ARM: added 'neon_src_rpixbuf_8888' fast path

With this optimization added, pixman assisted conversion from
non-premultiplied to premultiplied alpha format is now fully
NEON optimized (both with and without R/B color components
swapping in the process).

2 files changed, 62 insertions, 0 deletions

diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index cf014fa7..4dddde11 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -2072,6 +2072,63 @@ generate_composite_function \
 
 /******************************************************************************/
 
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_head
+    vmull.u8    q8, d3, d0
+    vmull.u8    q9, d3, d1
+    vmull.u8    q10, d3, d2
+.endm
+
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_tail
+    vrshr.u16   q11, q8, #8
+    vswp        d3, d31
+    vrshr.u16   q12, q9, #8
+    vrshr.u16   q13, q10, #8
+    vraddhn.u16 d28, q11, q8
+    vraddhn.u16 d29, q12, q9
+    vraddhn.u16 d30, q13, q10
+.endm
+
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_tail_head
+        vrshr.u16   q11, q8, #8
+        vswp        d3, d31
+        vrshr.u16   q12, q9, #8
+        vrshr.u16   q13, q10, #8
+    fetch_src_pixblock
+        vraddhn.u16 d28, q11, q8
+                                    PF add PF_X, PF_X, #8
+                                    PF tst PF_CTL, #0xF
+                                    PF addne PF_X, PF_X, #8
+                                    PF subne PF_CTL, PF_CTL, #1
+        vraddhn.u16 d29, q12, q9
+        vraddhn.u16 d30, q13, q10
+    vmull.u8    q8, d3, d0
+    vmull.u8    q9, d3, d1
+    vmull.u8    q10, d3, d2
+        vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
+                                    PF cmp PF_X, ORIG_W
+                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+                                    PF subge PF_X, PF_X, ORIG_W
+                                    PF subges PF_CTL, PF_CTL, #0x10
+                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+.endm
+
+generate_composite_function \
+    pixman_composite_src_rpixbuf_8888_asm_neon, 32, 0, 32, \
+    FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_src_rpixbuf_8888_process_pixblock_head, \
+    pixman_composite_src_rpixbuf_8888_process_pixblock_tail, \
+    pixman_composite_src_rpixbuf_8888_process_pixblock_tail_head, \
+    28, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
 .macro pixman_composite_over_0565_8_0565_process_pixblock_head
     /* mask is in d15 */
     convert_0565_to_x888 q4, d2, d1, d0
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index e3eca2b3..c28c481b 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -52,6 +52,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_0888_0565_rev,
                                    uint8_t, 3, uint16_t, 1)
 PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_pixbuf_8888,
                                    uint32_t, 1, uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_rpixbuf_8888,
+                                   uint32_t, 1, uint32_t, 1)
 PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, add_8_8,
                                    uint8_t, 1, uint8_t, 1)
 PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, add_8888_8888,
@@ -249,6 +251,9 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
     PIXMAN_STD_FAST_PATH (SRC,  b8g8r8,   null,     x8r8g8b8, neon_composite_src_0888_8888_rev),
     PIXMAN_STD_FAST_PATH (SRC,  b8g8r8,   null,     r5g6b5,   neon_composite_src_0888_0565_rev),
     PIXMAN_STD_FAST_PATH (SRC,  pixbuf,   pixbuf,   a8r8g8b8, neon_composite_src_pixbuf_8888),
+    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 (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),