aarch64 libunwind compatibility fix

Grid/GridStd.h

@@ -42,6 +42,7 @@ void GridAbort(void);
 #define ASSLOG(A) ::write(STDERR_FILENO,A,::strlen(A));
 #ifdef HAVE_EXECINFO_H
 #define GRID_ASSERT(b) if(!(b)) {					\
+    fflush(stdout); \
     ASSLOG(" GRID_ASSERT failure: ");					\
     ASSLOG(__FILE__);							\
     ASSLOG(" : ");							\

Grid/algorithms/FFT.h

@@ -36,6 +36,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <hipfft/hipfft.h>
 #endif
 
+#if !defined(GRID_CUDA) && !defined(GRID_HIP)
 #ifdef HAVE_FFTW
 #if defined(USE_MKL) || defined(GRID_SYCL)
 #include <fftw/fftw3.h>
@@ -43,6 +44,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
 #include <fftw3.h>
 #endif
 #endif
+#endif
 
 NAMESPACE_BEGIN(Grid);
 
@@ -138,7 +140,7 @@ public:
     
   inline static void fftw_execute_dft(const FFTW_plan p,FFTW_scalar *in,FFTW_scalar *out, int sign) {
     if ( sign == forward ) cufftExecZ2Z(p,in,out,CUFFT_FORWARD);
-    else                   cufftExecZ2Z(p,in,out,CUFFT_BACKWARD);
+    else                   cufftExecZ2Z(p,in,out,CUFFT_INVERSE);
     accelerator_barrier();
   }
   inline static void fftw_destroy_plan(const FFTW_plan p) {
@@ -165,7 +167,7 @@ public:
     
   inline static void fftw_execute_dft(const FFTW_plan p,FFTW_scalar *in,FFTW_scalar *out, int sign) {
     if ( sign == forward ) cufftExecC2C(p,in,out,CUFFT_FORWARD);
-    else                   cufftExecC2C(p,in,out,CUFFT_BACKWARD);
+    else                   cufftExecC2C(p,in,out,CUFFT_INVERSE);
     accelerator_barrier();
   }
   inline static void fftw_destroy_plan(const FFTW_plan p) {
@@ -174,6 +176,7 @@ public:
 };
 #endif
 
+#if !defined(GRID_CUDA) && !defined(GRID_HIP)
 #ifdef HAVE_FFTW
 template<> struct FFTW<ComplexD> {
 public:
@@ -216,6 +219,7 @@ public:
   }
 };
 #endif
+#endif
 
 class FFT {
 private:

Grid/communicator/Communicator_mpi3.cc

@@ -349,7 +349,6 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<MpiCommsRequest_t> &
 
   GRID_ASSERT(dest != _processor);
   GRID_ASSERT(from != _processor);
-  GRID_ASSERT(bytes/(sizeof(int32_t))<= 2*1024*1024*1024);
   int tag;
 
   tag= dir+from*32;

Grid/util/Init.cc

@@ -755,7 +755,7 @@ void Grid_generic_handler(int sig,siginfo_t *si,void * ptr)
   sig_print_uint(si->si_code);
   SIGLOG("\n");
 
-  ucontext_t *uc= (ucontext_t *)ptr;
+  unw_context_t *uc= (unw_context_t *)ptr;
   
   SIGLOG("Backtrace:\n");
 #ifdef HAVE_UNWIND

configure.ac

@@ -292,13 +292,14 @@ AC_ARG_ENABLE([accelerator],
 case ${ac_ACCELERATOR} in
     cuda)
       echo CUDA acceleration
-      LIBS="${LIBS} -lcuda -lcublas"
+      LIBS="${LIBS} -lcuda -lcublas -lcufft"
       AC_DEFINE([GRID_CUDA],[1],[Use CUDA offload]);;
     sycl)
       echo SYCL acceleration
       AC_DEFINE([GRID_SYCL],[1],[Use SYCL offload]);;
     hip)
       echo HIP acceleration
+      LIBS="${LIBS} -lhipblas -lrocblas -lhipfft"
       AC_DEFINE([GRID_HIP],[1],[Use HIP offload]);;
     none)
       echo NO acceleration    ;;
@@ -408,10 +409,16 @@ AC_SEARCH_LIBS([unw_backtrace], [unwind],
                [have_unwind=true],
 	             [AC_MSG_WARN(libunwind library was not found in your system.)])
 
-AC_SEARCH_LIBS([_Ux86_64_step], [unwind-x86_64],
+AS_CASE([$host_cpu], [x86_64],
-               [AC_DEFINE([HAVE_UNWIND_X86_64], [1], [Define to 1 if you have the `libunwind-x86_64' library])]
+  [AC_SEARCH_LIBS([_Ux86_64_step], [unwind-x86_64],
-               [have_unwind_x86_64=true],
+                [AC_DEFINE([HAVE_UNWIND_X86_64], [1], [Define to 1 if you have the `libunwind-x86_64' library])]
-	             [AC_MSG_WARN(libunwind library was not found in your system.)])
+                [have_unwind_x86_64=true],
+                [AC_MSG_WARN(libunwind library was not found in your system.)])],
+  [aarch64],
+  [AC_SEARCH_LIBS([_Uaarch64_step], [unwind-aarch64],
+                [AC_DEFINE([HAVE_UNWIND_AARCH64], [1], [Define to 1 if you have the `libunwind-aarch64' library])]
+                [have_unwind_aarch64=true],
+                [AC_MSG_WARN(libunwind library was not found in your system.)])])
 
 AC_SEARCH_LIBS([SHA256_Init], [crypto],
                [AC_DEFINE([HAVE_CRYPTO], [1], [Define to 1 if you have the `OpenSSL' library])]