i965/vec4: Add the ability for attributes to be interleaved.

When geometry shaders are operated in "single" or "dual instanced"
mode, a single set of geometry shader inputs is interleaved into the
thread payload (with each payload register containing a pair of
inputs) in order to save register space.

This patch modifies vec4_visitor::lower_attributes_to_hw_regs so that
it can handle the interleaved format.

Reviewed-by: Eric Anholt <[email protected]>

3 files changed, 27 insertions, 6 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
index ba7490053b1..589de48f33a 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
@@ -1184,12 +1184,32 @@ vec4_visitor::dump_instruction(backend_instruction *be_inst)
    printf("\n");
 }
 
+
+static inline struct brw_reg
+attribute_to_hw_reg(int attr, bool interleaved)
+{
+   if (interleaved)
+      return stride(brw_vec4_grf(attr / 2, (attr % 2) * 4), 0, 4, 1);
+   else
+      return brw_vec8_grf(attr, 0);
+}
+
+
 /**
  * Replace each register of type ATTR in this->instructions with a reference
  * to a fixed HW register.
+ *
+ * If interleaved is true, then each attribute takes up half a register, with
+ * register N containing attribute 2*N in its first half and attribute 2*N+1
+ * in its second half (this corresponds to the payload setup used by geometry
+ * shaders in "single" or "dual instanced" dispatch mode).  If interleaved is
+ * false, then each attribute takes up a whole register, with register N
+ * containing attribute N (this corresponds to the payload setup used by
+ * vertex shaders, and by geometry shaders in "dual object" dispatch mode).
  */
 void
-vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map)
+vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map,
+                                          bool interleaved)
 {
    foreach_list(node, &this->instructions) {
       vec4_instruction *inst = (vec4_instruction *)node;
@@ -1203,7 +1223,7 @@ vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map)
           */
          assert(grf != 0);
 
-	 struct brw_reg reg = brw_vec8_grf(grf, 0);
+	 struct brw_reg reg = attribute_to_hw_reg(grf, interleaved);
 	 reg.type = inst->dst.type;
 	 reg.dw1.bits.writemask = inst->dst.writemask;
 
@@ -1222,7 +1242,7 @@ vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map)
           */
          assert(grf != 0);
 
-	 struct brw_reg reg = brw_vec8_grf(grf, 0);
+	 struct brw_reg reg = attribute_to_hw_reg(grf, interleaved);
 	 reg.dw1.bits.swizzle = inst->src[i].swizzle;
          reg.type = inst->src[i].type;
 	 if (inst->src[i].abs)
@@ -1260,7 +1280,7 @@ vec4_vs_visitor::setup_attributes(int payload_reg)
       nr_attributes++;
    }
 
-   lower_attributes_to_hw_regs(attribute_map);
+   lower_attributes_to_hw_regs(attribute_map, false /* interleaved */);
 
    /* The BSpec says we always have to read at least one thing from
     * the VF, and it appears that the hardware wedges otherwise.
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index 854478ce06e..23e004ef397 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -515,7 +515,8 @@ public:
 
 protected:
    void emit_vertex();
-   void lower_attributes_to_hw_regs(const int *attribute_map);
+   void lower_attributes_to_hw_regs(const int *attribute_map,
+                                    bool interleaved);
    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
                                    int reg_node_count);
    virtual dst_reg *make_reg_for_system_value(ir_variable *ir) = 0;
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
index 0893c95b8fa..08a55a3fab9 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
@@ -110,7 +110,7 @@ vec4_gs_visitor::setup_payload()
 
    reg = setup_varying_inputs(reg, attribute_map);
 
-   lower_attributes_to_hw_regs(attribute_map);
+   lower_attributes_to_hw_regs(attribute_map, false /* interleaved */);
 
    this->first_non_payload_grf = reg;
 }