Hi Michael,
On 2023-11-24 09:30, Andrius Merkys wrote:
On 2023-11-23 19:25, Michael R. Crusoe wrote:
It won't fix the FTBFS on armel / armhf; that's due to them being
detected as arm64 and inappropriate NEON intrinsics being used
Thanks for spotting this, I will give it a look, although I understand
very little about vectorization.
I successfully patched the build on armel by leaving NEON vectorization
for arm64 only:
--- a/libraries/vecmath/src/vecmath.cpp
+++ b/libraries/vecmath/src/vecmath.cpp
@@ -1,4 +1,4 @@
-#if defined(__ARM__) || defined(__ARM64__)
+#if defined(__ARM64__)
#include "neon_mathfun.h"
#else
#if !defined(__PNACL__)
--- a/openmmapi/include/openmm/internal/vectorize.h
+++ b/openmmapi/include/openmm/internal/vectorize.h
@@ -32,7 +32,7 @@
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*
*
--------------------------------------------------------------------------
*/
-#if defined(__ARM__) || defined(__ARM64__)
+#if defined(__ARM64__)
#include "vectorize_neon.h"
#elif defined(__PPC__)
#include "vectorize_ppc.h"
Build on armel succeeded, however, some tests segfault:
The following tests FAILED:
10 - TestReferenceCustomCentroidBondForce (SEGFAULT)
11 - TestReferenceCustomCompoundBondForce (SEGFAULT)
12 - TestReferenceCustomExternalForce (SEGFAULT)
13 - TestReferenceCustomGBForce (SEGFAULT)
15 - TestReferenceCustomIntegrator (SEGFAULT)
16 - TestReferenceCustomManyParticleForce (SEGFAULT)
17 - TestReferenceCustomNonbondedForce (Subprocess aborted)
18 - TestReferenceCustomTorsionForce (SEGFAULT)
69 - TestReferenceRpmd (Timeout)
110 - TestParser (SEGFAULT)
I gave TestParser a look with gdb:
(gdb) run
Starting program:
/home/merkys/openmm-8.0.0+dfsg/obj-arm-linux-gnueabi/TestParser
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/arm-linux-gnueabi/libthread_db.so.1".
Program received signal SIGSEGV, Segmentation fault.
Lepton::CompiledExpression::generateTwoArgCall
(this=this@entry=0xbefff004, c=..., dest=..., arg1=..., arg2=...,
function=<optimized out>) at
./libraries/lepton/src/CompiledExpression.cpp:511
511 invoke->setArg(0, arg1);
(gdb) print arg1
$1 = (asmjit::_abi_1_9::arm::Vec &) @0x4186f8:
{<asmjit::_abi_1_9::arm::Reg> = {<asmjit::_abi_1_9::BaseReg> =
{<asmjit::_abi_1_9::Operand> = {<asmjit::_abi_1_9::Operand_> =
{_signature = {_bits = 0},
_baseId = 0, _data = {0, 0}}, <No data fields>}, <No data
fields>}, <No data fields>}, <No data fields>}
My wild guess is that asmjit is incompatible with armel/armhf as in
standalone asmjit (see #1049872). Building openmm without asmjit support
seems possible, I will try that.
Best,
Andrius
