Optimize Raster Grid Generation (Speed & Memory)

Author: e-kotov

R/backend_sequential.R

@@ -199,12 +199,22 @@ inspire_grid_from_extent_internal <- function(
 
     # 3. Generate IDs and Assign as Levels (Factor Raster)
     if (id_format != "none") {
-      # xyFromCell returns centroids
-      coords <- terra::xyFromCell(r, seq_len(terra::ncell(r)))
+      # xyFromCell returns centroids. For large rasters, this is memory-intensive.
+      # We replicate the logic using efficient vector generation.
+      # Rows (Y) are outer loop (slowest changing), Columns (X) are inner loop (fastest).
+      
+      nrows <- terra::nrow(r)
+      ncols <- terra::ncol(r)
 
-      # Calculate LLC for make_ids
-      x_llc <- coords[, 1] - (cellsize_m / 2)
-      y_llc <- coords[, 2] - (cellsize_m / 2)
+      # Calculate LLC for make_ids directly
+      # X LLC: seq from xmin
+      x_llc_seq <- seq.int(from = xmin, by = cellsize_m, length.out = ncols)
+      # Y LLC: seq from ymax - cellsize (top-left Y LLC) down to ymin
+      y_llc_seq <- seq.int(from = ymax - cellsize_m, by = -cellsize_m, length.out = nrows)
+      
+      # Create full vectors
+      x_llc <- rep(x_llc_seq, times = nrows)
+      y_llc <- rep(y_llc_seq, each = ncols)
 
       # Generate ID strings (using existing helper)
       ids_list <- make_ids(

R/stream_grid_raster.R

@@ -127,17 +127,7 @@ stream_grid_raster_terra <- function(
     )
 
     cells <- first_cell:last_cell
-    coords <- terra::xyFromCell(r_template, cells)
-
-    x_vals <- coords[, 1]
-    y_vals <- coords[, 2]
-
-    # Compute grid cell IDs (linear addressing, row-major order)
-    # Column: floor((x - xmin) / cellsize) + 1
-    # Row: floor((ymax - y) / cellsize) + 1 (Y axis is inverted)
-    col_idx <- floor((x_vals - xmin) / cellsize_m) + 1
-    row_idx <- floor((ymax - y_vals) / cellsize_m) + 1
-    cell_ids <- (row_idx - 1) * ncols + col_idx
+    cell_ids <- cells
 
     # Write chunk values
     terra::writeValues(

R/stream_grid_raster_parallel.R

@@ -41,14 +41,15 @@ NULL
 #' @keywords internal
 #' @noRd
 .compute_raster_chunk <- function(start_row, nrows, ncols) {
-  # Generate row/col indices for the chunk
-  # Generate continuous sequence of cell IDs directly (row-major order)
-  # Optimization: Avoid allocating large row/col vectors
-  start_id <- (start_row - 1) * ncols + 1
-  end_id <- start_id + (nrows * ncols) - 1
-  cell_ids <- start_id:end_id
-
-  list(values = as.integer(cell_ids), start = start_row, nrows = nrows)
+  # Compute first cell ID for this linear chunk
+  # Cell IDs are 1-based, row-major
+  first_cell <- (start_row - 1) * ncols + 1
+
+  # Create sequence directly
+  # Since we are processing full rows, the cell IDs are contiguous
+  values <- seq.int(from = first_cell, by = 1L, length.out = nrows * ncols)
+
+  list(values = values, start = start_row, nrows = nrows)
 }