Merge "Copy C code to not have multiplication overflow" into main

12 files changed, 109 insertions, 126 deletions

diff --git a/src/dsp/dec.c b/src/dsp/dec.c
index 33d8df8a..451d649d 100644
--- a/src/dsp/dec.c
+++ b/src/dsp/dec.c
@@ -37,9 +37,6 @@ static WEBP_INLINE uint8_t clip_8b(int v) {
   STORE(3, y, DC - (d));            \
 } while (0)
 
-#define MUL1(a) ((((a) * 20091) >> 16) + (a))
-#define MUL2(a) (((a) * 35468) >> 16)
-
 #if !WEBP_NEON_OMIT_C_CODE
 static void TransformOne_C(const int16_t* in, uint8_t* dst) {
   int C[4 * 4], *tmp;
@@ -48,8 +45,10 @@ static void TransformOne_C(const int16_t* in, uint8_t* dst) {
   for (i = 0; i < 4; ++i) {    // vertical pass
     const int a = in[0] + in[8];    // [-4096, 4094]
     const int b = in[0] - in[8];    // [-4095, 4095]
-    const int c = MUL2(in[4]) - MUL1(in[12]);   // [-3783, 3783]
-    const int d = MUL1(in[4]) + MUL2(in[12]);   // [-3785, 3781]
+    const int c = WEBP_TRANSFORM_AC3_MUL2(in[4]) -
+                  WEBP_TRANSFORM_AC3_MUL1(in[12]);  // [-3783, 3783]
+    const int d = WEBP_TRANSFORM_AC3_MUL1(in[4]) +
+                  WEBP_TRANSFORM_AC3_MUL2(in[12]);  // [-3785, 3781]
     tmp[0] = a + d;   // [-7881, 7875]
     tmp[1] = b + c;   // [-7878, 7878]
     tmp[2] = b - c;   // [-7878, 7878]
@@ -69,8 +68,10 @@ static void TransformOne_C(const int16_t* in, uint8_t* dst) {
     const int dc = tmp[0] + 4;
     const int a =  dc +  tmp[8];
     const int b =  dc -  tmp[8];
-    const int c = MUL2(tmp[4]) - MUL1(tmp[12]);
-    const int d = MUL1(tmp[4]) + MUL2(tmp[12]);
+    const int c =
+        WEBP_TRANSFORM_AC3_MUL2(tmp[4]) - WEBP_TRANSFORM_AC3_MUL1(tmp[12]);
+    const int d =
+        WEBP_TRANSFORM_AC3_MUL1(tmp[4]) + WEBP_TRANSFORM_AC3_MUL2(tmp[12]);
     STORE(0, 0, a + d);
     STORE(1, 0, b + c);
     STORE(2, 0, b - c);
@@ -83,17 +84,15 @@ static void TransformOne_C(const int16_t* in, uint8_t* dst) {
 // Simplified transform when only in[0], in[1] and in[4] are non-zero
 static void TransformAC3_C(const int16_t* in, uint8_t* dst) {
   const int a = in[0] + 4;
-  const int c4 = MUL2(in[4]);
-  const int d4 = MUL1(in[4]);
-  const int c1 = MUL2(in[1]);
-  const int d1 = MUL1(in[1]);
+  const int c4 = WEBP_TRANSFORM_AC3_MUL2(in[4]);
+  const int d4 = WEBP_TRANSFORM_AC3_MUL1(in[4]);
+  const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
+  const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
   STORE2(0, a + d4, d1, c1);
   STORE2(1, a + c4, d1, c1);
   STORE2(2, a - c4, d1, c1);
   STORE2(3, a - d4, d1, c1);
 }
-#undef MUL1
-#undef MUL2
 #undef STORE2
 
 static void TransformTwo_C(const int16_t* in, uint8_t* dst, int do_two) {
diff --git a/src/dsp/dec_mips32.c b/src/dsp/dec_mips32.c
index e4e70966..f0e7de4a 100644
--- a/src/dsp/dec_mips32.c
+++ b/src/dsp/dec_mips32.c
@@ -18,8 +18,8 @@
 
 #include "src/dsp/mips_macro.h"
 
-static const int kC1 = 20091 + (1 << 16);
-static const int kC2 = 35468;
+static const int kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int kC2 = WEBP_TRANSFORM_AC3_C2;
 
 static WEBP_INLINE int abs_mips32(int x) {
   const int sign = x >> 31;
@@ -219,7 +219,7 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
   int temp0, temp1, temp2, temp3, temp4;
   int temp5, temp6, temp7, temp8, temp9;
   int temp10, temp11, temp12, temp13, temp14;
-  int temp15, temp16, temp17, temp18;
+  int temp15, temp16, temp17, temp18, temp19;
   int16_t* p_in = (int16_t*)in;
 
   // loops unrolled and merged to avoid usage of tmp buffer
@@ -233,16 +233,14 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
     "addu     %[temp16], %[temp0],  %[temp8]           \n\t"
     "subu     %[temp0],  %[temp0],  %[temp8]           \n\t"
     "mul      %[temp8],  %[temp4],  %[kC2]             \n\t"
-    "mul      %[temp17], %[temp12], %[kC1]             \n\t"
-    "mul      %[temp4],  %[temp4],  %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp17, temp12)
+    MUL_SHIFT_C1_IO(temp4, temp19)
     "mul      %[temp12], %[temp12], %[kC2]             \n\t"
     "lh       %[temp1],  2(%[in])                      \n\t"
     "lh       %[temp5],  10(%[in])                     \n\t"
     "lh       %[temp9],  18(%[in])                     \n\t"
     "lh       %[temp13], 26(%[in])                     \n\t"
     "sra      %[temp8],  %[temp8],  16                 \n\t"
-    "sra      %[temp17], %[temp17], 16                 \n\t"
-    "sra      %[temp4],  %[temp4],  16                 \n\t"
     "sra      %[temp12], %[temp12], 16                 \n\t"
     "lh       %[temp2],  4(%[in])                      \n\t"
     "lh       %[temp6],  12(%[in])                     \n\t"
@@ -261,49 +259,43 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
     "addu     %[temp12], %[temp0],  %[temp17]          \n\t"
     "subu     %[temp0],  %[temp0],  %[temp17]          \n\t"
     "mul      %[temp9],  %[temp5],  %[kC2]             \n\t"
-    "mul      %[temp17], %[temp13], %[kC1]             \n\t"
-    "mul      %[temp5],  %[temp5],  %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp17, temp13)
+    MUL_SHIFT_C1_IO(temp5, temp19)
     "mul      %[temp13], %[temp13], %[kC2]             \n\t"
     "sra      %[temp9],  %[temp9],  16                 \n\t"
-    "sra      %[temp17], %[temp17], 16                 \n\t"
     "subu     %[temp17], %[temp9],  %[temp17]          \n\t"
-    "sra      %[temp5],  %[temp5],  16                 \n\t"
     "sra      %[temp13], %[temp13], 16                 \n\t"
     "addu     %[temp5],  %[temp5],  %[temp13]          \n\t"
     "addu     %[temp13], %[temp1],  %[temp17]          \n\t"
     "subu     %[temp1],  %[temp1],  %[temp17]          \n\t"
-    "mul      %[temp17], %[temp14], %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp17, temp14)
     "mul      %[temp14], %[temp14], %[kC2]             \n\t"
     "addu     %[temp9],  %[temp16], %[temp5]           \n\t"
     "subu     %[temp5],  %[temp16], %[temp5]           \n\t"
     "addu     %[temp16], %[temp2],  %[temp10]          \n\t"
     "subu     %[temp2],  %[temp2],  %[temp10]          \n\t"
     "mul      %[temp10], %[temp6],  %[kC2]             \n\t"
-    "mul      %[temp6],  %[temp6],  %[kC1]             \n\t"
-    "sra      %[temp17], %[temp17], 16                 \n\t"
+    MUL_SHIFT_C1_IO(temp6, temp19)
     "sra      %[temp14], %[temp14], 16                 \n\t"
     "sra      %[temp10], %[temp10], 16                 \n\t"
-    "sra      %[temp6],  %[temp6],  16                 \n\t"
     "subu     %[temp17], %[temp10], %[temp17]          \n\t"
     "addu     %[temp6],  %[temp6],  %[temp14]          \n\t"
     "addu     %[temp10], %[temp16], %[temp6]           \n\t"
     "subu     %[temp6],  %[temp16], %[temp6]           \n\t"
     "addu     %[temp14], %[temp2],  %[temp17]          \n\t"
     "subu     %[temp2],  %[temp2],  %[temp17]          \n\t"
-    "mul      %[temp17], %[temp15], %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp17, temp15)
     "mul      %[temp15], %[temp15], %[kC2]             \n\t"
     "addu     %[temp16], %[temp3],  %[temp11]          \n\t"
     "subu     %[temp3],  %[temp3],  %[temp11]          \n\t"
     "mul      %[temp11], %[temp7],  %[kC2]             \n\t"
-    "mul      %[temp7],  %[temp7],  %[kC1]             \n\t"
+    MUL_SHIFT_C1_IO(temp7, temp19)
     "addiu    %[temp8],  %[temp8],  4                  \n\t"
     "addiu    %[temp12], %[temp12], 4                  \n\t"
     "addiu    %[temp0],  %[temp0],  4                  \n\t"
     "addiu    %[temp4],  %[temp4],  4                  \n\t"
-    "sra      %[temp17], %[temp17], 16                 \n\t"
     "sra      %[temp15], %[temp15], 16                 \n\t"
     "sra      %[temp11], %[temp11], 16                 \n\t"
-    "sra      %[temp7],  %[temp7],  16                 \n\t"
     "subu     %[temp17], %[temp11], %[temp17]          \n\t"
     "addu     %[temp7],  %[temp7],  %[temp15]          \n\t"
     "addu     %[temp15], %[temp3],  %[temp17]          \n\t"
@@ -313,48 +305,40 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
     "addu     %[temp16], %[temp8],  %[temp10]          \n\t"
     "subu     %[temp8],  %[temp8],  %[temp10]          \n\t"
     "mul      %[temp10], %[temp9],  %[kC2]             \n\t"
-    "mul      %[temp17], %[temp11], %[kC1]             \n\t"
-    "mul      %[temp9],  %[temp9],  %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp17, temp11)
+    MUL_SHIFT_C1_IO(temp9, temp19)
     "mul      %[temp11], %[temp11], %[kC2]             \n\t"
     "sra      %[temp10], %[temp10], 16                 \n\t"
-    "sra      %[temp17], %[temp17], 16                 \n\t"
-    "sra      %[temp9],  %[temp9],  16                 \n\t"
     "sra      %[temp11], %[temp11], 16                 \n\t"
     "subu     %[temp17], %[temp10], %[temp17]          \n\t"
     "addu     %[temp11], %[temp9],  %[temp11]          \n\t"
     "addu     %[temp10], %[temp12], %[temp14]          \n\t"
     "subu     %[temp12], %[temp12], %[temp14]          \n\t"
     "mul      %[temp14], %[temp13], %[kC2]             \n\t"
-    "mul      %[temp9],  %[temp15], %[kC1]             \n\t"
-    "mul      %[temp13], %[temp13], %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp9, temp15)
+    MUL_SHIFT_C1_IO(temp13, temp19)
     "mul      %[temp15], %[temp15], %[kC2]             \n\t"
     "sra      %[temp14], %[temp14], 16                 \n\t"
-    "sra      %[temp9],  %[temp9],  16                 \n\t"
-    "sra      %[temp13], %[temp13], 16                 \n\t"
     "sra      %[temp15], %[temp15], 16                 \n\t"
     "subu     %[temp9],  %[temp14], %[temp9]           \n\t"
     "addu     %[temp15], %[temp13], %[temp15]          \n\t"
     "addu     %[temp14], %[temp0],  %[temp2]           \n\t"
     "subu     %[temp0],  %[temp0],  %[temp2]           \n\t"
     "mul      %[temp2],  %[temp1],  %[kC2]             \n\t"
-    "mul      %[temp13], %[temp3],  %[kC1]             \n\t"
-    "mul      %[temp1],  %[temp1],  %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp13, temp3)
+    MUL_SHIFT_C1_IO(temp1, temp19)
     "mul      %[temp3],  %[temp3],  %[kC2]             \n\t"
     "sra      %[temp2],  %[temp2],  16                 \n\t"
-    "sra      %[temp13], %[temp13], 16                 \n\t"
-    "sra      %[temp1],  %[temp1],  16                 \n\t"
     "sra      %[temp3],  %[temp3],  16                 \n\t"
     "subu     %[temp13], %[temp2],  %[temp13]          \n\t"
     "addu     %[temp3],  %[temp1],  %[temp3]           \n\t"
     "addu     %[temp2],  %[temp4],  %[temp6]           \n\t"
     "subu     %[temp4],  %[temp4],  %[temp6]           \n\t"
     "mul      %[temp6],  %[temp5],  %[kC2]             \n\t"
-    "mul      %[temp1],  %[temp7],  %[kC1]             \n\t"
-    "mul      %[temp5],  %[temp5],  %[kC1]             \n\t"
+    MUL_SHIFT_C1(temp1, temp7)
+    MUL_SHIFT_C1_IO(temp5, temp19)
     "mul      %[temp7],  %[temp7],  %[kC2]             \n\t"
     "sra      %[temp6],  %[temp6],  16                 \n\t"
-    "sra      %[temp1],  %[temp1],  16                 \n\t"
-    "sra      %[temp5],  %[temp5],  16                 \n\t"
     "sra      %[temp7],  %[temp7],  16                 \n\t"
     "subu     %[temp1],  %[temp6],  %[temp1]           \n\t"
     "addu     %[temp7],  %[temp5],  %[temp7]           \n\t"
@@ -542,7 +526,7 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
       [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
       [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
-      [temp18]"=&r"(temp18)
+      [temp18]"=&r"(temp18), [temp19]"=&r"(temp19)
     : [in]"r"(p_in), [kC1]"r"(kC1), [kC2]"r"(kC2), [dst]"r"(dst)
     : "memory", "hi", "lo"
   );
diff --git a/src/dsp/dec_mips_dsp_r2.c b/src/dsp/dec_mips_dsp_r2.c
index b0936bc4..0ba706a2 100644
--- a/src/dsp/dec_mips_dsp_r2.c
+++ b/src/dsp/dec_mips_dsp_r2.c
@@ -18,10 +18,8 @@
 
 #include "src/dsp/mips_macro.h"
 
-static const int kC1 = 20091 + (1 << 16);
-static const int kC2 = 35468;
-
-#define MUL(a, b) (((a) * (b)) >> 16)
+static const int kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int kC2 = WEBP_TRANSFORM_AC3_C2;
 
 static void TransformDC(const int16_t* in, uint8_t* dst) {
   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9, temp10;
@@ -49,10 +47,10 @@ static void TransformDC(const int16_t* in, uint8_t* dst) {
 
 static void TransformAC3(const int16_t* in, uint8_t* dst) {
   const int a = in[0] + 4;
-  int c4 = MUL(in[4], kC2);
-  const int d4 = MUL(in[4], kC1);
-  const int c1 = MUL(in[1], kC2);
-  const int d1 = MUL(in[1], kC1);
+  int c4 = WEBP_TRANSFORM_AC3_MUL2(in[4]);
+  const int d4 = WEBP_TRANSFORM_AC3_MUL1(in[4]);
+  const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
+  const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17, temp18;
 
@@ -479,8 +477,6 @@ static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
   FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
 }
 
-#undef MUL
-
 //------------------------------------------------------------------------------
 // Simple In-loop filtering (Paragraph 15.2)
 
diff --git a/src/dsp/dec_msa.c b/src/dsp/dec_msa.c
index 8090622b..58d17301 100644
--- a/src/dsp/dec_msa.c
+++ b/src/dsp/dec_msa.c
@@ -37,8 +37,6 @@
   d1_m = d_tmp1_m + d_tmp2_m;                                    \
   BUTTERFLY_4(a1_m, b1_m, c1_m, d1_m, out0, out1, out2, out3);   \
 }
-#define MULT1(a) ((((a) * 20091) >> 16) + (a))
-#define MULT2(a) (((a) * 35468) >> 16)
 
 static void TransformOne(const int16_t* in, uint8_t* dst) {
   v8i16 input0, input1;
@@ -124,10 +122,10 @@ static void TransformDC(const int16_t* in, uint8_t* dst) {
 
 static void TransformAC3(const int16_t* in, uint8_t* dst) {
   const int a = in[0] + 4;
-  const int c4 = MULT2(in[4]);
-  const int d4 = MULT1(in[4]);
-  const int in2 = MULT2(in[1]);
-  const int in3 = MULT1(in[1]);
+  const int c4 = WEBP_TRANSFORM_AC3_MUL2(in[4]);
+  const int d4 = WEBP_TRANSFORM_AC3_MUL1(in[4]);
+  const int in2 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
+  const int in3 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
   v4i32 tmp0 = { 0 };
   v4i32 out0 = __msa_fill_w(a + d4);
   v4i32 out1 = __msa_fill_w(a + c4);
diff --git a/src/dsp/dec_neon.c b/src/dsp/dec_neon.c
index 22784cf1..83b3a1f9 100644
--- a/src/dsp/dec_neon.c
+++ b/src/dsp/dec_neon.c
@@ -1000,8 +1000,9 @@ static void HFilter8i_NEON(uint8_t* u, uint8_t* v, int stride,
 // libwebp adds 1 << 16 to cospi8sqrt2minus1 (kC1). However, this causes the
 // same issue with kC1 and vqdmulh that we work around by down shifting kC2
 
-static const int16_t kC1 = 20091;
-static const int16_t kC2 = 17734;  // half of kC2, actually. See comment above.
+static const int16_t kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int16_t kC2 =
+    WEBP_TRANSFORM_AC3_C2 / 2;  // half of kC2, actually. See comment above.
 
 #if defined(WEBP_USE_INTRINSICS)
 static WEBP_INLINE void Transpose8x2_NEON(const int16x8_t in0,
@@ -1255,15 +1256,12 @@ static void TransformWHT_NEON(const int16_t* in, int16_t* out) {
 
 //------------------------------------------------------------------------------
 
-#define MUL(a, b) (((a) * (b)) >> 16)
 static void TransformAC3_NEON(const int16_t* in, uint8_t* dst) {
-  static const int kC1_full = 20091 + (1 << 16);
-  static const int kC2_full = 35468;
   const int16x4_t A = vld1_dup_s16(in);
-  const int16x4_t c4 = vdup_n_s16(MUL(in[4], kC2_full));
-  const int16x4_t d4 = vdup_n_s16(MUL(in[4], kC1_full));
-  const int c1 = MUL(in[1], kC2_full);
-  const int d1 = MUL(in[1], kC1_full);
+  const int16x4_t c4 = vdup_n_s16(WEBP_TRANSFORM_AC3_MUL2(in[4]));
+  const int16x4_t d4 = vdup_n_s16(WEBP_TRANSFORM_AC3_MUL1(in[4]));
+  const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
+  const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
   const uint64_t cd = (uint64_t)( d1 & 0xffff) <<  0 |
                       (uint64_t)( c1 & 0xffff) << 16 |
                       (uint64_t)(-c1 & 0xffff) << 32 |
@@ -1274,7 +1272,6 @@ static void TransformAC3_NEON(const int16_t* in, uint8_t* dst) {
   const int16x8_t m2_m3 = vcombine_s16(vqsub_s16(B, c4), vqsub_s16(B, d4));
   Add4x4_NEON(m0_m1, m2_m3, dst);
 }
-#undef MUL
 
 //------------------------------------------------------------------------------
 // 4x4
diff --git a/src/dsp/dec_sse2.c b/src/dsp/dec_sse2.c
index a0163c37..ff3a2855 100644
--- a/src/dsp/dec_sse2.c
+++ b/src/dsp/dec_sse2.c
@@ -196,15 +196,13 @@ static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
 }
 
 #if (USE_TRANSFORM_AC3 == 1)
-#define MUL(a, b) (((a) * (b)) >> 16)
+
 static void TransformAC3(const int16_t* in, uint8_t* dst) {
-  static const int kC1 = 20091 + (1 << 16);
-  static const int kC2 = 35468;
   const __m128i A = _mm_set1_epi16(in[0] + 4);
-  const __m128i c4 = _mm_set1_epi16(MUL(in[4], kC2));
-  const __m128i d4 = _mm_set1_epi16(MUL(in[4], kC1));
-  const int c1 = MUL(in[1], kC2);
-  const int d1 = MUL(in[1], kC1);
+  const __m128i c4 = _mm_set1_epi16(WEBP_TRANSFORM_AC3_MUL2(in[4]));
+  const __m128i d4 = _mm_set1_epi16(WEBP_TRANSFORM_AC3_MUL1(in[4]));
+  const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
+  const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
   const __m128i CD = _mm_set_epi16(0, 0, 0, 0, -d1, -c1, c1, d1);
   const __m128i B = _mm_adds_epi16(A, CD);
   const __m128i m0 = _mm_adds_epi16(B, d4);
@@ -238,7 +236,7 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
   WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
 }
-#undef MUL
+
 #endif   // USE_TRANSFORM_AC3
 
 //------------------------------------------------------------------------------
diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index d2000b8e..23bc2965 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -203,6 +203,11 @@ extern VP8DecIdct VP8TransformDC;
 extern VP8DecIdct VP8TransformDCUV;
 extern VP8WHT VP8TransformWHT;
 
+#define WEBP_TRANSFORM_AC3_C1 20091
+#define WEBP_TRANSFORM_AC3_C2 35468
+#define WEBP_TRANSFORM_AC3_MUL1(a) ((((a) * WEBP_TRANSFORM_AC3_C1) >> 16) + (a))
+#define WEBP_TRANSFORM_AC3_MUL2(a) (((a) * WEBP_TRANSFORM_AC3_C2) >> 16)
+
 // *dst is the destination block, with stride BPS. Boundary samples are
 // assumed accessible when needed.
 typedef void (*VP8PredFunc)(uint8_t* dst);
diff --git a/src/dsp/enc.c b/src/dsp/enc.c
index 2ba97ba8..395ad05b 100644
--- a/src/dsp/enc.c
+++ b/src/dsp/enc.c
@@ -109,10 +109,6 @@ static WEBP_TSAN_IGNORE_FUNCTION void InitTables(void) {
 #define STORE(x, y, v) \
   dst[(x) + (y) * BPS] = clip_8b(ref[(x) + (y) * BPS] + ((v) >> 3))
 
-static const int kC1 = 20091 + (1 << 16);
-static const int kC2 = 35468;
-#define MUL(a, b) (((a) * (b)) >> 16)
-
 static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
                                       uint8_t* dst) {
   int C[4 * 4], *tmp;
@@ -121,8 +117,10 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   for (i = 0; i < 4; ++i) {    // vertical pass
     const int a = in[0] + in[8];
     const int b = in[0] - in[8];
-    const int c = MUL(in[4], kC2) - MUL(in[12], kC1);
-    const int d = MUL(in[4], kC1) + MUL(in[12], kC2);
+    const int c =
+        WEBP_TRANSFORM_AC3_MUL2(in[4]) - WEBP_TRANSFORM_AC3_MUL1(in[12]);
+    const int d =
+        WEBP_TRANSFORM_AC3_MUL1(in[4]) + WEBP_TRANSFORM_AC3_MUL2(in[12]);
     tmp[0] = a + d;
     tmp[1] = b + c;
     tmp[2] = b - c;
@@ -134,10 +132,12 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   tmp = C;
   for (i = 0; i < 4; ++i) {    // horizontal pass
     const int dc = tmp[0] + 4;
-    const int a =  dc +  tmp[8];
-    const int b =  dc -  tmp[8];
-    const int c = MUL(tmp[4], kC2) - MUL(tmp[12], kC1);
-    const int d = MUL(tmp[4], kC1) + MUL(tmp[12], kC2);
+    const int a = dc + tmp[8];
+    const int b = dc - tmp[8];
+    const int c =
+        WEBP_TRANSFORM_AC3_MUL2(tmp[4]) - WEBP_TRANSFORM_AC3_MUL1(tmp[12]);
+    const int d =
+        WEBP_TRANSFORM_AC3_MUL1(tmp[4]) + WEBP_TRANSFORM_AC3_MUL2(tmp[12]);
     STORE(0, i, a + d);
     STORE(1, i, b + c);
     STORE(2, i, b - c);
@@ -222,7 +222,6 @@ static void FTransformWHT_C(const int16_t* in, int16_t* out) {
 }
 #endif  // !WEBP_NEON_OMIT_C_CODE
 
-#undef MUL
 #undef STORE
 
 //------------------------------------------------------------------------------
diff --git a/src/dsp/enc_mips32.c b/src/dsp/enc_mips32.c
index 618f0fc0..50518a5f 100644
--- a/src/dsp/enc_mips32.c
+++ b/src/dsp/enc_mips32.c
@@ -21,8 +21,8 @@
 #include "src/enc/vp8i_enc.h"
 #include "src/enc/cost_enc.h"
 
-static const int kC1 = 20091 + (1 << 16);
-static const int kC2 = 35468;
+static const int kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int kC2 = WEBP_TRANSFORM_AC3_C2;
 
 // macro for one vertical pass in ITransformOne
 // MUL macro inlined
@@ -30,7 +30,7 @@ static const int kC2 = 35468;
 // A..D - offsets in bytes to load from in buffer
 // TEMP0..TEMP3 - registers for corresponding tmp elements
 // TEMP4..TEMP5 - temporary registers
-#define VERTICAL_PASS(A, B, C, D, TEMP4, TEMP0, TEMP1, TEMP2, TEMP3)        \
+#define VERTICAL_PASS(A, B, C, D, TEMP4, TEMP0, TEMP1, TEMP2, TEMP3) \
   "lh      %[temp16],      " #A "(%[temp20])                 \n\t"          \
   "lh      %[temp18],      " #B "(%[temp20])                 \n\t"          \
   "lh      %[temp17],      " #C "(%[temp20])                 \n\t"          \
@@ -38,12 +38,10 @@ static const int kC2 = 35468;
   "addu    %[" #TEMP4 "],    %[temp16],      %[temp18]       \n\t"          \
   "subu    %[temp16],      %[temp16],      %[temp18]         \n\t"          \
   "mul     %[" #TEMP0 "],    %[temp17],      %[kC2]          \n\t"          \
-  "mul     %[temp18],      %[temp19],      %[kC1]            \n\t"          \
-  "mul     %[temp17],      %[temp17],      %[kC1]            \n\t"          \
+  MUL_SHIFT_C1_IO(temp17, temp18)                                           \
+  MUL_SHIFT_C1(temp18, temp19)                                              \
   "mul     %[temp19],      %[temp19],      %[kC2]            \n\t"          \
   "sra     %[" #TEMP0 "],    %[" #TEMP0 "],    16            \n\n"          \
-  "sra     %[temp18],      %[temp18],      16                \n\n"          \
-  "sra     %[temp17],      %[temp17],      16                \n\n"          \
   "sra     %[temp19],      %[temp19],      16                \n\n"          \
   "subu    %[" #TEMP2 "],    %[" #TEMP0 "],    %[temp18]     \n\t"          \
   "addu    %[" #TEMP3 "],    %[temp17],      %[temp19]       \n\t"          \
@@ -58,17 +56,15 @@ static const int kC2 = 35468;
 // temp0..temp15 holds tmp[0]..tmp[15]
 // A - offset in bytes to load from ref and store to dst buffer
 // TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
-#define HORIZONTAL_PASS(A, TEMP0, TEMP4, TEMP8, TEMP12)                       \
+#define HORIZONTAL_PASS(A, TEMP0, TEMP4, TEMP8, TEMP12) \
   "addiu   %[" #TEMP0 "],    %[" #TEMP0 "],    4               \n\t"          \
   "addu    %[temp16],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
   "subu    %[temp17],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
   "mul     %[" #TEMP0 "],    %[" #TEMP4 "],    %[kC2]          \n\t"          \
-  "mul     %[" #TEMP8 "],    %[" #TEMP12 "],   %[kC1]          \n\t"          \
-  "mul     %[" #TEMP4 "],    %[" #TEMP4 "],    %[kC1]          \n\t"          \
+  MUL_SHIFT_C1_IO(TEMP4, TEMP8)                                               \
+  MUL_SHIFT_C1(TEMP8, TEMP12)                                                 \
   "mul     %[" #TEMP12 "],   %[" #TEMP12 "],   %[kC2]          \n\t"          \
   "sra     %[" #TEMP0 "],    %[" #TEMP0 "],    16              \n\t"          \
-  "sra     %[" #TEMP8 "],    %[" #TEMP8 "],    16              \n\t"          \
-  "sra     %[" #TEMP4 "],    %[" #TEMP4 "],    16              \n\t"          \
   "sra     %[" #TEMP12 "],   %[" #TEMP12 "],   16              \n\t"          \
   "subu    %[temp18],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
   "addu    %[temp19],      %[" #TEMP4 "],    %[" #TEMP12 "]    \n\t"          \
diff --git a/src/dsp/enc_mips_dsp_r2.c b/src/dsp/enc_mips_dsp_r2.c
index 9ddd8950..e1431f3b 100644
--- a/src/dsp/enc_mips_dsp_r2.c
+++ b/src/dsp/enc_mips_dsp_r2.c
@@ -20,8 +20,8 @@
 #include "src/enc/cost_enc.h"
 #include "src/enc/vp8i_enc.h"
 
-static const int kC1 = 20091 + (1 << 16);
-static const int kC2 = 35468;
+static const int kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int kC2 = WEBP_TRANSFORM_AC3_C2;
 
 // O - output
 // I - input (macro doesn't change it)
diff --git a/src/dsp/enc_neon.c b/src/dsp/enc_neon.c
index 71480036..6f641c9a 100644
--- a/src/dsp/enc_neon.c
+++ b/src/dsp/enc_neon.c
@@ -27,8 +27,9 @@
 // This code is pretty much the same as TransformOne in the dec_neon.c, except
 // for subtraction to *ref. See the comments there for algorithmic explanations.
 
-static const int16_t kC1 = 20091;
-static const int16_t kC2 = 17734;  // half of kC2, actually. See comment above.
+static const int16_t kC1 = WEBP_TRANSFORM_AC3_C1;
+static const int16_t kC2 =
+    WEBP_TRANSFORM_AC3_C2 / 2;  // half of kC2, actually. See comment above.
 
 // This code works but is *slower* than the inlined-asm version below
 // (with gcc-4.6). So we disable it for now. Later, it'll be conditional to
diff --git a/src/dsp/mips_macro.h b/src/dsp/mips_macro.h
index 44aba9b7..e810d3d3 100644
--- a/src/dsp/mips_macro.h
+++ b/src/dsp/mips_macro.h
@@ -45,28 +45,38 @@
   "ulw    %[" #O2 "],    " #I3 "+" XSTR(I9) "*" #I7 "(%[" #I0 "])       \n\t"  \
   "ulw    %[" #O3 "],    " #I4 "+" XSTR(I9) "*" #I8 "(%[" #I0 "])       \n\t"
 
+
+// O - output
+// I - input (macro doesn't change it so it should be different from I)
+#define MUL_SHIFT_C1(O, I)                                                     \
+  "mul              %[" #O "],    %[" #I "],    %[kC1]        \n\t"            \
+  "sra              %[" #O "],    %[" #O "],    16            \n\t"            \
+  "addu             %[" #O "],    %[" #O "],    %[" #I "]     \n\t"
+#define MUL_SHIFT_C2(O, I) \
+  "mul              %[" #O "],    %[" #I "],    %[kC2]        \n\t"            \
+  "sra              %[" #O "],    %[" #O "],    16            \n\t"
+
+// Same as #define MUL_SHIFT_C1 but I and O are the same. It stores the
+// intermediary result in TMP.
+#define MUL_SHIFT_C1_IO(IO, TMP)                                               \
+  "mul              %[" #TMP "],  %[" #IO  "], %[kC1]     \n\t"                \
+  "sra              %[" #TMP "],  %[" #TMP "], 16         \n\t"                \
+  "addu             %[" #IO  "],  %[" #TMP "], %[" #IO "] \n\t"
+
 // O - output
 // IO - input/output
 // I - input (macro doesn't change it)
 #define MUL_SHIFT_SUM(O0, O1, O2, O3, O4, O5, O6, O7,                          \
                       IO0, IO1, IO2, IO3,                                      \
                       I0, I1, I2, I3, I4, I5, I6, I7)                          \
-  "mul              %[" #O0 "],   %[" #I0 "],   %[kC2]        \n\t"            \
-  "mul              %[" #O1 "],   %[" #I0 "],   %[kC1]        \n\t"            \
-  "mul              %[" #O2 "],   %[" #I1 "],   %[kC2]        \n\t"            \
-  "mul              %[" #O3 "],   %[" #I1 "],   %[kC1]        \n\t"            \
-  "mul              %[" #O4 "],   %[" #I2 "],   %[kC2]        \n\t"            \
-  "mul              %[" #O5 "],   %[" #I2 "],   %[kC1]        \n\t"            \
-  "mul              %[" #O6 "],   %[" #I3 "],   %[kC2]        \n\t"            \
-  "mul              %[" #O7 "],   %[" #I3 "],   %[kC1]        \n\t"            \
-  "sra              %[" #O0 "],   %[" #O0 "],   16            \n\t"            \
-  "sra              %[" #O1 "],   %[" #O1 "],   16            \n\t"            \
-  "sra              %[" #O2 "],   %[" #O2 "],   16            \n\t"            \
-  "sra              %[" #O3 "],   %[" #O3 "],   16            \n\t"            \
-  "sra              %[" #O4 "],   %[" #O4 "],   16            \n\t"            \
-  "sra              %[" #O5 "],   %[" #O5 "],   16            \n\t"            \
-  "sra              %[" #O6 "],   %[" #O6 "],   16            \n\t"            \
-  "sra              %[" #O7 "],   %[" #O7 "],   16            \n\t"            \
+  MUL_SHIFT_C2(O0, I0)                                                         \
+  MUL_SHIFT_C1(O1, I0)                                                         \
+  MUL_SHIFT_C2(O2, I1)                                                         \
+  MUL_SHIFT_C1(O3, I1)                                                         \
+  MUL_SHIFT_C2(O4, I2)                                                         \
+  MUL_SHIFT_C1(O5, I2)                                                         \
+  MUL_SHIFT_C2(O6, I3)                                                         \
+  MUL_SHIFT_C1(O7, I3)                                                         \
   "addu             %[" #IO0 "],  %[" #IO0 "],  %[" #I4 "]    \n\t"            \
   "addu             %[" #IO1 "],  %[" #IO1 "],  %[" #I5 "]    \n\t"            \
   "subu             %[" #IO2 "],  %[" #IO2 "],  %[" #I6 "]    \n\t"            \