AVX2 LowBD av1_convolve_2d_sr()
Speed Comparison with C-code:
SSE2 : 13.11x AVX2 : 17.61x
Change-Id: I5a5672e0f8489add988323f62f43c40adfb9db35
diff --git a/aom_dsp/x86/convolve_avx2.h b/aom_dsp/x86/convolve_avx2.h
index 8924033..86345bb 100644
--- a/aom_dsp/x86/convolve_avx2.h
+++ b/aom_dsp/x86/convolve_avx2.h
@@ -311,6 +311,167 @@
s[6] = s[7]; \
}
+#define CONVOLVE_SR_HORIZONTAL_FILTER_12TAP \
+ const __m256i v_zero = _mm256_setzero_si256(); \
+ __m256i s[12]; \
+ if (w <= 4) { \
+ for (i = 0; i < im_h; i += 2) { \
+ const __m256i data = _mm256_permute2x128_si256( \
+ _mm256_castsi128_si256( \
+ _mm_loadu_si128((__m128i *)(&src_ptr[i * src_stride + j]))), \
+ _mm256_castsi128_si256(_mm_loadu_si128( \
+ (__m128i *)(&src_ptr[i * src_stride + src_stride + j]))), \
+ 0x20); \
+ const __m256i s_16l = _mm256_unpacklo_epi8(data, v_zero); \
+ const __m256i s_16h = _mm256_unpackhi_epi8(data, v_zero); \
+ const __m256i s_ll = _mm256_unpacklo_epi16(s_16l, s_16l); \
+ const __m256i s_lh = _mm256_unpackhi_epi16(s_16l, s_16l); \
+ \
+ const __m256i s_hl = _mm256_unpacklo_epi16(s_16h, s_16h); \
+ const __m256i s_hh = _mm256_unpackhi_epi16(s_16h, s_16h); \
+ \
+ s[0] = _mm256_alignr_epi8(s_lh, s_ll, 2); \
+ s[1] = _mm256_alignr_epi8(s_lh, s_ll, 10); \
+ s[2] = _mm256_alignr_epi8(s_hl, s_lh, 2); \
+ s[3] = _mm256_alignr_epi8(s_hl, s_lh, 10); \
+ s[4] = _mm256_alignr_epi8(s_hh, s_hl, 2); \
+ s[5] = _mm256_alignr_epi8(s_hh, s_hl, 10); \
+ \
+ const __m256i res_lo = convolve_12taps(s, coeffs_h); \
+ \
+ __m256i res_32b_lo = _mm256_sra_epi32( \
+ _mm256_add_epi32(res_lo, round_const_h12), round_shift_h12); \
+ __m256i res_16b_lo = _mm256_packs_epi32(res_32b_lo, res_32b_lo); \
+ const __m128i res_0 = _mm256_extracti128_si256(res_16b_lo, 0); \
+ const __m128i res_1 = _mm256_extracti128_si256(res_16b_lo, 1); \
+ if (w > 2) { \
+ _mm_storel_epi64((__m128i *)&im_block[i * im_stride], res_0); \
+ _mm_storel_epi64((__m128i *)&im_block[i * im_stride + im_stride], \
+ res_1); \
+ } else { \
+ uint32_t horiz_2; \
+ horiz_2 = (uint32_t)_mm_cvtsi128_si32(res_0); \
+ im_block[i * im_stride] = (uint16_t)horiz_2; \
+ im_block[i * im_stride + 1] = (uint16_t)(horiz_2 >> 16); \
+ horiz_2 = (uint32_t)_mm_cvtsi128_si32(res_1); \
+ im_block[i * im_stride + im_stride] = (uint16_t)horiz_2; \
+ im_block[i * im_stride + im_stride + 1] = (uint16_t)(horiz_2 >> 16); \
+ } \
+ } \
+ } else { \
+ for (i = 0; i < im_h; i++) { \
+ const __m256i data = _mm256_permute2x128_si256( \
+ _mm256_castsi128_si256( \
+ _mm_loadu_si128((__m128i *)(&src_ptr[i * src_stride + j]))), \
+ _mm256_castsi128_si256( \
+ _mm_loadu_si128((__m128i *)(&src_ptr[i * src_stride + j + 4]))), \
+ 0x20); \
+ const __m256i s_16l = _mm256_unpacklo_epi8(data, v_zero); \
+ const __m256i s_16h = _mm256_unpackhi_epi8(data, v_zero); \
+ \
+ const __m256i s_ll = _mm256_unpacklo_epi16(s_16l, s_16l); \
+ const __m256i s_lh = _mm256_unpackhi_epi16(s_16l, s_16l); \
+ \
+ const __m256i s_hl = _mm256_unpacklo_epi16(s_16h, s_16h); \
+ const __m256i s_hh = _mm256_unpackhi_epi16(s_16h, s_16h); \
+ \
+ s[0] = _mm256_alignr_epi8(s_lh, s_ll, 2); \
+ s[1] = _mm256_alignr_epi8(s_lh, s_ll, 10); \
+ s[2] = _mm256_alignr_epi8(s_hl, s_lh, 2); \
+ s[3] = _mm256_alignr_epi8(s_hl, s_lh, 10); \
+ s[4] = _mm256_alignr_epi8(s_hh, s_hl, 2); \
+ s[5] = _mm256_alignr_epi8(s_hh, s_hl, 10); \
+ \
+ const __m256i res_lo = convolve_12taps(s, coeffs_h); \
+ \
+ __m256i res_32b_lo = _mm256_sra_epi32( \
+ _mm256_add_epi32(res_lo, round_const_h12), round_shift_h12); \
+ \
+ __m256i res_16b_lo = _mm256_packs_epi32(res_32b_lo, res_32b_lo); \
+ _mm_store_si128((__m128i *)&im_block[i * im_stride], \
+ _mm256_extracti128_si256( \
+ _mm256_permute4x64_epi64(res_16b_lo, 0x88), 0)); \
+ } \
+ }
+
+#define CONVOLVE_SR_VERTICAL_FILTER_12TAP \
+ __m256i src_0 = _mm256_loadu_si256((__m256i *)(im_block + 0 * im_stride)); \
+ __m256i src_1 = _mm256_loadu_si256((__m256i *)(im_block + 1 * im_stride)); \
+ __m256i src_2 = _mm256_loadu_si256((__m256i *)(im_block + 2 * im_stride)); \
+ __m256i src_3 = _mm256_loadu_si256((__m256i *)(im_block + 3 * im_stride)); \
+ __m256i src_4 = _mm256_loadu_si256((__m256i *)(im_block + 4 * im_stride)); \
+ __m256i src_5 = _mm256_loadu_si256((__m256i *)(im_block + 5 * im_stride)); \
+ __m256i src_6 = _mm256_loadu_si256((__m256i *)(im_block + 6 * im_stride)); \
+ __m256i src_7 = _mm256_loadu_si256((__m256i *)(im_block + 7 * im_stride)); \
+ __m256i src_8 = _mm256_loadu_si256((__m256i *)(im_block + 8 * im_stride)); \
+ __m256i src_9 = _mm256_loadu_si256((__m256i *)(im_block + 9 * im_stride)); \
+ \
+ s[0] = _mm256_unpacklo_epi16(src_0, src_1); \
+ s[1] = _mm256_unpacklo_epi16(src_2, src_3); \
+ s[2] = _mm256_unpacklo_epi16(src_4, src_5); \
+ s[3] = _mm256_unpacklo_epi16(src_6, src_7); \
+ s[4] = _mm256_unpacklo_epi16(src_8, src_9); \
+ \
+ s[6] = _mm256_unpackhi_epi16(src_0, src_1); \
+ s[7] = _mm256_unpackhi_epi16(src_2, src_3); \
+ s[8] = _mm256_unpackhi_epi16(src_4, src_5); \
+ s[9] = _mm256_unpackhi_epi16(src_6, src_7); \
+ s[10] = _mm256_unpackhi_epi16(src_8, src_9); \
+ \
+ for (i = 0; i < h; i += 2) { \
+ const int16_t *data = &im_block[i * im_stride]; \
+ \
+ const __m256i s6 = _mm256_loadu_si256((__m256i *)(data + 10 * im_stride)); \
+ const __m256i s7 = _mm256_loadu_si256((__m256i *)(data + 11 * im_stride)); \
+ \
+ s[5] = _mm256_unpacklo_epi16(s6, s7); \
+ s[11] = _mm256_unpackhi_epi16(s6, s7); \
+ \
+ __m256i res_a = convolve_12taps(s, coeffs_v); \
+ __m256i res_b = convolve_12taps(s + 6, coeffs_v); \
+ \
+ res_a = \
+ _mm256_sra_epi32(_mm256_add_epi32(res_a, sum_round_v), sum_shift_v); \
+ res_b = \
+ _mm256_sra_epi32(_mm256_add_epi32(res_b, sum_round_v), sum_shift_v); \
+ \
+ const __m256i res_a_round = _mm256_sra_epi32( \
+ _mm256_add_epi32(res_a, round_const_v), round_shift_v); \
+ const __m256i res_b_round = _mm256_sra_epi32( \
+ _mm256_add_epi32(res_b, round_const_v), round_shift_v); \
+ \
+ const __m256i res_16bit = _mm256_packs_epi32(res_a_round, res_b_round); \
+ const __m256i res_8b = _mm256_packus_epi16(res_16bit, res_16bit); \
+ \
+ const __m128i res_0 = _mm256_castsi256_si128(res_8b); \
+ const __m128i res_1 = _mm256_extracti128_si256(res_8b, 1); \
+ \
+ __m128i *const p_0 = (__m128i *)&dst[i * dst_stride + j]; \
+ __m128i *const p_1 = (__m128i *)&dst[i * dst_stride + j + dst_stride]; \
+ if (w - j > 4) { \
+ _mm_storel_epi64(p_0, res_0); \
+ _mm_storel_epi64(p_1, res_1); \
+ } else if (w == 4) { \
+ xx_storel_32(p_0, res_0); \
+ xx_storel_32(p_1, res_1); \
+ } else { \
+ *(uint16_t *)p_0 = (uint16_t)_mm_cvtsi128_si32(res_0); \
+ *(uint16_t *)p_1 = (uint16_t)_mm_cvtsi128_si32(res_1); \
+ } \
+ \
+ s[0] = s[1]; \
+ s[1] = s[2]; \
+ s[2] = s[3]; \
+ s[3] = s[4]; \
+ s[4] = s[5]; \
+ \
+ s[6] = s[7]; \
+ s[7] = s[8]; \
+ s[8] = s[9]; \
+ s[9] = s[10]; \
+ s[10] = s[11]; \
+ }
+
#define DIST_WTD_CONVOLVE_HORIZONTAL_FILTER_8TAP \
for (i = 0; i < im_h; i += 2) { \
__m256i data = _mm256_castsi128_si256(_mm_loadu_si128((__m128i *)src_h)); \
@@ -427,6 +588,7 @@
s[5] = s[6]; \
s[6] = s[7]; \
}
+
static INLINE void prepare_coeffs_lowbd(
const InterpFilterParams *const filter_params, const int subpel_q4,
__m256i *const coeffs /* [4] */) {
diff --git a/av1/common/x86/convolve_2d_avx2.c b/av1/common/x86/convolve_2d_avx2.c
index f78a7d0..b9f05a3 100644
--- a/av1/common/x86/convolve_2d_avx2.c
+++ b/av1/common/x86/convolve_2d_avx2.c
@@ -27,14 +27,44 @@
const int subpel_x_qn, const int subpel_y_qn,
ConvolveParams *conv_params) {
if (filter_params_x->taps > 8) {
- if (w < 8) {
- av1_convolve_2d_sr_c(src, src_stride, dst, dst_stride, w, h,
- filter_params_x, filter_params_y, subpel_x_qn,
- subpel_y_qn, conv_params);
- } else {
- av1_convolve_2d_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
- filter_params_x, filter_params_y,
- subpel_x_qn, subpel_y_qn, conv_params);
+ const int bd = 8;
+ int im_stride = 8, i;
+ DECLARE_ALIGNED(32, int16_t, im_block[(MAX_SB_SIZE + MAX_FILTER_TAP) * 8]);
+ const int bits =
+ FILTER_BITS * 2 - conv_params->round_0 - conv_params->round_1;
+ const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
+
+ assert(conv_params->round_0 > 0);
+
+ const __m256i round_const_h12 = _mm256_set1_epi32(
+ ((1 << (conv_params->round_0)) >> 1) + (1 << (bd + FILTER_BITS - 1)));
+ const __m128i round_shift_h12 = _mm_cvtsi32_si128(conv_params->round_0);
+
+ const __m256i sum_round_v = _mm256_set1_epi32(
+ (1 << offset_bits) + ((1 << conv_params->round_1) >> 1));
+ const __m128i sum_shift_v = _mm_cvtsi32_si128(conv_params->round_1);
+
+ const __m256i round_const_v = _mm256_set1_epi32(
+ ((1 << bits) >> 1) - (1 << (offset_bits - conv_params->round_1)) -
+ ((1 << (offset_bits - conv_params->round_1)) >> 1));
+ const __m128i round_shift_v = _mm_cvtsi32_si128(bits);
+
+ __m256i coeffs_h[6] = {}, coeffs_v[6] = {};
+
+ int horiz_tap = 12;
+ int vert_tap = 12;
+
+ prepare_coeffs_12taps(filter_params_x, subpel_x_qn, coeffs_h);
+ prepare_coeffs_12taps(filter_params_y, subpel_y_qn, coeffs_v);
+
+ int im_h = h + vert_tap - 1;
+ const int fo_vert = vert_tap / 2 - 1;
+ const int fo_horiz = horiz_tap / 2 - 1;
+ const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
+
+ for (int j = 0; j < w; j += 8) {
+ CONVOLVE_SR_HORIZONTAL_FILTER_12TAP
+ CONVOLVE_SR_VERTICAL_FILTER_12TAP
}
} else {
const int bd = 8;
