ref: a5d9416fd7da3f66485642af9a6f62f686d3b70a
parent: ce4c4b96e4ad28c2ba326629dd9ffdab95d63274
author: Parag Salasakar <img.mips1@gmail.com>
date: Wed Jul 29 05:40:26 EDT 2015
mips msa vp8 post proc optimization average improvement ~2x-4x Change-Id: I93abc15389649c169bb8b69127c0b95407d34692
--- a/test/pp_filter_test.cc
+++ b/test/pp_filter_test.cc
@@ -110,4 +110,9 @@
::testing::Values(vp8_post_proc_down_and_across_mb_row_sse2));
#endif
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, VP8PostProcessingFilterTest,
+ ::testing::Values(vp8_post_proc_down_and_across_mb_row_msa));
+#endif
+
} // namespace
--- /dev/null
+++ b/vp8/common/mips/msa/postproc_msa.c
@@ -1,0 +1,851 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdlib.h>
+#include "./vp8_rtcd.h"
+#include "vp8/common/mips/msa/vp8_macros_msa.h"
+
+static const int16_t vp8_rv_msa[] =
+{+ 8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
+ 0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
+ 10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
+ 8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
+ 8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
+ 1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
+ 3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
+ 11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
+ 14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
+ 4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
+ 7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
+ 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
+ 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
+ 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
+ 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
+ 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
+ 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
+ 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
+ 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
+ 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
+ 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
+ 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
+ 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
+ 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
+ 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
+ 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
+ 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
+ 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
+ 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
+ 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
+ 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
+ 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
+ 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
+ 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
+ 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
+ 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
+ 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
+ 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
+ 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
+ 3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
+ 11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
+ 14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
+ 5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
+ 0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
+};
+
+#define VP8_TRANSPOSE8x16_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3, \
+ out4, out5, out6, out7, \
+ out8, out9, out10, out11, \
+ out12, out13, out14, out15) \
+{ \+ v8i16 temp0, temp1, temp2, temp3, temp4; \
+ v8i16 temp5, temp6, temp7, temp8, temp9; \
+ \
+ ILVR_B4_SH(in1, in0, in3, in2, in5, in4, in7, in6, \
+ temp0, temp1, temp2, temp3); \
+ ILVR_H2_SH(temp1, temp0, temp3, temp2, temp4, temp5); \
+ ILVRL_W2_SH(temp5, temp4, temp6, temp7); \
+ ILVL_H2_SH(temp1, temp0, temp3, temp2, temp4, temp5); \
+ ILVRL_W2_SH(temp5, temp4, temp8, temp9); \
+ ILVL_B4_SH(in1, in0, in3, in2, in5, in4, in7, in6, \
+ temp0, temp1, temp2, temp3); \
+ ILVR_H2_SH(temp1, temp0, temp3, temp2, temp4, temp5); \
+ ILVRL_W2_UB(temp5, temp4, out8, out10); \
+ ILVL_H2_SH(temp1, temp0, temp3, temp2, temp4, temp5); \
+ ILVRL_W2_UB(temp5, temp4, out12, out14); \
+ out0 = (v16u8)temp6; \
+ out2 = (v16u8)temp7; \
+ out4 = (v16u8)temp8; \
+ out6 = (v16u8)temp9; \
+ out9 = (v16u8)__msa_ilvl_d((v2i64)out8, (v2i64)out8); \
+ out11 = (v16u8)__msa_ilvl_d((v2i64)out10, (v2i64)out10); \
+ out13 = (v16u8)__msa_ilvl_d((v2i64)out12, (v2i64)out12); \
+ out15 = (v16u8)__msa_ilvl_d((v2i64)out14, (v2i64)out14); \
+ out1 = (v16u8)__msa_ilvl_d((v2i64)out0, (v2i64)out0); \
+ out3 = (v16u8)__msa_ilvl_d((v2i64)out2, (v2i64)out2); \
+ out5 = (v16u8)__msa_ilvl_d((v2i64)out4, (v2i64)out4); \
+ out7 = (v16u8)__msa_ilvl_d((v2i64)out6, (v2i64)out6); \
+}
+
+#define VP8_AVER_IF_RETAIN(above2_in, above1_in, src_in, \
+ below1_in, below2_in, ref, out) \
+{ \+ v16u8 temp0, temp1; \
+ \
+ temp1 = __msa_aver_u_b(above2_in, above1_in); \
+ temp0 = __msa_aver_u_b(below2_in, below1_in); \
+ temp1 = __msa_aver_u_b(temp1, temp0); \
+ out = __msa_aver_u_b(src_in, temp1); \
+ temp0 = __msa_asub_u_b(src_in, above2_in); \
+ temp1 = __msa_asub_u_b(src_in, above1_in); \
+ temp0 = (temp0 < ref); \
+ temp1 = (temp1 < ref); \
+ temp0 = temp0 & temp1; \
+ temp1 = __msa_asub_u_b(src_in, below1_in); \
+ temp1 = (temp1 < ref); \
+ temp0 = temp0 & temp1; \
+ temp1 = __msa_asub_u_b(src_in, below2_in); \
+ temp1 = (temp1 < ref); \
+ temp0 = temp0 & temp1; \
+ out = __msa_bmz_v(out, src_in, temp0); \
+}
+
+#define TRANSPOSE12x16_B(in0, in1, in2, in3, in4, in5, in6, in7, \
+ in8, in9, in10, in11, in12, in13, in14, in15) \
+{ \+ v8i16 temp0, temp1, temp2, temp3, temp4; \
+ v8i16 temp5, temp6, temp7, temp8, temp9; \
+ \
+ ILVR_B2_SH(in1, in0, in3, in2, temp0, temp1); \
+ ILVRL_H2_SH(temp1, temp0, temp2, temp3); \
+ ILVR_B2_SH(in5, in4, in7, in6, temp0, temp1); \
+ ILVRL_H2_SH(temp1, temp0, temp4, temp5); \
+ ILVRL_W2_SH(temp4, temp2, temp0, temp1); \
+ ILVRL_W2_SH(temp5, temp3, temp2, temp3); \
+ ILVR_B2_SH(in9, in8, in11, in10, temp4, temp5); \
+ ILVR_B2_SH(in9, in8, in11, in10, temp4, temp5); \
+ ILVRL_H2_SH(temp5, temp4, temp6, temp7); \
+ ILVR_B2_SH(in13, in12, in15, in14, temp4, temp5); \
+ ILVRL_H2_SH(temp5, temp4, temp8, temp9); \
+ ILVRL_W2_SH(temp8, temp6, temp4, temp5); \
+ ILVRL_W2_SH(temp9, temp7, temp6, temp7); \
+ ILVL_B2_SH(in1, in0, in3, in2, temp8, temp9); \
+ ILVR_D2_UB(temp4, temp0, temp5, temp1, in0, in2); \
+ in1 = (v16u8)__msa_ilvl_d((v2i64)temp4, (v2i64)temp0); \
+ in3 = (v16u8)__msa_ilvl_d((v2i64)temp5, (v2i64)temp1); \
+ ILVL_B2_SH(in5, in4, in7, in6, temp0, temp1); \
+ ILVR_D2_UB(temp6, temp2, temp7, temp3, in4, in6); \
+ in5 = (v16u8)__msa_ilvl_d((v2i64)temp6, (v2i64)temp2); \
+ in7 = (v16u8)__msa_ilvl_d((v2i64)temp7, (v2i64)temp3); \
+ ILVL_B4_SH(in9, in8, in11, in10, in13, in12, in15, in14, \
+ temp2, temp3, temp4, temp5); \
+ ILVR_H4_SH(temp9, temp8, temp1, temp0, temp3, temp2, temp5, temp4, \
+ temp6, temp7, temp8, temp9); \
+ ILVR_W2_SH(temp7, temp6, temp9, temp8, temp0, temp1); \
+ in8 = (v16u8)__msa_ilvr_d((v2i64)temp1, (v2i64)temp0); \
+ in9 = (v16u8)__msa_ilvl_d((v2i64)temp1, (v2i64)temp0); \
+ ILVL_W2_SH(temp7, temp6, temp9, temp8, temp2, temp3); \
+ in10 = (v16u8)__msa_ilvr_d((v2i64)temp3, (v2i64)temp2); \
+ in11 = (v16u8)__msa_ilvl_d((v2i64)temp3, (v2i64)temp2); \
+}
+
+#define VP8_TRANSPOSE12x8_UB_UB(in0, in1, in2, in3, in4, in5, \
+ in6, in7, in8, in9, in10, in11) \
+{ \+ v8i16 temp0, temp1, temp2, temp3; \
+ v8i16 temp4, temp5, temp6, temp7; \
+ \
+ ILVR_B2_SH(in1, in0, in3, in2, temp0, temp1); \
+ ILVRL_H2_SH(temp1, temp0, temp2, temp3); \
+ ILVR_B2_SH(in5, in4, in7, in6, temp0, temp1); \
+ ILVRL_H2_SH(temp1, temp0, temp4, temp5); \
+ ILVRL_W2_SH(temp4, temp2, temp0, temp1); \
+ ILVRL_W2_SH(temp5, temp3, temp2, temp3); \
+ ILVL_B2_SH(in1, in0, in3, in2, temp4, temp5); \
+ temp4 = __msa_ilvr_h(temp5, temp4); \
+ ILVL_B2_SH(in5, in4, in7, in6, temp6, temp7); \
+ temp5 = __msa_ilvr_h(temp7, temp6); \
+ ILVRL_W2_SH(temp5, temp4, temp6, temp7); \
+ in0 = (v16u8)temp0; \
+ in2 = (v16u8)temp1; \
+ in4 = (v16u8)temp2; \
+ in6 = (v16u8)temp3; \
+ in8 = (v16u8)temp6; \
+ in10 = (v16u8)temp7; \
+ in1 = (v16u8)__msa_ilvl_d((v2i64)temp0, (v2i64)temp0); \
+ in3 = (v16u8)__msa_ilvl_d((v2i64)temp1, (v2i64)temp1); \
+ in5 = (v16u8)__msa_ilvl_d((v2i64)temp2, (v2i64)temp2); \
+ in7 = (v16u8)__msa_ilvl_d((v2i64)temp3, (v2i64)temp3); \
+ in9 = (v16u8)__msa_ilvl_d((v2i64)temp6, (v2i64)temp6); \
+ in11 = (v16u8)__msa_ilvl_d((v2i64)temp7, (v2i64)temp7); \
+}
+
+static void postproc_down_across_chroma_msa(uint8_t *src_ptr, uint8_t *dst_ptr,
+ int32_t src_stride,
+ int32_t dst_stride,
+ int32_t cols, uint8_t *f)
+{+ uint8_t *p_src = src_ptr;
+ uint8_t *p_dst = dst_ptr;
+ uint8_t *f_orig = f;
+ uint8_t *p_dst_st = dst_ptr;
+ uint16_t col;
+ uint64_t out0, out1, out2, out3;
+ v16u8 above2, above1, below2, below1, src, ref, ref_temp;
+ v16u8 inter0, inter1, inter2, inter3, inter4, inter5;
+ v16u8 inter6, inter7, inter8, inter9, inter10, inter11;
+
+ for (col = (cols / 16); col--;)
+ {+ ref = LD_UB(f);
+ LD_UB2(p_src - 2 * src_stride, src_stride, above2, above1);
+ src = LD_UB(p_src);
+ LD_UB2(p_src + 1 * src_stride, src_stride, below1, below2);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter0);
+ above2 = LD_UB(p_src + 3 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter1);
+ above1 = LD_UB(p_src + 4 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter2);
+ src = LD_UB(p_src + 5 * src_stride);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src, ref, inter3);
+ below1 = LD_UB(p_src + 6 * src_stride);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1, ref, inter4);
+ below2 = LD_UB(p_src + 7 * src_stride);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter5);
+ above2 = LD_UB(p_src + 8 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter6);
+ above1 = LD_UB(p_src + 9 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter7);
+ ST_UB8(inter0, inter1, inter2, inter3, inter4, inter5, inter6, inter7,
+ p_dst, dst_stride);
+
+ p_dst += 16;
+ p_src += 16;
+ f += 16;
+ }
+
+ if (0 != (cols / 16))
+ {+ ref = LD_UB(f);
+ LD_UB2(p_src - 2 * src_stride, src_stride, above2, above1);
+ src = LD_UB(p_src);
+ LD_UB2(p_src + 1 * src_stride, src_stride, below1, below2);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter0);
+ above2 = LD_UB(p_src + 3 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter1);
+ above1 = LD_UB(p_src + 4 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter2);
+ src = LD_UB(p_src + 5 * src_stride);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src, ref, inter3);
+ below1 = LD_UB(p_src + 6 * src_stride);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1, ref, inter4);
+ below2 = LD_UB(p_src + 7 * src_stride);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter5);
+ above2 = LD_UB(p_src + 8 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter6);
+ above1 = LD_UB(p_src + 9 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter7);
+ out0 = __msa_copy_u_d((v2i64)inter0, 0);
+ out1 = __msa_copy_u_d((v2i64)inter1, 0);
+ out2 = __msa_copy_u_d((v2i64)inter2, 0);
+ out3 = __msa_copy_u_d((v2i64)inter3, 0);
+ SD4(out0, out1, out2, out3, p_dst, dst_stride);
+
+ out0 = __msa_copy_u_d((v2i64)inter4, 0);
+ out1 = __msa_copy_u_d((v2i64)inter5, 0);
+ out2 = __msa_copy_u_d((v2i64)inter6, 0);
+ out3 = __msa_copy_u_d((v2i64)inter7, 0);
+ SD4(out0, out1, out2, out3, p_dst + 4 * dst_stride, dst_stride);
+ }
+
+ f = f_orig;
+ p_dst = dst_ptr - 2;
+ LD_UB8(p_dst, dst_stride,
+ inter0, inter1, inter2, inter3, inter4, inter5, inter6, inter7);
+
+ for (col = 0; col < (cols / 8); ++col)
+ {+ ref = LD_UB(f);
+ f += 8;
+ VP8_TRANSPOSE12x8_UB_UB(inter0, inter1, inter2, inter3,
+ inter4, inter5, inter6, inter7,
+ inter8, inter9, inter10, inter11);
+ if (0 == col)
+ {+ above2 = inter2;
+ above1 = inter2;
+ }
+ else
+ {+ above2 = inter0;
+ above1 = inter1;
+ }
+ src = inter2;
+ below1 = inter3;
+ below2 = inter4;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 0);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2,
+ ref_temp, inter2);
+ above2 = inter5;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 1);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2,
+ ref_temp, inter3);
+ above1 = inter6;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 2);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1,
+ ref_temp, inter4);
+ src = inter7;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 3);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src,
+ ref_temp, inter5);
+ below1 = inter8;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 4);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1,
+ ref_temp, inter6);
+ below2 = inter9;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 5);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2,
+ ref_temp, inter7);
+ if (col == (cols / 8 - 1))
+ {+ above2 = inter9;
+ }
+ else
+ {+ above2 = inter10;
+ }
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 6);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2,
+ ref_temp, inter8);
+ if (col == (cols / 8 - 1))
+ {+ above1 = inter9;
+ }
+ else
+ {+ above1 = inter11;
+ }
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 7);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1,
+ ref_temp, inter9);
+ TRANSPOSE8x8_UB_UB(inter2, inter3, inter4, inter5, inter6, inter7,
+ inter8, inter9, inter2, inter3, inter4, inter5,
+ inter6, inter7, inter8, inter9);
+ p_dst += 8;
+ LD_UB2(p_dst, dst_stride, inter0, inter1);
+ ST8x1_UB(inter2, p_dst_st);
+ ST8x1_UB(inter3, (p_dst_st + 1 * dst_stride));
+ LD_UB2(p_dst + 2 * dst_stride, dst_stride, inter2, inter3);
+ ST8x1_UB(inter4, (p_dst_st + 2 * dst_stride));
+ ST8x1_UB(inter5, (p_dst_st + 3 * dst_stride));
+ LD_UB2(p_dst + 4 * dst_stride, dst_stride, inter4, inter5);
+ ST8x1_UB(inter6, (p_dst_st + 4 * dst_stride));
+ ST8x1_UB(inter7, (p_dst_st + 5 * dst_stride));
+ LD_UB2(p_dst + 6 * dst_stride, dst_stride, inter6, inter7);
+ ST8x1_UB(inter8, (p_dst_st + 6 * dst_stride));
+ ST8x1_UB(inter9, (p_dst_st + 7 * dst_stride));
+ p_dst_st += 8;
+ }
+}
+
+static void postproc_down_across_luma_msa(uint8_t *src_ptr, uint8_t *dst_ptr,
+ int32_t src_stride,
+ int32_t dst_stride,
+ int32_t cols, uint8_t *f)
+{+ uint8_t *p_src = src_ptr;
+ uint8_t *p_dst = dst_ptr;
+ uint8_t *p_dst_st = dst_ptr;
+ uint8_t *f_orig = f;
+ uint16_t col;
+ v16u8 above2, above1, below2, below1;
+ v16u8 src, ref, ref_temp;
+ v16u8 inter0, inter1, inter2, inter3, inter4, inter5, inter6;
+ v16u8 inter7, inter8, inter9, inter10, inter11;
+ v16u8 inter12, inter13, inter14, inter15;
+
+ for (col = (cols / 16); col--;)
+ {+ ref = LD_UB(f);
+ LD_UB2(p_src - 2 * src_stride, src_stride, above2, above1);
+ src = LD_UB(p_src);
+ LD_UB2(p_src + 1 * src_stride, src_stride, below1, below2);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter0);
+ above2 = LD_UB(p_src + 3 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter1);
+ above1 = LD_UB(p_src + 4 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter2);
+ src = LD_UB(p_src + 5 * src_stride);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src, ref, inter3);
+ below1 = LD_UB(p_src + 6 * src_stride);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1, ref, inter4);
+ below2 = LD_UB(p_src + 7 * src_stride);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter5);
+ above2 = LD_UB(p_src + 8 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter6);
+ above1 = LD_UB(p_src + 9 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter7);
+ src = LD_UB(p_src + 10 * src_stride);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src, ref, inter8);
+ below1 = LD_UB(p_src + 11 * src_stride);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1, ref, inter9);
+ below2 = LD_UB(p_src + 12 * src_stride);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter10);
+ above2 = LD_UB(p_src + 13 * src_stride);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2, ref, inter11);
+ above1 = LD_UB(p_src + 14 * src_stride);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1, ref, inter12);
+ src = LD_UB(p_src + 15 * src_stride);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src, ref, inter13);
+ below1 = LD_UB(p_src + 16 * src_stride);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1, ref, inter14);
+ below2 = LD_UB(p_src + 17 * src_stride);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2, ref, inter15);
+ ST_UB8(inter0, inter1, inter2, inter3, inter4, inter5, inter6, inter7,
+ p_dst, dst_stride);
+ ST_UB8(inter8, inter9, inter10, inter11, inter12, inter13,
+ inter14, inter15, p_dst + 8 * dst_stride, dst_stride);
+ p_src += 16;
+ p_dst += 16;
+ f += 16;
+ }
+
+ f = f_orig;
+ p_dst = dst_ptr - 2;
+ LD_UB8(p_dst, dst_stride,
+ inter0, inter1, inter2, inter3, inter4, inter5, inter6, inter7);
+ LD_UB8(p_dst + 8 * dst_stride, dst_stride,
+ inter8, inter9, inter10, inter11, inter12, inter13,
+ inter14, inter15);
+
+ for (col = 0; col < cols / 8; ++col)
+ {+ ref = LD_UB(f);
+ f += 8;
+ TRANSPOSE12x16_B(inter0, inter1, inter2, inter3, inter4, inter5,
+ inter6, inter7, inter8, inter9, inter10, inter11,
+ inter12, inter13, inter14, inter15);
+ if (0 == col)
+ {+ above2 = inter2;
+ above1 = inter2;
+ }
+ else
+ {+ above2 = inter0;
+ above1 = inter1;
+ }
+
+ src = inter2;
+ below1 = inter3;
+ below2 = inter4;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 0);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2,
+ ref_temp, inter2);
+ above2 = inter5;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 1);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2,
+ ref_temp, inter3);
+ above1 = inter6;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 2);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1,
+ ref_temp, inter4);
+ src = inter7;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 3);
+ VP8_AVER_IF_RETAIN(below1, below2, above2, above1, src,
+ ref_temp, inter5);
+ below1 = inter8;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 4);
+ VP8_AVER_IF_RETAIN(below2, above2, above1, src, below1,
+ ref_temp, inter6);
+ below2 = inter9;
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 5);
+ VP8_AVER_IF_RETAIN(above2, above1, src, below1, below2,
+ ref_temp, inter7);
+ if (col == (cols / 8 - 1))
+ {+ above2 = inter9;
+ }
+ else
+ {+ above2 = inter10;
+ }
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 6);
+ VP8_AVER_IF_RETAIN(above1, src, below1, below2, above2,
+ ref_temp, inter8);
+ if (col == (cols / 8 - 1))
+ {+ above1 = inter9;
+ }
+ else
+ {+ above1 = inter11;
+ }
+ ref_temp = (v16u8)__msa_splati_b((v16i8)ref, 7);
+ VP8_AVER_IF_RETAIN(src, below1, below2, above2, above1,
+ ref_temp, inter9);
+ VP8_TRANSPOSE8x16_UB_UB(inter2, inter3, inter4, inter5,
+ inter6, inter7, inter8, inter9,
+ inter2, inter3, inter4, inter5,
+ inter6, inter7, inter8, inter9,
+ inter10, inter11, inter12, inter13,
+ inter14, inter15, above2, above1);
+
+ p_dst += 8;
+ LD_UB2(p_dst, dst_stride, inter0, inter1);
+ ST8x1_UB(inter2, p_dst_st);
+ ST8x1_UB(inter3, (p_dst_st + 1 * dst_stride));
+ LD_UB2(p_dst + 2 * dst_stride, dst_stride, inter2, inter3);
+ ST8x1_UB(inter4, (p_dst_st + 2 * dst_stride));
+ ST8x1_UB(inter5, (p_dst_st + 3 * dst_stride));
+ LD_UB2(p_dst + 4 * dst_stride, dst_stride, inter4, inter5);
+ ST8x1_UB(inter6, (p_dst_st + 4 * dst_stride));
+ ST8x1_UB(inter7, (p_dst_st + 5 * dst_stride));
+ LD_UB2(p_dst + 6 * dst_stride, dst_stride, inter6, inter7);
+ ST8x1_UB(inter8, (p_dst_st + 6 * dst_stride));
+ ST8x1_UB(inter9, (p_dst_st + 7 * dst_stride));
+ LD_UB2(p_dst + 8 * dst_stride, dst_stride, inter8, inter9);
+ ST8x1_UB(inter10, (p_dst_st + 8 * dst_stride));
+ ST8x1_UB(inter11, (p_dst_st + 9 * dst_stride));
+ LD_UB2(p_dst + 10 * dst_stride, dst_stride, inter10, inter11);
+ ST8x1_UB(inter12, (p_dst_st + 10 * dst_stride));
+ ST8x1_UB(inter13, (p_dst_st + 11 * dst_stride));
+ LD_UB2(p_dst + 12 * dst_stride, dst_stride, inter12, inter13);
+ ST8x1_UB(inter14, (p_dst_st + 12 * dst_stride));
+ ST8x1_UB(inter15, (p_dst_st + 13 * dst_stride));
+ LD_UB2(p_dst + 14 * dst_stride, dst_stride, inter14, inter15);
+ ST8x1_UB(above2, (p_dst_st + 14 * dst_stride));
+ ST8x1_UB(above1, (p_dst_st + 15 * dst_stride));
+ p_dst_st += 8;
+ }
+}
+
+void vp8_post_proc_down_and_across_mb_row_msa(uint8_t *src, uint8_t *dst,
+ int32_t src_stride,
+ int32_t dst_stride,
+ int32_t cols, uint8_t *f,
+ int32_t size)
+{+ if (8 == size)
+ {+ postproc_down_across_chroma_msa(src, dst, src_stride, dst_stride,
+ cols, f);
+ }
+ else if (16 == size)
+ {+ postproc_down_across_luma_msa(src, dst, src_stride, dst_stride,
+ cols, f);
+ }
+}
+
+void vp8_mbpost_proc_across_ip_msa(uint8_t *src_ptr, int32_t pitch,
+ int32_t rows, int32_t cols, int32_t flimit)
+{+ int32_t row, col, cnt;
+ uint8_t *src_dup = src_ptr;
+ v16u8 src0, src, tmp_orig;
+ v16u8 tmp = { 0 };+ v16i8 zero = { 0 };+ v8u16 sum_h, src_r_h, src_l_h;
+ v4u32 src_r_w, src_l_w;
+ v4i32 flimit_vec;
+
+ flimit_vec = __msa_fill_w(flimit);
+ for (row = rows; row--;)
+ {+ int32_t sum_sq = 0;
+ int32_t sum = 0;
+ src0 = (v16u8)__msa_fill_b(src_dup[0]);
+ ST8x1_UB(src0, (src_dup - 8));
+
+ src0 = (v16u8)__msa_fill_b(src_dup[cols - 1]);
+ ST_UB(src0, src_dup + cols);
+ src_dup[cols + 16] = src_dup[cols - 1];
+ tmp_orig = (v16u8)__msa_ldi_b(0);
+ tmp_orig[15] = tmp[15];
+ src = LD_UB(src_dup - 8);
+ src[15] = 0;
+ ILVRL_B2_UH(zero, src, src_r_h, src_l_h);
+ src_r_w = __msa_dotp_u_w(src_r_h, src_r_h);
+ src_l_w = __msa_dotp_u_w(src_l_h, src_l_h);
+ sum_sq = HADD_SW_S32(src_r_w);
+ sum_sq += HADD_SW_S32(src_l_w);
+ sum_h = __msa_hadd_u_h(src, src);
+ sum = HADD_UH_U32(sum_h);
+ {+ v16u8 src7, src8, src_r, src_l;
+ v16i8 mask;
+ v8u16 add_r, add_l;
+ v8i16 sub_r, sub_l, sum_r, sum_l, mask0, mask1;
+ v4i32 sum_sq0, sum_sq1, sum_sq2, sum_sq3;
+ v4i32 sub0, sub1, sub2, sub3;
+ v4i32 sum0_w, sum1_w, sum2_w, sum3_w;
+ v4i32 mul0, mul1, mul2, mul3;
+ v4i32 total0, total1, total2, total3;
+ v8i16 const8 = __msa_fill_h(8);
+
+ src7 = LD_UB(src_dup + 7);
+ src8 = LD_UB(src_dup - 8);
+ for (col = 0; col < (cols >> 4); ++col)
+ {+ ILVRL_B2_UB(src7, src8, src_r, src_l);
+ HSUB_UB2_SH(src_r, src_l, sub_r, sub_l);
+
+ sum_r[0] = sum + sub_r[0];
+ for (cnt = 0; cnt < 7; ++cnt)
+ {+ sum_r[cnt + 1] = sum_r[cnt] + sub_r[cnt + 1];
+ }
+ sum_l[0] = sum_r[7] + sub_l[0];
+ for (cnt = 0; cnt < 7; ++cnt)
+ {+ sum_l[cnt + 1] = sum_l[cnt] + sub_l[cnt + 1];
+ }
+ sum = sum_l[7];
+ src = LD_UB(src_dup + 16 * col);
+ ILVRL_B2_UH(zero, src, src_r_h, src_l_h);
+ src7 = (v16u8)((const8 + sum_r + (v8i16)src_r_h) >> 4);
+ src8 = (v16u8)((const8 + sum_l + (v8i16)src_l_h) >> 4);
+ tmp = (v16u8)__msa_pckev_b((v16i8)src8, (v16i8)src7);
+
+ HADD_UB2_UH(src_r, src_l, add_r, add_l);
+ UNPCK_SH_SW(sub_r, sub0, sub1);
+ UNPCK_SH_SW(sub_l, sub2, sub3);
+ ILVR_H2_SW(zero, add_r, zero, add_l, sum0_w, sum2_w);
+ ILVL_H2_SW(zero, add_r, zero, add_l, sum1_w, sum3_w);
+ MUL4(sum0_w, sub0, sum1_w, sub1, sum2_w, sub2, sum3_w, sub3,
+ mul0, mul1, mul2, mul3);
+ sum_sq0[0] = sum_sq + mul0[0];
+ for (cnt = 0; cnt < 3; ++cnt)
+ {+ sum_sq0[cnt + 1] = sum_sq0[cnt] + mul0[cnt + 1];
+ }
+ sum_sq1[0] = sum_sq0[3] + mul1[0];
+ for (cnt = 0; cnt < 3; ++cnt)
+ {+ sum_sq1[cnt + 1] = sum_sq1[cnt] + mul1[cnt + 1];
+ }
+ sum_sq2[0] = sum_sq1[3] + mul2[0];
+ for (cnt = 0; cnt < 3; ++cnt)
+ {+ sum_sq2[cnt + 1] = sum_sq2[cnt] + mul2[cnt + 1];
+ }
+ sum_sq3[0] = sum_sq2[3] + mul3[0];
+ for (cnt = 0; cnt < 3; ++cnt)
+ {+ sum_sq3[cnt + 1] = sum_sq3[cnt] + mul3[cnt + 1];
+ }
+ sum_sq = sum_sq3[3];
+
+ UNPCK_SH_SW(sum_r, sum0_w, sum1_w);
+ UNPCK_SH_SW(sum_l, sum2_w, sum3_w);
+ total0 = sum_sq0 * __msa_ldi_w(15);
+ total0 -= sum0_w * sum0_w;
+ total1 = sum_sq1 * __msa_ldi_w(15);
+ total1 -= sum1_w * sum1_w;
+ total2 = sum_sq2 * __msa_ldi_w(15);
+ total2 -= sum2_w * sum2_w;
+ total3 = sum_sq3 * __msa_ldi_w(15);
+ total3 -= sum3_w * sum3_w;
+ total0 = (total0 < flimit_vec);
+ total1 = (total1 < flimit_vec);
+ total2 = (total2 < flimit_vec);
+ total3 = (total3 < flimit_vec);
+ PCKEV_H2_SH(total1, total0, total3, total2, mask0, mask1);
+ mask = __msa_pckev_b((v16i8)mask1, (v16i8)mask0);
+ tmp = __msa_bmz_v(tmp, src, (v16u8)mask);
+
+ if (col == 0)
+ {+ uint64_t src_d;
+
+ src_d = __msa_copy_u_d((v2i64)tmp_orig, 1);
+ SD(src_d, (src_dup - 8));
+ }
+
+ src7 = LD_UB(src_dup + 16 * (col + 1) + 7);
+ src8 = LD_UB(src_dup + 16 * (col + 1) - 8);
+ ST_UB(tmp, (src_dup + (16 * col)));
+ }
+
+ src_dup += pitch;
+ }
+ }
+}
+
+void vp8_mbpost_proc_down_msa(uint8_t *dst_ptr, int32_t pitch, int32_t rows,
+ int32_t cols, int32_t flimit)
+{+ int32_t row, col, cnt, i;
+ const int16_t *rv3 = &vp8_rv_msa[63 & rand()];
+ v4i32 flimit_vec;
+ v16u8 dst7, dst8, dst_r_b, dst_l_b;
+ v16i8 mask;
+ v8u16 add_r, add_l;
+ v8i16 dst_r_h, dst_l_h, sub_r, sub_l, mask0, mask1;
+ v4i32 sub0, sub1, sub2, sub3, total0, total1, total2, total3;
+
+ flimit_vec = __msa_fill_w(flimit);
+
+ for (col = 0; col < (cols >> 4); ++col)
+ {+ uint8_t *dst_tmp = &dst_ptr[col << 4];
+ v16u8 dst;
+ v16i8 zero = { 0 };+ v16u8 tmp[16];
+ v8i16 mult0, mult1, rv2_0, rv2_1;
+ v8i16 sum0_h = { 0 };+ v8i16 sum1_h = { 0 };+ v4i32 mul0 = { 0 };+ v4i32 mul1 = { 0 };+ v4i32 mul2 = { 0 };+ v4i32 mul3 = { 0 };+ v4i32 sum0_w, sum1_w, sum2_w, sum3_w;
+ v4i32 add0, add1, add2, add3;
+ const int16_t *rv2[16];
+
+ dst = LD_UB(dst_tmp);
+ for (cnt = (col << 4), i = 0; i < 16; ++cnt)
+ {+ rv2[i] = rv3 + ((cnt * 17) & 127);
+ ++i;
+ }
+ for (cnt = -8; cnt < 0; ++cnt)
+ {+ ST_UB(dst, dst_tmp + cnt * pitch);
+ }
+
+ dst = LD_UB((dst_tmp + (rows - 1) * pitch));
+ for (cnt = rows; cnt < rows + 17; ++cnt)
+ {+ ST_UB(dst, dst_tmp + cnt * pitch);
+ }
+ for (cnt = -8; cnt <= 6; ++cnt)
+ {+ dst = LD_UB(dst_tmp + (cnt * pitch));
+ UNPCK_UB_SH(dst, dst_r_h, dst_l_h);
+ MUL2(dst_r_h, dst_r_h, dst_l_h, dst_l_h, mult0, mult1);
+ mul0 += (v4i32)__msa_ilvr_h((v8i16)zero, (v8i16)mult0);
+ mul1 += (v4i32)__msa_ilvl_h((v8i16)zero, (v8i16)mult0);
+ mul2 += (v4i32)__msa_ilvr_h((v8i16)zero, (v8i16)mult1);
+ mul3 += (v4i32)__msa_ilvl_h((v8i16)zero, (v8i16)mult1);
+ ADD2(sum0_h, dst_r_h, sum1_h, dst_l_h, sum0_h, sum1_h);
+ }
+
+ for (row = 0; row < (rows + 8); ++row)
+ {+ for (i = 0; i < 8; ++i)
+ {+ rv2_0[i] = *(rv2[i] + (row & 127));
+ rv2_1[i] = *(rv2[i + 8] + (row & 127));
+ }
+ dst7 = LD_UB(dst_tmp + (7 * pitch));
+ dst8 = LD_UB(dst_tmp - (8 * pitch));
+ ILVRL_B2_UB(dst7, dst8, dst_r_b, dst_l_b);
+
+ HSUB_UB2_SH(dst_r_b, dst_l_b, sub_r, sub_l);
+ UNPCK_SH_SW(sub_r, sub0, sub1);
+ UNPCK_SH_SW(sub_l, sub2, sub3);
+ sum0_h += sub_r;
+ sum1_h += sub_l;
+
+ HADD_UB2_UH(dst_r_b, dst_l_b, add_r, add_l);
+
+ ILVRL_H2_SW(zero, add_r, add0, add1);
+ ILVRL_H2_SW(zero, add_l, add2, add3);
+ mul0 += add0 * sub0;
+ mul1 += add1 * sub1;
+ mul2 += add2 * sub2;
+ mul3 += add3 * sub3;
+ dst = LD_UB(dst_tmp);
+ ILVRL_B2_SH(zero, dst, dst_r_h, dst_l_h);
+ dst7 = (v16u8)((rv2_0 + sum0_h + dst_r_h) >> 4);
+ dst8 = (v16u8)((rv2_1 + sum1_h + dst_l_h) >> 4);
+ tmp[row & 15] = (v16u8)__msa_pckev_b((v16i8)dst8, (v16i8)dst7);
+
+ UNPCK_SH_SW(sum0_h, sum0_w, sum1_w);
+ UNPCK_SH_SW(sum1_h, sum2_w, sum3_w);
+ total0 = mul0 * __msa_ldi_w(15);
+ total0 -= sum0_w * sum0_w;
+ total1 = mul1 * __msa_ldi_w(15);
+ total1 -= sum1_w * sum1_w;
+ total2 = mul2 * __msa_ldi_w(15);
+ total2 -= sum2_w * sum2_w;
+ total3 = mul3 * __msa_ldi_w(15);
+ total3 -= sum3_w * sum3_w;
+ total0 = (total0 < flimit_vec);
+ total1 = (total1 < flimit_vec);
+ total2 = (total2 < flimit_vec);
+ total3 = (total3 < flimit_vec);
+ PCKEV_H2_SH(total1, total0, total3, total2, mask0, mask1);
+ mask = __msa_pckev_b((v16i8)mask1, (v16i8)mask0);
+ tmp[row & 15] = __msa_bmz_v(tmp[row & 15], dst, (v16u8)mask);
+
+ if (row >= 8)
+ {+ ST_UB(tmp[(row - 8) & 15], (dst_tmp - 8 * pitch));
+ }
+
+ dst_tmp += pitch;
+ }
+ }
+}
+
+void vp8_plane_add_noise_msa(uint8_t *start_ptr, char *noise,
+ char blackclamp[16], char whiteclamp[16],
+ char bothclamp[16],
+ uint32_t width, uint32_t height,
+ int32_t pitch)
+{+ uint32_t i, j;
+
+ for (i = 0; i < height / 2; ++i)
+ {+ uint8_t *pos0_ptr = start_ptr + (2 * i) * pitch;
+ int8_t *ref0_ptr = (int8_t *) (noise + (rand() & 0xff));
+ uint8_t *pos1_ptr = start_ptr + (2 * i + 1) * pitch;
+ int8_t *ref1_ptr = (int8_t *) (noise + (rand() & 0xff));
+ for (j = width / 16; j--;)
+ {+ v16i8 temp00_s, temp01_s;
+ v16u8 temp00, temp01, black_clamp, white_clamp;
+ v16u8 pos0, ref0, pos1, ref1;
+ v16i8 const127 = __msa_ldi_b(127);
+
+ pos0 = LD_UB(pos0_ptr);
+ ref0 = LD_UB(ref0_ptr);
+ pos1 = LD_UB(pos1_ptr);
+ ref1 = LD_UB(ref1_ptr);
+ black_clamp = (v16u8)__msa_fill_b(blackclamp[0]);
+ white_clamp = (v16u8)__msa_fill_b(whiteclamp[0]);
+ temp00 = (pos0 < black_clamp);
+ pos0 = __msa_bmnz_v(pos0, black_clamp, temp00);
+ temp01 = (pos1 < black_clamp);
+ pos1 = __msa_bmnz_v(pos1, black_clamp, temp01);
+ XORI_B2_128_UB(pos0, pos1);
+ temp00_s = __msa_adds_s_b((v16i8)white_clamp, const127);
+ temp00 = (v16u8)(temp00_s < pos0);
+ pos0 = (v16u8)__msa_bmnz_v((v16u8)pos0, (v16u8)temp00_s, temp00);
+ temp01_s = __msa_adds_s_b((v16i8)white_clamp, const127);
+ temp01 = (temp01_s < pos1);
+ pos1 = (v16u8)__msa_bmnz_v((v16u8)pos1, (v16u8)temp01_s, temp01);
+ XORI_B2_128_UB(pos0, pos1);
+ pos0 += ref0;
+ ST_UB(pos0, pos0_ptr);
+ pos1 += ref1;
+ ST_UB(pos1, pos1_ptr);
+ pos0_ptr += 16;
+ pos1_ptr += 16;
+ ref0_ptr += 16;
+ ref1_ptr += 16;
+ }
+ }
+}
--- a/vp8/common/mips/msa/vp8_macros_msa.h
+++ b/vp8/common/mips/msa/vp8_macros_msa.h
@@ -435,6 +435,19 @@
ST4x4_UB(in1, in1, 0, 1, 2, 3, pblk_4x8 + 4 * stride, stride); \
}
+/* Description : Store 8x1 byte block to destination memory from input vector
+ Arguments : Inputs - in, pdst
+ Details : Index 0 double word element from 'in' vector is copied to the
+ GP register and stored to (pdst)
+*/
+#define ST8x1_UB(in, pdst) \
+{ \+ uint64_t out0_m; \
+ \
+ out0_m = __msa_copy_u_d((v2i64)in, 0); \
+ SD(out0_m, pdst); \
+}
+
/* Description : Store 8x2 byte block to destination memory from input vector
Arguments : Inputs - in, pdst, stride
Details : Index 0 double word element from 'in' vector is copied to the
@@ -479,6 +492,22 @@
SD4(out0_m, out1_m, out2_m, out3_m, pblk_8x4_m, stride); \
}
+/* Description : Immediate number of elements to slide with zero
+ Arguments : Inputs - in0, in1, slide_val
+ Outputs - out0, out1
+ Return Type - as per RTYPE
+ Details : Byte elements from 'zero_m' vector are slid into 'in0' by
+ value specified in the 'slide_val'
+*/
+#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val) \
+{ \+ v16i8 zero_m = { 0 }; \+ \
+ out0 = (RTYPE)__msa_sldi_b((v16i8)zero_m, (v16i8)in0, slide_val); \
+ out1 = (RTYPE)__msa_sldi_b((v16i8)zero_m, (v16i8)in1, slide_val); \
+}
+#define SLDI_B2_0_UB(...) SLDI_B2_0(v16u8, __VA_ARGS__)
+
/* Description : Immediate number of elements to slide
Arguments : Inputs - in0_0, in0_1, in1_0, in1_1, slide_val
Outputs - out0, out1
@@ -642,6 +671,76 @@
out_m; \
})
+/* Description : Horizontal addition of 4 signed word elements of input vector
+ Arguments : Input - in (signed word vector)
+ Output - sum_m (i32 sum)
+ Return Type - signed word (GP)
+ Details : 4 signed word elements of 'in' vector are added together and
+ the resulting integer sum is returned
+*/
+#define HADD_SW_S32(in) \
+({ \+ v2i64 res0_m, res1_m; \
+ int32_t sum_m; \
+ \
+ res0_m = __msa_hadd_s_d((v4i32)in, (v4i32)in); \
+ res1_m = __msa_splati_d(res0_m, 1); \
+ res0_m = res0_m + res1_m; \
+ sum_m = __msa_copy_s_w((v4i32)res0_m, 0); \
+ sum_m; \
+})
+
+/* Description : Horizontal addition of 8 unsigned halfword elements
+ Arguments : Inputs - in (unsigned halfword vector)
+ Outputs - sum_m (u32 sum)
+ Return Type - unsigned word
+ Details : 8 unsigned halfword elements of input vector are added
+ together and the resulting integer sum is returned
+*/
+#define HADD_UH_U32(in) \
+({ \+ v4u32 res_m; \
+ v2u64 res0_m, res1_m; \
+ uint32_t sum_m; \
+ \
+ res_m = __msa_hadd_u_w((v8u16)in, (v8u16)in); \
+ res0_m = __msa_hadd_u_d(res_m, res_m); \
+ res1_m = (v2u64)__msa_splati_d((v2i64)res0_m, 1); \
+ res0_m = res0_m + res1_m; \
+ sum_m = __msa_copy_u_w((v4i32)res0_m, 0); \
+ sum_m; \
+})
+
+/* Description : Horizontal addition of unsigned byte vector elements
+ Arguments : Inputs - in0, in1
+ Outputs - out0, out1
+ Return Type - as per RTYPE
+ Details : Each unsigned odd byte element from 'in0' is added to
+ even unsigned byte element from 'in0' (pairwise) and the
+ halfword result is written to 'out0'
+*/
+#define HADD_UB2(RTYPE, in0, in1, out0, out1) \
+{ \+ out0 = (RTYPE)__msa_hadd_u_h((v16u8)in0, (v16u8)in0); \
+ out1 = (RTYPE)__msa_hadd_u_h((v16u8)in1, (v16u8)in1); \
+}
+#define HADD_UB2_UH(...) HADD_UB2(v8u16, __VA_ARGS__)
+
+/* Description : Horizontal subtraction of unsigned byte vector elements
+ Arguments : Inputs - in0, in1
+ Outputs - out0, out1
+ Return Type - as per RTYPE
+ Details : Each unsigned odd byte element from 'in0' is subtracted from
+ even unsigned byte element from 'in0' (pairwise) and the
+ halfword result is written to 'out0'
+*/
+#define HSUB_UB2(RTYPE, in0, in1, out0, out1) \
+{ \+ out0 = (RTYPE)__msa_hsub_u_h((v16u8)in0, (v16u8)in0); \
+ out1 = (RTYPE)__msa_hsub_u_h((v16u8)in1, (v16u8)in1); \
+}
+#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
+
/* Description : Set element n input vector to GPR value
Arguments : Inputs - in0, in1, in2, in3
Output - out
@@ -864,6 +963,9 @@
out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
}
+#define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
+#define ILVRL_B2_SB(...) ILVRL_B2(v16i8, __VA_ARGS__)
+#define ILVRL_B2_UH(...) ILVRL_B2(v8u16, __VA_ARGS__)
#define ILVRL_B2_SH(...) ILVRL_B2(v8i16, __VA_ARGS__)
#define ILVRL_H2(RTYPE, in0, in1, out0, out1) \
@@ -1194,6 +1296,28 @@
out2 = in1 - in2; \
out3 = in0 - in3; \
}
+
+/* Description : Transpose input 8x8 byte block
+ Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
+ Outputs - out0, out1, out2, out3, out4, out5, out6, out7
+ Return Type - as per RTYPE
+*/
+#define TRANSPOSE8x8_UB(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3, out4, out5, out6, out7) \
+{ \+ v16i8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
+ v16i8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
+ \
+ ILVR_B4_SB(in2, in0, in3, in1, in6, in4, in7, in5, \
+ tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
+ ILVRL_B2_SB(tmp1_m, tmp0_m, tmp4_m, tmp5_m); \
+ ILVRL_B2_SB(tmp3_m, tmp2_m, tmp6_m, tmp7_m); \
+ ILVRL_W2(RTYPE, tmp6_m, tmp4_m, out0, out2); \
+ ILVRL_W2(RTYPE, tmp7_m, tmp5_m, out4, out6); \
+ SLDI_B2_0(RTYPE, out0, out2, out1, out3, 8); \
+ SLDI_B2_0(RTYPE, out4, out6, out5, out7, 8); \
+}
+#define TRANSPOSE8x8_UB_UB(...) TRANSPOSE8x8_UB(v16u8, __VA_ARGS__)
/* Description : Transpose 16x4 block into 4x16 with byte elements in vectors
Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7,
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -167,18 +167,18 @@
#
if (vpx_config("CONFIG_POSTPROC") eq "yes") {add_proto qw/void vp8_mbpost_proc_down/, "unsigned char *dst, int pitch, int rows, int cols,int flimit";
- specialize qw/vp8_mbpost_proc_down mmx sse2/;
+ specialize qw/vp8_mbpost_proc_down mmx sse2 msa/;
$vp8_mbpost_proc_down_sse2=vp8_mbpost_proc_down_xmm;
add_proto qw/void vp8_mbpost_proc_across_ip/, "unsigned char *dst, int pitch, int rows, int cols,int flimit";
- specialize qw/vp8_mbpost_proc_across_ip sse2/;
+ specialize qw/vp8_mbpost_proc_across_ip sse2 msa/;
$vp8_mbpost_proc_across_ip_sse2=vp8_mbpost_proc_across_ip_xmm;
add_proto qw/void vp8_post_proc_down_and_across_mb_row/, "unsigned char *src, unsigned char *dst, int src_pitch, int dst_pitch, int cols, unsigned char *flimits, int size";
- specialize qw/vp8_post_proc_down_and_across_mb_row sse2/;
+ specialize qw/vp8_post_proc_down_and_across_mb_row sse2 msa/;
add_proto qw/void vp8_plane_add_noise/, "unsigned char *s, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int w, unsigned int h, int pitch";
- specialize qw/vp8_plane_add_noise mmx sse2/;
+ specialize qw/vp8_plane_add_noise mmx sse2 msa/;
$vp8_plane_add_noise_sse2=vp8_plane_add_noise_wmt;
add_proto qw/void vp8_blend_mb_inner/, "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride";
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -124,6 +124,7 @@
ifeq ($(CONFIG_POSTPROC),yes)
VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/mfqe_msa.c
+VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/postproc_msa.c
endif
# common (c)
--
⑨