lattice / quda

QUDA is a library for performing calculations in lattice QCD on GPUs.
https://lattice.github.io/quda
Other
286 stars 94 forks source link

lib/blas_magma.cpp compilation bug #139

Closed detar closed 10 years ago

detar commented 10 years ago

The lib/Makefile compilation command on line 135 for blas_magma.cpp needs fixing.

maddyscientist commented 10 years ago

I think this stems from MAGMA_INCLUDE and MAGMA_FLAGS not being defined: they are referred to in lib/Makefile, but are not set anywhere.

Assigning to Alexei.

maddyscientist commented 10 years ago

Moreover, Alexei, you should have a configure option to enable/disable the use of MAGMA, with the default option being to have it disabled so that out of the box compilation should work.

detar commented 10 years ago

Here is what looks like one more needed fix --- Alexei?

index c1af46a..c529f3a 100644 --- a/make.inc.in +++ b/make.inc.in @@ -109,7 +109,7 @@ ifeq ($(strip $(BUILD_MPI)), yes) MPI_CFLAGS = MPI_LDFLAGS = MPI_LIBS =

On 6/9/2014 10:11 PM, mikeaclark wrote:

Moreover, Alexei, you should have a configure option to enable/disable the use of MAGMA, with the default option being to have it disabled so that out of the box compilation should work.

— Reply to this email directly or view it on GitHub https://github.com/lattice/quda/issues/139#issuecomment-45572016.

AlexVaq commented 10 years ago

Alexei is here in Cyprus, at the linksceem meeting. I spoke with him about these issues, he told me he prepared some fix, but apparently it has not been committed yet. I'll try to speak with him later.

On Tue, Jun 10, 2014 at 3:23 PM, Carleton DeTar notifications@github.com wrote:

Here is what looks like one more needed fix --- Alexei?

index c1af46a..c529f3a 100644 --- a/make.inc.in +++ b/make.inc.in @@ -109,7 +109,7 @@ ifeq ($(strip $(BUILD_MPI)), yes) MPI_CFLAGS = MPI_LDFLAGS = MPI_LIBS =

  • INC += -DMPI_COMMS $(MPI_CFLAGS) -I$(MPI_HOME)/include/mpi
  • INC += -DMPI_COMMS $(MPI_CFLAGS) -I$(MPI_HOME)/include LIB += $(MPI_LDFLAGS) $(MPI_LIBS) COMM_OBJS = comm_mpi.o endif

On 6/9/2014 10:11 PM, mikeaclark wrote:

Moreover, Alexei, you should have a configure option to enable/disable the use of MAGMA, with the default option being to have it disabled so that out of the box compilation should work.

— Reply to this email directly or view it on GitHub https://github.com/lattice/quda/issues/139#issuecomment-45572016.

— Reply to this email directly or view it on GitHub https://github.com/lattice/quda/issues/139#issuecomment-45606386.

alexstrel commented 10 years ago

Fixed, I disabled compilation with MAGMA library. Should work. Sorry for inconvenience, Alexei


From: Alejandro Vaquero [notifications@github.com] Sent: Tuesday, June 10, 2014 7:57 AM To: lattice/quda Cc: Alexei Strelchenko Subject: Re: [quda] lib/blas_magma.cpp compilation bug (#139)

Alexei is here in Cyprus, at the linksceem meeting. I spoke with him about these issues, he told me he prepared some fix, but apparently it has not been committed yet. I'll try to speak with him later.

On Tue, Jun 10, 2014 at 3:23 PM, Carleton DeTar notifications@github.com wrote:

Here is what looks like one more needed fix --- Alexei?

index c1af46a..c529f3a 100644 --- a/make.inc.in +++ b/make.inc.in @@ -109,7 +109,7 @@ ifeq ($(strip $(BUILD_MPI)), yes) MPI_CFLAGS = MPI_LDFLAGS = MPI_LIBS =

  • INC += -DMPI_COMMS $(MPI_CFLAGS) -I$(MPI_HOME)/include/mpi
  • INC += -DMPI_COMMS $(MPI_CFLAGS) -I$(MPI_HOME)/include LIB += $(MPI_LDFLAGS) $(MPI_LIBS) COMM_OBJS = comm_mpi.o endif

On 6/9/2014 10:11 PM, mikeaclark wrote:

Moreover, Alexei, you should have a configure option to enable/disable the use of MAGMA, with the default option being to have it disabled so that out of the box compilation should work.

— Reply to this email directly or view it on GitHub https://github.com/lattice/quda/issues/139#issuecomment-45572016.

— Reply to this email directly or view it on GitHub https://github.com/lattice/quda/issues/139#issuecomment-45606386.

— Reply to this email directly or view it on GitHubhttps://github.com/lattice/quda/issues/139#issuecomment-45609543.

maddyscientist commented 10 years ago

More fixes were needed: the flags in make.in.inc were not being set in any way by configure, there were bugs left in blas_magma.cpp when MAGMA_LIB was left undefined, and a bug in lib/Makefile meant the include directories were not being read. Should be fixed now.

To enable magma, you now should add the configure option --with-magma=path, where path is the location of the MAGMA installation. Not specifying this option will disable the use of MAGMA and blas_magma.cpp becomes a series of dummy functions.