shithub: libvpx

Info  •  Files  •  Log  •  Branches

Merge "vpx_dsp: added vpx_highbd_idct32x32_1_add_sse2()"

--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl

+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -695,7 +695,7 @@

   specialize qw/vpx_highbd_idct32x32_34_add/;

   add_proto qw/void vpx_highbd_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";

-  specialize qw/vpx_highbd_idct32x32_1_add/;

+  specialize qw/vpx_highbd_idct32x32_1_add sse2/;

   add_proto qw/void vpx_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";

   specialize qw/vpx_highbd_iwht4x4_1_add/;

--- a/vpx_dsp/x86/inv_txfm_sse2.c

+++ b/vpx_dsp/x86/inv_txfm_sse2.c

@@ -4020,4 +4020,32 @@

     }

   }

 }

+

+void vpx_highbd_idct32x32_1_add_sse2(const tran_low_t *input, uint8_t *dest8,

+                                     int stride, int bd) {

+  __m128i dc_value, d;

+  const __m128i zero = _mm_setzero_si128();

+  const __m128i one = _mm_set1_epi16(1);

+  const __m128i max = _mm_subs_epi16(_mm_slli_epi16(one, bd), one);

+  int a, i, j;

+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);

+  tran_low_t out;

+

+  out = highbd_dct_const_round_shift(input[0] * cospi_16_64);

+  out = highbd_dct_const_round_shift(out * cospi_16_64);

+  a = ROUND_POWER_OF_TWO(out, 6);

+

+  d = _mm_set1_epi32(a);

+  dc_value = _mm_packs_epi32(d, d);

+  for (i = 0; i < 32; ++i) {

+    for (j = 0; j < 4; ++j) {

+      d = _mm_loadu_si128((const __m128i *)(&dest[j * 8]));

+      d = _mm_adds_epi16(d, dc_value);

+      d = _mm_max_epi16(d, zero);

+      d = _mm_min_epi16(d, max);

+      _mm_storeu_si128((__m128i *)(&dest[j * 8]), d);

+    }

+    dest += stride;

+  }

+}

 #endif  // CONFIG_VP9_HIGHBITDEPTH