avcodec: loongson optimized mpegvideo dct unquantize with mmi
authorZhouXiaoyong <zhouxiaoyong@loongson.cn>
Thu, 9 Jul 2015 06:34:21 +0000 (14:34 +0800)
committerMichael Niedermayer <michael@niedermayer.cc>
Wed, 15 Jul 2015 13:51:07 +0000 (15:51 +0200)
Signed-off-by: ZhouXiaoyong <zhouxiaoyong@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavcodec/mips/Makefile
libavcodec/mips/mpegvideo_init_mips.c
libavcodec/mips/mpegvideo_mips.h [new file with mode: 0644]
libavcodec/mips/mpegvideo_mmi.c [new file with mode: 0644]

index c16c93a..9eb815a 100644 (file)
@@ -59,3 +59,4 @@ MSA-OBJS-$(CONFIG_ME_CMP)                 += mips/me_cmp_msa.o
 MMI-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_mmi.o
 MMI-OBJS-$(CONFIG_H264CHROMA)             += mips/h264chroma_mmi.o
 MMI-OBJS-$(CONFIG_H264PRED)               += mips/h264pred_mmi.o
+MMI-OBJS-$(CONFIG_MPEGVIDEO)              += mips/mpegvideo_mmi.o
index ee14b31..85a833c 100644 (file)
@@ -19,6 +19,7 @@
  */
 
 #include "h263dsp_mips.h"
+#include "mpegvideo_mips.h"
 
 #if HAVE_MSA
 static av_cold void dct_unquantize_init_msa(MpegEncContext *s)
@@ -29,9 +30,22 @@ static av_cold void dct_unquantize_init_msa(MpegEncContext *s)
 }
 #endif  // #if HAVE_MSA
 
+#if HAVE_MMI
+static av_cold void dct_unquantize_init_mmi(MpegEncContext *s)
+{
+    s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_mmi;
+    s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_mmi;
+    s->dct_unquantize_mpeg1_intra = ff_dct_unquantize_mpeg1_intra_mmi;
+    s->dct_unquantize_mpeg1_inter = ff_dct_unquantize_mpeg1_inter_mmi;
+}
+#endif /* HAVE_MMI */
+
 av_cold void ff_mpv_common_init_mips(MpegEncContext *s)
 {
 #if HAVE_MSA
     dct_unquantize_init_msa(s);
 #endif  // #if HAVE_MSA
+#if HAVE_MMI
+    dct_unquantize_init_mmi(s);
+#endif /* HAVE_MMI */
 }
diff --git a/libavcodec/mips/mpegvideo_mips.h b/libavcodec/mips/mpegvideo_mips.h
new file mode 100644 (file)
index 0000000..dbcea6a
--- /dev/null
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2015 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
+ *
+ * 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 MPEGVIDEO_MIPS_H
+#define MPEGVIDEO_MIPS_H
+
+#include "libavcodec/mpegvideo.h"
+
+void ff_dct_unquantize_h263_intra_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale);
+void ff_dct_unquantize_h263_inter_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale);
+void ff_dct_unquantize_mpeg1_intra_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale);
+void ff_dct_unquantize_mpeg1_inter_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale);
+
+#endif /* MPEGVIDEO_MIPS_H */
diff --git a/libavcodec/mips/mpegvideo_mmi.c b/libavcodec/mips/mpegvideo_mmi.c
new file mode 100644 (file)
index 0000000..b3d58bd
--- /dev/null
@@ -0,0 +1,303 @@
+/*
+ * Loongson SIMD optimized mpegvideo
+ *
+ * Copyright (c) 2015 Loongson Technology Corporation Limited
+ * Copyright (c) 2015 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
+ *                    Zhang Shuangshuang <zhangshuangshuang@ict.ac.cn>
+ *
+ * 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 "mpegvideo_mips.h"
+
+void ff_dct_unquantize_h263_intra_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale)
+{
+    int64_t level, qmul, qadd, nCoeffs;
+
+    qmul = qscale << 1;
+    assert(s->block_last_index[n]>=0 || s->h263_aic);
+
+    if (!s->h263_aic) {
+        if (n<4)
+            level = block[0] * s->y_dc_scale;
+        else
+            level = block[0] * s->c_dc_scale;
+        qadd = (qscale-1) | 1;
+    } else {
+        qadd = 0;
+        level = block[0];
+    }
+
+    if(s->ac_pred)
+        nCoeffs = 63;
+    else
+        nCoeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
+
+    __asm__ volatile (
+        "xor $f12, $f12, $f12           \r\n"
+        "lwc1 $f12, %1                  \n\r"
+        "xor $f10, $f10, $f10           \r\n"
+        "lwc1 $f10, %2                  \r\n"
+        "xor $f14, $f14, $f14           \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "packsswh $f10, $f10, $f10      \r\n"
+        "packsswh $f10, $f10, $f10      \r\n"
+        "psubh $f14, $f14, $f10         \r\n"
+        "xor $f8, $f8, $f8              \r\n"
+        ".p2align 4                     \r\n"
+        "1:                             \r\n"
+        "daddu $8, %0, %3               \r\n"
+        "gsldlc1 $f0, 7($8)             \r\n"
+        "gsldrc1 $f0, 0($8)             \r\n"
+        "gsldlc1 $f2, 15($8)            \r\n"
+        "gsldrc1 $f2, 8($8)             \r\n"
+        "mov.d $f4, $f0                 \r\n"
+        "mov.d $f6, $f2                 \r\n"
+        "pmullh $f0, $f0, $f12          \r\n"
+        "pmullh $f2, $f2, $f12          \r\n"
+        "pcmpgth $f4, $f4, $f8          \r\n"
+        "pcmpgth $f6, $f6, $f8          \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "paddh $f0, $f0, $f14           \r\n"
+        "paddh $f2, $f2, $f14           \r\n"
+        "xor $f4, $f4, $f0              \r\n"
+        "xor $f6, $f6, $f2              \r\n"
+        "pcmpeqh $f0, $f0, $f14         \r\n"
+        "pcmpeqh $f2, $f2, $f14         \r\n"
+        "pandn $f0, $f0, $f4            \r\n"
+        "pandn $f2, $f2, $f6            \r\n"
+        "gssdlc1 $f0, 7($8)             \r\n"
+        "gssdrc1 $f0, 0($8)             \r\n"
+        "gssdlc1 $f2, 15($8)            \r\n"
+        "gssdrc1 $f2, 8($8)             \r\n"
+        "addi %3, %3, 16                \r\n"
+        "blez %3, 1b                    \r\n"
+        ::"r"(block+nCoeffs),"m"(qmul),"m"(qadd),"r"(2*(-nCoeffs))
+        :"$8","memory"
+    );
+
+    block[0] = level;
+}
+
+void ff_dct_unquantize_h263_inter_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale)
+{
+    int64_t qmul, qadd, nCoeffs;
+
+    qmul = qscale << 1;
+    qadd = (qscale - 1) | 1;
+    assert(s->block_last_index[n]>=0 || s->h263_aic);
+    nCoeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
+
+    __asm__ volatile (
+        "xor $f12, $f12, $f12           \r\n"
+        "lwc1 $f12, %1                  \r\n"
+        "xor $f10, $f10, $f10           \r\n"
+        "lwc1 $f10, %2                  \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "xor $f14, $f14, $f14           \r\n"
+        "packsswh $f10, $f10, $f10      \r\n"
+        "packsswh $f10, $f10, $f10      \r\n"
+        "psubh $f14, $f14, $f10         \r\n"
+        "xor $f8, $f8, $f8              \r\n"
+        ".p2align 4                     \r\n"
+        "1:                             \r\n"
+        "daddu $8, %0, %3               \r\n"
+        "gsldlc1 $f0, 7($8)             \r\n"
+        "gsldrc1 $f0, 0($8)             \r\n"
+        "gsldlc1 $f2, 15($8)            \r\n"
+        "gsldrc1 $f2, 8($8)             \r\n"
+        "mov.d $f4, $f0                 \r\n"
+        "mov.d $f6, $f2                 \r\n"
+        "pmullh $f0, $f0, $f12          \r\n"
+        "pmullh $f2, $f2, $f12          \r\n"
+        "pcmpgth $f4, $f4, $f8          \r\n"
+        "pcmpgth $f6, $f6, $f8          \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "paddh $f0, $f0, $f14           \r\n"
+        "paddh $f2, $f2, $f14           \r\n"
+        "xor $f4, $f4, $f0              \r\n"
+        "xor $f6, $f6, $f2              \r\n"
+        "pcmpeqh $f0, $f0, $f14         \r\n"
+        "pcmpeqh $f2, $f2, $f14         \r\n"
+        "pandn $f0, $f0, $f4            \r\n"
+        "pandn $f2, $f2, $f6            \r\n"
+        "gssdlc1 $f0, 7($8)             \r\n"
+        "gssdrc1 $f0, 0($8)             \r\n"
+        "gssdlc1 $f2, 15($8)            \r\n"
+        "gssdrc1 $f2, 8($8)             \r\n"
+        "addi %3, %3, 16                \r\n"
+        "blez %3, 1b                    \r\n"
+        ::"r"(block+nCoeffs),"m"(qmul),"m"(qadd),"r"(2*(-nCoeffs))
+        : "$8","memory"
+    );
+}
+
+void ff_dct_unquantize_mpeg1_intra_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale)
+{
+    int64_t nCoeffs;
+    const uint16_t *quant_matrix;
+    int block0;
+
+    assert(s->block_last_index[n]>=0);
+    nCoeffs = s->intra_scantable.raster_end[s->block_last_index[n]] + 1;
+
+    if (n<4)
+        block0 = block[0] * s->y_dc_scale;
+    else
+        block0 = block[0] * s->c_dc_scale;
+
+    /* XXX: only mpeg1 */
+    quant_matrix = s->intra_matrix;
+
+    __asm__ volatile (
+        "pcmpeqh $f14, $f14, $f14       \r\n"
+        "dli $10, 15                    \r\n"
+        "dmtc1 $10, $f16                \r\n"
+        "xor $f12, $f12, $f12           \r\n"
+        "lwc1 $f12, %2                  \r\n"
+        "psrlh $f14, $f14, $f16         \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "or $8, %3, $0                  \r\n"
+        ".p2align 4                     \r\n"
+        "1:                             \r\n"
+        "gsldxc1 $f0, 0($8, %0)         \r\n"
+        "gsldxc1 $f2, 8($8, %0)         \r\n"
+        "mov.d $f16, $f0                \r\n"
+        "mov.d $f18, $f2                \r\n"
+        "gsldxc1 $f8, 0($8, %1)         \r\n"
+        "gsldxc1 $f10, 8($8, %1)        \r\n"
+        "pmullh $f8, $f8, $f12          \r\n"
+        "pmullh $f10, $f10, $f12        \r\n"
+        "xor $f4, $f4, $f4              \r\n"
+        "xor $f6, $f6, $f6              \r\n"
+        "pcmpgth $f4, $f4, $f0          \r\n"
+        "pcmpgth $f6, $f6, $f2          \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "psubh $f0, $f0, $f4            \r\n"
+        "psubh $f2, $f2, $f6            \r\n"
+        "pmullh $f0, $f0, $f8           \r\n"
+        "pmullh $f2, $f2, $f10          \r\n"
+        "xor $f8, $f8, $f8              \r\n"
+        "xor $f10, $f10, $f10           \r\n"
+        "pcmpeqh $f8, $f8, $f16         \r\n"
+        "pcmpeqh $f10, $f10, $f18       \r\n"
+        "dli $10, 3                     \r\n"
+        "dmtc1 $10, $f16                \r\n"
+        "psrah $f0, $f0, $f16           \r\n"
+        "psrah $f2, $f2, $f16           \r\n"
+        "psubh $f0, $f0, $f14           \r\n"
+        "psubh $f2, $f2, $f14           \r\n"
+        "or $f0, $f0, $f14              \r\n"
+        "or $f2, $f2, $f14              \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "psubh $f0, $f0, $f4            \r\n"
+        "psubh $f2, $f2, $f6            \r\n"
+        "pandn $f8, $f8, $f0            \r\n"
+        "pandn $f10, $f10, $f2          \r\n"
+        "gssdxc1 $f8, 0($8, %0)         \r\n"
+        "gssdxc1 $f10, 8($8, %0)        \r\n"
+        "addi $8, $8, 16                \r\n"
+        "bltz $8, 1b                    \r\n"
+        ::"r"(block+nCoeffs),"r"(quant_matrix+nCoeffs),"m"(qscale),
+          "g"(-2*nCoeffs)
+        : "$8","$10","memory"
+    );
+
+    block[0] = block0;
+}
+
+void ff_dct_unquantize_mpeg1_inter_mmi(MpegEncContext *s, int16_t *block,
+        int n, int qscale)
+{
+    int64_t nCoeffs;
+    const uint16_t *quant_matrix;
+
+    assert(s->block_last_index[n] >= 0);
+    nCoeffs = s->intra_scantable.raster_end[s->block_last_index[n]] + 1;
+    quant_matrix = s->inter_matrix;
+
+    __asm__ volatile (
+        "pcmpeqh $f14, $f14, $f14       \r\n"
+        "dli $10, 15                    \r\n"
+        "dmtc1 $10, $f16                \r\n"
+        "xor $f12, $f12, $f12           \r\n"
+        "lwc1 $f12, %2                  \r\n"
+        "psrlh $f14, $f14, $f16         \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "packsswh $f12, $f12, $f12      \r\n"
+        "or $8, %3, $0                  \r\n"
+        ".p2align 4                     \r\n"
+        "1:                             \r\n"
+        "gsldxc1 $f0, 0($8, %0)         \r\n"
+        "gsldxc1 $f2, 8($8, %0)         \r\n"
+        "mov.d $f16, $f0                \r\n"
+        "mov.d $f18, $f2                \r\n"
+        "gsldxc1 $f8, 0($8, %1)         \r\n"
+        "gsldxc1 $f10, 8($8, %1)        \r\n"
+        "pmullh $f8, $f8, $f12          \r\n"
+        "pmullh $f10, $f10, $f12        \r\n"
+        "xor $f4, $f4, $f4              \r\n"
+        "xor $f6, $f6, $f6              \r\n"
+        "pcmpgth $f4, $f4, $f0          \r\n"
+        "pcmpgth $f6, $f6, $f2          \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "psubh $f0, $f0, $f4            \r\n"
+        "psubh $f2, $f2, $f6            \r\n"
+        "paddh $f0, $f0, $f0            \r\n"
+        "paddh $f2, $f2, $f2            \r\n"
+        "paddh $f0, $f0, $f14           \r\n"
+        "paddh $f2, $f2, $f14           \r\n"
+        "pmullh $f0, $f0, $f8           \r\n"
+        "pmullh $f2, $f2, $f10          \r\n"
+        "xor $f8, $f8, $f8              \r\n"
+        "xor $f10, $f10, $f10           \r\n"
+        "pcmpeqh $f8, $f8, $f16         \r\n"
+        "pcmpeqh $f10, $f10, $f18       \r\n"
+        "dli $10, 4                     \r\n"
+        "dmtc1 $10, $f16                \r\n"
+        "psrah $f0, $f0, $f16           \r\n"
+        "psrah $f2, $f2, $f16           \r\n"
+        "psubh $f0, $f0, $f14           \r\n"
+        "psubh $f2, $f2, $f14           \r\n"
+        "or $f0, $f0, $f14              \r\n"
+        "or $f2, $f2, $f14              \r\n"
+        "xor $f0, $f0, $f4              \r\n"
+        "xor $f2, $f2, $f6              \r\n"
+        "psubh $f0, $f0, $f4            \r\n"
+        "psubh $f2, $f2, $f6            \r\n"
+        "pandn $f8, $f8, $f0            \r\n"
+        "pandn $f10, $f10, $f2          \r\n"
+        "gssdxc1 $f8, 0($8, %0)         \r\n"
+        "gssdxc1 $f10, 8($8, %0)        \r\n"
+        "addi $8, $8, 16                \r\n"
+        "bltz $8, 1b                    \r\n"
+        ::"r"(block+nCoeffs),"r"(quant_matrix+nCoeffs),"m"(qscale),
+          "g"(-2*nCoeffs)
+        :"$8","$10","memory"
+    );
+}