[Mesa-dev] [PATCH v3 (part2) 38/56] i965/nir/fs: Implement SSBO atomics

[Iago Toral Quiroga]

---
 src/mesa/drivers/dri/i965/brw_fs.h       |  2 +
 src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 77 ++++++++++++++++++++++++++++++++
 2 files changed, 79 insertions(+)

diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h
index ea204d3..844f813 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.h
+++ b/src/mesa/drivers/dri/i965/brw_fs.h
@@ -276,6 +276,8 @@ public:
                        nir_ssa_undef_instr *instr);
    void nir_emit_intrinsic(const brw::fs_builder &bld,
                            nir_intrinsic_instr *instr);
+   void nir_emit_ssbo_atomic(const brw::fs_builder &bld,
+                             int op, nir_intrinsic_instr *instr);
    void nir_emit_texture(const brw::fs_builder &bld,
                          nir_tex_instr *instr);
    void nir_emit_jump(const brw::fs_builder &bld,
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 9bf41f7..752742d 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -1742,6 +1742,37 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
       emit_barrier();
       break;
 
+   case nir_intrinsic_ssbo_atomic_add:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_ADD, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_min:
+      if (dest.type == BRW_REGISTER_TYPE_D)
+         nir_emit_ssbo_atomic(bld, BRW_AOP_IMIN, instr);
+      else
+         nir_emit_ssbo_atomic(bld, BRW_AOP_UMIN, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_max:
+      if (dest.type == BRW_REGISTER_TYPE_D)
+         nir_emit_ssbo_atomic(bld, BRW_AOP_IMAX, instr);
+      else
+         nir_emit_ssbo_atomic(bld, BRW_AOP_UMAX, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_and:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_AND, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_or:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_OR, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_xor:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_XOR, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_exchange:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_MOV, instr);
+      break;
+   case nir_intrinsic_ssbo_atomic_comp_swap:
+      nir_emit_ssbo_atomic(bld, BRW_AOP_CMPWR, instr);
+      break;
+
    case nir_intrinsic_ssbo_unsized_array_length: {
       nir_const_value *const_uniform_block = nir_src_as_const_value(instr->src[0]);
       unsigned ubo_index = const_uniform_block ? const_uniform_block->u[0] : 0;
@@ -1798,6 +1829,52 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
 }
 
 void
+fs_visitor::nir_emit_ssbo_atomic(const fs_builder &bld,
+                                 int op, nir_intrinsic_instr *instr)
+{
+   fs_reg dest;
+   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
+      dest = get_nir_dest(instr->dest);
+
+   fs_reg surface;
+   nir_const_value *const_surface = nir_src_as_const_value(instr->src[0]);
+   if (const_surface) {
+      unsigned surf_index = stage_prog_data->binding_table.ubo_start +
+                            const_surface->u[0];
+      surface = fs_reg(surf_index);
+      brw_mark_surface_used(prog_data, surf_index);
+   } else {
+      surface = vgrf(glsl_type::uint_type);
+      bld.ADD(surface, get_nir_src(instr->src[0]),
+              fs_reg(stage_prog_data->binding_table.ubo_start));
+
+      /* Assume this may touch any UBO. This is the same we do for other
+       * UBO/SSBO accesses with non-constant surface.
+       */
+      brw_mark_surface_used(prog_data,
+                            stage_prog_data->binding_table.ubo_start +
+                            shader_prog->NumUniformBlocks - 1);
+   }
+
+   fs_reg offset = get_nir_src(instr->src[1]);
+   fs_reg data1 = get_nir_src(instr->src[2]);
+   fs_reg data2;
+   if (op == BRW_AOP_CMPWR)
+      data2 = get_nir_src(instr->src[3]);
+
+   /* Emit the actual atomic operation operation */
+
+   fs_reg atomic_result =
+      surface_access::emit_untyped_atomic(bld, surface, offset,
+                                          data1, data2,
+                                          1 /* dims */, 1 /* rsize */,
+                                          op,
+                                          BRW_PREDICATE_NONE);
+   dest.type = atomic_result.type;
+   bld.MOV(dest, atomic_result);
+}
+
+void
 fs_visitor::nir_emit_texture(const fs_builder &bld, nir_tex_instr *instr)
 {
    unsigned sampler = instr->sampler_index;
-- 
1.9.1