api!: add image_span versions of ImageInput methods (#4748)

ImageInput methods that read scanlines, tiles, and image are given new
flavors that take image_span, much like the changes of PR #4727 did for
ImageOutput.

Generally, for each, there are three versions: (1) a low-level virtual
method base case that takes an `image_span<byte>` that describes the
underlying memory, and an explicit TypeDesc that specifies the data type
conversion (including allowing TypeUnknown to indicate keeping channels
in their native per-channel format from the file); (2) a templated
version taking an `image_span<T>` that infers the data type conversion
from `T` (must be a single type for all channels); (3) a templated
version taking a `span<T>` that further implies a contiguous buffer in
all dimensions.

For now, the default implementations of these new ImageInput methods are
just wrappers that call the old pointer-based ones. One by one, over
time, we can swap them, changing the format implementations to have a
full implementation of the new bounded versions, and make their raw
pointer versions call the wrappers. The raw pointer ones will be
understood to be "unsafe", merely assuming that the pointers always
refer to appropriately-sized memory areas. Meanwhile, the ones using
spans and image_spans will, due to assertions in their implementations,
make it easier to verify (at least in debug mode), that we never touch
memory outside these bounds.

---------

Signed-off-by: Larry Gritz <[email protected]>

Author: lgritz

CMakeLists.txt

@@ -4,7 +4,7 @@
 
 cmake_minimum_required (VERSION 3.18.2...4.0)
 
-set (OpenImageIO_VERSION "3.1.2.0")
+set (OpenImageIO_VERSION "3.1.3.0")
 set (OpenImageIO_VERSION_OVERRIDE "" CACHE STRING
      "Version override (use with caution)!")
 mark_as_advanced (OpenImageIO_VERSION_OVERRIDE)

src/doc/imageoutput.rst

@@ -246,26 +246,20 @@ individual plugin.
 
 .. tabs::
 
-   .. code-tab:: c++
-
-      unsigned char tile[tilesize*tilesize*channels];
-      int z = 0;   // Always zero for 2D images
-      for (int y = 0;  y < yres;  y += tilesize) {
-          for (int x = 0;  x < xres;  x += tilesize) {
-              ... generate data in tile[] ..
-              out->write_tile (x, y, z, TypeDesc::UINT8, tile);
-          }
-      }
-      out->close ();
+   .. tab:: C++
+      .. literalinclude:: ../../testsuite/docs-examples-cpp/src/docs-examples-imageoutput.cpp
+          :language: c++
+          :start-after: BEGIN-imageoutput-tiles
+          :end-before: END-imageoutput-tiles
+          :dedent: 4
 
-   .. code-tab:: py
+   .. tab:: Python
 
-      z = 0  # Always zero for 2D images
-      for y in range(0, yres, tilesize) :
-          for x in range(0, xres, tilesize) :
-              # ... generate data in tile[][][] ..
-              out.write_tile (x, y, z, tile)
-      out.close ()
+      .. literalinclude:: ../../testsuite/docs-examples-python/src/docs-examples-imageoutput.py
+          :language: py
+          :start-after: BEGIN-imageoutput-tiles
+          :end-before: END-imageoutput-tiles
+          :dedent: 8
 
 The first three arguments to ``write_tile()`` specify which tile is being
 written by the pixel coordinates of any pixel contained in the tile: *x*

src/include/OpenImageIO/imageio.h

@@ -286,6 +286,37 @@ ImageInput::read_scanline(int y, int z, TypeDesc format, void* data,
 
 
 
+bool
+ImageInput::read_scanlines(int subimage, int miplevel, int ybegin, int yend,
+                           int chbegin, int chend, TypeDesc format,
+                           const image_span<std::byte>& data)
+{
+    ImageSpec spec = spec_dimensions(subimage, miplevel);
+    if (chend < 0 || chend > spec.nchannels)
+        chend = spec.nchannels;
+    if (chbegin < 0 || chbegin >= chend) {
+        errorfmt("read_scanlines: invalid channel range [{},{})", chbegin,
+                 chend);
+        return false;
+    }
+    size_t isize = (format == TypeUnknown
+                        ? spec.pixel_bytes(chbegin, chend, true /*native*/)
+                        : format.size() * (chend - chbegin))
+                   * size_t(spec.width);
+    if (isize != data.size_bytes()) {
+        errorfmt(
+            "read_scanlines: Buffer size is incorrect ({} bytes vs {} needed)",
+            isize, data.size_bytes());
+        return false;
+    }
+
+    // Default implementation (for now): call the old pointer+stride
+    return read_scanlines(subimage, miplevel, ybegin, yend, 0, chbegin, chend,
+                          format, data.data(), data.xstride());
+}
+
+
+
 bool
 ImageInput::read_scanlines(int subimage, int miplevel, int ybegin, int yend,
                            int z, int chbegin, int chend, TypeDesc format,
@@ -611,6 +642,38 @@ ImageInput::read_tile(int x, int y, int z, TypeDesc format, void* data,
 
 
 
+bool
+ImageInput::read_tiles(int subimage, int miplevel, int xbegin, int xend,
+                       int ybegin, int yend, int zbegin, int zend, int chbegin,
+                       int chend, TypeDesc format,
+                       const image_span<std::byte>& data)
+{
+    ImageSpec spec = spec_dimensions(subimage, miplevel);
+    if (chend < 0 || chend > spec.nchannels)
+        chend = spec.nchannels;
+    if (chbegin < 0 || chbegin >= chend) {
+        errorfmt("read_tiles: invalid channel range [{},{})", chbegin, chend);
+        return false;
+    }
+    size_t isize = (format == TypeUnknown
+                        ? spec.pixel_bytes(chbegin, chend, true /*native*/)
+                        : format.size() * (chend - chbegin))
+                   * size_t(xend - xbegin) * size_t(yend - ybegin)
+                   * size_t(zend - zbegin);
+    if (isize != data.size_bytes()) {
+        errorfmt("read_tiles: Buffer size is incorrect ({} bytes vs {} needed)",
+                 isize, data.size_bytes());
+        return false;
+    }
+
+    // Default implementation (for now): call the old pointer+stride
+    return read_tiles(subimage, miplevel, ybegin, yend, xbegin, xend, zbegin,
+                      zend, chbegin, chend, format, data.data(),
+                      data.xstride());
+}
+
+
+
 bool
 ImageInput::read_tiles(int subimage, int miplevel, int xbegin, int xend,
                        int ybegin, int yend, int zbegin, int zend, int chbegin,
@@ -1096,6 +1159,110 @@ ImageInput::read_image(int subimage, int miplevel, int chbegin, int chend,
 
 
 
+bool
+ImageInput::read_image(int subimage, int miplevel, int chbegin, int chend,
+                       TypeDesc format, const image_span<std::byte>& data)
+{
+#if 0
+    ImageSpec spec = spec_dimensions(subimage, miplevel);
+    if (chend < 0 || chend > spec.nchannels)
+        chend = spec.nchannels;
+    size_t isize = (format == TypeUnknown
+                        ? spec.pixel_bytes(chbegin, chend, true /*native*/)
+                        : format.size() * (chend - chbegin))
+                   * spec.image_pixels();
+    if (isize != data.size_bytes()) {
+        errorfmt("read_image: Buffer size is incorrect ({} bytes vs {} needed)",
+                 sz, data.size_bytes());
+        return false;
+    }
+
+    // Default implementation (for now): call the old pointer+stride
+    return read_image(subimage, miplevel, chbegin, chend, format, data.data(),
+                      data.xstride(), data.ystride(), data.zstride());
+#else
+    pvt::LoggedTimer logtime("II::read_image");
+    ImageSpec spec;
+    int rps = 0;
+    {
+        // We need to lock briefly to retrieve rps from the spec
+        lock_guard lock(*this);
+        if (!seek_subimage(subimage, miplevel))
+            return false;
+        // Copying the dimensions of the designated subimage/miplevel to a
+        // local `spec` means that we can release the lock!  (Calls to
+        // read_native_* will internally lock again if necessary.)
+        spec.copy_dimensions(m_spec);
+        // For scanline files, we also need one piece of metadata
+        if (!spec.tile_width)
+            rps = m_spec.get_int_attribute("tiff:RowsPerStrip", 64);
+    }
+    if (spec.image_bytes() < 1) {
+        errorfmt("Invalid image size {} x {} ({} chans)", m_spec.width,
+                 m_spec.height, m_spec.nchannels);
+        return false;
+    }
+
+    if (chend < 0 || chend > spec.nchannels)
+        chend = spec.nchannels;
+    if (chbegin < 0 || chbegin >= chend) {
+        errorfmt("read_image: invalid channel range [{},{})", chbegin, chend);
+        return false;
+    }
+    int nchans         = chend - chbegin;
+    bool native        = (format == TypeUnknown);
+    size_t pixel_bytes = native ? spec.pixel_bytes(chbegin, chend, native)
+                                : (format.size() * nchans);
+    size_t isize       = pixel_bytes * spec.image_pixels();
+    if (isize != data.size_bytes()) {
+        errorfmt("read_image: Buffer size is incorrect ({} bytes vs {} needed)",
+                 isize, data.size_bytes());
+        return false;
+    }
+
+    bool ok = true;
+    if (spec.tile_width) {  // Tiled image -- rely on read_tiles
+        // Read in chunks of a whole row of tiles at once. If tiles are
+        // 64x64, a 2k image has 32 tiles across. That's fine for now (for
+        // parallelization purposes), but as typical core counts increase,
+        // we may someday want to revisit this to batch multiple rows.
+        for (int z = 0; z < spec.depth; z += spec.tile_depth) {
+            int zend = std::min(z + spec.z + spec.tile_depth,
+                                spec.z + spec.depth);
+            for (int y = 0; y < spec.height && ok; y += spec.tile_height) {
+                int yend = std::min(y + spec.y + spec.tile_height,
+                                    spec.y + spec.height);
+                ok &= read_tiles(subimage, miplevel, spec.x,
+                                 spec.x + spec.width, y + spec.y, yend,
+                                 z + spec.z, zend, chbegin, chend, format,
+                                 data.subspan(spec.x, spec.x + spec.width,
+                                              y + spec.y, yend, z + spec.z,
+                                              zend));
+            }
+        }
+    } else {  // Scanline image -- rely on read_scanlines.
+        // Split into reasonable chunks -- try to use around 64 MB or the
+        // oiio_read_chunk value, which ever is bigger, but also round up to
+        // a multiple of the TIFF rows per strip (or 64).
+        int chunk = std::max(1, (1 << 26) / int(spec.scanline_bytes(true)));
+        chunk     = std::max(chunk, int(oiio_read_chunk));
+        chunk     = round_to_multiple(chunk, rps);
+        for (int z = 0; z < spec.depth; ++z) {
+            for (int y = 0; y < spec.height && ok; y += chunk) {
+                int yend = std::min(y + spec.y + chunk, spec.y + spec.height);
+                ok &= read_scanlines(subimage, miplevel, y + spec.y, yend,
+                                     chbegin, chend, format,
+                                     data.subspan(spec.x, spec.x + spec.width,
+                                                  y + spec.y, yend));
+            }
+        }
+    }
+    return ok;
+#endif
+}
+
+
+
 bool
 ImageInput::read_native_deep_scanlines(int /*subimage*/, int /*miplevel*/,
                                        int /*ybegin*/, int /*yend*/, int /*z*/,

src/libOpenImageIO/imageinput.cpp

@@ -144,3 +144,5 @@ Comparing "simple.tif" and "ref/simple.tif"
 PASS
 Comparing "scanlines.tif" and "ref/scanlines.tif"
 PASS
+Comparing "tiles.tif" and "ref/tiles.tif"
+PASS

testsuite/docs-examples-cpp/ref/out-arm.txt

@@ -144,3 +144,5 @@ Comparing "simple.tif" and "ref/simple.tif"
 PASS
 Comparing "scanlines.tif" and "ref/scanlines.tif"
 PASS
+Comparing "tiles.tif" and "ref/tiles.tif"
+PASS

testsuite/docs-examples-cpp/ref/out.txt

@@ -100,7 +100,7 @@
 # and need the images checked into the ref directory.
 outputs = [
     # Outputs from the ImageOutput chapter:
-    "simple.tif", "scanlines.tif",
+    "simple.tif", "scanlines.tif", "tiles.tif",
     # Outputs from the ImageInput chapter:
 
     # Outputs from the ImageBuf chapter:

testsuite/docs-examples-cpp/ref/tiles.tif

@@ -49,13 +49,14 @@ simple_read()
     int xres              = spec.width;
     int yres              = spec.height;
     int nchannels         = spec.nchannels;
-    auto pixels           = std::unique_ptr<unsigned char[]>(
-        new unsigned char[xres * yres * nchannels]);
-    inp->read_image(0, 0, 0, nchannels, TypeDesc::UINT8, &pixels[0]);
+    std::vector<uint8_t> pixels(xres * yres * nchannels);
+    inp->read_image(0 /*subimage*/, 0 /*miplevel*/, 0 /*chbegin*/,
+                    nchannels /*chend*/, make_span(pixels));
     inp->close();
 }
 // END-imageinput-simple
 
+
 void
 scanlines_read()
 {
@@ -65,10 +66,11 @@ scanlines_read()
     auto inp              = ImageInput::open(filename);
     const ImageSpec& spec = inp->spec();
     if (spec.tile_width == 0) {
-        auto scanline = std::unique_ptr<unsigned char[]>(
-            new unsigned char[spec.width * spec.nchannels]);
-        for (int y = 0; y < spec.height; ++y) {
-            inp->read_scanline(y, 0, TypeDesc::UINT8, &scanline[0]);
+        std::vector<uint8_t> scanline(spec.width * spec.nchannels);
+        for (int y = spec.y; y < spec.y + spec.height; ++y) {
+            inp->read_scanlines(0 /*subimage*/, 0 /*miplevel*/, y, y + 1,
+                                0 /*chbegin*/, spec.nchannels /*chend*/,
+                                make_span(scanline));
             // ... process data in scanline[0..width*channels-1] ...
         }
     } else {
@@ -93,19 +95,31 @@ tiles_read()
     } else {
         // Tiles
         int tilesize = spec.tile_width * spec.tile_height;
-        auto tile    = std::unique_ptr<unsigned char[]>(
-            new unsigned char[tilesize * spec.nchannels]);
-        for (int y = 0; y < spec.height; y += spec.tile_height) {
-            for (int x = 0; x < spec.width; x += spec.tile_width) {
-                inp->read_tile(x, y, 0, TypeDesc::UINT8, &tile[0]);
+        std::vector<uint8_t> tile(tilesize * spec.nchannels);
+        for (int y = spec.y; y < spec.y + spec.height; y += spec.tile_height) {
+            for (int x = spec.x; x < spec.x + spec.width;
+                 x += spec.tile_width) {
+                inp->read_tiles(0 /*subimage*/, 0 /*miplevel*/, x,
+                                std::min(x + spec.tile_width, spec.width), y,
+                                std::min(y + spec.tile_height, spec.height), 0,
+                                1, 0 /*chbegin*/, spec.nchannels /*chend*/,
+                                make_span(tile));
                 // ... process the pixels in tile[] ..
+                // Watch out for "edge tiles" that are smaller than the full
+                // tile size.
+                // For example, if the image is 100x100 and the tile size is
+                // 32x32, the last tile in each row will be 4x32, the bottom
+                // row of tiles will be 32x4, and the very last
+                // tile of the whole images will be 4x4.
             }
         }
     }
     inp->close();
     // END-imageinput-tiles
 }
 
+
+
 void
 unassociatedalpha()
 {

testsuite/docs-examples-cpp/run.py

@@ -65,9 +65,9 @@ scanlines_write()
     std::unique_ptr<ImageOutput> out = ImageOutput::create(filename);
     if (!out)
         return;  // error
-    ImageSpec spec(xres, yres, channels, TypeDesc::UINT8);
 
     // BEGIN-imageoutput-scanlines
+    ImageSpec spec(xres, yres, channels, TypeDesc::UINT8);
     std::vector<unsigned char> scanline(xres * channels);
     out->open(filename, spec);
     for (int y = 0; y < yres; ++y) {
@@ -80,10 +80,56 @@ scanlines_write()
 
 
 
+void
+tiles_write()
+{
+    const char* filename = "tiles.tif";
+    const int xres = 320, yres = 240, channels = 3;
+    const int tilesize = 64;
+
+    // BEGIN-imageoutput-tiles-create
+    std::unique_ptr<ImageOutput> out = ImageOutput::create(filename);
+    if (!out)
+        return;  // error: could not create output at all
+    if (!out->supports("tiles")) {
+        // Tiles are not supported
+    }
+    // END-imageoutput-tiles-create
+
+    // BEGIN-imageoutput-tiles-make-spec-open
+    ImageSpec spec(xres, yres, channels, TypeDesc::UINT8);
+    spec.tile_width  = tilesize;
+    spec.tile_height = tilesize;
+    out->open(filename, spec);
+    // END-imageoutput-tiles-make-spec-open
+
+    // BEGIN-imageoutput-tiles
+    std::vector<uint8_t> tile(tilesize * tilesize * spec.nchannels);
+    for (int y = 0; y < yres; y += tilesize) {
+        for (int x = 0; x < xres; x += tilesize) {
+            out->write_tiles(x, std::min(x + spec.tile_width, spec.width), y,
+                             std::min(y + spec.tile_height, spec.height), 0, 1,
+                             make_span(tile));
+            // ... process the pixels in tile[] ..
+            // Watch out for "edge tiles" that are smaller than the full
+            // tile size.
+            // For example, if the image is 100x100 and the tile size is
+            // 32x32, the last tile in each row will be 4x32, the bottom
+            // row of tiles will be 32x4, and the very last
+            // tile of the whole images will be 4x4.
+        }
+    }
+    out->close();
+    // END-imageoutput-tiles
+}
+
+
+
 int
 main(int /*argc*/, char** /*argv*/)
 {
     simple_write();
     scanlines_write();
+    tiles_write();
     return 0;
 }