[FFmpeg-cvslog] avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for pixblock functions

Shivraj Patil git at videolan.org
Mon Jun 29 12:24:36 CEST 2015


ffmpeg | branch: master | Shivraj Patil <shivraj.patil at imgtec.com> | Sun Jun 14 23:44:26 2015 +0530| [d9deae04a78b6b698b90d050a67a3bd9155aba74] | committer: Michael Niedermayer

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

This patch adds MSA (MIPS-SIMD-Arch) optimizations for pixblock functions in new file pixblockdsp_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>
Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d9deae04a78b6b698b90d050a67a3bd9155aba74
---

 libavcodec/mips/Makefile                |    2 +
 libavcodec/mips/pixblockdsp_init_mips.c |   53 ++++++++++++
 libavcodec/mips/pixblockdsp_mips.h      |   33 +++++++
 libavcodec/mips/pixblockdsp_msa.c       |  143 +++++++++++++++++++++++++++++++
 libavcodec/pixblockdsp.c                |    2 +
 libavcodec/pixblockdsp.h                |    2 +
 libavutil/mips/generic_macros_msa.h     |    8 ++
 7 files changed, 243 insertions(+)

diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile
index 823a2c5..c0ecb15 100644
--- a/libavcodec/mips/Makefile
+++ b/libavcodec/mips/Makefile
@@ -28,6 +28,7 @@ OBJS-$(CONFIG_H263DSP)                    += mips/h263dsp_init_mips.o
 OBJS-$(CONFIG_QPELDSP)                    += mips/qpeldsp_init_mips.o
 OBJS-$(CONFIG_HPELDSP)                    += mips/hpeldsp_init_mips.o
 OBJS-$(CONFIG_BLOCKDSP)                   += mips/blockdsp_init_mips.o
+OBJS-$(CONFIG_PIXBLOCKDSP)                += mips/pixblockdsp_init_mips.o
 MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevc_mc_uni_msa.o        \
                                              mips/hevc_mc_uniw_msa.o       \
@@ -45,5 +46,6 @@ MSA-OBJS-$(CONFIG_H263DSP)                += mips/h263dsp_msa.o
 MSA-OBJS-$(CONFIG_QPELDSP)                += mips/qpeldsp_msa.o
 MSA-OBJS-$(CONFIG_HPELDSP)                += mips/hpeldsp_msa.o
 MSA-OBJS-$(CONFIG_BLOCKDSP)               += mips/blockdsp_msa.o
+MSA-OBJS-$(CONFIG_PIXBLOCKDSP)            += mips/pixblockdsp_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o
 LOONGSON3-OBJS-$(CONFIG_H264CHROMA)       += mips/h264chroma_mmi.o
diff --git a/libavcodec/mips/pixblockdsp_init_mips.c b/libavcodec/mips/pixblockdsp_init_mips.c
new file mode 100644
index 0000000..0f2fb15
--- /dev/null
+++ b/libavcodec/mips/pixblockdsp_init_mips.c
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil 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 "pixblockdsp_mips.h"
+
+#if HAVE_MSA
+static av_cold void pixblockdsp_init_msa(PixblockDSPContext *c,
+                                         AVCodecContext *avctx,
+                                         unsigned high_bit_depth)
+{
+    c->diff_pixels = ff_diff_pixels_msa;
+
+    switch (avctx->bits_per_raw_sample) {
+    case 9:
+    case 10:
+    case 12:
+    case 14:
+        c->get_pixels = ff_get_pixels_16_msa;
+        break;
+    default:
+        if (avctx->bits_per_raw_sample <= 8 || avctx->codec_type !=
+            AVMEDIA_TYPE_VIDEO) {
+            c->get_pixels = ff_get_pixels_8_msa;
+        }
+        break;
+    }
+}
+#endif  // #if HAVE_MSA
+
+void ff_pixblockdsp_init_mips(PixblockDSPContext *c, AVCodecContext *avctx,
+                              unsigned high_bit_depth)
+{
+#if HAVE_MSA
+    pixblockdsp_init_msa(c, avctx, high_bit_depth);
+#endif  // #if HAVE_MSA
+}
diff --git a/libavcodec/mips/pixblockdsp_mips.h b/libavcodec/mips/pixblockdsp_mips.h
new file mode 100644
index 0000000..3eee6e0
--- /dev/null
+++ b/libavcodec/mips/pixblockdsp_mips.h
@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil 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
+ */
+
+#ifndef AVCODEC_MIPS_PIXBLOCKDSP_MIPS_H
+#define AVCODEC_MIPS_PIXBLOCKDSP_MIPS_H
+
+#include "../mpegvideo.h"
+
+void ff_diff_pixels_msa(int16_t *av_restrict block, const uint8_t *src1,
+                        const uint8_t *src2, int stride);
+void ff_get_pixels_16_msa(int16_t *restrict dst, const uint8_t *src,
+                          ptrdiff_t stride);
+void ff_get_pixels_8_msa(int16_t *restrict dst, const uint8_t *src,
+                         ptrdiff_t stride);
+
+#endif  // #ifndef AVCODEC_MIPS_PIXBLOCKDSP_MIPS_H
diff --git a/libavcodec/mips/pixblockdsp_msa.c b/libavcodec/mips/pixblockdsp_msa.c
new file mode 100644
index 0000000..966e11a
--- /dev/null
+++ b/libavcodec/mips/pixblockdsp_msa.c
@@ -0,0 +1,143 @@
+/*
+ * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil 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 "pixblockdsp_mips.h"
+
+static void diff_pixels_msa(int16_t *block, const uint8_t *src1,
+                            const uint8_t *src2, int32_t stride)
+{
+    v16u8 in10, in11, in12, in13, in14, in15, in16, in17;
+    v16u8 in20, in21, in22, in23, in24, in25, in26, in27;
+    v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
+
+    LD_UB8(src1, stride, in10, in11, in12, in13, in14, in15, in16, in17);
+    LD_UB8(src2, stride, in20, in21, in22, in23, in24, in25, in26, in27);
+    ILVR_B4_SH(in10, in20, in11, in21, in12, in22, in13, in23,
+               out0, out1, out2, out3);
+    ILVR_B4_SH(in14, in24, in15, in25, in16, in26, in17, in27,
+               out4, out5, out6, out7);
+    HSUB_UB4_SH(out0, out1, out2, out3, out0, out1, out2, out3);
+    HSUB_UB4_SH(out4, out5, out6, out7, out4, out5, out6, out7);
+    ST_SH8(out0, out1, out2, out3, out4, out5, out6, out7, block, 8);
+}
+
+static void copy_8bit_to_16bit_width8_msa(const uint8_t *src, int32_t src_stride,
+                                          int16_t *dst, int32_t dst_stride,
+                                          int32_t height)
+{
+    uint8_t *dst_ptr;
+    int32_t cnt;
+    v16u8 src0, src1, src2, src3;
+    v16i8 zero = { 0 };
+
+    dst_ptr = (uint8_t *) dst;
+
+    for (cnt = (height >> 2); cnt--;) {
+        LD_UB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+
+        ILVR_B4_UB(zero, src0, zero, src1, zero, src2, zero, src3,
+                   src0, src1, src2, src3);
+
+        ST_UB4(src0, src1, src2, src3, dst_ptr, (dst_stride * 2));
+        dst_ptr += (4 * 2 * dst_stride);
+    }
+}
+
+static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
+                                  uint8_t *dst, int32_t dst_stride,
+                                  int32_t height, int32_t width)
+{
+    int32_t cnt, loop_cnt;
+    const uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+    for (cnt = (width >> 4); cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        for (loop_cnt = (height >> 3); loop_cnt--;) {
+            LD_UB8(src_tmp, src_stride,
+                   src0, src1, src2, src3, src4, src5, src6, src7);
+            src_tmp += (8 * src_stride);
+
+            ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7,
+                   dst_tmp, dst_stride);
+            dst_tmp += (8 * dst_stride);
+        }
+
+        src += 16;
+        dst += 16;
+    }
+}
+
+static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
+                             uint8_t *dst, int32_t dst_stride,
+                             int32_t height)
+{
+    int32_t cnt;
+    v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+    if (0 == height % 12) {
+        for (cnt = (height / 12); cnt--;) {
+            LD_UB8(src, src_stride,
+                   src0, src1, src2, src3, src4, src5, src6, src7);
+            src += (8 * src_stride);
+            ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7,
+                   dst, dst_stride);
+            dst += (8 * dst_stride);
+
+            LD_UB4(src, src_stride, src0, src1, src2, src3);
+            src += (4 * src_stride);
+            ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+            dst += (4 * dst_stride);
+        }
+    } else if (0 == height % 8) {
+        copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
+    } else if (0 == height % 4) {
+        for (cnt = (height >> 2); cnt--;) {
+            LD_UB4(src, src_stride, src0, src1, src2, src3);
+            src += (4 * src_stride);
+
+            ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+            dst += (4 * dst_stride);
+        }
+    }
+}
+
+void ff_get_pixels_16_msa(int16_t *av_restrict dest, const uint8_t *src,
+                          ptrdiff_t stride)
+{
+    copy_width16_msa(src, stride, (uint8_t *) dest, 16, 8);
+}
+
+void ff_get_pixels_8_msa(int16_t *av_restrict dest, const uint8_t *src,
+                         ptrdiff_t stride)
+{
+    copy_8bit_to_16bit_width8_msa(src, stride, dest, 8, 8);
+}
+
+void ff_diff_pixels_msa(int16_t *av_restrict block, const uint8_t *src1,
+                        const uint8_t *src2, int stride)
+{
+    diff_pixels_msa(block, src1, src2, stride);
+}
diff --git a/libavcodec/pixblockdsp.c b/libavcodec/pixblockdsp.c
index ebde68b..322e1dd 100644
--- a/libavcodec/pixblockdsp.c
+++ b/libavcodec/pixblockdsp.c
@@ -79,4 +79,6 @@ av_cold void ff_pixblockdsp_init(PixblockDSPContext *c, AVCodecContext *avctx)
         ff_pixblockdsp_init_ppc(c, avctx, high_bit_depth);
     if (ARCH_X86)
         ff_pixblockdsp_init_x86(c, avctx, high_bit_depth);
+    if (ARCH_MIPS)
+        ff_pixblockdsp_init_mips(c, avctx, high_bit_depth);
 }
diff --git a/libavcodec/pixblockdsp.h b/libavcodec/pixblockdsp.h
index d4b8590..79ed86c 100644
--- a/libavcodec/pixblockdsp.h
+++ b/libavcodec/pixblockdsp.h
@@ -42,5 +42,7 @@ void ff_pixblockdsp_init_ppc(PixblockDSPContext *c, AVCodecContext *avctx,
                              unsigned high_bit_depth);
 void ff_pixblockdsp_init_x86(PixblockDSPContext *c, AVCodecContext *avctx,
                              unsigned high_bit_depth);
+void ff_pixblockdsp_init_mips(PixblockDSPContext *c, AVCodecContext *avctx,
+                              unsigned high_bit_depth);
 
 #endif /* AVCODEC_PIXBLOCKDSP_H */
diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h
index 272a423..1738c52 100644
--- a/libavutil/mips/generic_macros_msa.h
+++ b/libavutil/mips/generic_macros_msa.h
@@ -1206,6 +1206,14 @@
 #define HSUB_UB2_UH(...) HSUB_UB2(v8u16, __VA_ARGS__)
 #define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
 
+#define HSUB_UB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3)  \
+{                                                                    \
+    HSUB_UB2(RTYPE, in0, in1, out0, out1);                           \
+    HSUB_UB2(RTYPE, in2, in3, out2, out3);                           \
+}
+#define HSUB_UB4_UH(...) HSUB_UB4(v8u16, __VA_ARGS__)
+#define HSUB_UB4_SH(...) HSUB_UB4(v8i16, __VA_ARGS__)
+
 /* Description : Insert specified word elements from input vectors to 1
                  destination vector
    Arguments   : Inputs  - in0, in1, in2, in3 (4 input vectors)



More information about the ffmpeg-cvslog mailing list