recipes/xorg-lib/pixman-0.17.8/3-composite.patch

   1 From 19fc0c4c8d28796f0fbc2c8f47edc130851ae3a5 Mon Sep 17 00:00:00 2001
   2 From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
   3 Date: Wed, 24 Feb 2010 02:14:45 +0000
   4 Subject: ARM: added 'neon_composite_over_reverse_n_8888' fast path
   5
   6 This fast path function improves performance of 'poppler' cairo-perf trace.
   7
   8 Benchmark from ARM Cortex-A8 @720MHz
   9
  10 before:
  11
  12 [ # ]  backend                         test   min(s) median(s) stddev. count
  13 [  0]    image                      poppler   38.986   39.158   0.23%    6/6
  14
  15 after:
  16
  17 [ # ]  backend                         test   min(s) median(s) stddev. count
  18 [  0]    image                      poppler   24.981   25.136   0.28%    6/6
  19 ---
  20 diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
  21 index f2643de..eb8cc4c 100644
  22 --- a/pixman/pixman-arm-neon-asm.S
  23 +++ b/pixman/pixman-arm-neon-asm.S
  24 @@ -680,6 +680,61 @@ generate_composite_function \
  25
  26  /******************************************************************************/
  27
  28 +.macro pixman_composite_over_reverse_n_8888_process_pixblock_tail_head
  29 +        vrshr.u16   q14, q8, #8
  30 +                                    PF add PF_X, PF_X, #8
  31 +                                    PF tst PF_CTL, #0xF
  32 +        vrshr.u16   q15, q9, #8
  33 +        vrshr.u16   q12, q10, #8
  34 +        vrshr.u16   q13, q11, #8
  35 +                                    PF addne PF_X, PF_X, #8
  36 +                                    PF subne PF_CTL, PF_CTL, #1
  37 +        vraddhn.u16 d28, q14, q8
  38 +        vraddhn.u16 d29, q15, q9
  39 +                                    PF cmp PF_X, ORIG_W
  40 +        vraddhn.u16 d30, q12, q10
  41 +        vraddhn.u16 d31, q13, q11
  42 +        vqadd.u8    q14, q0, q14
  43 +        vqadd.u8    q15, q1, q15
  44 +    vld4.8      {d0, d1, d2, d3}, [DST_R, :128]!
  45 +    vmvn.8      d22, d3
  46 +                                    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
  47 +        vst4.8      {d28, d29, d30, d31}, [DST_W, :128]!
  48 +                                    PF subge PF_X, PF_X, ORIG_W
  49 +    vmull.u8    q8, d22, d4
  50 +                                    PF subges PF_CTL, PF_CTL, #0x10
  51 +    vmull.u8    q9, d22, d5
  52 +    vmull.u8    q10, d22, d6
  53 +                                    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
  54 +    vmull.u8    q11, d22, d7
  55 +.endm
  56 +
  57 +.macro pixman_composite_over_reverse_n_8888_init
  58 +    add         DUMMY, sp, #ARGS_STACK_OFFSET
  59 +    vld1.32     {d7[0]}, [DUMMY]
  60 +    vdup.8      d4, d7[0]
  61 +    vdup.8      d5, d7[1]
  62 +    vdup.8      d6, d7[2]
  63 +    vdup.8      d7, d7[3]
  64 +.endm
  65 +
  66 +generate_composite_function \
  67 +    pixman_composite_over_reverse_n_8888_asm_neon, 0, 0, 32, \
  68 +    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
  69 +    8, /* number of pixels, processed in a single block */ \
  70 +    5, /* prefetch distance */ \
  71 +    pixman_composite_over_reverse_n_8888_init, \
  72 +    default_cleanup, \
  73 +    pixman_composite_over_8888_8888_process_pixblock_head, \
  74 +    pixman_composite_over_8888_8888_process_pixblock_tail, \
  75 +    pixman_composite_over_reverse_n_8888_process_pixblock_tail_head, \
  76 +    28, /* dst_w_basereg */ \
  77 +    0,  /* dst_r_basereg */ \
  78 +    4,  /* src_basereg   */ \
  79 +    24  /* mask_basereg  */
  80 +
  81 +/******************************************************************************/
  82 +
  83  .macro pixman_composite_over_n_8_0565_process_pixblock_head
  84      /* in */
  85      vmull.u8    q0, d24, d8
  86 diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
  87 index 0a7ac31..30c5056 100644
  88 --- a/pixman/pixman-arm-neon.c
  89 +++ b/pixman/pixman-arm-neon.c
  90 @@ -264,6 +264,7 @@ BIND_SRC_NULL_DST(add_8888_8888, uint32_t, 1, uint32_t, 1)
  91
  92  BIND_N_NULL_DST(over_n_0565, uint16_t, 1)
  93  BIND_N_NULL_DST(over_n_8888, uint32_t, 1)
  94 +BIND_N_NULL_DST(over_reverse_n_8888, uint32_t, 1)
  95
  96  BIND_SRC_NULL_DST(over_8888_0565, uint32_t, 1, uint16_t, 1)
  97  BIND_SRC_NULL_DST(over_8888_8888, uint32_t, 1, uint32_t, 1)
  98 @@ -442,6 +443,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
  99      PIXMAN_STD_FAST_PATH (ADD,  a8,       null,     a8,       neon_composite_add_8000_8000),
 100      PIXMAN_STD_FAST_PATH (ADD,  a8r8g8b8, null,     a8r8g8b8, neon_composite_add_8888_8888),
 101      PIXMAN_STD_FAST_PATH (ADD,  a8b8g8r8, null,     a8b8g8r8, neon_composite_add_8888_8888),
 102 +    PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8r8g8b8, neon_composite_over_reverse_n_8888),
 103 +    PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8b8g8r8, neon_composite_over_reverse_n_8888),
 104
 105      { PIXMAN_OP_NONE },
 106  };
 107 --
 108 cgit v0.8.3-6-g21f6