@scrypted/libav 1.0.180 → 1.0.181

package/openh264/codec/decoder/core/mips/dct_mmi.c

@@ -0,0 +1,786 @@
+/*!
+ * \copy
+ *     Copyright (c)  2009-2018, Cisco Systems
+ *     All rights reserved.
+ *
+ *     Redistribution and use in source and binary forms, with or without
+ *     modification, are permitted provided that the following conditions
+ *     are met:
+ *
+ *        * Redistributions of source code must retain the above copyright
+ *          notice, this list of conditions and the following disclaimer.
+ *
+ *        * Redistributions in binary form must reproduce the above copyright
+ *          notice, this list of conditions and the following disclaimer in
+ *          the documentation and/or other materials provided with the
+ *          distribution.
+ *
+ *     THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *     "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *     LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ *     FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ *     COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ *     INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ *     BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ *     LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ *     CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ *     LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ *     ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ *     POSSIBILITY OF SUCH DAMAGE.
+ *
+ *
+ * \file    dct_mmi.c
+ *
+ * \brief   Loongson optimization
+ *
+ * \date    17/07/2018 Created
+ *
+ *************************************************************************************
+ */
+#include <stdint.h>
+#include "asmdefs_mmi.h"
+
+#define LOAD_2_LEFT_AND_ADD                                   \
+  PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t" \
+  "lbu        $9, -0x1(%[pPred])                        \n\t" \
+  PTR_ADDU   "$8, $8, $9                                \n\t" \
+  PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t" \
+  "lbu        $9, -0x1(%[pPred])                        \n\t" \
+  PTR_ADDU   "$8, $8, $9                                \n\t"
+
+unsigned char mmi_dc_0x80[16] __attribute__((aligned(16))) = {
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80};
+
+short mmi_wd_0x02[8] __attribute__((aligned(16))) = {2, 2, 2, 2, 2, 2, 2, 2};
+short mmi_plane_inc_minus[8]__attribute__((aligned(16))) = {-7, -6, -5, -4, -3, -2, -1, 0};
+short mmi_plane_inc[8]__attribute__((aligned(16))) = {1, 2, 3, 4, 5, 6, 7, 8};
+short mmi_plane_dec[8]__attribute__((aligned(16))) = {8, 7, 6, 5, 4, 3, 2, 1};
+
+short mmi_plane_inc_c[4]__attribute__((aligned(16))) = {1, 2, 3, 4};
+short mmi_plane_dec_c[4]__attribute__((aligned(16))) = {4, 3, 2, 1};
+short mmi_plane_mul_b_c[8]__attribute__((aligned(16))) = {-3, -2, -1, 0, 1, 2, 3, 4};
+
+unsigned char mmi_01bytes[16]__attribute__((aligned(16))) = {1, 1, 1, 1, 1, 1, 1, 1,
+                                                             1, 1, 1, 1, 1, 1, 1, 1};
+
+void IdctResAddPred_mmi(uint8_t *pPred, const int32_t kiStride, int16_t *pRs) {
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    "dli        $8, 0x1                                   \n\t"
+    "gsldxc1    $f0, 0x0(%[pRs], $0)                      \n\t"
+    "gsldxc1    $f2, 0x8(%[pRs], $0)                      \n\t"
+    "gsldxc1    $f4, 0x10(%[pRs], $0)                     \n\t"
+    "gsldxc1    $f6, 0x18(%[pRs], $0)                     \n\t"
+    "dmtc1      $8, $f14                                  \n\t"
+
+    MMI_Trans4x4H_SINGLE($f0, $f2, $f4, $f6, $f8)
+    MMI_IDCT_SINGLE($f2, $f4, $f6, $f8, $f0, $f12, $f14)
+    MMI_Trans4x4H_SINGLE($f2, $f6, $f0, $f8, $f4)
+    MMI_IDCT_SINGLE($f6, $f0, $f8, $f4, $f2, $f12, $f14)
+
+    "dli        $8, 0x20                                  \n\t"
+    "xor        $f14, $f14, $f14                          \n\t"
+    "dmtc1      $8, $f12                                  \n\t"
+    "pshufh     $f12, $f12, $f14                          \n\t"
+    "dli        $8, 0x6                                   \n\t"
+    "dmtc1      $8, $f16                                  \n\t"
+
+    MMI_StoreDiff4P_SINGLE($f6, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    MMI_StoreDiff4P_SINGLE($f8, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    MMI_StoreDiff4P_SINGLE($f2, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    MMI_StoreDiff4P_SINGLE($f4, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+    : [pPred]"+&r"((unsigned char *)pPred)
+    : [pRs]"r"((unsigned char *)pRs), [kiStride]"r"((int)kiStride)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+      "$f14", "$f16"
+  );
+}
+
+void WelsDecoderI16x16LumaPredDc_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile(
+    ".set       arch=loongson3a                           \n\t"
+    "dli        $8, 0x5                                   \n\t"
+    "gsldxc1    $f10, 0x0(%[mmi_01bytes], $0)             \n\t"
+    "dmtc1      $8, $f8                                   \n\t"
+
+    "move       $10, %[pPred]                             \n\t"
+    PTR_SUBU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gslqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    "xor        $f4, $f4, $f4                             \n\t"
+    "pasubub    $f0, $f0, $f4                             \n\t"
+    "pasubub    $f2, $f2, $f4                             \n\t"
+    "biadd      $f0, $f0                                  \n\t"
+    "biadd      $f2, $f2                                  \n\t"
+    "paddh      $f0, $f0, $f2                             \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $8, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+    LOAD_2_LEFT_AND_ADD
+
+    PTR_ADDIU  "$8, $8, 0x10                              \n\t"
+    "dmtc1      $8, $f4                                   \n\t"
+    "paddh      $f0, $f0, $f4                             \n\t"
+    "psrlw      $f0, $f0, $f8                             \n\t"
+    "pmuluw     $f0, $f0, $f10                            \n\t"
+    "punpcklwd  $f0, $f0, $f0                             \n\t"
+    "mov.d      $f2, $f0                                  \n\t"
+
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssqc1     $f2, $f0, 0x0($10)                        \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride),
+      [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10"
+  );
+}
+
+void WelsDecoderI16x16LumaPredPlane_mmi(uint8_t *pPred, const int32_t kiStride) {
+  BACKUP_REG;
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    "move       $10, %[pPred]                             \n\t"
+    PTR_ADDIU  "%[pPred], %[pPred], -0x1                  \n\t"
+    PTR_SUBU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+
+    "gsldlc1    $f0, 0x7(%[pPred])                        \n\t"
+    "xor        $f28, $f28, $f28                          \n\t"
+    "gsldrc1    $f0, 0x0(%[pPred])                        \n\t"
+    "gslqc1     $f22, $f20, 0x0(%[mmi_plane_dec])         \n\t"
+    "punpckhbh  $f2, $f0, $f28                            \n\t"
+    "punpcklbh  $f0, $f0, $f28                            \n\t"
+    "pmullh     $f0, $f0, $f20                            \n\t"
+    "gsldlc1    $f4, 0x10(%[pPred])                       \n\t"
+    "pmullh     $f2, $f2, $f22                            \n\t"
+    "gsldrc1    $f4, 0x9(%[pPred])                        \n\t"
+    "gslqc1     $f26, $f24, 0x0(%[mmi_plane_inc])         \n\t"
+    "punpckhbh  $f6, $f4, $f28                            \n\t"
+    "punpcklbh  $f4, $f4, $f28                            \n\t"
+    "pmullh     $f4, $f4, $f24                            \n\t"
+    "pmullh     $f6, $f6, $f26                            \n\t"
+    "psubh      $f4, $f4, $f0                             \n\t"
+    "psubh      $f6, $f6, $f2                             \n\t"
+
+    SUMH_HORIZON($f4, $f6, $f0, $f2, $f8)
+    "dmfc1      $8, $f4                                   \n\t"
+    "seh        $8, $8                                    \n\t"
+    "mul        $8, $8, 0x5                               \n\t"
+    PTR_ADDIU  "$8, $8, 0x20                              \n\t"
+    "sra        $8, $8, 0x6                               \n\t"
+    MMI_Copy8Times($f4, $f6, $f28, $8)
+
+    "lbu        $9, 0x10(%[pPred])                        \n\t"
+    PTR_ADDIU  "%[pPred], %[pPred], -0x3                  \n\t"
+    LOAD_COLUMN($f0, $f2, $f8, $f10, $f12, $f14, $f16, $f18, %[pPred],
+                %[kiStride], $11)
+
+    PTR_ADDIU  "%[pPred], %[pPred], 0x3                   \n\t"
+    "dsll       $11, %[kiStride], 0x3                     \n\t"
+    PTR_ADDU   "$11, $11, %[pPred]                        \n\t"
+    "lbu        $8, 0x0($11)                              \n\t"
+    PTR_ADDU   "$9, $9, $8                                \n\t"
+    "dsll       $9, $9, 0x4                               \n\t"
+
+    PTR_ADDIU  "%[pPred], %[pPred], -0x3                  \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    LOAD_COLUMN($f28, $f30, $f8, $f10, $f12, $f14, $f16, $f18, %[pPred],
+                %[kiStride], $11)
+
+    "xor        $f16, $f16, $f16                          \n\t"
+    "punpcklbh  $f0, $f2, $f16                            \n\t"
+    "punpckhbh  $f2, $f2, $f16                            \n\t"
+    "pmullh     $f0, $f0, $f20                            \n\t"
+    "pmullh     $f2, $f2, $f22                            \n\t"
+    "punpcklbh  $f28, $f30, $f16                          \n\t"
+    "punpckhbh  $f30, $f30, $f16                          \n\t"
+    "pmullh     $f28, $f28, $f24                          \n\t"
+    "pmullh     $f30, $f30, $f26                          \n\t"
+    "psubh      $f28, $f28, $f0                           \n\t"
+    "psubh      $f30, $f30, $f2                           \n\t"
+
+    "xor        $f8, $f8, $f8                             \n\t"
+
+    SUMH_HORIZON($f28, $f30, $f0, $f2, $f8)
+    "dmfc1      $8, $f28                                  \n\t"
+    "seh        $8, $8                                    \n\t"
+
+    "mul        $8, $8, 0x5                               \n\t"
+    PTR_ADDIU  "$8, $8, 0x20                              \n\t"
+    "sra        $8, $8, 0x6                               \n\t"
+    MMI_Copy8Times($f16, $f18, $f8, $8)
+
+    "move       %[pPred], $10                             \n\t"
+    PTR_ADDIU  "$9, $9, 0x10                              \n\t"
+    "mul        $8, $8, -0x7                              \n\t"
+    PTR_ADDU   "$9, $9, $8                                \n\t"
+    MMI_Copy8Times($f0, $f2, $f8, $9)
+
+    "xor        $8, $8, $8                                \n\t"
+    "gslqc1     $f22, $f20, 0x0(%[mmi_plane_inc_minus])   \n\t"
+
+    "dli        $11, 0x5                                  \n\t"
+    "dmtc1      $11, $f30                                 \n\t"
+    "1:                                                   \n\t"
+    "pmullh     $f8, $f4, $f20                            \n\t"
+    "pmullh     $f10, $f6, $f22                           \n\t"
+    "paddh      $f8, $f8, $f0                             \n\t"
+    "paddh      $f10, $f10, $f2                           \n\t"
+    "psrah      $f8, $f8, $f30                            \n\t"
+    "psrah      $f10, $f10, $f30                          \n\t"
+    "pmullh     $f12, $f4, $f24                           \n\t"
+    "pmullh     $f14, $f6, $f26                           \n\t"
+    "paddh      $f12, $f12, $f0                           \n\t"
+    "paddh      $f14, $f14, $f2                           \n\t"
+    "psrah      $f12, $f12, $f30                          \n\t"
+    "psrah      $f14, $f14, $f30                          \n\t"
+    "packushb   $f8, $f8, $f10                            \n\t"
+    "packushb   $f10, $f12, $f14                          \n\t"
+    "gssqc1     $f10, $f8, 0x0(%[pPred])                  \n\t"
+    "paddh      $f0, $f0, $f16                            \n\t"
+    "paddh      $f2, $f2, $f18                            \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    PTR_ADDIU  "$8, $8, 0x1                               \n\t"
+    PTR_ADDIU  "$11, $8, -0x10                            \n\t"
+    "bnez       $11, 1b                                   \n\t"
+    "nop                                                  \n\t"
+    : [pPred]"+&r"((unsigned char *)pPred)
+    : [kiStride]"r"((int)kiStride), [mmi_plane_inc_minus]"r"(mmi_plane_inc_minus),
+      [mmi_plane_inc]"r"(mmi_plane_inc), [mmi_plane_dec]"r"(mmi_plane_dec)
+    : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8",
+      "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26",
+      "$f28", "$f30"
+  );
+  RECOVER_REG;
+}
+
+#define COPY_16_TIMES(r0, f0, f2, f4, f6, f8)                 \
+  "gslqc1     "#f2", "#f0", -0x10("#r0")                \n\t" \
+  "dsrl       "#f0", "#f2", "#f4"                       \n\t" \
+  "pmuluw     "#f0", "#f0", "#f6"                       \n\t" \
+  "punpcklwd  "#f0", "#f0", "#f0"                       \n\t" \
+  "mov.d      "#f2", "#f0"                              \n\t"
+
+#define MMI_PRED_H_16X16_TWO_LINE_DEC                         \
+  PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t" \
+  COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)            \
+  "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t" \
+  PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t" \
+  COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)            \
+  "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+
+void WelsDecoderI16x16LumaPredH_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    "dli        $8, 56                                    \n\t"
+    "dmtc1      $8, $f4                                   \n\t"
+    "gsldxc1    $f6, 0x0(%[mmi_01bytes], $0)              \n\t"
+    "xor        $f8, $f8, $f8                             \n\t"
+
+    COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    MMI_PRED_H_16X16_TWO_LINE_DEC
+    : [pPred]"+&r"((unsigned char *)pPred)
+    : [kiStride]"r"((int)kiStride),
+      [mmi_01bytes]"r"((unsigned char *)mmi_01bytes)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8"
+  );
+}
+
+void WelsDecoderI16x16LumaPredV_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile(
+    ".set       arch=loongson3a                           \n\t"
+    PTR_SUBU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gslqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride)
+    : "memory", "$f0", "$f2"
+  );
+}
+
+void WelsDecoderI16x16LumaPredDcTop_mmi(uint8_t *pPred, const int32_t kiStride) {
+  BACKUP_REG;
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    PTR_SUBU   "$8, %[pPred], %[kiStride]                 \n\t"
+    "gslqc1     $f2, $f0, 0x0($8)                         \n\t"
+    "xor        $f28, $f28, $f28                          \n\t"
+    "pasubub    $f0, $f0, $f28                            \n\t"
+    "pasubub    $f2, $f2, $f28                            \n\t"
+    "biadd      $f0, $f0                                  \n\t"
+    "biadd      $f2, $f2                                  \n\t"
+    "paddh      $f0, $f0, $f2                             \n\t"
+    "dmfc1      $8, $f0                                   \n\t"
+
+    PTR_ADDIU  "$8, $8, 0x8                               \n\t"
+    "dsra       $8, $8, 0x4                               \n\t"
+    MMI_Copy16Times($f4, $f6, $f28, $8)
+    "mov.d      $f0, $f4                                  \n\t"
+    "mov.d      $f2, $f6                                  \n\t"
+
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    : [pPred]"+&r"((unsigned char *)pPred)
+    : [kiStride]"r"((int)kiStride)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+  );
+  RECOVER_REG;
+}
+
+void WelsDecoderI16x16LumaPredDcNA_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile(
+    ".set       arch=loongson3a                           \n\t"
+    "gslqc1     $f2, $f0, 0x0(%[mmi_dc_0x80])             \n\t"
+    "mov.d      $f4, $f0                                  \n\t"
+    "mov.d      $f6, $f2                                  \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f2, $f0, 0x0(%[pPred])                   \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssqc1     $f6, $f4, 0x0(%[pPred])                   \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride), [mmi_dc_0x80] "r"(mmi_dc_0x80)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+  );
+}
+
+void WelsDecoderIChromaPredPlane_mmi(uint8_t *pPred, const int32_t kiStride) {
+  BACKUP_REG;
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    "move       $10, %[pPred]                             \n\t"
+    PTR_ADDIU  "%[pPred], %[pPred], -0x1                  \n\t"
+    PTR_SUBU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+
+    "gsldlc1    $f0, 0x7(%[pPred])                        \n\t"
+    "xor        $f28, $f28, $f28                          \n\t"
+    "gsldrc1    $f0, 0x0(%[pPred])                        \n\t"
+    "gsldxc1    $f20, 0x0(%[mmi_plane_dec_c], $0)         \n\t"
+    "punpcklbh  $f0, $f0, $f28                            \n\t"
+    "gsldlc1    $f4, 0xc(%[pPred])                        \n\t"
+    "pmullh     $f0, $f0, $f20                            \n\t"
+    "gsldrc1    $f4, 0x5(%[pPred])                        \n\t"
+    "gsldxc1    $f24, 0x0(%[mmi_plane_inc_c], $0)         \n\t"
+    "punpcklbh  $f4, $f4, $f28                            \n\t"
+    "pmullh     $f4, $f4, $f24                            \n\t"
+    "psubh      $f4, $f4, $f0                             \n\t"
+
+    "xor        $f6, $f6, $f6                             \n\t"
+    "xor        $f8, $f8, $f8                             \n\t"
+    SUMH_HORIZON($f4, $f6, $f0, $f2, $f8)
+    "dmfc1      $8, $f4                                   \n\t"
+    "seh        $8, $8                                    \n\t"
+    "mul        $8, $8, 0x11                              \n\t"
+    PTR_ADDIU  "$8, $8, 0x10                              \n\t"
+    "sra        $8, $8, 0x5                               \n\t"
+    MMI_Copy8Times($f4, $f6, $f8, $8)
+
+    "lbu        $9, 0x8(%[pPred])                         \n\t"
+    PTR_ADDIU  "%[pPred], %[pPred], -0x3                  \n\t"
+    LOAD_COLUMN_C($f0, $f8, $f12, $f16, %[pPred], %[kiStride], $11)
+
+    PTR_ADDIU  "%[pPred], %[pPred], 0x3                   \n\t"
+    "dsll       $11, %[kiStride], 0x2                     \n\t"
+    PTR_ADDU   "$11, $11, %[pPred]                        \n\t"
+    "lbu        $8, 0x0($11)                              \n\t"
+    PTR_ADDU   "$9, $9, $8                                \n\t"
+    "dsll       $9, $9, 0x4                               \n\t"
+
+    PTR_ADDIU  "%[pPred], %[pPred], -0x3                  \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    LOAD_COLUMN_C($f28, $f8, $f12, $f16, %[pPred], %[kiStride], $11)
+    "xor        $f16, $f16, $f16                          \n\t"
+    "punpckhbh  $f0, $f0, $f16                            \n\t"
+    "pmullh     $f0, $f0, $f20                            \n\t"
+    "punpckhbh  $f28, $f28, $f16                          \n\t"
+    "pmullh     $f28, $f28, $f24                          \n\t"
+    "psubh      $f28, $f28, $f0                           \n\t"
+
+    "xor        $f30, $f30, $f30                          \n\t"
+    "xor        $f8, $f8, $f8                             \n\t"
+    SUMH_HORIZON($f28, $f30, $f0, $f2, $f8)
+    "dmfc1      $8, $f28                                  \n\t"
+    "seh        $8, $8                                    \n\t"
+
+    "mul        $8, $8, 0x11                              \n\t"
+    PTR_ADDIU  "$8, $8, 0x10                              \n\t"
+    "sra        $8, $8, 0x5                               \n\t"
+    MMI_Copy8Times($f16, $f18, $f8, $8)
+
+    "move       %[pPred], $10                             \n\t"
+    PTR_ADDIU  "$9, $9, 0x10                              \n\t"
+    "mul        $8, $8, -0x3                              \n\t"
+    PTR_ADDU   "$9, $9, $8                                \n\t"
+    MMI_Copy8Times($f0, $f2, $f8, $9)
+
+    "xor        $8, $8, $8                                \n\t"
+    "gslqc1     $f22, $f20, 0x0(%[mmi_plane_mul_b_c])     \n\t"
+
+    "dli        $11, 0x5                                  \n\t"
+    "dmtc1      $11, $f30                                 \n\t"
+    "1:                                                   \n\t"
+    "pmullh     $f8, $f4, $f20                            \n\t"
+    "pmullh     $f10, $f6, $f22                           \n\t"
+    "paddh      $f8, $f8, $f0                             \n\t"
+    "paddh      $f10, $f10, $f2                           \n\t"
+    "psrah      $f8, $f8, $f30                            \n\t"
+    "psrah      $f10, $f10, $f30                          \n\t"
+    "packushb   $f8, $f8, $f10                            \n\t"
+    "gssdxc1    $f8, 0x0(%[pPred], $0)                    \n\t"
+    "paddh      $f0, $f0, $f16                            \n\t"
+    "paddh      $f2, $f2, $f18                            \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    PTR_ADDIU  "$8, $8, 0x1                               \n\t"
+    PTR_ADDIU  "$11, $8, -0x8                             \n\t"
+    "bnez       $11, 1b                                   \n\t"
+    "nop                                                  \n\t"
+    : [pPred]"+&r"((unsigned char *)pPred)
+    : [kiStride]"r"((int)kiStride), [mmi_plane_mul_b_c]"r"(mmi_plane_mul_b_c),
+      [mmi_plane_inc_c]"r"(mmi_plane_inc_c), [mmi_plane_dec_c]"r"(mmi_plane_dec_c)
+    : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10",
+      "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+  );
+  RECOVER_REG;
+}
+
+void WelsDecoderIChromaPredDc_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile(
+    ".set       arch=loongson3a                           \n\t"
+    "move       $10, %[pPred]                             \n\t"
+
+    PTR_SUBU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gsldxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $8, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    "dmtc1      $8, $f2                                   \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $8, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "lbu        $9, -0x1(%[pPred])                        \n\t"
+    PTR_ADDU   "$8, $8, $9                                \n\t"
+    "dmtc1      $8, $f4                                   \n\t"
+
+    "xor        $f8, $f8, $f8                             \n\t"
+    "punpcklwd  $f6, $f0, $f8                             \n\t"
+    "punpckhwd  $f0, $f0, $f8                             \n\t"
+    "pasubub    $f0, $f0, $f8                             \n\t"
+    "pasubub    $f6, $f6, $f8                             \n\t"
+    "biadd      $f0, $f0                                  \n\t"
+    "biadd      $f6, $f6                                  \n\t"
+
+    "paddd      $f6, $f6, $f2                             \n\t"
+    "paddd      $f2, $f4, $f0                             \n\t"
+
+    "dli        $8, 0x2                                   \n\t"
+    "dmtc1      $8, $f8                                   \n\t"
+    "gsldxc1    $f12, 0x0(%[mmi_01bytes], $0)             \n\t"
+    "dli        $8, 0x3                                   \n\t"
+    "dmtc1      $8, $f10                                  \n\t"
+
+    "paddd      $f0, $f0, $f8                             \n\t"
+    "dsrl       $f0, $f0, $f8                             \n\t"
+
+    "paddd      $f4, $f4, $f8                             \n\t"
+    "dsrl       $f4, $f4, $f8                             \n\t"
+
+    "paddd      $f6, $f6, $f8                             \n\t"
+    "paddd      $f6, $f6, $f8                             \n\t"
+    "dsrl       $f6, $f6, $f10                            \n\t"
+
+    "paddd      $f2, $f2, $f8                             \n\t"
+    "paddd      $f2, $f2, $f8                             \n\t"
+    "dsrl       $f2, $f2, $f10                            \n\t"
+
+    "dli        $8, 0x20                                  \n\t"
+    "dmtc1      $8, $f8                                   \n\t"
+    "pmuluw     $f0, $f0, $f12                            \n\t"
+    "pmuluw     $f6, $f6, $f12                            \n\t"
+    "dsll       $f0, $f0, $f8                             \n\t"
+    "xor        $f0, $f0, $f6                             \n\t"
+
+    "pmuluw     $f4, $f4, $f12                            \n\t"
+    "pmuluw     $f2, $f2, $f12                            \n\t"
+    "dsll       $f2, $f2, $f8                             \n\t"
+    "xor        $f2, $f2, $f4                             \n\t"
+
+    "gssdxc1    $f0, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f0, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f0, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f0, 0x0($10, $0)                         \n\t"
+
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f2, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f2, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f2, 0x0($10, $0)                         \n\t"
+    PTR_ADDU   "$10, $10, %[kiStride]                     \n\t"
+    "gssdxc1    $f2, 0x0($10, $0)                         \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride),
+      [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+    : "memory", "$8", "$9", "$10", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10",
+      "$f12"
+  );
+}
+
+void WelsDecoderIChromaPredDcTop_mmi(uint8_t *pPred, const int32_t kiStride) {
+  BACKUP_REG;
+  __asm__ volatile (
+    ".set       arch=loongson3a                           \n\t"
+    "dli        $8, 0x4e                                  \n\t"
+    "dmtc1      $8, $f16                                  \n\t"
+    "dli        $8, 0xb1                                  \n\t"
+    "dmtc1      $8, $f18                                  \n\t"
+    "dli        $8, 0x2                                   \n\t"
+    "dmtc1      $8, $f20                                  \n\t"
+    PTR_SUBU   "$8, %[pPred], %[kiStride]                 \n\t"
+    "gsldxc1    $f0, 0x0($8, $0)                          \n\t"
+    "xor        $f28, $f28, $f28                          \n\t"
+    "punpckhbh  $f2, $f0, $f28                            \n\t"
+    "punpcklbh  $f0, $f0, $f28                            \n\t"
+    "pshufh     $f4, $f0, $f16                            \n\t"
+    "pshufh     $f6, $f2, $f16                            \n\t"
+    "paddh      $f0, $f0, $f4                             \n\t"
+    "paddh      $f2, $f2, $f6                             \n\t"
+
+    "pshufh     $f8, $f0, $f18                            \n\t"
+    "pshufh     $f14, $f2, $f18                           \n\t"
+    "paddh      $f2, $f2, $f14                            \n\t"
+    "paddh      $f0, $f0, $f8                             \n\t"
+
+    "gslqc1     $f26, $f24, 0x0(%[mmi_wd_0x02])           \n\t"
+    "paddh      $f0, $f0, $f24                            \n\t"
+    "paddh      $f2, $f2, $f26                            \n\t"
+    "psrah      $f0, $f0, $f20                            \n\t"
+    "psrah      $f2, $f2, $f20                            \n\t"
+    "packushb   $f0, $f0, $f2                             \n\t"
+
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    PTR_ADDU   "%[pPred], %[pPred], %[kiStride]           \n\t"
+    "gssdxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride), [mmi_wd_0x02] "r"((short *)mmi_wd_0x02)
+    : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+  );
+  RECOVER_REG;
+}
+
+void WelsDecoderI4x4LumaPredH_mmi(uint8_t *pPred, const int32_t kiStride) {
+  __asm__ volatile(
+    ".set       arch=loongson3a                           \n\t"
+    "gsldxc1    $f8, 0x0(%[mmi_01bytes], $0)              \n\t"
+    "lbu        $8, -0x1(%[pPred])                        \n\t"
+    "dmtc1      $8, $f0                                   \n\t"
+    "pmuluw     $f0, $f0, $f8                             \n\t"
+
+    PTR_ADDU   "$9, %[pPred], %[kiStride]                 \n\t"
+    "lbu        $8, -0x1($9)                              \n\t"
+    "dmtc1      $8, $f2                                   \n\t"
+    "pmuluw     $f2, $f2, $f8                             \n\t"
+
+    PTR_ADDU   "$10, $9, %[kiStride]                      \n\t"
+    "lbu        $8, -0x1($10)                             \n\t"
+    "dmtc1      $8, $f4                                   \n\t"
+    "pmuluw     $f4, $f4, $f8                             \n\t"
+
+    PTR_ADDU   "$11, $10, %[kiStride]                     \n\t"
+    "lbu        $8, -0x1($11)                             \n\t"
+    "dmtc1      $8, $f6                                   \n\t"
+    "pmuluw     $f6, $f6, $f8                             \n\t"
+
+    "gsswxc1    $f0, 0x0(%[pPred], $0)                    \n\t"
+    "gsswxc1    $f2, 0x0($9, $0)                          \n\t"
+    "gsswxc1    $f4, 0x0($10, $0)                         \n\t"
+    "gsswxc1    $f6, 0x0($11, $0)                         \n\t"
+    : [pPred] "+&r"((unsigned char *)pPred)
+    : [kiStride] "r"((int)kiStride),
+      [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+    : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8"
+  );
+}