Opened 11 years ago
Closed 11 years ago
#38576 closed defect (duplicate)
Install of atlas fails - ESTIMATING THE NUMBER OF USEABLE REGISTERS FOR GEMM
| Reported by: | watsodw | Owned by: | Veence (Vincent) |
|---|---|---|---|
| Priority: | Normal | Milestone: | |
| Component: | ports | Version: | 2.1.3 |
| Keywords: | Cc: | cooljeanius (Eric Gallager), cmutel (Chris Mutel) | |
| Port: | atlas |
Description
see log.
Attachments (5)
Change History (18)
Changed 11 years ago by watsodw
comment:1 Changed 11 years ago by cooljeanius (Eric Gallager)
comment:2 Changed 11 years ago by watsodw
"uname -a " returns: Darwin m965611-mac.msfc.nasa.gov 10.8.0 Darwin Kernel Version 10.8.0: Tue Jun 7 16:33:36 PDT 2011; root:xnu-1504.15.3~1/RELEASE_I386 i386
comment:3 follow-up: 6 Changed 11 years ago by Veence (Vincent)
Well, this is one of isolated cases hard to fix. It might be a bug in the compiler – you could try to use the newest gcc48 instead or maybe clang-3.3.
The culprit is definitely here:
ERROR IN COMMAND: make mmcase pre=s loopO=JIK ta=T tb=N mb=80 nb=80 kb=80 lda=80 ldb=80 ldc=0 mu=1 nu=2 ku=80 alpha=1 beta=1 muladd=0 lat=2 cleanup=1 pfA=512 casnam="res/sJIKmmTN80_80x80x80_80x80x0_1x2x80_a1_b1_0x2_1_pf512_cf-1" > /dev/null 2>&1
But without further information, I can’t guess much. Can you try to located the file ERROR.LOG inside atlas build directory and see what is written inside?
comment:5 Changed 11 years ago by jmroot (Joshua Root)
| Cc: | vince@… removed |
|---|---|
| Owner: | changed from macports-tickets@… to vince@… |
Changed 11 years ago by watsodw
Changed 11 years ago by watsodw
| Attachment: | main.2.log added |
|---|
Changed 11 years ago by watsodw
| Attachment: | Stage1.log added |
|---|
comment:6 follow-up: 7 Changed 11 years ago by watsodw
Replying to vince@…:
Well, this is one of isolated cases hard to fix. It might be a bug in the compiler – you could try to use the newest gcc48 instead or maybe clang-3.3.
The culprit is definitely here:
ERROR IN COMMAND: make mmcase pre=s loopO=JIK ta=T tb=N mb=80 nb=80 kb=80 lda=80 ldb=80 ldc=0 mu=1 nu=2 ku=80 alpha=1 beta=1 muladd=0 lat=2 cleanup=1 pfA=512 casnam="res/sJIKmmTN80_80x80x80_80x80x0_1x2x80_a1_b1_0x2_1_pf512_cf-1" > /dev/null 2>&1But without further information, I can’t guess much. Can you try to located the file ERROR.LOG inside atlas build directory and see what is written inside?
I have attached the error.log, stage1.log and a new main.log. i tried again with gcc46 and gcc47, but no luck. I actually got it built with mpclang-3.2, but qrupdate keeps trying to update itself with the gcc46 atlas variant.
comment:7 Changed 11 years ago by watsodw
Replying to david.w.watson@…:
Replying to vince@…:
Well, this is one of isolated cases hard to fix. It might be a bug in the compiler – you could try to use the newest gcc48 instead or maybe clang-3.3.
The culprit is definitely here:
ERROR IN COMMAND: make mmcase pre=s loopO=JIK ta=T tb=N mb=80 nb=80 kb=80 lda=80 ldb=80 ldc=0 mu=1 nu=2 ku=80 alpha=1 beta=1 muladd=0 lat=2 cleanup=1 pfA=512 casnam="res/sJIKmmTN80_80x80x80_80x80x0_1x2x80_a1_b1_0x2_1_pf512_cf-1" > /dev/null 2>&1But without further information, I can’t guess much. Can you try to located the file ERROR.LOG inside atlas build directory and see what is written inside?
I have attached the error.log, stage1.log and a new main.log. i tried again with gcc46 and gcc47, but no luck. I actually got it built with mpclang-3.2, but qrupdate keeps trying to update itself with the gcc46 atlas variant.
I tried again and it looks, from my experience, that the issue is trying to build a universal build. I just successfully built atlas with gcc46, non-universal build.
comment:8 Changed 11 years ago by Veence (Vincent)
I see. There is a -msse4.2 flag inserted in the compiler options while you are compiling for an i386 arch (-m32), which is obviously an explosive mix. This is most certainly a bug in the Portfile that I’ll try to fix as soon as I’m back home Friday. Thanks for posting the extra information!
comment:10 follow-up: 11 Changed 11 years ago by Veence (Vincent)
Please try again. I have fixed some major bugs in the Portfile, it should now behave correctly while building universal
comment:11 Changed 11 years ago by watsodw
Replying to vince@…:
Please try again. I have fixed some major bugs in the Portfile, it should now behave correctly while building universal
Tried again, but fails again, but different reason. See main3.log.
Changed 11 years ago by watsodw
comment:12 Changed 11 years ago by Veence (Vincent)
Good. I’ve the same error at the same point, which means that:
- It is not a bug due to fancy transient stuff;
- I’ll be able to investigate on my own build.
I’ll have a look tonight.
comment:13 Changed 11 years ago by Veence (Vincent)
| Resolution: | → duplicate |
|---|---|
| Status: | new → closed |
This is a dup of #38770 that should be fixed now.
Relevant part of log:
OP, could you post some of your system information? "
uname -a" and that sort of thing...