[core] DEMData: do decode on GPU (port mapbox/mapbox-gl-js#8694)

This is first part of work on porting mapbox/mapbox-gl-js#8694 - in follow up patch(es) it is required to remove CPU side copy using 2d canvas support on all supported platforms, similar to approach taken in gl.js  https://github.com/mapbox/mapbox-gl-js/pull/8694/files#diff-34dbe5f7de34dc4b9a8745dcde9bdc37R48

Decoding on CPU removed.
Padding is still done in DEMData() but, instead od doing it wwhile decoding, it is using memcpy to pad original values.

Rebase to latest mapbox-gl-js master and re-generate shaders.

Partly fixes: #15503

1 files changed, 12 insertions, 16 deletions

diff --git a/src/mbgl/geometry/dem_data.hpp b/src/mbgl/geometry/dem_data.hpp
index 21146bc468..537575873f 100644
--- a/src/mbgl/geometry/dem_data.hpp
+++ b/src/mbgl/geometry/dem_data.hpp
@@ -16,13 +16,8 @@ public:
     DEMData(const PremultipliedImage& image, Tileset::DEMEncoding encoding);
     void backfillBorder(const DEMData& borderTileData, int8_t dx, int8_t dy);
 
-    void set(const int32_t x, const int32_t y, const int32_t value) {
-        reinterpret_cast<int32_t*>(image.data.get())[idx(x, y)] = value + 65536;
-    }
-
-    int32_t get(const int32_t x, const int32_t y) const {
-        return reinterpret_cast<const int32_t*>(image.data.get())[idx(x, y)] - 65536;
-    }
+    int32_t get(const int32_t x, const int32_t y) const;
+    const std::array<float, 4>& getUnpackVector() const;
 
     const PremultipliedImage* getImage() const {
         return &image;
@@ -32,16 +27,17 @@ public:
     const int32_t stride;
 
 
-    private:
-        PremultipliedImage image;
+private:
+    Tileset::DEMEncoding encoding;
+    PremultipliedImage image;
 
-        size_t idx(const int32_t x, const int32_t y) const {
-            assert(x >= -1);
-            assert(x < dim + 1);
-            assert(y >= -1);
-            assert(y < dim + 1);
-            return (y + 1) * stride + (x + 1);
-        }
+    size_t idx(const int32_t x, const int32_t y) const {
+        assert(x >= -1);
+        assert(x < dim + 1);
+        assert(y >= -1);
+        assert(y < dim + 1);
+        return (y + 1) * stride + (x + 1);
+    }
 
 };