Improve JPEG encoding loop (#71)

Cargo.toml

@@ -35,6 +35,7 @@ wide = "0.7"
 log = "0.4"
 simple_logger ="4.0"
 rayon = "1.10"
+unroll="*"
 
 [target.'cfg(windows)'.dependencies]
 cpu-time = "1.0"

src/enabled_features.rs

@@ -18,6 +18,9 @@ pub struct EnabledFeatures {
 
     /// Sadly C++ version has a bug where it uses 16 bit math in the SIMD path and 32 bit math in the scalar path
     pub use_16bit_adv_predict: bool,
+
+    /// Accept JPEG files that have invalid DHT tables
+    pub accept_invalid_dht: bool,
 }
 
 impl EnabledFeatures {
@@ -31,6 +34,7 @@ impl EnabledFeatures {
             max_jpeg_width: 16386,
             use_16bit_dc_estimate: true,
             use_16bit_adv_predict: true,
+            accept_invalid_dht: false,
         }
     }
 
@@ -45,6 +49,7 @@ impl EnabledFeatures {
             max_jpeg_width: i32::MAX,
             use_16bit_dc_estimate: false,
             use_16bit_adv_predict: false,
+            accept_invalid_dht: true,
         }
     }
 
@@ -59,6 +64,7 @@ impl EnabledFeatures {
             max_jpeg_width: i32::MAX,
             use_16bit_dc_estimate: true,
             use_16bit_adv_predict: true,
+            accept_invalid_dht: true,
         }
     }
 }

src/structs/block_based_image.rs

@@ -10,6 +10,8 @@ use crate::consts::ZIGZAG_TO_RASTER;
 
 use super::{block_context::BlockContext, jpeg_header::JPegHeader};
 
+use unroll::unroll_for_loops;
+
 /// holds the 8x8 blocks for a given component. Since we do multithreaded encoding,
 /// the image may only hold a subset of the components (specified by dpos_offset),
 /// but they can be merged
@@ -203,6 +205,17 @@ impl AlignedBlock {
     }
 
     /// gets underlying array of 64 coefficients (guaranteed to be 32-byte aligned)
+    #[unroll_for_loops]
+    pub fn zigzag(&self) -> AlignedBlock {
+        let mut block = AlignedBlock::default();
+        for i in 0..64 {
+            block.raw_data[i] = self.raw_data[usize::from(ZIGZAG_TO_RASTER[i])];
+        }
+        return block;
+    }
+
+    // used for debugging
+    #[allow(dead_code)]
     pub fn get_block(&self) -> &[i16; 64] {
         return &self.raw_data;
     }