compiler: add SYSTEM_VALUE_USER_DATA_AMD

for internal radeonsi shaders

src/amd/common/ac_nir_to_llvm.c

@@ -3313,6 +3313,10 @@ static void visit_intrinsic(struct ac_nir_context *ctx,
 	case nir_intrinsic_load_color1:
 		result = ctx->abi->color1;
 		break;
+	case nir_intrinsic_load_user_data_amd:
+		assert(LLVMTypeOf(ctx->abi->user_data) == ctx->ac.v4i32);
+		result = ctx->abi->user_data;
+		break;
 	case nir_intrinsic_load_instance_id:
 		result = ctx->abi->instance_id;
 		break;

src/amd/common/ac_shader_abi.h

@@ -65,6 +65,7 @@ struct ac_shader_abi {
 	LLVMValueRef prim_mask;
 	LLVMValueRef color0;
 	LLVMValueRef color1;
+	LLVMValueRef user_data;
 	/* CS */
 	LLVMValueRef local_invocation_ids;
 	LLVMValueRef num_work_groups;

src/compiler/nir/nir.c

@@ -2022,6 +2022,8 @@ nir_intrinsic_from_system_value(gl_system_value val)
       return nir_intrinsic_load_global_invocation_index;
    case SYSTEM_VALUE_WORK_DIM:
       return nir_intrinsic_load_work_dim;
+   case SYSTEM_VALUE_USER_DATA_AMD:
+      return nir_intrinsic_load_user_data_amd;
    default:
       unreachable("system value does not directly correspond to intrinsic");
    }
@@ -2109,6 +2111,8 @@ nir_system_value_from_intrinsic(nir_intrinsic_op intrin)
       return SYSTEM_VALUE_LOCAL_GROUP_SIZE;
    case nir_intrinsic_load_global_invocation_id:
       return SYSTEM_VALUE_GLOBAL_INVOCATION_ID;
+   case nir_intrinsic_load_user_data_amd:
+      return SYSTEM_VALUE_USER_DATA_AMD;
    default:
       unreachable("intrinsic doesn't produce a system value");
    }

src/compiler/nir/nir_intrinsics.py

@@ -614,6 +614,9 @@ system_value("blend_const_color_aaaa8888_unorm", 1)
 system_value("color0", 4)
 system_value("color1", 4)
 
+# System value for internal compute shaders in radeonsi.
+system_value("user_data_amd", 4)
+
 # Barycentric coordinate intrinsics.
 #
 # These set up the barycentric coordinates for a particular interpolation.

src/compiler/shader_enums.h

@@ -609,6 +609,7 @@ typedef enum
    SYSTEM_VALUE_LOCAL_GROUP_SIZE,
    SYSTEM_VALUE_GLOBAL_GROUP_SIZE,
    SYSTEM_VALUE_WORK_DIM,
+   SYSTEM_VALUE_USER_DATA_AMD,
    /*@}*/
 
    /** Required for VK_KHR_device_group */

src/gallium/auxiliary/nir/tgsi_to_nir.c

@@ -628,6 +628,10 @@ ttn_src_for_file_and_index(struct ttn_compile *c, unsigned file, unsigned index,
          op = nir_intrinsic_load_work_group_id;
          load = nir_load_work_group_id(b);
          break;
+      case TGSI_SEMANTIC_CS_USER_DATA_AMD:
+         op = nir_intrinsic_load_user_data_amd;
+         load = nir_load_user_data_amd(b);
+         break;
       default:
          unreachable("bad system value");
       }

src/gallium/drivers/radeonsi/si_shader_nir.c

@@ -1240,6 +1240,12 @@ bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir)
 
 		ctx->abi.interp_at_sample_force_center =
 			ctx->shader->key.mono.u.ps.interpolate_at_sample_force_center;
+	} else if (nir->info.stage == MESA_SHADER_COMPUTE) {
+		if (nir->info.cs.user_data_components_amd) {
+			ctx->abi.user_data = LLVMGetParam(ctx->main_fn, ctx->param_cs_user_data);
+			ctx->abi.user_data = ac_build_expand_to_vec4(&ctx->ac, ctx->abi.user_data,
+								     nir->info.cs.user_data_components_amd);
+		}
 	}
 
 	ctx->abi.inputs = &ctx->inputs[0];