pixman 0.18.2: add 2 more neon patches

3 files changed, 201 insertions, 1 deletions

diff --git a/recipes/xorg-lib/pixman-0.18.2/565-over-neon.patch b/recipes/xorg-lib/pixman-0.18.2/565-over-neon.patch
new file mode 100644
index 0000000000..c8156c4d78
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.18.2/565-over-neon.patch
@@ -0,0 +1,23 @@
+From d297443267de0f5fab49ec245df9055a0dddddaf Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Tue, 25 May 2010 17:21:26 +0000
+Subject: ARM: NEON: don't hit general path for r5g6b5 OVER r5g6b5 operation
+
+OVER can't be reduced to SRC because the source image is not
+considered opaque when repeat is set to PIXMAN_REPEAT_NONE.
+An additional fast path table entry solves the problem.
+---
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 64aa908..394dcea 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -246,6 +246,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
+     PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null,     x8b8g8r8, neon_composite_over_8888_8888),
+     PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, null,     a8r8g8b8, neon_composite_src_x888_8888),
+     PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, null,     a8b8g8r8, neon_composite_src_x888_8888),
++    PIXMAN_STD_FAST_PATH (OVER, r5g6b5,   null,     r5g6b5,   neon_composite_src_0565_0565),
+     PIXMAN_STD_FAST_PATH (ADD,  solid,    a8,       a8,       neon_composite_add_n_8_8),
+     PIXMAN_STD_FAST_PATH (ADD,  a8,       a8,       a8,       neon_composite_add_8_8_8),
+     PIXMAN_STD_FAST_PATH (ADD,  a8r8g8b8, a8r8g8b8, a8r8g8b8, neon_composite_add_8888_8888_8888),
+--
+cgit v0.8.3-6-g21f6
diff --git a/recipes/xorg-lib/pixman-0.18.2/neon-reverse-u.patch b/recipes/xorg-lib/pixman-0.18.2/neon-reverse-u.patch
new file mode 100644
index 0000000000..76726ce1ae
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.18.2/neon-reverse-u.patch
@@ -0,0 +1,175 @@
+From e0017c2a676b267900e48c6f32a6e973395c83d3 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Mon, 31 May 2010 16:24:43 +0000
+Subject: ARM: 'neon_combine_out_reverse_u' combiner
+
+---
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index f30869e..44fbfce 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -597,7 +597,7 @@ generate_composite_function_single_scanline \
+ 
+ /******************************************************************************/
+ 
+-.macro pixman_composite_over_8888_8888_process_pixblock_head
++.macro pixman_composite_out_reverse_8888_8888_process_pixblock_head
+     vmvn.8      d24, d3  /* get inverted alpha */
+     /* do alpha blending */
+     vmull.u8    q8, d24, d4
+@@ -606,7 +606,7 @@ generate_composite_function_single_scanline \
+     vmull.u8    q11, d24, d7
+ .endm
+ 
+-.macro pixman_composite_over_8888_8888_process_pixblock_tail
++.macro pixman_composite_out_reverse_8888_8888_process_pixblock_tail
+     vrshr.u16   q14, q8, #8
+     vrshr.u16   q15, q9, #8
+     vrshr.u16   q12, q10, #8
+@@ -615,6 +615,56 @@ generate_composite_function_single_scanline \
+     vraddhn.u16 d29, q15, q9
+     vraddhn.u16 d30, q12, q10
+     vraddhn.u16 d31, q13, q11
++.endm
++
++.macro pixman_composite_out_reverse_8888_8888_process_pixblock_tail_head
++    vld4.8      {d4, d5, d6, d7}, [DST_R, :128]!
++        vrshr.u16   q14, q8, #8
++                                    PF add PF_X, PF_X, #8
++                                    PF tst PF_CTL, #0xF
++        vrshr.u16   q15, q9, #8
++        vrshr.u16   q12, q10, #8
++        vrshr.u16   q13, q11, #8
++                                    PF addne PF_X, PF_X, #8
++                                    PF subne PF_CTL, PF_CTL, #1
++        vraddhn.u16 d28, q14, q8
++        vraddhn.u16 d29, q15, q9
++                                    PF cmp PF_X, ORIG_W
++        vraddhn.u16 d30, q12, q10
++        vraddhn.u16 d31, q13, q11
++    vld4.8      {d0, d1, d2, d3}, [SRC]!
++                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
++    vmvn.8      d22, d3
++                                    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
++        vst4.8      {d28, d29, d30, d31}, [DST_W, :128]!
++                                    PF subge PF_X, PF_X, ORIG_W
++    vmull.u8    q8, d22, d4
++                                    PF subges PF_CTL, PF_CTL, #0x10
++    vmull.u8    q9, d22, d5
++                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
++    vmull.u8    q10, d22, d6
++                                    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
++    vmull.u8    q11, d22, d7
++.endm
++
++generate_composite_function_single_scanline \
++    pixman_composite_scanline_out_reverse_asm_neon, 32, 0, 32, \
++    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
++    8, /* number of pixels, processed in a single block */ \
++    default_init, \
++    default_cleanup, \
++    pixman_composite_out_reverse_8888_8888_process_pixblock_head, \
++    pixman_composite_out_reverse_8888_8888_process_pixblock_tail, \
++    pixman_composite_out_reverse_8888_8888_process_pixblock_tail_head
++
++/******************************************************************************/
++
++.macro pixman_composite_over_8888_8888_process_pixblock_head
++    pixman_composite_out_reverse_8888_8888_process_pixblock_head
++.endm
++
++.macro pixman_composite_over_8888_8888_process_pixblock_tail
++    pixman_composite_out_reverse_8888_8888_process_pixblock_tail
+     vqadd.u8    q14, q0, q14
+     vqadd.u8    q15, q1, q15
+ .endm
+@@ -1416,7 +1466,7 @@ generate_composite_function_single_scanline \
+ 
+ /******************************************************************************/
+ 
+-.macro pixman_composite_over_8888_n_8888_process_pixblock_head
++.macro pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
+     /* expecting source data in {d0, d1, d2, d3} */
+     /* destination data in {d4, d5, d6, d7} */
+     /* solid mask is in d15 */
+@@ -1442,7 +1492,7 @@ generate_composite_function_single_scanline \
+     vmull.u8    q11, d24, d7
+ .endm
+ 
+-.macro pixman_composite_over_8888_n_8888_process_pixblock_tail
++.macro pixman_composite_out_reverse_8888_n_8888_process_pixblock_tail
+     vrshr.u16   q14, q8, #8
+     vrshr.u16   q15, q9, #8
+     vrshr.u16   q12, q10, #8
+@@ -1451,6 +1501,49 @@ generate_composite_function_single_scanline \
+     vraddhn.u16 d29, q15, q9
+     vraddhn.u16 d30, q12, q10
+     vraddhn.u16 d31, q13, q11
++.endm
++
++.macro pixman_composite_out_reverse_8888_8888_8888_init
++    vpush       {d8-d15}
++.endm
++
++.macro pixman_composite_out_reverse_8888_8888_8888_cleanup
++    vpop        {d8-d15}
++.endm
++
++/* TODO: expand macros and do better instructions scheduling */
++.macro pixman_composite_out_reverse_8888_8888_8888_process_pixblock_tail_head
++    vld4.8     {d4, d5, d6, d7}, [DST_R, :128]!
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_tail
++    vld4.8     {d0, d1, d2, d3}, [SRC]!
++    cache_preload 8, 8
++    vld4.8     {d12, d13, d14, d15}, [MASK]!
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
++    vst4.8     {d28, d29, d30, d31}, [DST_W, :128]!
++.endm
++
++generate_composite_function_single_scanline \
++    pixman_composite_scanline_out_reverse_mask_asm_neon, 32, 32, 32, \
++    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
++    8, /* number of pixels, processed in a single block */ \
++    pixman_composite_out_reverse_8888_8888_8888_init, \
++    pixman_composite_out_reverse_8888_8888_8888_cleanup, \
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_head, \
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_tail, \
++    pixman_composite_out_reverse_8888_8888_8888_process_pixblock_tail_head \
++    28, /* dst_w_basereg */ \
++    4,  /* dst_r_basereg */ \
++    0,  /* src_basereg   */ \
++    12  /* mask_basereg  */
++
++/******************************************************************************/
++
++.macro pixman_composite_over_8888_n_8888_process_pixblock_head
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
++.endm
++
++.macro pixman_composite_over_8888_n_8888_process_pixblock_tail
++    pixman_composite_out_reverse_8888_n_8888_process_pixblock_tail
+     vqadd.u8    q14, q0, q14
+     vqadd.u8    q15, q1, q15
+ .endm
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 394dcea..1be9606 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -335,6 +335,7 @@ neon_combine_##name##_u (pixman_implementation_t *imp,                   \
+ 
+ BIND_COMBINE_U (over)
+ BIND_COMBINE_U (add)
++BIND_COMBINE_U (out_reverse)
+ 
+ void
+ pixman_fetch_scanline_r5g6b5_asm_neon (int             width,
+@@ -382,6 +383,7 @@ _pixman_implementation_create_arm_neon (void)
+ 
+     imp->combine_32[PIXMAN_OP_OVER] = neon_combine_over_u;
+     imp->combine_32[PIXMAN_OP_ADD] = neon_combine_add_u;
++    imp->combine_32[PIXMAN_OP_OUT_REVERSE] = neon_combine_out_reverse_u;
+ 
+     _pixman_bits_override_accessors (PIXMAN_r5g6b5,
+                                      neon_fetch_scanline_r5g6b5,
+--
+cgit v0.8.3-6-g21f6
diff --git a/recipes/xorg-lib/pixman_0.18.2.bb b/recipes/xorg-lib/pixman_0.18.2.bb
index d07084015e..9b37cb3740 100644
--- a/recipes/xorg-lib/pixman_0.18.2.bb
+++ b/recipes/xorg-lib/pixman_0.18.2.bb
@@ -1,5 +1,5 @@
 require pixman.inc
-PR = "${INC_PR}.0"
+PR = "${INC_PR}.1"
 
 DEFAULT_PREFERENCE = "-1"
 DEFAULT_PREFERENCE_angstrom = "2"
@@ -15,6 +15,8 @@ SRC_URI += "\
            file://0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch\
            file://0006-A-copy-paste-version-of-16bpp-bilinear-scanline-fetc.patch\
            file://0007-ARM-added-missing-cache-preload.patch\
+           file://565-over-neon.patch \
+           file://neon-reverse-u.patch \
 "
 
 NEON = " --disable-arm-neon "