| /****************************************************************************** |
| * |
| * Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore |
| * |
| * Licensed under the Apache License, Version 2.0 (the "License"); |
| * you may not use this file except in compliance with the License. |
| * You may obtain a copy of the License at: |
| * |
| * http://www.apache.org/licenses/LICENSE-2.0 |
| * |
| * Unless required by applicable law or agreed to in writing, software |
| * distributed under the License is distributed on an "AS IS" BASIS, |
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| * See the License for the specific language governing permissions and |
| * limitations under the License. |
| * |
| ******************************************************************************/ |
| |
| /** |
| ******************************************************************************* |
| * @file |
| * ihevc_inter_pred_filters_atom_intr.c |
| * |
| * @brief |
| * Contains function definitions for inter prediction interpolation filters |
| * coded in x86 intrinsics |
| * |
| * |
| * @author |
| * |
| * |
| * @par List of Functions: |
| * - ihevc_inter_pred_luma_copy_ssse3() |
| * - ihevc_inter_pred_luma_horz_ssse3() |
| * - ihevc_inter_pred_luma_vert_ssse3() |
| * - ihevc_inter_pred_luma_copy_w16out_ssse3() |
| * - ihevc_inter_pred_luma_horz_w16out_ssse3() |
| * - ihevc_inter_pred_luma_vert_w16out_ssse3() |
| * - ihevc_inter_pred_luma_vert_w16inp_ssse3() |
| * - ihevc_inter_pred_luma_vert_w16inp_w16out_ssse3() |
| * - ihevc_inter_pred_chroma_copy_ssse3() |
| * - ihevc_inter_pred_chroma_horz_ssse3() |
| * - ihevc_inter_pred_chroma_vert_ssse3() |
| * - ihevc_inter_pred_chroma_copy_w16out_ssse3() |
| * - ihevc_inter_pred_chroma_horz_w16out_ssse3() |
| * - ihevc_inter_pred_chroma_vert_w16out_ssse3() |
| * - ihevc_inter_pred_chroma_vert_w16inp_ssse3() |
| * - ihevc_inter_pred_chroma_vert_w16inp_w16out_ssse3() |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| |
| |
| /*****************************************************************************/ |
| /* File Includes */ |
| /*****************************************************************************/ |
| #include <assert.h> |
| |
| #include "ihevc_debug.h" |
| #include "ihevc_typedefs.h" |
| #include "ihevc_defs.h" |
| #include "ihevc_inter_pred.h" |
| #include "ihevc_platform_macros.h" |
| #include "ihevc_macros.h" |
| #include "ihevc_func_selector.h" |
| |
| #include <immintrin.h> |
| |
| /*****************************************************************************/ |
| /* Function Definitions */ |
| /*****************************************************************************/ |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma function for copy |
| * |
| * @par Description: |
| * Copies the array of width 'wd' and height 'ht' from the location pointed |
| * by 'src' to the location pointed by 'dst' |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| * Assumption : ht%4 == 0, wd%4 == 0 |
| * |
| ******************************************************************************* |
| */ |
| |
| |
| void ihevc_inter_pred_luma_copy_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| |
| WORD32 row, col; |
| __m128i src0_16x8b, src1_16x8b, src2_16x8b, src3_16x8b; |
| UNUSED(pi1_coeff); |
| ASSERT(wd % 4 == 0); /* checking assumption*/ |
| ASSERT(ht % 4 == 0); /* checking assumption*/ |
| |
| /* outer for loop starts from here */ |
| if(0 == (wd & 15)) /* wd multiple of 16 case */ |
| { |
| for(row = 0; row < ht; row += 4) |
| { |
| for(col = 0; col < wd; col += 16) |
| { |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| src0_16x8b = _mm_loadu_si128((__m128i *)(pu1_src)); /* row =0 */ |
| src1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + 1 * src_strd)); /* row =1 */ |
| src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + 2 * src_strd)); /* row =2 */ |
| src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + 3 * src_strd)); /* row =3 */ |
| |
| /* storing 16 8-bit output values */ |
| _mm_storeu_si128((__m128i *)(pu1_dst), src0_16x8b); /* row =0 */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + 1 * dst_strd), src1_16x8b); /* row =1 */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + 2 * dst_strd), src2_16x8b); /* row =2 */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + 3 * dst_strd), src3_16x8b); /* row =3 */ |
| |
| pu1_src += 16; /* pointer update */ |
| pu1_dst += 16; /* pointer update */ |
| } /* inner for loop ends here(16-output values in single iteration) */ |
| |
| pu1_src += 4 * src_strd - wd; /* pointer update */ |
| pu1_dst += 4 * dst_strd - wd; /* pointer update */ |
| } |
| |
| } |
| else if(0 == (wd & 7)) /* multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row += 4) |
| { |
| for(col = 0; col < wd; col += 8) |
| { |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| src0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); /* row =0 */ |
| src1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 1 * src_strd)); /* row =1 */ |
| src2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 2 * src_strd)); /* row =2 */ |
| src3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 3 * src_strd)); /* row =3 */ |
| |
| /* storing 16 8-bit output values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst), src0_16x8b); /* row =0 */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + 1 * dst_strd), src1_16x8b); /* row =1 */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + 2 * dst_strd), src2_16x8b); /* row =2 */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + 3 * dst_strd), src3_16x8b); /* row =3 */ |
| |
| pu1_src += 8; /* pointer update */ |
| pu1_dst += 8; /* pointer update */ |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src += 4 * src_strd - wd; /* pointer update */ |
| pu1_dst += 4 * dst_strd - wd; /* pointer update */ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| WORD32 dst0, dst1, dst2, dst3; |
| for(row = 0; row < ht; row += 4) |
| { |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| src0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); /* row =0 */ |
| src1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 1 * src_strd)); /* row =1 */ |
| src2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 2 * src_strd)); /* row =2 */ |
| src3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + 3 * src_strd)); /* row =3 */ |
| |
| dst0 = _mm_cvtsi128_si32(src0_16x8b); |
| dst1 = _mm_cvtsi128_si32(src1_16x8b); |
| dst2 = _mm_cvtsi128_si32(src2_16x8b); |
| dst3 = _mm_cvtsi128_si32(src3_16x8b); |
| |
| /* storing 4 8-bit output values */ |
| *(WORD32 *)(&pu1_dst[0 * dst_strd]) = dst0; /* row =0 */ |
| *(WORD32 *)(&pu1_dst[1 * dst_strd]) = dst1; /* row =1 */ |
| *(WORD32 *)(&pu1_dst[2 * dst_strd]) = dst2; /* row =2 */ |
| *(WORD32 *)(&pu1_dst[3 * dst_strd]) = dst3; /* row =3 */ |
| |
| pu1_src += 4; /* pointer update */ |
| pu1_dst += 4; /* pointer update */ |
| } /* inner for loop ends here(4- output values in single iteration) */ |
| |
| pu1_src += 4 * src_strd - wd; /* pointer update */ |
| pu1_dst += 4 * dst_strd - wd; /* pointer update */ |
| } |
| } |
| } |
| |
| /* INTER_PRED_LUMA_COPY */ |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma filter for horizontal input |
| * |
| * @par Description: |
| * Applies a horizontal filter with coefficients pointed to by 'pi1_coeff' |
| * to the elements pointed by 'pu1_src' and writes to the location pointed |
| * by 'pu1_dst' The output is downshifted by 6 and clipped to 8 bits |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_horz_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| |
| /* all 128 bit registers are named with a suffix mxnb, where m is the */ |
| /* number of n bits packed in the register */ |
| __m128i zero_8x16b, offset_8x16b, mask_low_32b, mask_high_96b; |
| __m128i src_temp1_16x8b, src_temp2_16x8b, src_temp3_16x8b, src_temp4_16x8b, src_temp5_16x8b, src_temp6_16x8b; |
| __m128i src_temp11_16x8b, src_temp12_16x8b, src_temp13_16x8b, src_temp14_16x8b, src_temp15_16x8b, src_temp16_16x8b; |
| __m128i res_temp1_8x16b, res_temp2_8x16b, res_temp3_8x16b, res_temp4_8x16b, res_temp5_8x16b, res_temp6_8x16b, res_temp7_8x16b, res_temp8_8x16b; |
| __m128i res_temp11_8x16b, res_temp12_8x16b, res_temp13_8x16b, res_temp14_8x16b, res_temp15_8x16b, res_temp16_8x16b, res_temp17_8x16b, res_temp18_8x16b; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b, control_mask_3_8x16b, control_mask_4_8x16b; |
| |
| ASSERT(wd % 4 == 0); /* checking assumption*/ |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| src_temp1_16x8b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| zero_8x16b = _mm_set1_epi32(0); |
| offset_8x16b = _mm_set1_epi16(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| |
| mask_low_32b = _mm_cmpeq_epi16(zero_8x16b, zero_8x16b); |
| mask_high_96b = _mm_srli_si128(mask_low_32b, 12); |
| mask_low_32b = _mm_slli_si128(mask_low_32b, 4); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| control_mask_3_8x16b = _mm_set1_epi32(0x05040504); /* Control Mask register */ |
| control_mask_4_8x16b = _mm_set1_epi32(0x07060706); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_3_8x16b); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_4_8x16b); /* pi1_coeff[4] */ |
| |
| if(0 == (ht & 1)) /* ht multiple of 2 case */ |
| { |
| |
| if(0 == (wd & 7)) /* wd = multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| /*load 16 pixel values from row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| |
| /*load 16 pixel values from row 1*/ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| res_temp6_8x16b = _mm_adds_epi16(res_temp5_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp6_8x16b = _mm_srai_epi16(res_temp6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp5_8x16b = _mm_packus_epi16(res_temp6_8x16b, res_temp6_8x16b); /* row = 0 */ |
| |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp5_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row =1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row =1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row =1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row =1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row =1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| res_temp16_8x16b = _mm_adds_epi16(res_temp15_8x16b, offset_8x16b); /* row = 1 */ |
| res_temp16_8x16b = _mm_srai_epi16(res_temp16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 1 */ |
| res_temp15_8x16b = _mm_packus_epi16(res_temp16_8x16b, res_temp16_8x16b); /* row = 1 */ |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 8; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* pointer updates*/ |
| pu1_dst += 2 * dst_strd; /* pointer updates*/ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| res_temp6_8x16b = _mm_adds_epi16(res_temp5_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp6_8x16b = _mm_srai_epi16(res_temp6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp5_8x16b = _mm_packus_epi16(res_temp6_8x16b, res_temp6_8x16b); /* row = 0 */ |
| |
| res_temp7_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + offset)); |
| res_temp8_8x16b = _mm_and_si128(res_temp7_8x16b, mask_low_32b); |
| res_temp7_8x16b = _mm_and_si128(res_temp5_8x16b, mask_high_96b); |
| res_temp5_8x16b = _mm_or_si128(res_temp7_8x16b, res_temp8_8x16b); |
| |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp5_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row = 1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| res_temp16_8x16b = _mm_adds_epi16(res_temp15_8x16b, offset_8x16b); /* row = 1 */ |
| res_temp16_8x16b = _mm_srai_epi16(res_temp16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 1 */ |
| res_temp15_8x16b = _mm_packus_epi16(res_temp16_8x16b, res_temp16_8x16b); /* row = 1 */ |
| |
| res_temp17_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd + offset)); |
| res_temp18_8x16b = _mm_and_si128(res_temp17_8x16b, mask_low_32b); |
| res_temp17_8x16b = _mm_and_si128(res_temp15_8x16b, mask_high_96b); |
| res_temp15_8x16b = _mm_or_si128(res_temp17_8x16b, res_temp18_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* Pointer update */ |
| pu1_dst += 2 * dst_strd; /* Pointer update */ |
| } |
| } |
| } |
| else /* odd ht */ |
| { |
| if(0 == (wd & 7)) /* multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row++) |
| { |
| int offset = 0; |
| |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| res_temp6_8x16b = _mm_adds_epi16(res_temp5_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp6_8x16b = _mm_srai_epi16(res_temp6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp5_8x16b = _mm_packus_epi16(res_temp6_8x16b, res_temp6_8x16b); /* row = 0 */ |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp5_8x16b); |
| |
| offset += 8; /* To pointer updates*/ |
| } |
| pu1_src += src_strd; /* pointer updates*/ |
| pu1_dst += dst_strd; /* pointer updates*/ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| for(row = 0; row < (ht - 1); row += 2) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| res_temp6_8x16b = _mm_adds_epi16(res_temp5_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp6_8x16b = _mm_srai_epi16(res_temp6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp5_8x16b = _mm_packus_epi16(res_temp6_8x16b, res_temp6_8x16b); /* row = 0 */ |
| |
| res_temp7_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + offset)); |
| res_temp8_8x16b = _mm_and_si128(res_temp7_8x16b, mask_low_32b); |
| res_temp7_8x16b = _mm_and_si128(res_temp5_8x16b, mask_high_96b); |
| res_temp5_8x16b = _mm_or_si128(res_temp7_8x16b, res_temp8_8x16b); |
| |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp5_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row = 1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| res_temp16_8x16b = _mm_adds_epi16(res_temp15_8x16b, offset_8x16b); /* row = 1 */ |
| res_temp16_8x16b = _mm_srai_epi16(res_temp16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 1 */ |
| res_temp15_8x16b = _mm_packus_epi16(res_temp16_8x16b, res_temp16_8x16b); /* row = 1 */ |
| |
| res_temp17_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd + offset)); |
| res_temp18_8x16b = _mm_and_si128(res_temp17_8x16b, mask_low_32b); |
| res_temp17_8x16b = _mm_and_si128(res_temp15_8x16b, mask_high_96b); |
| res_temp15_8x16b = _mm_or_si128(res_temp17_8x16b, res_temp18_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* Pointer update */ |
| pu1_dst += 2 * dst_strd; /* Pointer update */ |
| } |
| { /* last repeat at outside the loop */ |
| int offset = 0; |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| res_temp6_8x16b = _mm_adds_epi16(res_temp5_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp6_8x16b = _mm_srai_epi16(res_temp6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp5_8x16b = _mm_packus_epi16(res_temp6_8x16b, res_temp6_8x16b); /* row = 0 */ |
| |
| res_temp7_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + offset)); |
| res_temp8_8x16b = _mm_and_si128(res_temp7_8x16b, mask_low_32b); |
| res_temp7_8x16b = _mm_and_si128(res_temp5_8x16b, mask_high_96b); |
| res_temp5_8x16b = _mm_or_si128(res_temp7_8x16b, res_temp8_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp5_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| } |
| } |
| } |
| } |
| |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma filter for vertical input |
| * |
| * @par Description: |
| * Applies a vertcal filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' The output is downshifted by 6 and clipped to 8 bits |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_vert_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| UWORD8 *pu1_src_copy; |
| UWORD8 *pu1_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i s0_8x16b, s1_8x16b, s2_8x16b, s3_8x16b, s4_8x16b, s5_8x16b, s6_8x16b, s7_8x16b, s8_8x16b, s9_8x16b; |
| __m128i s2_0_16x8b, s2_1_16x8b, s2_2_16x8b, s2_3_16x8b, s2_4_16x8b, s2_5_16x8b, s2_6_16x8b, s2_7_16x8b, s2_8_16x8b, s2_9_16x8b, s2_10_16x8b; |
| __m128i s3_0_16x8b, s3_1_16x8b, s3_2_16x8b, s3_3_16x8b, s3_4_16x8b; |
| __m128i s4_0_16x8b, s4_1_16x8b, s4_2_16x8b, s4_3_16x8b, s4_4_16x8b; |
| __m128i s10_8x16b, s11_8x16b, s12_8x16b, s13_8x16b, s14_8x16b, s15_8x16b, s16_8x16b, s17_8x16b, s18_8x16b, s19_8x16b; |
| __m128i s20_8x16b, s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b, s26_8x16b, s27_8x16b, s28_8x16b, s29_8x16b; |
| __m128i s30_8x16b, s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b, s36_8x16b, s37_8x16b, s38_8x16b, s39_8x16b; |
| |
| __m128i zero_8x16b, offset_8x16b, mask_low_32b, mask_high_96b; |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b, control_mask_3_8x16b, control_mask_4_8x16b; |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| control_mask_3_8x16b = _mm_set1_epi32(0x05040504); /* Control Mask register */ |
| control_mask_4_8x16b = _mm_set1_epi32(0x07060706); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_3_8x16b); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_4_8x16b); /* pi1_coeff[4] */ |
| |
| /* seting values in register */ |
| zero_8x16b = _mm_setzero_si128(); /* for saturated clipping */ |
| offset_8x16b = _mm_set1_epi16(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| mask_low_32b = _mm_set_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000); |
| mask_high_96b = _mm_set_epi32(0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF); |
| |
| /* outer for loop starts from here */ |
| if(wd % 8 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pu1_src_copy = pu1_src; |
| pu1_dst_copy = pu1_dst; |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pu1_dst = pu1_dst_copy + col; |
| |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| /*load 8 pixel values.*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-3 * src_strd))); |
| |
| /*load 8 pixel values*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 8 pixel values*/ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values*/ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi8(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values*/ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*load 8 pixel values*/ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| s7_8x16b = _mm_add_epi16(s6_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi16(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values*/ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (5 * src_strd))); |
| |
| /*load 8 pixel values*/ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| s27_8x16b = _mm_add_epi16(s26_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi16(s27_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s29_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi8(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi8(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| s17_8x16b = _mm_add_epi16(s16_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi16(s17_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (dst_strd)), s19_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values*/ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| s37_8x16b = _mm_add_epi16(s36_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi16(s37_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s39_8x16b); |
| |
| pu1_src += (8 * src_strd); |
| pu1_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| s7_8x16b = _mm_add_epi16(s6_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi16(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); |
| |
| /*load 8 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| s27_8x16b = _mm_add_epi16(s26_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi16(s27_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s29_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| s17_8x16b = _mm_add_epi16(s16_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi16(s17_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s19_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| s37_8x16b = _mm_add_epi16(s36_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi16(s37_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s39_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| pu1_src += 4 * src_strd; /* pointer update */ |
| pu1_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| else /* wd = multiple of 8 case */ |
| { |
| |
| pu1_src_copy = pu1_src; |
| pu1_dst_copy = pu1_dst; |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pu1_dst = pu1_dst_copy + col; |
| |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values */ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi8(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| s7_8x16b = _mm_add_epi16(s6_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi16(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| s5_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s6_8x16b = _mm_and_si128(s5_8x16b, mask_low_32b); |
| s7_8x16b = _mm_and_si128(s9_8x16b, mask_high_96b); |
| s8_8x16b = _mm_or_si128(s6_8x16b, s7_8x16b); |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s8_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (5 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| s27_8x16b = _mm_add_epi16(s26_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi16(s27_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (2 * dst_strd))); |
| s26_8x16b = _mm_and_si128(s25_8x16b, mask_low_32b); |
| s27_8x16b = _mm_and_si128(s29_8x16b, mask_high_96b); |
| s28_8x16b = _mm_or_si128(s26_8x16b, s27_8x16b); |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s28_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi8(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi8(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| s17_8x16b = _mm_add_epi16(s16_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi16(s17_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| s15_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd)); |
| s16_8x16b = _mm_and_si128(s15_8x16b, mask_low_32b); |
| s17_8x16b = _mm_and_si128(s19_8x16b, mask_high_96b); |
| s18_8x16b = _mm_or_si128(s16_8x16b, s17_8x16b); |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (dst_strd)), s18_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| s37_8x16b = _mm_add_epi16(s36_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi16(s37_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| s35_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (3 * dst_strd))); |
| s36_8x16b = _mm_and_si128(s35_8x16b, mask_low_32b); |
| s37_8x16b = _mm_and_si128(s39_8x16b, mask_high_96b); |
| s38_8x16b = _mm_or_si128(s36_8x16b, s37_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s38_8x16b); |
| |
| pu1_src += (8 * src_strd); |
| pu1_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 16 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| s7_8x16b = _mm_add_epi16(s6_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi16(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| |
| s5_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s6_8x16b = _mm_and_si128(s5_8x16b, mask_low_32b); |
| s7_8x16b = _mm_and_si128(s9_8x16b, mask_high_96b); |
| s8_8x16b = _mm_or_si128(s6_8x16b, s7_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s8_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 16 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); |
| |
| /*load 16 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| s27_8x16b = _mm_add_epi16(s26_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi16(s27_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (2 * dst_strd))); |
| s26_8x16b = _mm_and_si128(s25_8x16b, mask_low_32b); |
| s27_8x16b = _mm_and_si128(s29_8x16b, mask_high_96b); |
| s28_8x16b = _mm_or_si128(s26_8x16b, s27_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s28_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| s17_8x16b = _mm_add_epi16(s16_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi16(s17_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| |
| s15_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd)); |
| s16_8x16b = _mm_and_si128(s15_8x16b, mask_low_32b); |
| s17_8x16b = _mm_and_si128(s19_8x16b, mask_high_96b); |
| s18_8x16b = _mm_or_si128(s16_8x16b, s17_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s18_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 16 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| s37_8x16b = _mm_add_epi16(s36_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi16(s37_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| s35_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (3 * dst_strd))); |
| s36_8x16b = _mm_and_si128(s35_8x16b, mask_low_32b); |
| s37_8x16b = _mm_and_si128(s39_8x16b, mask_high_96b); |
| s38_8x16b = _mm_or_si128(s36_8x16b, s37_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s38_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| pu1_src += 4 * src_strd; /* pointer update */ |
| pu1_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| } |
| |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma filter for copy 16bit output |
| * |
| * @par Description: |
| * Copies the array of width 'wd' and height 'ht' from the location pointed |
| * by 'src' to the location pointed by 'dst' The output is upshifted by 6 |
| * bits and is used as input for vertical filtering or weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| |
| void ihevc_inter_pred_luma_copy_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| __m128i s3, zero_8x16b; |
| |
| ASSERT(wd % 2 == 0); /* checking assumption*/ |
| ASSERT(ht % 2 == 0); /* checking assumption*/ |
| UNUSED(pi1_coeff); |
| zero_8x16b = _mm_setzero_si128(); |
| /* outer for loop starts from here */ |
| if(wd % 8 == 0) /* wd = multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < wd; col += 8) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_store_si128((__m128i *)(pi2_dst + offset), s3); |
| |
| /* row =1 */ |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| |
| offset += 8; /* To pointer update */ |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pi2_dst += 2 * dst_strd; /* pointer update */ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < wd; col += 4) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), s3); |
| |
| /* row =1 */ |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| offset += 4; /* To pointer update */ |
| } /* inner for loop ends here(4-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pi2_dst += 2 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma filter for horizontal 16bit output |
| * |
| * @par Description: |
| * Applies a horizontal filter with coefficients pointed to by 'pi1_coeff' |
| * to the elements pointed by 'pu1_src' and writes to the location pointed |
| * by 'pu1_dst' No downshifting or clipping is done and the output is used |
| * as an input for vertical filtering or weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_horz_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| |
| /* all 128 bit registers are named with a suffix mxnb, where m is the */ |
| /* number of n bits packed in the register */ |
| |
| __m128i src_temp1_16x8b, src_temp2_16x8b, src_temp3_16x8b, src_temp4_16x8b, src_temp5_16x8b, src_temp6_16x8b; |
| __m128i src_temp11_16x8b, src_temp12_16x8b, src_temp13_16x8b, src_temp14_16x8b, src_temp15_16x8b, src_temp16_16x8b; |
| __m128i res_temp1_8x16b, res_temp2_8x16b, res_temp3_8x16b, res_temp4_8x16b, res_temp5_8x16b, res_temp6_8x16b; |
| __m128i res_temp11_8x16b, res_temp12_8x16b, res_temp13_8x16b, res_temp14_8x16b, res_temp15_8x16b, res_temp16_8x16b; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b, control_mask_3_8x16b, control_mask_4_8x16b; |
| |
| ASSERT(wd % 4 == 0); /* checking assumption*/ |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| src_temp1_16x8b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| control_mask_3_8x16b = _mm_set1_epi32(0x05040504); /* Control Mask register */ |
| control_mask_4_8x16b = _mm_set1_epi32(0x07060706); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_3_8x16b); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_4_8x16b); /* pi1_coeff[4] */ |
| |
| if(0 == (ht & 1)) /* ht multiple of 2 case */ |
| { |
| |
| if(0 == (wd & 7)) /* wd = multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row = 1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_store_si128((__m128i *)(pi2_dst + offset), res_temp5_8x16b); |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 8; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* pointer updates*/ |
| pi2_dst += 2 * dst_strd; /* pointer updates*/ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row = 1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), res_temp5_8x16b); |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* Pointer update */ |
| pi2_dst += 2 * dst_strd; /* Pointer update */ |
| } |
| } |
| } |
| else /* odd ht */ |
| { |
| if(0 == (wd & 7)) /* multiple of 8 case */ |
| { |
| for(row = 0; row < ht; row++) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_store_si128((__m128i *)(pi2_dst + offset), res_temp5_8x16b); |
| |
| offset += 8; /* To pointer updates*/ |
| } |
| pu1_src += src_strd; /* pointer updates*/ |
| pi2_dst += dst_strd; /* pointer updates*/ |
| } |
| } |
| else /* wd = multiple of 4 case */ |
| { |
| for(row = 0; row < (ht - 1); row += 2) |
| { |
| int offset = 0; |
| |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src + src_strd - 3 + offset)); /* row = 1 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 1); /* row = 1 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp13_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp13_16x8b, coeff0_1_8x16b); /* row = 1 */ |
| /* row = 1 */ |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp14_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp14_16x8b, coeff2_3_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp13_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff4_5_8x16b); /* row = 1 */ |
| |
| src_temp11_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); /* row = 1 */ |
| src_temp12_16x8b = _mm_srli_si128(src_temp12_16x8b, 2); /* row = 1 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 1 */ |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); /* row = 1 */ |
| res_temp14_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff6_7_8x16b); /* row = 1 */ |
| |
| res_temp15_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| res_temp16_8x16b = _mm_add_epi16(res_temp13_8x16b, res_temp14_8x16b); |
| res_temp15_8x16b = _mm_add_epi16(res_temp15_8x16b, res_temp16_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), res_temp5_8x16b); |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), res_temp15_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| pu1_src += 2 * src_strd; /* Pointer update */ |
| pi2_dst += 2 * dst_strd; /* Pointer update */ |
| } |
| { /* last repeat at outside the loop */ |
| int offset = 0; |
| for(col = 0; col < wd; col += 4) |
| { |
| /*load 16 pixel values from 12:-3 pos. relative to cur. pos.*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 3 + offset)); /* row = 0 */ |
| |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 1); /* row = 0 */ |
| /* pix. |5:-2|4:-3| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp3_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp3_16x8b, coeff0_1_8x16b); /* row = 0 */ |
| /* row = 0 */ |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp4_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp4_16x8b, coeff2_3_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp3_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff4_5_8x16b); /* row = 0 */ |
| |
| src_temp1_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); /* row = 0 */ |
| src_temp2_16x8b = _mm_srli_si128(src_temp2_16x8b, 2); /* row = 0 */ |
| /* pix. |7:0|6:-1| to do two dot-products at same time*/ /* row = 0 */ |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); /* row = 0 */ |
| res_temp4_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff6_7_8x16b); /* row = 0 */ |
| |
| res_temp5_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| res_temp6_8x16b = _mm_add_epi16(res_temp3_8x16b, res_temp4_8x16b); |
| res_temp5_8x16b = _mm_add_epi16(res_temp5_8x16b, res_temp6_8x16b); |
| |
| /* to store the 1st 4 pixels res. */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), res_temp5_8x16b); |
| |
| offset += 4; /* To pointer updates*/ |
| } |
| } |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction luma filter for vertical 16bit output |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' No downshifting or clipping is done and the output is used as |
| * an input for weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_vert_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| UWORD8 *pu1_src_copy; |
| WORD16 *pi2_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i s0_8x16b, s1_8x16b, s2_8x16b, s3_8x16b, s4_8x16b, s5_8x16b, s6_8x16b; |
| __m128i s2_0_16x8b, s2_1_16x8b, s2_2_16x8b, s2_3_16x8b, s2_4_16x8b, s2_5_16x8b, s2_6_16x8b, s2_7_16x8b, s2_8_16x8b, s2_9_16x8b, s2_10_16x8b; |
| __m128i s3_0_16x8b, s3_1_16x8b, s3_2_16x8b, s3_3_16x8b, s3_4_16x8b; |
| __m128i s4_0_16x8b, s4_1_16x8b, s4_2_16x8b, s4_3_16x8b, s4_4_16x8b; |
| __m128i s10_8x16b, s11_8x16b, s12_8x16b, s13_8x16b, s14_8x16b, s15_8x16b, s16_8x16b; |
| __m128i s20_8x16b, s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b, s26_8x16b; |
| __m128i s30_8x16b, s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b, s36_8x16b; |
| |
| |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b, control_mask_3_8x16b, control_mask_4_8x16b; |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| control_mask_3_8x16b = _mm_set1_epi32(0x05040504); /* Control Mask register */ |
| control_mask_4_8x16b = _mm_set1_epi32(0x07060706); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_3_8x16b); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_4_8x16b); /* pi1_coeff[4] */ |
| |
| |
| /* outer for loop starts from here */ |
| if((wd % 8) == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pu1_src_copy = pu1_src; |
| pi2_dst_copy = pi2_dst; |
| |
| for(col = 0; col < wd; col += 8) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pi2_dst = pi2_dst_copy + col; |
| |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| /*load 8 pixel values */ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi8(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_store_si128((__m128i *)(pi2_dst), s6_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (5 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_store_si128((__m128i *)(pi2_dst + (2 * dst_strd)), s26_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi8(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi8(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_store_si128((__m128i *)(pi2_dst + (dst_strd)), s16_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_store_si128((__m128i *)(pi2_dst + (3 * dst_strd)), s36_8x16b); |
| |
| pu1_src += (8 * src_strd); |
| pi2_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_store_si128((__m128i *)(pi2_dst), s6_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); |
| |
| /*load 8 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_store_si128((__m128i *)(pi2_dst + (2 * dst_strd)), s26_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd), s16_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_store_si128((__m128i *)(pi2_dst + (3 * dst_strd)), s36_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| |
| pu1_src += 4 * src_strd; /* pointer update */ |
| pi2_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| else /* wd = multiple of 8 case */ |
| { |
| |
| pu1_src_copy = pu1_src; |
| pi2_dst_copy = pi2_dst; |
| |
| for(col = 0; col < wd; col += 4) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pi2_dst = pi2_dst_copy + col; |
| |
| PREFETCH((char const *)(pu1_src + (8 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (9 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (10 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (11 * src_strd)), _MM_HINT_T0) |
| |
| /*load 8 pixel values */ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi8(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s6_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (5 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (2 * dst_strd)), s26_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi8(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi8(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (dst_strd)), s16_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values */ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (3 * dst_strd)), s36_8x16b); |
| |
| pu1_src += (8 * src_strd); |
| pi2_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_maddubs_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi8(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi16(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi16(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi16(s4_8x16b, s5_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s6_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_maddubs_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_maddubs_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_maddubs_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); |
| |
| /*load 8 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi8(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_maddubs_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi16(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi16(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi16(s24_8x16b, s25_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (2 * dst_strd)), s26_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_maddubs_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi8(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi16(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi16(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi16(s14_8x16b, s15_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd), s16_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_maddubs_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_maddubs_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_maddubs_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 8 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi8(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_maddubs_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi16(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi16(s34_8x16b, s35_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (3 * dst_strd)), s36_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| pu1_src += 4 * src_strd; /* pointer update */ |
| pi2_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * |
| * Luma vertical filter for 16bit input. |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' Input is 16 bits The filter output is downshifted by 12 and |
| * clipped to lie between 0 and 255 |
| * |
| * @param[in] pi2_src |
| * WORD16 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_vert_w16inp_ssse3(WORD16 *pi2_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| WORD16 *pi2_src_copy; |
| UWORD8 *pu1_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i s0_8x16b, s1_8x16b, s2_8x16b, s3_8x16b, s4_8x16b, s5_8x16b, s6_8x16b, s8_8x16b, s9_8x16b; |
| __m128i s2_0_16x8b, s2_1_16x8b, s2_2_16x8b, s2_3_16x8b, s2_4_16x8b, s2_5_16x8b, s2_6_16x8b, s2_7_16x8b, s2_8_16x8b, s2_9_16x8b, s2_10_16x8b; |
| __m128i s3_0_16x8b, s3_1_16x8b, s3_2_16x8b, s3_3_16x8b, s3_4_16x8b; |
| __m128i s4_0_16x8b, s4_1_16x8b, s4_2_16x8b, s4_3_16x8b, s4_4_16x8b; |
| __m128i s10_8x16b, s11_8x16b, s12_8x16b, s13_8x16b, s14_8x16b, s15_8x16b, s16_8x16b, s18_8x16b, s19_8x16b; |
| __m128i s20_8x16b, s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b, s26_8x16b, s28_8x16b, s29_8x16b; |
| __m128i s30_8x16b, s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b, s36_8x16b, s38_8x16b, s39_8x16b; |
| |
| __m128i zero_8x16b, offset_8x16b, mask_low_32b, mask_high_96b, sign_reg; |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| zero_8x16b = _mm_setzero_si128(); |
| sign_reg = _mm_cmpgt_epi8(zero_8x16b, s4_8x16b); |
| s5_8x16b = _mm_unpacklo_epi8(s4_8x16b, sign_reg); |
| |
| coeff0_1_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(0, 0, 0, 0)); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(1, 1, 1, 1)); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(2, 2, 2, 2)); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(3, 3, 3, 3)); /* pi1_coeff[4] */ |
| |
| |
| /* seting values in register */ |
| offset_8x16b = _mm_set1_epi32(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| mask_low_32b = _mm_set_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000); |
| mask_high_96b = _mm_set_epi32(0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF); |
| |
| |
| pi2_src_copy = pi2_src; |
| pu1_dst_copy = pu1_dst; |
| |
| /* outer for loop starts from here */ |
| for(col = 0; col < wd; col += 4) |
| { |
| |
| pi2_src = pi2_src_copy + col; |
| pu1_dst = pu1_dst_copy + col; |
| |
| /*load 4 pixel values */ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-3 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi16(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_madd_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 4 pixel values */ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi16(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 4 pixel values */ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi16(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values */ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi16(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi32(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi32(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi32(s4_8x16b, s5_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s9_8x16b = _mm_add_epi32(s8_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s9_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s8_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| |
| s4_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s5_8x16b = _mm_and_si128(s4_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s9_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values */ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (5 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi16(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_madd_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi32(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi32(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi32(s24_8x16b, s25_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s26_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s29_8x16b = _mm_add_epi32(s28_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s29_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s28_8x16b = _mm_packs_epi32(s28_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (2 * dst_strd))); |
| s25_8x16b = _mm_and_si128(s24_8x16b, mask_low_32b); |
| s26_8x16b = _mm_and_si128(s29_8x16b, mask_high_96b); |
| s29_8x16b = _mm_or_si128(s25_8x16b, s26_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s29_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi16(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_madd_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi16(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi16(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi16(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi32(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi32(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi32(s14_8x16b, s15_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s19_8x16b = _mm_add_epi32(s18_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s19_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s18_8x16b = _mm_packs_epi32(s18_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| |
| s14_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (dst_strd))); |
| s15_8x16b = _mm_and_si128(s14_8x16b, mask_low_32b); |
| s16_8x16b = _mm_and_si128(s19_8x16b, mask_high_96b); |
| s19_8x16b = _mm_or_si128(s15_8x16b, s16_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (dst_strd)), s19_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values */ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi16(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_madd_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi32(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi32(s34_8x16b, s35_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s36_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s39_8x16b = _mm_add_epi32(s38_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s39_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s38_8x16b = _mm_packs_epi32(s38_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| s34_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (3 * dst_strd))); |
| s35_8x16b = _mm_and_si128(s34_8x16b, mask_low_32b); |
| s36_8x16b = _mm_and_si128(s39_8x16b, mask_high_96b); |
| s39_8x16b = _mm_or_si128(s35_8x16b, s36_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s39_8x16b); |
| |
| pi2_src += (8 * src_strd); |
| pu1_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_madd_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi16(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi32(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi32(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi32(s4_8x16b, s5_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s9_8x16b = _mm_add_epi32(s8_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s9_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s8_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s9_8x16b = _mm_packus_epi16(s8_8x16b, zero_8x16b); |
| |
| s4_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s5_8x16b = _mm_and_si128(s4_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s9_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + src_strd)); |
| |
| /*load 4 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi16(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_madd_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi32(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi32(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi32(s24_8x16b, s25_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s26_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s29_8x16b = _mm_add_epi32(s28_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s29_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s28_8x16b = _mm_packs_epi32(s28_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s29_8x16b = _mm_packus_epi16(s28_8x16b, zero_8x16b); |
| |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (2 * dst_strd))); |
| s25_8x16b = _mm_and_si128(s24_8x16b, mask_low_32b); |
| s26_8x16b = _mm_and_si128(s29_8x16b, mask_high_96b); |
| s29_8x16b = _mm_or_si128(s25_8x16b, s26_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (2 * dst_strd)), s29_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_madd_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi16(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi32(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi32(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi32(s14_8x16b, s15_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s19_8x16b = _mm_add_epi32(s18_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s19_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s18_8x16b = _mm_packs_epi32(s18_8x16b, zero_8x16b); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s19_8x16b = _mm_packus_epi16(s18_8x16b, zero_8x16b); |
| |
| s14_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd)); |
| s15_8x16b = _mm_and_si128(s14_8x16b, mask_low_32b); |
| s16_8x16b = _mm_and_si128(s19_8x16b, mask_high_96b); |
| s19_8x16b = _mm_or_si128(s15_8x16b, s16_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s19_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi16(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_madd_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi32(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi32(s34_8x16b, s35_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s36_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s39_8x16b = _mm_add_epi32(s38_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s39_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s38_8x16b = _mm_packs_epi32(s38_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s39_8x16b = _mm_packus_epi16(s38_8x16b, zero_8x16b); |
| |
| s34_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + (3 * dst_strd))); |
| s35_8x16b = _mm_and_si128(s34_8x16b, mask_low_32b); |
| s36_8x16b = _mm_and_si128(s39_8x16b, mask_high_96b); |
| s39_8x16b = _mm_or_si128(s35_8x16b, s36_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_storel_epi64((__m128i *)(pu1_dst + (3 * dst_strd)), s39_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| pi2_src += 4 * src_strd; /* pointer update */ |
| pu1_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| } |
| |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Luma prediction filter for vertical 16bit input & output |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' Input is 16 bits The filter output is downshifted by 6 and |
| * 8192 is subtracted to store it as a 16 bit number The output is used as |
| * a input to weighted prediction |
| * |
| * @param[in] pi2_src |
| * WORD16 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_luma_vert_w16inp_w16out_ssse3(WORD16 *pi2_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| WORD16 *pi2_src_copy; |
| WORD16 *pi2_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, coeff4_5_8x16b, coeff6_7_8x16b; |
| __m128i s0_8x16b, s1_8x16b, s2_8x16b, s3_8x16b, s4_8x16b, s5_8x16b, s6_8x16b, s8_8x16b, s9_8x16b; |
| __m128i s2_0_16x8b, s2_1_16x8b, s2_2_16x8b, s2_3_16x8b, s2_4_16x8b, s2_5_16x8b, s2_6_16x8b, s2_7_16x8b, s2_8_16x8b, s2_9_16x8b, s2_10_16x8b; |
| __m128i s3_0_16x8b, s3_1_16x8b, s3_2_16x8b, s3_3_16x8b, s3_4_16x8b; |
| __m128i s4_0_16x8b, s4_1_16x8b, s4_2_16x8b, s4_3_16x8b, s4_4_16x8b; |
| __m128i s10_8x16b, s11_8x16b, s12_8x16b, s13_8x16b, s14_8x16b, s15_8x16b, s16_8x16b, s18_8x16b, s19_8x16b; |
| __m128i s20_8x16b, s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b, s26_8x16b, s28_8x16b, s29_8x16b; |
| __m128i s30_8x16b, s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b, s36_8x16b, s38_8x16b, s39_8x16b; |
| |
| __m128i zero_8x16b, offset_8x16b, sign_reg; |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| zero_8x16b = _mm_setzero_si128(); |
| sign_reg = _mm_cmpgt_epi8(zero_8x16b, s4_8x16b); |
| s5_8x16b = _mm_unpacklo_epi8(s4_8x16b, sign_reg); |
| |
| coeff0_1_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(0, 0, 0, 0)); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(1, 1, 1, 1)); /* pi1_coeff[4] */ |
| |
| coeff4_5_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(2, 2, 2, 2)); /* pi1_coeff[4] */ |
| coeff6_7_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(3, 3, 3, 3)); /* pi1_coeff[4] */ |
| |
| |
| /* seting values in register */ |
| offset_8x16b = _mm_set1_epi32(OFFSET14); /* for offset addition */ |
| |
| pi2_src_copy = pi2_src; |
| pi2_dst_copy = pi2_dst; |
| |
| /* outer for loop starts from here */ |
| for(col = 0; col < wd; col += 4) |
| { |
| |
| pi2_src = pi2_src_copy + col; |
| pi2_dst = pi2_dst_copy + col; |
| |
| /*load 4 pixel values*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-3 * src_strd))); |
| |
| /*load 4 pixel values*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-2 * src_strd))); |
| |
| s3_0_16x8b = _mm_unpacklo_epi16(s2_0_16x8b, s2_1_16x8b); |
| |
| s0_8x16b = _mm_madd_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| |
| /*load 4 pixel values*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 4 pixel values*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| s3_1_16x8b = _mm_unpacklo_epi16(s2_2_16x8b, s2_3_16x8b); |
| |
| s1_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| |
| /*load 4 pixel values*/ |
| s2_4_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 4 pixel values*/ |
| s2_5_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s3_2_16x8b = _mm_unpacklo_epi16(s2_4_16x8b, s2_5_16x8b); |
| |
| s2_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values*/ |
| s2_6_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| /*load 4 pixel values*/ |
| s2_7_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (4 * src_strd))); |
| |
| s3_3_16x8b = _mm_unpacklo_epi16(s2_6_16x8b, s2_7_16x8b); |
| |
| s3_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_8x16b = _mm_add_epi32(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi32(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi32(s4_8x16b, s5_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s9_8x16b = _mm_sub_epi32(s8_8x16b, offset_8x16b); |
| |
| s8_8x16b = _mm_packs_epi32(s9_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 0*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s8_8x16b); |
| |
| /* ROW 2*/ |
| s20_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values*/ |
| s2_8_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (5 * src_strd))); |
| |
| /*load 4 pixel values*/ |
| s2_9_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (6 * src_strd))); |
| |
| s3_4_16x8b = _mm_unpacklo_epi16(s2_8_16x8b, s2_9_16x8b); |
| |
| s23_8x16b = _mm_madd_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi32(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi32(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi32(s24_8x16b, s25_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s26_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s29_8x16b = _mm_sub_epi32(s28_8x16b, offset_8x16b); |
| |
| s28_8x16b = _mm_packs_epi32(s29_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (2 * dst_strd)), s28_8x16b); |
| |
| |
| /*ROW 1*/ |
| s4_0_16x8b = _mm_unpacklo_epi16(s2_1_16x8b, s2_2_16x8b); |
| |
| s10_8x16b = _mm_madd_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| |
| s4_1_16x8b = _mm_unpacklo_epi16(s2_3_16x8b, s2_4_16x8b); |
| |
| s11_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| |
| s4_2_16x8b = _mm_unpacklo_epi16(s2_5_16x8b, s2_6_16x8b); |
| |
| s12_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| s4_3_16x8b = _mm_unpacklo_epi16(s2_7_16x8b, s2_8_16x8b); |
| |
| s13_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi32(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi32(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi32(s14_8x16b, s15_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s19_8x16b = _mm_sub_epi32(s18_8x16b, offset_8x16b); |
| |
| s18_8x16b = _mm_packs_epi32(s19_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 1*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (dst_strd)), s18_8x16b); |
| |
| |
| /* ROW 3*/ |
| s30_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values*/ |
| s2_10_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (7 * src_strd))); |
| |
| s4_4_16x8b = _mm_unpacklo_epi16(s2_9_16x8b, s2_10_16x8b); |
| |
| s33_8x16b = _mm_madd_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi32(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi32(s34_8x16b, s35_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s36_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s39_8x16b = _mm_sub_epi32(s38_8x16b, offset_8x16b); |
| |
| s38_8x16b = _mm_packs_epi32(s39_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 2*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (3 * dst_strd)), s38_8x16b); |
| |
| pi2_src += (8 * src_strd); |
| pi2_dst += (4 * dst_strd); |
| |
| for(row = 4; row < ht; row += 4) |
| { |
| |
| s3_0_16x8b = s3_2_16x8b; |
| s3_1_16x8b = s3_3_16x8b; |
| s3_2_16x8b = s3_4_16x8b; |
| |
| s0_8x16b = _mm_madd_epi16(s3_0_16x8b, coeff0_1_8x16b); |
| s1_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff2_3_8x16b); |
| s2_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 4)th row*/ |
| s2_0_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src)); |
| |
| s3_3_16x8b = _mm_unpacklo_epi16(s2_10_16x8b, s2_0_16x8b); |
| s3_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff6_7_8x16b); |
| |
| s4_0_16x8b = s4_2_16x8b; |
| s4_1_16x8b = s4_3_16x8b; |
| s4_2_16x8b = s4_4_16x8b; |
| |
| s4_8x16b = _mm_add_epi32(s0_8x16b, s1_8x16b); |
| s5_8x16b = _mm_add_epi32(s2_8x16b, s3_8x16b); |
| s6_8x16b = _mm_add_epi32(s4_8x16b, s5_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s6_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s9_8x16b = _mm_sub_epi32(s8_8x16b, offset_8x16b); |
| |
| s8_8x16b = _mm_packs_epi32(s9_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of row 4*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s8_8x16b); |
| |
| /* row + 2*/ |
| s20_8x16b = _mm_madd_epi16(s3_1_16x8b, coeff0_1_8x16b); |
| s21_8x16b = _mm_madd_epi16(s3_2_16x8b, coeff2_3_8x16b); |
| s22_8x16b = _mm_madd_epi16(s3_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 5)th row*/ |
| s2_1_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + src_strd)); |
| |
| /*load 4 pixel values from (cur_row + 6)th row*/ |
| s2_2_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| /*unpacking (cur_row + 5)th row and (cur_row + 6)th row*/ |
| s3_4_16x8b = _mm_unpacklo_epi16(s2_1_16x8b, s2_2_16x8b); |
| |
| s23_8x16b = _mm_madd_epi16(s3_4_16x8b, coeff6_7_8x16b); |
| |
| s24_8x16b = _mm_add_epi32(s20_8x16b, s21_8x16b); |
| s25_8x16b = _mm_add_epi32(s22_8x16b, s23_8x16b); |
| s26_8x16b = _mm_add_epi32(s24_8x16b, s25_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s28_8x16b = _mm_srai_epi32(s26_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s29_8x16b = _mm_sub_epi32(s28_8x16b, offset_8x16b); |
| |
| s28_8x16b = _mm_packs_epi32(s29_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+2)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (2 * dst_strd)), s28_8x16b); |
| |
| |
| /*row + 1*/ |
| s10_8x16b = _mm_madd_epi16(s4_0_16x8b, coeff0_1_8x16b); |
| s11_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff2_3_8x16b); |
| s12_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff4_5_8x16b); |
| |
| /*unpacking (cur_row + 4)th row and (cur_row + 5)th row*/ |
| s4_3_16x8b = _mm_unpacklo_epi16(s2_0_16x8b, s2_1_16x8b); |
| s13_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff6_7_8x16b); |
| |
| s14_8x16b = _mm_add_epi32(s10_8x16b, s11_8x16b); |
| s15_8x16b = _mm_add_epi32(s12_8x16b, s13_8x16b); |
| s16_8x16b = _mm_add_epi32(s14_8x16b, s15_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s18_8x16b = _mm_srai_epi32(s16_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s19_8x16b = _mm_sub_epi32(s18_8x16b, offset_8x16b); |
| |
| s18_8x16b = _mm_packs_epi32(s19_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row + 1)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd), s18_8x16b); |
| |
| |
| /* row + 3*/ |
| s30_8x16b = _mm_madd_epi16(s4_1_16x8b, coeff0_1_8x16b); |
| s31_8x16b = _mm_madd_epi16(s4_2_16x8b, coeff2_3_8x16b); |
| s32_8x16b = _mm_madd_epi16(s4_3_16x8b, coeff4_5_8x16b); |
| |
| /*load 4 pixel values from (cur_row + 7)th row*/ |
| s2_3_16x8b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| /*unpacking (cur_row + 6)th row and (cur_row + 7)th row*/ |
| s4_4_16x8b = _mm_unpacklo_epi16(s2_2_16x8b, s2_3_16x8b); |
| |
| s33_8x16b = _mm_madd_epi16(s4_4_16x8b, coeff6_7_8x16b); |
| |
| s34_8x16b = _mm_add_epi32(s30_8x16b, s31_8x16b); |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s33_8x16b); |
| s36_8x16b = _mm_add_epi32(s34_8x16b, s35_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s38_8x16b = _mm_srai_epi32(s36_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s39_8x16b = _mm_sub_epi32(s38_8x16b, offset_8x16b); |
| |
| s38_8x16b = _mm_packs_epi32(s39_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* Store the output pixels of (cur_row+3)*/ |
| _mm_storel_epi64((__m128i *)(pi2_dst + (3 * dst_strd)), s38_8x16b); |
| |
| s2_10_16x8b = s2_3_16x8b; |
| |
| pi2_src += 4 * src_strd; /* pointer update */ |
| pi2_dst += 4 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Chroma interprediction filter for copy |
| * |
| * @par Description: |
| * Copies the array of width 'wd' and height 'ht' from the location pointed |
| * by 'src' to the location pointed by 'dst' |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| |
| void ihevc_inter_pred_chroma_copy_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| __m128i s3, mask_4x32b; |
| UNUSED(pi1_coeff); |
| ASSERT(wd % 2 == 0); /* checking assumption*/ |
| ASSERT(ht % 2 == 0); /* checking assumption*/ |
| |
| mask_4x32b = _mm_set_epi32(0, 0, 0, 0x80808080); /* Mask register */ |
| |
| /* for loop starts from here */ |
| if(wd % 8 == 0) |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < 2 * wd; col += 16) |
| { |
| /* row =0 */ |
| |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col]; */ |
| /* storing 16 8-bit output values */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + offset), s3); /* pu1_dst[col] = pu1_src[col]; */ |
| |
| /* row =1 */ |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); /* pu1_src[col]; */ |
| /* storing 8 8-bit output values */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd + offset), s3); /* pu1_dst[col] = pu1_src[col]*/ |
| |
| offset += 16; /*To pointer update */ |
| } /* inner for loop ends here(16-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pu1_dst += 2 * dst_strd; /* pointer update */ |
| } |
| } |
| else if(wd % 4 == 0) |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col]; */ |
| /* storing 8 8-bit output values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), s3); /* pu1_dst[col] = pu1_src[col]; */ |
| /* row =1 */ |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); /* pu1_src[col]; */ |
| /* storing 8 8-bit output values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), s3); /* pu1_dst[col] = pu1_src[col]; */ |
| |
| offset += 8; /* To pointer update */ |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pu1_dst += 2 * dst_strd; /* pointer update */ |
| } |
| } |
| else |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| /* row =0 */ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col] */ |
| /* storing four 8-bit output values */ |
| _mm_maskmoveu_si128(s3, mask_4x32b, (char *)(pu1_dst + offset)); /* pu1_dst[col] = pu1_src[col]; */ |
| /* row =1 */ |
| /* pu1_src[col] */ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); |
| |
| /* storing four 8-bit output values */ |
| _mm_maskmoveu_si128(s3, mask_4x32b, (char *)(pu1_dst + dst_strd + offset)); /* pu1_dst[col] = pu1_src[col]; */ |
| |
| offset += 4; /* To pointer update */ |
| } /* inner for loop ends here(4-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer increment */ |
| pu1_dst += 2 * dst_strd; /* pointer increment */ |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Chroma interprediction filter for horizontal input |
| * |
| * @par Description: |
| * Applies a horizontal filter with coefficients pointed to by 'pi1_coeff' |
| * to the elements pointed by 'pu1_src' and writes to the location pointed |
| * by 'pu1_dst' The output is downshifted by 6 and clipped to 8 bits |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_horz_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, control_mask_1_8x16b, control_mask_2_8x16b, offset_8x16b, mask_low_32b, mask_high_96b; |
| __m128i src_temp1_16x8b, src_temp2_16x8b, src_temp3_16x8b, src_temp4_16x8b, src_temp5_16x8b, src_temp6_16x8b; |
| __m128i src_temp11_16x8b, src_temp12_16x8b, src_temp13_16x8b, src_temp14_16x8b, src_temp15_16x8b, src_temp16_16x8b; |
| __m128i res_temp1_8x16b, res_temp2_8x16b, res_temp3_8x16b, res_temp4_8x16b, res_temp5_8x16b, res_temp6_8x16b, res_temp7_8x16b; |
| __m128i res_temp11_8x16b, res_temp12_8x16b, res_temp13_8x16b, res_temp14_8x16b, res_temp15_8x16b, res_temp16_8x16b, res_temp17_8x16b; |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| ASSERT(wd % 2 == 0); /* checking assumption*/ |
| |
| /* loading four 8-bit coefficients */ |
| src_temp1_16x8b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| offset_8x16b = _mm_set1_epi16(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| mask_low_32b = _mm_cmpeq_epi16(offset_8x16b, offset_8x16b); |
| mask_high_96b = _mm_srli_si128(mask_low_32b, 12); |
| mask_low_32b = _mm_slli_si128(mask_low_32b, 4); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| /* outer for loop starts from here */ |
| if(wd % 2 == 0 && wd % 4 != 0) |
| { |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| |
| /*load 16 pixel values from row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*load 16 pixel values from row 1*/ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + src_strd + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| /*Derive the source pixels for processing the 2nd pixel*/ |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); |
| |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel*/ |
| src_temp13_16x8b = _mm_srli_si128(src_temp11_16x8b, 4); |
| /*Derive the source pixels for processing the 4th pixel*/ |
| src_temp14_16x8b = _mm_srli_si128(src_temp11_16x8b, 6); |
| |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp13_16x8b, src_temp14_16x8b); |
| |
| res_temp1_8x16b = _mm_unpacklo_epi64(src_temp5_16x8b, src_temp15_16x8b); |
| res_temp2_8x16b = _mm_unpacklo_epi64(src_temp6_16x8b, src_temp16_16x8b); |
| res_temp11_8x16b = _mm_maddubs_epi16(res_temp1_8x16b, coeff0_1_8x16b); |
| res_temp12_8x16b = _mm_maddubs_epi16(res_temp2_8x16b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp13_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| |
| res_temp14_8x16b = _mm_adds_epi16(res_temp13_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp15_8x16b = _mm_srai_epi16(res_temp14_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp13_8x16b = _mm_packus_epi16(res_temp15_8x16b, res_temp15_8x16b); /* row = 0 */ |
| |
| res_temp3_8x16b = _mm_srli_si128(res_temp13_8x16b, 4); |
| |
| res_temp4_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + offset)); |
| res_temp5_8x16b = _mm_and_si128(res_temp4_8x16b, mask_low_32b); |
| res_temp6_8x16b = _mm_and_si128(res_temp13_8x16b, mask_high_96b); |
| res_temp7_8x16b = _mm_or_si128(res_temp5_8x16b, res_temp6_8x16b); |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp7_8x16b); /* pu1_dst[col] = i2_tmp_u */ |
| |
| res_temp14_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd + offset)); |
| res_temp15_8x16b = _mm_and_si128(res_temp14_8x16b, mask_low_32b); |
| res_temp16_8x16b = _mm_and_si128(res_temp3_8x16b, mask_high_96b); |
| res_temp17_8x16b = _mm_or_si128(res_temp15_8x16b, res_temp16_8x16b); |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), res_temp17_8x16b); /* pu1_dst[col] = i2_tmp_u */ |
| |
| |
| offset += 4; /* To pointer update*/ |
| |
| } /* inner loop ends here(8- output values in single iteration)*/ |
| |
| pu1_src += 2 * src_strd; /*pointer update*/ |
| pu1_dst += 2 * dst_strd; /*pointer update*/ |
| } |
| } |
| else |
| { |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| /*load 16 pixel values from row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*load 16 pixel values from row 1*/ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + src_strd + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff0_1_8x16b); |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp3_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| |
| res_temp4_8x16b = _mm_adds_epi16(res_temp3_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp5_8x16b = _mm_srai_epi16(res_temp4_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp6_8x16b = _mm_packus_epi16(res_temp5_8x16b, res_temp5_8x16b); /* row = 0 */ |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + offset), res_temp6_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 1*/ |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); |
| |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 1*/ |
| src_temp13_16x8b = _mm_srli_si128(src_temp11_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 1*/ |
| src_temp14_16x8b = _mm_srli_si128(src_temp11_16x8b, 6); |
| |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp13_16x8b, src_temp14_16x8b); |
| |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff0_1_8x16b); |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp13_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| |
| res_temp14_8x16b = _mm_adds_epi16(res_temp13_8x16b, offset_8x16b); /* row = 0 */ |
| res_temp15_8x16b = _mm_srai_epi16(res_temp14_8x16b, SHIFT_14_MINUS_BIT_DEPTH); /* row = 0 */ |
| res_temp16_8x16b = _mm_packus_epi16(res_temp15_8x16b, res_temp15_8x16b); /* row = 0 */ |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd + offset), res_temp16_8x16b); /* pu1_dst[col] = i2_tmp_u */ |
| |
| |
| offset += 8; /* To pointer update*/ |
| |
| } /* inner loop ends here(8- output values in single iteration)*/ |
| |
| pu1_src += 2 * src_strd; /*pointer update*/ |
| pu1_dst += 2 * dst_strd; /*pointer update*/ |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Chroma interprediction filter for vertical input |
| * |
| * @par Description: |
| * Applies a vertcal filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' The output is downshifted by 6 and clipped to 8 bits |
| * |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_vert_ssse3(UWORD8 *pu1_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| UWORD8 *pu1_src_copy; |
| UWORD8 *pu1_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b; |
| __m128i s4_8x16b, s5_8x16b, s6_8x16b, s7_8x16b, s8_8x16b, s9_8x16b; |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b; |
| __m128i s11_8x16b, s12_8x16b, s15_8x16b, s16_8x16b; |
| __m128i zero_8x16b, offset_8x16b, mask_low_32b, mask_high_96b; |
| __m128i s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b; |
| __m128i s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b; |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| |
| /* seting values in register */ |
| zero_8x16b = _mm_setzero_si128(); /* for saturated clipping */ |
| offset_8x16b = _mm_set1_epi16(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| mask_low_32b = _mm_set_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000); |
| mask_high_96b = _mm_set_epi32(0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF); |
| |
| /* outer for loop starts from here */ |
| if(wd % 8 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pu1_src_copy = pu1_src; |
| pu1_dst_copy = pu1_dst; |
| |
| for(col = 0; col < 2 * wd; col += 16) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pu1_dst = pu1_dst_copy + col; |
| |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values from -751:-768 pos. relative to cur. pos.*/ |
| s21_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values from -495:-512 pos. relative to cur. pos.*/ |
| s22_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| |
| /*load 8 pixel values from -239:-256 pos. relative to cur. pos.*/ |
| s23_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s24_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi8(s21_8x16b, s22_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi8(s23_8x16b, s24_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_maddubs_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_maddubs_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s34_8x16b); |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| s31_8x16b = _mm_add_epi16(s35_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi16(s31_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| s33_8x16b = _mm_packus_epi16(s32_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi64(s7_8x16b, s33_8x16b); |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storeu_si128((__m128i *)(pu1_dst), s7_8x16b); |
| |
| |
| s25_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi8(s22_8x16b, s23_8x16b); |
| |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_maddubs_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi8(s24_8x16b, s25_8x16b); |
| |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_maddubs_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s34_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| s31_8x16b = _mm_add_epi16(s35_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi16(s31_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| s33_8x16b = _mm_packus_epi16(s32_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi64(s7_8x16b, s33_8x16b); |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), s7_8x16b); |
| |
| pu1_src += 2 * src_strd; |
| pu1_dst += 2 * dst_strd; |
| |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| } |
| } |
| else if(wd % 4 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| pu1_src_copy = pu1_src; |
| pu1_dst_copy = pu1_dst; |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values from -751:-768 pos. relative to cur. pos.*/ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values from -495:-512 pos. relative to cur. pos.*/ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values from -239:-256 pos. relative to cur. pos.*/ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s7_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s7_8x16b); |
| |
| pu1_src += 8; /* To pointer update */ |
| pu1_dst += 8; |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src = pu1_src_copy + 2 * src_strd; /* pointer update */ |
| pu1_dst = pu1_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| else |
| { /* wd = multiple of 4 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| pu1_src_copy = pu1_src; |
| pu1_dst_copy = pu1_dst; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values from -751:-768 pos. relative to cur. pos.*/ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values from -495:-512 pos. relative to cur. pos.*/ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values from -239:-256 pos. relative to cur. pos.*/ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| s9_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s5_8x16b = _mm_and_si128(s9_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s7_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s5_8x16b = _mm_add_epi16(s8_8x16b, offset_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi16(s5_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s6_8x16b, zero_8x16b); |
| |
| s9_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd)); |
| s5_8x16b = _mm_and_si128(s9_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s7_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s9_8x16b); |
| |
| pu1_src += 4; /* To pointer update */ |
| pu1_dst += 4; |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src = pu1_src_copy + 2 * src_strd; /* pointer update */ |
| pu1_dst = pu1_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * chroma interprediction filter for copying 16bit output |
| * |
| * @par Description: |
| * Copies the array of width 'wd' and height 'ht' from the location pointed |
| * by 'src' to the location pointed by 'dst' The output is upshifted by 6 |
| * bits and is used as input for vertical filtering or weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| |
| void ihevc_inter_pred_chroma_copy_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| __m128i s3, zero_8x16b; |
| |
| ASSERT(wd % 2 == 0); /* checking assumption*/ |
| ASSERT(ht % 2 == 0); /* checking assumption*/ |
| |
| UNUSED(pi1_coeff); |
| zero_8x16b = _mm_setzero_si128(); |
| /* outer for loop starts from here */ |
| if(wd == 2) /* for wd =2 */ |
| { |
| for(row = 0; row < ht; row += 2) |
| { |
| int offset = 0; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), s3); |
| |
| /* row =1 */ |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| offset += 4; /* To pointer update */ |
| } /* inner for loop ends here */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pi2_dst += 2 * dst_strd; /* pointer update */ |
| } |
| } |
| else if(wd % 2 == 0 && wd % 4 != 0) |
| { |
| for(row = 0; row < ht / 2; row++) |
| { |
| int offset = 0; |
| int count = (2 * wd) / 8; |
| for(col = 0; col < count; col++) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col]*/ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + offset), s3); |
| |
| /*row=1*/ /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| |
| offset += 8; /* To pointer update*/ |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| /* finding last four values */ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col] */ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), s3); |
| |
| /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pi2_dst += 2 * dst_strd; |
| } |
| } |
| else |
| { |
| for(row = 0; row < ht / 2; row++) |
| { |
| int offset = 0; |
| for(col = 0; col < 2 * wd / 8; col++) |
| { |
| /* row =0 */ |
| /*load 16 pixel values from 15:0 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + offset)); /* pu1_src[col]*/ |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* pi2_dst[col] = (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH); */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + offset), s3); |
| |
| /*row=1*/ /*load 16 pixel values from 271:256 pos. relative to cur. pos.*/ |
| s3 = _mm_loadu_si128((__m128i *)(pu1_src + src_strd + offset)); |
| s3 = _mm_unpacklo_epi8(s3, zero_8x16b); |
| |
| s3 = _mm_slli_epi16(s3, SHIFT_14_MINUS_BIT_DEPTH); /* (pu1_src[col] << SHIFT_14_MINUS_BIT_DEPTH) */ |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd + offset), s3); |
| |
| offset += 8; /* To pointer update*/ |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src += 2 * src_strd; /* pointer update */ |
| pi2_dst += 2 * dst_strd; |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * chroma interprediction filter to store horizontal 16bit ouput |
| * |
| * @par Description: |
| * Applies a horizontal filter with coefficients pointed to by 'pi1_coeff' |
| * to the elements pointed by 'pu1_src' and writes to the location pointed |
| * by 'pu1_dst' No downshifting or clipping is done and the output is used |
| * as an input for vertical filtering or weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_horz_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b, control_mask_1_8x16b, control_mask_2_8x16b, all_zero; |
| __m128i src_temp1_16x8b, src_temp2_16x8b, src_temp3_16x8b, src_temp4_16x8b, src_temp5_16x8b, src_temp6_16x8b; |
| __m128i src_temp11_16x8b, src_temp12_16x8b, src_temp13_16x8b, src_temp14_16x8b, src_temp15_16x8b, src_temp16_16x8b; |
| __m128i res_temp1_8x16b, res_temp2_8x16b, res_temp3_8x16b; |
| __m128i res_temp11_8x16b, res_temp12_8x16b, res_temp13_8x16b; |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| ASSERT(wd % 2 == 0); /* checking assumption*/ |
| |
| /* loading four 8-bit coefficients and convert 8-bit into 16-bit */ |
| src_temp1_16x8b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| all_zero = _mm_setzero_si128(); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(src_temp1_16x8b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| /* outer for loop starts from here */ |
| if(wd % 2 == 0 && wd % 4 != 0) |
| { |
| int offset = 0; |
| for(row = ht; row >= 2; row -= 2) |
| { |
| offset = 0; |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| /*load 16 pixel values of row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*load 16 pixel values of row 1*/ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + src_strd + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 0*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 0*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 0*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 1*/ |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); |
| |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 1*/ |
| src_temp13_16x8b = _mm_srli_si128(src_temp11_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 1*/ |
| src_temp14_16x8b = _mm_srli_si128(src_temp11_16x8b, 6); |
| |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp13_16x8b, src_temp14_16x8b); |
| |
| res_temp1_8x16b = _mm_unpacklo_epi64(src_temp5_16x8b, src_temp15_16x8b); |
| res_temp2_8x16b = _mm_unpacklo_epi64(src_temp6_16x8b, src_temp16_16x8b); |
| res_temp11_8x16b = _mm_maddubs_epi16(res_temp1_8x16b, coeff0_1_8x16b); |
| res_temp12_8x16b = _mm_maddubs_epi16(res_temp2_8x16b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp13_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| |
| res_temp3_8x16b = _mm_srli_si128(res_temp13_8x16b, 8); |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), res_temp13_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd + offset), res_temp3_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| |
| offset += 4; /* To pointer update*/ |
| |
| } /* inner loop ends here(8- output values in single iteration)*/ |
| |
| pu1_src += 2 * src_strd; /*pointer update*/ |
| pi2_dst += 2 * dst_strd; /*pointer update*/ |
| } |
| |
| /*Epilogue to handle ht= odd case*/ |
| if(row) |
| { |
| offset = 0; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| /*load 16 pixel values of row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 0*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 0*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 0*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| res_temp1_8x16b = _mm_unpacklo_epi64(src_temp5_16x8b, all_zero); |
| res_temp2_8x16b = _mm_unpacklo_epi64(src_temp6_16x8b, all_zero); |
| res_temp11_8x16b = _mm_maddubs_epi16(res_temp1_8x16b, coeff0_1_8x16b); |
| res_temp12_8x16b = _mm_maddubs_epi16(res_temp2_8x16b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp13_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| |
| //res_temp3_8x16b = _mm_srli_si128 (res_temp13_8x16b, 8); |
| |
| /* store 4 16-bit values */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + offset), res_temp13_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| offset += 4; /* To pointer update*/ |
| |
| } |
| } |
| |
| } |
| else |
| { |
| int offset = 0; |
| |
| for(row = ht; row >= 2; row -= 2) |
| { |
| offset = 0; |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| /*load 16 pixel values of row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*load 16 pixel values of row 1*/ |
| src_temp11_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + src_strd + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 0*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 0*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 0*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff0_1_8x16b); |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp3_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| |
| /* store 8 16-bit values */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + offset), res_temp3_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 1*/ |
| src_temp12_16x8b = _mm_srli_si128(src_temp11_16x8b, 2); |
| |
| src_temp15_16x8b = _mm_unpacklo_epi8(src_temp11_16x8b, src_temp12_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 1*/ |
| src_temp13_16x8b = _mm_srli_si128(src_temp11_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 1*/ |
| src_temp14_16x8b = _mm_srli_si128(src_temp11_16x8b, 6); |
| |
| src_temp16_16x8b = _mm_unpacklo_epi8(src_temp13_16x8b, src_temp14_16x8b); |
| |
| res_temp11_8x16b = _mm_maddubs_epi16(src_temp15_16x8b, coeff0_1_8x16b); |
| res_temp12_8x16b = _mm_maddubs_epi16(src_temp16_16x8b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp13_8x16b = _mm_add_epi16(res_temp11_8x16b, res_temp12_8x16b); |
| |
| /* store 8 16-bit values */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + dst_strd + offset), res_temp13_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| |
| offset += 8; /* To pointer update*/ |
| |
| } /* inner loop ends here(8- output values in single iteration)*/ |
| |
| pu1_src += 2 * src_strd; /*pointer update*/ |
| pi2_dst += 2 * dst_strd; /*pointer update*/ |
| } |
| |
| /*Epilogue to take care of odd ht*/ |
| if(row) |
| { |
| offset = 0; |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| /*load 16 pixel values of row 0*/ |
| src_temp1_16x8b = _mm_loadu_si128((__m128i *)(pu1_src - 2 + offset)); /* pu1_src[col + (i-1) * 2]*/ |
| |
| /*Derive the source pixels for processing the 2nd pixel of row 0*/ |
| src_temp2_16x8b = _mm_srli_si128(src_temp1_16x8b, 2); |
| |
| src_temp5_16x8b = _mm_unpacklo_epi8(src_temp1_16x8b, src_temp2_16x8b); |
| |
| /*Derive the source pixels for processing the 3rd pixel of row 0*/ |
| src_temp3_16x8b = _mm_srli_si128(src_temp1_16x8b, 4); |
| |
| /*Derive the source pixels for processing the 4th pixel of row 0*/ |
| src_temp4_16x8b = _mm_srli_si128(src_temp1_16x8b, 6); |
| |
| src_temp6_16x8b = _mm_unpacklo_epi8(src_temp3_16x8b, src_temp4_16x8b); |
| |
| res_temp1_8x16b = _mm_maddubs_epi16(src_temp5_16x8b, coeff0_1_8x16b); |
| res_temp2_8x16b = _mm_maddubs_epi16(src_temp6_16x8b, coeff2_3_8x16b); |
| |
| /* i4_tmp += pi1_coeff[i] * pi2_src[col + (i-1) * 2] */ |
| res_temp3_8x16b = _mm_add_epi16(res_temp1_8x16b, res_temp2_8x16b); |
| |
| /* store 8 16-bit values */ |
| _mm_storeu_si128((__m128i *)(pi2_dst + offset), res_temp3_8x16b); /* pi2_dst[col] = i2_tmp_u */ |
| |
| offset += 8; /* To pointer update*/ |
| |
| } |
| } |
| |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * Interprediction chroma filter to store vertical 16bit ouput |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' No downshifting or clipping is done and the output is used as |
| * an input for weighted prediction |
| * |
| * @param[in] pu1_src |
| * UWORD8 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_vert_w16out_ssse3(UWORD8 *pu1_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| UWORD8 *pu1_src_copy; |
| WORD16 *pi2_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b; |
| __m128i s4_8x16b, s5_8x16b, s6_8x16b, s8_8x16b; |
| __m128i control_mask_1_8x16b, control_mask_2_8x16b; |
| __m128i s11_8x16b, s12_8x16b, s15_8x16b, s16_8x16b; |
| __m128i s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b; |
| __m128i s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b; |
| |
| |
| PREFETCH((char const *)(pu1_src + (0 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (1 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (2 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (3 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (4 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (5 * src_strd)), _MM_HINT_T0) |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| control_mask_1_8x16b = _mm_set1_epi32(0x01000100); /* Control Mask register */ |
| control_mask_2_8x16b = _mm_set1_epi32(0x03020302); /* Control Mask register */ |
| |
| coeff0_1_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_1_8x16b); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi8(s4_8x16b, control_mask_2_8x16b); /* pi1_coeff[4] */ |
| |
| |
| |
| /* outer for loop starts from here */ |
| if(wd % 8 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pu1_src_copy = pu1_src; |
| pi2_dst_copy = pi2_dst; |
| |
| for(col = 0; col < 2 * wd; col += 16) |
| { |
| |
| pu1_src = pu1_src_copy + col; |
| pi2_dst = pi2_dst_copy + col; |
| |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 16 pixel values */ |
| s21_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s22_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| |
| /*load 16 pixel values */ |
| s23_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s24_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi8(s21_8x16b, s22_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi8(s23_8x16b, s24_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_maddubs_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_maddubs_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s34_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storeu_si128((__m128i *)(pi2_dst), s8_8x16b); |
| |
| _mm_storeu_si128((__m128i *)(pi2_dst + 8), s35_8x16b); |
| |
| |
| s25_8x16b = _mm_loadu_si128((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi8(s22_8x16b, s23_8x16b); |
| |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_maddubs_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi8(s24_8x16b, s25_8x16b); |
| |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_maddubs_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi16(s32_8x16b, s34_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd), s8_8x16b); |
| |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd + 8), s35_8x16b); |
| |
| |
| pu1_src += 2 * src_strd; |
| pi2_dst += 2 * dst_strd; |
| |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| } |
| } |
| |
| else if(wd % 4 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| pu1_src_copy = pu1_src; |
| pi2_dst_copy = pi2_dst; |
| |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values */ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| _mm_storeu_si128((__m128i *)(pi2_dst), s8_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd), s8_8x16b); |
| |
| pu1_src += 8; /* To pointer update */ |
| pi2_dst += 8; |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src = pu1_src_copy + 2 * src_strd; /* pointer update */ |
| pi2_dst = pi2_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| else |
| { /* wd = multiple of 4 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| pu1_src_copy = pu1_src; |
| pi2_dst_copy = pi2_dst; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| PREFETCH((char const *)(pu1_src + (6 * src_strd)), _MM_HINT_T0) |
| PREFETCH((char const *)(pu1_src + (7 * src_strd)), _MM_HINT_T0) |
| |
| |
| /*load 8 pixel values */ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s8_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pu1_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi8(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_maddubs_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi8(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_maddubs_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi16(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd), s8_8x16b); |
| |
| pu1_src += 4; /* To pointer update */ |
| pi2_dst += 4; |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pu1_src = pu1_src_copy + 2 * src_strd; /* pointer update */ |
| pi2_dst = pi2_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * chroma interprediction filter for vertical 16bit input |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' Input is 16 bits The filter output is downshifted by 12 and |
| * clipped to lie between 0 and 255 |
| * |
| * @param[in] pi2_src |
| * WORD16 pointer to the source |
| * |
| * @param[out] pu1_dst |
| * UWORD8 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_vert_w16inp_ssse3(WORD16 *pi2_src, |
| UWORD8 *pu1_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| WORD16 *pi2_src_copy; |
| UWORD8 *pu1_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b; |
| __m128i s4_8x16b, s5_8x16b, s6_8x16b, s7_8x16b, s8_8x16b, s9_8x16b; |
| __m128i s11_8x16b, s12_8x16b, s15_8x16b, s16_8x16b; |
| __m128i zero_8x16b, offset_8x16b, mask_low_32b, mask_high_96b, sign_reg; |
| __m128i s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b; |
| __m128i s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b; |
| |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| zero_8x16b = _mm_setzero_si128(); |
| sign_reg = _mm_cmpgt_epi8(zero_8x16b, s4_8x16b); |
| s5_8x16b = _mm_unpacklo_epi8(s4_8x16b, sign_reg); |
| |
| coeff0_1_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(0, 0, 0, 0)); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(1, 1, 1, 1)); /* pi1_coeff[4] */ |
| |
| /* seting values in register */ |
| offset_8x16b = _mm_set1_epi32(OFFSET_14_MINUS_BIT_DEPTH); /* for offset addition */ |
| mask_low_32b = _mm_set_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000); |
| mask_high_96b = _mm_set_epi32(0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF); |
| |
| /* outer for loop starts from here */ |
| if(wd % 4 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pi2_src_copy = pi2_src; |
| pu1_dst_copy = pu1_dst; |
| |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| pi2_src = pi2_src_copy + col; |
| pu1_dst = pu1_dst_copy + col; |
| |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| /*load 16 pixel values */ |
| s21_8x16b = _mm_load_si128((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s22_8x16b = _mm_load_si128((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| |
| /*load 16 pixel values */ |
| s23_8x16b = _mm_load_si128((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s24_8x16b = _mm_load_si128((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s21_8x16b, s22_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi16(s21_8x16b, s22_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s23_8x16b, s24_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi16(s23_8x16b, s24_8x16b); |
| |
| s11_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_madd_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s12_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_madd_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s34_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi32(s35_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s7_8x16b = _mm_add_epi32(s6_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s33_8x16b = _mm_add_epi32(s32_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s34_8x16b = _mm_srai_epi32(s33_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s35_8x16b = _mm_packs_epi32(s34_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s9_8x16b, zero_8x16b); |
| |
| s33_8x16b = _mm_packus_epi16(s35_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi32(s7_8x16b, s33_8x16b); |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s7_8x16b); |
| |
| |
| s25_8x16b = _mm_load_si128((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s22_8x16b, s23_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi16(s22_8x16b, s23_8x16b); |
| |
| s15_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_madd_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s24_8x16b, s25_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi16(s24_8x16b, s25_8x16b); |
| |
| s16_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_madd_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s34_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi32(s35_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s7_8x16b = _mm_add_epi32(s6_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s33_8x16b = _mm_add_epi32(s32_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s34_8x16b = _mm_srai_epi32(s33_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s35_8x16b = _mm_packs_epi32(s34_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s9_8x16b, zero_8x16b); |
| |
| s33_8x16b = _mm_packus_epi16(s35_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi32(s7_8x16b, s33_8x16b); |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s7_8x16b); |
| |
| pi2_src += 2 * src_strd; |
| pu1_dst += 2 * dst_strd; |
| |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| } |
| } |
| else |
| { /* wd = multiple of 4 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| pi2_src_copy = pi2_src; |
| pu1_dst_copy = pu1_dst; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| /*load 8 pixel values */ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 8 pixel values */ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 8 pixel values */ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s7_8x16b = _mm_add_epi32(s6_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s9_8x16b, zero_8x16b); |
| |
| s9_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst)); |
| s5_8x16b = _mm_and_si128(s9_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s7_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst), s9_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| /* (i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) */ |
| s7_8x16b = _mm_add_epi32(s6_8x16b, offset_8x16b); |
| |
| /* i4_tmp = ((i4_tmp >> SHIFT_14_MINUS_BIT_DEPTH) + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s8_8x16b = _mm_srai_epi32(s7_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s8_8x16b, zero_8x16b); |
| |
| /* i2_tmp = CLIP_U8(i2_tmp);*/ |
| s7_8x16b = _mm_packus_epi16(s9_8x16b, zero_8x16b); |
| |
| s9_8x16b = _mm_loadl_epi64((__m128i *)(pu1_dst + dst_strd)); |
| s5_8x16b = _mm_and_si128(s9_8x16b, mask_low_32b); |
| s6_8x16b = _mm_and_si128(s7_8x16b, mask_high_96b); |
| s9_8x16b = _mm_or_si128(s5_8x16b, s6_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pu1_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pu1_dst + dst_strd), s9_8x16b); |
| |
| pi2_src += 4; /* To pointer update */ |
| pu1_dst += 4; |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pi2_src = pi2_src_copy + 2 * src_strd; /* pointer update */ |
| pu1_dst = pu1_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| } |
| |
| /** |
| ******************************************************************************* |
| * |
| * @brief |
| * |
| * Chroma interprediction filter for 16bit vertical input and output. |
| * |
| * @par Description: |
| * Applies a vertical filter with coefficients pointed to by 'pi1_coeff' to |
| * the elements pointed by 'pu1_src' and writes to the location pointed by |
| * 'pu1_dst' Input is 16 bits The filter output is downshifted by 6 and |
| * 8192 is subtracted to store it as a 16 bit number The output is used as |
| * a input to weighted prediction |
| * |
| * @param[in] pi2_src |
| * WORD16 pointer to the source |
| * |
| * @param[out] pi2_dst |
| * WORD16 pointer to the destination |
| * |
| * @param[in] src_strd |
| * integer source stride |
| * |
| * @param[in] dst_strd |
| * integer destination stride |
| * |
| * @param[in] pi1_coeff |
| * WORD8 pointer to the filter coefficients |
| * |
| * @param[in] ht |
| * integer height of the array |
| * |
| * @param[in] wd |
| * integer width of the array |
| * |
| * @returns |
| * |
| * @remarks |
| * None |
| * |
| ******************************************************************************* |
| */ |
| void ihevc_inter_pred_chroma_vert_w16inp_w16out_ssse3(WORD16 *pi2_src, |
| WORD16 *pi2_dst, |
| WORD32 src_strd, |
| WORD32 dst_strd, |
| WORD8 *pi1_coeff, |
| WORD32 ht, |
| WORD32 wd) |
| { |
| WORD32 row, col; |
| WORD16 *pi2_src_copy; |
| WORD16 *pi2_dst_copy; |
| __m128i coeff0_1_8x16b, coeff2_3_8x16b; |
| __m128i s4_8x16b, s5_8x16b, s6_8x16b, s7_8x16b, s8_8x16b, s9_8x16b; |
| __m128i s11_8x16b, s12_8x16b, s15_8x16b, s16_8x16b; |
| __m128i zero_8x16b, sign_reg; |
| __m128i s21_8x16b, s22_8x16b, s23_8x16b, s24_8x16b, s25_8x16b; |
| __m128i s31_8x16b, s32_8x16b, s33_8x16b, s34_8x16b, s35_8x16b; |
| |
| |
| /* load 8 8-bit coefficients and convert 8-bit into 16-bit */ |
| s4_8x16b = _mm_loadl_epi64((__m128i *)pi1_coeff); |
| |
| zero_8x16b = _mm_setzero_si128(); |
| sign_reg = _mm_cmpgt_epi8(zero_8x16b, s4_8x16b); |
| s5_8x16b = _mm_unpacklo_epi8(s4_8x16b, sign_reg); |
| |
| coeff0_1_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(0, 0, 0, 0)); /* pi1_coeff[4] */ |
| coeff2_3_8x16b = _mm_shuffle_epi32(s5_8x16b, _MM_SHUFFLE(1, 1, 1, 1)); /* pi1_coeff[4] */ |
| |
| |
| /* outer for loop starts from here */ |
| if(wd % 4 == 0) |
| { /* wd = multiple of 8 case */ |
| |
| pi2_src_copy = pi2_src; |
| pi2_dst_copy = pi2_dst; |
| |
| for(col = 0; col < 2 * wd; col += 8) |
| { |
| |
| pi2_src = pi2_src_copy + col; |
| pi2_dst = pi2_dst_copy + col; |
| |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| |
| /*load 16 pixel values */ |
| s21_8x16b = _mm_load_si128((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s22_8x16b = _mm_load_si128((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| |
| /*load 16 pixel values */ |
| s23_8x16b = _mm_load_si128((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 16 pixel values */ |
| s24_8x16b = _mm_load_si128((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s21_8x16b, s22_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi16(s21_8x16b, s22_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s23_8x16b, s24_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi16(s23_8x16b, s24_8x16b); |
| |
| s11_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_madd_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s12_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_madd_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s34_8x16b); |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi32(s35_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s6_8x16b, zero_8x16b); |
| |
| s35_8x16b = _mm_packs_epi32(s32_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi64(s9_8x16b, s35_8x16b); |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_store_si128((__m128i *)(pi2_dst), s7_8x16b); |
| |
| |
| s25_8x16b = _mm_loadu_si128((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s22_8x16b, s23_8x16b); |
| |
| s31_8x16b = _mm_unpackhi_epi16(s22_8x16b, s23_8x16b); |
| |
| s15_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s32_8x16b = _mm_madd_epi16(s31_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s24_8x16b, s25_8x16b); |
| |
| s33_8x16b = _mm_unpackhi_epi16(s24_8x16b, s25_8x16b); |
| |
| s16_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s34_8x16b = _mm_madd_epi16(s33_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| s35_8x16b = _mm_add_epi32(s32_8x16b, s34_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s32_8x16b = _mm_srai_epi32(s35_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s6_8x16b, zero_8x16b); |
| |
| s35_8x16b = _mm_packs_epi32(s32_8x16b, zero_8x16b); |
| |
| s7_8x16b = _mm_unpacklo_epi64(s9_8x16b, s35_8x16b); |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_store_si128((__m128i *)(pi2_dst + dst_strd), s7_8x16b); |
| |
| pi2_src += 2 * src_strd; |
| pi2_dst += 2 * dst_strd; |
| |
| |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| } |
| } |
| else |
| { /* wd = multiple of 4 case */ |
| |
| for(row = 0; row < ht; row += 2) |
| { |
| pi2_src_copy = pi2_src; |
| pi2_dst_copy = pi2_dst; |
| for(col = 0; col < 2 * wd; col += 4) |
| { |
| |
| /*load 4 pixel values */ |
| s21_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (-1 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s22_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (0 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s21_8x16b, s22_8x16b); |
| |
| s11_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| /*load 4 pixel values */ |
| s23_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (1 * src_strd))); |
| |
| /*load 4 pixel values */ |
| s24_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (2 * src_strd))); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s23_8x16b, s24_8x16b); |
| |
| s12_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s11_8x16b, s12_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s6_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pi2_dst), s9_8x16b); |
| |
| s25_8x16b = _mm_loadl_epi64((__m128i *)(pi2_src + (3 * src_strd))); |
| |
| s5_8x16b = _mm_unpacklo_epi16(s22_8x16b, s23_8x16b); |
| s15_8x16b = _mm_madd_epi16(s5_8x16b, coeff0_1_8x16b); |
| |
| s6_8x16b = _mm_unpacklo_epi16(s24_8x16b, s25_8x16b); |
| s16_8x16b = _mm_madd_epi16(s6_8x16b, coeff2_3_8x16b); |
| |
| s8_8x16b = _mm_add_epi32(s15_8x16b, s16_8x16b); /* (i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) */ |
| |
| /*(i2_tmp + OFFSET_14_MINUS_BIT_DEPTH) >> SHIFT_14_MINUS_BIT_DEPTH */ |
| s6_8x16b = _mm_srai_epi32(s8_8x16b, SHIFT_14_MINUS_BIT_DEPTH); |
| |
| s9_8x16b = _mm_packs_epi32(s6_8x16b, zero_8x16b); |
| |
| /* store 8 8-bit output values */ |
| /* pi2_dst[col] = (UWORD8)i2_tmp; */ |
| _mm_storel_epi64((__m128i *)(pi2_dst + dst_strd), s9_8x16b); |
| |
| pi2_src += 4; /* To pointer update */ |
| pi2_dst += 4; |
| } /* inner for loop ends here(8-output values in single iteration) */ |
| |
| pi2_src = pi2_src_copy + 2 * src_strd; /* pointer update */ |
| pi2_dst = pi2_dst_copy + 2 * dst_strd; /* pointer update */ |
| } |
| } |
| |
| } |
