;
 
+      case GS_OPCODE_URB_WRITE_ALLOCATE:
+         generate_gs_urb_write_allocate(inst);
+         break;
+
       case GS_OPCODE_THREAD_END:
          generate_gs_thread_end(inst);
          break;