radeonsi: implement nir_intrinsic_load_streamout_buffer_amd

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17456>

src/amd/llvm/ac_nir_to_llvm.c

@@ -3617,6 +3617,9 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
    case nir_intrinsic_load_user_clip_plane:
       result = ctx->abi->load_user_clip_plane(ctx->abi, nir_intrinsic_ucp_id(instr));
       break;
+   case nir_intrinsic_load_streamout_buffer_amd:
+      result = ctx->abi->load_streamout_buffer(ctx->abi, nir_intrinsic_base(instr));
+      break;
    case nir_intrinsic_load_ordered_id_amd:
       result = ac_unpack_param(&ctx->ac, ac_get_arg(&ctx->ac, ctx->args->gs_tg_info), 0, 12);
       break;

src/amd/llvm/ac_shader_abi.h

@@ -111,6 +111,8 @@ struct ac_shader_abi {
 
    LLVMValueRef (*load_user_clip_plane)(struct ac_shader_abi *abi, unsigned ucp_id);
 
+   LLVMValueRef (*load_streamout_buffer)(struct ac_shader_abi *abi, unsigned buffer);
+
    LLVMValueRef (*emit_fbfetch)(struct ac_shader_abi *abi);
 
    LLVMValueRef (*intrinsic_load)(struct ac_shader_abi *abi, nir_intrinsic_op op);

src/gallium/drivers/radeonsi/si_shader_llvm.c

@@ -902,6 +902,15 @@ static LLVMValueRef si_llvm_load_user_clip_plane(struct ac_shader_abi *abi, unsi
                                ctx->ac.f32, 0, true, true);
 }
 
+static LLVMValueRef si_llvm_load_streamout_buffer(struct ac_shader_abi *abi, unsigned buffer)
+{
+   struct si_shader_context *ctx = si_shader_context_from_abi(abi);
+   LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
+
+   return ac_build_load_to_sgpr(
+      &ctx->ac, buf_ptr, LLVMConstInt(ctx->ac.i32, SI_VS_STREAMOUT_BUF0 + buffer, false));
+}
+
 bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shader,
                            struct nir_shader *nir, bool free_nir, bool ngg_cull_shader)
 {
@@ -919,6 +928,7 @@ bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shad
 
    ctx->abi.intrinsic_load = si_llvm_load_intrinsic;
    ctx->abi.load_user_clip_plane = si_llvm_load_user_clip_plane;
+   ctx->abi.load_streamout_buffer = si_llvm_load_streamout_buffer;
 
    si_llvm_init_resource_callbacks(ctx);
    si_llvm_create_main_func(ctx, ngg_cull_shader);