radv,aco: implement iadd_sat

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12617>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 32b54c8..f5753e8 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -1809,6 +1809,22 @@ visit_alu_instr(isel_context* ctx, nir_alu_instr* instr)
       }
       break;
    }
+   case nir_op_iadd_sat: {
+      Temp src0 = get_alu_src(ctx, instr->src[0]);
+      Temp src1 = as_vgpr(ctx, get_alu_src(ctx, instr->src[1]));
+      if (dst.regClass() == v2b) {
+         Instruction* add_instr =
+            bld.vop3(aco_opcode::v_add_i16, Definition(dst), src0, src1).instr;
+         add_instr->vop3().clamp = 1;
+      } else if (dst.regClass() == v1) {
+         Instruction* add_instr =
+            bld.vop3(aco_opcode::v_add_i32, Definition(dst), src0, src1).instr;
+         add_instr->vop3().clamp = 1;
+      } else {
+         isel_err(&instr->instr, "Unimplemented NIR instr bit size");
+      }
+      break;
+   }
    case nir_op_uadd_carry: {
       Temp src0 = get_alu_src(ctx, instr->src[0]);
       Temp src1 = get_alu_src(ctx, instr->src[1]);

diff --git a/src/amd/compiler/aco_instruction_selection_setup.cpp b/src/amd/compiler/aco_instruction_selection_setup.cpp
index 28ce822..5371628 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection_setup.cpp
+++ b/src/amd/compiler/aco_instruction_selection_setup.cpp
@@ -593,7 +593,8 @@ init_context(isel_context* ctx, nir_shader* shader)
                case nir_op_frexp_exp:
                case nir_op_cube_face_index_amd:
                case nir_op_cube_face_coord_amd:
-               case nir_op_sad_u8x4: type = RegType::vgpr; break;
+               case nir_op_sad_u8x4:
+               case nir_op_iadd_sat: type = RegType::vgpr; break;
                case nir_op_f2i16:
                case nir_op_f2u16:
                case nir_op_f2i32:

diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index 3b01545..414ede8 100644 (file)
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -3224,6 +3224,9 @@ lower_bit_size_callback(const nir_instr *instr, void *_)
       case nir_op_uadd_sat:
          return (bit_size == 8 || !(chip >= GFX8 && nir_dest_is_divergent(alu->dest.dest))) ? 32
                                                                                             : 0;
+      case nir_op_iadd_sat:
+         return bit_size == 8 || !nir_dest_is_divergent(alu->dest.dest) ? 32 : 0;
+
       default:
          return 0;
       }

diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index 5952f95..0b9d3d8 100644 (file)
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -81,6 +81,7 @@ radv_get_nir_options(struct radv_physical_device *device)
       .lower_fpow = true,
       .lower_mul_2x32_64 = true,
       .lower_rotate = true,
+      .lower_iadd_sat = device->rad_info.chip_class <= GFX8,
       .has_fsub = true,
       .has_isub = true,
       .use_scoped_barrier = true,