ARM: NEON: Standard fast path src_n_8_8888

Performance numbers of before/after on cortex-a8 @ 1GHz

- before
L1:  32.39  L2:  31.79  M: 30.84 ( 13.77%)  HT: 21.58  VT: 19.75  R: 18.83  RT: 10.46 ( 106Kops/s)

- after
L1: 516.25  L2: 372.00  M:193.49 ( 85.59%)  HT:136.93  VT:109.10  R:104.48  RT: 34.77 ( 253Kops/s)

2 files changed, 80 insertions, 0 deletions

diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 3fcd07d..1db02db 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1219,6 +1219,79 @@ generate_composite_function \
 
 /******************************************************************************/
 
+.macro pixman_composite_src_n_8_8888_process_pixblock_head
+    /* expecting solid source in {d0, d1, d2, d3} */
+    /* mask is in d24 (d25, d26, d27 are unused) */
+
+    /* in */
+    vmull.u8    q8, d24, d0
+    vmull.u8    q9, d24, d1
+    vmull.u8    q10, d24, d2
+    vmull.u8    q11, d24, d3
+    vrsra.u16   q8, q8, #8
+    vrsra.u16   q9, q9, #8
+    vrsra.u16   q10, q10, #8
+    vrsra.u16   q11, q11, #8
+.endm
+
+.macro pixman_composite_src_n_8_8888_process_pixblock_tail
+    vrshrn.u16  d28, q8, #8
+    vrshrn.u16  d29, q9, #8
+    vrshrn.u16  d30, q10, #8
+    vrshrn.u16  d31, q11, #8
+.endm
+
+.macro pixman_composite_src_n_8_8888_process_pixblock_tail_head
+    fetch_mask_pixblock
+                                    PF add PF_X, PF_X, #8
+        vrshrn.u16  d28, q8, #8
+                                    PF tst PF_CTL, #0x0F
+        vrshrn.u16  d29, q9, #8
+                                    PF addne PF_X, PF_X, #8
+        vrshrn.u16  d30, q10, #8
+                                    PF subne PF_CTL, PF_CTL, #1
+        vrshrn.u16  d31, q11, #8
+                                    PF cmp PF_X, ORIG_W
+    vmull.u8    q8, d24, d0
+                                    PF pld, [PF_MASK, PF_X, lsl #mask_bpp_shift]
+    vmull.u8    q9, d24, d1
+                                    PF subge PF_X, PF_X, ORIG_W
+    vmull.u8    q10, d24, d2
+                                    PF subges PF_CTL, PF_CTL, #0x10
+    vmull.u8    q11, d24, d3
+                                    PF ldrgeb DUMMY, [PF_MASK, MASK_STRIDE, lsl #mask_bpp_shift]!
+        vst4.8      {d28, d29, d30, d31}, [DST_W :128]!
+    vrsra.u16   q8, q8, #8
+    vrsra.u16   q9, q9, #8
+    vrsra.u16   q10, q10, #8
+    vrsra.u16   q11, q11, #8
+.endm
+
+.macro pixman_composite_src_n_8_8888_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vld1.32     {d3[0]}, [DUMMY]
+    vdup.8      d0, d3[0]
+    vdup.8      d1, d3[1]
+    vdup.8      d2, d3[2]
+    vdup.8      d3, d3[3]
+.endm
+
+.macro pixman_composite_src_n_8_8888_cleanup
+.endm
+
+generate_composite_function \
+    pixman_composite_src_n_8_8888_asm_neon, 0, 8, 32, \
+    FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_src_n_8_8888_init, \
+    pixman_composite_src_n_8_8888_cleanup, \
+    pixman_composite_src_n_8_8888_process_pixblock_head, \
+    pixman_composite_src_n_8_8888_process_pixblock_tail, \
+    pixman_composite_src_n_8_8888_process_pixblock_tail_head, \
+
+/******************************************************************************/
+
 .macro pixman_composite_over_n_8_8888_process_pixblock_head
     /* expecting deinterleaved source data in {d8, d9, d10, d11} */
     /* d8 - blue, d9 - green, d10 - red, d11 - alpha */
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index effb50b..3db9adf 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -90,6 +90,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, add_n_8_8,
                                       uint8_t, 1, uint8_t, 1)
 PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, add_n_8_8888,
                                       uint8_t, 1, uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (0, neon, src_n_8_8888,
+                                      uint8_t, 1, uint32_t, 1)
 
 PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, neon, over_8888_n_8888,
                                      uint32_t, 1, uint32_t, 1)
@@ -289,6 +291,11 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
     PIXMAN_STD_FAST_PATH (SRC,  pixbuf,   pixbuf,   a8b8g8r8, neon_composite_src_rpixbuf_8888),
     PIXMAN_STD_FAST_PATH (SRC,  rpixbuf,  rpixbuf,  a8r8g8b8, neon_composite_src_rpixbuf_8888),
     PIXMAN_STD_FAST_PATH (SRC,  rpixbuf,  rpixbuf,  a8b8g8r8, neon_composite_src_pixbuf_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  solid,    a8,       a8r8g8b8, neon_composite_src_n_8_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  solid,    a8,       x8r8g8b8, neon_composite_src_n_8_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  solid,    a8,       a8b8g8r8, neon_composite_src_n_8_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  solid,    a8,       x8b8g8r8, neon_composite_src_n_8_8888),
+
     PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       a8,       neon_composite_over_n_8_8),
     PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       r5g6b5,   neon_composite_over_n_8_0565),
     PIXMAN_STD_FAST_PATH (OVER, solid,    a8,       b5g6r5,   neon_composite_over_n_8_0565),