[Mesa-dev] [PATCH 26/37] i965/gen6/gs: implement GS_OPCODE_FF_SYNC_SET_PRIMITIVES opcode

[Iago Toral Quiroga]

From: Samuel Iglesias Gonsalvez <siglesias at igalia.com>

This opcode will be used when filling FF_SYNC header before
emitting vertices and their data.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias at igalia.com>
---
 src/mesa/drivers/dri/i965/brw_defines.h          | 15 +++++++++++++++
 src/mesa/drivers/dri/i965/brw_shader.cpp         |  2 ++
 src/mesa/drivers/dri/i965/brw_vec4.h             |  4 ++++
 src/mesa/drivers/dri/i965/brw_vec4_generator.cpp | 24 ++++++++++++++++++++++++
 4 files changed, 45 insertions(+)

diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 7095c39..6e8b998 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -1061,6 +1061,21 @@ enum opcode {
     * - src is the register that holds the destination indices value.
     */
    GS_OPCODE_SVB_SET_DST_INDEX,
+
+   /**
+    * Prepare Mx.0 subregister for being used in the FF_SYNC message header.
+    * Used in gen6 for transform feedback.
+    *
+    * - dst will hold the register with the final Mx.0 value.
+    *
+    * - src0 has the number of vertices emitted in SO (NumSOVertsToWrite)
+    *
+    * - src1 has the number of needed primitives for SO (NumSOPrimsNeeded)
+    *
+    * - src2 is the value to hold in M0: number of SO vertices to write
+    *   and number of SO primitives needed.
+    */
+   GS_OPCODE_FF_SYNC_SET_PRIMITIVES,
 };
 
 enum brw_urb_write_flags {
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index bf625a5..7328fdc 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -540,6 +540,8 @@ brw_instruction_name(enum opcode op)
       return "gs_svb_write";
    case GS_OPCODE_SVB_SET_DST_INDEX:
       return "gs_svb_set_dst_index";
+   case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
+      return "gs_ff_sync_set_primitives";
 
    default:
       /* Yes, this leaks.  It's in debug code, it should never occur, and if
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index ea3967d..763cb23 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -673,6 +673,10 @@ private:
    void generate_gs_prepare_channel_masks(struct brw_reg dst);
    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
    void generate_gs_get_instance_id(struct brw_reg dst);
+   void generate_gs_ff_sync_set_primitives(struct brw_reg dst,
+                                           struct brw_reg src0,
+                                           struct brw_reg src1,
+                                           struct brw_reg src2);
    void generate_gs_ff_sync(struct brw_reg dst,
                             struct brw_reg src0,
                             struct brw_reg src1);
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
index d914a52..d4554f5 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
@@ -712,6 +712,26 @@ vec4_generator::generate_gs_get_instance_id(struct brw_reg dst)
 }
 
 void
+vec4_generator::generate_gs_ff_sync_set_primitives(struct brw_reg dst,
+                                              struct brw_reg src0,
+                                              struct brw_reg src1,
+                                              struct brw_reg src2)
+{
+   brw_push_insn_state(p);
+   brw_set_default_access_mode(p, BRW_ALIGN_1);
+   /* Save src0 data in 16:31 bits of dst.0 */
+   brw_AND(p, suboffset(vec1(dst), 0), suboffset(vec1(src0), 0), brw_imm_ud(0xffffu));
+   brw_SHL(p, suboffset(vec1(dst), 0), suboffset(vec1(dst), 0), brw_imm_ud(16));
+   /* Save src1 data in 0:15 bits of dst.0 */
+   brw_AND(p, suboffset(vec1(src2), 0), suboffset(vec1(src1), 0), brw_imm_ud(0xffffu));
+   brw_OR(p, suboffset(vec1(dst), 0),
+          suboffset(vec1(dst), 0),
+          suboffset(vec1(src2), 0));
+   brw_set_default_access_mode(p, BRW_ALIGN_16);
+   brw_pop_insn_state(p);
+}
+
+void
 vec4_generator::generate_gs_ff_sync(struct brw_reg dst,
                                     struct brw_reg src0,
                                     struct brw_reg src1)
@@ -1357,6 +1377,10 @@ vec4_generator::generate_vec4_instruction(vec4_instruction *instruction,
       generate_gs_ff_sync(dst, src[0], src[1]);
       break;
 
+   case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
+      generate_gs_ff_sync_set_primitives(dst, src[0], src[1], src[2]);
+      break;
+
    case GS_OPCODE_SET_PRIMITIVE_ID:
       generate_gs_set_primitive_id(dst);
       break;
-- 
1.9.1