v210enc: Add SIMD optimised 8-bit and 10-bit encoders

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Vittorio Giovara <vittorio.giovara@gmail.com>

8 files changed, 389 insertions, 55 deletions

diff --git a/libavcodec/v210enc.c b/libavcodec/v210enc.c
index ef0d6ab2a4..cb887887ee 100644
--- a/libavcodec/v210enc.c
+++ b/libavcodec/v210enc.c
@@ -24,84 +24,188 @@
 #include "avcodec.h"
 #include "bytestream.h"
 #include "internal.h"
+#include "v210enc.h"
+
+#define CLIP(v) av_clip(v, 4, 1019)
+#define CLIP8(v) av_clip(v, 1, 254)
+
+#define WRITE_PIXELS(a, b, c)           \
+    do {                                \
+        val  =  CLIP(*a++);             \
+        val |= (CLIP(*b++) << 10) |     \
+               (CLIP(*c++) << 20);      \
+        AV_WL32(dst, val);              \
+        dst += 4;                       \
+    } while (0)
+
+#define WRITE_PIXELS8(a, b, c)          \
+    do {                                \
+        val  = (CLIP8(*a++) << 2);      \
+        val |= (CLIP8(*b++) << 12) |    \
+               (CLIP8(*c++) << 22);     \
+        AV_WL32(dst, val);              \
+        dst += 4;                       \
+    } while (0)
+
+static void v210_planar_pack_8_c(const uint8_t *y, const uint8_t *u,
+                                 const uint8_t *v, uint8_t *dst,
+                                 ptrdiff_t width)
+{
+    uint32_t val;
+    int i;
+
+    /* unroll this to match the assembly */
+    for (i = 0; i < width - 11; i += 12) {
+        WRITE_PIXELS8(u, y, v);
+        WRITE_PIXELS8(y, u, y);
+        WRITE_PIXELS8(v, y, u);
+        WRITE_PIXELS8(y, v, y);
+        WRITE_PIXELS8(u, y, v);
+        WRITE_PIXELS8(y, u, y);
+        WRITE_PIXELS8(v, y, u);
+        WRITE_PIXELS8(y, v, y);
+    }
+}
+
+static void v210_planar_pack_10_c(const uint16_t *y, const uint16_t *u,
+                                  const uint16_t *v, uint8_t *dst,
+                                  ptrdiff_t width)
+{
+    uint32_t val;
+    int i;
+
+    for (i = 0; i < width - 5; i += 6) {
+        WRITE_PIXELS(u, y, v);
+        WRITE_PIXELS(y, u, y);
+        WRITE_PIXELS(v, y, u);
+        WRITE_PIXELS(y, v, y);
+    }
+}
 
 static av_cold int encode_init(AVCodecContext *avctx)
 {
+    V210EncContext *s = avctx->priv_data;
+
     if (avctx->width & 1) {
         av_log(avctx, AV_LOG_ERROR, "v210 needs even width\n");
         return AVERROR(EINVAL);
     }
 
-    if (avctx->bits_per_raw_sample != 10)
-        av_log(avctx, AV_LOG_WARNING, "bits per raw sample: %d != 10-bit\n",
-               avctx->bits_per_raw_sample);
-
     avctx->coded_frame = av_frame_alloc();
     if (!avctx->coded_frame)
         return AVERROR(ENOMEM);
 
     avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
 
+    s->pack_line_8  = v210_planar_pack_8_c;
+    s->pack_line_10 = v210_planar_pack_10_c;
+
+    if (ARCH_X86)
+        ff_v210enc_init_x86(s);
+
     return 0;
 }
 
 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
                         const AVFrame *pic, int *got_packet)
 {
+    V210EncContext *s = avctx->priv_data;
     int aligned_width = ((avctx->width + 47) / 48) * 48;
     int stride = aligned_width * 8 / 3;
     int line_padding = stride - ((avctx->width * 8 + 11) / 12) * 4;
     int h, w, ret;
-    const uint16_t *y = (const uint16_t*)pic->data[0];
-    const uint16_t *u = (const uint16_t*)pic->data[1];
-    const uint16_t *v = (const uint16_t*)pic->data[2];
-    PutByteContext p;
+    uint8_t *dst;
 
-    if ((ret = ff_alloc_packet(pkt, avctx->height * stride)) < 0) {
+    ret = ff_alloc_packet(pkt, avctx->height * stride);
+    if (ret < 0) {
         av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
         return ret;
     }
-
-    bytestream2_init_writer(&p, pkt->data, pkt->size);
-
-#define CLIP(v) av_clip(v, 4, 1019)
-
-#define WRITE_PIXELS(a, b, c)           \
-    do {                                \
-        val =   CLIP(*a++);             \
-        val |= (CLIP(*b++) << 10) |     \
-               (CLIP(*c++) << 20);      \
-        bytestream2_put_le32u(&p, val); \
-    } while (0)
-
-    for (h = 0; h < avctx->height; h++) {
-        uint32_t val;
-        for (w = 0; w < avctx->width - 5; w += 6) {
-            WRITE_PIXELS(u, y, v);
-            WRITE_PIXELS(y, u, y);
-            WRITE_PIXELS(v, y, u);
-            WRITE_PIXELS(y, v, y);
+    dst = pkt->data;
+
+    if (pic->format == AV_PIX_FMT_YUV422P10) {
+        const uint16_t *y = (const uint16_t *)pic->data[0];
+        const uint16_t *u = (const uint16_t *)pic->data[1];
+        const uint16_t *v = (const uint16_t *)pic->data[2];
+        for (h = 0; h < avctx->height; h++) {
+            uint32_t val;
+            w = (avctx->width / 6) * 6;
+            s->pack_line_10(y, u, v, dst, w);
+
+            y += w;
+            u += w >> 1;
+            v += w >> 1;
+            dst += (w / 6) * 16;
+            if (w < avctx->width - 1) {
+                WRITE_PIXELS(u, y, v);
+
+                val = CLIP(*y++);
+                if (w == avctx->width - 2) {
+                    AV_WL32(dst, val);
+                    dst += 4;
+                }
+            }
+            if (w < avctx->width - 3) {
+                val |= (CLIP(*u++) << 10) | (CLIP(*y++) << 20);
+                AV_WL32(dst, val);
+                dst += 4;
+
+                val = CLIP(*v++) | (CLIP(*y++) << 10);
+                AV_WL32(dst, val);
+                dst += 4;
+            }
+
+            memset(dst, 0, line_padding);
+            dst += line_padding;
+            y += pic->linesize[0] / 2 - avctx->width;
+            u += pic->linesize[1] / 2 - avctx->width / 2;
+            v += pic->linesize[2] / 2 - avctx->width / 2;
         }
-        if (w < avctx->width - 1) {
-            WRITE_PIXELS(u, y, v);
-
-            val = CLIP(*y++);
-            if (w == avctx->width - 2)
-                bytestream2_put_le32u(&p, val);
+    } else if(pic->format == AV_PIX_FMT_YUV422P) {
+        const uint8_t *y = pic->data[0];
+        const uint8_t *u = pic->data[1];
+        const uint8_t *v = pic->data[2];
+        for (h = 0; h < avctx->height; h++) {
+            uint32_t val;
+            w = (avctx->width / 12) * 12;
+            s->pack_line_8(y, u, v, dst, w);
+
+            y += w;
+            u += w >> 1;
+            v += w >> 1;
+            dst += (w / 12) * 32;
+
+            for (; w < avctx->width - 5; w += 6) {
+                WRITE_PIXELS8(u, y, v);
+                WRITE_PIXELS8(y, u, y);
+                WRITE_PIXELS8(v, y, u);
+                WRITE_PIXELS8(y, v, y);
+            }
+            if (w < avctx->width - 1) {
+                WRITE_PIXELS8(u, y, v);
+
+                val = CLIP8(*y++) << 2;
+                if (w == avctx->width - 2) {
+                    AV_WL32(dst, val);
+                    dst += 4;
+                }
+            }
+            if (w < avctx->width - 3) {
+                val |= (CLIP8(*u++) << 12) | (CLIP8(*y++) << 22);
+                AV_WL32(dst, val);
+                dst += 4;
+
+                val = (CLIP8(*v++) << 2) | (CLIP8(*y++) << 12);
+                AV_WL32(dst, val);
+                dst += 4;
+            }
+            memset(dst, 0, line_padding);
+            dst += line_padding;
+
+            y += pic->linesize[0] - avctx->width;
+            u += pic->linesize[1] - avctx->width / 2;
+            v += pic->linesize[2] - avctx->width / 2;
         }
-        if (w < avctx->width - 3) {
-            val |= (CLIP(*u++) << 10) | (CLIP(*y++) << 20);
-            bytestream2_put_le32u(&p, val);
-
-            val = CLIP(*v++) | (CLIP(*y++) << 10);
-            bytestream2_put_le32u(&p, val);
-        }
-
-        bytestream2_set_buffer(&p, 0, line_padding);
-
-        y += pic->linesize[0] / 2 - avctx->width;
-        u += pic->linesize[1] / 2 - avctx->width / 2;
-        v += pic->linesize[2] / 2 - avctx->width / 2;
     }
 
     pkt->flags |= AV_PKT_FLAG_KEY;
@@ -121,8 +225,9 @@ AVCodec ff_v210_encoder = {
     .long_name      = NULL_IF_CONFIG_SMALL("Uncompressed 4:2:2 10-bit"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_V210,
+    .priv_data_size = sizeof(V210EncContext),
     .init           = encode_init,
     .encode2        = encode_frame,
     .close          = encode_close,
-    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV422P10, AV_PIX_FMT_NONE },
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV422P, AV_PIX_FMT_NONE },
 };
diff --git a/libavcodec/v210enc.h b/libavcodec/v210enc.h
new file mode 100644
index 0000000000..f950328d30
--- /dev/null
+++ b/libavcodec/v210enc.h
@@ -0,0 +1,35 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCOENC_V210ENC_H
+#define AVCOENC_V210ENC_H
+
+#include "libavutil/log.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixfmt.h"
+
+typedef struct {
+    void (*pack_line_8)(const uint8_t *y, const uint8_t *u,
+                        const uint8_t *v, uint8_t *dst, ptrdiff_t width);
+    void (*pack_line_10)(const uint16_t *y, const uint16_t *u,
+                         const uint16_t *v, uint8_t *dst, ptrdiff_t width);
+} V210EncContext;
+
+void ff_v210enc_init_x86(V210EncContext *s);
+
+#endif /* AVCOENC_V210ENC_H */
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 3cec000e1d..7caa256c43 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -47,6 +47,7 @@ OBJS-$(CONFIG_RV40_DECODER)            += x86/rv34dsp_init.o            \
                                           x86/rv40dsp_init.o
 OBJS-$(CONFIG_SVQ1_ENCODER)            += x86/svq1enc.o
 OBJS-$(CONFIG_TRUEHD_DECODER)          += x86/mlpdsp.o
+OBJS-$(CONFIG_V210_ENCODER)            += x86/v210enc_init.o
 OBJS-$(CONFIG_VC1_DECODER)             += x86/vc1dsp_init.o
 OBJS-$(CONFIG_VORBIS_DECODER)          += x86/vorbisdsp_init.o
 OBJS-$(CONFIG_VP6_DECODER)             += x86/vp6dsp_init.o
@@ -107,6 +108,7 @@ YASM-OBJS-$(CONFIG_PIXBLOCKDSP)        += x86/pixblockdsp.o
 YASM-OBJS-$(CONFIG_QPELDSP)            += x86/qpeldsp.o                 \
                                           x86/fpel.o                    \
                                           x86/qpel.o
+YASM-OBJS-$(CONFIG_V210_ENCODER)       += x86/v210enc.o
 YASM-OBJS-$(CONFIG_VIDEODSP)           += x86/videodsp.o
 YASM-OBJS-$(CONFIG_VP3DSP)             += x86/vp3dsp.o
 
diff --git a/libavcodec/x86/v210enc.asm b/libavcodec/x86/v210enc.asm
new file mode 100644
index 0000000000..595c8907b3
--- /dev/null
+++ b/libavcodec/x86/v210enc.asm
@@ -0,0 +1,145 @@
+;******************************************************************************
+;* V210 SIMD pack
+;* Copyright (c) 2014 Kieran Kunhya <kierank@obe.tv>
+;*
+;* This file is part of Libav.
+;*
+;* Libav 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.
+;*
+;* Libav 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 Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+v210_enc_min_10: times 8 dw 0x4
+v210_enc_max_10: times 8 dw 0x3fb
+
+v210_enc_luma_mult_10: dw 4,1,16,4,1,16,0,0
+v210_enc_luma_shuf_10: db -1,0,1,-1,2,3,4,5,-1,6,7,-1,8,9,10,11
+
+v210_enc_chroma_mult_10: dw 1,4,16,0,16,1,4,0
+v210_enc_chroma_shuf_10: db 0,1,8,9,-1,2,3,-1,10,11,4,5,-1,12,13,-1
+
+v210_enc_min_8: times 16 db 0x1
+v210_enc_max_8: times 16 db 0xfe
+
+v210_enc_luma_shuf_8: db 6,-1,7,-1,8,-1,9,-1,10,-1,11,-1,-1,-1,-1,-1
+v210_enc_luma_mult_8: dw 16,4,64,16,4,64,0,0
+
+v210_enc_chroma_shuf1_8: db 0,-1,1,-1,2,-1,3,-1,8,-1,9,-1,10,-1,11,-1
+v210_enc_chroma_shuf2_8: db 3,-1,4,-1,5,-1,7,-1,11,-1,12,-1,13,-1,15,-1
+
+v210_enc_chroma_mult_8: dw 4,16,64,0,64,4,16,0
+
+SECTION .text
+
+%macro v210_planar_pack_10 0
+
+; v210_planar_pack_10(const uint16_t *y, const uint16_t *u, const uint16_t *v, uint8_t *dst, ptrdiff_t width)
+cglobal v210_planar_pack_10, 5, 5, 4, y, u, v, dst, width
+    lea     r0, [yq+2*widthq]
+    add     uq, widthq
+    add     vq, widthq
+    neg     widthq
+
+    mova    m2, [v210_enc_min_10]
+    mova    m3, [v210_enc_max_10]
+
+.loop
+    movu    m0, [yq+2*widthq]
+    CLIPW   m0, m2, m3
+
+    movq    m1, [uq+widthq]
+    movhps  m1, [vq+widthq]
+    CLIPW   m1, m2, m3
+
+    pmullw  m0, [v210_enc_luma_mult_10]
+    pshufb  m0, [v210_enc_luma_shuf_10]
+
+    pmullw  m1, [v210_enc_chroma_mult_10]
+    pshufb  m1, [v210_enc_chroma_shuf_10]
+
+    por     m0, m1
+
+    movu    [dstq], m0
+
+    add     dstq, mmsize
+    add     widthq, 6
+    jl .loop
+
+    RET
+%endmacro
+
+INIT_XMM ssse3
+v210_planar_pack_10
+
+%macro v210_planar_pack_8 0
+
+; v210_planar_pack_8(const uint8_t *y, const uint8_t *u, const uint8_t *v, uint8_t *dst, ptrdiff_t width)
+cglobal v210_planar_pack_8, 5, 5, 7, y, u, v, dst, width
+    add     yq, widthq
+    shr     widthq, 1
+    add     uq, widthq
+    add     vq, widthq
+    neg     widthq
+
+    mova    m4, [v210_enc_min_8]
+    mova    m5, [v210_enc_max_8]
+    pxor    m6, m6
+
+.loop
+    movu    m1, [yq+2*widthq]
+    CLIPUB  m1, m4, m5
+
+    punpcklbw m0, m1, m6
+    ; can't unpack high bytes in the same way because we process
+    ; only six bytes at a time
+    pshufb  m1, [v210_enc_luma_shuf_8]
+
+    pmullw  m0, [v210_enc_luma_mult_8]
+    pmullw  m1, [v210_enc_luma_mult_8]
+    pshufb  m0, [v210_enc_luma_shuf_10]
+    pshufb  m1, [v210_enc_luma_shuf_10]
+
+    movq    m3, [uq+widthq]
+    movhps  m3, [vq+widthq]
+    CLIPUB  m3, m4, m5
+
+    ; shuffle and multiply to get the same packing as in 10-bit
+    pshufb  m2, m3, [v210_enc_chroma_shuf1_8]
+    pshufb  m3, [v210_enc_chroma_shuf2_8]
+
+    pmullw  m2, [v210_enc_chroma_mult_8]
+    pmullw  m3, [v210_enc_chroma_mult_8]
+    pshufb  m2, [v210_enc_chroma_shuf_10]
+    pshufb  m3, [v210_enc_chroma_shuf_10]
+
+    por     m0, m2
+    por     m1, m3
+
+    movu    [dstq], m0
+    movu    [dstq+mmsize], m1
+
+    add     dstq, 2*mmsize
+    add     widthq, 6
+    jl .loop
+
+    RET
+%endmacro
+
+INIT_XMM ssse3
+v210_planar_pack_8
+INIT_XMM avx
+v210_planar_pack_8
diff --git a/libavcodec/x86/v210enc_init.c b/libavcodec/x86/v210enc_init.c
new file mode 100644
index 0000000000..95b999bc05
--- /dev/null
+++ b/libavcodec/x86/v210enc_init.c
@@ -0,0 +1,42 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/x86/cpu.h"
+#include "libavcodec/v210enc.h"
+
+void ff_v210_planar_pack_8_ssse3(const uint8_t *y, const uint8_t *u,
+                                 const uint8_t *v, uint8_t *dst,
+                                 ptrdiff_t width);
+void ff_v210_planar_pack_8_avx(const uint8_t *y, const uint8_t *u,
+                               const uint8_t *v, uint8_t *dst, ptrdiff_t width);
+void ff_v210_planar_pack_10_ssse3(const uint16_t *y, const uint16_t *u,
+                                  const uint16_t *v, uint8_t *dst,
+                                  ptrdiff_t width);
+
+av_cold void ff_v210enc_init_x86(V210EncContext *s)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (EXTERNAL_SSSE3(cpu_flags)) {
+        s->pack_line_8 = ff_v210_planar_pack_8_ssse3;
+        s->pack_line_10 = ff_v210_planar_pack_10_ssse3;
+    }
+
+    if (EXTERNAL_AVX(cpu_flags))
+        s->pack_line_8 = ff_v210_planar_pack_8_avx;
+}
diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm
index 11779cf07f..9f64dd13e1 100644
--- a/libavutil/x86/x86util.asm
+++ b/libavutil/x86/x86util.asm
@@ -584,6 +584,11 @@
 %endif
 %endmacro
 
+%macro CLIPUB 3 ;(dst, min, max)
+    pmaxub %1, %2
+    pminub %1, %3
+%endmacro
+
 %macro CLIPW 3 ;(dst, min, max)
     pmaxsw %1, %2
     pminsw %1, %3
diff --git a/tests/ref/vsynth/vsynth1-v210 b/tests/ref/vsynth/vsynth1-v210
index defd2f658b..e2f4588c54 100644
--- a/tests/ref/vsynth/vsynth1-v210
+++ b/tests/ref/vsynth/vsynth1-v210
@@ -1,4 +1,4 @@
-054d7e903fb4d9e68700c56b93d45916 *tests/data/fate/vsynth1-v210.avi
+a38de84bfef53ed69204480c979dd6c4 *tests/data/fate/vsynth1-v210.avi
 14752448 tests/data/fate/vsynth1-v210.avi
-50973792d3f1abe04a51ee0121f077f2 *tests/data/fate/vsynth1-v210.out.rawvideo
-stddev:    1.85 PSNR: 42.78 MAXDIFF:   29 bytes:  7603200/  7603200
+2ba7f4ca302f3c4147860b9dfb12b6e4 *tests/data/fate/vsynth1-v210.out.rawvideo
+stddev:    1.84 PSNR: 42.81 MAXDIFF:   29 bytes:  7603200/  7603200
diff --git a/tests/ref/vsynth/vsynth2-v210 b/tests/ref/vsynth/vsynth2-v210
index 5e934de41b..245e2822a9 100644
--- a/tests/ref/vsynth/vsynth2-v210
+++ b/tests/ref/vsynth/vsynth2-v210
@@ -1,4 +1,4 @@
-87bb634932b3f5cacd4d08142798db17 *tests/data/fate/vsynth2-v210.avi
+3aae4d5b25a4d91e4812a297c6a9ef8a *tests/data/fate/vsynth2-v210.avi
 14752448 tests/data/fate/vsynth2-v210.avi
-8bb1c449e1a2a94fd0d98841c04246bb *tests/data/fate/vsynth2-v210.out.rawvideo
-stddev:    0.39 PSNR: 56.17 MAXDIFF:    9 bytes:  7603200/  7603200
+99e367a50da75c2c187230889bee8e2e *tests/data/fate/vsynth2-v210.out.rawvideo
+stddev:    0.40 PSNR: 56.06 MAXDIFF:    9 bytes:  7603200/  7603200