Merge "dsp: reuse wht transform from dec in encoder"

src/dsp/dsp.h

@@ -67,7 +67,6 @@ typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out);
 typedef void (*VP8WHT)(const int16_t* in, int16_t* out);
 extern VP8Idct VP8ITransform;
 extern VP8Fdct VP8FTransform;
-extern VP8WHT VP8ITransformWHT;
 extern VP8WHT VP8FTransformWHT;
 // Predictions
 // *dst is the destination block. *top and *left can be NULL.

src/dsp/enc.c

@@ -159,33 +159,6 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   }
 }
 
-static void ITransformWHT(const int16_t* in, int16_t* out) {
-  int tmp[16];
-  int i;
-  for (i = 0; i < 4; ++i) {
-    const int a0 = in[0 + i] + in[12 + i];
-    const int a1 = in[4 + i] + in[ 8 + i];
-    const int a2 = in[4 + i] - in[ 8 + i];
-    const int a3 = in[0 + i] - in[12 + i];
-    tmp[0  + i] = a0 + a1;
-    tmp[8  + i] = a0 - a1;
-    tmp[4  + i] = a3 + a2;
-    tmp[12 + i] = a3 - a2;
-  }
-  for (i = 0; i < 4; ++i) {
-    const int dc = tmp[0 + i * 4] + 3;    // w/ rounder
-    const int a0 = dc             + tmp[3 + i * 4];
-    const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
-    const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
-    const int a3 = dc             - tmp[3 + i * 4];
-    out[ 0] = (a0 + a1) >> 3;
-    out[16] = (a3 + a2) >> 3;
-    out[32] = (a0 - a1) >> 3;
-    out[48] = (a3 - a2) >> 3;
-    out += 64;
-  }
-}
-
 static void FTransformWHT(const int16_t* in, int16_t* out) {
   // input is 12b signed
   int32_t tmp[16];
@@ -699,7 +672,6 @@ static void Copy4x4(const uint8_t* src, uint8_t* dst) { Copy(src, dst, 4); }
 VP8CHisto VP8CollectHistogram;
 VP8Idct VP8ITransform;
 VP8Fdct VP8FTransform;
-VP8WHT VP8ITransformWHT;
 VP8WHT VP8FTransformWHT;
 VP8Intra4Preds VP8EncPredLuma4;
 VP8IntraPreds VP8EncPredLuma16;
@@ -718,13 +690,13 @@ extern void VP8EncDspInitSSE2(void);
 extern void VP8EncDspInitNEON(void);
 
 void VP8EncDspInit(void) {
+  VP8DspInit();  // common inverse transforms
   InitTables();
 
   // default C implementations
   VP8CollectHistogram = CollectHistogram;
   VP8ITransform = ITransform;
   VP8FTransform = FTransform;
-  VP8ITransformWHT = ITransformWHT;
   VP8FTransformWHT = FTransformWHT;
   VP8EncPredLuma4 = Intra4Preds;
   VP8EncPredLuma16 = Intra16Preds;

src/dsp/enc_neon.c

@@ -145,74 +145,6 @@ static void ITransform(const uint8_t* ref,
   }
 }
 
-// Same code as dec_neon.c
-static void ITransformWHT(const int16_t* in, int16_t* out) {
-  const int kStep = 32;  // The store is only incrementing the pointer as if we
-                         // had stored a single byte.
-  __asm__ volatile (
-    // part 1
-    // load data into q0, q1
-    "vld1.16         {q0, q1}, [%[in]]           \n"
-
-    "vaddl.s16       q2, d0, d3                  \n" // a0 = in[0] + in[12]
-    "vaddl.s16       q3, d1, d2                  \n" // a1 = in[4] + in[8]
-    "vsubl.s16       q4, d1, d2                  \n" // a2 = in[4] - in[8]
-    "vsubl.s16       q5, d0, d3                  \n" // a3 = in[0] - in[12]
-
-    "vadd.s32        q0, q2, q3                  \n" // tmp[0] = a0 + a1
-    "vsub.s32        q2, q2, q3                  \n" // tmp[8] = a0 - a1
-    "vadd.s32        q1, q5, q4                  \n" // tmp[4] = a3 + a2
-    "vsub.s32        q3, q5, q4                  \n" // tmp[12] = a3 - a2
-
-    // Transpose
-    // q0 = tmp[0, 4, 8, 12], q1 = tmp[2, 6, 10, 14]
-    // q2 = tmp[1, 5, 9, 13], q3 = tmp[3, 7, 11, 15]
-    "vswp            d1, d4                      \n" // vtrn.64 q0, q2
-    "vswp            d3, d6                      \n" // vtrn.64 q1, q3
-    "vtrn.32         q0, q1                      \n"
-    "vtrn.32         q2, q3                      \n"
-
-    "vmov.s32        q4, #3                      \n" // dc = 3
-    "vadd.s32        q0, q0, q4                  \n" // dc = tmp[0] + 3
-    "vadd.s32        q6, q0, q3                  \n" // a0 = dc + tmp[3]
-    "vadd.s32        q7, q1, q2                  \n" // a1 = tmp[1] + tmp[2]
-    "vsub.s32        q8, q1, q2                  \n" // a2 = tmp[1] - tmp[2]
-    "vsub.s32        q9, q0, q3                  \n" // a3 = dc - tmp[3]
-
-    "vadd.s32        q0, q6, q7                  \n"
-    "vshrn.s32       d0, q0, #3                  \n" // (a0 + a1) >> 3
-    "vadd.s32        q1, q9, q8                  \n"
-    "vshrn.s32       d1, q1, #3                  \n" // (a3 + a2) >> 3
-    "vsub.s32        q2, q6, q7                  \n"
-    "vshrn.s32       d2, q2, #3                  \n" // (a0 - a1) >> 3
-    "vsub.s32        q3, q9, q8                  \n"
-    "vshrn.s32       d3, q3, #3                  \n" // (a3 - a2) >> 3
-
-    // set the results to output
-    "vst1.16         d0[0], [%[out]], %[kStep]      \n"
-    "vst1.16         d1[0], [%[out]], %[kStep]      \n"
-    "vst1.16         d2[0], [%[out]], %[kStep]      \n"
-    "vst1.16         d3[0], [%[out]], %[kStep]      \n"
-    "vst1.16         d0[1], [%[out]], %[kStep]      \n"
-    "vst1.16         d1[1], [%[out]], %[kStep]      \n"
-    "vst1.16         d2[1], [%[out]], %[kStep]      \n"
-    "vst1.16         d3[1], [%[out]], %[kStep]      \n"
-    "vst1.16         d0[2], [%[out]], %[kStep]      \n"
-    "vst1.16         d1[2], [%[out]], %[kStep]      \n"
-    "vst1.16         d2[2], [%[out]], %[kStep]      \n"
-    "vst1.16         d3[2], [%[out]], %[kStep]      \n"
-    "vst1.16         d0[3], [%[out]], %[kStep]      \n"
-    "vst1.16         d1[3], [%[out]], %[kStep]      \n"
-    "vst1.16         d2[3], [%[out]], %[kStep]      \n"
-    "vst1.16         d3[3], [%[out]], %[kStep]      \n"
-
-    : [out] "+r"(out)  // modified registers
-    : [in] "r"(in), [kStep] "r"(kStep)  // constants
-    : "memory", "q0", "q1", "q2", "q3", "q4",
-      "q5", "q6", "q7", "q8", "q9" // clobbered
-  );
-}
-
 // Forward transform.
 
 // adapted from vp8/encoder/arm/neon/shortfdct_neon.asm
@@ -622,7 +554,6 @@ void VP8EncDspInitNEON(void) {
   VP8ITransform = ITransform;
   VP8FTransform = FTransform;
 
-  VP8ITransformWHT = ITransformWHT;
   VP8FTransformWHT = FTransformWHT;
 
   VP8TDisto4x4 = Disto4x4;

src/enc/quant.c

@@ -751,7 +751,7 @@ static int ReconstructIntra16(VP8EncIterator* const it,
   }
 
   // Transform back
-  VP8ITransformWHT(dc_tmp, tmp[0]);
+  VP8TransformWHT(dc_tmp, tmp[0]);
   for (n = 0; n < 16; n += 2) {
     VP8ITransform(ref + VP8Scan[n], tmp[n], yuv_out + VP8Scan[n], 1);
   }