shithub: libvpx

Info  •  Files  •  Log  •  Branches

vpx_fdct32x32_1_msa: fix accumulator overflow

Change-Id: I33a5432eda3416382e1cea06b45082c0c65faa75

--- a/vpx_dsp/mips/fwd_dct32x32_msa.c

+++ b/vpx_dsp/mips/fwd_dct32x32_msa.c

@@ -933,23 +933,21 @@

 }

 void vpx_fdct32x32_1_msa(const int16_t *input, int16_t *out, int32_t stride) {

-  out[1] = 0;

-

-  out[0] = LD_HADD(input, stride);

-  out[0] += LD_HADD(input + 8, stride);

-  out[0] += LD_HADD(input + 16, stride);

-  out[0] += LD_HADD(input + 24, stride);

-  out[0] += LD_HADD(input + 32 * 8, stride);

-  out[0] += LD_HADD(input + 32 * 8 + 8, stride);

-  out[0] += LD_HADD(input + 32 * 8 + 16, stride);

-  out[0] += LD_HADD(input + 32 * 8 + 24, stride);

-  out[0] += LD_HADD(input + 32 * 16, stride);

-  out[0] += LD_HADD(input + 32 * 16 + 8, stride);

-  out[0] += LD_HADD(input + 32 * 16 + 16, stride);

-  out[0] += LD_HADD(input + 32 * 16 + 24, stride);

-  out[0] += LD_HADD(input + 32 * 24, stride);

-  out[0] += LD_HADD(input + 32 * 24 + 8, stride);

-  out[0] += LD_HADD(input + 32 * 24 + 16, stride);

-  out[0] += LD_HADD(input + 32 * 24 + 24, stride);

-  out[0] >>= 3;

+  int sum = LD_HADD(input, stride);

+  sum += LD_HADD(input + 8, stride);

+  sum += LD_HADD(input + 16, stride);

+  sum += LD_HADD(input + 24, stride);

+  sum += LD_HADD(input + 32 * 8, stride);

+  sum += LD_HADD(input + 32 * 8 + 8, stride);

+  sum += LD_HADD(input + 32 * 8 + 16, stride);

+  sum += LD_HADD(input + 32 * 8 + 24, stride);

+  sum += LD_HADD(input + 32 * 16, stride);

+  sum += LD_HADD(input + 32 * 16 + 8, stride);

+  sum += LD_HADD(input + 32 * 16 + 16, stride);

+  sum += LD_HADD(input + 32 * 16 + 24, stride);

+  sum += LD_HADD(input + 32 * 24, stride);

+  sum += LD_HADD(input + 32 * 24 + 8, stride);

+  sum += LD_HADD(input + 32 * 24 + 16, stride);

+  sum += LD_HADD(input + 32 * 24 + 24, stride);

+  out[0] = (int16_t)(sum >> 3);

 }