diff --git a/App/lib/Phase/Phase1/Phase1Main.vp b/App/lib/Phase/Phase1/Phase1Main.vp index 4676166..c484cb3 100644 --- a/App/lib/Phase/Phase1/Phase1Main.vp +++ b/App/lib/Phase/Phase1/Phase1Main.vp @@ -922,8 +922,8 @@ def RunPhase1(mb: memhub.MemBuddy*, includes_dir: str, temp_dir: str, tf: rawptr[fileio.File] = fileio.File(text_path, fileio.MODE.R) if not tf.closed: tf.close() - # sidecar（{sha1}.meta）也必须存在：缺失则重新翻译以补写， - # 否则 Phase B+/C 的「纯声明文件」判定拿不到结构化清单。 + # sidecar（{sha1}.meta）与在线 FULL 产物（{sha1}.full.ll）也必须存在： + # 缺失则重新翻译以补写（否则 Phase B+/C 的声明判定与编译拿不到数据）。 meta_path_1b: str = StubMergerCollect._sliced_path(temp_dir, dir_len, sha1, "meta") meta_ok_1b: int = 0 if meta_path_1b is not None: @@ -932,7 +932,15 @@ def RunPhase1(mb: memhub.MemBuddy*, includes_dir: str, temp_dir: str, mf_chk.close() meta_ok_1b = 1 stdlib.free(meta_path_1b) - if meta_ok_1b != 0: + full_path_1b: str = StubMergerCollect._sliced_path(temp_dir, dir_len, sha1, "full.ll") + full_ok_1b: int = 0 + if full_path_1b is not None: + ff_chk: rawptr[fileio.File] = fileio.File(full_path_1b, fileio.MODE.R) + if not ff_chk.closed: + ff_chk.close() + full_ok_1b = 1 + stdlib.free(full_path_1b) + if meta_ok_1b != 0 and full_ok_1b != 0: stdlib.free(text_path) stdlib.free(stub_path) continue @@ -1093,7 +1101,9 @@ def RunPhase1(mb: memhub.MemBuddy*, includes_dir: str, temp_dir: str, # dump stub IR(declarations only) # 1MB 不够：ast/parser.vp 的 text.ll 超 1MB 被截断 → 合并 IR 尾部丢字 # （opt: "end of file in global variable name"），提至 8MB。 - PHASE1_IR_SIZE: sizet = 8388608 + PHASE1_IR_SIZE: sizet = 8388608 + # FULL dump = stub + text（同一模块），留 2× 余量 + PHASE1_FULL_SIZE: sizet = 16777216 stub_buf: bytes = stdlib.malloc(PHASE1_IR_SIZE) if stub_buf is None: stdlib.free(src_buf) @@ -1143,6 +1153,22 @@ def RunPhase1(mb: memhub.MemBuddy*, includes_dir: str, temp_dir: str, # save {sha1}.meta（结构化 sidecar：函数清单/结构体布局/虚表，替代从 IR 文本反推） StubMergerIR.WriteModuleMeta(tr.Module, temp_dir, dir_len_p1, sha1) + # save {sha1}.full.ll（在线 FULL 产物：模型在线时的自足单模块 IR，供 Phase B 直接编译， + # 取代 stub + 依赖 stub/text 文本合并） + full_buf_p1: bytes = stdlib.malloc(PHASE1_FULL_SIZE) + if full_buf_p1 is None: + VLogger.error("full.ll 缓冲区分配失败", "Phase1") + tr.dump_ir(full_buf_p1, PHASE1_FULL_SIZE, llvmlite.OUTPUT_FULL) + full_len_p1: sizet = string.strlen(full_buf_p1) + full_path_p1: str = StubMergerCollect._sliced_path(temp_dir, dir_len_p1, sha1, "full.ll") + if full_path_p1 is not None: + ff_p1: rawptr[fileio.File] = fileio.File(full_path_p1, fileio.MODE.W) + if not ff_p1.closed: + ff_p1.write(full_buf_p1, full_len_p1) + ff_p1.close() + stdlib.free(full_path_p1) + stdlib.free(full_buf_p1) + # 计时点 3：dump（stub+text）完成；输出单文件三段耗时 t_f3: uint64 = w32.win32base.GetTickCount64() VLogger.info(f"[{i+1}/{loop_n_b}] 翻译: {file_path} (parse={t_f1-t_f0} ms, translate={t_f2-t_f1} ms, dump={t_f3-t_f2} ms, total={t_f3-t_f0} ms)", "Phase1") diff --git a/App/lib/Phase/Phase2/__init__.vp b/App/lib/Phase/Phase2/__init__.vp index 6987ae3..e2baca7 100644 --- a/App/lib/Phase/Phase2/__init__.vp +++ b/App/lib/Phase/Phase2/__init__.vp @@ -965,6 +965,8 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, # Phase A IR 缓冲：1MB 不够，ast/parser.vp 的 text.ll 超 1MB 被截断 → # 合并 IR 尾部丢字（opt: "end of file in global variable name"）。 PHASE_A_IR_SIZE: sizet = 8388608 + # FULL dump = stub + text（同一模块），留 2× 余量 + PHASE_A_FULL_SIZE: sizet = 16777216 td_len_pa: sizet = string.strlen(temp_dir) # 设置 temp 目录路径，使 HandlesExprCall 能读取依赖模块的 text.ll @@ -1088,6 +1090,21 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, # save {sha1}.meta（结构化 sidecar：函数清单/结构体布局/虚表） StubMergerIR.WriteModuleMeta(tr_a.Module, temp_dir, td_len_pa, ent.Sha1) + # save {sha1}.full.ll（在线 FULL 产物：模型在线时的自足单模块 IR） + full_buf_a: bytes = stdlib.malloc(PHASE_A_FULL_SIZE) + if full_buf_a is None: + VLogger.error("full.ll 缓冲区分配失败", "PhaseA") + tr_a.dump_ir(full_buf_a, PHASE_A_FULL_SIZE, llvmlite.OUTPUT_FULL) + full_len_a: sizet = string.strlen(full_buf_a) + full_path_a: str = StubMergerCollect._sliced_path(temp_dir, td_len_pa, ent.Sha1, "full.ll") + if full_path_a is not None: + ffa: rawptr[fileio.File] = fileio.File(full_path_a, fileio.MODE.W) + if not ffa.closed: + ffa.write(full_buf_a, full_len_a) + ffa.close() + stdlib.free(full_path_a) + stdlib.free(full_buf_a) + # save dependencies (_imported_modules) for Phase B (切片路径) deps_path_a: str = StubMergerCollect._sliced_path(temp_dir, td_len_pa, ent.Sha1, "deps.txt") if deps_path_a is not None: @@ -1165,8 +1182,8 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, main_obj_path[0] = '\0' compiled_count: int = 0 - # 组合 IR 缓冲区大小（4MB，足够容纳本地 stub + 所有依赖 stub + 本地 text） - COMBINED_IR_SIZE: sizet = 4194304 + # IR 缓冲区大小（16MB，容纳在线 FULL 产物 {sha1}.full.ll；FULL = stub + text） + COMBINED_IR_SIZE: sizet = 16777216 # llc 任务表（stdlib.malloc，跨线程存活）：连续 LlcTask 记录数组， # 循环内只写 .ll + 构造任务，循环后 run_llc_parallel 分批并行跑 llc @@ -1183,17 +1200,17 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, continue # 单文件三步计时起点（GetTickCount64，ms） t_pb0: uint64 = w32.win32base.GetTickCount64() - # 组合本地 stub + 所有依赖 stub + 本地 text → 完整 IR + # 读在线 FULL 产物 {sha1}.full.ll → 完整 IR（取代 stub + 依赖 stub/text 合并） combined_ir: bytes = stdlib.malloc(COMBINED_IR_SIZE) if combined_ir is None: VLogger.error(f"combined_ir 分配失败: {ent.Path}", "PhaseB") continue - combined_len: sizet = StubMergerBuild.BuildCombinedIR(temp_dir, ent.Sha1, combined_ir, COMBINED_IR_SIZE) + combined_len: sizet = StubMergerBuild.LoadFullIR(temp_dir, ent.Sha1, combined_ir, COMBINED_IR_SIZE) if combined_len == 0: - VLogger.error(f"BuildCombinedIR 失败: {ent.Path}", "PhaseB") + VLogger.error(f"读 full.ll 失败: {ent.Path}", "PhaseB") stdlib.free(combined_ir) continue - # 计时点 1：BuildCombinedIR 完成 + # 计时点 1：读 full.ll 完成 t_pb1: uint64 = w32.win32base.GetTickCount64() # 写 .ll（llc 线程只读此文件）；llc 本身改为循环后批量并行 @@ -1433,13 +1450,13 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, stdlib.free(src_fp_mi) continue - # 尝试 BuildCombinedIR（stub/text 应已由 Phase1 生成） + # 尝试读在线 FULL 产物 {sha1}.full.ll（应已由 Phase1/PhaseA 生成） inc_combined_mi: bytes = stdlib.malloc(COMBINED_IR_SIZE) inc_combined_len_mi: sizet = 0 if inc_combined_mi is not None: - inc_combined_len_mi = StubMergerBuild.BuildCombinedIR(temp_dir, inc_sha1_mi, inc_combined_mi, COMBINED_IR_SIZE) + inc_combined_len_mi = StubMergerBuild.LoadFullIR(temp_dir, inc_sha1_mi, inc_combined_mi, COMBINED_IR_SIZE) - # 如果 stub/text 不存在，翻译源文件并保存 stub + text，然后重试 + # 如果 full.ll 不存在，翻译源文件并保存 stub/text/meta/full，然后重试 if inc_combined_len_mi == 0 and inc_combined_mi is not None: # 计算 includes 文件的包名（相对 includes_dir 的目录部分） inc_pkg_mi: str = None @@ -1450,6 +1467,8 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, if tr_mi is not None: # PBP IR 缓冲：同 PHASE_A_IR_SIZE（parser.vp text.ll >1MB 截断根因） PBP_IR_SIZE: sizet = 8388608 + # FULL dump = stub + text，留 2× 余量 + PBP_FULL_SIZE: sizet = 16777216 # 保存 stub.ll (切片路径) inc_stub_buf: bytes = stdlib.malloc(PBP_IR_SIZE) if inc_stub_buf is not None: @@ -1478,18 +1497,40 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, stdlib.free(inc_text_buf) # save {sha1}.meta（结构化 sidecar：函数清单/结构体布局/虚表） StubMergerIR.WriteModuleMeta(tr_mi.Module, temp_dir, td_len_mi, inc_sha1_mi) - # 重试 BuildCombinedIR - inc_combined_len_mi = StubMergerBuild.BuildCombinedIR(temp_dir, inc_sha1_mi, inc_combined_mi, COMBINED_IR_SIZE) - # OOM 清理：stub/text 已写盘，回收本 includes 文件翻译瞬时数据 - # （AST/tokens/lexer/IR/Handler）；BuildCombinedIR 只读文件，不受影响 + # save {sha1}.full.ll（在线 FULL 产物） + inc_full_buf: bytes = stdlib.malloc(PBP_FULL_SIZE) + if inc_full_buf is None: + VLogger.error("full.ll 缓冲区分配失败", "PhaseB+") + tr_mi.dump_ir(inc_full_buf, PBP_FULL_SIZE, llvmlite.OUTPUT_FULL) + inc_full_len: sizet = string.strlen(inc_full_buf) + inc_full_path: str = StubMergerCollect._sliced_path(temp_dir, td_len_mi, inc_sha1_mi, "full.ll") + if inc_full_path is not None: + ifff: rawptr[fileio.File] = fileio.File(inc_full_path, fileio.MODE.W) + if not ifff.closed: + ifff.write(inc_full_buf, inc_full_len) + ifff.close() + stdlib.free(inc_full_path) + stdlib.free(inc_full_buf) + # 重试读 full.ll + inc_combined_len_mi = StubMergerBuild.LoadFullIR(temp_dir, inc_sha1_mi, inc_combined_mi, COMBINED_IR_SIZE) + # OOM 清理：stub/text/meta/full 已写盘，回收本 includes 文件翻译瞬时数据 + # （AST/tokens/lexer/IR/Handler）；读取只碰文件，不受影响 mb.reset_to(mark2) if inc_combined_len_mi == 0: - # 诊断：检查 stub.ll/text.ll 是否存在 + # 诊断：检查 full.ll/stub.ll/text.ll 是否存在 + diag_full: str = StubMergerCollect._sliced_path(temp_dir, td_len_mi, inc_sha1_mi, "full.ll") diag_stub: str = StubMergerCollect._sliced_path(temp_dir, td_len_mi, inc_sha1_mi, "stub.ll") diag_text: str = StubMergerCollect._sliced_path(temp_dir, td_len_mi, inc_sha1_mi, "text.ll") + diag_full_ex: int = 0 diag_stub_ex: int = 0 diag_text_ex: int = 0 + if diag_full is not None: + df_diag_f: rawptr[fileio.File] = fileio.File(diag_full, fileio.MODE.R) + if not df_diag_f.closed: + diag_full_ex = 1 + df_diag_f.close() + stdlib.free(diag_full) if diag_stub is not None: df_diag_s: rawptr[fileio.File] = fileio.File(diag_stub, fileio.MODE.R) if not df_diag_s.closed: @@ -1502,7 +1543,7 @@ def RunMultiFileProject(mb: memhub.MemBuddy*, diag_text_ex = 1 df_diag_t.close() stdlib.free(diag_text) - VLogger.error(f"BuildCombinedIR 失败: {src_fp_mi} (stub={diag_stub_ex} text={diag_text_ex} is_decl={is_decl_mi})", "PhaseB+") + VLogger.error(f"读 full.ll 失败: {src_fp_mi} (full={diag_full_ex} stub={diag_stub_ex} text={diag_text_ex} is_decl={is_decl_mi})", "PhaseB+") if inc_combined_mi is not None: stdlib.free(inc_combined_mi) stdlib.free(check_pat_mi) diff --git a/App/lib/Stub/StubMerger/StubMergerBuild.vp b/App/lib/Stub/StubMerger/StubMergerBuild.vp index 1de400a..d5b70d1 100644 --- a/App/lib/Stub/StubMerger/StubMergerBuild.vp +++ b/App/lib/Stub/StubMerger/StubMergerBuild.vp @@ -1389,6 +1389,40 @@ def HoistTypeDefsBeforeDeclares(buf: bytes, length: sizet) -> sizet: stdlib.free(tmp) return new_len +# ============================================================ +# LoadFullIR - 读 {temp_dir}/{sha1}.full.ll 到 out_buf（④-c：合并期改用在线 FULL 产物） +# +# 与 BuildCombinedIR 同契约：返回写入长度（sizet；0=失败），out_buf 以 '\0' 结尾。 +# full.ll 由翻译点在模型在线时 dump_ir(OUTPUT_FULL) 产出 —— 自足的单模块 IR +# （类型定义 + external 声明 + 定义 + DI 一次齐全，类型定义天然在声明之前）， +# 无需再把 stub + 依赖 stub/text 文本合并。 +# ============================================================ +def LoadFullIR(temp_dir: str, sha1: str, + out_buf: bytes, out_size: sizet) -> sizet: + """读取 {sha1}.full.ll 到 out_buf；返回长度（0=失败）""" + if temp_dir is None or sha1 is None or out_buf is None or out_size == 0: + return 0 + td_len: sizet = string.strlen(temp_dir) + p: str = StubMergerCollect._sliced_path(temp_dir, td_len, sha1, "full.ll") + if p is None: + return 0 + f: rawptr[fileio.File] = fileio.File(p, fileio.MODE.R) + stdlib.free(p) + out_buf[0] = '\0' + if f.closed: + return 0 + if f.size() >= out_size: + f.close() + VLogger.error("full.ll 超过 IR 缓冲上限", "StubMerger") + return 0 + rd: int64 = f.read_all(out_buf, out_size) + f.close() + if rd <= 0: + return 0 + out_buf[rd] = '\0' + return sizet(rd) + + # ============================================================ # BuildCombinedIR - 组合本地 stub + 依赖 stubs + 本地 text → 完整 IR #