gfx/ycbcr/yuv_convert_arm.cpp

   1 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 // contributor Siarhei Siamashka <siarhei.siamashka@gmail.com>
   6
   7 #include "yuv_convert.h"
   8
   9 void __attribute((noinline)) yv12_to_rgb565_neon(uint16 *dst, const uint8 *y, const uint8 *u, const uint8 *v, int n, int oddflag)
  10 {
  11     static __attribute__((aligned(16))) uint16 acc_r[8] = {
  12         22840, 22840, 22840, 22840, 22840, 22840, 22840, 22840,
  13     };
  14     static __attribute__((aligned(16))) uint16 acc_g[8] = {
  15         17312, 17312, 17312, 17312, 17312, 17312, 17312, 17312,
  16     };
  17     static __attribute__((aligned(16))) uint16 acc_b[8] = {
  18         28832, 28832, 28832, 28832, 28832, 28832, 28832, 28832,
  19     };
  20     /*
  21      * Registers:
  22      * q0, q1 : d0, d1, d2, d3  - are used for initial loading of YUV data
  23      * q2     : d4, d5          - are used for storing converted RGB data
  24      * q3     : d6, d7          - are used for temporary storage
  25      *
  26      * q4-q7 - reserved
  27      *
  28      * q8, q9 : d16, d17, d18, d19  - are used for expanded Y data
  29      * q10    : d20, d21
  30      * q11    : d22, d23
  31      * q12    : d24, d25
  32      * q13    : d26, d27
  33      * q13, q14, q15            - various constants (#16, #149, #204, #50, #104, #154)
  34      */
  35     asm volatile (
  36 ".fpu neon\n"
  37 ".macro convert_macroblock size\n"
  38 /* load up to 16 source pixels */
  39         ".if \\size == 16\n"
  40             "pld [%[y], #64]\n"
  41             "pld [%[u], #64]\n"
  42             "pld [%[v], #64]\n"
  43             "vld1.8 {d1}, [%[y]]!\n"
  44             "vld1.8 {d3}, [%[y]]!\n"
  45             "vld1.8 {d0}, [%[u]]!\n"
  46             "vld1.8 {d2}, [%[v]]!\n"
  47         ".elseif \\size == 8\n"
  48             "vld1.8 {d1}, [%[y]]!\n"
  49             "vld1.8 {d0[0]}, [%[u]]!\n"
  50             "vld1.8 {d0[1]}, [%[u]]!\n"
  51             "vld1.8 {d0[2]}, [%[u]]!\n"
  52             "vld1.8 {d0[3]}, [%[u]]!\n"
  53             "vld1.8 {d2[0]}, [%[v]]!\n"
  54             "vld1.8 {d2[1]}, [%[v]]!\n"
  55             "vld1.8 {d2[2]}, [%[v]]!\n"
  56             "vld1.8 {d2[3]}, [%[v]]!\n"
  57         ".elseif \\size == 4\n"
  58             "vld1.8 {d1[0]}, [%[y]]!\n"
  59             "vld1.8 {d1[1]}, [%[y]]!\n"
  60             "vld1.8 {d1[2]}, [%[y]]!\n"
  61             "vld1.8 {d1[3]}, [%[y]]!\n"
  62             "vld1.8 {d0[0]}, [%[u]]!\n"
  63             "vld1.8 {d0[1]}, [%[u]]!\n"
  64             "vld1.8 {d2[0]}, [%[v]]!\n"
  65             "vld1.8 {d2[1]}, [%[v]]!\n"
  66         ".elseif \\size == 2\n"
  67             "vld1.8 {d1[0]}, [%[y]]!\n"
  68             "vld1.8 {d1[1]}, [%[y]]!\n"
  69             "vld1.8 {d0[0]}, [%[u]]!\n"
  70             "vld1.8 {d2[0]}, [%[v]]!\n"
  71         ".elseif \\size == 1\n"
  72             "vld1.8 {d1[0]}, [%[y]]!\n"
  73             "vld1.8 {d0[0]}, [%[u]]!\n"
  74             "vld1.8 {d2[0]}, [%[v]]!\n"
  75         ".else\n"
  76             ".error \"unsupported macroblock size\"\n"
  77         ".endif\n"
  78
  79         /* d1 - Y data (first 8 bytes) */
  80         /* d3 - Y data (next 8 bytes) */
  81         /* d0 - U data, d2 - V data */
  82
  83         /* split even and odd Y color components */
  84         "vuzp.8      d1, d3\n"                       /* d1 - evenY, d3 - oddY */
  85         /* clip upper and lower boundaries */
  86         "vqadd.u8    q0, q0, q4\n"
  87         "vqadd.u8    q1, q1, q4\n"
  88         "vqsub.u8    q0, q0, q5\n"
  89         "vqsub.u8    q1, q1, q5\n"
  90
  91         "vshr.u8     d4, d2, #1\n"                   /* d4 = V >> 1 */
  92
  93         "vmull.u8    q8, d1, d27\n"                  /* q8 = evenY * 149 */
  94         "vmull.u8    q9, d3, d27\n"                  /* q9 = oddY * 149 */
  95
  96         "vld1.16     {d20, d21}, [%[acc_r], :128]\n" /* q10 - initialize accumulator for red */
  97         "vsubw.u8    q10, q10, d4\n"                 /* red acc -= (V >> 1) */
  98         "vmlsl.u8    q10, d2, d28\n"                 /* red acc -= V * 204 */
  99         "vld1.16     {d22, d23}, [%[acc_g], :128]\n" /* q11 - initialize accumulator for green */
 100         "vmlsl.u8    q11, d2, d30\n"                 /* green acc -= V * 104 */
 101         "vmlsl.u8    q11, d0, d29\n"                 /* green acc -= U * 50 */
 102         "vld1.16     {d24, d25}, [%[acc_b], :128]\n" /* q12 - initialize accumulator for blue */
 103         "vmlsl.u8    q12, d0, d30\n"                 /* blue acc -= U * 104 */
 104         "vmlsl.u8    q12, d0, d31\n"                 /* blue acc -= U * 154 */
 105
 106         "vhsub.s16   q3, q8, q10\n"                  /* calculate even red components */
 107         "vhsub.s16   q10, q9, q10\n"                 /* calculate odd red components */
 108         "vqshrun.s16 d0, q3, #6\n"                   /* right shift, narrow and saturate even red components */
 109         "vqshrun.s16 d3, q10, #6\n"                  /* right shift, narrow and saturate odd red components */
 110
 111         "vhadd.s16   q3, q8, q11\n"                  /* calculate even green components */
 112         "vhadd.s16   q11, q9, q11\n"                 /* calculate odd green components */
 113         "vqshrun.s16 d1, q3, #6\n"                   /* right shift, narrow and saturate even green components */
 114         "vqshrun.s16 d4, q11, #6\n"                  /* right shift, narrow and saturate odd green components */
 115
 116         "vhsub.s16   q3, q8, q12\n"                  /* calculate even blue components */
 117         "vhsub.s16   q12, q9, q12\n"                 /* calculate odd blue components */
 118         "vqshrun.s16 d2, q3, #6\n"                   /* right shift, narrow and saturate even blue components */
 119         "vqshrun.s16 d5, q12, #6\n"                  /* right shift, narrow and saturate odd blue components */
 120
 121         "vzip.8      d0, d3\n"                       /* join even and odd red components */
 122         "vzip.8      d1, d4\n"                       /* join even and odd green components */
 123         "vzip.8      d2, d5\n"                       /* join even and odd blue components */
 124
 125         "vshll.u8    q3, d0, #8\n\t"
 126         "vshll.u8    q8, d1, #8\n\t"
 127         "vshll.u8    q9, d2, #8\n\t"
 128         "vsri.u16    q3, q8, #5\t\n"
 129         "vsri.u16    q3, q9, #11\t\n"
 130         /* store pixel data to memory */
 131         ".if \\size == 16\n"
 132         "    vst1.16 {d6, d7}, [%[dst]]!\n"
 133         "    vshll.u8    q3, d3, #8\n\t"
 134         "    vshll.u8    q8, d4, #8\n\t"
 135         "    vshll.u8    q9, d5, #8\n\t"
 136         "    vsri.u16    q3, q8, #5\t\n"
 137         "    vsri.u16    q3, q9, #11\t\n"
 138         "    vst1.16 {d6, d7}, [%[dst]]!\n"
 139         ".elseif \\size == 8\n"
 140         "    vst1.16 {d6, d7}, [%[dst]]!\n"
 141         ".elseif \\size == 4\n"
 142         "    vst1.16 {d6}, [%[dst]]!\n"
 143         ".elseif \\size == 2\n"
 144         "    vst1.16 {d6[0]}, [%[dst]]!\n"
 145         "    vst1.16 {d6[1]}, [%[dst]]!\n"
 146         ".elseif \\size == 1\n"
 147         "    vst1.16 {d6[0]}, [%[dst]]!\n"
 148         ".endif\n"
 149         ".endm\n"
 150
 151         "vmov.u8     d8, #15\n" /* add this to U/V to saturate upper boundary */
 152         "vmov.u8     d9, #20\n" /* add this to Y to saturate upper boundary */
 153         "vmov.u8     d10, #31\n" /* sub this from U/V to saturate lower boundary */
 154         "vmov.u8     d11, #36\n" /* sub this from Y to saturate lower boundary */
 155
 156         "vmov.u8     d26, #16\n"
 157         "vmov.u8     d27, #149\n"
 158         "vmov.u8     d28, #204\n"
 159         "vmov.u8     d29, #50\n"
 160         "vmov.u8     d30, #104\n"
 161         "vmov.u8     d31, #154\n"
 162
 163         "cmp         %[oddflag], #0\n"
 164         "beq         1f\n"
 165         "convert_macroblock 1\n"
 166         "sub         %[n], %[n], #1\n"
 167     "1:\n"
 168         "subs        %[n], %[n], #16\n"
 169         "blt         2f\n"
 170     "1:\n"
 171         "convert_macroblock 16\n"
 172         "subs        %[n], %[n], #16\n"
 173         "bge         1b\n"
 174     "2:\n"
 175         "tst         %[n], #8\n"
 176         "beq         3f\n"
 177         "convert_macroblock 8\n"
 178     "3:\n"
 179         "tst         %[n], #4\n"
 180         "beq         4f\n"
 181         "convert_macroblock 4\n"
 182     "4:\n"
 183         "tst         %[n], #2\n"
 184         "beq         5f\n"
 185         "convert_macroblock 2\n"
 186     "5:\n"
 187         "tst         %[n], #1\n"
 188         "beq         6f\n"
 189         "convert_macroblock 1\n"
 190     "6:\n"
 191         ".purgem convert_macroblock\n"
 192         : [y] "+&r" (y), [u] "+&r" (u), [v] "+&r" (v), [dst] "+&r" (dst), [n] "+&r" (n)
 193         : [acc_r] "r" (&acc_r[0]), [acc_g] "r" (&acc_g[0]), [acc_b] "r" (&acc_b[0]),
 194           [oddflag] "r" (oddflag)
 195         : "cc", "memory",
 196           "d0",  "d1",  "d2",  "d3",  "d4",  "d5",  "d6",  "d7",
 197           "d8",  "d9",  "d10", "d11", /* "d12", "d13", "d14", "d15", */
 198           "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23",
 199           "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31"
 200     );
 201 }