如何修复Gem5环境下LAPACK的Fortran文本解析错误？

数据老陈

2026-7-20

LAPACK在Gem5仿真中输入解析错误的修复方案

问题背景

在Windows 10的WSL2 Docker环境中执行LAPACK的xlintstrfz二进制并通过< ztest_rfp.in重定向输入时，程序能正常完成测试；但将配置迁移到Gem5的Python脚本后，程序读取NRHS参数时触发Fortran运行时错误：

Fortran runtime error: Bad integer for item 3 in list input

错误位于zchkrfp.f第166行的READ( NIN, FMT = * )( NSVAL( I ), I = 1, NNS )语句，仿真仅能处理输入文件前几行就崩溃。

相关信息

Docker中正常运行命令及输出

./LIN/xlintstrfz < ztest_rfp.in

输出：

Tests of the COMPLEX*16 LAPACK RFP routines 
 LAPACK VERSION 3.*.0

 The following parameter values will be used:
    N   :       0     1     2     3     5     6    10    11    50
    NRHS:       1     2    15
    TYPE:       1     2     3     4     5     6     7     8     9

 Routines pass computational tests if test ratio is less than   30.00

 Relative machine underflow is taken to be    0.222507-307
 Relative machine overflow  is taken to be    0.179769+309
 Relative machine precision is taken to be    0.111022D-15

 COMPLEX*16 RFP routines passed the tests of the error exits

 All tests for ZPF drivers  passed the threshold (   2304 tests run)
 All tests for ZLANHF auxiliary routine passed the threshold (   384 tests run)
 All tests for the RFP conversion routines passed (   72 tests run)
 All tests for ZTFSM auxiliary routine passed the threshold (  7776 tests run)
 All tests for ZHFRK auxiliary routine passed the threshold (   2592 tests run)

 End of tests
 Total time used =         0.71 seconds

Gem5的Python配置代码

binary = os.path.join(
    thispath,
    "../../../",
    "tests/test-progs/lapack/TESTING/LIN/xlintstrfz",
)

process = Process()
process.cmd = [binary]
process.input = 'tests/test-progs/lapack/TESTING/ztest_rfp.in'

Gem5错误输出

gem5 Simulator System.  https://www.gem5.org
gem5 is copyrighted software; use the --copyright option for details.

gem5 version 22.1.0.0
gem5 compiled May 18 2023 15:42:00
gem5 started May 24 2023 00:42:19
gem5 executing on 8fa8d1b910d8, pid 26054
command line: build/GCN3_X86/gem5.opt configs/learning_gem5/part1/simple.py

Global frequency set at 1000000000000 ticks per second
build/GCN3_X86/base/statistics.hh:280: warn: One of the stats is a legacy stat. Legacy stat is deprecated.
0: system.remote_gdb: listening for remote gdb on port 7008
Beginning simulation!
build/GCN3_X86/sim/simulate.cc:192: info: Entering event queue @ 0.  Starting simulation...

 Tests of the COMPLEX*16 LAPACK RFP routines 
 LAPACK VERSION 3.*.0

 The following parameter values will be used:
    N   :       0     1     2     3     5     6    10    11    50
build/GCN3_X86/sim/mem_state.cc:443: info: Increasing stack size by one page.
At line 166 of file zchkrfp.f (unit = 5, file = 'stdin')
Fortran runtime error: Bad integer for item 3 in list input

Error termination.

关键Fortran代码片段（zchkrfp.f第153行起）

*
*     Read the values of NRHS
*
      READ( NIN, FMT = * )NNS
      IF( NNS.LT.1 ) THEN
         WRITE( NOUT, FMT = 9996 )' NNS', NNS, 1
         NNS = 0
         FATAL = .TRUE.
      ELSE IF( NNS.GT.MAXIN ) THEN
         WRITE( NOUT, FMT = 9995 )' NNS', NNS, MAXIN
         NNS = 0
         FATAL = .TRUE.
      END IF
      READ( NIN, FMT = * )( NSVAL( I ), I = 1, NNS )
      DO 30 I = 1, NNS
         IF( NSVAL( I ).LT.0 ) THEN
            WRITE( NOUT, FMT = 9996 )'NRHS', NSVAL( I ), 0
            FATAL = .TRUE.
         ELSE IF( NSVAL( I ).GT.MAXRHS ) THEN
            WRITE( NOUT, FMT = 9995 )'NRHS', NSVAL( I ), MAXRHS
            FATAL = .TRUE.
         END IF
   30 CONTINUE
      IF( NNS.GT.0 )
     $   WRITE( NOUT, FMT = 9993 )'NRHS', ( NSVAL( I ), I = 1, NNS )
*

输入文件ztest_rfp.in内容

Data file for testing COMPLEX*16 LAPACK linear equation routines RFP format
9                              Number of values of N (at most 9)
0 1 2 3 5 6 10 11 50           Values of N
3                              Number of values of NRHS (at most 9)
1 2 15                         Values of NRHS (number of right hand sides)
9                              Number of matrix types (list types on next line if 0 < NTYPES <  9)
1 2 3 4 5 6 7 8 9              Matrix Types
30.0                           Threshold value of test ratio
T                              Put T to test the error exits

问题分析

核心原因是Gem5的process.input机制与shell的输入重定向<处理逻辑不同：

Shell重定向是将文件作为标准输入流式传递，Fortran的自由格式读取FMT=*能按预期识别换行和空白分隔的数值；
Gem5的process.input直接传递文件内容时，可能存在行尾格式（如CRLF转LF不彻底）、缓冲区读取方式差异，导致Fortran读取到的内容不是预期的结构化数值流，进而在解析NRHS的第三个数值时出错。

修复方案

方案1：通过Shell命令执行（推荐）

修改Gem5的Python配置，让shell处理输入重定向，和Docker中的执行逻辑完全一致：

import os
binary = os.path.join(
    thispath,
    "../../../",
    "tests/test-progs/lapack/TESTING/LIN/xlintstrfz",
)
input_file = os.path.join(
    thispath,
    "../../../",
    "tests/test-progs/lapack/TESTING/ztest_rfp.in"
)

process = Process()
# 用bash -c执行带重定向的命令
process.cmd = ['bash', '-c', f'{binary} < {input_file}']

这种方式完全复用shell的输入处理逻辑，避免Gem5输入机制带来的兼容性问题。

方案2：修正输入文件换行格式

如果输入文件是Windows CRLF换行格式，WSL2和Gem5可能对换行解析不一致，导致Fortran读取时混入\r字符干扰数值解析。执行以下命令转换为Unix LF格式：

dos2unix tests/test-progs/lapack/TESTING/ztest_rfp.in

方案3：修改Fortran代码强制逐行解析

修改zchkrfp.f中的读取逻辑，先逐行读取字符串再解析数值，避免流式读取的格式问题：

*
*     Read the values of NRHS
*
      CHARACTER*256 line
      READ( NIN, '(A)' ) line
      READ( line, * )NNS
      IF( NNS.LT.1 ) THEN
         WRITE( NOUT, FMT = 9996 )' NNS', NNS, 1
         NNS = 0
         FATAL = .TRUE.
      ELSE IF( NNS.GT.MAXIN ) THEN
         WRITE( NOUT, FMT = 9995 )' NNS', NNS, MAXIN
         NNS = 0
         FATAL = .TRUE.
      END IF
      READ( NIN, '(A)' ) line
      READ( line, * )( NSVAL( I ), I = 1, NNS )
      DO 30 I = 1, NNS
         IF( NSVAL( I ).LT.0 ) THEN
            WRITE( NOUT, FMT = 9996 )'NRHS', NSVAL( I ), 0
            FATAL = .TRUE.
         ELSE IF( NSVAL( I ).GT.MAXRHS ) THEN
            WRITE( NOUT, FMT = 9995 )'NRHS', NSVAL( I ), MAXRHS
            FATAL = .TRUE.
         END IF
   30 CONTINUE
      IF( NNS.GT.0 )
     $   WRITE( NOUT, FMT = 9993 )'NRHS', ( NSVAL( I ), I = 1, NNS )
*

修改后重新编译LAPACK测试程序，再在Gem5中运行。

内容的提问来源于stack exchange，提问作者Nick Felker

如何修复Gem5环境下LAPACK的Fortran文本解析错误？

问题背景 #

相关信息 #