avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for me_cmp functions

This patch adds MSA (MIPS-SIMD-Arch) optimizations for me_cmp functions in new file me_cmp_msa.c
Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/me_cmp.c

@@ -991,4 +991,6 @@ av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)
         ff_me_cmp_init_ppc(c, avctx);
     if (ARCH_X86)
         ff_me_cmp_init_x86(c, avctx);
+    if (ARCH_MIPS)
+        ff_me_cmp_init_mips(c, avctx);
 }

libavcodec/me_cmp.h

@@ -87,6 +87,7 @@ void ff_me_cmp_init_alpha(MECmpContext *c, AVCodecContext *avctx);
 void ff_me_cmp_init_arm(MECmpContext *c, AVCodecContext *avctx);
 void ff_me_cmp_init_ppc(MECmpContext *c, AVCodecContext *avctx);
 void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx);
+void ff_me_cmp_init_mips(MECmpContext *c, AVCodecContext *avctx);
 
 void ff_set_cmp(MECmpContext *c, me_cmp_func *cmp, int type);
 

libavcodec/mips/Makefile

@@ -31,6 +31,7 @@ OBJS-$(CONFIG_BLOCKDSP)                   += mips/blockdsp_init_mips.o
 OBJS-$(CONFIG_PIXBLOCKDSP)                += mips/pixblockdsp_init_mips.o
 OBJS-$(CONFIG_MPEGVIDEO)                  += mips/mpegvideo_init_mips.o
 OBJS-$(CONFIG_MPEGVIDEOENC)               += mips/mpegvideoencdsp_init_mips.o
+OBJS-$(CONFIG_ME_CMP)                     += mips/me_cmp_init_mips.o
 MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevc_mc_uni_msa.o        \
                                              mips/hevc_mc_uniw_msa.o       \
@@ -51,5 +52,6 @@ MSA-OBJS-$(CONFIG_BLOCKDSP)               += mips/blockdsp_msa.o
 MSA-OBJS-$(CONFIG_PIXBLOCKDSP)            += mips/pixblockdsp_msa.o
 MSA-OBJS-$(CONFIG_MPEGVIDEO)              += mips/mpegvideo_msa.o
 MSA-OBJS-$(CONFIG_MPEGVIDEOENC)           += mips/mpegvideoencdsp_msa.o
+MSA-OBJS-$(CONFIG_ME_CMP)                 += mips/me_cmp_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o
 LOONGSON3-OBJS-$(CONFIG_H264CHROMA)       += mips/h264chroma_mmi.o

libavcodec/mips/me_cmp_init_mips.c

+/*
+ * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@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 "me_cmp_mips.h"
+
+#if HAVE_MSA
+static av_cold void me_cmp_msa(MECmpContext *c, AVCodecContext *avctx)
+{
+#if BIT_DEPTH == 8
+    c->pix_abs[0][0] = ff_pix_abs16_msa;
+    c->pix_abs[0][1] = ff_pix_abs16_x2_msa;
+    c->pix_abs[0][2] = ff_pix_abs16_y2_msa;
+    c->pix_abs[0][3] = ff_pix_abs16_xy2_msa;
+    c->pix_abs[1][0] = ff_pix_abs8_msa;
+    c->pix_abs[1][1] = ff_pix_abs8_x2_msa;
+    c->pix_abs[1][2] = ff_pix_abs8_y2_msa;
+    c->pix_abs[1][3] = ff_pix_abs8_xy2_msa;
+
+    c->hadamard8_diff[0] = ff_hadamard8_diff16_msa;
+    c->hadamard8_diff[1] = ff_hadamard8_diff8x8_msa;
+
+    c->hadamard8_diff[4] = ff_hadamard8_intra16_msa;
+    c->hadamard8_diff[5] = ff_hadamard8_intra8x8_msa;
+
+    c->sad[0] = ff_pix_abs16_msa;
+    c->sad[1] = ff_pix_abs8_msa;
+    c->sse[0] = ff_sse16_msa;
+    c->sse[1] = ff_sse8_msa;
+    c->sse[2] = ff_sse4_msa;
+#endif
+}
+#endif  // #if HAVE_MSA
+
+av_cold void ff_me_cmp_init_mips(MECmpContext *c, AVCodecContext *avctx)
+{
+#if HAVE_MSA
+    me_cmp_msa(c, avctx);
+#endif  // #if HAVE_MSA
+}

libavcodec/mips/me_cmp_mips.h

+/*
+ * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@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
+ */
+
+#ifndef AVCODEC_MIPS_ME_CMP_MIPS_H
+#define AVCODEC_MIPS_ME_CMP_MIPS_H
+
+#include "../mpegvideo.h"
+#include "libavcodec/bit_depth_template.c"
+
+int ff_hadamard8_diff8x8_msa(MpegEncContext *s, uint8_t *dst, uint8_t *src,
+                             ptrdiff_t stride, int h);
+int ff_hadamard8_intra8x8_msa(MpegEncContext *s, uint8_t *dst, uint8_t *src,
+                              ptrdiff_t stride, int h);
+int ff_hadamard8_diff16_msa(MpegEncContext *s, uint8_t *dst, uint8_t *src,
+                            ptrdiff_t stride, int h);
+int ff_hadamard8_intra16_msa(MpegEncContext *s, uint8_t *dst, uint8_t *src,
+                             ptrdiff_t stride, int h);
+int ff_pix_abs16_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                     ptrdiff_t stride, int h);
+int ff_pix_abs16_x2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                        ptrdiff_t stride, int h);
+int ff_pix_abs16_y2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                        ptrdiff_t stride, int h);
+int ff_pix_abs16_xy2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                         ptrdiff_t stride, int h);
+int ff_pix_abs8_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                    ptrdiff_t stride, int h);
+int ff_pix_abs8_x2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                       ptrdiff_t stride, int h);
+int ff_pix_abs8_y2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                       ptrdiff_t stride, int h);
+int ff_pix_abs8_xy2_msa(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
+                        ptrdiff_t stride, int h);
+int ff_sse16_msa(MpegEncContext *v, uint8_t *pu8Src, uint8_t *pu8Ref,
+                 ptrdiff_t stride, int i32Height);
+int ff_sse8_msa(MpegEncContext *v, uint8_t *pu8Src, uint8_t *pu8Ref,
+                ptrdiff_t stride, int i32Height);
+int ff_sse4_msa(MpegEncContext *v, uint8_t *pu8Src, uint8_t *pu8Ref,
+                ptrdiff_t stride, int i32Height);
+void ff_add_pixels8_msa(uint8_t *av_restrict pixels, int16_t *block,
+                        ptrdiff_t stride);
+
+#endif  // #ifndef AVCODEC_MIPS_ME_CMP_MIPS_H

libavutil/mips/generic_macros_msa.h

@@ -1295,6 +1295,29 @@
 #define HSUB_UB4_UH(...) HSUB_UB4(v8u16, __VA_ARGS__)
 #define HSUB_UB4_SH(...) HSUB_UB4(v8i16, __VA_ARGS__)
 
+/* Description : SAD (Sum of Absolute Difference)
+   Arguments   : Inputs  - in0, in1, ref0, ref1  (unsigned byte src & ref)
+                 Outputs - sad_m                 (halfword vector with sad)
+                 Return Type - unsigned halfword
+   Details     : Absolute difference of all the byte elements from 'in0' with
+                 'ref0' is calculated and preserved in 'diff0'. From the 16
+                 unsigned absolute diff values, even-odd pairs are added
+                 together to generate 8 halfword results.
+*/
+#define SAD_UB2_UH(in0, in1, ref0, ref1)                        \
+( {                                                             \
+    v16u8 diff0_m, diff1_m;                                     \
+    v8u16 sad_m = { 0 };                                        \
+                                                                \
+    diff0_m = __msa_asub_u_b((v16u8) in0, (v16u8) ref0);        \
+    diff1_m = __msa_asub_u_b((v16u8) in1, (v16u8) ref1);        \
+                                                                \
+    sad_m += __msa_hadd_u_h((v16u8) diff0_m, (v16u8) diff0_m);  \
+    sad_m += __msa_hadd_u_h((v16u8) diff1_m, (v16u8) diff1_m);  \
+                                                                \
+    sad_m;                                                      \
+} )
+
 /* Description : Insert specified word elements from input vectors to 1
                  destination vector
    Arguments   : Inputs  - in0, in1, in2, in3 (4 input vectors)
@@ -2429,6 +2452,42 @@
 }
 #define TRANSPOSE8x8_UB_UB(...) TRANSPOSE8x8_UB(v16u8, __VA_ARGS__)
 #define TRANSPOSE8x8_UB_UH(...) TRANSPOSE8x8_UB(v8u16, __VA_ARGS__)
+
+/* Description : Transposes 16x4 block into 4x16 with byte elements in vectors
+   Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
+                           in8, in9, in10, in11, in12, in13, in14, in15
+                 Outputs - out0, out1, out2, out3
+                 Return Type - unsigned byte
+   Details     :
+*/
+#define TRANSPOSE16x4_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7,        \
+                            in8, in9, in10, in11, in12, in13, in14, in15,  \
+                            out0, out1, out2, out3)                        \
+{                                                                          \
+    v2i64 tmp0_m, tmp1_m, tmp2_m, tmp3_m;                                  \
+                                                                           \
+    ILVEV_W2_SD(in0, in4, in8, in12, tmp0_m, tmp1_m);                      \
+    out1 = (v16u8) __msa_ilvev_d(tmp1_m, tmp0_m);                          \
+                                                                           \
+    ILVEV_W2_SD(in1, in5, in9, in13, tmp0_m, tmp1_m);                      \
+    out3 = (v16u8) __msa_ilvev_d(tmp1_m, tmp0_m);                          \
+                                                                           \
+    ILVEV_W2_SD(in2, in6, in10, in14, tmp0_m, tmp1_m);                     \
+                                                                           \
+    tmp2_m = __msa_ilvev_d(tmp1_m, tmp0_m);                                \
+    ILVEV_W2_SD(in3, in7, in11, in15, tmp0_m, tmp1_m);                     \
+                                                                           \
+    tmp3_m = __msa_ilvev_d(tmp1_m, tmp0_m);                                \
+    ILVEV_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m);               \
+    out0 = (v16u8) __msa_ilvev_h((v8i16) tmp1_m, (v8i16) tmp0_m);          \
+    out2 = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m);          \
+                                                                           \
+    tmp0_m = (v2i64) __msa_ilvod_b((v16i8) out3, (v16i8) out1);            \
+    tmp1_m = (v2i64) __msa_ilvod_b((v16i8) tmp3_m, (v16i8) tmp2_m);        \
+    out1 = (v16u8) __msa_ilvev_h((v8i16) tmp1_m, (v8i16) tmp0_m);          \
+    out3 = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m);          \
+}
+
 /* Description : Transposes 16x8 block into 8x16 with byte elements in vectors
    Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
                            in8, in9, in10, in11, in12, in13, in14, in15