Floating point support

rawspec.c

@@ -615,6 +615,7 @@ int main(int argc, char *argv[])
         fprintf(stderr, "OBSNCHAN = %d\n",  raw_hdr.obsnchan);
         fprintf(stderr, "NANTS    = %d\n",  raw_hdr.nants);
         fprintf(stderr, "NBITS    = %d\n",  raw_hdr.nbits);
+        fprintf(stderr, "DATATYPE = %s\n",  raw_hdr.float_data ? "FLOAT" : "INTEGER");
         fprintf(stderr, "NPOL     = %d\n",  raw_hdr.npol);
         fprintf(stderr, "OBSFREQ  = %g\n",  raw_hdr.obsfreq);
         fprintf(stderr, "OBSBW    = %g\n",  raw_hdr.obsbw);
@@ -749,6 +750,7 @@ int main(int argc, char *argv[])
           ctx.Ntpb = Ntpb;
           ctx.Nbps = Nbps;
           ctx.input_conjugated = input_conjugated;
+          ctx.float_data = raw_hdr.float_data;
 
           // Initialize for new dimensions and/or conjugation
           ctx.Nb = 0;           // auto-calculate

rawspec.h

@@ -114,6 +114,9 @@ struct rawspec_context_s {
   // the next call to rawspec_initialize().
   int input_conjugated;
 
+  // Flag indicating that the input data is floating point data (not integer).
+  int float_data;
+
   // Flag indicating the concurrent output of the output data's incoherent-sum.
   int incoherently_sum;
   int Naws;

rawspec_gpu.cu

@@ -505,11 +505,16 @@ int rawspec_initialize(rawspec_context * ctx)
   // NbpsIsExpanded = 1 (true) if Nbps==4 else 0 (false).
   if(ctx->Nbps == 0)
     ctx->Nbps = 8;
-  if(ctx->Nbps != 4 && ctx->Nbps != 8 && ctx->Nbps != 16) {
+  if(!ctx->float_data && ctx->Nbps != 4 && ctx->Nbps != 8 && ctx->Nbps != 16) {
     fprintf(stderr, "Number of bits per sample in raw header must be 4, 0/8, or 16\n");
     fprintf(stderr, "Observed a value of %d\n", ctx->Nbps);
     fflush(stderr);
     return 1;
+  } else if(ctx->float_data && ctx->Nbps != 16 && ctx->Nbps != 32) {
+    fprintf(stderr, "Number of bits per float sample in raw header must be 16 or 32\n");
+    fprintf(stderr, "Observed a value of %d\n", ctx->Nbps);
+    fflush(stderr);
+    return 1;
   }
   NbpsIsExpanded = ctx->Nbps == 4;
   if(ctx->Nbps == 4)
@@ -831,13 +836,15 @@ int rawspec_initialize(rawspec_context * ctx)
   }
   fflush(stderr);
 
+  fprintf(stdout, "Detected %s data.\n", ctx->float_data ? "Float" : "Integer");
+
   // Create texture object for device input buffer
   // res_desc describes input resource
   // Width is 32K elements, height is buf_size/32K elements, pitch is 32K elements
   memset(&res_desc, 0, sizeof(res_desc));
   res_desc.resType = cudaResourceTypePitch2D;
   res_desc.res.pitch2D.devPtr = gpu_ctx->d_fft_in;
-  res_desc.res.pitch2D.desc.f = cudaChannelFormatKindSigned;
+  res_desc.res.pitch2D.desc.f = ctx->float_data ? cudaChannelFormatKindFloat : cudaChannelFormatKindSigned;
   res_desc.res.pitch2D.desc.x = ctx->Nbps; // bits per sample
   res_desc.res.pitch2D.width = 1<<LOAD_TEXTURE_WIDTH_POWER;         // elements
   res_desc.res.pitch2D.height = buf_size>>LOAD_TEXTURE_WIDTH_POWER; // elements
@@ -852,7 +859,7 @@ int rawspec_initialize(rawspec_context * ctx)
   // Not sure whether filter_mode matters for cudaReadModeNormalizedFloat
   tex_desc.filter_mode = cudaFilterModePoint;
 #endif // 0
-  tex_desc.readMode = cudaReadModeNormalizedFloat;
+  tex_desc.readMode = ctx->float_data ? cudaReadModeElementType : cudaReadModeNormalizedFloat;
 
   cuda_rc = cudaCreateTextureObject(&gpu_ctx->tex_obj,
                                     &res_desc, &tex_desc, NULL);

rawspec_rawutils.c

@@ -171,6 +171,9 @@ void rawspec_raw_parse_header(const char * buf, rawspec_raw_hdr_t * raw_hdr)
   raw_hdr->nbeam    = rawspec_raw_get_s32(buf, "NBEAM",   -1);
   raw_hdr->nants    = rawspec_raw_get_u32(buf, "NANTS",    1);
 
+  rawspec_raw_get_str(buf, "DATATYPE", "INTEGER", tmp, 80);
+  raw_hdr->float_data = strncmp(tmp, "FLOAT", 5) == 0;
+
   rawspec_raw_get_str(buf, "RA_STR", "0.0", tmp, 80);
   raw_hdr->ra = rawspec_raw_hmsstr_to_h(tmp);
 

rawspec_rawutils.h

@@ -24,6 +24,7 @@ typedef struct {
   char telescop[81];
   off_t hdr_pos; // Offset of start of header
   size_t hdr_size; // Size of header in bytes (not including DIRECTIO padding)
+  int float_data; // flag float (not integer) data
 } rawspec_raw_hdr_t;
 
 // Multiple of 80 and 512