Add sse2 version of aom_h_predictor_wxh()
for width or height equal to 64.
Change-Id: Idb173dc76aed06fb71037125421bd0a554764481
diff --git a/aom_dsp/aom_dsp_rtcd_defs.pl b/aom_dsp/aom_dsp_rtcd_defs.pl
index 66828f7..041dbe6 100755
--- a/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -153,6 +153,10 @@
specialize qw/aom_h_predictor_16x64 sse2/;
specialize qw/aom_h_predictor_32x16 sse2/;
specialize qw/aom_h_predictor_32x32 neon msa sse2 avx2/;
+specialize qw/aom_h_predictor_32x64 sse2/;
+specialize qw/aom_h_predictor_64x64 sse2/;
+specialize qw/aom_h_predictor_64x32 sse2/;
+specialize qw/aom_h_predictor_64x16 sse2/;
specialize qw/aom_paeth_predictor_4x4 ssse3/;
specialize qw/aom_paeth_predictor_4x8 ssse3/;
specialize qw/aom_paeth_predictor_8x4 ssse3/;
diff --git a/aom_dsp/x86/intrapred_sse2.c b/aom_dsp/x86/intrapred_sse2.c
index 5f5a20f..7c19fab 100644
--- a/aom_dsp/x86/intrapred_sse2.c
+++ b/aom_dsp/x86/intrapred_sse2.c
@@ -1019,3 +1019,83 @@
dst += stride << 2;
h_prediction_32x8_2(&left_col_8p, dst, stride);
}
+
+static INLINE void h_predictor_32xh(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *left, int height) {
+ int i = height >> 2;
+ do {
+ __m128i left4 = _mm_cvtsi32_si128(((uint32_t *)left)[0]);
+ left4 = _mm_unpacklo_epi8(left4, left4);
+ left4 = _mm_unpacklo_epi8(left4, left4);
+ const __m128i r0 = _mm_shuffle_epi32(left4, 0x0);
+ const __m128i r1 = _mm_shuffle_epi32(left4, 0x55);
+ _mm_store_si128((__m128i *)dst, r0);
+ _mm_store_si128((__m128i *)(dst + 16), r0);
+ _mm_store_si128((__m128i *)(dst + stride), r1);
+ _mm_store_si128((__m128i *)(dst + stride + 16), r1);
+ const __m128i r2 = _mm_shuffle_epi32(left4, 0xaa);
+ const __m128i r3 = _mm_shuffle_epi32(left4, 0xff);
+ _mm_store_si128((__m128i *)(dst + stride * 2), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 2 + 16), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 3), r3);
+ _mm_store_si128((__m128i *)(dst + stride * 3 + 16), r3);
+ left += 4;
+ dst += stride * 4;
+ } while (--i);
+}
+
+void aom_h_predictor_32x64_sse2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {
+ (void)above;
+ h_predictor_32xh(dst, stride, left, 64);
+}
+
+static INLINE void h_predictor_64xh(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *left, int height) {
+ int i = height >> 2;
+ do {
+ __m128i left4 = _mm_cvtsi32_si128(((uint32_t *)left)[0]);
+ left4 = _mm_unpacklo_epi8(left4, left4);
+ left4 = _mm_unpacklo_epi8(left4, left4);
+ const __m128i r0 = _mm_shuffle_epi32(left4, 0x0);
+ const __m128i r1 = _mm_shuffle_epi32(left4, 0x55);
+ _mm_store_si128((__m128i *)dst, r0);
+ _mm_store_si128((__m128i *)(dst + 16), r0);
+ _mm_store_si128((__m128i *)(dst + 32), r0);
+ _mm_store_si128((__m128i *)(dst + 48), r0);
+ _mm_store_si128((__m128i *)(dst + stride), r1);
+ _mm_store_si128((__m128i *)(dst + stride + 16), r1);
+ _mm_store_si128((__m128i *)(dst + stride + 32), r1);
+ _mm_store_si128((__m128i *)(dst + stride + 48), r1);
+ const __m128i r2 = _mm_shuffle_epi32(left4, 0xaa);
+ const __m128i r3 = _mm_shuffle_epi32(left4, 0xff);
+ _mm_store_si128((__m128i *)(dst + stride * 2), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 2 + 16), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 2 + 32), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 2 + 48), r2);
+ _mm_store_si128((__m128i *)(dst + stride * 3), r3);
+ _mm_store_si128((__m128i *)(dst + stride * 3 + 16), r3);
+ _mm_store_si128((__m128i *)(dst + stride * 3 + 32), r3);
+ _mm_store_si128((__m128i *)(dst + stride * 3 + 48), r3);
+ left += 4;
+ dst += stride * 4;
+ } while (--i);
+}
+
+void aom_h_predictor_64x64_sse2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {
+ (void)above;
+ h_predictor_64xh(dst, stride, left, 64);
+}
+
+void aom_h_predictor_64x32_sse2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {
+ (void)above;
+ h_predictor_64xh(dst, stride, left, 32);
+}
+
+void aom_h_predictor_64x16_sse2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {
+ (void)above;
+ h_predictor_64xh(dst, stride, left, 16);
+}
diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index 3b5e2cc..ccfbb0f 100644
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -701,7 +701,7 @@
aom_dc_left_predictor_32x64_sse2,
aom_dc_top_predictor_32x64_sse2,
aom_dc_128_predictor_32x64_sse2, aom_v_predictor_32x64_sse2,
- NULL, NULL, NULL, NULL, NULL)
+ aom_h_predictor_32x64_sse2, NULL, NULL, NULL, NULL)
#endif // HAVE_SSE2
#if HAVE_SSSE3
@@ -779,17 +779,17 @@
aom_dc_left_predictor_64x64_sse2,
aom_dc_top_predictor_64x64_sse2,
aom_dc_128_predictor_64x64_sse2, aom_v_predictor_64x64_sse2,
- NULL, NULL, NULL, NULL, NULL)
+ aom_h_predictor_64x64_sse2, NULL, NULL, NULL, NULL)
INTRA_PRED_TEST(SSE2_5, TX_64X32, aom_dc_predictor_64x32_sse2,
aom_dc_left_predictor_64x32_sse2,
aom_dc_top_predictor_64x32_sse2,
aom_dc_128_predictor_64x32_sse2, aom_v_predictor_64x32_sse2,
- NULL, NULL, NULL, NULL, NULL)
+ aom_h_predictor_64x32_sse2, NULL, NULL, NULL, NULL)
INTRA_PRED_TEST(SSE2_6, TX_64X16, aom_dc_predictor_64x16_sse2,
aom_dc_left_predictor_64x16_sse2,
aom_dc_top_predictor_64x16_sse2,
aom_dc_128_predictor_64x16_sse2, aom_v_predictor_64x16_sse2,
- NULL, NULL, NULL, NULL, NULL)
+ aom_h_predictor_64x16_sse2, NULL, NULL, NULL, NULL)
#endif
#if HAVE_SSSE3
