ref: 16dcf013d37d57ced6e97b74b46f083375a29d50
parent: a15edeb76d83ba60c1a3c416e128bf0db22ed18e
parent: fb73ceae854bd7a3be1f9203844cecb24c5e543d
author: Parag Salasakar <[email protected]>
date: Fri Jul 24 14:16:23 EDT 2015
Merge "mips msa vp8 bilinear filter optimization"
--- /dev/null
+++ b/vp8/common/mips/msa/bilinear_filter_msa.c
@@ -1,0 +1,911 @@
+/*
+ * 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 "./vp8_rtcd.h"
+#include "vpx_ports/mem.h"
+#include "vp8/common/filter.h"
+#include "vp8/common/mips/msa/vp8_macros_msa.h"
+
+DECLARE_ALIGNED(16, static const int8_t, vp8_bilinear_filters_msa[7][2]) =
+{+ { 112, 16 },+ { 96, 32 },+ { 80, 48 },+ { 64, 64 },+ { 48, 80 },+ { 32, 96 },+ { 16, 112 }+};
+
+static const uint8_t vp8_mc_filt_mask_arr[16 * 3] =
+{+ /* 8 width cases */
+ 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
+ /* 4 width cases */
+ 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20,
+ /* 4 width cases */
+ 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
+};
+
+static void common_hz_2t_4x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16i8 src0, src1, src2, src3, mask;
+ v16u8 filt0, vec0, vec1, res0, res1;
+ v8u16 vec2, vec3, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[16]);
+
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
+ SRARI_H2_UH(vec2, vec3, VP8_FILTER_SHIFT);
+ PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_4x8_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16u8 vec0, vec1, vec2, vec3, filt0;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16i8 res0, res1, res2, res3;
+ v8u16 vec4, vec5, vec6, vec7, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[16]);
+
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec4, vec5, vec6, vec7);
+ SRARI_H4_UH(vec4, vec5, vec6, vec7, VP8_FILTER_SHIFT);
+ PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7,
+ res0, res1, res2, res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_4w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ if (4 == height)
+ {+ common_hz_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ }
+ else if (8 == height)
+ {+ common_hz_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_hz_2t_8x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16u8 filt0;
+ v16i8 src0, src1, src2, src3, mask;
+ v8u16 vec0, vec1, vec2, vec3, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec0, vec1, vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, src0, src1);
+ ST8x4_UB(src0, src1, dst, dst_stride);
+}
+
+static void common_hz_2t_8x8mult_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ v16u8 filt0;
+ v16i8 src0, src1, src2, src3, mask, out0, out1;
+ v8u16 vec0, vec1, vec2, vec3, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec0, vec1, vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, VP8_FILTER_SHIFT);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec0, vec1, vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ if (16 == height)
+ {+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec0, vec1, vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, VP8_FILTER_SHIFT);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ vec0, vec1, vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst + 4 * dst_stride, dst_stride);
+ }
+}
+
+static void common_hz_2t_8w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ if (4 == height)
+ {+ common_hz_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ }
+ else
+ {+ common_hz_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
+ }
+}
+
+static void common_hz_2t_16w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ loop_cnt = (height >> 2) - 1;
+
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ out0, out1, out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0,
+ out4, out5, out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, VP8_FILTER_SHIFT);
+ SRARI_H4_UH(out4, out5, out6, out7, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(out0, out1, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out2, out3, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out4, out5, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out6, out7, dst);
+ dst += dst_stride;
+
+ for (; loop_cnt--;)
+ {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ out0, out1, out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0,
+ out4, out5, out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, VP8_FILTER_SHIFT);
+ SRARI_H4_UH(out4, out5, out6, out7, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(out0, out1, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out2, out3, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out4, out5, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out6, out7, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_vt_2t_4x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16i8 src0, src1, src2, src3, src4;
+ v16i8 src10_r, src32_r, src21_r, src43_r, src2110, src4332;
+ v16u8 filt0;
+ v8i16 filt;
+ v8u16 tmp0, tmp1;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ src += (5 * src_stride);
+
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3,
+ src10_r, src21_r, src32_r, src43_r);
+ ILVR_D2_SB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
+ DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, VP8_FILTER_SHIFT);
+ src2110 = __msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
+ ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_vt_2t_4x8_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src2110, src4332, src6554, src8776;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v16u8 filt0;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ src8 = LD_SB(src);
+ src += src_stride;
+
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
+ src32_r, src43_r);
+ ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
+ src76_r, src87_r);
+ ILVR_D4_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
+ src87_r, src76_r, src2110, src4332, src6554, src8776);
+ DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
+ ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
+ ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst + 4 * dst_stride, dst_stride);
+}
+
+static void common_vt_2t_4w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ if (4 == height)
+ {+ common_vt_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ }
+ else if (8 == height)
+ {+ common_vt_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_vt_2t_8x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter)
+{+ v16u8 src0, src1, src2, src3, src4, vec0, vec1, vec2, vec3, filt0;
+ v16i8 out0, out1;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
+ ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+}
+
+static void common_vt_2t_8x8mult_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v16i8 out0, out1;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 3); loop_cnt--;)
+ {+ LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
+ src += (8 * src_stride);
+
+ ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3,
+ vec0, vec1, vec2, vec3);
+ ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7,
+ vec4, vec5, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src0 = src8;
+ }
+}
+
+static void common_vt_2t_8w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ if (4 == height)
+ {+ common_vt_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ }
+ else
+ {+ common_vt_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter,
+ height);
+ }
+}
+
+static void common_vt_2t_16w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter, int32_t height)
+{+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;)
+ {+ LD_UB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+ dst += dst_stride;
+
+ ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
+ ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp2, tmp3, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp2, tmp3, dst);
+ dst += dst_stride;
+
+ src0 = src4;
+ }
+}
+
+static void common_hv_2ht_2vt_4x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert)
+{+ v16i8 src0, src1, src2, src3, src4, mask;
+ v16u8 filt_vt, filt_hz, vec0, vec1, res0, res1;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, filt, tmp0, tmp1;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[16]);
+
+ filt = LD_UH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
+ filt = LD_UH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
+ hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
+
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, VP8_FILTER_SHIFT);
+ PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_4x8_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert)
+{+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
+ v16i8 res0, res1, res2, res3;
+ v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8u16 hz_out7, hz_out8, vec4, vec5, vec6, vec7, filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[16]);
+
+ filt = LD_UH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
+ filt = LD_UH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ src8 = LD_SB(src);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, VP8_FILTER_SHIFT);
+ SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
+ hz_out3, hz_out5, 8);
+ hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
+
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt,
+ vec4, vec5, vec6, vec7);
+ SRARI_H4_UH(vec4, vec5, vec6, vec7, VP8_FILTER_SHIFT);
+ PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7,
+ res0, res1, res2, res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_4w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert,
+ int32_t height)
+{+ if (4 == height)
+ {+ common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ }
+ else if (8 == height)
+ {+ common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ }
+}
+
+static void common_hv_2ht_2vt_8x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert)
+{+ v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
+ v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
+ v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp0 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, VP8_FILTER_SHIFT);
+ vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp1 = __msa_dotp_u_h(vec1, filt_vt);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, VP8_FILTER_SHIFT);
+ vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp2 = __msa_dotp_u_h(vec2, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, VP8_FILTER_SHIFT);
+ vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp3 = __msa_dotp_u_h(vec3, filt_vt);
+
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_8x8mult_msa(uint8_t *RESTRICT src,
+ int32_t src_stride,
+ uint8_t *RESTRICT dst,
+ int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert,
+ int32_t height)
+{+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
+ v16u8 filt_hz, filt_vt, vec0;
+ v8u16 hz_out0, hz_out1, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
+ v8i16 filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_SB(src);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, VP8_FILTER_SHIFT);
+
+ for (loop_cnt = (height >> 3); loop_cnt--;)
+ {+ LD_SB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp1 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp2 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp1, tmp2, VP8_FILTER_SHIFT);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp3 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ LD_SB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp4 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp3, tmp4, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp2, tmp1, tmp4, tmp3, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp5 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp6 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp7 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp8 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H4_UH(tmp5, tmp6, tmp7, tmp8, VP8_FILTER_SHIFT);
+ PCKEV_B2_SB(tmp6, tmp5, tmp8, tmp7, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void common_hv_2ht_2vt_8w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert,
+ int32_t height)
+{+ if (4 == height)
+ {+ common_hv_2ht_2vt_8x4_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ }
+ else
+ {+ common_hv_2ht_2vt_8x8mult_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ }
+}
+
+static void common_hv_2ht_2vt_16w_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ uint8_t *RESTRICT dst, int32_t dst_stride,
+ const int8_t *filter_horiz,
+ const int8_t *filter_vert,
+ int32_t height)
+{+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt_hz, filt_vt, vec0, vec1;
+ v8u16 tmp1, tmp2, hz_out0, hz_out1, hz_out2, hz_out3;
+ v8i16 filt;
+
+ mask = LD_SB(&vp8_mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB2(src, 8, src0, src1);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, VP8_FILTER_SHIFT);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, VP8_FILTER_SHIFT);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;)
+ {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz,
+ VP8_FILTER_SHIFT);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, VP8_FILTER_SHIFT);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+ }
+}
+
+void vp8_bilinear_predict4x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ int32_t xoffset, int32_t yoffset,
+ uint8_t *RESTRICT dst, int32_t dst_stride)
+{+ const int8_t *h_filter = vp8_bilinear_filters_msa[xoffset - 1];
+ const int8_t *v_filter = vp8_bilinear_filters_msa[yoffset - 1];
+
+ if (yoffset)
+ {+ if (xoffset)
+ {+ common_hv_2ht_2vt_4w_msa(src, src_stride, dst, dst_stride,
+ h_filter, v_filter, 4);
+ }
+ else
+ {+ common_vt_2t_4w_msa(src, src_stride, dst, dst_stride, v_filter, 4);
+ }
+ }
+ else
+ {+ if (xoffset)
+ {+ common_hz_2t_4w_msa(src, src_stride, dst, dst_stride, h_filter, 4);
+ }
+ else
+ {+ uint32_t tp0, tp1, tp2, tp3;
+
+ LW4(src, src_stride, tp0, tp1, tp2, tp3);
+ SW4(tp0, tp1, tp2, tp3, dst, dst_stride);
+ }
+ }
+}
+
+void vp8_bilinear_predict8x4_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ int32_t xoffset, int32_t yoffset,
+ uint8_t *RESTRICT dst, int32_t dst_stride)
+{+ const int8_t *h_filter = vp8_bilinear_filters_msa[xoffset - 1];
+ const int8_t *v_filter = vp8_bilinear_filters_msa[yoffset - 1];
+
+ if (yoffset)
+ {+ if (xoffset)
+ {+ common_hv_2ht_2vt_8w_msa(src, src_stride, dst, dst_stride,
+ h_filter, v_filter, 4);
+ }
+ else
+ {+ common_vt_2t_8w_msa(src, src_stride, dst, dst_stride, v_filter, 4);
+ }
+ }
+ else
+ {+ if (xoffset)
+ {+ common_hz_2t_8w_msa(src, src_stride, dst, dst_stride, h_filter, 4);
+ }
+ else
+ {+ vp8_copy_mem8x4(src, src_stride, dst, dst_stride);
+ }
+ }
+}
+
+void vp8_bilinear_predict8x8_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ int32_t xoffset, int32_t yoffset,
+ uint8_t *RESTRICT dst, int32_t dst_stride)
+{+ const int8_t *h_filter = vp8_bilinear_filters_msa[xoffset - 1];
+ const int8_t *v_filter = vp8_bilinear_filters_msa[yoffset - 1];
+
+ if (yoffset)
+ {+ if (xoffset)
+ {+ common_hv_2ht_2vt_8w_msa(src, src_stride, dst, dst_stride,
+ h_filter, v_filter, 8);
+ }
+ else
+ {+ common_vt_2t_8w_msa(src, src_stride, dst, dst_stride, v_filter, 8);
+ }
+ }
+ else
+ {+ if (xoffset)
+ {+ common_hz_2t_8w_msa(src, src_stride, dst, dst_stride, h_filter, 8);
+ }
+ else
+ {+ vp8_copy_mem8x8(src, src_stride, dst, dst_stride);
+ }
+ }
+}
+
+void vp8_bilinear_predict16x16_msa(uint8_t *RESTRICT src, int32_t src_stride,
+ int32_t xoffset, int32_t yoffset,
+ uint8_t *RESTRICT dst, int32_t dst_stride)
+{+ const int8_t *h_filter = vp8_bilinear_filters_msa[xoffset - 1];
+ const int8_t *v_filter = vp8_bilinear_filters_msa[yoffset - 1];
+
+ if (yoffset)
+ {+ if (xoffset)
+ {+ common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride,
+ h_filter, v_filter, 16);
+ }
+ else
+ {+ common_vt_2t_16w_msa(src, src_stride, dst, dst_stride, v_filter,
+ 16);
+ }
+ }
+ else
+ {+ if (xoffset)
+ {+ common_hz_2t_16w_msa(src, src_stride, dst, dst_stride, h_filter,
+ 16);
+ }
+ else
+ {+ vp8_copy_mem16x16(src, src_stride, dst, dst_stride);
+ }
+ }
+}
--- a/vp8/common/mips/msa/vp8_macros_msa.h
+++ b/vp8/common/mips/msa/vp8_macros_msa.h
@@ -460,6 +460,27 @@
SD4(out0_m, out1_m, out2_m, out3_m, pblk_8x4_m, stride); \
}
+/* Description : Immediate number of elements to slide
+ Arguments : Inputs - in0_0, in0_1, in1_0, in1_1, slide_val
+ Outputs - out0, out1
+ Return Type - as per RTYPE
+ Details : Byte elements from 'in0_0' vector are slid into 'in1_0' by
+ value specified in the 'slide_val'
+*/
+#define SLDI_B2(RTYPE, in0_0, in0_1, in1_0, in1_1, out0, out1, slide_val) \
+{ \+ out0 = (RTYPE)__msa_sldi_b((v16i8)in0_0, (v16i8)in1_0, slide_val); \
+ out1 = (RTYPE)__msa_sldi_b((v16i8)in0_1, (v16i8)in1_1, slide_val); \
+}
+
+#define SLDI_B3(RTYPE, in0_0, in0_1, in0_2, in1_0, in1_1, in1_2, \
+ out0, out1, out2, slide_val) \
+{ \+ SLDI_B2(RTYPE, in0_0, in0_1, in1_0, in1_1, out0, out1, slide_val); \
+ out2 = (RTYPE)__msa_sldi_b((v16i8)in0_2, (v16i8)in1_2, slide_val); \
+}
+#define SLDI_B3_UH(...) SLDI_B3(v8u16, __VA_ARGS__)
+
/* Description : Shuffle byte vector elements as per mask vector
Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
Outputs - out0, out1
@@ -472,7 +493,9 @@
out0 = (RTYPE)__msa_vshf_b((v16i8)mask0, (v16i8)in1, (v16i8)in0); \
out1 = (RTYPE)__msa_vshf_b((v16i8)mask1, (v16i8)in3, (v16i8)in2); \
}
+#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
#define VSHF_B2_SB(...) VSHF_B2(v16i8, __VA_ARGS__)
+#define VSHF_B2_UH(...) VSHF_B2(v8u16, __VA_ARGS__)
#define VSHF_B3(RTYPE, in0, in1, in2, in3, in4, in5, mask0, mask1, mask2, \
out0, out1, out2) \
@@ -486,6 +509,32 @@
Arguments : Inputs - mult0, mult1, cnst0, cnst1
Outputs - out0, out1
Return Type - as per RTYPE
+ Details : Unsigned byte elements from 'mult0' are multiplied with
+ unsigned byte elements from 'cnst0' producing a result
+ twice the size of input i.e. unsigned halfword.
+ The multiplication result of adjacent odd-even elements
+ are added together and written to the 'out0' vector
+*/
+#define DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
+{ \+ out0 = (RTYPE)__msa_dotp_u_h((v16u8)mult0, (v16u8)cnst0); \
+ out1 = (RTYPE)__msa_dotp_u_h((v16u8)mult1, (v16u8)cnst1); \
+}
+#define DOTP_UB2_UH(...) DOTP_UB2(v8u16, __VA_ARGS__)
+
+#define DOTP_UB4(RTYPE, mult0, mult1, mult2, mult3, \
+ cnst0, cnst1, cnst2, cnst3, \
+ out0, out1, out2, out3) \
+{ \+ DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
+ DOTP_UB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
+}
+#define DOTP_UB4_UH(...) DOTP_UB4(v8u16, __VA_ARGS__)
+
+/* Description : Dot product of byte vector elements
+ Arguments : Inputs - mult0, mult1, cnst0, cnst1
+ Outputs - out0, out1
+ Return Type - as per RTYPE
Details : Signed byte elements from 'mult0' are multiplied with
signed byte elements from 'cnst0' producing a result
twice the size of input i.e. signed halfword.
@@ -768,6 +817,7 @@
ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1); \
ILVR_D2(RTYPE, in4, in5, in6, in7, out2, out3); \
}
+#define ILVR_D4_SB(...) ILVR_D4(v16i8, __VA_ARGS__)
#define ILVR_D4_UB(...) ILVR_D4(v16u8, __VA_ARGS__)
/* Description : Interleave both left and right half of input vectors
@@ -1251,5 +1301,31 @@
out_m = (v16u8)__msa_pckev_b((v16i8)in1, (v16i8)in0); \
out_m = (v16u8)__msa_xori_b((v16u8)out_m, 128); \
out_m; \
+})
+
+/* Description : Pack even byte elements and store byte vector in destination
+ memory
+ Arguments : Inputs - in0, in1, pdst
+*/
+#define PCKEV_ST_SB(in0, in1, pdst) \
+{ \+ v16i8 tmp_m; \
+ tmp_m = __msa_pckev_b((v16i8)in1, (v16i8)in0); \
+ ST_SB(tmp_m, (pdst)); \
+}
+
+/* Description : Horizontal 2 tap filter kernel code
+ Arguments : Inputs - in0, in1, mask, coeff, shift
+*/
+#define HORIZ_2TAP_FILT_UH(in0, in1, mask, coeff, shift) \
+({ \+ v16i8 tmp0_m; \
+ v8u16 tmp1_m; \
+ \
+ tmp0_m = __msa_vshf_b((v16i8)mask, (v16i8)in1, (v16i8)in0); \
+ tmp1_m = __msa_dotp_u_h((v16u8)tmp0_m, (v16u8)coeff); \
+ tmp1_m = (v8u16)__msa_srari_h((v8i16)tmp1_m, shift); \
+ \
+ tmp1_m; \
})
#endif /* VP8_COMMON_MIPS_MSA_VP8_MACROS_MSA_H_ */
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -225,20 +225,20 @@
$vp8_sixtap_predict4x4_dspr2=vp8_sixtap_predict4x4_dspr2;
add_proto qw/void vp8_bilinear_predict16x16/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon/;
+specialize qw/vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon msa/;
$vp8_bilinear_predict16x16_media=vp8_bilinear_predict16x16_armv6;
add_proto qw/void vp8_bilinear_predict8x8/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_bilinear_predict8x8 mmx sse2 ssse3 media neon/;
+specialize qw/vp8_bilinear_predict8x8 mmx sse2 ssse3 media neon msa/;
$vp8_bilinear_predict8x8_media=vp8_bilinear_predict8x8_armv6;
add_proto qw/void vp8_bilinear_predict8x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_bilinear_predict8x4 mmx media neon/;
+specialize qw/vp8_bilinear_predict8x4 mmx media neon msa/;
$vp8_bilinear_predict8x4_media=vp8_bilinear_predict8x4_armv6;
add_proto qw/void vp8_bilinear_predict4x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch";
#TODO(johannkoenig): fix the neon version https://code.google.com/p/webm/issues/detail?id=892
-specialize qw/vp8_bilinear_predict4x4 mmx media/;
+specialize qw/vp8_bilinear_predict4x4 mmx media msa/;
$vp8_bilinear_predict4x4_media=vp8_bilinear_predict4x4_armv6;
#
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -114,6 +114,7 @@
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/dequantize_dspr2.c
# common (c)
+VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/bilinear_filter_msa.c
VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/copymem_msa.c
VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/idct_msa.c
VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/loopfilter_filters_msa.c