Re: Help with new version of iqtree needed
Hi Andreas, I've update the sse3 patch to not enable sse3 and not define __SSE3, and added an sse2 code path to emulate the _mm_hadd_pd instruction where needed. It compiles, and the alignment sequence test reports the same "failures "and "passed" like the completely unpatched version. However, later I get errors like: ERROR: Checkpoint (example.phy.ckp.gz) indicates that a previous run successfully finished Use `-redo` option if you really want to redo the analysis and overwrite all output files. Command exited with non-zero status 2 with and without patching the code. best, Gert
Re: Help with new version of iqtree needed
Hello, > _mm_hadd_pd is a sse3 intrinsic so you have to pass -msse3 to allow > GCC to use it. However, this will cause a SIGILL on any amd64/i386 > processor without sse3 so instead the code should be replaced with > something more portable. Actually, the compile command defines -D__SSE3, and the according headers are only included if this define is set. I've tried to add a patch that removes the according tests in the CMakeLists.txt, but then I got /home/gerddie/debian-med/iqtree/phylokernel.h:19:2: error: #error "You must compile with SSE3 enabled!" within a function that does a so called horizontal add. I'll think of a patch to emulate the according code with sse2 instructions. Best, Gert
Re: Help with new version of iqtree needed
Hi James, thanks for the fast response: On Thu, Mar 10, 2016 at 01:49:23PM +, James Cowgill wrote: > On Thu, 2016-03-10 at 14:39 +0100, Andreas Tille wrote: > > Hi, > > > > I#m facing a C++ problem with the new version of iqtree. If I build the > > current state in Git[1] I get: > > > > ... > > [ 7%] Building C object pll/CMakeFiles/pll.dir/evaluateGenericSpecial.c.o > > cd /build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu/pll && /usr/bin/cc > > -DIQ_TREE -D_USE_PTHREADS -D__SSE3 -I/build/iqtree-1.4.0+dfsg > > -I/build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu -g -O2 > > -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time > > -D_FORTIFY_SOURCE=2 -pthread-o > > CMakeFiles/pll.dir/evaluateGenericSpecial.c.o -c > > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c > > In file included from > > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:46:0: > > /build/iqtree-1.4.0+dfsg/pll/pllInternal.h:150:30: warning: inline function > > 'bitcount_64_bit' declared but never defined > > extern __inline unsigned int bitcount_64_bit(uint64_t i); > > ^ > > In file included from /build/iqtree-1.4.0+dfsg/pll/pll.h:79:0, > > from /build/iqtree-1.4.0+dfsg/pll/mem_alloc.h:16, > > from > > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:31: > > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c: In function > > 'evaluateGTRCATPROT': > > /usr/lib/gcc/x86_64-linux-gnu/5/include/pmmintrin.h:86:1: error: inlining > > failed in call to always_inline '_mm_hadd_pd': target specific option > > mismatch > > _mm_hadd_pd (__m128d __X, __m128d __Y) > > ^ > > _mm_hadd_pd is a sse3 intrinsic so you have to pass -msse3 to allow GCC > to use it. However, this will cause a SIGILL on any amd64/i386 > processor without sse3 so instead the code should be replaced with > something more portable. Since I have no idea about SSE my attempt to fix #813436 obviously triggered this problem. I admit I have no idea how to deal with this sensibly. Kind regards Andreas. -- http://fam-tille.de
Re: Help with new version of iqtree needed
On Thu, 2016-03-10 at 14:39 +0100, Andreas Tille wrote: > Hi, > > I#m facing a C++ problem with the new version of iqtree. If I build the > current state in Git[1] I get: > > ... > [ 7%] Building C object pll/CMakeFiles/pll.dir/evaluateGenericSpecial.c.o > cd /build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu/pll && /usr/bin/cc > -DIQ_TREE -D_USE_PTHREADS -D__SSE3 -I/build/iqtree-1.4.0+dfsg > -I/build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu -g -O2 > -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time > -D_FORTIFY_SOURCE=2 -pthread-o > CMakeFiles/pll.dir/evaluateGenericSpecial.c.o -c > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c > In file included from > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:46:0: > /build/iqtree-1.4.0+dfsg/pll/pllInternal.h:150:30: warning: inline function > 'bitcount_64_bit' declared but never defined > extern __inline unsigned int bitcount_64_bit(uint64_t i); > ^ > In file included from /build/iqtree-1.4.0+dfsg/pll/pll.h:79:0, > from /build/iqtree-1.4.0+dfsg/pll/mem_alloc.h:16, > from > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:31: > /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c: In function > 'evaluateGTRCATPROT': > /usr/lib/gcc/x86_64-linux-gnu/5/include/pmmintrin.h:86:1: error: inlining > failed in call to always_inline '_mm_hadd_pd': target specific option mismatch > _mm_hadd_pd (__m128d __X, __m128d __Y) > ^ _mm_hadd_pd is a sse3 intrinsic so you have to pass -msse3 to allow GCC to use it. However, this will cause a SIGILL on any amd64/i386 processor without sse3 so instead the code should be replaced with something more portable. James signature.asc Description: This is a digitally signed message part
Help with new version of iqtree needed
Hi, I#m facing a C++ problem with the new version of iqtree. If I build the current state in Git[1] I get: ... [ 7%] Building C object pll/CMakeFiles/pll.dir/evaluateGenericSpecial.c.o cd /build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu/pll && /usr/bin/cc -DIQ_TREE -D_USE_PTHREADS -D__SSE3 -I/build/iqtree-1.4.0+dfsg -I/build/iqtree-1.4.0+dfsg/obj-x86_64-linux-gnu -g -O2 -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -pthread-o CMakeFiles/pll.dir/evaluateGenericSpecial.c.o -c /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c In file included from /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:46:0: /build/iqtree-1.4.0+dfsg/pll/pllInternal.h:150:30: warning: inline function 'bitcount_64_bit' declared but never defined extern __inline unsigned int bitcount_64_bit(uint64_t i); ^ In file included from /build/iqtree-1.4.0+dfsg/pll/pll.h:79:0, from /build/iqtree-1.4.0+dfsg/pll/mem_alloc.h:16, from /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:31: /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c: In function 'evaluateGTRCATPROT': /usr/lib/gcc/x86_64-linux-gnu/5/include/pmmintrin.h:86:1: error: inlining failed in call to always_inline '_mm_hadd_pd': target specific option mismatch _mm_hadd_pd (__m128d __X, __m128d __Y) ^ /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:2667:10: error: called from here tv = _mm_hadd_pd(tv, tv); ^ In file included from /build/iqtree-1.4.0+dfsg/pll/pll.h:79:0, from /build/iqtree-1.4.0+dfsg/pll/mem_alloc.h:16, from /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:31: /usr/lib/gcc/x86_64-linux-gnu/5/include/pmmintrin.h:86:1: error: inlining failed in call to always_inline '_mm_hadd_pd': target specific option mismatch _mm_hadd_pd (__m128d __X, __m128d __Y) ^ /build/iqtree-1.4.0+dfsg/pll/evaluateGenericSpecial.c:2700:10: error: called from here tv = _mm_hadd_pd(tv, tv); ^ pll/CMakeFiles/pll.dir/build.make:113: recipe for target 'pll/CMakeFiles/pll.dir/evaluateGenericSpecial.c.o' failed ... Any hint how to fix this? Kind regards Andreas. [1] https://anonscm.debian.org/git/debian-med/iqtree.git -- http://fam-tille.de
