support for nativeRaster (to avoid epensive integer or numeric conversion)

[mdsumner]

• internalize in C for gdal_raster_nativeRaster() function
• explore relation to native buffers from image files for gdal
• compare performance
• add attributes to nativeRaster ??

very very basics of nativeRaster

with this C code (R_RGB and R_RGBA direct from R source)

#include <Rinternals.h>


#define R_RGB(r,g,b)	  ((r)|((g)<<8)|((b)<<16)|0xFF000000)
#define R_RGBA(r,g,b,a)	((r)|((g)<<8)|((b)<<16)|((a)<<24))

SEXP C_native(SEXP b0, SEXP b1, SEXP b2) {
  SEXP res_ = PROTECT(allocVector(INTSXP, length(b0)));
  for (int i = 0; i < length(b0); i++) {
      INTEGER(res_)[i] = (int)R_RGB(RAW(b0)[i], RAW(b1)[i], RAW(b2)[i]);
  }

  UNPROTECT(1);
  return res_;
}

read bytes from a GDAL source, convert to nativeRaster and plot

bytes <- gdal_raster_data(sds::wms_arcgis_mapserver_ESRI.WorldImagery_tms(), target_dim = c(360, 180), band_output_type = "Byte", bands = 1:3)

cl <- .Call("C_native", bytes[[1]], bytes[[2]], bytes[[3]])

plot(0:1, 0:1); rasterImage(structure(matrix(cl, 180), class = "nativeRaster", channels = 3L), 0, 0, 1, 1, interpolate = F)

[mdsumner]

bundling it all in so there's one function call to return the nativeRaster:

#include <Rinternals.h>


#define R_RGB(r,g,b)	  ((r)|((g)<<8)|((b)<<16)|0xFF000000)
#define R_RGBA(r,g,b,a)	((r)|((g)<<8)|((b)<<16)|((a)<<24))

SEXP C_native(SEXP b0, SEXP b1, SEXP b2, SEXP dm) {
  SEXP res_ = PROTECT(allocVector(INTSXP, length(b0)));
  for (int i = 0; i < length(b0); i++) {
      INTEGER(res_)[i] = (int)R_RGB(RAW(b0)[i], RAW(b1)[i], RAW(b2)[i]);
  }
  SEXP dim;
  dim = allocVector(INTSXP, 2);
	    INTEGER(dim)[0] = INTEGER(dm)[1];
	    INTEGER(dim)[1] = INTEGER(dm)[0];
	    setAttrib(res_, R_DimSymbol, dim);
	    setAttrib(res_, R_ClassSymbol, mkString("nativeRaster"));
	    {
		SEXP chsym = install("channels");
		setAttrib(res_, chsym, ScalarInteger(3));
	    }
  UNPROTECT(1);
  return res_;
}

and ximage already supports the type

bytes <- gdal_raster_data(sds::wms_arcgis_mapserver_ESRI.WorldImagery_tms(), target_dim = c(1024, 0), band_output_type = "Byte", bands = 1:3, target_crs = "+proj=laea")
pryr::object_size(bytes)
cl <- .Call("C_native", bytes[[1]], bytes[[2]], bytes[[3]], as.integer(attr(bytes, "dimension")))
pryr::object_size(cl)
ximage::ximage(cl, extent = attr(bytes, "extent"), asp = 1)

[mdsumner]

and here's how to draw onto it, convert from sf to silicate and use the extent to determine the scaling range on the native raster

nr_draw <- function(r, x, ex, ...) {
  v <- as.matrix(silicate::sc_vertex(x))
  from1 <- vaster::col_from_x(dim(r)[2:1], ex, range(v[,1]))
  from2 <- nrow(r) - vaster::row_from_y(dim(r)[2:1], ex, range(v[,2]))

  v[,1] <- scales::rescale(v[,1], from1)
  v[,2] <- scales::rescale(v[,2], from2)
  top <- do.call(rbind, x$object$topology_)
  .v0 <- top$.vx0
  .v1 <- top$.vx1
  nara::nr_line(r, v[.v0,1], v[.v0, 2], v[.v1, 1], v[.v1, 2], ...)
}

crs <- sprintf("+proj=laea +lon_0=%i +lat_0=%i", 180, -30)

cg <- reproj::reproj(silicate::SC0(silicate::inlandwaters), 
                      crs)
library(vapour)
library(nara)
#library(dsn)
library(ximage)
par(mar = rep(0, 4))
bytes <- gdal_raster_data(sds::wms_arcgis_mapserver_ESRI.WorldImagery_tms(), 
                          target_ext = c(-1, 1, -1, 1) * 6378137,
                          target_dim = c(0, dev.size("px")[2]), band_output_type = "Byte", bands = 1:3, 
                          target_crs = crs, resample = "near")

cl <- .Call("C_native", bytes[[1]], bytes[[2]], bytes[[3]], as.integer(attr(bytes, "dimension")))
ximage(nr_draw(cl, cg, attr(bytes, "extent"), col = "green"), asp = 1)

very fast

[mdsumner]

benchmark on Windows

library(vapour)
library(ximage)
dsn <- sds::wms_arcgis_mapserver_ESRI.WorldImagery_tms()
system.time({
 im <- gdal_raster_image(dsn, target_dim = c(360, 180), target_crs = "OGC:CRS84", target_ext = c(-180, 180, -90, 90))
 ximage(im, asp = 1)

})

2.7 seconds

[mdsumner]

fleshed out here

https://github.com/mdsumner/gdalnara

I think it will just get embedded and replace _image