epeg 1.0

data/epeg/vendor/libjpeg-turbo-2.0.4/simd/mips/jsimd_dspr2.S

@@ -0,0 +1,4479 @@
+/*
+ * MIPS DSPr2 optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2013-2014, MIPS Technologies, Inc., California.
+ *                          All Rights Reserved.
+ * Authors:  Teodora Novkovic <teodora.novkovic@imgtec.com>
+ *           Darko Laus       <darko.laus@imgtec.com>
+ * Copyright (C) 2015, D. R. Commander.  All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+#include "jsimd_dspr2_asm.h"
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_c_null_convert_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = input_buf
+ * a2     = output_buf
+ * a3     = output_row
+ * 16(sp) = num_rows
+ * 20(sp) = cinfo->num_components
+ *
+ * Null conversion for compression
+ */
+    SAVE_REGS_ON_STACK 8, s0, s1
+
+    lw          t9, 24(sp)      // t9 = num_rows
+    lw          s0, 28(sp)      // s0 = cinfo->num_components
+    andi        t0, a0, 3       // t0 = cinfo->image_width & 3
+    beqz        t0, 4f          // no residual
+     nop
+0:
+    addiu       t9, t9, -1
+    bltz        t9, 7f
+     li         t1, 0
+1:
+    sll         t3, t1, 2
+    lwx         t5, t3(a2)      // t5 = outptr = output_buf[ci]
+    lw          t2, 0(a1)       // t2 = inptr = *input_buf
+    sll         t4, a3, 2
+    lwx         t5, t4(t5)      // t5 = outptr = output_buf[ci][output_row]
+    addu        t2, t2, t1
+    addu        s1, t5, a0
+    addu        t6, t5, t0
+2:
+    lbu         t3, 0(t2)
+    addiu       t5, t5, 1
+    sb          t3, -1(t5)
+    bne         t6, t5, 2b
+     addu       t2, t2, s0
+3:
+    lbu         t3, 0(t2)
+    addu        t4, t2, s0
+    addu        t7, t4, s0
+    addu        t8, t7, s0
+    addu        t2, t8, s0
+    lbu         t4, 0(t4)
+    lbu         t7, 0(t7)
+    lbu         t8, 0(t8)
+    addiu       t5, t5, 4
+    sb          t3, -4(t5)
+    sb          t4, -3(t5)
+    sb          t7, -2(t5)
+    bne         s1, t5, 3b
+     sb         t8, -1(t5)
+    addiu       t1, t1, 1
+    bne         t1, s0, 1b
+     nop
+    addiu       a1, a1, 4
+    bgez        t9, 0b
+     addiu      a3, a3, 1
+    b           7f
+     nop
+4:
+    addiu       t9, t9, -1
+    bltz        t9, 7f
+     li         t1, 0
+5:
+    sll         t3, t1, 2
+    lwx         t5, t3(a2)      // t5 = outptr = output_buf[ci]
+    lw          t2, 0(a1)       // t2 = inptr = *input_buf
+    sll         t4, a3, 2
+    lwx         t5, t4(t5)      // t5 = outptr = output_buf[ci][output_row]
+    addu        t2, t2, t1
+    addu        s1, t5, a0
+    addu        t6, t5, t0
+6:
+    lbu         t3, 0(t2)
+    addu        t4, t2, s0
+    addu        t7, t4, s0
+    addu        t8, t7, s0
+    addu        t2, t8, s0
+    lbu         t4, 0(t4)
+    lbu         t7, 0(t7)
+    lbu         t8, 0(t8)
+    addiu       t5, t5, 4
+    sb          t3, -4(t5)
+    sb          t4, -3(t5)
+    sb          t7, -2(t5)
+    bne         s1, t5, 6b
+     sb         t8, -1(t5)
+    addiu       t1, t1, 1
+    bne         t1, s0, 5b
+     nop
+    addiu       a1, a1, 4
+    bgez        t9, 4b
+     addiu      a3, a3, 1
+7:
+    RESTORE_REGS_FROM_STACK 8, s0, s1
+
+    j           ra
+     nop
+
+END(jsimd_c_null_convert_dspr2)
+
+
+/*****************************************************************************/
+/*
+ * jsimd_extrgb_ycc_convert_dspr2
+ * jsimd_extbgr_ycc_convert_dspr2
+ * jsimd_extrgbx_ycc_convert_dspr2
+ * jsimd_extbgrx_ycc_convert_dspr2
+ * jsimd_extxbgr_ycc_convert_dspr2
+ * jsimd_extxrgb_ycc_convert_dspr2
+ *
+ * Colorspace conversion RGB -> YCbCr
+ */
+
+.macro GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2  colorid, pixel_size, \
+                                             r_offs, g_offs, b_offs
+
+.macro DO_RGB_TO_YCC  r, g, b, inptr
+    lbu         \r, \r_offs(\inptr)
+    lbu         \g, \g_offs(\inptr)
+    lbu         \b, \b_offs(\inptr)
+    addiu       \inptr, \pixel_size
+.endm
+
+LEAF_DSPR2(jsimd_\colorid\()_ycc_convert_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = input_buf
+ * a2     = output_buf
+ * a3     = output_row
+ * 16(sp) = num_rows
+ */
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    lw          t7, 48(sp)      // t7 = num_rows
+    li          s0, 0x4c8b      // FIX(0.29900)
+    li          s1, 0x9646      // FIX(0.58700)
+    li          s2, 0x1d2f      // FIX(0.11400)
+    li          s3, 0xffffd4cd  // -FIX(0.16874)
+    li          s4, 0xffffab33  // -FIX(0.33126)
+    li          s5, 0x8000      // FIX(0.50000)
+    li          s6, 0xffff94d1  // -FIX(0.41869)
+    li          s7, 0xffffeb2f  // -FIX(0.08131)
+    li          t8, 0x807fff    // CBCR_OFFSET + ONE_HALF-1
+
+0:
+    addiu       t7, -1          // --num_rows
+    lw          t6, 0(a1)       // t6 = input_buf[0]
+    lw          t0, 0(a2)
+    lw          t1, 4(a2)
+    lw          t2, 8(a2)
+    sll         t3, a3, 2
+    lwx         t0, t3(t0)      // t0 = output_buf[0][output_row]
+    lwx         t1, t3(t1)      // t1 = output_buf[1][output_row]
+    lwx         t2, t3(t2)      // t2 = output_buf[2][output_row]
+
+    addu        t9, t2, a0      // t9 = end address
+    addiu       a3, 1
+
+1:
+    DO_RGB_TO_YCC t3, t4, t5, t6
+
+    mtlo        s5, $ac0
+    mtlo        t8, $ac1
+    mtlo        t8, $ac2
+    maddu       $ac0, s2, t5
+    maddu       $ac1, s5, t5
+    maddu       $ac2, s5, t3
+    maddu       $ac0, s0, t3
+    maddu       $ac1, s3, t3
+    maddu       $ac2, s6, t4
+    maddu       $ac0, s1, t4
+    maddu       $ac1, s4, t4
+    maddu       $ac2, s7, t5
+    extr.w      t3, $ac0, 16
+    extr.w      t4, $ac1, 16
+    extr.w      t5, $ac2, 16
+    sb          t3, 0(t0)
+    sb          t4, 0(t1)
+    sb          t5, 0(t2)
+    addiu       t0, 1
+    addiu       t2, 1
+    bne         t2, t9, 1b
+     addiu      t1, 1
+    bgtz        t7, 0b
+     addiu      a1, 4
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+END(jsimd_\colorid\()_ycc_convert_dspr2)
+
+.purgem DO_RGB_TO_YCC
+
+.endm
+
+/*-------------------------------------id -- pix R  G  B */
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extrgb,  3, 0, 1, 2
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extbgr,  3, 2, 1, 0
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extrgbx, 4, 0, 1, 2
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extbgrx, 4, 2, 1, 0
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extxbgr, 4, 3, 2, 1
+GENERATE_JSIMD_RGB_YCC_CONVERT_DSPR2 extxrgb, 4, 1, 2, 3
+
+
+/*****************************************************************************/
+/*
+ * jsimd_ycc_extrgb_convert_dspr2
+ * jsimd_ycc_extbgr_convert_dspr2
+ * jsimd_ycc_extrgbx_convert_dspr2
+ * jsimd_ycc_extbgrx_convert_dspr2
+ * jsimd_ycc_extxbgr_convert_dspr2
+ * jsimd_ycc_extxrgb_convert_dspr2
+ *
+ * Colorspace conversion YCbCr -> RGB
+ */
+
+.macro GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2  colorid, pixel_size, \
+                                             r_offs, g_offs, b_offs, a_offs
+
+.macro STORE_YCC_TO_RGB  scratch0 scratch1 scratch2 outptr
+    sb          \scratch0, \r_offs(\outptr)
+    sb          \scratch1, \g_offs(\outptr)
+    sb          \scratch2, \b_offs(\outptr)
+.if (\pixel_size == 4)
+    li          t0, 0xFF
+    sb          t0, \a_offs(\outptr)
+.endif
+    addiu       \outptr, \pixel_size
+.endm
+
+LEAF_DSPR2(jsimd_ycc_\colorid\()_convert_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = input_buf
+ * a2     = input_row
+ * a3     = output_buf
+ * 16(sp) = num_rows
+ */
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    lw          s1, 48(sp)
+    li          t3, 0x8000
+    li          t4, 0x166e9     // FIX(1.40200)
+    li          t5, 0x1c5a2     // FIX(1.77200)
+    li          t6, 0xffff492e  // -FIX(0.71414)
+    li          t7, 0xffffa7e6  // -FIX(0.34414)
+    repl.ph     t8, 128
+
+0:
+    lw          s0, 0(a3)
+    lw          t0, 0(a1)
+    lw          t1, 4(a1)
+    lw          t2, 8(a1)
+    sll         s5, a2, 2
+    addiu       s1, -1
+    lwx         s2, s5(t0)
+    lwx         s3, s5(t1)
+    lwx         s4, s5(t2)
+    addu        t9, s2, a0
+    addiu       a2, 1
+
+1:
+    lbu         s7, 0(s4)       // cr
+    lbu         s6, 0(s3)       // cb
+    lbu         s5, 0(s2)       // y
+    addiu       s2, 1
+    addiu       s4, 1
+    addiu       s7, -128
+    addiu       s6, -128
+    mul         t2, t7, s6
+    mul         t0, t6, s7      // Crgtab[cr]
+    sll         s7, 15
+    mulq_rs.w   t1, t4, s7      // Crrtab[cr]
+    sll         s6, 15
+    addu        t2, t3          // Cbgtab[cb]
+    addu        t2, t0
+
+    mulq_rs.w   t0, t5, s6      // Cbbtab[cb]
+    sra         t2, 16
+    addu        t1, s5
+    addu        t2, s5          // add y
+    ins         t2, t1, 16, 16
+    subu.ph     t2, t2, t8
+    addu        t0, s5
+    shll_s.ph   t2, t2, 8
+    subu        t0, 128
+    shra.ph     t2, t2, 8
+    shll_s.w    t0, t0, 24
+    addu.ph     t2, t2, t8      // clip & store
+    sra         t0, t0, 24
+    sra         t1, t2, 16
+    addiu       t0, 128
+
+    STORE_YCC_TO_RGB t1, t2, t0, s0
+
+    bne         s2, t9, 1b
+     addiu      s3, 1
+    bgtz        s1, 0b
+     addiu      a3, 4
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+END(jsimd_ycc_\colorid\()_convert_dspr2)
+
+.purgem STORE_YCC_TO_RGB
+
+.endm
+
+/*-------------------------------------id -- pix R  G  B  A */
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extrgb,  3, 0, 1, 2, 3
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extbgr,  3, 2, 1, 0, 3
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extrgbx, 4, 0, 1, 2, 3
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extbgrx, 4, 2, 1, 0, 3
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extxbgr, 4, 3, 2, 1, 0
+GENERATE_JSIMD_YCC_RGB_CONVERT_DSPR2 extxrgb, 4, 1, 2, 3, 0
+
+
+/*****************************************************************************/
+/*
+ * jsimd_extrgb_gray_convert_dspr2
+ * jsimd_extbgr_gray_convert_dspr2
+ * jsimd_extrgbx_gray_convert_dspr2
+ * jsimd_extbgrx_gray_convert_dspr2
+ * jsimd_extxbgr_gray_convert_dspr2
+ * jsimd_extxrgb_gray_convert_dspr2
+ *
+ * Colorspace conversion RGB -> GRAY
+ */
+
+.macro GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2  colorid, pixel_size, \
+                                              r_offs, g_offs, b_offs
+
+.macro DO_RGB_TO_GRAY  r, g, b, inptr
+    lbu         \r, \r_offs(\inptr)
+    lbu         \g, \g_offs(\inptr)
+    lbu         \b, \b_offs(\inptr)
+    addiu       \inptr, \pixel_size
+.endm
+
+LEAF_DSPR2(jsimd_\colorid\()_gray_convert_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = input_buf
+ * a2     = output_buf
+ * a3     = output_row
+ * 16(sp) = num_rows
+ */
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    li          s0, 0x4c8b      // s0 = FIX(0.29900)
+    li          s1, 0x9646      // s1 = FIX(0.58700)
+    li          s2, 0x1d2f      // s2 = FIX(0.11400)
+    li          s7, 0x8000      // s7 = FIX(0.50000)
+    lw          s6, 48(sp)
+    andi        t7, a0, 3
+
+0:
+    addiu       s6, -1          // s6 = num_rows
+    lw          t0, 0(a1)
+    lw          t1, 0(a2)
+    sll         t3, a3, 2
+    lwx         t1, t3(t1)
+    addiu       a3, 1
+    addu        t9, t1, a0
+    subu        t8, t9, t7
+    beq         t1, t8, 2f
+     nop
+
+1:
+    DO_RGB_TO_GRAY t3, t4, t5, t0
+    DO_RGB_TO_GRAY s3, s4, s5, t0
+
+    mtlo        s7, $ac0
+    maddu       $ac0, s2, t5
+    maddu       $ac0, s1, t4
+    maddu       $ac0, s0, t3
+    mtlo        s7, $ac1
+    maddu       $ac1, s2, s5
+    maddu       $ac1, s1, s4
+    maddu       $ac1, s0, s3
+    extr.w      t6, $ac0, 16
+
+    DO_RGB_TO_GRAY t3, t4, t5, t0
+    DO_RGB_TO_GRAY s3, s4, s5, t0
+
+    mtlo        s7, $ac0
+    maddu       $ac0, s2, t5
+    maddu       $ac0, s1, t4
+    extr.w      t2, $ac1, 16
+    maddu       $ac0, s0, t3
+    mtlo        s7, $ac1
+    maddu       $ac1, s2, s5
+    maddu       $ac1, s1, s4
+    maddu       $ac1, s0, s3
+    extr.w      t5, $ac0, 16
+    sb          t6, 0(t1)
+    sb          t2, 1(t1)
+    extr.w      t3, $ac1, 16
+    addiu       t1, 4
+    sb          t5, -2(t1)
+    sb          t3, -1(t1)
+    bne         t1, t8, 1b
+     nop
+
+2:
+    beqz        t7, 4f
+     nop
+
+3:
+    DO_RGB_TO_GRAY t3, t4, t5, t0
+
+    mtlo        s7, $ac0
+    maddu       $ac0, s2, t5
+    maddu       $ac0, s1, t4
+    maddu       $ac0, s0, t3
+    extr.w      t6, $ac0, 16
+    sb          t6, 0(t1)
+    addiu       t1, 1
+    bne         t1, t9, 3b
+     nop
+
+4:
+    bgtz        s6, 0b
+     addiu      a1, 4
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+END(jsimd_\colorid\()_gray_convert_dspr2)
+
+.purgem DO_RGB_TO_GRAY
+
+.endm
+
+/*-------------------------------------id --  pix R  G  B */
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extrgb,  3, 0, 1, 2
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extbgr,  3, 2, 1, 0
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extrgbx, 4, 0, 1, 2
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extbgrx, 4, 2, 1, 0
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extxbgr, 4, 3, 2, 1
+GENERATE_JSIMD_RGB_GRAY_CONVERT_DSPR2 extxrgb, 4, 1, 2, 3
+
+
+/*****************************************************************************/
+/*
+ * jsimd_h2v2_merged_upsample_dspr2
+ * jsimd_h2v2_extrgb_merged_upsample_dspr2
+ * jsimd_h2v2_extrgbx_merged_upsample_dspr2
+ * jsimd_h2v2_extbgr_merged_upsample_dspr2
+ * jsimd_h2v2_extbgrx_merged_upsample_dspr2
+ * jsimd_h2v2_extxbgr_merged_upsample_dspr2
+ * jsimd_h2v2_extxrgb_merged_upsample_dspr2
+ *
+ * Merged h2v2 upsample routines
+ */
+.macro GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2  colorid, pixel_size, \
+                                            r1_offs, g1_offs, \
+                                            b1_offs, a1_offs, \
+                                            r2_offs, g2_offs, \
+                                            b2_offs, a2_offs
+
+.macro STORE_H2V2_2_PIXELS  scratch0 scratch1 scratch2 scratch3 scratch4 \
+                            scratch5 outptr
+    sb          \scratch0, \r1_offs(\outptr)
+    sb          \scratch1, \g1_offs(\outptr)
+    sb          \scratch2, \b1_offs(\outptr)
+    sb          \scratch3, \r2_offs(\outptr)
+    sb          \scratch4, \g2_offs(\outptr)
+    sb          \scratch5, \b2_offs(\outptr)
+.if (\pixel_size == 8)
+    li          \scratch0, 0xFF
+    sb          \scratch0, \a1_offs(\outptr)
+    sb          \scratch0, \a2_offs(\outptr)
+.endif
+    addiu       \outptr, \pixel_size
+.endm
+
+.macro STORE_H2V2_1_PIXEL  scratch0 scratch1 scratch2 outptr
+    sb          \scratch0, \r1_offs(\outptr)
+    sb          \scratch1, \g1_offs(\outptr)
+    sb          \scratch2, \b1_offs(\outptr)
+
+.if (\pixel_size == 8)
+    li          t0, 0xFF
+    sb          t0, \a1_offs(\outptr)
+.endif
+.endm
+
+LEAF_DSPR2(jsimd_h2v2_\colorid\()_merged_upsample_dspr2)
+/*
+ * a0     = cinfo->output_width
+ * a1     = input_buf
+ * a2     = in_row_group_ctr
+ * a3     = output_buf
+ * 16(sp) = cinfo->sample_range_limit
+ */
+    SAVE_REGS_ON_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, ra
+
+    lw          t9, 56(sp)      // cinfo->sample_range_limit
+    lw          v0, 0(a1)
+    lw          v1, 4(a1)
+    lw          t0, 8(a1)
+    sll         t1, a2, 3
+    addiu       t2, t1, 4
+    sll         t3, a2, 2
+    lw          t4, 0(a3)       // t4 = output_buf[0]
+    lwx         t1, t1(v0)      // t1 = input_buf[0][in_row_group_ctr*2]
+    lwx         t2, t2(v0)      // t2 = input_buf[0][in_row_group_ctr*2 + 1]
+    lwx         t5, t3(v1)      // t5 = input_buf[1][in_row_group_ctr]
+    lwx         t6, t3(t0)      // t6 = input_buf[2][in_row_group_ctr]
+    lw          t7, 4(a3)       // t7 = output_buf[1]
+    li          s1, 0xe6ea
+    addiu       t8, s1, 0x7fff    // t8 = 0x166e9 [FIX(1.40200)]
+    addiu       s0, t8, 0x5eb9    // s0 = 0x1c5a2 [FIX(1.77200)]
+    addiu       s1, zero, 0xa7e6  // s4 = 0xffffa7e6 [-FIX(0.34414)]
+    xori        s2, s1, 0xeec8    // s3 = 0xffff492e [-FIX(0.71414)]
+    srl         t3, a0, 1
+    blez        t3, 2f
+     addu       t0, t5, t3      // t0 = end address
+ 1:
+    lbu         t3, 0(t5)
+    lbu         s3, 0(t6)
+    addiu       t5, t5, 1
+    addiu       t3, t3, -128    // (cb - 128)
+    addiu       s3, s3, -128    // (cr - 128)
+    mult        $ac1, s1, t3
+    madd        $ac1, s2, s3
+    sll         s3, s3, 15
+    sll         t3, t3, 15
+    mulq_rs.w   s4, t8, s3      // s4 = (C1 * cr + ONE_HALF)>> SCALEBITS
+    extr_r.w    s5, $ac1, 16
+    mulq_rs.w   s6, s0, t3      // s6 = (C2 * cb + ONE_HALF)>> SCALEBITS
+    lbu         v0, 0(t1)
+    addiu       t6, t6, 1
+    addiu       t1, t1, 2
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         AT, 0(t3)
+    lbu         s7, 0(s3)
+    lbu         ra, 0(v1)
+    lbu         v0, -1(t1)
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         t3, 0(t3)
+    lbu         s3, 0(s3)
+    lbu         v1, 0(v1)
+    lbu         v0, 0(t2)
+
+    STORE_H2V2_2_PIXELS AT, s7, ra, t3, s3, v1, t4
+
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         AT, 0(t3)
+    lbu         s7, 0(s3)
+    lbu         ra, 0(v1)
+    lbu         v0, 1(t2)
+    addiu       t2, t2, 2
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         t3, 0(t3)
+    lbu         s3, 0(s3)
+    lbu         v1, 0(v1)
+
+    STORE_H2V2_2_PIXELS AT, s7, ra, t3, s3, v1, t7
+
+    bne         t0, t5, 1b
+     nop
+2:
+    andi        t0, a0, 1
+    beqz        t0, 4f
+     lbu        t3, 0(t5)
+    lbu         s3, 0(t6)
+    addiu       t3, t3, -128    // (cb - 128)
+    addiu       s3, s3, -128    // (cr - 128)
+    mult        $ac1, s1, t3
+    madd        $ac1, s2, s3
+    sll         s3, s3, 15
+    sll         t3, t3, 15
+    lbu         v0, 0(t1)
+    extr_r.w    s5, $ac1, 16
+    mulq_rs.w   s4, t8, s3      // s4 = (C1 * cr + ONE_HALF)>> SCALEBITS
+    mulq_rs.w   s6, s0, t3      // s6 = (C2 * cb + ONE_HALF)>> SCALEBITS
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         t3, 0(t3)
+    lbu         s3, 0(s3)
+    lbu         v1, 0(v1)
+    lbu         v0, 0(t2)
+
+    STORE_H2V2_1_PIXEL t3, s3, v1, t4
+
+    addu        t3, v0, s4      // y+cred
+    addu        s3, v0, s5      // y+cgreen
+    addu        v1, v0, s6      // y+cblue
+    addu        t3, t9, t3      // y+cred
+    addu        s3, t9, s3      // y+cgreen
+    addu        v1, t9, v1      // y+cblue
+    lbu         t3, 0(t3)
+    lbu         s3, 0(s3)
+    lbu         v1, 0(v1)
+
+    STORE_H2V2_1_PIXEL t3, s3, v1, t7
+4:
+    RESTORE_REGS_FROM_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, ra
+
+    j           ra
+     nop
+
+END(jsimd_h2v2_\colorid\()_merged_upsample_dspr2)
+
+.purgem STORE_H2V2_1_PIXEL
+.purgem STORE_H2V2_2_PIXELS
+.endm
+
+/*------------------------------------id -- pix R1 G1 B1 A1 R2 G2 B2 A2 */
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extrgb,  6, 0, 1, 2, 6, 3, 4, 5, 6
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extbgr,  6, 2, 1, 0, 3, 5, 4, 3, 6
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extrgbx, 8, 0, 1, 2, 3, 4, 5, 6, 7
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extbgrx, 8, 2, 1, 0, 3, 6, 5, 4, 7
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extxbgr, 8, 3, 2, 1, 0, 7, 6, 5, 4
+GENERATE_H2V2_MERGED_UPSAMPLE_DSPR2 extxrgb, 8, 1, 2, 3, 0, 5, 6, 7, 4
+
+
+/*****************************************************************************/
+/*
+ * jsimd_h2v1_merged_upsample_dspr2
+ * jsimd_h2v1_extrgb_merged_upsample_dspr2
+ * jsimd_h2v1_extrgbx_merged_upsample_dspr2
+ * jsimd_h2v1_extbgr_merged_upsample_dspr2
+ * jsimd_h2v1_extbgrx_merged_upsample_dspr2
+ * jsimd_h2v1_extxbgr_merged_upsample_dspr2
+ * jsimd_h2v1_extxrgb_merged_upsample_dspr2
+ *
+ * Merged h2v1 upsample routines
+ */
+
+.macro GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2  colorid, pixel_size, \
+                                            r1_offs, g1_offs, \
+                                            b1_offs, a1_offs, \
+                                            r2_offs, g2_offs, \
+                                            b2_offs, a2_offs
+
+.macro STORE_H2V1_2_PIXELS  scratch0 scratch1 scratch2 scratch3 scratch4 \
+                            scratch5 outptr
+    sb          \scratch0, \r1_offs(\outptr)
+    sb          \scratch1, \g1_offs(\outptr)
+    sb          \scratch2, \b1_offs(\outptr)
+    sb          \scratch3, \r2_offs(\outptr)
+    sb          \scratch4, \g2_offs(\outptr)
+    sb          \scratch5, \b2_offs(\outptr)
+.if (\pixel_size == 8)
+    li          t0, 0xFF
+    sb          t0, \a1_offs(\outptr)
+    sb          t0, \a2_offs(\outptr)
+.endif
+    addiu       \outptr, \pixel_size
+.endm
+
+.macro STORE_H2V1_1_PIXEL  scratch0 scratch1 scratch2 outptr
+    sb          \scratch0, \r1_offs(\outptr)
+    sb          \scratch1, \g1_offs(\outptr)
+    sb          \scratch2, \b1_offs(\outptr)
+.if (\pixel_size == 8)
+    li          t0, 0xFF
+    sb          t0, \a1_offs(\outptr)
+.endif
+.endm
+
+LEAF_DSPR2(jsimd_h2v1_\colorid\()_merged_upsample_dspr2)
+/*
+ * a0     = cinfo->output_width
+ * a1     = input_buf
+ * a2     = in_row_group_ctr
+ * a3     = output_buf
+ * 16(sp) = range_limit
+ */
+    SAVE_REGS_ON_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, ra
+
+    li          t0, 0xe6ea
+    lw          t1, 0(a1)         // t1 = input_buf[0]
+    lw          t2, 4(a1)         // t2 = input_buf[1]
+    lw          t3, 8(a1)         // t3 = input_buf[2]
+    lw          t8, 56(sp)        // t8 = range_limit
+    addiu       s1, t0, 0x7fff    // s1 = 0x166e9 [FIX(1.40200)]
+    addiu       s2, s1, 0x5eb9    // s2 = 0x1c5a2 [FIX(1.77200)]
+    addiu       s0, t0, 0x9916    // s0 = 0x8000
+    addiu       s4, zero, 0xa7e6  // s4 = 0xffffa7e6 [-FIX(0.34414)]
+    xori        s3, s4, 0xeec8    // s3 = 0xffff492e [-FIX(0.71414)]
+    srl         t0, a0, 1
+    sll         t4, a2, 2
+    lwx         s5, t4(t1)      // s5 = inptr0
+    lwx         s6, t4(t2)      // s6 = inptr1
+    lwx         s7, t4(t3)      // s7 = inptr2
+    lw          t7, 0(a3)       // t7 = outptr
+    blez        t0, 2f
+     addu       t9, s6, t0      // t9 = end address
+1:
+    lbu         t2, 0(s6)       // t2 = cb
+    lbu         t0, 0(s7)       // t0 = cr
+    lbu         t1, 0(s5)       // t1 = y
+    addiu       t2, t2, -128    // t2 = cb - 128
+    addiu       t0, t0, -128    // t0 = cr - 128
+    mult        $ac1, s4, t2
+    madd        $ac1, s3, t0
+    sll         t0, t0, 15
+    sll         t2, t2, 15
+    mulq_rs.w   t0, s1, t0      // t0 = (C1*cr + ONE_HALF)>> SCALEBITS
+    extr_r.w    t5, $ac1, 16
+    mulq_rs.w   t6, s2, t2      // t6 = (C2*cb + ONE_HALF)>> SCALEBITS
+    addiu       s7, s7, 1
+    addiu       s6, s6, 1
+    addu        t2, t1, t0      // t2 = y + cred
+    addu        t3, t1, t5      // t3 = y + cgreen
+    addu        t4, t1, t6      // t4 = y + cblue
+    addu        t2, t8, t2
+    addu        t3, t8, t3
+    addu        t4, t8, t4
+    lbu         t1, 1(s5)
+    lbu         v0, 0(t2)
+    lbu         v1, 0(t3)
+    lbu         ra, 0(t4)
+    addu        t2, t1, t0
+    addu        t3, t1, t5
+    addu        t4, t1, t6
+    addu        t2, t8, t2
+    addu        t3, t8, t3
+    addu        t4, t8, t4
+    lbu         t2, 0(t2)
+    lbu         t3, 0(t3)
+    lbu         t4, 0(t4)
+
+    STORE_H2V1_2_PIXELS v0, v1, ra, t2, t3, t4, t7
+
+    bne         t9, s6, 1b
+     addiu      s5, s5, 2
+2:
+    andi        t0, a0, 1
+    beqz        t0, 4f
+     nop
+3:
+    lbu         t2, 0(s6)
+    lbu         t0, 0(s7)
+    lbu         t1, 0(s5)
+    addiu       t2, t2, -128    // (cb - 128)
+    addiu       t0, t0, -128    // (cr - 128)
+    mul         t3, s4, t2
+    mul         t4, s3, t0
+    sll         t0, t0, 15
+    sll         t2, t2, 15
+    mulq_rs.w   t0, s1, t0      // (C1*cr + ONE_HALF)>> SCALEBITS
+    mulq_rs.w   t6, s2, t2      // (C2*cb + ONE_HALF)>> SCALEBITS
+    addu        t3, t3, s0
+    addu        t3, t4, t3
+    sra         t5, t3, 16      // (C4*cb + ONE_HALF + C3*cr)>> SCALEBITS
+    addu        t2, t1, t0      // y + cred
+    addu        t3, t1, t5      // y + cgreen
+    addu        t4, t1, t6      // y + cblue
+    addu        t2, t8, t2
+    addu        t3, t8, t3
+    addu        t4, t8, t4
+    lbu         t2, 0(t2)
+    lbu         t3, 0(t3)
+    lbu         t4, 0(t4)
+
+    STORE_H2V1_1_PIXEL t2, t3, t4, t7
+4:
+    RESTORE_REGS_FROM_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, ra
+
+    j           ra
+     nop
+
+END(jsimd_h2v1_\colorid\()_merged_upsample_dspr2)
+
+.purgem STORE_H2V1_1_PIXEL
+.purgem STORE_H2V1_2_PIXELS
+.endm
+
+/*------------------------------------id -- pix R1 G1 B1 A1 R2 G2 B2 A2 */
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extrgb,  6, 0, 1, 2, 6, 3, 4, 5, 6
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extbgr,  6, 2, 1, 0, 3, 5, 4, 3, 6
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extrgbx, 8, 0, 1, 2, 3, 4, 5, 6, 7
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extbgrx, 8, 2, 1, 0, 3, 6, 5, 4, 7
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extxbgr, 8, 3, 2, 1, 0, 7, 6, 5, 4
+GENERATE_H2V1_MERGED_UPSAMPLE_DSPR2 extxrgb, 8, 1, 2, 3, 0, 5, 6, 7, 4
+
+
+/*****************************************************************************/
+/*
+ * jsimd_h2v2_fancy_upsample_dspr2
+ *
+ * Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
+ */
+LEAF_DSPR2(jsimd_h2v2_fancy_upsample_dspr2)
+/*
+ * a0 = cinfo->max_v_samp_factor
+ * a1 = downsampled_width
+ * a2 = input_data
+ * a3 = output_data_ptr
+ */
+    SAVE_REGS_ON_STACK 24, s0, s1, s2, s3, s4, s5
+
+    li            s4, 0
+    lw            s2, 0(a3)       // s2 = *output_data_ptr
+0:
+    li            t9, 2
+    lw            s1, -4(a2)      // s1 = inptr1
+
+1:
+    lw            s0, 0(a2)       // s0 = inptr0
+    lwx           s3, s4(s2)
+    addiu         s5, a1, -2      // s5 = downsampled_width - 2
+    srl           t4, s5, 1
+    sll           t4, t4, 1
+    lbu           t0, 0(s0)
+    lbu           t1, 1(s0)
+    lbu           t2, 0(s1)
+    lbu           t3, 1(s1)
+    addiu         s0, 2
+    addiu         s1, 2
+    addu          t8, s0, t4      // t8 = end address
+    andi          s5, s5, 1       // s5 = residual
+    sll           t4, t0, 1
+    sll           t6, t1, 1
+    addu          t0, t0, t4      // t0 = (*inptr0++) * 3
+    addu          t1, t1, t6      // t1 = (*inptr0++) * 3
+    addu          t7, t0, t2      // t7 = thiscolsum
+    addu          t6, t1, t3      // t5 = nextcolsum
+    sll           t0, t7, 2       // t0 = thiscolsum * 4
+    subu          t1, t0, t7      // t1 = thiscolsum * 3
+    shra_r.w      t0, t0, 4
+    addiu         t1, 7
+    addu          t1, t1, t6
+    srl           t1, t1, 4
+    sb            t0, 0(s3)
+    sb            t1, 1(s3)
+    beq           t8, s0, 22f     // skip to final iteration if width == 3
+     addiu        s3, 2
+2:
+    lh            t0, 0(s0)       // t0 = A3|A2
+    lh            t2, 0(s1)       // t2 = B3|B2
+    addiu         s0, 2
+    addiu         s1, 2
+    preceu.ph.qbr t0, t0          // t0 = 0|A3|0|A2
+    preceu.ph.qbr t2, t2          // t2 = 0|B3|0|B2
+    shll.ph       t1, t0, 1
+    sll           t3, t6, 1
+    addu.ph       t0, t1, t0      // t0 = A3*3|A2*3
+    addu          t3, t3, t6      // t3 = this * 3
+    addu.ph       t0, t0, t2      // t0 = next2|next1
+    addu          t1, t3, t7
+    andi          t7, t0, 0xFFFF  // t7 = next1
+    sll           t2, t7, 1
+    addu          t2, t7, t2      // t2 = next1*3
+    addu          t4, t2, t6
+    srl           t6, t0, 16      // t6 = next2
+    shra_r.w      t1, t1, 4       // t1 = (this*3 + last + 8) >> 4
+    addu          t0, t3, t7
+    addiu         t0, 7
+    srl           t0, t0, 4       // t0 = (this*3 + next1 + 7) >> 4
+    shra_r.w      t4, t4, 4       // t3 = (next1*3 + this + 8) >> 4
+    addu          t2, t2, t6
+    addiu         t2, 7
+    srl           t2, t2, 4       // t2 = (next1*3 + next2 + 7) >> 4
+    sb            t1, 0(s3)
+    sb            t0, 1(s3)
+    sb            t4, 2(s3)
+    sb            t2, 3(s3)
+    bne           t8, s0, 2b
+     addiu        s3, 4
+22:
+    beqz          s5, 4f
+     addu         t8, s0, s5
+3:
+    lbu           t0, 0(s0)
+    lbu           t2, 0(s1)
+    addiu         s0, 1
+    addiu         s1, 1
+    sll           t3, t6, 1
+    sll           t1, t0, 1
+    addu          t1, t0, t1      // t1 = inptr0 * 3
+    addu          t3, t3, t6      // t3 = thiscolsum * 3
+    addu          t5, t1, t2
+    addu          t1, t3, t7
+    shra_r.w      t1, t1, 4
+    addu          t0, t3, t5
+    addiu         t0, 7
+    srl           t0, t0, 4
+    sb            t1, 0(s3)
+    sb            t0, 1(s3)
+    addiu         s3, 2
+    move          t7, t6
+    bne           t8, s0, 3b
+     move         t6, t5
+4:
+    sll           t0, t6, 2       // t0 = thiscolsum * 4
+    subu          t1, t0, t6      // t1 = thiscolsum * 3
+    addu          t1, t1, t7
+    addiu         s4, 4
+    shra_r.w      t1, t1, 4
+    addiu         t0, 7
+    srl           t0, t0, 4
+    sb            t1, 0(s3)
+    sb            t0, 1(s3)
+    addiu         t9, -1
+    addiu         s3, 2
+    bnez          t9, 1b
+     lw           s1, 4(a2)
+    srl           t0, s4, 2
+    subu          t0, a0, t0
+    bgtz          t0, 0b
+     addiu        a2, 4
+
+    RESTORE_REGS_FROM_STACK 24, s0, s1, s2, s3, s4, s5
+
+    j             ra
+     nop
+END(jsimd_h2v2_fancy_upsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v1_fancy_upsample_dspr2)
+/*
+ * a0 = cinfo->max_v_samp_factor
+ * a1 = downsampled_width
+ * a2 = input_data
+ * a3 = output_data_ptr
+ */
+    SAVE_REGS_ON_STACK 16, s0, s1, s2, s3
+
+    .set at
+
+    beqz          a0, 3f
+     sll          t0, a0, 2
+    lw            s1, 0(a3)
+    li            s3, 0x10001
+    addu          s0, s1, t0
+0:
+    addiu         t8, a1, -2
+    srl           t9, t8, 2
+    lw            t7, 0(a2)
+    lw            s2, 0(s1)
+    lbu           t0, 0(t7)
+    lbu           t1, 1(t7)       // t1 = inptr[1]
+    sll           t2, t0, 1
+    addu          t2, t2, t0      // t2 = invalue*3
+    addu          t2, t2, t1
+    shra_r.w      t2, t2, 2
+    sb            t0, 0(s2)
+    sb            t2, 1(s2)
+    beqz          t9, 11f
+     addiu        s2, 2
+1:
+    ulw           t0, 0(t7)       // t0 = |P3|P2|P1|P0|
+    ulw           t1, 1(t7)
+    ulh           t2, 4(t7)       // t2 = |0|0|P5|P4|
+    preceu.ph.qbl t3, t0          // t3 = |0|P3|0|P2|
+    preceu.ph.qbr t0, t0          // t0 = |0|P1|0|P0|
+    preceu.ph.qbr t2, t2          // t2 = |0|P5|0|P4|
+    preceu.ph.qbl t4, t1          // t4 = |0|P4|0|P3|
+    preceu.ph.qbr t1, t1          // t1 = |0|P2|0|P1|
+    shll.ph       t5, t4, 1
+    shll.ph       t6, t1, 1
+    addu.ph       t5, t5, t4      // t5 = |P4*3|P3*3|
+    addu.ph       t6, t6, t1      // t6 = |P2*3|P1*3|
+    addu.ph       t4, t3, s3
+    addu.ph       t0, t0, s3
+    addu.ph       t4, t4, t5
+    addu.ph       t0, t0, t6
+    shrl.ph       t4, t4, 2       // t4 = |0|P3|0|P2|
+    shrl.ph       t0, t0, 2       // t0 = |0|P1|0|P0|
+    addu.ph       t2, t2, t5
+    addu.ph       t3, t3, t6
+    shra_r.ph     t2, t2, 2       // t2 = |0|P5|0|P4|
+    shra_r.ph     t3, t3, 2       // t3 = |0|P3|0|P2|
+    shll.ph       t2, t2, 8
+    shll.ph       t3, t3, 8
+    or            t2, t4, t2
+    or            t3, t3, t0
+    addiu         t9, -1
+    usw           t3, 0(s2)
+    usw           t2, 4(s2)
+    addiu         s2, 8
+    bgtz          t9, 1b
+     addiu        t7, 4
+11:
+    andi          t8, 3
+    beqz          t8, 22f
+     addiu        t7, 1
+
+2:
+    lbu           t0, 0(t7)
+    addiu         t7, 1
+    sll           t1, t0, 1
+    addu          t2, t0, t1      // t2 = invalue
+    lbu           t3, -2(t7)
+    lbu           t4, 0(t7)
+    addiu         t3, 1
+    addiu         t4, 2
+    addu          t3, t3, t2
+    addu          t4, t4, t2
+    srl           t3, 2
+    srl           t4, 2
+    sb            t3, 0(s2)
+    sb            t4, 1(s2)
+    addiu         t8, -1
+    bgtz          t8, 2b
+     addiu        s2, 2
+
+22:
+    lbu           t0, 0(t7)
+    lbu           t2, -1(t7)
+    sll           t1, t0, 1
+    addu          t1, t1, t0      // t1 = invalue * 3
+    addu          t1, t1, t2
+    addiu         t1, 1
+    srl           t1, t1, 2
+    sb            t1, 0(s2)
+    sb            t0, 1(s2)
+    addiu         s1, 4
+    bne           s1, s0, 0b
+     addiu        a2, 4
+3:
+    RESTORE_REGS_FROM_STACK 16, s0, s1, s2, s3
+
+    j             ra
+     nop
+END(jsimd_h2v1_fancy_upsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v1_downsample_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = cinfo->max_v_samp_factor
+ * a2     = compptr->v_samp_factor
+ * a3     = compptr->width_in_blocks
+ * 16(sp) = input_data
+ * 20(sp) = output_data
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 24, s0, s1, s2, s3, s4
+
+    beqz        a2, 7f
+     lw         s1, 44(sp)      // s1 = output_data
+    lw          s0, 40(sp)      // s0 = input_data
+    srl         s2, a0, 2
+    andi        t9, a0, 2
+    srl         t7, t9, 1
+    addu        s2, t7, s2
+    sll         t0, a3, 3       // t0 = width_in_blocks*DCT
+    srl         t7, t0, 1
+    subu        s2, t7, s2
+0:
+    andi        t6, a0, 1       // t6 = temp_index
+    addiu       t6, -1
+    lw          t4, 0(s1)       // t4 = outptr
+    lw          t5, 0(s0)       // t5 = inptr0
+    li          s3, 0           // s3 = bias
+    srl         t7, a0, 1       // t7 = image_width1
+    srl         s4, t7, 2
+    andi        t8, t7, 3
+1:
+    ulhu        t0, 0(t5)
+    ulhu        t1, 2(t5)
+    ulhu        t2, 4(t5)
+    ulhu        t3, 6(t5)
+    raddu.w.qb  t0, t0
+    raddu.w.qb  t1, t1
+    raddu.w.qb  t2, t2
+    raddu.w.qb  t3, t3
+    shra.ph     t0, t0, 1
+    shra_r.ph   t1, t1, 1
+    shra.ph     t2, t2, 1
+    shra_r.ph   t3, t3, 1
+    sb          t0, 0(t4)
+    sb          t1, 1(t4)
+    sb          t2, 2(t4)
+    sb          t3, 3(t4)
+    addiu       s4, -1
+    addiu       t4, 4
+    bgtz        s4, 1b
+     addiu      t5, 8
+    beqz        t8, 3f
+     addu       s4, t4, t8
+2:
+    ulhu        t0, 0(t5)
+    raddu.w.qb  t0, t0
+    addqh.w     t0, t0, s3
+    xori        s3, s3, 1
+    sb          t0, 0(t4)
+    addiu       t4, 1
+    bne         t4, s4, 2b
+     addiu      t5, 2
+3:
+    lbux        t1, t6(t5)
+    sll         t1, 1
+    addqh.w     t2, t1, s3      // t2 = pixval1
+    xori        s3, s3, 1
+    addqh.w     t3, t1, s3      // t3 = pixval2
+    blez        s2, 5f
+     append     t3, t2,  8
+    addu        t5, t4, s2      // t5 = loop_end2
+4:
+    ush         t3, 0(t4)
+    addiu       s2, -1
+    bgtz        s2, 4b
+     addiu      t4,  2
+5:
+    beqz        t9, 6f
+     nop
+    sb          t2, 0(t4)
+6:
+    addiu       s1, 4
+    addiu       a2, -1
+    bnez        a2, 0b
+     addiu      s0, 4
+7:
+    RESTORE_REGS_FROM_STACK 24, s0, s1, s2, s3, s4
+
+    j           ra
+    nop
+END(jsimd_h2v1_downsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v2_downsample_dspr2)
+/*
+ * a0     = cinfo->image_width
+ * a1     = cinfo->max_v_samp_factor
+ * a2     = compptr->v_samp_factor
+ * a3     = compptr->width_in_blocks
+ * 16(sp) = input_data
+ * 20(sp) = output_data
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    beqz        a2, 8f
+     lw         s1, 52(sp)      // s1 = output_data
+    lw          s0, 48(sp)      // s0 = input_data
+
+    andi        t6, a0, 1       // t6 = temp_index
+    addiu       t6, -1
+    srl         t7, a0, 1       // t7 = image_width1
+    srl         s4, t7, 2
+    andi        t8, t7, 3
+    andi        t9, a0, 2
+    srl         s2, a0, 2
+    srl         t7, t9, 1
+    addu        s2, t7, s2
+    sll         t0, a3, 3       // s2 = width_in_blocks*DCT
+    srl         t7, t0, 1
+    subu        s2, t7, s2
+0:
+    lw          t4, 0(s1)       // t4 = outptr
+    lw          t5, 0(s0)       // t5 = inptr0
+    lw          s7, 4(s0)       // s7 = inptr1
+    li          s6, 1           // s6 = bias
+2:
+    ulw         t0, 0(t5)       // t0 = |P3|P2|P1|P0|
+    ulw         t1, 0(s7)       // t1 = |Q3|Q2|Q1|Q0|
+    ulw         t2, 4(t5)
+    ulw         t3, 4(s7)
+    precrq.ph.w t7, t0, t1      // t2 = |P3|P2|Q3|Q2|
+    ins         t0, t1, 16, 16  // t0 = |Q1|Q0|P1|P0|
+    raddu.w.qb  t1, t7
+    raddu.w.qb  t0, t0
+    shra_r.w    t1, t1, 2
+    addiu       t0, 1
+    srl         t0, 2
+    precrq.ph.w t7, t2, t3
+    ins         t2, t3, 16, 16
+    raddu.w.qb  t7, t7
+    raddu.w.qb  t2, t2
+    shra_r.w    t7, t7, 2
+    addiu       t2, 1
+    srl         t2, 2
+    sb          t0, 0(t4)
+    sb          t1, 1(t4)
+    sb          t2, 2(t4)
+    sb          t7, 3(t4)
+    addiu       t4, 4
+    addiu       t5, 8
+    addiu       s4, s4, -1
+    bgtz        s4, 2b
+     addiu      s7, 8
+    beqz        t8, 4f
+     addu       t8, t4, t8
+3:
+    ulhu        t0, 0(t5)
+    ulhu        t1, 0(s7)
+    ins         t0, t1, 16, 16
+    raddu.w.qb  t0, t0
+    addu        t0, t0, s6
+    srl         t0, 2
+    xori        s6, s6, 3
+    sb          t0, 0(t4)
+    addiu       t5, 2
+    addiu       t4, 1
+    bne         t8, t4, 3b
+     addiu      s7, 2
+4:
+    lbux        t1, t6(t5)
+    sll         t1, 1
+    lbux        t0, t6(s7)
+    sll         t0, 1
+    addu        t1, t1, t0
+    addu        t3, t1, s6
+    srl         t0, t3, 2       // t2 = pixval1
+    xori        s6, s6, 3
+    addu        t2, t1, s6
+    srl         t1, t2, 2       // t3 = pixval2
+    blez        s2, 6f
+     append     t1, t0, 8
+5:
+    ush         t1, 0(t4)
+    addiu       s2, -1
+    bgtz        s2, 5b
+     addiu      t4, 2
+6:
+    beqz        t9, 7f
+     nop
+    sb          t0, 0(t4)
+7:
+    addiu       s1, 4
+    addiu       a2, -1
+    bnez        a2, 0b
+     addiu      s0, 8
+8:
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+END(jsimd_h2v2_downsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v2_smooth_downsample_dspr2)
+/*
+ * a0     = input_data
+ * a1     = output_data
+ * a2     = compptr->v_samp_factor
+ * a3     = cinfo->max_v_samp_factor
+ * 16(sp) = cinfo->smoothing_factor
+ * 20(sp) = compptr->width_in_blocks
+ * 24(sp) = cinfo->image_width
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    lw          s7, 52(sp)      // compptr->width_in_blocks
+    lw          s0, 56(sp)      // cinfo->image_width
+    lw          s6, 48(sp)      // cinfo->smoothing_factor
+    sll         s7, 3           // output_cols = width_in_blocks * DCTSIZE
+    sll         v0, s7, 1
+    subu        v0, v0, s0
+    blez        v0, 2f
+    move        v1, zero
+    addiu       t0, a3, 2       // t0 = cinfo->max_v_samp_factor + 2
+0:
+    addiu       t1, a0, -4
+    sll         t2, v1, 2
+    lwx         t1, t2(t1)
+    move        t3, v0
+    addu        t1, t1, s0
+    lbu         t2, -1(t1)
+1:
+    addiu       t3, t3, -1
+    sb          t2, 0(t1)
+    bgtz        t3, 1b
+    addiu       t1, t1, 1
+    addiu       v1, v1, 1
+    bne         v1, t0, 0b
+    nop
+2:
+    li          v0, 80
+    mul         v0, s6, v0
+    li          v1, 16384
+    move        t4, zero
+    move        t5, zero
+    subu        t6, v1, v0      // t6 = 16384 - tmp_smoot_f * 80
+    sll         t7, s6, 4       // t7 = tmp_smoot_f * 16
+3:
+/* Special case for first column: pretend column -1 is same as column 0 */
+    sll         v0, t4, 2
+    lwx         t8, v0(a1)      //  outptr = output_data[outrow]
+    sll         v1, t5, 2
+    addiu       t9, v1, 4
+    addiu       s0, v1, -4
+    addiu       s1, v1, 8
+    lwx         s2, v1(a0)      // inptr0 = input_data[inrow]
+    lwx         t9, t9(a0)      // inptr1 = input_data[inrow+1]
+    lwx         s0, s0(a0)      // above_ptr = input_data[inrow-1]
+    lwx         s1, s1(a0)      // below_ptr = input_data[inrow+2]
+    lh          v0, 0(s2)
+    lh          v1, 0(t9)
+    lh          t0, 0(s0)
+    lh          t1, 0(s1)
+    ins         v0, v1, 16, 16
+    ins         t0, t1, 16, 16
+    raddu.w.qb  t2, v0
+    raddu.w.qb  s3, t0
+    lbu         v0, 0(s2)
+    lbu         v1, 2(s2)
+    lbu         t0, 0(t9)
+    lbu         t1, 2(t9)
+    addu        v0, v0, v1
+    mult        $ac1, t2, t6
+    addu        t0, t0, t1
+    lbu         t2, 2(s0)
+    addu        t0, t0, v0
+    lbu         t3, 2(s1)
+    addu        s3, t0, s3
+    lbu         v0, 0(s0)
+    lbu         t0, 0(s1)
+    sll         s3, s3, 1
+    addu        v0, v0, t2
+    addu        t0, t0, t3
+    addu        t0, t0, v0
+    addu        s3, t0, s3
+    madd        $ac1, s3, t7
+    extr_r.w    v0, $ac1, 16
+    addiu       t8, t8, 1
+    addiu       s2, s2, 2
+    addiu       t9, t9, 2
+    addiu       s0, s0, 2
+    addiu       s1, s1, 2
+    sb          v0, -1(t8)
+    addiu       s4, s7, -2
+    and         s4, s4, 3
+    addu        s5, s4, t8      // end address
+4:
+    lh          v0, 0(s2)
+    lh          v1, 0(t9)
+    lh          t0, 0(s0)
+    lh          t1, 0(s1)
+    ins         v0, v1, 16, 16
+    ins         t0, t1, 16, 16
+    raddu.w.qb  t2, v0
+    raddu.w.qb  s3, t0
+    lbu         v0, -1(s2)
+    lbu         v1, 2(s2)
+    lbu         t0, -1(t9)
+    lbu         t1, 2(t9)
+    addu        v0, v0, v1
+    mult        $ac1, t2, t6
+    addu        t0, t0, t1
+    lbu         t2, 2(s0)
+    addu        t0, t0, v0
+    lbu         t3, 2(s1)
+    addu        s3, t0, s3
+    lbu         v0, -1(s0)
+    lbu         t0, -1(s1)
+    sll         s3, s3, 1
+    addu        v0, v0, t2
+    addu        t0, t0, t3
+    addu        t0, t0, v0
+    addu        s3, t0, s3
+    madd        $ac1, s3, t7
+    extr_r.w    t2, $ac1, 16
+    addiu       t8, t8, 1
+    addiu       s2, s2, 2
+    addiu       t9, t9, 2
+    addiu       s0, s0, 2
+    sb          t2, -1(t8)
+    bne         s5, t8, 4b
+    addiu       s1, s1, 2
+    addiu       s5, s7, -2
+    subu        s5, s5, s4
+    addu        s5, s5, t8      // end address
+5:
+    lh          v0, 0(s2)
+    lh          v1, 0(t9)
+    lh          t0, 0(s0)
+    lh          t1, 0(s1)
+    ins         v0, v1, 16, 16
+    ins         t0, t1, 16, 16
+    raddu.w.qb  t2, v0
+    raddu.w.qb  s3, t0
+    lbu         v0, -1(s2)
+    lbu         v1, 2(s2)
+    lbu         t0, -1(t9)
+    lbu         t1, 2(t9)
+    addu        v0, v0, v1
+    mult        $ac1, t2, t6
+    addu        t0, t0, t1
+    lbu         t2, 2(s0)
+    addu        t0, t0, v0
+    lbu         t3, 2(s1)
+    addu        s3, t0, s3
+    lbu         v0, -1(s0)
+    lbu         t0, -1(s1)
+    sll         s3, s3, 1
+    addu        v0, v0, t2
+    addu        t0, t0, t3
+    lh          v1, 2(t9)
+    addu        t0, t0, v0
+    lh          v0, 2(s2)
+    addu        s3, t0, s3
+    lh          t0, 2(s0)
+    lh          t1, 2(s1)
+    madd        $ac1, s3, t7
+    extr_r.w    t2, $ac1, 16
+    ins         t0, t1, 16, 16
+    ins         v0, v1, 16, 16
+    raddu.w.qb  s3, t0
+    lbu         v1, 4(s2)
+    lbu         t0, 1(t9)
+    lbu         t1, 4(t9)
+    sb          t2, 0(t8)
+    raddu.w.qb  t3, v0
+    lbu         v0, 1(s2)
+    addu        t0, t0, t1
+    mult        $ac1, t3, t6
+    addu        v0, v0, v1
+    lbu         t2, 4(s0)
+    addu        t0, t0, v0
+    lbu         v0, 1(s0)
+    addu        s3, t0, s3
+    lbu         t0, 1(s1)
+    lbu         t3, 4(s1)
+    addu        v0, v0, t2
+    sll         s3, s3, 1
+    addu        t0, t0, t3
+    lh          v1, 4(t9)
+    addu        t0, t0, v0
+    lh          v0, 4(s2)
+    addu        s3, t0, s3
+    lh          t0, 4(s0)
+    lh          t1, 4(s1)
+    madd        $ac1, s3, t7
+    extr_r.w    t2, $ac1, 16
+    ins         t0, t1, 16, 16
+    ins         v0, v1, 16, 16
+    raddu.w.qb  s3, t0
+    lbu         v1, 6(s2)
+    lbu         t0, 3(t9)
+    lbu         t1, 6(t9)
+    sb          t2, 1(t8)
+    raddu.w.qb  t3, v0
+    lbu         v0, 3(s2)
+    addu        t0, t0, t1
+    mult        $ac1, t3, t6
+    addu        v0, v0, v1
+    lbu         t2, 6(s0)
+    addu        t0, t0, v0
+    lbu         v0, 3(s0)
+    addu        s3, t0, s3
+    lbu         t0, 3(s1)
+    lbu         t3, 6(s1)
+    addu        v0, v0, t2
+    sll         s3, s3, 1
+    addu        t0, t0, t3
+    lh          v1, 6(t9)
+    addu        t0, t0, v0
+    lh          v0, 6(s2)
+    addu        s3, t0, s3
+    lh          t0, 6(s0)
+    lh          t1, 6(s1)
+    madd        $ac1, s3, t7
+    extr_r.w    t3, $ac1, 16
+    ins         t0, t1, 16, 16
+    ins         v0, v1, 16, 16
+    raddu.w.qb  s3, t0
+    lbu         v1, 8(s2)
+    lbu         t0, 5(t9)
+    lbu         t1, 8(t9)
+    sb          t3, 2(t8)
+    raddu.w.qb  t2, v0
+    lbu         v0, 5(s2)
+    addu        t0, t0, t1
+    mult        $ac1, t2, t6
+    addu        v0, v0, v1
+    lbu         t2, 8(s0)
+    addu        t0, t0, v0
+    lbu         v0, 5(s0)
+    addu        s3, t0, s3
+    lbu         t0, 5(s1)
+    lbu         t3, 8(s1)
+    addu        v0, v0, t2
+    sll         s3, s3, 1
+    addu        t0, t0, t3
+    addiu       t8, t8, 4
+    addu        t0, t0, v0
+    addiu       s2, s2, 8
+    addu        s3, t0, s3
+    addiu       t9, t9, 8
+    madd        $ac1, s3, t7
+    extr_r.w    t1, $ac1, 16
+    addiu       s0, s0, 8
+    addiu       s1, s1, 8
+    bne         s5, t8, 5b
+    sb          t1, -1(t8)
+/* Special case for last column */
+    lh          v0, 0(s2)
+    lh          v1, 0(t9)
+    lh          t0, 0(s0)
+    lh          t1, 0(s1)
+    ins         v0, v1, 16, 16
+    ins         t0, t1, 16, 16
+    raddu.w.qb  t2, v0
+    raddu.w.qb  s3, t0
+    lbu         v0, -1(s2)
+    lbu         v1, 1(s2)
+    lbu         t0, -1(t9)
+    lbu         t1, 1(t9)
+    addu        v0, v0, v1
+    mult        $ac1, t2, t6
+    addu        t0, t0, t1
+    lbu         t2, 1(s0)
+    addu        t0, t0, v0
+    lbu         t3, 1(s1)
+    addu        s3, t0, s3
+    lbu         v0, -1(s0)
+    lbu         t0, -1(s1)
+    sll         s3, s3, 1
+    addu        v0, v0, t2
+    addu        t0, t0, t3
+    addu        t0, t0, v0
+    addu        s3, t0, s3
+    madd        $ac1, s3, t7
+    extr_r.w    t0, $ac1, 16
+    addiu       t5, t5, 2
+    sb          t0, 0(t8)
+    addiu       t4, t4, 1
+    bne         t4, a2, 3b
+    addiu       t5, t5, 2
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+
+END(jsimd_h2v2_smooth_downsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_int_upsample_dspr2)
+/*
+ * a0     = upsample->h_expand[compptr->component_index]
+ * a1     = upsample->v_expand[compptr->component_index]
+ * a2     = input_data
+ * a3     = output_data_ptr
+ * 16(sp) = cinfo->output_width
+ * 20(sp) = cinfo->max_v_samp_factor
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 16, s0, s1, s2, s3
+
+    lw          s0, 0(a3)       // s0 = output_data
+    lw          s1, 32(sp)      // s1 = cinfo->output_width
+    lw          s2, 36(sp)      // s2 = cinfo->max_v_samp_factor
+    li          t6, 0           // t6 = inrow
+    beqz        s2, 10f
+     li         s3, 0           // s3 = outrow
+0:
+    addu        t0, a2, t6
+    addu        t7, s0, s3
+    lw          t3, 0(t0)       // t3 = inptr
+    lw          t8, 0(t7)       // t8 = outptr
+    beqz        s1, 4f
+     addu       t5, t8, s1      // t5 = outend
+1:
+    lb          t2, 0(t3)       // t2 = invalue = *inptr++
+    addiu       t3, 1
+    beqz        a0, 3f
+     move       t0, a0          // t0 = h_expand
+2:
+    sb          t2, 0(t8)
+    addiu       t0, -1
+    bgtz        t0, 2b
+     addiu      t8, 1
+3:
+    bgt         t5, t8, 1b
+     nop
+4:
+    addiu       t9, a1, -1      // t9 = v_expand - 1
+    blez        t9, 9f
+     nop
+5:
+    lw          t3, 0(s0)
+    lw          t4, 4(s0)
+    subu        t0, s1, 0xF
+    blez        t0, 7f
+     addu       t5, t3, s1      // t5 = end address
+    andi        t7, s1, 0xF     // t7 = residual
+    subu        t8, t5, t7
+6:
+    ulw         t0, 0(t3)
+    ulw         t1, 4(t3)
+    ulw         t2, 8(t3)
+    usw         t0, 0(t4)
+    ulw         t0, 12(t3)
+    usw         t1, 4(t4)
+    usw         t2, 8(t4)
+    usw         t0, 12(t4)
+    addiu       t3, 16
+    bne         t3, t8, 6b
+     addiu      t4, 16
+    beqz        t7, 8f
+     nop
+7:
+    lbu         t0, 0(t3)
+    sb          t0, 0(t4)
+    addiu       t3, 1
+    bne         t3, t5, 7b
+     addiu      t4, 1
+8:
+    addiu       t9, -1
+    bgtz        t9, 5b
+     addiu      s0, 8
+9:
+    addu        s3, s3, a1
+    bne         s3, s2, 0b
+     addiu      t6, 1
+10:
+    RESTORE_REGS_FROM_STACK 16, s0, s1, s2, s3
+
+    j           ra
+     nop
+END(jsimd_int_upsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v1_upsample_dspr2)
+/*
+ * a0 = cinfo->max_v_samp_factor
+ * a1 = cinfo->output_width
+ * a2 = input_data
+ * a3 = output_data_ptr
+ */
+    lw          t7, 0(a3)       // t7 = output_data
+    andi        t8, a1, 0xf     // t8 = residual
+    sll         t0, a0, 2
+    blez        a0, 4f
+     addu       t9, t7, t0      // t9 = output_data end address
+0:
+    lw          t5, 0(t7)       // t5 = outptr
+    lw          t6, 0(a2)       // t6 = inptr
+    addu        t3, t5, a1      // t3 = outptr + output_width (end address)
+    subu        t3, t8          // t3 = end address - residual
+    beq         t5, t3, 2f
+     move       t4, t8
+1:
+    ulw         t0, 0(t6)       // t0 = |P3|P2|P1|P0|
+    ulw         t2, 4(t6)       // t2 = |P7|P6|P5|P4|
+    srl         t1, t0, 16      // t1 = |X|X|P3|P2|
+    ins         t0, t0, 16, 16  // t0 = |P1|P0|P1|P0|
+    ins         t1, t1, 16, 16  // t1 = |P3|P2|P3|P2|
+    ins         t0, t0, 8, 16   // t0 = |P1|P1|P0|P0|
+    ins         t1, t1, 8, 16   // t1 = |P3|P3|P2|P2|
+    usw         t0, 0(t5)
+    usw         t1, 4(t5)
+    srl         t0, t2, 16      // t0 = |X|X|P7|P6|
+    ins         t2, t2, 16, 16  // t2 = |P5|P4|P5|P4|
+    ins         t0, t0, 16, 16  // t0 = |P7|P6|P7|P6|
+    ins         t2, t2, 8, 16   // t2 = |P5|P5|P4|P4|
+    ins         t0, t0, 8, 16   // t0 = |P7|P7|P6|P6|
+    usw         t2, 8(t5)
+    usw         t0, 12(t5)
+    addiu       t5, 16
+    bne         t5, t3, 1b
+     addiu      t6, 8
+    beqz        t8, 3f
+     move       t4, t8
+2:
+    lbu         t1, 0(t6)
+    sb          t1, 0(t5)
+    sb          t1, 1(t5)
+    addiu       t4, -2
+    addiu       t6, 1
+    bgtz        t4, 2b
+     addiu      t5, 2
+3:
+    addiu       t7, 4
+    bne         t9, t7, 0b
+     addiu      a2, 4
+4:
+    j           ra
+     nop
+END(jsimd_h2v1_upsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_h2v2_upsample_dspr2)
+/*
+ * a0 = cinfo->max_v_samp_factor
+ * a1 = cinfo->output_width
+ * a2 = input_data
+ * a3 = output_data_ptr
+ */
+    lw          t7, 0(a3)
+    blez        a0, 7f
+     andi       t9, a1, 0xf     // t9 = residual
+0:
+    lw          t6, 0(a2)       // t6 = inptr
+    lw          t5, 0(t7)       // t5 = outptr
+    addu        t8, t5, a1      // t8 = outptr end address
+    subu        t8, t9          // t8 = end address - residual
+    beq         t5, t8, 2f
+     move       t4, t9
+1:
+    ulw         t0, 0(t6)
+    srl         t1, t0, 16
+    ins         t0, t0, 16, 16
+    ins         t0, t0, 8, 16
+    ins         t1, t1, 16, 16
+    ins         t1, t1, 8, 16
+    ulw         t2, 4(t6)
+    usw         t0, 0(t5)
+    usw         t1, 4(t5)
+    srl         t3, t2, 16
+    ins         t2, t2, 16, 16
+    ins         t2, t2, 8, 16
+    ins         t3, t3, 16, 16
+    ins         t3, t3, 8, 16
+    usw         t2, 8(t5)
+    usw         t3, 12(t5)
+    addiu       t5, 16
+    bne         t5, t8, 1b
+     addiu      t6, 8
+    beqz        t9, 3f
+     move       t4, t9
+2:
+    lbu         t0, 0(t6)
+    sb          t0, 0(t5)
+    sb          t0, 1(t5)
+    addiu       t4, -2
+    addiu       t6, 1
+    bgtz        t4, 2b
+     addiu      t5, 2
+3:
+    lw          t6, 0(t7)       // t6 = outptr[0]
+    lw          t5, 4(t7)       // t5 = outptr[1]
+    addu        t4, t6, a1      // t4 = new end address
+    beq         a1, t9, 5f
+     subu       t8, t4, t9
+4:
+    ulw         t0, 0(t6)
+    ulw         t1, 4(t6)
+    ulw         t2, 8(t6)
+    usw         t0, 0(t5)
+    ulw         t0, 12(t6)
+    usw         t1, 4(t5)
+    usw         t2, 8(t5)
+    usw         t0, 12(t5)
+    addiu       t6, 16
+    bne         t6, t8, 4b
+     addiu      t5, 16
+    beqz        t9, 6f
+     nop
+5:
+    lbu         t0, 0(t6)
+    sb          t0, 0(t5)
+    addiu       t6, 1
+    bne         t6, t4, 5b
+     addiu      t5, 1
+6:
+    addiu       t7, 8
+    addiu       a0, -2
+    bgtz        a0, 0b
+     addiu      a2, 4
+7:
+    j           ra
+     nop
+END(jsimd_h2v2_upsample_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_islow_dspr2)
+/*
+ * a0 = coef_block
+ * a1 = compptr->dcttable
+ * a2 = output
+ * a3 = range_limit
+ */
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    addiu       sp, sp, -256
+    move        v0, sp
+    addiu       v1, zero, 8     // v1 = DCTSIZE = 8
+1:
+    lh          s4, 32(a0)      // s4 = inptr[16]
+    lh          s5, 64(a0)      // s5 = inptr[32]
+    lh          s6, 96(a0)      // s6 = inptr[48]
+    lh          t1, 112(a0)     // t1 = inptr[56]
+    lh          t7, 16(a0)      // t7 = inptr[8]
+    lh          t5, 80(a0)      // t5 = inptr[40]
+    lh          t3, 48(a0)      // t3 = inptr[24]
+    or          s4, s4, t1
+    or          s4, s4, t3
+    or          s4, s4, t5
+    or          s4, s4, t7
+    or          s4, s4, s5
+    or          s4, s4, s6
+    bnez        s4, 2f
+     addiu      v1, v1, -1
+    lh          s5, 0(a1)       // quantptr[DCTSIZE*0]
+    lh          s6, 0(a0)       // inptr[DCTSIZE*0]
+    mul         s5, s5, s6      // DEQUANTIZE(inptr[0], quantptr[0])
+    sll         s5, s5, 2
+    sw          s5, 0(v0)
+    sw          s5, 32(v0)
+    sw          s5, 64(v0)
+    sw          s5, 96(v0)
+    sw          s5, 128(v0)
+    sw          s5, 160(v0)
+    sw          s5, 192(v0)
+    b           3f
+     sw         s5, 224(v0)
+2:
+    lh          t0, 112(a1)
+    lh          t2, 48(a1)
+    lh          t4, 80(a1)
+    lh          t6, 16(a1)
+    mul         t0, t0, t1      // DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7])
+    mul         t1, t2, t3      // DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3])
+    mul         t2, t4, t5      // DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5])
+    mul         t3, t6, t7      // DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1])
+    lh          t4, 32(a1)
+    lh          t5, 32(a0)
+    lh          t6, 96(a1)
+    lh          t7, 96(a0)
+    addu        s0, t0, t1       // z3 = tmp0 + tmp2
+    addu        s1, t1, t2       // z2 = tmp1 + tmp2
+    addu        s2, t2, t3       // z4 = tmp1 + tmp3
+    addu        s3, s0, s2       // z3 + z4
+    addiu       t9, zero, 9633   // FIX_1_175875602
+    mul         s3, s3, t9       // z5 = MULTIPLY(z3 + z4, FIX_1_175875602)
+    addu        t8, t0, t3       // z1 = tmp0 + tmp3
+    addiu       t9, zero, 2446   // FIX_0_298631336
+    mul         t0, t0, t9       // tmp0 = MULTIPLY(tmp0, FIX_0_298631336)
+    addiu       t9, zero, 16819  // FIX_2_053119869
+    mul         t2, t2, t9       // tmp1 = MULTIPLY(tmp1, FIX_2_053119869)
+    addiu       t9, zero, 25172  // FIX_3_072711026
+    mul         t1, t1, t9       // tmp2 = MULTIPLY(tmp2, FIX_3_072711026)
+    addiu       t9, zero, 12299  // FIX_1_501321110
+    mul         t3, t3, t9       // tmp3 = MULTIPLY(tmp3, FIX_1_501321110)
+    addiu       t9, zero, 16069  // FIX_1_961570560
+    mul         s0, s0, t9       // -z3 = MULTIPLY(z3, FIX_1_961570560)
+    addiu       t9, zero, 3196   // FIX_0_390180644
+    mul         s2, s2, t9       // -z4 = MULTIPLY(z4, FIX_0_390180644)
+    addiu       t9, zero, 7373   // FIX_0_899976223
+    mul         t8, t8, t9       // -z1 = MULTIPLY(z1, FIX_0_899976223)
+    addiu       t9, zero, 20995  // FIX_2_562915447
+    mul         s1, s1, t9       // -z2 = MULTIPLY(z2, FIX_2_562915447)
+    subu        s0, s3, s0       // z3 += z5
+    addu        t0, t0, s0       // tmp0 += z3
+    addu        t1, t1, s0       // tmp2 += z3
+    subu        s2, s3, s2       // z4 += z5
+    addu        t2, t2, s2       // tmp1 += z4
+    addu        t3, t3, s2       // tmp3 += z4
+    subu        t0, t0, t8       // tmp0 += z1
+    subu        t1, t1, s1       // tmp2 += z2
+    subu        t2, t2, s1       // tmp1 += z2
+    subu        t3, t3, t8       // tmp3 += z1
+    mul         s0, t4, t5       // DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2])
+    addiu       t9, zero, 6270   // FIX_0_765366865
+    mul         s1, t6, t7       // DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6])
+    lh          t4, 0(a1)
+    lh          t5, 0(a0)
+    lh          t6, 64(a1)
+    lh          t7, 64(a0)
+    mul         s2, t9, s0       // MULTIPLY(z2, FIX_0_765366865)
+    mul         t5, t4, t5       // DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0])
+    mul         t6, t6, t7       // DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4])
+    addiu       t9, zero, 4433   // FIX_0_541196100
+    addu        s3, s0, s1       // z2 + z3
+    mul         s3, s3, t9       // z1 = MULTIPLY(z2 + z3, FIX_0_541196100)
+    addiu       t9, zero, 15137  // FIX_1_847759065
+    mul         t8, s1, t9       // MULTIPLY(z3, FIX_1_847759065)
+    addu        t4, t5, t6
+    subu        t5, t5, t6
+    sll         t4, t4, 13      // tmp0 = (z2 + z3) << CONST_BITS
+    sll         t5, t5, 13      // tmp1 = (z2 - z3) << CONST_BITS
+    addu        t7, s3, s2      // tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865)
+    subu        t6, s3, t8      // tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065)
+    addu        s0, t4, t7
+    subu        s1, t4, t7
+    addu        s2, t5, t6
+    subu        s3, t5, t6
+    addu        t4, s0, t3
+    subu        s0, s0, t3
+    addu        t3, s2, t1
+    subu        s2, s2, t1
+    addu        t1, s3, t2
+    subu        s3, s3, t2
+    addu        t2, s1, t0
+    subu        s1, s1, t0
+    shra_r.w    t4, t4, 11
+    shra_r.w    t3, t3, 11
+    shra_r.w    t1, t1, 11
+    shra_r.w    t2, t2, 11
+    shra_r.w    s1, s1, 11
+    shra_r.w    s3, s3, 11
+    shra_r.w    s2, s2, 11
+    shra_r.w    s0, s0, 11
+    sw          t4, 0(v0)
+    sw          t3, 32(v0)
+    sw          t1, 64(v0)
+    sw          t2, 96(v0)
+    sw          s1, 128(v0)
+    sw          s3, 160(v0)
+    sw          s2, 192(v0)
+    sw          s0, 224(v0)
+3:
+    addiu       a1, a1, 2
+    addiu       a0, a0, 2
+    bgtz        v1, 1b
+     addiu      v0, v0, 4
+    move        v0, sp
+    addiu       v1, zero, 8
+4:
+    lw          t0, 8(v0)       // z2 = (JLONG)wsptr[2]
+    lw          t1, 24(v0)      // z3 = (JLONG)wsptr[6]
+    lw          t2, 0(v0)       // (JLONG)wsptr[0]
+    lw          t3, 16(v0)      // (JLONG)wsptr[4]
+    lw          s4, 4(v0)       // (JLONG)wsptr[1]
+    lw          s5, 12(v0)      // (JLONG)wsptr[3]
+    lw          s6, 20(v0)      // (JLONG)wsptr[5]
+    lw          s7, 28(v0)      // (JLONG)wsptr[7]
+    or          s4, s4, t0
+    or          s4, s4, t1
+    or          s4, s4, t3
+    or          s4, s4, s7
+    or          s4, s4, s5
+    or          s4, s4, s6
+    bnez        s4, 5f
+     addiu      v1, v1, -1
+    shra_r.w    s5, t2, 5
+    andi        s5, s5, 0x3ff
+    lbux        s5, s5(a3)
+    lw          s1, 0(a2)
+    replv.qb    s5, s5
+    usw         s5, 0(s1)
+    usw         s5, 4(s1)
+    b           6f
+     nop
+5:
+    addu        t4, t0, t1       // z2 + z3
+    addiu       t8, zero, 4433   // FIX_0_541196100
+    mul         t5, t4, t8       // z1 = MULTIPLY(z2 + z3, FIX_0_541196100)
+    addiu       t8, zero, 15137  // FIX_1_847759065
+    mul         t1, t1, t8       // MULTIPLY(z3, FIX_1_847759065)
+    addiu       t8, zero, 6270   // FIX_0_765366865
+    mul         t0, t0, t8       // MULTIPLY(z2, FIX_0_765366865)
+    addu        t4, t2, t3       // (JLONG)wsptr[0] + (JLONG)wsptr[4]
+    subu        t2, t2, t3       // (JLONG)wsptr[0] - (JLONG)wsptr[4]
+    sll         t4, t4, 13       // tmp0 = (wsptr[0] + wsptr[4]) << CONST_BITS
+    sll         t2, t2, 13       // tmp1 = (wsptr[0] - wsptr[4]) << CONST_BITS
+    subu        t1, t5, t1       // tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065)
+    subu        t3, t2, t1       // tmp12 = tmp1 - tmp2
+    addu        t2, t2, t1       // tmp11 = tmp1 + tmp2
+    addu        t5, t5, t0       // tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865)
+    subu        t1, t4, t5       // tmp13 = tmp0 - tmp3
+    addu        t0, t4, t5       // tmp10 = tmp0 + tmp3
+    lw          t4, 28(v0)       // tmp0 = (JLONG)wsptr[7]
+    lw          t6, 12(v0)       // tmp2 = (JLONG)wsptr[3]
+    lw          t5, 20(v0)       // tmp1 = (JLONG)wsptr[5]
+    lw          t7, 4(v0)        // tmp3 = (JLONG)wsptr[1]
+    addu        s0, t4, t6       // z3 = tmp0 + tmp2
+    addiu       t8, zero, 9633   // FIX_1_175875602
+    addu        s1, t5, t7       // z4 = tmp1 + tmp3
+    addu        s2, s0, s1       // z3 + z4
+    mul         s2, s2, t8       // z5 = MULTIPLY(z3 + z4, FIX_1_175875602)
+    addu        s3, t4, t7       // z1 = tmp0 + tmp3
+    addu        t9, t5, t6       // z2 = tmp1 + tmp2
+    addiu       t8, zero, 16069  // FIX_1_961570560
+    mul         s0, s0, t8       // -z3 = MULTIPLY(z3, FIX_1_961570560)
+    addiu       t8, zero, 3196   // FIX_0_390180644
+    mul         s1, s1, t8       // -z4 = MULTIPLY(z4, FIX_0_390180644)
+    addiu       t8, zero, 2446   // FIX_0_298631336
+    mul         t4, t4, t8       // tmp0 = MULTIPLY(tmp0, FIX_0_298631336)
+    addiu       t8, zero, 7373   // FIX_0_899976223
+    mul         s3, s3, t8       // -z1 = MULTIPLY(z1, FIX_0_899976223)
+    addiu       t8, zero, 16819  // FIX_2_053119869
+    mul         t5, t5, t8       // tmp1 = MULTIPLY(tmp1, FIX_2_053119869)
+    addiu       t8, zero, 20995  // FIX_2_562915447
+    mul         t9, t9, t8       // -z2 = MULTIPLY(z2, FIX_2_562915447)
+    addiu       t8, zero, 25172  // FIX_3_072711026
+    mul         t6, t6, t8       // tmp2 = MULTIPLY(tmp2, FIX_3_072711026)
+    addiu       t8, zero, 12299  // FIX_1_501321110
+    mul         t7, t7, t8       // tmp3 = MULTIPLY(tmp3, FIX_1_501321110)
+    subu        s0, s2, s0       // z3 += z5
+    subu        s1, s2, s1       // z4 += z5
+    addu        t4, t4, s0
+    subu        t4, t4, s3      // tmp0
+    addu        t5, t5, s1
+    subu        t5, t5, t9      // tmp1
+    addu        t6, t6, s0
+    subu        t6, t6, t9      // tmp2
+    addu        t7, t7, s1
+    subu        t7, t7, s3      // tmp3
+    addu        s0, t0, t7
+    subu        t0, t0, t7
+    addu        t7, t2, t6
+    subu        t2, t2, t6
+    addu        t6, t3, t5
+    subu        t3, t3, t5
+    addu        t5, t1, t4
+    subu        t1, t1, t4
+    shra_r.w    s0, s0, 18
+    shra_r.w    t7, t7, 18
+    shra_r.w    t6, t6, 18
+    shra_r.w    t5, t5, 18
+    shra_r.w    t1, t1, 18
+    shra_r.w    t3, t3, 18
+    shra_r.w    t2, t2, 18
+    shra_r.w    t0, t0, 18
+    andi        s0, s0, 0x3ff
+    andi        t7, t7, 0x3ff
+    andi        t6, t6, 0x3ff
+    andi        t5, t5, 0x3ff
+    andi        t1, t1, 0x3ff
+    andi        t3, t3, 0x3ff
+    andi        t2, t2, 0x3ff
+    andi        t0, t0, 0x3ff
+    lw          s1, 0(a2)
+    lbux        s0, s0(a3)
+    lbux        t7, t7(a3)
+    lbux        t6, t6(a3)
+    lbux        t5, t5(a3)
+    lbux        t1, t1(a3)
+    lbux        t3, t3(a3)
+    lbux        t2, t2(a3)
+    lbux        t0, t0(a3)
+    sb          s0, 0(s1)
+    sb          t7, 1(s1)
+    sb          t6, 2(s1)
+    sb          t5, 3(s1)
+    sb          t1, 4(s1)
+    sb          t3, 5(s1)
+    sb          t2, 6(s1)
+    sb          t0, 7(s1)
+6:
+    addiu       v0, v0, 32
+    bgtz        v1, 4b
+     addiu      a2, a2, 4
+    addiu       sp, sp, 256
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+
+END(jsimd_idct_islow_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_ifast_cols_dspr2)
+/*
+ * a0 = inptr
+ * a1 = quantptr
+ * a2 = wsptr
+ * a3 = mips_idct_ifast_coefs
+ */
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    addiu         t9, a0, 16      // end address
+    or            AT, a3, zero
+
+0:
+    lw            s0, 0(a1)       // quantptr[DCTSIZE*0]
+    lw            t0, 0(a0)       // inptr[DCTSIZE*0]
+    lw            t1, 16(a0)      // inptr[DCTSIZE*1]
+    muleq_s.w.phl v0, t0, s0      // tmp0 ...
+    lw            t2, 32(a0)      // inptr[DCTSIZE*2]
+    lw            t3, 48(a0)      // inptr[DCTSIZE*3]
+    lw            t4, 64(a0)      // inptr[DCTSIZE*4]
+    lw            t5, 80(a0)      // inptr[DCTSIZE*5]
+    muleq_s.w.phr t0, t0, s0      // ... tmp0 ...
+    lw            t6, 96(a0)      // inptr[DCTSIZE*6]
+    lw            t7, 112(a0)     // inptr[DCTSIZE*7]
+    or            s4, t1, t2
+    or            s5, t3, t4
+    bnez          s4, 1f
+     ins          t0, v0, 16, 16  // ... tmp0
+    bnez          s5, 1f
+     or           s6, t5, t6
+    or            s6, s6, t7
+    bnez          s6, 1f
+     sw           t0, 0(a2)       // wsptr[DCTSIZE*0]
+    sw            t0, 16(a2)      // wsptr[DCTSIZE*1]
+    sw            t0, 32(a2)      // wsptr[DCTSIZE*2]
+    sw            t0, 48(a2)      // wsptr[DCTSIZE*3]
+    sw            t0, 64(a2)      // wsptr[DCTSIZE*4]
+    sw            t0, 80(a2)      // wsptr[DCTSIZE*5]
+    sw            t0, 96(a2)      // wsptr[DCTSIZE*6]
+    sw            t0, 112(a2)     // wsptr[DCTSIZE*7]
+    addiu         a0, a0, 4
+    b             2f
+     addiu        a1, a1, 4
+
+1:
+    lw            s1, 32(a1)      // quantptr[DCTSIZE*2]
+    lw            s2, 64(a1)      // quantptr[DCTSIZE*4]
+    muleq_s.w.phl v0, t2, s1      // tmp1 ...
+    muleq_s.w.phr t2, t2, s1      // ... tmp1 ...
+    lw            s0, 16(a1)      // quantptr[DCTSIZE*1]
+    lw            s1, 48(a1)      // quantptr[DCTSIZE*3]
+    lw            s3, 96(a1)      // quantptr[DCTSIZE*6]
+    muleq_s.w.phl v1, t4, s2      // tmp2 ...
+    muleq_s.w.phr t4, t4, s2      // ... tmp2 ...
+    lw            s2, 80(a1)      // quantptr[DCTSIZE*5]
+    lw            t8, 4(AT)       // FIX(1.414213562)
+    ins           t2, v0, 16, 16  // ... tmp1
+    muleq_s.w.phl v0, t6, s3      // tmp3 ...
+    muleq_s.w.phr t6, t6, s3      // ... tmp3 ...
+    ins           t4, v1, 16, 16  // ... tmp2
+    addq.ph       s4, t0, t4      // tmp10
+    subq.ph       s5, t0, t4      // tmp11
+    ins           t6, v0, 16, 16  // ... tmp3
+    subq.ph       s6, t2, t6      // tmp12 ...
+    addq.ph       s7, t2, t6      // tmp13
+    mulq_s.ph     s6, s6, t8      // ... tmp12 ...
+    addq.ph       t0, s4, s7      // tmp0
+    subq.ph       t6, s4, s7      // tmp3
+    muleq_s.w.phl v0, t1, s0      // tmp4 ...
+    muleq_s.w.phr t1, t1, s0      // ... tmp4 ...
+    shll_s.ph     s6, s6, 1       // x2
+    lw            s3, 112(a1)     // quantptr[DCTSIZE*7]
+    subq.ph       s6, s6, s7      // ... tmp12
+    muleq_s.w.phl v1, t7, s3      // tmp7 ...
+    muleq_s.w.phr t7, t7, s3      // ... tmp7 ...
+    ins           t1, v0, 16, 16  // ... tmp4
+    addq.ph       t2, s5, s6      // tmp1
+    subq.ph       t4, s5, s6      // tmp2
+    muleq_s.w.phl v0, t5, s2      // tmp6 ...
+    muleq_s.w.phr t5, t5, s2      // ... tmp6 ...
+    ins           t7, v1, 16, 16  // ... tmp7
+    addq.ph       s5, t1, t7      // z11
+    subq.ph       s6, t1, t7      // z12
+    muleq_s.w.phl v1, t3, s1      // tmp5 ...
+    muleq_s.w.phr t3, t3, s1      // ... tmp5 ...
+    ins           t5, v0, 16, 16  // ... tmp6
+    ins           t3, v1, 16, 16  // ... tmp5
+    addq.ph       s7, t5, t3      // z13
+    subq.ph       v0, t5, t3      // z10
+    addq.ph       t7, s5, s7      // tmp7
+    subq.ph       s5, s5, s7      // tmp11 ...
+    addq.ph       v1, v0, s6      // z5 ...
+    mulq_s.ph     s5, s5, t8      // ... tmp11
+    lw            t8, 8(AT)       // FIX(1.847759065)
+    lw            s4, 0(AT)       // FIX(1.082392200)
+    addq.ph       s0, t0, t7
+    subq.ph       s1, t0, t7
+    mulq_s.ph     v1, v1, t8      // ... z5
+    shll_s.ph     s5, s5, 1       // x2
+    lw            t8, 12(AT)      // FIX(-2.613125930)
+    sw            s0, 0(a2)       // wsptr[DCTSIZE*0]
+    shll_s.ph     v0, v0, 1       // x4
+    mulq_s.ph     v0, v0, t8      // tmp12 ...
+    mulq_s.ph     s4, s6, s4      // tmp10 ...
+    shll_s.ph     v1, v1, 1       // x2
+    addiu         a0, a0, 4
+    addiu         a1, a1, 4
+    sw            s1, 112(a2)     // wsptr[DCTSIZE*7]
+    shll_s.ph     s6, v0, 1       // x4
+    shll_s.ph     s4, s4, 1       // x2
+    addq.ph       s6, s6, v1      // ... tmp12
+    subq.ph       t5, s6, t7      // tmp6
+    subq.ph       s4, s4, v1      // ... tmp10
+    subq.ph       t3, s5, t5      // tmp5
+    addq.ph       s2, t2, t5
+    addq.ph       t1, s4, t3      // tmp4
+    subq.ph       s3, t2, t5
+    sw            s2, 16(a2)      // wsptr[DCTSIZE*1]
+    sw            s3, 96(a2)      // wsptr[DCTSIZE*6]
+    addq.ph       v0, t4, t3
+    subq.ph       v1, t4, t3
+    sw            v0, 32(a2)      // wsptr[DCTSIZE*2]
+    sw            v1, 80(a2)      // wsptr[DCTSIZE*5]
+    addq.ph       v0, t6, t1
+    subq.ph       v1, t6, t1
+    sw            v0, 64(a2)      // wsptr[DCTSIZE*4]
+    sw            v1, 48(a2)      // wsptr[DCTSIZE*3]
+
+2:
+    bne           a0, t9, 0b
+     addiu        a2, a2, 4
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j             ra
+     nop
+
+END(jsimd_idct_ifast_cols_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_ifast_rows_dspr2)
+/*
+ * a0 = wsptr
+ * a1 = output_buf
+ * a2 = output_col
+ * a3 = mips_idct_ifast_coefs
+ */
+    SAVE_REGS_ON_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, s8, a3
+
+    addiu         t9, a0, 128     // end address
+    lui           s8, 0x8080
+    ori           s8, s8, 0x8080
+
+0:
+    lw            AT, 36(sp)      // restore $a3 (mips_idct_ifast_coefs)
+    lw            t0, 0(a0)       // wsptr[DCTSIZE*0+0/1]  b a
+    lw            s0, 16(a0)      // wsptr[DCTSIZE*1+0/1]  B A
+    lw            t2, 4(a0)       // wsptr[DCTSIZE*0+2/3]  d c
+    lw            s2, 20(a0)      // wsptr[DCTSIZE*1+2/3]  D C
+    lw            t4, 8(a0)       // wsptr[DCTSIZE*0+4/5]  f e
+    lw            s4, 24(a0)      // wsptr[DCTSIZE*1+4/5]  F E
+    lw            t6, 12(a0)      // wsptr[DCTSIZE*0+6/7]  h g
+    lw            s6, 28(a0)      // wsptr[DCTSIZE*1+6/7]  H G
+    precrq.ph.w   t1, s0, t0      // B b
+    ins           t0, s0, 16, 16  // A a
+    bnez          t1, 1f
+     or           s0, t2, s2
+    bnez          s0, 1f
+     or           s0, t4, s4
+    bnez          s0, 1f
+     or           s0, t6, s6
+    bnez          s0, 1f
+     shll_s.ph    s0, t0, 2       // A a
+    lw            a3, 0(a1)
+    lw            AT, 4(a1)
+    precrq.ph.w   t0, s0, s0      // A A
+    ins           s0, s0, 16, 16  // a a
+    addu          a3, a3, a2
+    addu          AT, AT, a2
+    precrq.qb.ph  t0, t0, t0      // A A A A
+    precrq.qb.ph  s0, s0, s0      // a a a a
+    addu.qb       s0, s0, s8
+    addu.qb       t0, t0, s8
+    sw            s0, 0(a3)
+    sw            s0, 4(a3)
+    sw            t0, 0(AT)
+    sw            t0, 4(AT)
+    addiu         a0, a0, 32
+    bne           a0, t9, 0b
+     addiu        a1, a1, 8
+    b             2f
+     nop
+
+1:
+    precrq.ph.w   t3, s2, t2
+    ins           t2, s2, 16, 16
+    precrq.ph.w   t5, s4, t4
+    ins           t4, s4, 16, 16
+    precrq.ph.w   t7, s6, t6
+    ins           t6, s6, 16, 16
+    lw            t8, 4(AT)       // FIX(1.414213562)
+    addq.ph       s4, t0, t4      // tmp10
+    subq.ph       s5, t0, t4      // tmp11
+    subq.ph       s6, t2, t6      // tmp12 ...
+    addq.ph       s7, t2, t6      // tmp13
+    mulq_s.ph     s6, s6, t8      // ... tmp12 ...
+    addq.ph       t0, s4, s7      // tmp0
+    subq.ph       t6, s4, s7      // tmp3
+    shll_s.ph     s6, s6, 1       // x2
+    subq.ph       s6, s6, s7      // ... tmp12
+    addq.ph       t2, s5, s6      // tmp1
+    subq.ph       t4, s5, s6      // tmp2
+    addq.ph       s5, t1, t7      // z11
+    subq.ph       s6, t1, t7      // z12
+    addq.ph       s7, t5, t3      // z13
+    subq.ph       v0, t5, t3      // z10
+    addq.ph       t7, s5, s7      // tmp7
+    subq.ph       s5, s5, s7      // tmp11 ...
+    addq.ph       v1, v0, s6      // z5 ...
+    mulq_s.ph     s5, s5, t8      // ... tmp11
+    lw            t8, 8(AT)       // FIX(1.847759065)
+    lw            s4, 0(AT)       // FIX(1.082392200)
+    addq.ph       s0, t0, t7      // tmp0 + tmp7
+    subq.ph       s7, t0, t7      // tmp0 - tmp7
+    mulq_s.ph     v1, v1, t8      // ... z5
+    lw            a3, 0(a1)
+    lw            t8, 12(AT)      // FIX(-2.613125930)
+    shll_s.ph     s5, s5, 1       // x2
+    addu          a3, a3, a2
+    shll_s.ph     v0, v0, 1       // x4
+    mulq_s.ph     v0, v0, t8      // tmp12 ...
+    mulq_s.ph     s4, s6, s4      // tmp10 ...
+    shll_s.ph     v1, v1, 1       // x2
+    addiu         a0, a0, 32
+    addiu         a1, a1, 8
+    shll_s.ph     s6, v0, 1       // x4
+    shll_s.ph     s4, s4, 1       // x2
+    addq.ph       s6, s6, v1      // ... tmp12
+    shll_s.ph     s0, s0, 2
+    subq.ph       t5, s6, t7      // tmp6
+    subq.ph       s4, s4, v1      // ... tmp10
+    subq.ph       t3, s5, t5      // tmp5
+    shll_s.ph     s7, s7, 2
+    addq.ph       t1, s4, t3      // tmp4
+    addq.ph       s1, t2, t5      // tmp1 + tmp6
+    subq.ph       s6, t2, t5      // tmp1 - tmp6
+    addq.ph       s2, t4, t3      // tmp2 + tmp5
+    subq.ph       s5, t4, t3      // tmp2 - tmp5
+    addq.ph       s4, t6, t1      // tmp3 + tmp4
+    subq.ph       s3, t6, t1      // tmp3 - tmp4
+    shll_s.ph     s1, s1, 2
+    shll_s.ph     s2, s2, 2
+    shll_s.ph     s3, s3, 2
+    shll_s.ph     s4, s4, 2
+    shll_s.ph     s5, s5, 2
+    shll_s.ph     s6, s6, 2
+    precrq.ph.w   t0, s1, s0      // B A
+    ins           s0, s1, 16, 16  // b a
+    precrq.ph.w   t2, s3, s2      // D C
+    ins           s2, s3, 16, 16  // d c
+    precrq.ph.w   t4, s5, s4      // F E
+    ins           s4, s5, 16, 16  // f e
+    precrq.ph.w   t6, s7, s6      // H G
+    ins           s6, s7, 16, 16  // h g
+    precrq.qb.ph  t0, t2, t0      // D C B A
+    precrq.qb.ph  s0, s2, s0      // d c b a
+    precrq.qb.ph  t4, t6, t4      // H G F E
+    precrq.qb.ph  s4, s6, s4      // h g f e
+    addu.qb       s0, s0, s8
+    addu.qb       s4, s4, s8
+    sw            s0, 0(a3)       // outptr[0/1/2/3]       d c b a
+    sw            s4, 4(a3)       // outptr[4/5/6/7]       h g f e
+    lw            a3, -4(a1)
+    addu.qb       t0, t0, s8
+    addu          a3, a3, a2
+    addu.qb       t4, t4, s8
+    sw            t0, 0(a3)       // outptr[0/1/2/3]       D C B A
+    bne           a0, t9, 0b
+     sw           t4, 4(a3)       // outptr[4/5/6/7]       H G F E
+
+2:
+
+    RESTORE_REGS_FROM_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, s8, a3
+
+    j             ra
+     nop
+
+END(jsimd_idct_ifast_rows_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_fdct_islow_dspr2)
+/*
+ * a0 = data
+ */
+    SAVE_REGS_ON_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, s8
+
+    lui         t0, 6437
+    ori         t0, 2260
+    lui         t1, 9633
+    ori         t1, 11363
+    lui         t2, 0xd39e
+    ori         t2, 0xe6dc
+    lui         t3, 0xf72d
+    ori         t3, 9633
+    lui         t4, 2261
+    ori         t4, 9633
+    lui         t5, 0xd39e
+    ori         t5, 6437
+    lui         t6, 9633
+    ori         t6, 0xd39d
+    lui         t7, 0xe6dc
+    ori         t7, 2260
+    lui         t8, 4433
+    ori         t8, 10703
+    lui         t9, 0xd630
+    ori         t9, 4433
+    li          s8, 8
+    move        a1, a0
+1:
+    lw          s0, 0(a1)       // tmp0 = 1|0
+    lw          s1, 4(a1)       // tmp1 = 3|2
+    lw          s2, 8(a1)       // tmp2 = 5|4
+    lw          s3, 12(a1)      // tmp3 = 7|6
+    packrl.ph   s1, s1, s1      // tmp1 = 2|3
+    packrl.ph   s3, s3, s3      // tmp3 = 6|7
+    subq.ph     s7, s1, s2      // tmp7 = 2-5|3-4 = t5|t4
+    subq.ph     s5, s0, s3      // tmp5 = 1-6|0-7 = t6|t7
+    mult        $0, $0          // ac0  = 0
+    dpa.w.ph    $ac0, s7, t0    // ac0 += t5*  6437 + t4*  2260
+    dpa.w.ph    $ac0, s5, t1    // ac0 += t6*  9633 + t7* 11363
+    mult        $ac1, $0, $0    // ac1  = 0
+    dpa.w.ph    $ac1, s7, t2    // ac1 += t5*-11362 + t4* -6436
+    dpa.w.ph    $ac1, s5, t3    // ac1 += t6* -2259 + t7*  9633
+    mult        $ac2, $0, $0    // ac2  = 0
+    dpa.w.ph    $ac2, s7, t4    // ac2 += t5*  2261 + t4*  9633
+    dpa.w.ph    $ac2, s5, t5    // ac2 += t6*-11362 + t7*  6437
+    mult        $ac3, $0, $0    // ac3  = 0
+    dpa.w.ph    $ac3, s7, t6    // ac3 += t5*  9633 + t4*-11363
+    dpa.w.ph    $ac3, s5, t7    // ac3 += t6* -6436 + t7*  2260
+    addq.ph     s6, s1, s2      // tmp6 = 2+5|3+4 = t2|t3
+    addq.ph     s4, s0, s3      // tmp4 = 1+6|0+7 = t1|t0
+    extr_r.w    s0, $ac0, 11    // tmp0 = (ac0 + 1024) >> 11
+    extr_r.w    s1, $ac1, 11    // tmp1 = (ac1 + 1024) >> 11
+    extr_r.w    s2, $ac2, 11    // tmp2 = (ac2 + 1024) >> 11
+    extr_r.w    s3, $ac3, 11    // tmp3 = (ac3 + 1024) >> 11
+    addq.ph     s5, s4, s6      // tmp5 = t1+t2|t0+t3 = t11|t10
+    subq.ph     s7, s4, s6      // tmp7 = t1-t2|t0-t3 = t12|t13
+    sh          s0, 2(a1)
+    sh          s1, 6(a1)
+    sh          s2, 10(a1)
+    sh          s3, 14(a1)
+    mult        $0, $0          // ac0  = 0
+    dpa.w.ph    $ac0, s7, t8    // ac0 += t12*  4433 + t13* 10703
+    mult        $ac1, $0, $0    // ac1  = 0
+    dpa.w.ph    $ac1, s7, t9    // ac1 += t12*-10704 + t13*  4433
+    sra         s4, s5, 16      // tmp4 = t11
+    addiu       a1, a1, 16
+    addiu       s8, s8, -1
+    extr_r.w    s0, $ac0, 11    // tmp0 = (ac0 + 1024) >> 11
+    extr_r.w    s1, $ac1, 11    // tmp1 = (ac1 + 1024) >> 11
+    addu        s2, s5, s4      // tmp2 = t10 + t11
+    subu        s3, s5, s4      // tmp3 = t10 - t11
+    sll         s2, s2, 2       // tmp2 = (t10 + t11) << 2
+    sll         s3, s3, 2       // tmp3 = (t10 - t11) << 2
+    sh          s2, -16(a1)
+    sh          s3, -8(a1)
+    sh          s0, -12(a1)
+    bgtz        s8, 1b
+     sh         s1, -4(a1)
+    li          t0, 2260
+    li          t1, 11363
+    li          t2, 9633
+    li          t3, 6436
+    li          t4, 6437
+    li          t5, 2261
+    li          t6, 11362
+    li          t7, 2259
+    li          t8, 4433
+    li          t9, 10703
+    li          a1, 10704
+    li          s8, 8
+
+2:
+    lh          a2, 0(a0)       // 0
+    lh          a3, 16(a0)      // 8
+    lh          v0, 32(a0)      // 16
+    lh          v1, 48(a0)      // 24
+    lh          s4, 64(a0)      // 32
+    lh          s5, 80(a0)      // 40
+    lh          s6, 96(a0)      // 48
+    lh          s7, 112(a0)     // 56
+    addu        s2, v0, s5      // tmp2 = 16 + 40
+    subu        s5, v0, s5      // tmp5 = 16 - 40
+    addu        s3, v1, s4      // tmp3 = 24 + 32
+    subu        s4, v1, s4      // tmp4 = 24 - 32
+    addu        s0, a2, s7      // tmp0 =  0 + 56
+    subu        s7, a2, s7      // tmp7 =  0 - 56
+    addu        s1, a3, s6      // tmp1 =  8 + 48
+    subu        s6, a3, s6      // tmp6 =  8 - 48
+    addu        a2, s0, s3      // tmp10 = tmp0 + tmp3
+    subu        v1, s0, s3      // tmp13 = tmp0 - tmp3
+    addu        a3, s1, s2      // tmp11 = tmp1 + tmp2
+    subu        v0, s1, s2      // tmp12 = tmp1 - tmp2
+    mult        s7, t1          // ac0  = tmp7 * c1
+    madd        s4, t0          // ac0 += tmp4 * c0
+    madd        s5, t4          // ac0 += tmp5 * c4
+    madd        s6, t2          // ac0 += tmp6 * c2
+    mult        $ac1, s7, t2    // ac1  = tmp7 * c2
+    msub        $ac1, s4, t3    // ac1 -= tmp4 * c3
+    msub        $ac1, s5, t6    // ac1 -= tmp5 * c6
+    msub        $ac1, s6, t7    // ac1 -= tmp6 * c7
+    mult        $ac2, s7, t4    // ac2  = tmp7 * c4
+    madd        $ac2, s4, t2    // ac2 += tmp4 * c2
+    madd        $ac2, s5, t5    // ac2 += tmp5 * c5
+    msub        $ac2, s6, t6    // ac2 -= tmp6 * c6
+    mult        $ac3, s7, t0    // ac3  = tmp7 * c0
+    msub        $ac3, s4, t1    // ac3 -= tmp4 * c1
+    madd        $ac3, s5, t2    // ac3 += tmp5 * c2
+    msub        $ac3, s6, t3    // ac3 -= tmp6 * c3
+    extr_r.w    s0, $ac0, 15    // tmp0 = (ac0 + 16384) >> 15
+    extr_r.w    s1, $ac1, 15    // tmp1 = (ac1 + 16384) >> 15
+    extr_r.w    s2, $ac2, 15    // tmp2 = (ac2 + 16384) >> 15
+    extr_r.w    s3, $ac3, 15    // tmp3 = (ac3 + 16384) >> 15
+    addiu       s8, s8, -1
+    addu        s4, a2, a3      // tmp4 = tmp10 + tmp11
+    subu        s5, a2, a3      // tmp5 = tmp10 - tmp11
+    sh          s0, 16(a0)
+    sh          s1, 48(a0)
+    sh          s2, 80(a0)
+    sh          s3, 112(a0)
+    mult        v0, t8          // ac0  = tmp12 * c8
+    madd        v1, t9          // ac0 += tmp13 * c9
+    mult        $ac1, v1, t8    // ac1  = tmp13 * c8
+    msub        $ac1, v0, a1    // ac1 -= tmp12 * c10
+    addiu       a0, a0, 2
+    extr_r.w    s6, $ac0, 15    // tmp6 = (ac0 + 16384) >> 15
+    extr_r.w    s7, $ac1, 15    // tmp7 = (ac1 + 16384) >> 15
+    shra_r.w    s4, s4, 2       // tmp4 = (tmp4 + 2) >> 2
+    shra_r.w    s5, s5, 2       // tmp5 = (tmp5 + 2) >> 2
+    sh          s4, -2(a0)
+    sh          s5, 62(a0)
+    sh          s6, 30(a0)
+    bgtz        s8, 2b
+     sh         s7, 94(a0)
+
+    RESTORE_REGS_FROM_STACK 40, s0, s1, s2, s3, s4, s5, s6, s7, s8
+
+    jr          ra
+     nop
+
+END(jsimd_fdct_islow_dspr2)
+
+
+/**************************************************************************/
+LEAF_DSPR2(jsimd_fdct_ifast_dspr2)
+/*
+ * a0 = data
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 8, s0, s1
+
+    li          a1, 0x014e014e  // FIX_1_306562965 (334 << 16)|(334 & 0xffff)
+    li          a2, 0x008b008b  // FIX_0_541196100 (139 << 16)|(139 & 0xffff)
+    li          a3, 0x00620062  // FIX_0_382683433 (98 << 16) |(98 & 0xffff)
+    li          s1, 0x00b500b5  // FIX_0_707106781 (181 << 16)|(181 & 0xffff)
+
+    move        v0, a0
+    addiu       v1, v0, 128     // end address
+
+0:
+    lw          t0, 0(v0)       // tmp0 = 1|0
+    lw          t1, 4(v0)       // tmp1 = 3|2
+    lw          t2, 8(v0)       // tmp2 = 5|4
+    lw          t3, 12(v0)      // tmp3 = 7|6
+    packrl.ph   t1, t1, t1      // tmp1 = 2|3
+    packrl.ph   t3, t3, t3      // tmp3 = 6|7
+    subq.ph     t7, t1, t2      // tmp7 = 2-5|3-4 = t5|t4
+    subq.ph     t5, t0, t3      // tmp5 = 1-6|0-7 = t6|t7
+    addq.ph     t6, t1, t2      // tmp6 = 2+5|3+4 = t2|t3
+    addq.ph     t4, t0, t3      // tmp4 = 1+6|0+7 = t1|t0
+    addq.ph     t8, t4, t6      // tmp5 = t1+t2|t0+t3 = t11|t10
+    subq.ph     t9, t4, t6      // tmp7 = t1-t2|t0-t3 = t12|t13
+    sra         t4, t8, 16      // tmp4 = t11
+    mult        $0, $0          // ac0  = 0
+    dpa.w.ph    $ac0, t9, s1
+    mult        $ac1, $0, $0    // ac1  = 0
+    dpa.w.ph    $ac1, t7, a3    // ac1 += t4*98 + t5*98
+    dpsx.w.ph   $ac1, t5, a3    // ac1 += t6*98 + t7*98
+    mult        $ac2, $0, $0    // ac2  = 0
+    dpa.w.ph    $ac2, t7, a2    // ac2 += t4*139 + t5*139
+    mult        $ac3, $0, $0    // ac3  = 0
+    dpa.w.ph    $ac3, t5, a1    // ac3 += t6*334 + t7*334
+    precrq.ph.w t0, t5, t7      // t0 = t5|t6
+    addq.ph     t2, t8, t4      // tmp2 = t10 + t11
+    subq.ph     t3, t8, t4      // tmp3 = t10 - t11
+    extr.w      t4, $ac0, 8
+    mult        $0, $0          // ac0  = 0
+    dpa.w.ph    $ac0, t0, s1    // ac0 += t5*181 + t6*181
+    extr.w      t0, $ac1, 8     // t0 = z5
+    extr.w      t1, $ac2, 8     // t1 = MULTIPLY(tmp10, 139)
+    extr.w      t7, $ac3, 8     // t2 = MULTIPLY(tmp12, 334)
+    extr.w      t8, $ac0, 8     // t8 = z3 = MULTIPLY(tmp11, 181)
+    add         t6, t1, t0      // t6 = z2
+    add         t7, t7, t0      // t7 = z4
+    subq.ph     t0, t5, t8      // t0 = z13 = tmp7 - z3
+    addq.ph     t8, t5, t8      // t9 = z11 = tmp7 + z3
+    addq.ph     t1, t0, t6      // t1 = z13 + z2
+    subq.ph     t6, t0, t6      // t6 = z13 - z2
+    addq.ph     t0, t8, t7      // t0 = z11 + z4
+    subq.ph     t7, t8, t7      // t7 = z11 - z4
+    addq.ph     t5, t4, t9
+    subq.ph     t4, t9, t4
+    sh          t2, 0(v0)
+    sh          t5, 4(v0)
+    sh          t3, 8(v0)
+    sh          t4, 12(v0)
+    sh          t1, 10(v0)
+    sh          t6, 6(v0)
+    sh          t0, 2(v0)
+    sh          t7, 14(v0)
+    addiu       v0, 16
+    bne         v1, v0, 0b
+     nop
+    move        v0, a0
+    addiu       v1, v0, 16
+
+1:
+    lh          t0, 0(v0)       // 0
+    lh          t1, 16(v0)      // 8
+    lh          t2, 32(v0)      // 16
+    lh          t3, 48(v0)      // 24
+    lh          t4, 64(v0)      // 32
+    lh          t5, 80(v0)      // 40
+    lh          t6, 96(v0)      // 48
+    lh          t7, 112(v0)     // 56
+    add         t8, t0, t7      // t8 = tmp0
+    sub         t7, t0, t7      // t7 = tmp7
+    add         t0, t1, t6      // t0 = tmp1
+    sub         t1, t1, t6      // t1 = tmp6
+    add         t6, t2, t5      // t6 = tmp2
+    sub         t5, t2, t5      // t5 = tmp5
+    add         t2, t3, t4      // t2 = tmp3
+    sub         t3, t3, t4      // t3 = tmp4
+    add         t4, t8, t2      // t4 = tmp10 = tmp0 + tmp3
+    sub         t8, t8, t2      // t8 = tmp13 = tmp0 - tmp3
+    sub         s0, t0, t6      // s0 = tmp12 = tmp1 - tmp2
+    ins         t8, s0, 16, 16  // t8 = tmp12|tmp13
+    add         t2, t0, t6      // t2 = tmp11 = tmp1 + tmp2
+    mult        $0, $0          // ac0  = 0
+    dpa.w.ph    $ac0, t8, s1    // ac0 += t12*181 + t13*181
+    add         s0, t4, t2      // t8 = tmp10+tmp11
+    sub         t4, t4, t2      // t4 = tmp10-tmp11
+    sh          s0, 0(v0)
+    sh          t4, 64(v0)
+    extr.w      t2, $ac0, 8     // z1 = MULTIPLY(tmp12+tmp13, FIX_0_707106781)
+    addq.ph     t4, t8, t2      // t9 = tmp13 + z1
+    subq.ph     t8, t8, t2      // t2 = tmp13 - z1
+    sh          t4, 32(v0)
+    sh          t8, 96(v0)
+    add         t3, t3, t5      // t3 = tmp10 = tmp4 + tmp5
+    add         t0, t5, t1      // t0 = tmp11 = tmp5 + tmp6
+    add         t1, t1, t7      // t1 = tmp12 = tmp6 + tmp7
+    andi        t4, a1, 0xffff
+    mul         s0, t1, t4
+    sra         s0, s0, 8       // s0 = z4 = MULTIPLY(tmp12, FIX_1_306562965)
+    ins         t1, t3, 16, 16  // t1 = tmp10|tmp12
+    mult        $0, $0          // ac0  = 0
+    mulsa.w.ph  $ac0, t1, a3    // ac0 += t10*98 - t12*98
+    extr.w      t8, $ac0, 8     // z5 = MULTIPLY(tmp10-tmp12, FIX_0_382683433)
+    add         t2, t7, t8      // t2 = tmp7 + z5
+    sub         t7, t7, t8      // t7 = tmp7 - z5
+    andi        t4, a2, 0xffff
+    mul         t8, t3, t4
+    sra         t8, t8, 8       // t8 = z2 = MULTIPLY(tmp10, FIX_0_541196100)
+    andi        t4, s1, 0xffff
+    mul         t6, t0, t4
+    sra         t6, t6, 8       // t6 = z3 = MULTIPLY(tmp11, FIX_0_707106781)
+    add         t0, t6, t8      // t0 = z3 + z2
+    sub         t1, t6, t8      // t1 = z3 - z2
+    add         t3, t6, s0      // t3 = z3 + z4
+    sub         t4, t6, s0      // t4 = z3 - z4
+    sub         t5, t2, t1      // t5 = dataptr[5]
+    sub         t6, t7, t0      // t6 = dataptr[3]
+    add         t3, t2, t3      // t3 = dataptr[1]
+    add         t4, t7, t4      // t4 = dataptr[7]
+    sh          t5, 80(v0)
+    sh          t6, 48(v0)
+    sh          t3, 16(v0)
+    sh          t4, 112(v0)
+    addiu       v0, 2
+    bne         v0, v1, 1b
+     nop
+
+    RESTORE_REGS_FROM_STACK 8, s0, s1
+
+    j           ra
+     nop
+END(jsimd_fdct_ifast_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_quantize_dspr2)
+/*
+ * a0 = coef_block
+ * a1 = divisors
+ * a2 = workspace
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 16, s0, s1, s2
+
+    addiu       v0, a2, 124     // v0 = workspace_end
+    lh          t0, 0(a2)
+    lh          t1, 0(a1)
+    lh          t2, 128(a1)
+    sra         t3, t0, 15
+    sll         t3, t3, 1
+    addiu       t3, t3, 1
+    mul         t0, t0, t3
+    lh          t4, 384(a1)
+    lh          t5, 130(a1)
+    lh          t6, 2(a2)
+    lh          t7, 2(a1)
+    lh          t8, 386(a1)
+
+1:
+    andi        t1, 0xffff
+    add         t9, t0, t2
+    andi        t9, 0xffff
+    mul         v1, t9, t1
+    sra         s0, t6, 15
+    sll         s0, s0, 1
+    addiu       s0, s0, 1
+    addiu       t9, t4, 16
+    srav        v1, v1, t9
+    mul         v1, v1, t3
+    mul         t6, t6, s0
+    andi        t7, 0xffff
+    addiu       a2, a2, 4
+    addiu       a1, a1, 4
+    add         s1, t6, t5
+    andi        s1, 0xffff
+    sh          v1, 0(a0)
+
+    mul         s2, s1, t7
+    addiu       s1, t8, 16
+    srav        s2, s2, s1
+    mul         s2, s2, s0
+    lh          t0, 0(a2)
+    lh          t1, 0(a1)
+    sra         t3, t0, 15
+    sll         t3, t3, 1
+    addiu       t3, t3, 1
+    mul         t0, t0, t3
+    lh          t2, 128(a1)
+    lh          t4, 384(a1)
+    lh          t5, 130(a1)
+    lh          t8, 386(a1)
+    lh          t6, 2(a2)
+    lh          t7, 2(a1)
+    sh          s2, 2(a0)
+    lh          t0, 0(a2)
+    sra         t3, t0, 15
+    sll         t3, t3, 1
+    addiu       t3, t3, 1
+    mul         t0, t0, t3
+    bne         a2, v0, 1b
+     addiu      a0, a0, 4
+
+    andi        t1, 0xffff
+    add         t9, t0, t2
+    andi        t9, 0xffff
+    mul         v1, t9, t1
+    sra         s0, t6, 15
+    sll         s0, s0, 1
+    addiu       s0, s0, 1
+    addiu       t9, t4, 16
+    srav        v1, v1, t9
+    mul         v1, v1, t3
+    mul         t6, t6, s0
+    andi        t7, 0xffff
+    sh          v1, 0(a0)
+    add         s1, t6, t5
+    andi        s1, 0xffff
+    mul         s2, s1, t7
+    addiu       s1, t8, 16
+    addiu       a2, a2, 4
+    addiu       a1, a1, 4
+    srav        s2, s2, s1
+    mul         s2, s2, s0
+    sh          s2, 2(a0)
+
+    RESTORE_REGS_FROM_STACK 16, s0, s1, s2
+
+    j           ra
+     nop
+
+END(jsimd_quantize_dspr2)
+
+
+#ifndef __mips_soft_float
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_quantize_float_dspr2)
+/*
+ * a0 = coef_block
+ * a1 = divisors
+ * a2 = workspace
+ */
+    .set at
+
+    li          t1, 0x46800100  // integer representation 16384.5
+    mtc1        t1, f0
+    li          t0, 63
+0:
+    lwc1        f2, 0(a2)
+    lwc1        f10, 0(a1)
+    lwc1        f4, 4(a2)
+    lwc1        f12, 4(a1)
+    lwc1        f6, 8(a2)
+    lwc1        f14, 8(a1)
+    lwc1        f8, 12(a2)
+    lwc1        f16, 12(a1)
+    madd.s      f2, f0, f2, f10
+    madd.s      f4, f0, f4, f12
+    madd.s      f6, f0, f6, f14
+    madd.s      f8, f0, f8, f16
+    lwc1        f10, 16(a1)
+    lwc1        f12, 20(a1)
+    trunc.w.s   f2, f2
+    trunc.w.s   f4, f4
+    trunc.w.s   f6, f6
+    trunc.w.s   f8, f8
+    lwc1        f14, 24(a1)
+    lwc1        f16, 28(a1)
+    mfc1        t1, f2
+    mfc1        t2, f4
+    mfc1        t3, f6
+    mfc1        t4, f8
+    lwc1        f2, 16(a2)
+    lwc1        f4, 20(a2)
+    lwc1        f6, 24(a2)
+    lwc1        f8, 28(a2)
+    madd.s      f2, f0, f2, f10
+    madd.s      f4, f0, f4, f12
+    madd.s      f6, f0, f6, f14
+    madd.s      f8, f0, f8, f16
+    addiu       t1, t1, -16384
+    addiu       t2, t2, -16384
+    addiu       t3, t3, -16384
+    addiu       t4, t4, -16384
+    trunc.w.s   f2, f2
+    trunc.w.s   f4, f4
+    trunc.w.s   f6, f6
+    trunc.w.s   f8, f8
+    sh          t1, 0(a0)
+    sh          t2, 2(a0)
+    sh          t3, 4(a0)
+    sh          t4, 6(a0)
+    mfc1        t1, f2
+    mfc1        t2, f4
+    mfc1        t3, f6
+    mfc1        t4, f8
+    addiu       t0, t0, -8
+    addiu       a2, a2, 32
+    addiu       a1, a1, 32
+    addiu       t1, t1, -16384
+    addiu       t2, t2, -16384
+    addiu       t3, t3, -16384
+    addiu       t4, t4, -16384
+    sh          t1, 8(a0)
+    sh          t2, 10(a0)
+    sh          t3, 12(a0)
+    sh          t4, 14(a0)
+    bgez        t0, 0b
+     addiu      a0, a0, 16
+
+    j           ra
+     nop
+
+END(jsimd_quantize_float_dspr2)
+
+#endif
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_2x2_dspr2)
+/*
+ * a0 = compptr->dct_table
+ * a1 = coef_block
+ * a2 = output_buf
+ * a3 = output_col
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 24, s0, s1, s2, s3, s4, s5
+
+    addiu       sp, sp, -40
+    move        v0, sp
+    addiu       s2, zero, 29692
+    addiu       s3, zero, -10426
+    addiu       s4, zero, 6967
+    addiu       s5, zero, -5906
+    lh          t0, 0(a1)       // t0 = inptr[DCTSIZE*0]
+    lh          t5, 0(a0)       // t5 = quantptr[DCTSIZE*0]
+    lh          t1, 48(a1)      // t1 = inptr[DCTSIZE*3]
+    lh          t6, 48(a0)      // t6 = quantptr[DCTSIZE*3]
+    mul         t4, t5, t0
+    lh          t0, 16(a1)      // t0 = inptr[DCTSIZE*1]
+    lh          t5, 16(a0)      // t5 = quantptr[DCTSIZE*1]
+    mul         t6, t6, t1
+    mul         t5, t5, t0
+    lh          t2, 80(a1)      // t2 = inptr[DCTSIZE*5]
+    lh          t7, 80(a0)      // t7 = quantptr[DCTSIZE*5]
+    lh          t3, 112(a1)     // t3 = inptr[DCTSIZE*7]
+    lh          t8, 112(a0)     // t8 = quantptr[DCTSIZE*7]
+    mul         t7, t7, t2
+    mult        zero, zero
+    mul         t8, t8, t3
+    li          s0, 0x73FCD746  // s0 = (29692 << 16) | (-10426 & 0xffff)
+    li          s1, 0x1B37E8EE  // s1 = (6967 << 16) | (-5906 & 0xffff)
+    ins         t6, t5, 16, 16  // t6 = t5|t6
+    sll         t4, t4, 15
+    dpa.w.ph    $ac0, t6, s0
+    lh          t1, 2(a1)
+    lh          t6, 2(a0)
+    ins         t8, t7, 16, 16  // t8 = t7|t8
+    dpa.w.ph    $ac0, t8, s1
+    mflo        t0, $ac0
+    mul         t5, t6, t1
+    lh          t1, 18(a1)
+    lh          t6, 18(a0)
+    lh          t2, 50(a1)
+    lh          t7, 50(a0)
+    mul         t6, t6, t1
+    subu        t8, t4, t0
+    mul         t7, t7, t2
+    addu        t0, t4, t0
+    shra_r.w    t0, t0, 13
+    lh          t1, 82(a1)
+    lh          t2, 82(a0)
+    lh          t3, 114(a1)
+    lh          t4, 114(a0)
+    shra_r.w    t8, t8, 13
+    mul         t1, t1, t2
+    mul         t3, t3, t4
+    sw          t0, 0(v0)
+    sw          t8, 20(v0)
+    sll         t4, t5, 15
+    ins         t7, t6, 16, 16
+    mult        zero, zero
+    dpa.w.ph    $ac0, t7, s0
+    ins         t3, t1, 16, 16
+    lh          t1, 6(a1)
+    lh          t6, 6(a0)
+    dpa.w.ph    $ac0, t3, s1
+    mflo        t0, $ac0
+    mul         t5, t6, t1
+    lh          t1, 22(a1)
+    lh          t6, 22(a0)
+    lh          t2, 54(a1)
+    lh          t7, 54(a0)
+    mul         t6, t6, t1
+    subu        t8, t4, t0
+    mul         t7, t7, t2
+    addu        t0, t4, t0
+    shra_r.w    t0, t0, 13
+    lh          t1, 86(a1)
+    lh          t2, 86(a0)
+    lh          t3, 118(a1)
+    lh          t4, 118(a0)
+    shra_r.w    t8, t8, 13
+    mul         t1, t1, t2
+    mul         t3, t3, t4
+    sw          t0, 4(v0)
+    sw          t8, 24(v0)
+    sll         t4, t5, 15
+    ins         t7, t6, 16, 16
+    mult        zero, zero
+    dpa.w.ph    $ac0, t7, s0
+    ins         t3, t1, 16, 16
+    lh          t1, 10(a1)
+    lh          t6, 10(a0)
+    dpa.w.ph    $ac0, t3, s1
+    mflo        t0, $ac0
+    mul         t5, t6, t1
+    lh          t1, 26(a1)
+    lh          t6, 26(a0)
+    lh          t2, 58(a1)
+    lh          t7, 58(a0)
+    mul         t6, t6, t1
+    subu        t8, t4, t0
+    mul         t7, t7, t2
+    addu        t0, t4, t0
+    shra_r.w    t0, t0, 13
+    lh          t1, 90(a1)
+    lh          t2, 90(a0)
+    lh          t3, 122(a1)
+    lh          t4, 122(a0)
+    shra_r.w    t8, t8, 13
+    mul         t1, t1, t2
+    mul         t3, t3, t4
+    sw          t0, 8(v0)
+    sw          t8, 28(v0)
+    sll         t4, t5, 15
+    ins         t7, t6, 16, 16
+    mult        zero, zero
+    dpa.w.ph    $ac0, t7, s0
+    ins         t3, t1, 16, 16
+    lh          t1, 14(a1)
+    lh          t6, 14(a0)
+    dpa.w.ph    $ac0, t3, s1
+    mflo        t0, $ac0
+    mul         t5, t6, t1
+    lh          t1, 30(a1)
+    lh          t6, 30(a0)
+    lh          t2, 62(a1)
+    lh          t7, 62(a0)
+    mul         t6, t6, t1
+    subu        t8, t4, t0
+    mul         t7, t7, t2
+    addu        t0, t4, t0
+    shra_r.w    t0, t0, 13
+    lh          t1, 94(a1)
+    lh          t2, 94(a0)
+    lh          t3, 126(a1)
+    lh          t4, 126(a0)
+    shra_r.w    t8, t8, 13
+    mul         t1, t1, t2
+    mul         t3, t3, t4
+    sw          t0, 12(v0)
+    sw          t8, 32(v0)
+    sll         t4, t5, 15
+    ins         t7, t6, 16, 16
+    mult        zero, zero
+    dpa.w.ph    $ac0, t7, s0
+    ins         t3, t1, 16, 16
+    dpa.w.ph    $ac0, t3, s1
+    mflo        t0, $ac0
+    lw          t9, 0(a2)
+    lw          t3, 0(v0)
+    lw          t7, 4(v0)
+    lw          t1, 8(v0)
+    addu        t9, t9, a3
+    sll         t3, t3, 15
+    subu        t8, t4, t0
+    addu        t0, t4, t0
+    shra_r.w    t0, t0, 13
+    shra_r.w    t8, t8, 13
+    sw          t0, 16(v0)
+    sw          t8, 36(v0)
+    lw          t5, 12(v0)
+    lw          t6, 16(v0)
+    mult        t7, s2
+    madd        t1, s3
+    madd        t5, s4
+    madd        t6, s5
+    lw          t5, 24(v0)
+    lw          t7, 28(v0)
+    mflo        t0, $ac0
+    lw          t8, 32(v0)
+    lw          t2, 36(v0)
+    mult        $ac1, t5, s2
+    madd        $ac1, t7, s3
+    madd        $ac1, t8, s4
+    madd        $ac1, t2, s5
+    addu        t1, t3, t0
+    subu        t6, t3, t0
+    shra_r.w    t1, t1, 20
+    shra_r.w    t6, t6, 20
+    mflo        t4, $ac1
+    shll_s.w    t1, t1, 24
+    shll_s.w    t6, t6, 24
+    sra         t1, t1, 24
+    sra         t6, t6, 24
+    addiu       t1, t1, 128
+    addiu       t6, t6, 128
+    lw          t0, 20(v0)
+    sb          t1, 0(t9)
+    sb          t6, 1(t9)
+    sll         t0, t0, 15
+    lw          t9, 4(a2)
+    addu        t1, t0, t4
+    subu        t6, t0, t4
+    addu        t9, t9, a3
+    shra_r.w    t1, t1, 20
+    shra_r.w    t6, t6, 20
+    shll_s.w    t1, t1, 24
+    shll_s.w    t6, t6, 24
+    sra         t1, t1, 24
+    sra         t6, t6, 24
+    addiu       t1, t1, 128
+    addiu       t6, t6, 128
+    sb          t1, 0(t9)
+    sb          t6, 1(t9)
+    addiu       sp, sp, 40
+
+    RESTORE_REGS_FROM_STACK 24, s0, s1, s2, s3, s4, s5
+
+    j           ra
+     nop
+
+END(jsimd_idct_2x2_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_4x4_dspr2)
+/*
+ * a0     = compptr->dct_table
+ * a1     = coef_block
+ * a2     = output_buf
+ * a3     = output_col
+ * 16(sp) = workspace[DCTSIZE*4];  // buffers data between passes
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    lw          v1, 48(sp)
+    move        t0, a1
+    move        t1, v1
+    li          t9, 4
+    li          s0, 0x2e75f93e
+    li          s1, 0x21f9ba79
+    li          s2, 0xecc2efb0
+    li          s3, 0x52031ccd
+
+0:
+    lh          s6, 32(t0)      // inptr[DCTSIZE*2]
+    lh          t6, 32(a0)      // quantptr[DCTSIZE*2]
+    lh          s7, 96(t0)      // inptr[DCTSIZE*6]
+    lh          t7, 96(a0)      // quantptr[DCTSIZE*6]
+    mul         t6, s6, t6      // z2 = (inptr[DCTSIZE*2] * quantptr[DCTSIZE*2])
+    lh          s4, 0(t0)       // inptr[DCTSIZE*0]
+    mul         t7, s7, t7      // z3 = (inptr[DCTSIZE*6] * quantptr[DCTSIZE*6])
+    lh          s5, 0(a0)       // quantptr[0]
+    li          s6, 15137
+    li          s7, 6270
+    mul         t2, s4, s5      // tmp0 = (inptr[0] * quantptr[0])
+    mul         t6, s6, t6      // z2 = (inptr[DCTSIZE*2] * quantptr[DCTSIZE*2])
+    lh          t5, 112(t0)     // inptr[DCTSIZE*7]
+    mul         t7, s7, t7      // z3 = (inptr[DCTSIZE*6] * quantptr[DCTSIZE*6])
+    lh          s4, 112(a0)     // quantptr[DCTSIZE*7]
+    lh          v0, 80(t0)      // inptr[DCTSIZE*5]
+    lh          s5, 80(a0)      // quantptr[DCTSIZE*5]
+    lh          s6, 48(a0)      // quantptr[DCTSIZE*3]
+    sll         t2, t2, 14      // tmp0 <<= (CONST_BITS+1)
+    lh          s7, 16(a0)      // quantptr[DCTSIZE*1]
+    lh          t8, 16(t0)      // inptr[DCTSIZE*1]
+    subu        t6, t6, t7      // tmp2 = MULTIPLY(z2, t5) - MULTIPLY(z3, t6)
+    lh          t7, 48(t0)      // inptr[DCTSIZE*3]
+    mul         t5, s4, t5      // z1 = (inptr[DCTSIZE*7] * quantptr[DCTSIZE*7])
+    mul         v0, s5, v0      // z2 = (inptr[DCTSIZE*5] * quantptr[DCTSIZE*5])
+    mul         t7, s6, t7      // z3 = (inptr[DCTSIZE*3] * quantptr[DCTSIZE*3])
+    mul         t8, s7, t8      // z4 = (inptr[DCTSIZE*1] * quantptr[DCTSIZE*1])
+    addu        t3, t2, t6      // tmp10 = tmp0 + z2
+    subu        t4, t2, t6      // tmp10 = tmp0 - z2
+    mult        $ac0, zero, zero
+    mult        $ac1, zero, zero
+    ins         t5, v0, 16, 16
+    ins         t7, t8, 16, 16
+    addiu       t9, t9, -1
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    mflo        s4, $ac0
+    mflo        s5, $ac1
+    addiu       a0, a0, 2
+    addiu       t1, t1, 4
+    addiu       t0, t0, 2
+    addu        t6, t4, s4
+    subu        t5, t4, s4
+    addu        s6, t3, s5
+    subu        s7, t3, s5
+    shra_r.w    t6, t6, 12      // DESCALE(tmp12 + temp1, 12)
+    shra_r.w    t5, t5, 12      // DESCALE(tmp12 - temp1, 12)
+    shra_r.w    s6, s6, 12      // DESCALE(tmp10 + temp2, 12)
+    shra_r.w    s7, s7, 12      // DESCALE(tmp10 - temp2, 12)
+    sw          t6, 28(t1)
+    sw          t5, 60(t1)
+    sw          s6, -4(t1)
+    bgtz        t9, 0b
+     sw         s7, 92(t1)
+    // second loop three pass
+    li          t9, 3
+1:
+    lh          s6, 34(t0)      // inptr[DCTSIZE*2]
+    lh          t6, 34(a0)      // quantptr[DCTSIZE*2]
+    lh          s7, 98(t0)      // inptr[DCTSIZE*6]
+    lh          t7, 98(a0)      // quantptr[DCTSIZE*6]
+    mul         t6, s6, t6      // z2 = (inptr[DCTSIZE*2] * quantptr[DCTSIZE*2])
+    lh          s4, 2(t0)       // inptr[DCTSIZE*0]
+    mul         t7, s7, t7      // z3 = (inptr[DCTSIZE*6] * quantptr[DCTSIZE*6])
+    lh          s5, 2(a0)       // quantptr[DCTSIZE*0]
+    li          s6, 15137
+    li          s7, 6270
+    mul         t2, s4, s5      // tmp0 = (inptr[0] * quantptr[0])
+    mul         v0, s6, t6      // z2 = (inptr[DCTSIZE*2] * quantptr[DCTSIZE*2])
+    lh          t5, 114(t0)     // inptr[DCTSIZE*7]
+    mul         t7, s7, t7      // z3 = (inptr[DCTSIZE*6] * quantptr[DCTSIZE*6])
+    lh          s4, 114(a0)     // quantptr[DCTSIZE*7]
+    lh          s5, 82(a0)      // quantptr[DCTSIZE*5]
+    lh          t6, 82(t0)      // inptr[DCTSIZE*5]
+    sll         t2, t2, 14      // tmp0 <<= (CONST_BITS+1)
+    lh          s6, 50(a0)      // quantptr[DCTSIZE*3]
+    lh          t8, 18(t0)      // inptr[DCTSIZE*1]
+    subu        v0, v0, t7      // tmp2 = MULTIPLY(z2, t5) - MULTIPLY(z3, t6)
+    lh          t7, 50(t0)      // inptr[DCTSIZE*3]
+    lh          s7, 18(a0)      // quantptr[DCTSIZE*1]
+    mul         t5, s4, t5      // z1 = (inptr[DCTSIZE*7] * quantptr[DCTSIZE*7])
+    mul         t6, s5, t6      // z2 = (inptr[DCTSIZE*5] * quantptr[DCTSIZE*5])
+    mul         t7, s6, t7      // z3 = (inptr[DCTSIZE*3] * quantptr[DCTSIZE*3])
+    mul         t8, s7, t8      // z4 = (inptr[DCTSIZE*1] * quantptr[DCTSIZE*1])
+    addu        t3, t2, v0      // tmp10 = tmp0 + z2
+    subu        t4, t2, v0      // tmp10 = tmp0 - z2
+    mult        $ac0, zero, zero
+    mult        $ac1, zero, zero
+    ins         t5, t6, 16, 16
+    ins         t7, t8, 16, 16
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    mflo        t5, $ac0
+    mflo        t6, $ac1
+    addiu       t9, t9, -1
+    addiu       t0, t0, 2
+    addiu       a0, a0, 2
+    addiu       t1, t1, 4
+    addu        s5, t4, t5
+    subu        s4, t4, t5
+    addu        s6, t3, t6
+    subu        s7, t3, t6
+    shra_r.w    s5, s5, 12      // DESCALE(tmp12 + temp1, 12)
+    shra_r.w    s4, s4, 12      // DESCALE(tmp12 - temp1, 12)
+    shra_r.w    s6, s6, 12      // DESCALE(tmp10 + temp2, 12)
+    shra_r.w    s7, s7, 12      // DESCALE(tmp10 - temp2, 12)
+    sw          s5, 32(t1)
+    sw          s4, 64(t1)
+    sw          s6, 0(t1)
+    bgtz        t9, 1b
+     sw         s7, 96(t1)
+    move        t1, v1
+    li          s4, 15137
+    lw          s6, 8(t1)       // wsptr[2]
+    li          s5, 6270
+    lw          s7, 24(t1)      // wsptr[6]
+    mul         s4, s4, s6      // MULTIPLY((JLONG)wsptr[2], FIX_1_847759065)
+    lw          t2, 0(t1)       // wsptr[0]
+    mul         s5, s5, s7      // MULTIPLY((JLONG)wsptr[6], -FIX_0_765366865)
+    lh          t5, 28(t1)      // wsptr[7]
+    lh          t6, 20(t1)      // wsptr[5]
+    lh          t7, 12(t1)      // wsptr[3]
+    lh          t8, 4(t1)       // wsptr[1]
+    ins         t5, t6, 16, 16
+    ins         t7, t8, 16, 16
+    mult        $ac0, zero, zero
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    mult        $ac1, zero, zero
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    sll         t2, t2, 14      // tmp0 = ((JLONG)wsptr[0]) << (CONST_BITS+1)
+    mflo        s6, $ac0
+    // MULTIPLY(wsptr[2], FIX_1_847759065 + MULTIPLY(wsptr[6], -FIX_0_765366865)
+    subu        s4, s4, s5
+    addu        t3, t2, s4      // tmp10 = tmp0 + z2
+    mflo        s7, $ac1
+    subu        t4, t2, s4      // tmp10 = tmp0 - z2
+    addu        t7, t4, s6
+    subu        t8, t4, s6
+    addu        t5, t3, s7
+    subu        t6, t3, s7
+    shra_r.w    t5, t5, 19      // DESCALE(tmp10 + temp2, 19)
+    shra_r.w    t6, t6, 19      // DESCALE(tmp10 - temp2, 19)
+    shra_r.w    t7, t7, 19      // DESCALE(tmp12 + temp1, 19)
+    shra_r.w    t8, t8, 19      // DESCALE(tmp12 - temp1, 19)
+    sll         s4, t9, 2
+    lw          v0, 0(a2)       // output_buf[ctr]
+    shll_s.w    t5, t5, 24
+    shll_s.w    t6, t6, 24
+    shll_s.w    t7, t7, 24
+    shll_s.w    t8, t8, 24
+    sra         t5, t5, 24
+    sra         t6, t6, 24
+    sra         t7, t7, 24
+    sra         t8, t8, 24
+    addu        v0, v0, a3      // outptr = output_buf[ctr] + output_col
+    addiu       t5, t5, 128
+    addiu       t6, t6, 128
+    addiu       t7, t7, 128
+    addiu       t8, t8, 128
+    sb          t5, 0(v0)
+    sb          t7, 1(v0)
+    sb          t8, 2(v0)
+    sb          t6, 3(v0)
+    // 2
+    li          s4, 15137
+    lw          s6, 40(t1)      // wsptr[2]
+    li          s5, 6270
+    lw          s7, 56(t1)      // wsptr[6]
+    mul         s4, s4, s6      // MULTIPLY((JLONG)wsptr[2], FIX_1_847759065)
+    lw          t2, 32(t1)      // wsptr[0]
+    mul         s5, s5, s7      // MULTIPLY((JLONG)wsptr[6], -FIX_0_765366865)
+    lh          t5, 60(t1)      // wsptr[7]
+    lh          t6, 52(t1)      // wsptr[5]
+    lh          t7, 44(t1)      // wsptr[3]
+    lh          t8, 36(t1)      // wsptr[1]
+    ins         t5, t6, 16, 16
+    ins         t7, t8, 16, 16
+    mult        $ac0, zero, zero
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    mult        $ac1, zero, zero
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    sll         t2, t2, 14      // tmp0 = ((JLONG)wsptr[0]) << (CONST_BITS+1)
+    mflo        s6, $ac0
+    // MULTIPLY(wsptr[2], FIX_1_847759065 + MULTIPLY(wsptr[6], -FIX_0_765366865)
+    subu        s4, s4, s5
+    addu        t3, t2, s4      // tmp10 = tmp0 + z2
+    mflo        s7, $ac1
+    subu        t4, t2, s4      // tmp10 = tmp0 - z2
+    addu        t7, t4, s6
+    subu        t8, t4, s6
+    addu        t5, t3, s7
+    subu        t6, t3, s7
+    shra_r.w    t5, t5, 19      // DESCALE(tmp10 + temp2, CONST_BITS-PASS1_BITS+1)
+    shra_r.w    t6, t6, 19      // DESCALE(tmp10 - temp2, CONST_BITS-PASS1_BITS+1)
+    shra_r.w    t7, t7, 19      // DESCALE(tmp12 + temp1, CONST_BITS-PASS1_BITS+1)
+    shra_r.w    t8, t8, 19      // DESCALE(tmp12 - temp1, CONST_BITS-PASS1_BITS+1)
+    sll         s4, t9, 2
+    lw          v0, 4(a2)       // output_buf[ctr]
+    shll_s.w    t5, t5, 24
+    shll_s.w    t6, t6, 24
+    shll_s.w    t7, t7, 24
+    shll_s.w    t8, t8, 24
+    sra         t5, t5, 24
+    sra         t6, t6, 24
+    sra         t7, t7, 24
+    sra         t8, t8, 24
+    addu        v0, v0, a3      // outptr = output_buf[ctr] + output_col
+    addiu       t5, t5, 128
+    addiu       t6, t6, 128
+    addiu       t7, t7, 128
+    addiu       t8, t8, 128
+    sb          t5, 0(v0)
+    sb          t7, 1(v0)
+    sb          t8, 2(v0)
+    sb          t6, 3(v0)
+    // 3
+    li          s4, 15137
+    lw          s6, 72(t1)      // wsptr[2]
+    li          s5, 6270
+    lw          s7, 88(t1)      // wsptr[6]
+    mul         s4, s4, s6      // MULTIPLY((JLONG)wsptr[2], FIX_1_847759065)
+    lw          t2, 64(t1)      // wsptr[0]
+    mul         s5, s5, s7      // MULTIPLY((JLONG)wsptr[6], -FIX_0_765366865)
+    lh          t5, 92(t1)      // wsptr[7]
+    lh          t6, 84(t1)      // wsptr[5]
+    lh          t7, 76(t1)      // wsptr[3]
+    lh          t8, 68(t1)      // wsptr[1]
+    ins         t5, t6, 16, 16
+    ins         t7, t8, 16, 16
+    mult        $ac0, zero, zero
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    mult        $ac1, zero, zero
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    sll         t2, t2, 14      // tmp0 = ((JLONG)wsptr[0]) << (CONST_BITS+1)
+    mflo        s6, $ac0
+    // MULTIPLY(wsptr[2], FIX_1_847759065 + MULTIPLY(wsptr[6], -FIX_0_765366865)
+    subu        s4, s4, s5
+    addu        t3, t2, s4      // tmp10 = tmp0 + z2
+    mflo        s7, $ac1
+    subu        t4, t2, s4      // tmp10 = tmp0 - z2
+    addu        t7, t4, s6
+    subu        t8, t4, s6
+    addu        t5, t3, s7
+    subu        t6, t3, s7
+    shra_r.w    t5, t5, 19      // DESCALE(tmp10 + temp2, 19)
+    shra_r.w    t6, t6, 19      // DESCALE(tmp10 - temp2, 19)
+    shra_r.w    t7, t7, 19      // DESCALE(tmp12 + temp1, 19)
+    shra_r.w    t8, t8, 19      // DESCALE(tmp12 - temp1, 19)
+    sll         s4, t9, 2
+    lw          v0, 8(a2)       // output_buf[ctr]
+    shll_s.w    t5, t5, 24
+    shll_s.w    t6, t6, 24
+    shll_s.w    t7, t7, 24
+    shll_s.w    t8, t8, 24
+    sra         t5, t5, 24
+    sra         t6, t6, 24
+    sra         t7, t7, 24
+    sra         t8, t8, 24
+    addu        v0, v0, a3      // outptr = output_buf[ctr] + output_col
+    addiu       t5, t5, 128
+    addiu       t6, t6, 128
+    addiu       t7, t7, 128
+    addiu       t8, t8, 128
+    sb          t5, 0(v0)
+    sb          t7, 1(v0)
+    sb          t8, 2(v0)
+    sb          t6, 3(v0)
+    li          s4, 15137
+    lw          s6, 104(t1)     // wsptr[2]
+    li          s5, 6270
+    lw          s7, 120(t1)     // wsptr[6]
+    mul         s4, s4, s6      // MULTIPLY((JLONG)wsptr[2], FIX_1_847759065)
+    lw          t2, 96(t1)      // wsptr[0]
+    mul         s5, s5, s7      // MULTIPLY((JLONG)wsptr[6], -FIX_0_765366865)
+    lh          t5, 124(t1)     // wsptr[7]
+    lh          t6, 116(t1)     // wsptr[5]
+    lh          t7, 108(t1)     // wsptr[3]
+    lh          t8, 100(t1)     // wsptr[1]
+    ins         t5, t6, 16, 16
+    ins         t7, t8, 16, 16
+    mult        $ac0, zero, zero
+    dpa.w.ph    $ac0, t5, s0
+    dpa.w.ph    $ac0, t7, s1
+    mult        $ac1, zero, zero
+    dpa.w.ph    $ac1, t5, s2
+    dpa.w.ph    $ac1, t7, s3
+    sll         t2, t2, 14      // tmp0 = ((JLONG)wsptr[0]) << (CONST_BITS+1)
+    mflo        s6, $ac0
+    // MULTIPLY(wsptr[2], FIX_1_847759065 + MULTIPLY(wsptr[6], -FIX_0_765366865)
+    subu        s4, s4, s5
+    addu        t3, t2, s4      // tmp10 = tmp0 + z2;
+    mflo        s7, $ac1
+    subu        t4, t2, s4      // tmp10 = tmp0 - z2;
+    addu        t7, t4, s6
+    subu        t8, t4, s6
+    addu        t5, t3, s7
+    subu        t6, t3, s7
+    shra_r.w    t5, t5, 19      // DESCALE(tmp10 + temp2, 19)
+    shra_r.w    t6, t6, 19      // DESCALE(tmp10 - temp2, 19)
+    shra_r.w    t7, t7, 19      // DESCALE(tmp12 + temp1, 19)
+    shra_r.w    t8, t8, 19      // DESCALE(tmp12 - temp1, 19)
+    sll         s4, t9, 2
+    lw          v0, 12(a2)      // output_buf[ctr]
+    shll_s.w    t5, t5, 24
+    shll_s.w    t6, t6, 24
+    shll_s.w    t7, t7, 24
+    shll_s.w    t8, t8, 24
+    sra         t5, t5, 24
+    sra         t6, t6, 24
+    sra         t7, t7, 24
+    sra         t8, t8, 24
+    addu        v0, v0, a3      // outptr = output_buf[ctr] + output_col
+    addiu       t5, t5, 128
+    addiu       t6, t6, 128
+    addiu       t7, t7, 128
+    addiu       t8, t8, 128
+    sb          t5, 0(v0)
+    sb          t7, 1(v0)
+    sb          t8, 2(v0)
+    sb          t6, 3(v0)
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+END(jsimd_idct_4x4_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_6x6_dspr2)
+/*
+ * a0 = compptr->dct_table
+ * a1 = coef_block
+ * a2 = output_buf
+ * a3 = output_col
+ */
+    .set at
+
+    SAVE_REGS_ON_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    addiu       sp, sp, -144
+    move        v0, sp
+    addiu       v1, v0, 24
+    addiu       t9, zero, 5793
+    addiu       s0, zero, 10033
+    addiu       s1, zero, 2998
+
+1:
+    lh          s2, 0(a0)       // q0 = quantptr[ 0]
+    lh          s3, 32(a0)      // q1 = quantptr[16]
+    lh          s4, 64(a0)      // q2 = quantptr[32]
+    lh          t2, 64(a1)      // tmp2 = inptr[32]
+    lh          t1, 32(a1)      // tmp1 = inptr[16]
+    lh          t0, 0(a1)       // tmp0 = inptr[ 0]
+    mul         t2, t2, s4      // tmp2 = tmp2 * q2
+    mul         t1, t1, s3      // tmp1 = tmp1 * q1
+    mul         t0, t0, s2      // tmp0 = tmp0 * q0
+    lh          t6, 16(a1)      // z1 = inptr[ 8]
+    lh          t8, 80(a1)      // z3 = inptr[40]
+    lh          t7, 48(a1)      // z2 = inptr[24]
+    lh          s2, 16(a0)      // q0 = quantptr[ 8]
+    lh          s4, 80(a0)      // q2 = quantptr[40]
+    lh          s3, 48(a0)      // q1 = quantptr[24]
+    mul         t2, t2, t9      // tmp2 = tmp2 * 5793
+    mul         t1, t1, s0      // tmp1 = tmp1 * 10033
+    sll         t0, t0, 13      // tmp0 = tmp0 << 13
+    mul         t6, t6, s2      // z1 = z1 * q0
+    mul         t8, t8, s4      // z3 = z3 * q2
+    mul         t7, t7, s3      // z2 = z2 * q1
+    addu        t3, t0, t2      // tmp10 = tmp0 + tmp2
+    sll         t2, t2, 1       // tmp2 = tmp2 << 2
+    subu        t4, t0, t2      // tmp11 = tmp0 - tmp2;
+    subu        t5, t3, t1      // tmp12 = tmp10 - tmp1
+    addu        t3, t3, t1      // tmp10 = tmp10 + tmp1
+    addu        t1, t6, t8      // tmp1 = z1 + z3
+    mul         t1, t1, s1      // tmp1 = tmp1 * 2998
+    shra_r.w    t4, t4, 11      // tmp11 = (tmp11 + 1024) >> 11
+    subu        t2, t6, t8      // tmp2 = z1 - z3
+    subu        t2, t2, t7      // tmp2 = tmp2 - z2
+    sll         t2, t2, 2       // tmp2 = tmp2 << 2
+    addu        t0, t6, t7      // tmp0 = z1 + z2
+    sll         t0, t0, 13      // tmp0 = tmp0 << 13
+    subu        s2, t8, t7      // q0 = z3 - z2
+    sll         s2, s2, 13      // q0 = q0 << 13
+    addu        t0, t0, t1      // tmp0 = tmp0 + tmp1
+    addu        t1, s2, t1      // tmp1 = q0 + tmp1
+    addu        s2, t4, t2      // q0 = tmp11 + tmp2
+    subu        s3, t4, t2      // q1 = tmp11 - tmp2
+    addu        t6, t3, t0      // z1 = tmp10 + tmp0
+    subu        t7, t3, t0      // z2 = tmp10 - tmp0
+    addu        t4, t5, t1      // tmp11 = tmp12 + tmp1
+    subu        t5, t5, t1      // tmp12 = tmp12 - tmp1
+    shra_r.w    t6, t6, 11      // z1 = (z1 + 1024) >> 11
+    shra_r.w    t7, t7, 11      // z2 = (z2 + 1024) >> 11
+    shra_r.w    t4, t4, 11      // tmp11 = (tmp11 + 1024) >> 11
+    shra_r.w    t5, t5, 11      // tmp12 = (tmp12 + 1024) >> 11
+    sw          s2, 24(v0)
+    sw          s3, 96(v0)
+    sw          t6, 0(v0)
+    sw          t7, 120(v0)
+    sw          t4, 48(v0)
+    sw          t5, 72(v0)
+    addiu       v0, v0, 4
+    addiu       a1, a1, 2
+    bne         v0, v1, 1b
+     addiu      a0, a0, 2
+
+    /* Pass 2: process 6 rows from work array, store into output array. */
+    move        v0, sp
+    addiu       v1, v0, 144
+
+2:
+    lw          t0, 0(v0)
+    lw          t2, 16(v0)
+    lw          s5, 0(a2)
+    addiu       t0, t0, 16
+    sll         t0, t0, 13
+    mul         t3, t2, t9
+    lw          t6, 4(v0)
+    lw          t8, 20(v0)
+    lw          t7, 12(v0)
+    addu        s5, s5, a3
+    addu        s6, t6, t8
+    mul         s6, s6, s1
+    addu        t1, t0, t3
+    subu        t4, t0, t3
+    subu        t4, t4, t3
+    lw          t3, 8(v0)
+    mul         t0, t3, s0
+    addu        s7, t6, t7
+    sll         s7, s7, 13
+    addu        s7, s6, s7
+    subu        t2, t8, t7
+    sll         t2, t2, 13
+    addu        t2, s6, t2
+    subu        s6, t6, t7
+    subu        s6, s6, t8
+    sll         s6, s6, 13
+    addu        t3, t1, t0
+    subu        t5, t1, t0
+    addu        t6, t3, s7
+    subu        t3, t3, s7
+    addu        t7, t4, s6
+    subu        t4, t4, s6
+    addu        t8, t5, t2
+    subu        t5, t5, t2
+    shll_s.w    t6, t6, 6
+    shll_s.w    t3, t3, 6
+    shll_s.w    t7, t7, 6
+    shll_s.w    t4, t4, 6
+    shll_s.w    t8, t8, 6
+    shll_s.w    t5, t5, 6
+    sra         t6, t6, 24
+    addiu       t6, t6, 128
+    sra         t3, t3, 24
+    addiu       t3, t3, 128
+    sb          t6, 0(s5)
+    sra         t7, t7, 24
+    addiu       t7, t7, 128
+    sb          t3, 5(s5)
+    sra         t4, t4, 24
+    addiu       t4, t4, 128
+    sb          t7, 1(s5)
+    sra         t8, t8, 24
+    addiu       t8, t8, 128
+    sb          t4, 4(s5)
+    addiu       v0, v0, 24
+    sra         t5, t5, 24
+    addiu       t5, t5, 128
+    sb          t8, 2(s5)
+    addiu       a2, a2,  4
+    bne         v0, v1, 2b
+     sb         t5, 3(s5)
+
+    addiu       sp, sp, 144
+
+    RESTORE_REGS_FROM_STACK 32, s0, s1, s2, s3, s4, s5, s6, s7
+
+    j           ra
+     nop
+
+END(jsimd_idct_6x6_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_12x12_pass1_dspr2)
+/*
+ * a0 = compptr->dct_table
+ * a1 = coef_block
+ * a2 = workspace
+ */
+    SAVE_REGS_ON_STACK 16, s0, s1, s2, s3
+
+    li          a3, 8
+
+1:
+    // odd part
+    lh          t0, 48(a1)
+    lh          t1, 48(a0)
+    lh          t2, 16(a1)
+    lh          t3, 16(a0)
+    lh          t4, 80(a1)
+    lh          t5, 80(a0)
+    lh          t6, 112(a1)
+    lh          t7, 112(a0)
+    mul         t0, t0, t1      // z2
+    mul         t1, t2, t3      // z1
+    mul         t2, t4, t5      // z3
+    mul         t3, t6, t7      // z4
+    li          t4, 10703       // FIX(1.306562965)
+    li          t5, 4433        // FIX_0_541196100
+    li          t6, 7053        // FIX(0.860918669)
+    mul         t4, t0, t4      // tmp11
+    mul         t5, t0, t5      // -tmp14
+    addu        t7, t1, t2      // tmp10
+    addu        t8, t7, t3      // tmp10 + z4
+    mul         t6, t6, t8      // tmp15
+    li          t8, 2139        // FIX(0.261052384)
+    mul         t8, t7, t8      // MULTIPLY(tmp10, FIX(0.261052384))
+    li          t7, 2295        // FIX(0.280143716)
+    mul         t7, t1, t7      // MULTIPLY(z1, FIX(0.280143716))
+    addu        t9, t2, t3      // z3 + z4
+    li          s0, 8565        // FIX(1.045510580)
+    mul         t9, t9, s0      // -tmp13
+    li          s0, 12112       // FIX(1.478575242)
+    mul         s0, t2, s0      // MULTIPLY(z3, FIX(1.478575242)
+    li          s1, 12998       // FIX(1.586706681)
+    mul         s1, t3, s1      // MULTIPLY(z4, FIX(1.586706681))
+    li          s2, 5540        // FIX(0.676326758)
+    mul         s2, t1, s2      // MULTIPLY(z1, FIX(0.676326758))
+    li          s3, 16244       // FIX(1.982889723)
+    mul         s3, t3, s3      // MULTIPLY(z4, FIX(1.982889723))
+    subu        t1, t1, t3      // z1-=z4
+    subu        t0, t0, t2      // z2-=z3
+    addu        t2, t0, t1      // z1+z2
+    li          t3, 4433        // FIX_0_541196100
+    mul         t2, t2, t3      // z3
+    li          t3, 6270        // FIX_0_765366865
+    mul         t1, t1, t3      // MULTIPLY(z1, FIX_0_765366865)
+    li          t3, 15137       // FIX_0_765366865
+    mul         t0, t0, t3      // MULTIPLY(z2, FIX_1_847759065)
+    addu        t8, t6, t8      // tmp12
+    addu        t3, t8, t4      // tmp12 + tmp11
+    addu        t3, t3, t7      // tmp10
+    subu        t8, t8, t9      // tmp12 + tmp13
+    addu        s0, t5, s0
+    subu        t8, t8, s0      // tmp12
+    subu        t9, t6, t9
+    subu        s1, s1, t4
+    addu        t9, t9, s1      // tmp13
+    subu        t6, t6, t5
+    subu        t6, t6, s2
+    subu        t6, t6, s3      // tmp15
+    // even part start
+    lh          t4, 64(a1)
+    lh          t5, 64(a0)
+    lh          t7, 32(a1)
+    lh          s0, 32(a0)
+    lh          s1, 0(a1)
+    lh          s2, 0(a0)
+    lh          s3, 96(a1)
+    lh          v0, 96(a0)
+    mul         t4, t4, t5      // DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4])
+    mul         t5, t7, s0      // DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2])
+    mul         t7, s1, s2      // DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0])
+    mul         s0, s3, v0      // DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6])
+    // odd part end
+    addu        t1, t2, t1      // tmp11
+    subu        t0, t2, t0      // tmp14
+    // update counter and pointers
+    addiu       a3, a3, -1
+    addiu       a0, a0, 2
+    addiu       a1, a1, 2
+    // even part rest
+    li          s1, 10033
+    li          s2, 11190
+    mul         t4, t4, s1      // z4
+    mul         s1, t5, s2      // z4
+    sll         t5, t5, 13      // z1
+    sll         t7, t7, 13
+    addiu       t7, t7, 1024    // z3
+    sll         s0, s0, 13      // z2
+    addu        s2, t7, t4      // tmp10
+    subu        t4, t7, t4      // tmp11
+    subu        s3, t5, s0      // tmp12
+    addu        t2, t7, s3      // tmp21
+    subu        s3, t7, s3      // tmp24
+    addu        t7, s1, s0      // tmp12
+    addu        v0, s2, t7      // tmp20
+    subu        s2, s2, t7      // tmp25
+    subu        s1, s1, t5      // z4 - z1
+    subu        s1, s1, s0      // tmp12
+    addu        s0, t4, s1      // tmp22
+    subu        t4, t4, s1      // tmp23
+    // final output stage
+    addu        t5, v0, t3
+    subu        v0, v0, t3
+    addu        t3, t2, t1
+    subu        t2, t2, t1
+    addu        t1, s0, t8
+    subu        s0, s0, t8
+    addu        t8, t4, t9
+    subu        t4, t4, t9
+    addu        t9, s3, t0
+    subu        s3, s3, t0
+    addu        t0, s2, t6
+    subu        s2, s2, t6
+    sra         t5, t5, 11
+    sra         t3, t3, 11
+    sra         t1, t1, 11
+    sra         t8, t8, 11
+    sra         t9, t9, 11
+    sra         t0, t0, 11
+    sra         s2, s2, 11
+    sra         s3, s3, 11
+    sra         t4, t4, 11
+    sra         s0, s0, 11
+    sra         t2, t2, 11
+    sra         v0, v0, 11
+    sw          t5, 0(a2)
+    sw          t3, 32(a2)
+    sw          t1, 64(a2)
+    sw          t8, 96(a2)
+    sw          t9, 128(a2)
+    sw          t0, 160(a2)
+    sw          s2, 192(a2)
+    sw          s3, 224(a2)
+    sw          t4, 256(a2)
+    sw          s0, 288(a2)
+    sw          t2, 320(a2)
+    sw          v0, 352(a2)
+    bgtz        a3, 1b
+     addiu      a2, a2, 4
+
+    RESTORE_REGS_FROM_STACK 16, s0, s1, s2, s3
+
+    j           ra
+     nop
+
+END(jsimd_idct_12x12_pass1_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_idct_12x12_pass2_dspr2)
+/*
+ * a0 = workspace
+ * a1 = output
+ */
+    SAVE_REGS_ON_STACK 16, s0, s1, s2, s3
+
+    li          a3, 12
+
+1:
+    // Odd part
+    lw          t0, 12(a0)
+    lw          t1, 4(a0)
+    lw          t2, 20(a0)
+    lw          t3, 28(a0)
+    li          t4, 10703       // FIX(1.306562965)
+    li          t5, 4433        // FIX_0_541196100
+    mul         t4, t0, t4      // tmp11
+    mul         t5, t0, t5      // -tmp14
+    addu        t6, t1, t2      // tmp10
+    li          t7, 2139        // FIX(0.261052384)
+    mul         t7, t6, t7      // MULTIPLY(tmp10, FIX(0.261052384))
+    addu        t6, t6, t3      // tmp10 + z4
+    li          t8, 7053        // FIX(0.860918669)
+    mul         t6, t6, t8      // tmp15
+    li          t8, 2295        // FIX(0.280143716)
+    mul         t8, t1, t8      // MULTIPLY(z1, FIX(0.280143716))
+    addu        t9, t2, t3      // z3 + z4
+    li          s0, 8565        // FIX(1.045510580)
+    mul         t9, t9, s0      // -tmp13
+    li          s0, 12112       // FIX(1.478575242)
+    mul         s0, t2, s0      // MULTIPLY(z3, FIX(1.478575242))
+    li          s1, 12998       // FIX(1.586706681)
+    mul         s1, t3, s1      // MULTIPLY(z4, FIX(1.586706681))
+    li          s2, 5540        // FIX(0.676326758)
+    mul         s2, t1, s2      // MULTIPLY(z1, FIX(0.676326758))
+    li          s3, 16244       // FIX(1.982889723)
+    mul         s3, t3, s3      // MULTIPLY(z4, FIX(1.982889723))
+    subu        t1, t1, t3      // z1 -= z4
+    subu        t0, t0, t2      // z2 -= z3
+    addu        t2, t1, t0      // z1 + z2
+    li          t3, 4433        // FIX_0_541196100
+    mul         t2, t2, t3      // z3
+    li          t3, 6270        // FIX_0_765366865
+    mul         t1, t1, t3      // MULTIPLY(z1, FIX_0_765366865)
+    li          t3, 15137       // FIX_1_847759065
+    mul         t0, t0, t3      // MULTIPLY(z2, FIX_1_847759065)
+    addu        t3, t6, t7      // tmp12
+    addu        t7, t3, t4
+    addu        t7, t7, t8      // tmp10
+    subu        t3, t3, t9
+    subu        t3, t3, t5
+    subu        t3, t3, s0      // tmp12
+    subu        t9, t6, t9
+    subu        t9, t9, t4
+    addu        t9, t9, s1      // tmp13
+    subu        t6, t6, t5
+    subu        t6, t6, s2
+    subu        t6, t6, s3      // tmp15
+    addu        t1, t2, t1      // tmp11
+    subu        t0, t2, t0      // tmp14
+    // even part
+    lw          t2, 16(a0)      // z4
+    lw          t4, 8(a0)       // z1
+    lw          t5, 0(a0)       // z3
+    lw          t8, 24(a0)      // z2
+    li          s0, 10033       // FIX(1.224744871)
+    li          s1, 11190       // FIX(1.366025404)
+    mul         t2, t2, s0      // z4
+    mul         s0, t4, s1      // z4
+    addiu       t5, t5, 0x10
+    sll         t5, t5, 13      // z3
+    sll         t4, t4, 13      // z1
+    sll         t8, t8, 13      // z2
+    subu        s1, t4, t8      // tmp12
+    addu        s2, t5, t2      // tmp10
+    subu        t2, t5, t2      // tmp11
+    addu        s3, t5, s1      // tmp21
+    subu        s1, t5, s1      // tmp24
+    addu        t5, s0, t8      // tmp12
+    addu        v0, s2, t5      // tmp20
+    subu        t5, s2, t5      // tmp25
+    subu        t4, s0, t4
+    subu        t4, t4, t8      // tmp12
+    addu        t8, t2, t4      // tmp22
+    subu        t2, t2, t4      // tmp23
+    // increment counter and pointers
+    addiu       a3, a3, -1
+    addiu       a0, a0, 32
+    // Final stage
+    addu        t4, v0, t7
+    subu        v0, v0, t7
+    addu        t7, s3, t1
+    subu        s3, s3, t1
+    addu        t1, t8, t3
+    subu        t8, t8, t3
+    addu        t3, t2, t9
+    subu        t2, t2, t9
+    addu        t9, s1, t0
+    subu        s1, s1, t0
+    addu        t0, t5, t6
+    subu        t5, t5, t6
+    sll         t4, t4, 4
+    sll         t7, t7, 4
+    sll         t1, t1, 4
+    sll         t3, t3, 4
+    sll         t9, t9, 4
+    sll         t0, t0, 4
+    sll         t5, t5, 4
+    sll         s1, s1, 4
+    sll         t2, t2, 4
+    sll         t8, t8, 4
+    sll         s3, s3, 4
+    sll         v0, v0, 4
+    shll_s.w    t4, t4, 2
+    shll_s.w    t7, t7, 2
+    shll_s.w    t1, t1, 2
+    shll_s.w    t3, t3, 2
+    shll_s.w    t9, t9, 2
+    shll_s.w    t0, t0, 2
+    shll_s.w    t5, t5, 2
+    shll_s.w    s1, s1, 2
+    shll_s.w    t2, t2, 2
+    shll_s.w    t8, t8, 2
+    shll_s.w    s3, s3, 2
+    shll_s.w    v0, v0, 2
+    srl         t4, t4, 24
+    srl         t7, t7, 24
+    srl         t1, t1, 24
+    srl         t3, t3, 24
+    srl         t9, t9, 24
+    srl         t0, t0, 24
+    srl         t5, t5, 24
+    srl         s1, s1, 24
+    srl         t2, t2, 24
+    srl         t8, t8, 24
+    srl         s3, s3, 24
+    srl         v0, v0, 24
+    lw          t6, 0(a1)
+    addiu       t4, t4, 0x80
+    addiu       t7, t7, 0x80
+    addiu       t1, t1, 0x80
+    addiu       t3, t3, 0x80
+    addiu       t9, t9, 0x80
+    addiu       t0, t0, 0x80
+    addiu       t5, t5, 0x80
+    addiu       s1, s1, 0x80
+    addiu       t2, t2, 0x80
+    addiu       t8, t8, 0x80
+    addiu       s3, s3, 0x80
+    addiu       v0, v0, 0x80
+    sb          t4, 0(t6)
+    sb          t7, 1(t6)
+    sb          t1, 2(t6)
+    sb          t3, 3(t6)
+    sb          t9, 4(t6)
+    sb          t0, 5(t6)
+    sb          t5, 6(t6)
+    sb          s1, 7(t6)
+    sb          t2, 8(t6)
+    sb          t8, 9(t6)
+    sb          s3, 10(t6)
+    sb          v0, 11(t6)
+    bgtz        a3, 1b
+     addiu      a1, a1, 4
+
+    RESTORE_REGS_FROM_STACK 16, s0, s1, s2, s3
+
+    jr          ra
+     nop
+
+END(jsimd_idct_12x12_pass2_dspr2)
+
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_convsamp_dspr2)
+/*
+ * a0 = sample_data
+ * a1 = start_col
+ * a2 = workspace
+ */
+    lw            t0, 0(a0)
+    li            t7, 0xff80ff80
+    addu          t0, t0, a1
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    lw            t0, 4(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 0(a2)
+    usw           t4, 4(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 8(a2)
+    usw           t6, 12(a2)
+
+    lw            t0, 8(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 16(a2)
+    usw           t4, 20(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 24(a2)
+    usw           t6, 28(a2)
+
+    lw            t0, 12(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 32(a2)
+    usw           t4, 36(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 40(a2)
+    usw           t6, 44(a2)
+
+    lw            t0, 16(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 48(a2)
+    usw           t4, 52(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 56(a2)
+    usw           t6, 60(a2)
+
+    lw            t0, 20(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 64(a2)
+    usw           t4, 68(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 72(a2)
+    usw           t6, 76(a2)
+
+    lw            t0, 24(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 80(a2)
+    usw           t4, 84(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 88(a2)
+    usw           t6, 92(a2)
+
+    lw            t0, 28(a0)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu          t0, t0, a1
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    ulw           t1, 0(t0)
+    ulw           t2, 4(t0)
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 96(a2)
+    usw           t4, 100(a2)
+    preceu.ph.qbr t3, t1
+    preceu.ph.qbl t4, t1
+    usw           t5, 104(a2)
+    usw           t6, 108(a2)
+    preceu.ph.qbr t5, t2
+    preceu.ph.qbl t6, t2
+    addu.ph       t3, t3, t7
+    addu.ph       t4, t4, t7
+    addu.ph       t5, t5, t7
+    addu.ph       t6, t6, t7
+    usw           t3, 112(a2)
+    usw           t4, 116(a2)
+    usw           t5, 120(a2)
+    usw           t6, 124(a2)
+
+    j             ra
+     nop
+
+END(jsimd_convsamp_dspr2)
+
+
+#ifndef __mips_soft_float
+
+/*****************************************************************************/
+LEAF_DSPR2(jsimd_convsamp_float_dspr2)
+/*
+ * a0 = sample_data
+ * a1 = start_col
+ * a2 = workspace
+ */
+    .set at
+
+    lw          t0, 0(a0)
+    addu        t0, t0, a1
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 4(a0)
+    swc1        f2, 0(a2)
+    swc1        f4, 4(a2)
+    swc1        f6, 8(a2)
+    addu        t0, t0, a1
+    swc1        f8, 12(a2)
+    swc1        f10, 16(a2)
+    swc1        f12, 20(a2)
+    swc1        f14, 24(a2)
+    swc1        f16, 28(a2)
+    // elemr 1
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 8(a0)
+    swc1        f2, 32(a2)
+    swc1        f4, 36(a2)
+    swc1        f6, 40(a2)
+    addu        t0, t0, a1
+    swc1        f8, 44(a2)
+    swc1        f10, 48(a2)
+    swc1        f12, 52(a2)
+    swc1        f14, 56(a2)
+    swc1        f16, 60(a2)
+    // elemr 2
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 12(a0)
+    swc1        f2, 64(a2)
+    swc1        f4, 68(a2)
+    swc1        f6, 72(a2)
+    addu        t0, t0, a1
+    swc1        f8, 76(a2)
+    swc1        f10, 80(a2)
+    swc1        f12, 84(a2)
+    swc1        f14, 88(a2)
+    swc1        f16, 92(a2)
+    //  elemr 3
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 16(a0)
+    swc1        f2, 96(a2)
+    swc1        f4, 100(a2)
+    swc1        f6, 104(a2)
+    addu        t0, t0, a1
+    swc1        f8, 108(a2)
+    swc1        f10, 112(a2)
+    swc1        f12, 116(a2)
+    swc1        f14, 120(a2)
+    swc1        f16, 124(a2)
+    // elemr 4
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 20(a0)
+    swc1        f2, 128(a2)
+    swc1        f4, 132(a2)
+    swc1        f6, 136(a2)
+    addu        t0, t0, a1
+    swc1        f8, 140(a2)
+    swc1        f10, 144(a2)
+    swc1        f12, 148(a2)
+    swc1        f14, 152(a2)
+    swc1        f16, 156(a2)
+    // elemr 5
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 24(a0)
+    swc1        f2, 160(a2)
+    swc1        f4, 164(a2)
+    swc1        f6, 168(a2)
+    addu        t0, t0, a1
+    swc1        f8, 172(a2)
+    swc1        f10, 176(a2)
+    swc1        f12, 180(a2)
+    swc1        f14, 184(a2)
+    swc1        f16, 188(a2)
+    // elemr 6
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    lw          t0, 28(a0)
+    swc1        f2, 192(a2)
+    swc1        f4, 196(a2)
+    swc1        f6, 200(a2)
+    addu        t0, t0, a1
+    swc1        f8, 204(a2)
+    swc1        f10, 208(a2)
+    swc1        f12, 212(a2)
+    swc1        f14, 216(a2)
+    swc1        f16, 220(a2)
+    // elemr 7
+    lbu         t1, 0(t0)
+    lbu         t2, 1(t0)
+    lbu         t3, 2(t0)
+    lbu         t4, 3(t0)
+    lbu         t5, 4(t0)
+    lbu         t6, 5(t0)
+    lbu         t7, 6(t0)
+    lbu         t8, 7(t0)
+    addiu       t1, t1, -128
+    addiu       t2, t2, -128
+    addiu       t3, t3, -128
+    addiu       t4, t4, -128
+    addiu       t5, t5, -128
+    addiu       t6, t6, -128
+    addiu       t7, t7, -128
+    addiu       t8, t8, -128
+    mtc1        t1, f2
+    mtc1        t2, f4
+    mtc1        t3, f6
+    mtc1        t4, f8
+    mtc1        t5, f10
+    mtc1        t6, f12
+    mtc1        t7, f14
+    mtc1        t8, f16
+    cvt.s.w     f2, f2
+    cvt.s.w     f4, f4
+    cvt.s.w     f6, f6
+    cvt.s.w     f8, f8
+    cvt.s.w     f10, f10
+    cvt.s.w     f12, f12
+    cvt.s.w     f14, f14
+    cvt.s.w     f16, f16
+    swc1        f2, 224(a2)
+    swc1        f4, 228(a2)
+    swc1        f6, 232(a2)
+    swc1        f8, 236(a2)
+    swc1        f10, 240(a2)
+    swc1        f12, 244(a2)
+    swc1        f14, 248(a2)
+    swc1        f16, 252(a2)
+
+    j           ra
+     nop
+
+END(jsimd_convsamp_float_dspr2)
+
+#endif
+
+/*****************************************************************************/