Configurable chroma dc_qi and ac_qi interpolation

Reduce the resolution to Q8; cover half of combinations.
Drop the bound on qi offset at low rates, for completeness.
Add a CLI flag to support stat collection.

src/api/mod.rs

@@ -111,6 +111,7 @@ pub struct EncoderConfig {
   pub speed_settings: SpeedSettings,
   pub show_psnr: bool,
   pub train_rdo: bool,
+  pub uv_m_q8: u8,
 }
 
 /// Default preset for EncoderConfig: it is a balance between quality and speed.
@@ -163,6 +164,7 @@ impl EncoderConfig {
       speed_settings: SpeedSettings::from_preset(speed),
       show_psnr: false,
       train_rdo: false,
+      uv_m_q8: 81,
     }
   }
 

src/bin/common.rs

@@ -303,6 +303,12 @@ pub fn parse_cli() -> Result<CliOptions, CliError> {
       Arg::with_name("train-rdo")
         .long("train-rdo")
     )
+    .arg(
+      Arg::with_name("UV_M_Q8")
+        .long("uv-m-q8")
+        .takes_value(true)
+        .default_value("81")
+    )
     .subcommand(SubCommand::with_name("advanced")
                 .setting(AppSettings::Hidden)
                 .about("Advanced features")
@@ -547,6 +553,7 @@ fn parse_config(matches: &ArgMatches<'_>) -> Result<EncoderConfig, CliError> {
 
   cfg.low_latency = matches.is_present("LOW_LATENCY");
   cfg.train_rdo = train_rdo;
+  cfg.uv_m_q8 = matches.value_of("UV_M_Q8").unwrap().parse().unwrap();
 
   Ok(cfg)
 }

src/rate.rs

@@ -559,21 +559,21 @@ const Q57_SQUARE_EXP_SCALE: f64 =
   (2.0 * ::std::f64::consts::LN_2) / ((1i64 << 57) as f64);
 
 // Daala style log-offset for chroma quantizers
-fn chroma_offset(log_target_q: i64) -> (i64, i64) {
+fn chroma_offset(log_target_q: i64, m_q8: u8) -> (i64, i64) {
   let x = log_target_q.max(0);
-  // Gradient 0.266 optimized for CIEDE2000+PSNR on subset3
-  let y = (x >> 2) + (x >> 6);
+  let m_q12 = (m_q8 as i64) << 4;
+  let y = m_q12 * (x >> 12);
   // blog64(7) - blog64(4); blog64(5) - blog64(4)
   (0x19D_5D9F_D501_0B37 - y, 0xA4_D3C2_5E68_DC58 - y)
 }
 
 impl QuantizerParameters {
   fn new_from_log_q(
-    log_base_q: i64, log_target_q: i64, bit_depth: usize,
+    log_base_q: i64, log_target_q: i64, bit_depth: usize, uv_m_q8: u8
   ) -> QuantizerParameters {
     let scale = q57(QSCALE + bit_depth as i32 - 8);
     let quantizer = bexp64(log_target_q + scale);
-    let (offset_u, offset_v) = chroma_offset(log_target_q);
+    let (offset_u, offset_v) = chroma_offset(log_target_q, uv_m_q8);
     let log_target_q_u = log_target_q + offset_u + scale;
     let log_target_q_v = log_target_q + offset_v + scale;
     let quantizer_u = bexp64(log_target_q_u);
@@ -749,6 +749,7 @@ impl RCState {
     &self, ctx: &ContextInner<T>, output_frameno: u64, fti: usize,
     maybe_prev_log_base_q: Option<i64>,
   ) -> QuantizerParameters {
+    let uv_m_q8 = ctx.config.uv_m_q8;
     // Is rate control active?
     if self.target_bitrate <= 0 {
       // Rate control is not active.
@@ -772,7 +773,7 @@ impl RCState {
       // Adjust the quantizer for the frame type, result is Q57:
       let log_q = ((log_base_q + (1i64 << 11)) >> 12) * (MQP_Q12[fti] as i64)
         + DQP_Q57[fti];
-      QuantizerParameters::new_from_log_q(log_base_q, log_q, bit_depth)
+      QuantizerParameters::new_from_log_q(log_base_q, log_q, bit_depth, uv_m_q8)
     } else {
       let mut nframes: [i32; FRAME_NSUBTYPES + 1] = [0; FRAME_NSUBTYPES + 1];
       let mut log_scale: [i64; FRAME_NSUBTYPES] = self.log_scale;
@@ -790,6 +791,7 @@ impl RCState {
             self.pass1_log_base_q,
             log_q,
             ctx.config.bit_depth,
+            uv_m_q8,
           );
         }
         // Second pass of 2-pass mode: we know exactly how much of each frame
@@ -1052,7 +1054,7 @@ impl RCState {
           // If that target is unreasonable, oh well; we'll have to drop.
         }
       }
-      QuantizerParameters::new_from_log_q(log_base_q, log_q, bit_depth)
+      QuantizerParameters::new_from_log_q(log_base_q, log_q, bit_depth, uv_m_q8)
     }
   }