ac/nir: gs and nogs use ac_nir_export_primitive

Mesh shader primitive export is left unchanged because it needs
extra changes for per primitive output export when export
primitive.

Mesh shader will use second channel of primitive export.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20691>

src/amd/common/ac_nir.c

@@ -22,6 +22,7 @@
  */
 
 #include "ac_nir.h"
+#include "sid.h"
 #include "nir_builder.h"
 #include "nir_xfb_info.h"
 
@@ -44,6 +45,17 @@ ac_nir_unpack_arg(nir_builder *b, const struct ac_shader_args *ac_args, struct a
    return nir_ubfe_imm(b, value, rshift, bitwidth);
 }
 
+void
+ac_nir_export_primitive(nir_builder *b, nir_ssa_def *prim)
+{
+   unsigned write_mask = BITFIELD_MASK(prim->num_components);
+
+   nir_export_amd(b, nir_pad_vec4(b, prim),
+                  .base = V_008DFC_SQ_EXP_PRIM,
+                  .flags = AC_EXP_FLAG_DONE,
+                  .write_mask = write_mask);
+}
+
 /**
  * This function takes an I/O intrinsic like load/store_input,
  * and emits a sequence that calculates the full offset of that instruction,

src/amd/common/ac_nir.h

@@ -73,6 +73,9 @@ nir_ssa_def *
 ac_nir_unpack_arg(nir_builder *b, const struct ac_shader_args *ac_args, struct ac_arg arg,
                   unsigned rshift, unsigned bitwidth);
 
+void
+ac_nir_export_primitive(nir_builder *b, nir_ssa_def *prim);
+
 nir_ssa_def *
 ac_nir_calc_io_offset(nir_builder *b,
                       nir_intrinsic_instr *intrin,

src/amd/common/ac_nir_lower_ngg.c

@@ -545,7 +545,7 @@ emit_ngg_nogs_prim_export(nir_builder *b, lower_ngg_nogs_state *st, nir_ssa_def
          nir_pop_if(b, if_shader_query);
       }
 
-      nir_export_primitive_amd(b, arg);
+      ac_nir_export_primitive(b, arg);
    }
    nir_pop_if(b, if_gs_thread);
 }
@@ -2840,7 +2840,7 @@ ngg_gs_export_primitives(nir_builder *b, nir_ssa_def *max_num_out_prims, nir_ssa
    }
 
    nir_ssa_def *arg = emit_pack_ngg_prim_exp_arg(b, s->num_vertices_per_primitive, vtx_indices, is_null_prim, false);
-   nir_export_primitive_amd(b, arg);
+   ac_nir_export_primitive(b, arg);
    nir_pop_if(b, if_prim_export_thread);
 }