I have an old Fortran program from 1983 I could never get working of X86, getting an unexpected NaN when it should have converted an input digit to a real*8. I could never track down the exact point of failure (the lack of debugger support didn't help). This morning I decided to attack it again with Fortran X8.5-7 and got it to not fail if I added a couple extraneous type* statements in 2 of the routines, the failure went away. After a couple more cycles adding/removing output statements, I compiled the code with /real_size=64 and it worked without the extra statements. All the routines set implicit real*8 or explicitly declare the variables, so /real_size=64 shouldn't have been necessary. Compiling with /list/show=all/real_size=32 didn't show any r*4 variables.
This morning I installed V8.5-8 (OpenVMS 9.2-2) and got the same behavior (albeit better debugger support).
Dusty deck Fortran, strange real*8 /real_size=32 interaction.
Re: Dusty deck Fortran, strange real*8 /real_size=32 interaction.
Can you give a short example? So no difference between the default, /REAL=32, and /REAL=64 listings but you say that using /REAL=64 "fixed" it? I'll assume that if you looked at the generated code with ANAL/OBJ/DISA, the "bad" vs "good" modules have different code?
-
jonesd
Topic author - Valued Contributor
- Posts: 78
- Joined: Mon Aug 09, 2021 7:59 pm
- Reputation: 0
- Status: Offline
Re: Dusty deck Fortran, strange real*8 /real_size=32 interaction.
I didn't until you mentioned it, and yes they do have different code (and I had to learn about /page=section and ignore=header=3). The section size is 4 bytes longer for the /real_size=32 compile, which incurs several differences just due to jump offsets. The extra 4 bytes appear to be a cvtsi2ssl instruction at line 225 in the rs32 object. Another notable difference is the movq at line 65 where the constant changes from $3FB99999A0000000 to $3FB999999999999A.jreagan wrote: ↑Thu Feb 08, 2024 9:41 amCan you give a short example? So no difference between the default, /REAL=32, and /REAL=64 listings but you say that using /REAL=64 "fixed" it? I'll assume that if you looked at the generated code with ANAL/OBJ/DISA, the "bad" vs "good" modules have different code?
https://drive.google.com/file/d/15nSbvv ... sp=sharing
Code: Select all
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
File USER:[JONESD.UTILITY.BOMAR]OPEXEC-rs32.anl;5 | File USER:[JONESD.UTILITY.BOMAR]OPEXEC-rs64.anl;5
----------------------------------------------- 11 -------------------------------------------------------------------------------------------- 11 --------------------------------------------
Section size 1032 bytes | Section size 1028 bytes
|
CPI0_0: | CPI0_0:
00 00 00000000: addb %al,(%rax) | 00 00 00000000: addb %al,(%rax)
----------------------------------------------- 20 -------------------------------------------------------------------------------------------- 20 --------------------------------------------
00 00 00000008: addb %al,(%rax) | 9A 00000008: ??
3F B9 99 99 A0 00 0000000A: addb %ah,3FB99999(%rax) | 99 00000009: clq
| 99 0000000A: clq
| 99 0000000B: clq
| 99 0000000C: clq
| 99 0000000D: clq
| 89 48 55 3F B9 0000000E: movl $8948553F,%ecx
OPEXEC:: | OPEXEC::
55 00000010: pushq %rbp | 55 00000010: pushq %rbp
E5 89 48 00000011: movq %rsp,%rbp | E5 89 48 00000011: movq %rsp,%rbp
----------------------------------------------- 65 -------------------------------------------------------------------------------------------- 70 --------------------------------------------
3F B9 99 99 A0 00 00 00 B9 48 000000C9: movq $3FB99999A0000000,%rcx | 3F B9 99 99 99 99 99 9A B9 48 000000C9: movq $3FB999999999999A,%rcx
08 89 48 000000D3: movq %rcx,(%rax) | 08 89 48 000000D3: movq %rcx,(%rax)
00 00 00 00 05 8B 48 000000D6: movq var$0003@GOTPCREL(%rip),%rax | 00 00 00 00 05 8B 48 000000D6: movq var$0003@GOTPCREL(%rip),%rax
00 63 48 000000DD: movslq (%rax),%rax | 00 63 48 000000DD: movslq (%rax),%rax
----------------------------------------------- 72 -------------------------------------------------------------------------------------------- 77 --------------------------------------------
00 00 02 DD E9 000000F3: jmpq 00000000000003D5 | 00 00 02 D9 E9 000000F3: jmpq 00000000000003D1
00 00 00 00 1D 8B 48 000000F8: movq INSTRC@GOTPCREL(%rip),%rbx | 00 00 00 00 1D 8B 48 000000F8: movq INSTRC@GOTPCREL(%rip),%rbx
00 00 00 00 25 8B 4C 000000FF: movq IDUM@GOTPCREL(%rip),%r12 | 00 00 00 00 25 8B 4C 000000FF: movq IDUM@GOTPCREL(%rip),%r12
00 00 00 7D E9 00000106: jmpq 0000000000000188 | 00 00 00 7D E9 00000106: jmpq 0000000000000188
----------------------------------------------- 126 ------------------------------------------------------------------------------------------- 131 -------------------------------------------
00 00 02 2C 0D 8D 48 000001C5: leaq JTI0_0(%rip),%rcx | 00 00 02 28 0D 8D 48 000001C5: leaq JTI0_0(%rip),%rcx
81 04 63 48 000001CC: movslq (%rcx,%rax,4),%rax | 81 04 63 48 000001CC: movslq (%rcx,%rax,4),%rax
C8 01 48 000001D0: addq %rcx,%rax | C8 01 48 000001D0: addq %rcx,%rax
E0 FF 000001D3: jmpq *%rax | E0 FF 000001D3: jmpq *%rax
----------------------------------------------- 211 ------------------------------------------------------------------------------------------- 216 -------------------------------------------
00 00 00 88 8C 0F 00000349: jl 00000000000003D7 | 00 00 00 84 8C 0F 00000349: jl 00000000000003D3
00 00 00 00 35 8B 48 0000034F: movq HVALU@GOTPCREL(%rip),%rsi | 00 00 00 00 35 8B 48 0000034F: movq HVALU@GOTPCREL(%rip),%rsi
DF 89 48 00000356: movq %rbx,%rdi | DF 89 48 00000356: movq %rbx,%rdi
02 00 B8 66 00000359: movw $0200,%ax | 02 00 B8 66 00000359: movw $0200,%ax
----------------------------------------------- 218 ------------------------------------------------------------------------------------------- 223 -------------------------------------------
67 EB 0000036C: jmpb 00000000000003D5 | 63 EB 0000036C: jmpb 00000000000003D1
00 00 00 00 0D 8B 48 0000036E: movq var$0005@GOTPCREL(%rip),%rcx | 00 00 00 00 0D 8B 48 0000036E: movq var$0005@GOTPCREL(%rip),%rcx
F8 C1 54 10 0F F2 00000375: movsd -08(%rcx,%rax,8),%xmm2 | F8 C1 54 10 0F F2 00000375: movsd -08(%rcx,%rax,8),%xmm2
D0 75 8B 48 0000037B: movq -30(%rbp),%rsi | D0 75 8B 48 0000037B: movq -30(%rbp),%rsi
----------------------------------------------- 225 ------------------------------------------------------------------------------------------- 230 -------------------------------------------
0E 2A 0F F3 00000391: cvtsi2ssl (%rsi),%xmm1 | C0 57 0F 00000391: xorps %xmm0,%xmm0
C0 57 0F 00000395: xorps %xmm0,%xmm0 | 06 2A 0F F2 00000394: cvtsi2sdl (%rsi),%xmm0
C1 5A 0F F3 00000398: cvtss2sd %xmm1,%xmm0 | 00 00 00 00 0D 8B 48 00000398: movq var$0006@GOTPCREL(%rip),%rcx
00 00 00 00 0D 8B 48 0000039C: movq var$0006@GOTPCREL(%rip),%rcx | 09 10 0F F2 0000039F: movsd (%rcx),%xmm1
09 10 0F F2 000003A3: movsd (%rcx),%xmm1 | C1 59 0F F2 000003A3: mulsd %xmm1,%xmm0
C1 59 0F F2 000003A7: mulsd %xmm1,%xmm0 | C2 58 0F F2 000003A7: addsd %xmm2,%xmm0
C2 58 0F F2 000003AB: addsd %xmm2,%xmm0 | F8 C2 44 11 0F F2 000003AB: movsd %xmm0,-08(%rdx,%rax,8)
F8 C2 44 11 0F F2 000003AF: movsd %xmm0,-08(%rdx,%rax,8) | FF FF FC 47 15 10 0F F2 000003B1: movsd CPI0_0(%rip),%xmm2
FF FF FC 43 15 10 0F F2 000003B5: movsd CPI0_0(%rip),%xmm2 | D1 2E 0F 66 000003B9: ucomisd %xmm1,%xmm2
D1 2E 0F 66 000003BD: ucomisd %xmm1,%xmm2 | 0C 76 000003BD: jbe 00000000000003CB
0C 76 000003C1: jbe 00000000000003CF | FF FF FC 41 0D 59 0F F2 000003BF: mulsd CPI0_1(%rip),%xmm1
FF FF FC 3D 0D 59 0F F2 000003C3: mulsd CPI0_1(%rip),%xmm1 | 09 11 0F F2 000003C7: movsd %xmm1,(%rcx)
09 11 0F F2 000003CB: movsd %xmm1,(%rcx) | 00 00 00 23 06 C7 000003CB: movl $00000023,(%rsi)
00 00 00 23 06 C7 000003CF: movl $00000023,(%rsi) | 03 8B 000003D1: movl (%rbx),%eax
03 8B 000003D5: movl (%rbx),%eax | F9 C0 83 000003D3: addl $-07,%eax
F9 C0 83 000003D7: addl $-07,%eax | 03 F8 83 000003D6: cmpl $03,%eax
03 F8 83 000003DA: cmpl $03,%eax | 00 00 00 00 05 8B 48 000003D9: movq NEWENT@GOTPCREL(%rip),%rax
00 00 00 00 05 8B 48 000003DD: movq NEWENT@GOTPCREL(%rip),%rax | 00 92 0F 000003E0: setb (%rax)
00 92 0F 000003E4: setb (%rax) | 18 C4 83 48 000003E3: addq $18,%rsp
18 C4 83 48 000003E7: addq $18,%rsp | 5B 000003E7: popq %rbx
5B 000003EB: popq %rbx | 5C 41 000003E8: popq %r12
5C 41 000003EC: popq %r12 | 5D 41 000003EA: popq %r13
5D 41 000003EE: popq %r13 | 5E 41 000003EC: popq %r14
5E 41 000003F0: popq %r14 | 5F 41 000003EE: popq %r15
5F 41 000003F2: popq %r15 | 5D 000003F0: popq %rbp
5D 000003F4: popq %rbp | C3 000003F1: retq
C3 000003F5: retq | 90 90 000003F2: .byte 0x90,0x90
90 90 000003F6: .byte 0x90,0x90 | JTI0_0:
JTI0_0: | FF FF FD E1 000003F4: .byte 0xE1,0xFD,0xFF,0xFF
FF FF FE 2A FF FF FD DD 000003F8: .byte 0xDD,0xFD,0xFF,0xFF,0x2A,0xFE,0xFF,0x | FF FF FE 55 FF FF FE 2E 000003F8: .byte 0x2E,0xFE,0xFF,0xFF,0x55,0xFE,0xFF,0x
FF FF FD 46 FF FF FE 51 00000400: .byte 0x51,0xFE,0xFF,0xFF,0x46,0xFD,0xFF,0x | FF FF FD 4A 00000400: .byte 0x4A,0xFD,0xFF,0xFF
<FF> | <FF>
|
The analysis uncovered NO errors. | The analysis uncovered NO errors.
----------------------------------------------- 264 ------------------------------------------------------------------------------------------- 269 -------------------------------------------
ANALY/OBJECT/DISASSEM/PAGE=SEPARATE OPEXEC-rs32.obj/OUT=OPEXEC-rs32.anl | ANALY/OBJECT/DISASSEM/PAGE=SEPARATE OPEXEC-rs64.obj/OUT=OPEXEC-rs64.anl
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Number of difference sections found: 9
Number of difference records found: 45
DIFFERENCES /IGNORE=(HEADER=3)/WIDTH=192/MATCH=3/OUTPUT=USER:[JONESD.UTILITY.BOMAR]OPEXEC-rs32-64.diff;5/PARALLEL=3-
USER:[JONESD.UTILITY.BOMAR]OPEXEC-rs32.anl;5-
USER:[JONESD.UTILITY.BOMAR]OPEXEC-rs64.anl;5
Last edited by jonesd on Fri Feb 09, 2024 2:24 am, edited 1 time in total.