[FFmpeg-devel] [PATCH 3/4] avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for AVC idct functions

shivraj.patil at imgtec.com shivraj.patil at imgtec.com
Wed Jun 10 16:18:40 CEST 2015


From: Shivraj Patil <shivraj.patil at imgtec.com>

This patch adds MSA (MIPS-SIMD-Arch) optimizations for AVC idct functions in new file h264idct_msa.c
Adds new generic macros (needed for this patch) in libavutil/mips/generic_macros_msa.h

Signed-off-by: Shivraj Patil <shivraj.patil at imgtec.com>
---
 libavcodec/mips/Makefile            |   3 +-
 libavcodec/mips/h264dsp_init_mips.c |  15 ++
 libavcodec/mips/h264dsp_mips.h      |  24 ++
 libavcodec/mips/h264idct_msa.c      | 469 ++++++++++++++++++++++++++++++++++++
 libavutil/mips/generic_macros_msa.h |  96 ++++++++
 5 files changed, 606 insertions(+), 1 deletion(-)
 create mode 100644 libavcodec/mips/h264idct_msa.c

diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile
index 0a97e7c..993c649 100644
--- a/libavcodec/mips/Makefile
+++ b/libavcodec/mips/Makefile
@@ -31,7 +31,8 @@ MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevc_idct_msa.o          \
                                              mips/hevc_lpf_sao_msa.o       \
                                              mips/hevcpred_msa.o
-MSA-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_msa.o
+MSA-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_msa.o            \
+                                             mips/h264idct_msa.o
 MSA-OBJS-$(CONFIG_H264CHROMA)             += mips/h264chroma_msa.o
 MSA-OBJS-$(CONFIG_H264PRED)               += mips/h264pred_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o
diff --git a/libavcodec/mips/h264dsp_init_mips.c b/libavcodec/mips/h264dsp_init_mips.c
index d9182f2..7f74adf 100644
--- a/libavcodec/mips/h264dsp_init_mips.c
+++ b/libavcodec/mips/h264dsp_init_mips.c
@@ -62,6 +62,21 @@ static av_cold void h264dsp_init_msa(H264DSPContext *c,
         c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_8_msa;
         c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_msa;
         c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_msa;
+
+        c->h264_idct_add = ff_h264_idct_add_msa;
+        c->h264_idct8_add = ff_h264_idct8_addblk_msa;
+        c->h264_idct_dc_add = ff_h264_idct4x4_addblk_dc_msa;
+        c->h264_idct8_dc_add = ff_h264_idct8_dc_addblk_msa;
+        c->h264_idct_add16 = ff_h264_idct_add16_msa;
+        c->h264_idct8_add4 = ff_h264_idct8_add4_msa;
+
+        if (chroma_format_idc <= 1)
+            c->h264_idct_add8 = ff_h264_idct_add8_msa;
+        else
+            c->h264_idct_add8 = ff_h264_idct_add8_422_msa;
+
+        c->h264_idct_add16intra = ff_h264_idct_add16_intra_msa;
+        c->h264_luma_dc_dequant_idct = ff_h264_deq_idct_luma_dc_msa;
     }  // if (8 == bit_depth)
 }
 #endif  // #if HAVE_MSA
diff --git a/libavcodec/mips/h264dsp_mips.h b/libavcodec/mips/h264dsp_mips.h
index ef380fa..0e39057 100644
--- a/libavcodec/mips/h264dsp_mips.h
+++ b/libavcodec/mips/h264dsp_mips.h
@@ -41,6 +41,30 @@ void ff_h264_h_loop_filter_luma_mbaff_msa(uint8_t *src, int32_t stride,
                                           int32_t alpha, int32_t beta,
                                           int8_t *tc0);
 
+void ff_h264_idct_add_msa(uint8_t *dst, int16_t *src, int32_t dst_stride);
+void ff_h264_idct4x4_addblk_dc_msa(uint8_t *dst, int16_t *src,
+                                   int32_t dst_stride);
+void ff_h264_deq_idct_luma_dc_msa(int16_t *dst, int16_t *src,
+                                  int32_t de_q_val);
+void ff_h264_idct_add16_msa(uint8_t *dst, const int32_t *blk_offset,
+                            int16_t *block, int32_t stride,
+                            const uint8_t nnzc[15 * 8]);
+void ff_h264_idct_add16_intra_msa(uint8_t *dst, const int32_t *blk_offset,
+                                  int16_t *block, int32_t dst_stride,
+                                  const uint8_t nnzc[15 * 8]);
+void ff_h264_idct_add8_msa(uint8_t **dst, const int32_t *blk_offset,
+                           int16_t *block, int32_t dst_stride,
+                           const uint8_t nnzc[15 * 8]);
+void ff_h264_idct_add8_422_msa(uint8_t **dst, const int32_t *blk_offset,
+                               int16_t *block, int32_t dst_stride,
+                               const uint8_t nnzc[15 * 8]);
+void ff_h264_idct8_addblk_msa(uint8_t *dst, int16_t *src, int32_t dst_stride);
+void ff_h264_idct8_dc_addblk_msa(uint8_t *dst, int16_t *src,
+                                 int32_t dst_stride);
+void ff_h264_idct8_add4_msa(uint8_t *dst, const int *blk_offset,
+                            int16_t *blk, int dst_stride,
+                            const uint8_t nnzc[15 * 8]);
+
 void ff_h264_h_lpf_luma_intra_msa(uint8_t *src, int stride,
                                   int alpha, int beta);
 void ff_h264_v_lpf_luma_intra_msa(uint8_t *src, int stride,
diff --git a/libavcodec/mips/h264idct_msa.c b/libavcodec/mips/h264idct_msa.c
new file mode 100644
index 0000000..fac1e7a
--- /dev/null
+++ b/libavcodec/mips/h264idct_msa.c
@@ -0,0 +1,469 @@
+/*
+ * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale at imgtec.com)
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/mips/generic_macros_msa.h"
+#include "h264dsp_mips.h"
+#include "libavcodec/bit_depth_template.c"
+
+#define AVC_ITRANS_H(in0, in1, in2, in3, out0, out1, out2, out3)          \
+{                                                                         \
+    v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m;                                 \
+                                                                          \
+    tmp0_m = in0 + in2;                                                   \
+    tmp1_m = in0 - in2;                                                   \
+    tmp2_m = in1 >> 1;                                                    \
+    tmp2_m = tmp2_m - in3;                                                \
+    tmp3_m = in3 >> 1;                                                    \
+    tmp3_m = in1 + tmp3_m;                                                \
+                                                                          \
+    BUTTERFLY_4(tmp0_m, tmp1_m, tmp2_m, tmp3_m, out0, out1, out2, out3);  \
+}
+
+static void avc_idct4x4_addblk_msa(uint8_t *dst, int16_t *src,
+                                   int32_t dst_stride)
+{
+    v8i16 src0, src1, src2, src3;
+    v8i16 hres0, hres1, hres2, hres3;
+    v8i16 vres0, vres1, vres2, vres3;
+    v8i16 zeros = { 0 };
+
+    LD4x4_SH(src, src0, src1, src2, src3);
+    AVC_ITRANS_H(src0, src1, src2, src3, hres0, hres1, hres2, hres3);
+    TRANSPOSE4x4_SH_SH(hres0, hres1, hres2, hres3, hres0, hres1, hres2, hres3);
+    AVC_ITRANS_H(hres0, hres1, hres2, hres3, vres0, vres1, vres2, vres3);
+    SRARI_H4_SH(vres0, vres1, vres2, vres3, 6);
+    ADDBLK_ST4x4_UB(vres0, vres1, vres2, vres3, dst, dst_stride);
+    ST_SH2(zeros, zeros, src, 8);
+}
+
+static void avc_idct4x4_addblk_dc_msa(uint8_t *dst, int16_t *src,
+                                      int32_t dst_stride)
+{
+    int16_t dc;
+    uint32_t src0, src1, src2, src3;
+    v16u8 pred = { 0 };
+    v16i8 out;
+    v8i16 input_dc, pred_r, pred_l;
+
+    dc = (src[0] + 32) >> 6;
+    input_dc = __msa_fill_h(dc);
+    src[0] = 0;
+
+    LW4(dst, dst_stride, src0, src1, src2, src3);
+    INSERT_W4_UB(src0, src1, src2, src3, pred);
+    UNPCK_UB_SH(pred, pred_r, pred_l);
+
+    pred_r += input_dc;
+    pred_l += input_dc;
+
+    CLIP_SH2_0_255(pred_r, pred_l);
+    out = __msa_pckev_b((v16i8) pred_l, (v16i8) pred_r);
+    ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void avc_deq_idct_luma_dc_msa(int16_t *dst, int16_t *src,
+                                     int32_t de_q_val)
+{
+#define DC_DEST_STRIDE 16
+    int16_t out0, out1, out2, out3;
+    v8i16 src0, src1, src2, src3;
+    v8i16 vec0, vec1, vec2, vec3;
+    v8i16 hres0, hres1, hres2, hres3;
+    v8i16 vres0, vres1, vres2, vres3;
+    v4i32 vres0_r, vres1_r, vres2_r, vres3_r;
+    v4i32 de_q_vec = __msa_fill_w(de_q_val);
+
+    LD4x4_SH(src, src0, src1, src2, src3);
+    TRANSPOSE4x4_SH_SH(src0, src1, src2, src3, src0, src1, src2, src3);
+    BUTTERFLY_4(src0, src2, src3, src1, vec0, vec3, vec2, vec1);
+    BUTTERFLY_4(vec0, vec1, vec2, vec3, hres0, hres3, hres2, hres1);
+    TRANSPOSE4x4_SH_SH(hres0, hres1, hres2, hres3, hres0, hres1, hres2, hres3);
+    BUTTERFLY_4(hres0, hres1, hres3, hres2, vec0, vec3, vec2, vec1);
+    BUTTERFLY_4(vec0, vec1, vec2, vec3, vres0, vres1, vres2, vres3);
+    UNPCK_R_SH_SW(vres0, vres0_r);
+    UNPCK_R_SH_SW(vres1, vres1_r);
+    UNPCK_R_SH_SW(vres2, vres2_r);
+    UNPCK_R_SH_SW(vres3, vres3_r);
+
+    vres0_r *= de_q_vec;
+    vres1_r *= de_q_vec;
+    vres2_r *= de_q_vec;
+    vres3_r *= de_q_vec;
+
+    SRARI_W4_SW(vres0_r, vres1_r, vres2_r, vres3_r, 8);
+    PCKEV_H2_SH(vres1_r, vres0_r, vres3_r, vres2_r, vec0, vec1);
+
+    out0 = __msa_copy_s_h(vec0, 0);
+    out1 = __msa_copy_s_h(vec0, 1);
+    out2 = __msa_copy_s_h(vec0, 2);
+    out3 = __msa_copy_s_h(vec0, 3);
+    SH(out0, dst);
+    SH(out1, (dst + 2 * DC_DEST_STRIDE));
+    SH(out2, (dst + 8 * DC_DEST_STRIDE));
+    SH(out3, (dst + 10 * DC_DEST_STRIDE));
+    dst += DC_DEST_STRIDE;
+
+    out0 = __msa_copy_s_h(vec0, 4);
+    out1 = __msa_copy_s_h(vec0, 5);
+    out2 = __msa_copy_s_h(vec0, 6);
+    out3 = __msa_copy_s_h(vec0, 7);
+    SH(out0, dst);
+    SH(out1, (dst + 2 * DC_DEST_STRIDE));
+    SH(out2, (dst + 8 * DC_DEST_STRIDE));
+    SH(out3, (dst + 10 * DC_DEST_STRIDE));
+    dst += (3 * DC_DEST_STRIDE);
+
+    out0 = __msa_copy_s_h(vec1, 0);
+    out1 = __msa_copy_s_h(vec1, 1);
+    out2 = __msa_copy_s_h(vec1, 2);
+    out3 = __msa_copy_s_h(vec1, 3);
+    SH(out0, dst);
+    SH(out1, (dst + 2 * DC_DEST_STRIDE));
+    SH(out2, (dst + 8 * DC_DEST_STRIDE));
+    SH(out3, (dst + 10 * DC_DEST_STRIDE));
+    dst += DC_DEST_STRIDE;
+
+    out0 = __msa_copy_s_h(vec1, 4);
+    out1 = __msa_copy_s_h(vec1, 5);
+    out2 = __msa_copy_s_h(vec1, 6);
+    out3 = __msa_copy_s_h(vec1, 7);
+    SH(out0, dst);
+    SH(out1, (dst + 2 * DC_DEST_STRIDE));
+    SH(out2, (dst + 8 * DC_DEST_STRIDE));
+    SH(out3, (dst + 10 * DC_DEST_STRIDE));
+
+#undef DC_DEST_STRIDE
+}
+
+static void avc_idct8_addblk_msa(uint8_t *dst, int16_t *src, int32_t dst_stride)
+{
+    v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
+    v8i16 vec0, vec1, vec2, vec3;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+    v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
+    v4i32 tmp0_r, tmp1_r, tmp2_r, tmp3_r, tmp4_r, tmp5_r, tmp6_r, tmp7_r;
+    v4i32 tmp0_l, tmp1_l, tmp2_l, tmp3_l, tmp4_l, tmp5_l, tmp6_l, tmp7_l;
+    v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec0_l, vec1_l, vec2_l, vec3_l;
+    v4i32 res0_r, res1_r, res2_r, res3_r, res4_r, res5_r, res6_r, res7_r;
+    v4i32 res0_l, res1_l, res2_l, res3_l, res4_l, res5_l, res6_l, res7_l;
+    v16i8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+    v16i8 zeros = { 0 };
+
+    src[0] += 32;
+
+    LD_SH8(src, 8, src0, src1, src2, src3, src4, src5, src6, src7);
+
+    vec0 = src0 + src4;
+    vec1 = src0 - src4;
+    vec2 = src2 >> 1;
+    vec2 = vec2 - src6;
+    vec3 = src6 >> 1;
+    vec3 = src2 + vec3;
+
+    BUTTERFLY_4(vec0, vec1, vec2, vec3, tmp0, tmp1, tmp2, tmp3);
+
+    vec0 = src7 >> 1;
+    vec0 = src5 - vec0 - src3 - src7;
+    vec1 = src3 >> 1;
+    vec1 = src1 - vec1 + src7 - src3;
+    vec2 = src5 >> 1;
+    vec2 = vec2 - src1 + src7 + src5;
+    vec3 = src1 >> 1;
+    vec3 = vec3 + src3 + src5 + src1;
+    tmp4 = vec3 >> 2;
+    tmp4 += vec0;
+    tmp5 = vec2 >> 2;
+    tmp5 += vec1;
+    tmp6 = vec1 >> 2;
+    tmp6 -= vec2;
+    tmp7 = vec0 >> 2;
+    tmp7 = vec3 - tmp7;
+
+    BUTTERFLY_8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
+                res0, res1, res2, res3, res4, res5, res6, res7);
+    TRANSPOSE8x8_SH_SH(res0, res1, res2, res3, res4, res5, res6, res7,
+                       res0, res1, res2, res3, res4, res5, res6, res7);
+    UNPCK_SH_SW(res0, tmp0_r, tmp0_l);
+    UNPCK_SH_SW(res1, tmp1_r, tmp1_l);
+    UNPCK_SH_SW(res2, tmp2_r, tmp2_l);
+    UNPCK_SH_SW(res3, tmp3_r, tmp3_l);
+    UNPCK_SH_SW(res4, tmp4_r, tmp4_l);
+    UNPCK_SH_SW(res5, tmp5_r, tmp5_l);
+    UNPCK_SH_SW(res6, tmp6_r, tmp6_l);
+    UNPCK_SH_SW(res7, tmp7_r, tmp7_l);
+    BUTTERFLY_4(tmp0_r, tmp0_l, tmp4_l, tmp4_r, vec0_r, vec0_l, vec1_l, vec1_r);
+
+    vec2_r = tmp2_r >> 1;
+    vec2_l = tmp2_l >> 1;
+    vec2_r -= tmp6_r;
+    vec2_l -= tmp6_l;
+    vec3_r = tmp6_r >> 1;
+    vec3_l = tmp6_l >> 1;
+    vec3_r += tmp2_r;
+    vec3_l += tmp2_l;
+
+    BUTTERFLY_4(vec0_r, vec1_r, vec2_r, vec3_r, tmp0_r, tmp2_r, tmp4_r, tmp6_r);
+    BUTTERFLY_4(vec0_l, vec1_l, vec2_l, vec3_l, tmp0_l, tmp2_l, tmp4_l, tmp6_l);
+
+    vec0_r = tmp7_r >> 1;
+    vec0_l = tmp7_l >> 1;
+    vec0_r = tmp5_r - vec0_r - tmp3_r - tmp7_r;
+    vec0_l = tmp5_l - vec0_l - tmp3_l - tmp7_l;
+    vec1_r = tmp3_r >> 1;
+    vec1_l = tmp3_l >> 1;
+    vec1_r = tmp1_r - vec1_r + tmp7_r - tmp3_r;
+    vec1_l = tmp1_l - vec1_l + tmp7_l - tmp3_l;
+    vec2_r = tmp5_r >> 1;
+    vec2_l = tmp5_l >> 1;
+    vec2_r = vec2_r - tmp1_r + tmp7_r + tmp5_r;
+    vec2_l = vec2_l - tmp1_l + tmp7_l + tmp5_l;
+    vec3_r = tmp1_r >> 1;
+    vec3_l = tmp1_l >> 1;
+    vec3_r = vec3_r + tmp3_r + tmp5_r + tmp1_r;
+    vec3_l = vec3_l + tmp3_l + tmp5_l + tmp1_l;
+    tmp1_r = vec3_r >> 2;
+    tmp1_l = vec3_l >> 2;
+    tmp1_r += vec0_r;
+    tmp1_l += vec0_l;
+    tmp3_r = vec2_r >> 2;
+    tmp3_l = vec2_l >> 2;
+    tmp3_r += vec1_r;
+    tmp3_l += vec1_l;
+    tmp5_r = vec1_r >> 2;
+    tmp5_l = vec1_l >> 2;
+    tmp5_r -= vec2_r;
+    tmp5_l -= vec2_l;
+    tmp7_r = vec0_r >> 2;
+    tmp7_l = vec0_l >> 2;
+    tmp7_r = vec3_r - tmp7_r;
+    tmp7_l = vec3_l - tmp7_l;
+
+    BUTTERFLY_4(tmp0_r, tmp0_l, tmp7_l, tmp7_r, res0_r, res0_l, res7_l, res7_r);
+    BUTTERFLY_4(tmp2_r, tmp2_l, tmp5_l, tmp5_r, res1_r, res1_l, res6_l, res6_r);
+    BUTTERFLY_4(tmp4_r, tmp4_l, tmp3_l, tmp3_r, res2_r, res2_l, res5_l, res5_r);
+    BUTTERFLY_4(tmp6_r, tmp6_l, tmp1_l, tmp1_r, res3_r, res3_l, res4_l, res4_r);
+    SRA_4V(res0_r, res0_l, res1_r, res1_l, 6);
+    SRA_4V(res2_r, res2_l, res3_r, res3_l, 6);
+    SRA_4V(res4_r, res4_l, res5_r, res5_l, 6);
+    SRA_4V(res6_r, res6_l, res7_r, res7_l, 6);
+    PCKEV_H4_SH(res0_l, res0_r, res1_l, res1_r, res2_l, res2_r, res3_l, res3_r,
+                res0, res1, res2, res3);
+    PCKEV_H4_SH(res4_l, res4_r, res5_l, res5_r, res6_l, res6_r, res7_l, res7_r,
+                res4, res5, res6, res7);
+    LD_SB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+    ILVR_B4_SH(zeros, dst0, zeros, dst1, zeros, dst2, zeros, dst3,
+               tmp0, tmp1, tmp2, tmp3);
+    ILVR_B4_SH(zeros, dst4, zeros, dst5, zeros, dst6, zeros, dst7,
+               tmp4, tmp5, tmp6, tmp7);
+    ADD4(res0, tmp0, res1, tmp1, res2, tmp2, res3, tmp3,
+         res0, res1, res2, res3);
+    ADD4(res4, tmp4, res5, tmp5, res6, tmp6, res7, tmp7,
+         res4, res5, res6, res7);
+    CLIP_SH4_0_255(res0, res1, res2, res3);
+    CLIP_SH4_0_255(res4, res5, res6, res7);
+    PCKEV_B4_SB(res1, res0, res3, res2, res5, res4, res7, res6,
+                dst0, dst1, dst2, dst3);
+    ST8x4_UB(dst0, dst1, dst, dst_stride);
+    dst += (4 * dst_stride);
+    ST8x4_UB(dst2, dst3, dst, dst_stride);
+}
+
+static void avc_idct8_dc_addblk_msa(uint8_t *dst, int16_t *src,
+                                    int32_t dst_stride)
+{
+    int32_t dc_val;
+    v16i8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+    v8i16 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
+    v8i16 dc;
+    v16i8 zeros = { 0 };
+
+    dc_val = (src[0] + 32) >> 6;
+    dc = __msa_fill_h(dc_val);
+
+    src[0] = 0;
+
+    LD_SB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+    ILVR_B4_SH(zeros, dst0, zeros, dst1, zeros, dst2, zeros, dst3,
+               dst0_r, dst1_r, dst2_r, dst3_r);
+    ILVR_B4_SH(zeros, dst4, zeros, dst5, zeros, dst6, zeros, dst7,
+               dst4_r, dst5_r, dst6_r, dst7_r);
+    ADD4(dst0_r, dc, dst1_r, dc, dst2_r, dc, dst3_r, dc,
+         dst0_r, dst1_r, dst2_r, dst3_r);
+    ADD4(dst4_r, dc, dst5_r, dc, dst6_r, dc, dst7_r, dc,
+         dst4_r, dst5_r, dst6_r, dst7_r);
+    CLIP_SH4_0_255(dst0_r, dst1_r, dst2_r, dst3_r);
+    CLIP_SH4_0_255(dst4_r, dst5_r, dst6_r, dst7_r);
+    PCKEV_B4_SB(dst1_r, dst0_r, dst3_r, dst2_r, dst5_r, dst4_r, dst7_r, dst6_r,
+                dst0, dst1, dst2, dst3);
+    ST8x4_UB(dst0, dst1, dst, dst_stride);
+    dst += (4 * dst_stride);
+    ST8x4_UB(dst2, dst3, dst, dst_stride);
+}
+
+void ff_h264_idct_add_msa(uint8_t *dst, int16_t *src,
+                          int32_t dst_stride)
+{
+    avc_idct4x4_addblk_msa(dst, src, dst_stride);
+    memset(src, 0, 16 * sizeof(dctcoef));
+}
+
+void ff_h264_idct8_addblk_msa(uint8_t *dst, int16_t *src,
+                              int32_t dst_stride)
+{
+    avc_idct8_addblk_msa(dst, src, dst_stride);
+    memset(src, 0, 64 * sizeof(dctcoef));
+}
+
+void ff_h264_idct4x4_addblk_dc_msa(uint8_t *dst, int16_t *src,
+                                   int32_t dst_stride)
+{
+    avc_idct4x4_addblk_dc_msa(dst, src, dst_stride);
+}
+
+void ff_h264_idct8_dc_addblk_msa(uint8_t *dst, int16_t *src,
+                                 int32_t dst_stride)
+{
+    avc_idct8_dc_addblk_msa(dst, src, dst_stride);
+}
+
+void ff_h264_idct_add16_msa(uint8_t *dst,
+                            const int32_t *blk_offset,
+                            int16_t *block, int32_t dst_stride,
+                            const uint8_t nzc[15 * 8])
+{
+    int32_t i;
+
+    for (i = 0; i < 16; i++) {
+        int32_t nnz = nzc[scan8[i]];
+
+        if (nnz) {
+            if (nnz == 1 && ((dctcoef *) block)[i * 16])
+                ff_h264_idct4x4_addblk_dc_msa(dst + blk_offset[i],
+                                              block + i * 16 * sizeof(pixel),
+                                              dst_stride);
+            else
+                ff_h264_idct_add_msa(dst + blk_offset[i],
+                                     block + i * 16 * sizeof(pixel),
+                                     dst_stride);
+        }
+    }
+}
+
+void ff_h264_idct8_add4_msa(uint8_t *dst, const int32_t *blk_offset,
+                            int16_t *block, int32_t dst_stride,
+                            const uint8_t nzc[15 * 8])
+{
+    int32_t cnt;
+
+    for (cnt = 0; cnt < 16; cnt += 4) {
+        int32_t nnz = nzc[scan8[cnt]];
+
+        if (nnz) {
+            if (nnz == 1 && ((dctcoef *) block)[cnt * 16])
+                ff_h264_idct8_dc_addblk_msa(dst + blk_offset[cnt],
+                                            block + cnt * 16 * sizeof(pixel),
+                                            dst_stride);
+            else
+                ff_h264_idct8_addblk_msa(dst + blk_offset[cnt],
+                                         block + cnt * 16 * sizeof(pixel),
+                                         dst_stride);
+        }
+    }
+}
+
+void ff_h264_idct_add8_msa(uint8_t **dst,
+                           const int32_t *blk_offset,
+                           int16_t *block, int32_t dst_stride,
+                           const uint8_t nzc[15 * 8])
+{
+    int32_t i, j;
+
+    for (j = 1; j < 3; j++) {
+        for (i = (j * 16); i < (j * 16 + 4); i++) {
+            if (nzc[scan8[i]])
+                ff_h264_idct_add_msa(dst[j - 1] + blk_offset[i],
+                                     block + i * 16 * sizeof(pixel),
+                                     dst_stride);
+            else if (((dctcoef *) block)[i * 16])
+                ff_h264_idct4x4_addblk_dc_msa(dst[j - 1] + blk_offset[i],
+                                              block + i * 16 * sizeof(pixel),
+                                              dst_stride);
+        }
+    }
+}
+
+void ff_h264_idct_add8_422_msa(uint8_t **dst,
+                               const int32_t *blk_offset,
+                               int16_t *block, int32_t dst_stride,
+                               const uint8_t nzc[15 * 8])
+{
+    int32_t i, j;
+
+    for (j = 1; j < 3; j++) {
+        for (i = (j * 16); i < (j * 16 + 4); i++) {
+            if (nzc[scan8[i]])
+                ff_h264_idct_add_msa(dst[j - 1] + blk_offset[i],
+                                     block + i * 16 * sizeof(pixel),
+                                     dst_stride);
+            else if (((dctcoef *) block)[i * 16])
+                ff_h264_idct4x4_addblk_dc_msa(dst[j - 1] + blk_offset[i],
+                                              block + i * 16 * sizeof(pixel),
+                                              dst_stride);
+        }
+    }
+
+    for (j = 1; j < 3; j++) {
+        for (i = (j * 16 + 4); i < (j * 16 + 8); i++) {
+            if (nzc[scan8[i + 4]])
+                ff_h264_idct_add_msa(dst[j - 1] + blk_offset[i + 4],
+                                     block + i * 16 * sizeof(pixel),
+                                     dst_stride);
+            else if (((dctcoef *) block)[i * 16])
+                ff_h264_idct4x4_addblk_dc_msa(dst[j - 1] + blk_offset[i + 4],
+                                              block + i * 16 * sizeof(pixel),
+                                              dst_stride);
+        }
+    }
+}
+
+void ff_h264_idct_add16_intra_msa(uint8_t *dst,
+                                  const int32_t *blk_offset,
+                                  int16_t *block,
+                                  int32_t dst_stride,
+                                  const uint8_t nzc[15 * 8])
+{
+    int32_t i;
+
+    for (i = 0; i < 16; i++) {
+        if (nzc[scan8[i]])
+            ff_h264_idct_add_msa(dst + blk_offset[i],
+                                 block + i * 16 * sizeof(pixel), dst_stride);
+        else if (((dctcoef *) block)[i * 16])
+            ff_h264_idct4x4_addblk_dc_msa(dst + blk_offset[i],
+                                          block + i * 16 * sizeof(pixel),
+                                          dst_stride);
+    }
+}
+
+void ff_h264_deq_idct_luma_dc_msa(int16_t *dst, int16_t *src,
+                                  int32_t de_qval)
+{
+    avc_deq_idct_luma_dc_msa(dst, src, de_qval);
+}
diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h
index 0f06c80..1d2d1d0 100644
--- a/libavutil/mips/generic_macros_msa.h
+++ b/libavutil/mips/generic_macros_msa.h
@@ -403,6 +403,19 @@
 }
 #define LD_SH16(...) LD_H16(v8i16, __VA_ARGS__)
 
+/* Description : Load as 4x4 block of signed halfword elements from 1D source
+                 data into 4 vectors (Each vector with 4 signed halfwords)
+   Arguments   : Inputs  - psrc
+                 Outputs - out0, out1, out2, out3
+*/
+#define LD4x4_SH(psrc, out0, out1, out2, out3)                \
+{                                                             \
+    out0 = LD_SH(psrc);                                       \
+    out2 = LD_SH(psrc + 8);                                   \
+    out1 = (v8i16) __msa_ilvl_d((v2i64) out0, (v2i64) out0);  \
+    out3 = (v8i16) __msa_ilvl_d((v2i64) out2, (v2i64) out2);  \
+}
+
 /* Description : Load 2 vectors of signed word elements with stride
    Arguments   : Inputs  - psrc    (source pointer to load from)
                          - stride
@@ -2025,6 +2038,22 @@
     out1 = in2 - in3;                         \
 }
 
+/* Description : Sign extend halfword elements from right half of the vector
+   Arguments   : Inputs  - in    (input halfword vector)
+                 Outputs - out   (sign extended word vectors)
+                 Return Type - signed word
+   Details     : Sign bit of halfword elements from input vector 'in' is
+                 extracted and interleaved with same vector 'in0' to generate
+                 4 word elements keeping sign intact
+*/
+#define UNPCK_R_SH_SW(in, out)                       \
+{                                                    \
+    v8i16 sign_m;                                    \
+                                                     \
+    sign_m = __msa_clti_s_h((v8i16) in, 0);          \
+    out = (v4i32) __msa_ilvr_h(sign_m, (v8i16) in);  \
+}
+
 /* Description : Sign extend byte elements from input vector and return
                  halfword results in pair of vectors
    Arguments   : Inputs  - in           (1 input byte vector)
@@ -2103,6 +2132,25 @@
     out3 = in0 - in3;                                            \
 }
 
+/* Description : Butterfly of 8 input vectors
+   Arguments   : Inputs  - in0 ...  in7
+                 Outputs - out0 .. out7
+   Details     : Butterfly operation
+*/
+#define BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,          \
+                    out0, out1, out2, out3, out4, out5, out6, out7)  \
+{                                                                    \
+    out0 = in0 + in7;                                                \
+    out1 = in1 + in6;                                                \
+    out2 = in2 + in5;                                                \
+    out3 = in3 + in4;                                                \
+                                                                     \
+    out4 = in3 - in4;                                                \
+    out5 = in2 - in5;                                                \
+    out6 = in1 - in6;                                                \
+    out7 = in0 - in7;                                                \
+}
+
 /* Description : Transposes input 4x4 byte block
    Arguments   : Inputs  - in0, in1, in2, in3      (input 4x4 byte block)
                  Outputs - out0, out1, out2, out3  (output 4x4 byte block)
@@ -2222,6 +2270,22 @@
     out7 = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m);            \
 }
 
+/* Description : Transposes 4x4 block with half word elements in vectors
+   Arguments   : Inputs  - in0, in1, in2, in3
+                 Outputs - out0, out1, out2, out3
+                 Return Type - signed halfword
+   Details     :
+*/
+#define TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, out0, out1, out2, out3)  \
+{                                                                       \
+    v8i16 s0_m, s1_m;                                                   \
+                                                                        \
+    ILVR_H2_SH(in1, in0, in3, in2, s0_m, s1_m);                         \
+    ILVRL_W2_SH(s1_m, s0_m, out0, out2);                                \
+    out1 = (v8i16) __msa_ilvl_d((v2i64) out0, (v2i64) out0);            \
+    out3 = (v8i16) __msa_ilvl_d((v2i64) out0, (v2i64) out2);            \
+}
+
 /* Description : Transposes 8x8 block with half word elements in vectors
    Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7
                  Outputs - out0, out1, out2, out3, out4, out5, out6, out7
@@ -2272,6 +2336,38 @@
     out3 = (v4i32) __msa_ilvl_d((v2i64) s3_m, (v2i64) s1_m);            \
 }
 
+/* Description : Add block 4x4
+   Arguments   : Inputs  - in0, in1, in2, in3, pdst, stride
+                 Outputs -
+                 Return Type - unsigned bytes
+   Details     : Least significant 4 bytes from each input vector are added to
+                 the destination bytes, clipped between 0-255 and then stored.
+*/
+#define ADDBLK_ST4x4_UB(in0, in1, in2, in3, pdst, stride)         \
+{                                                                 \
+    uint32_t src0_m, src1_m, src2_m, src3_m;                      \
+    uint32_t out0_m, out1_m, out2_m, out3_m;                      \
+    v8i16 inp0_m, inp1_m, res0_m, res1_m;                         \
+    v16i8 dst0_m = { 0 };                                         \
+    v16i8 dst1_m = { 0 };                                         \
+    v16i8 zero_m = { 0 };                                         \
+                                                                  \
+    ILVR_D2_SH(in1, in0, in3, in2, inp0_m, inp1_m)                \
+    LW4(pdst, stride,  src0_m, src1_m, src2_m, src3_m);           \
+    INSERT_W2_SB(src0_m, src1_m, dst0_m);                         \
+    INSERT_W2_SB(src2_m, src3_m, dst1_m);                         \
+    ILVR_B2_SH(zero_m, dst0_m, zero_m, dst1_m, res0_m, res1_m);   \
+    ADD2(res0_m, inp0_m, res1_m, inp1_m, res0_m, res1_m);         \
+    CLIP_SH2_0_255(res0_m, res1_m);                               \
+    PCKEV_B2_SB(res0_m, res0_m, res1_m, res1_m, dst0_m, dst1_m);  \
+                                                                  \
+    out0_m = __msa_copy_u_w((v4i32) dst0_m, 0);                   \
+    out1_m = __msa_copy_u_w((v4i32) dst0_m, 1);                   \
+    out2_m = __msa_copy_u_w((v4i32) dst1_m, 0);                   \
+    out3_m = __msa_copy_u_w((v4i32) dst1_m, 1);                   \
+    SW4(out0_m, out1_m, out2_m, out3_m, pdst, stride);            \
+}
+
 /* Description : Pack even elements of input vectors & xor with 128
    Arguments   : Inputs  - in0, in1
                  Outputs - out_m
-- 
2.3.7



More information about the ffmpeg-devel mailing list