i965/gs: Add GS_OPCODE_SET_WRITE_OFFSET.

v2: Added a comment to vec4_generator::generate_gs_set_write_offset().

Reviewed-by: Ian Romanick <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Matt Turner <[email protected]>

4 files changed, 57 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 52009e204fc..ff270da1536 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -817,6 +817,22 @@ enum opcode {
     * for Slot {0,1}" fields in the message header.
     */
    GS_OPCODE_THREAD_END,
+
+   /**
+    * Set the "Slot {0,1} Offset" fields of a URB_WRITE message header.
+    *
+    * - dst is the MRF containing the message header.
+    *
+    * - src0.x indicates which portion of the URB should be written to (e.g. a
+    *   vertex number)
+    *
+    * - src1 is an immediate multiplier which will be applied to src0
+    *   (e.g. the size of a single vertex in the URB).
+    *
+    * Note: the hardware will apply this offset *in addition to* the offset in
+    * vec4_instruction::offset.
+    */
+   GS_OPCODE_SET_WRITE_OFFSET,
 };
 
 #define BRW_PREDICATE_NONE             0
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 689e908d37d..e5d939af24f 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -501,6 +501,8 @@ brw_instruction_name(enum opcode op)
       return "gs_urb_write";
    case GS_OPCODE_THREAD_END:
       return "gs_thread_end";
+   case GS_OPCODE_SET_WRITE_OFFSET:
+      return "set_write_offset";
 
    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 a95f61fb75a..484e5787e5d 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -630,6 +630,9 @@ private:
    void generate_vs_urb_write(vec4_instruction *inst);
    void generate_gs_urb_write(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,
+                                     struct brw_reg src1);
    void generate_oword_dual_block_offsets(struct brw_reg m1,
 					  struct brw_reg index);
    void generate_scratch_write(vec4_instruction *inst,
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp b/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
index 23a3b675138..c487ac85a23 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_emit.cpp
@@ -443,6 +443,38 @@ vec4_generator::generate_gs_thread_end(vec4_instruction *inst)
 }
 
 void
+vec4_generator::generate_gs_set_write_offset(struct brw_reg dst,
+                                             struct brw_reg src0,
+                                             struct brw_reg src1)
+{
+   /* From p22 of volume 4 part 2 of the Ivy Bridge PRM (2.4.3.1 Message
+    * Header: M0.3):
+    *
+    *     Slot 0 Offset. This field, after adding to the Global Offset field
+    *     in the message descriptor, specifies the offset (in 256-bit units)
+    *     from the start of the URB entry, as referenced by URB Handle 0, at
+    *     which the data will be accessed.
+    *
+    * Similar text describes DWORD M0.4, which is slot 1 offset.
+    *
+    * Therefore, we want to multiply DWORDs 0 and 4 of src0 (the x components
+    * of the register for geometry shader invocations 0 and 1) by the
+    * immediate value in src1, and store the result in DWORDs 3 and 4 of dst.
+    *
+    * We can do this with the following EU instruction:
+    *
+    *     mul(2) dst.3<1>UD src0<8;2,4>UD src1   { Align1 WE_all }
+    */
+   brw_push_insn_state(p);
+   brw_set_access_mode(p, BRW_ALIGN_1);
+   brw_set_mask_control(p, BRW_MASK_DISABLE);
+   brw_MUL(p, suboffset(stride(dst, 2, 2, 1), 3), stride(src0, 8, 2, 4),
+           src1);
+   brw_set_access_mode(p, BRW_ALIGN_16);
+   brw_pop_insn_state(p);
+}
+
+void
 vec4_generator::generate_oword_dual_block_offsets(struct brw_reg m1,
                                                   struct brw_reg index)
 {
@@ -918,6 +950,10 @@ vec4_generator::generate_vec4_instruction(vec4_instruction *instruction,
       generate_gs_thread_end(inst);
       break;
 
+   case GS_OPCODE_SET_WRITE_OFFSET:
+      generate_gs_set_write_offset(dst, src[0], src[1]);
+      break;
+
    case SHADER_OPCODE_SHADER_TIME_ADD:
       brw_shader_time_add(p, src[0], SURF_INDEX_VS_SHADER_TIME);
       mark_surface_used(SURF_INDEX_VS_SHADER_TIME);