recipes/xorg-lib/pixman-0.21.2/0005-ARM-added-neon_composite_over_n_8_8-fast-path.patch

   1 From 98d08b37f17a3379d0ceff8bb7de8f943873fbd8 Mon Sep 17 00:00:00 2001
   2 From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
   3 Date: Fri, 26 Nov 2010 08:55:49 +0200
   4 Subject: [PATCH 05/24] ARM: added 'neon_composite_over_n_8_8' fast path
   5
   6 ---
   7  pixman/pixman-arm-neon-asm.S |   68 ++++++++++++++++++++++++++++++++++++++++++
   8  pixman/pixman-arm-neon.c     |    3 ++
   9  2 files changed, 71 insertions(+), 0 deletions(-)
  10
  11 diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
  12 index 91ec27d..a3875ee 100644
  13 --- a/pixman/pixman-arm-neon-asm.S
  14 +++ b/pixman/pixman-arm-neon-asm.S
  15 @@ -1203,6 +1203,74 @@ generate_composite_function \
  16
  17  /******************************************************************************/
  18
  19 +.macro pixman_composite_over_n_8_8_process_pixblock_head
  20 +    vmull.u8    q0,  d24, d8
  21 +    vmull.u8    q1,  d25, d8
  22 +    vmull.u8    q6,  d26, d8
  23 +    vmull.u8    q7,  d27, d8
  24 +    vrshr.u16   q10, q0,  #8
  25 +    vrshr.u16   q11, q1,  #8
  26 +    vrshr.u16   q12, q6,  #8
  27 +    vrshr.u16   q13, q7,  #8
  28 +    vraddhn.u16 d0,  q0,  q10
  29 +    vraddhn.u16 d1,  q1,  q11
  30 +    vraddhn.u16 d2,  q6,  q12
  31 +    vraddhn.u16 d3,  q7,  q13
  32 +    vmvn.8      q12, q0
  33 +    vmvn.8      q13, q1
  34 +    vmull.u8    q8,  d24, d4
  35 +    vmull.u8    q9,  d25, d5
  36 +    vmull.u8    q10, d26, d6
  37 +    vmull.u8    q11, d27, d7
  38 +.endm
  39 +
  40 +.macro pixman_composite_over_n_8_8_process_pixblock_tail
  41 +    vrshr.u16   q14, q8,  #8
  42 +    vrshr.u16   q15, q9,  #8
  43 +    vrshr.u16   q12, q10, #8
  44 +    vrshr.u16   q13, q11, #8
  45 +    vraddhn.u16 d28, q14, q8
  46 +    vraddhn.u16 d29, q15, q9
  47 +    vraddhn.u16 d30, q12, q10
  48 +    vraddhn.u16 d31, q13, q11
  49 +    vqadd.u8    q14, q0,  q14
  50 +    vqadd.u8    q15, q1,  q15
  51 +.endm
  52 +
  53 +/* TODO: expand macros and do better instructions scheduling */
  54 +.macro pixman_composite_over_n_8_8_process_pixblock_tail_head
  55 +    vld1.8      {d4, d5, d6, d7}, [DST_R, :128]!
  56 +    pixman_composite_over_n_8_8_process_pixblock_tail
  57 +    vld1.8      {d24, d25, d26, d27}, [MASK]!
  58 +    cache_preload 32, 32
  59 +    vst1.8      {d28, d29, d30, d31}, [DST_W, :128]!
  60 +    pixman_composite_over_n_8_8_process_pixblock_head
  61 +.endm
  62 +
  63 +.macro pixman_composite_over_n_8_8_init
  64 +    add         DUMMY, sp, #ARGS_STACK_OFFSET
  65 +    vpush       {d8-d15}
  66 +    vld1.32     {d8[0]}, [DUMMY]
  67 +    vdup.8      d8, d8[3]
  68 +.endm
  69 +
  70 +.macro pixman_composite_over_n_8_8_cleanup
  71 +    vpop        {d8-d15}
  72 +.endm
  73 +
  74 +generate_composite_function \
  75 +    pixman_composite_over_n_8_8_asm_neon, 0, 8, 8, \
  76 +    FLAG_DST_READWRITE, \
  77 +    32, /* number of pixels, processed in a single block */ \
  78 +    5, /* prefetch distance */ \
  79 +    pixman_composite_over_n_8_8_init, \
  80 +    pixman_composite_over_n_8_8_cleanup, \
  81 +    pixman_composite_over_n_8_8_process_pixblock_head, \
  82 +    pixman_composite_over_n_8_8_process_pixblock_tail, \
  83 +    pixman_composite_over_n_8_8_process_pixblock_tail_head
  84 +
  85 +/******************************************************************************/
  86 +
  87  .macro pixman_composite_over_n_8888_8888_ca_process_pixblock_head
  88      /*
  89       * 'combine_mask_ca' replacement
  90 diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
  91 index 2f82069..72ef75e 100644
  92 --- a/pixman/pixman-arm-neon.c
  93 +++ b/pixman/pixman-arm-neon.c
  94 @@ -76,6 +76,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8888,
  95                                        uint8_t, 1, uint32_t, 1)
  96  PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8888_8888_ca,
  97                                        uint32_t, 1, uint32_t, 1)
  98 +PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8,
  99 +                                      uint8_t, 1, uint8_t, 1)
 100  PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8,
 101                                        uint8_t, 1, uint8_t, 1)
 102
 103 @@ -235,6 +237,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
 104      PIXMAN_STD_FAST_PATH (SRC,  b8g8r8,   null,     x8r8g8b8, neon_composite_src_0888_8888_rev),
 105      PIXMAN_STD_FAST_PATH (SRC,  b8g8r8,   null,     r5g6b5,   neon_composite_src_0888_0565_rev),
 106      PIXMAN_STD_FAST_PATH (SRC,  pixbuf,   pixbuf,   a8r8g8b8, neon_composite_src_pixbuf_8888),
 107 +    PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       a8,       neon_composite_over_n_8_8),
 108      PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       r5g6b5,   neon_composite_over_n_8_0565),
 109      PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       b5g6r5,   neon_composite_over_n_8_0565),
 110      PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       a8r8g8b8, neon_composite_over_n_8_8888),
 111 --
 112 1.6.6.1
 113