shithub: libvpx

Info  •  Files  •  Log  •  Branches

Merge "Change C and msa to match results from sse2."

--- a/test/pp_filter_test.cc

+++ b/test/pp_filter_test.cc

@@ -272,7 +272,7 @@

   delete[] c_mem;

 }

-TEST_P(VpxMbPostProcDownTest, DISABLED_CheckCvsAssembly) {

+TEST_P(VpxMbPostProcDownTest, CheckCvsAssembly) {

   const int rows = 16;

   const int cols = 16;

   const int src_pitch = cols;

@@ -308,6 +308,10 @@

 INSTANTIATE_TEST_CASE_P(C, VpxMbPostProcAcrossIpTest,

                         ::testing::Values(vpx_mbpost_proc_across_ip_c));

+

+INSTANTIATE_TEST_CASE_P(C, VpxMbPostProcDownTest,

+                        ::testing::Values(vpx_mbpost_proc_down_c));

+

 #if HAVE_SSE2

 INSTANTIATE_TEST_CASE_P(

     SSE2, VpxPostProcDownAndAcrossMbRowTest,

@@ -327,6 +331,9 @@

 INSTANTIATE_TEST_CASE_P(MSA, VpxMbPostProcAcrossIpTest,

                         ::testing::Values(vpx_mbpost_proc_across_ip_msa));

+

+INSTANTIATE_TEST_CASE_P(MSA, VpxMbPostProcDownTest,

+                        ::testing::Values(vpx_mbpost_proc_down_msa));

 #endif

 }  // namespace

--- a/vpx_dsp/deblock.c

+++ b/vpx_dsp/deblock.c

@@ -156,7 +156,6 @@

 void vpx_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols,

                             int flimit) {

   int r, c, i;

-  const int16_t *rv3 = &vpx_rv[63 & rand()];

   for (c = 0; c < cols; c++) {

     unsigned char *s = &dst[c];

@@ -163,7 +162,6 @@

     int sumsq = 0;

     int sum = 0;

     unsigned char d[16];

-    const int16_t *rv2 = rv3 + ((c * 17) & 127);

     for (i = -8; i < 0; i++) s[i * pitch] = s[0];

@@ -183,7 +181,7 @@

       d[r & 15] = s[0];

       if (sumsq * 15 - sum * sum < flimit) {

-        d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;

+        d[r & 15] = (vpx_rv[(r & 127) + (c & 7)] + sum + s[0]) >> 4;

       }

       if (r >= 8) s[-8 * pitch] = d[(r - 8) & 15];

       s += pitch;

--- a/vpx_dsp/mips/deblock_msa.c

+++ b/vpx_dsp/mips/deblock_msa.c

@@ -573,7 +573,6 @@

 void vpx_mbpost_proc_down_msa(uint8_t *dst_ptr, int32_t pitch, int32_t rows,

                               int32_t cols, int32_t flimit) {

   int32_t row, col, cnt, i;

-  const int16_t *rv3 = &vpx_rv[63 & rand()];

   v4i32 flimit_vec;

   v16u8 dst7, dst8, dst_r_b, dst_l_b;

   v16i8 mask;

@@ -601,7 +600,7 @@

     dst = LD_UB(dst_tmp);

     for (cnt = (col << 4), i = 0; i < 16; ++cnt) {

-      rv2[i] = rv3 + ((cnt * 17) & 127);

+      rv2[i] = vpx_rv + (i & 7);

       ++i;

     }

     for (cnt = -8; cnt < 0; ++cnt) {