:13:14: error: unknown token in expression
vinsertf128 , %xmm15, %ymm15, %ymm15
./acados/external/blasfeo/kernel/avx/kernel_dgemv_4_lib4.S:4087:2: note: while in macro instantiation
INNER_STORE_4_VS_LIB4
^
make[2]: *** [external/blasfeo/CMakeFiles/blasfeo.dir/kernel/avx/kernel_dgemv_4_lib4.S.o] Error 1
make[1]: *** [external/blasfeo/CMakeFiles/blasfeo.dir/all] Error 2
I’ve also tried to compile blasfeo directly (pull from blasfeo GitHub). It has failed too. So is there any suggestion for solving this problem. Thanks.
other users have reported similar issues compiling BLASFEO assembly code on macOS (e.g. https://github.com/giaf/hpipm/issues/95), but since I don’t have access to a macOS system I couldn’t investigate the issue myself.
If you are willing to help, we can try to figure out the issue together.
In the specific case of the line you report as an example, in the BLASFEO source code it looks like vinsertf128 $1, %xmm15, %ymm15, %ymm15
instead of vinsertf128 , %xmm15, %ymm15, %ymm15
as in the error print you report.
As you can see, the “immediate” (i.e. numerical constant) has been somehow removed.
In case of exadecimal immediate, there is a similar issue, with e.g. $0x12 becoming x12.
Now, I could not find many hints of the issue googling around, but we can start trying this:
try to put a space in between the dollar symbol and the following number, e.g. $1 becomes $ 1.
If this still fails, can you double check that the compiler is set up to expect the correct assembly syntax? More specifically, BLASFEO uses AT&T syntax (which is the standard in linux/gcc), as opposite to Intel syntax with is more common in windows/intel compilers.
In case of gcc or clang compilers, this can be enforced by passing the compiler flag -masm=att.
The short term solution for MAC would be to set the BLASFEO_TARGET to GENERIC when compiling acados.
It would of course be great if you could help giaf to fix this issue properly.
Ok great, thanks for trying it out.
Don’t worry about correcting any file, I will do with some automatic bash script when I find some time one of these days, I just needed to know what works
Once done, I will ping you here to test it on your PC, on my machines it already works smooth so I won’t notice the difference.
And don’t worry about the -masm=att, it was just another hypothesis to test out, but since the space after the dollar makes the job, that was not the issue.
Hi,
thanks. When I change to master, it works now for blasfeo.
However, I’ve got new error
In file included from ./acados/external/hpipm/cond/d_cond_aux.c:107:
./external/hpipm/cond/x_cond_aux.c:807:108: error: too many arguments to function call, expected 12, have 13
…SYMV_L(nx[nn], nx[nn], 1.0, RSQrq+nn, nu[nn], nu[nn], Gammab+nn-1, 0, 1.0, rqz+nn, nu[nn], tmp_nuxM, 0);
Then I pull hpipm as master, then I’ve got errors
./acados/acados/ocp_nlp/ocp_nlp_cost_ls.c:796:58: error: too many arguments to function call, expected 12, have 13
0, 0.0, &work->tmp_ny, 0, &work->tmp_ny, 0);
^
./acados/external/blasfeo/include/blasfeo_d_blasfeo_api.h:120:1: note: ‘blasfeo_dsymv_l’ declared here
void blasfeo_dsymv_l(int m, double alpha, struct blasfeo_dmat *sA, int ai, int aj, struct blasfeo_dvec *sx, int xi, double beta, struct blas…
^
.acados/acados/ocp_nlp/ocp_nlp_cost_ls.c:817:63: error: too many arguments to function call, expected 12, have 13
0.0, &work->tmp_ny, 0, &work->tmp_ny, 0);
