--- /dev/null
+From c29c9fa826b7112156fd6150b5f1564227935c05 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Thu, 22 Oct 2009 05:27:33 +0300
+Subject: [PATCH 3/6] Generic C implementation of pixman_blt with overlapping support
+
+Uses memcpy/memmove functions to copy pixels, can handle the
+case when both source and destination areas are in the same
+image (this is useful for scrolling).
+
+It is assumed that copying direction is only important when
+using the same image for both source and destination (and
+src_stride == dst_stride). Copying direction is undefined
+for the images with different source and destination stride
+which happen to be in the overlapped areas (but this is an
+unrealistic case anyway).
+---
+ pixman/pixman-general.c | 21 ++++++++++++++++++---
+ pixman/pixman-private.h | 43 +++++++++++++++++++++++++++++++++++++++++++
+ 2 files changed, 61 insertions(+), 3 deletions(-)
+
+diff --git a/pixman/pixman-general.c b/pixman/pixman-general.c
+index c96a3f9..d71a299 100644
+--- a/pixman/pixman-general.c
++++ b/pixman/pixman-general.c
+@@ -300,9 +300,24 @@ general_blt (pixman_implementation_t *imp,
+ int width,
+ int height)
+ {
+- /* We can't blit unless we have sse2 or mmx */
+-
+- return FALSE;
++ uint8_t *dst_bytes = (uint8_t *)dst_bits;
++ uint8_t *src_bytes = (uint8_t *)src_bits;
++ int bpp;
++
++ if (src_bpp != dst_bpp || src_bpp & 7)
++ return FALSE;
++
++ bpp = src_bpp >> 3;
++ width *= bpp;
++ src_stride *= 4;
++ dst_stride *= 4;
++ pixman_blt_helper (src_bytes + src_y * src_stride + src_x * bpp,
++ dst_bytes + dst_y * dst_stride + dst_x * bpp,
++ src_stride,
++ dst_stride,
++ width,
++ height);
++ return TRUE;
+ }
+
+ static pixman_bool_t
+diff --git a/pixman/pixman-private.h b/pixman/pixman-private.h
+index 5000f91..8c5d4fd 100644
+--- a/pixman/pixman-private.h
++++ b/pixman/pixman-private.h
+@@ -10,6 +10,7 @@
+
+ #include "pixman.h"
+ #include <time.h>
++#include <string.h>
+ #include <assert.h>
+
+ #include "pixman-compiler.h"
+@@ -794,4 +795,46 @@ void pixman_timer_register (pixman_timer_t *timer);
+
+ #endif /* PIXMAN_TIMERS */
+
++/* a helper function, can blit 8-bit images with src/dst overlapping support */
++static inline void
++pixman_blt_helper (uint8_t *src_bytes,
++ uint8_t *dst_bytes,
++ int src_stride,
++ int dst_stride,
++ int width,
++ int height)
++{
++ /*
++ * The second part of this check is not strictly needed, but it prevents
++ * unnecessary upside-down processing of areas which belong to different
++ * images. Upside-down processing can be slower with fixed-distance-ahead
++ * prefetch and perceived as having more tearing.
++ */
++ if (src_bytes < dst_bytes + width &&
++ src_bytes + src_stride * height > dst_bytes)
++ {
++ src_bytes += src_stride * height - src_stride;
++ dst_bytes += dst_stride * height - dst_stride;
++ dst_stride = -dst_stride;
++ src_stride = -src_stride;
++ /* Horizontal scrolling to the left needs memmove */
++ if (src_bytes + width > dst_bytes)
++ {
++ while (--height >= 0)
++ {
++ memmove (dst_bytes, src_bytes, width);
++ dst_bytes += dst_stride;
++ src_bytes += src_stride;
++ }
++ return;
++ }
++ }
++ while (--height >= 0)
++ {
++ memcpy (dst_bytes, src_bytes, width);
++ dst_bytes += dst_stride;
++ src_bytes += src_stride;
++ }
++}
++
+ #endif /* PIXMAN_PRIVATE_H */
+--
+1.6.2.4
+
--- /dev/null
+From 7ca32542c957ff308a6ca7e3715e6552a65ae395 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Thu, 22 Oct 2009 05:45:47 +0300
+Subject: [PATCH 4/6] Support of overlapping src/dst for pixman_blt_mmx
+
+---
+ pixman/pixman-mmx.c | 55 +++++++++++++++++++++++++++++---------------------
+ 1 files changed, 32 insertions(+), 23 deletions(-)
+
+diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
+index 819e3a0..dcccadb 100644
+--- a/pixman/pixman-mmx.c
++++ b/pixman/pixman-mmx.c
+@@ -3002,34 +3002,43 @@ pixman_blt_mmx (uint32_t *src_bits,
+ {
+ uint8_t * src_bytes;
+ uint8_t * dst_bytes;
+- int byte_width;
++ int bpp;
+
+- if (src_bpp != dst_bpp)
++ if (src_bpp != dst_bpp || src_bpp & 7)
+ return FALSE;
+
+- if (src_bpp == 16)
+- {
+- src_stride = src_stride * (int) sizeof (uint32_t) / 2;
+- dst_stride = dst_stride * (int) sizeof (uint32_t) / 2;
+- src_bytes = (uint8_t *)(((uint16_t *)src_bits) + src_stride * (src_y) + (src_x));
+- dst_bytes = (uint8_t *)(((uint16_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+- byte_width = 2 * width;
+- src_stride *= 2;
+- dst_stride *= 2;
+- }
+- else if (src_bpp == 32)
++ bpp = src_bpp >> 3;
++ width *= bpp;
++ src_stride *= 4;
++ dst_stride *= 4;
++ src_bytes = (uint8_t *)src_bits + src_y * src_stride + src_x * bpp;
++ dst_bytes = (uint8_t *)dst_bits + dst_y * dst_stride + dst_x * bpp;
++
++ if (src_bpp != 16 && src_bpp != 32)
+ {
+- src_stride = src_stride * (int) sizeof (uint32_t) / 4;
+- dst_stride = dst_stride * (int) sizeof (uint32_t) / 4;
+- src_bytes = (uint8_t *)(((uint32_t *)src_bits) + src_stride * (src_y) + (src_x));
+- dst_bytes = (uint8_t *)(((uint32_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+- byte_width = 4 * width;
+- src_stride *= 4;
+- dst_stride *= 4;
++ pixman_blt_helper (src_bytes, dst_bytes, src_stride, dst_stride,
++ width, height);
++ return TRUE;
+ }
+- else
++
++ if (src_bytes < dst_bytes && src_bytes + src_stride * height > dst_bytes)
+ {
+- return FALSE;
++ src_bytes += src_stride * height - src_stride;
++ dst_bytes += dst_stride * height - dst_stride;
++ dst_stride = -dst_stride;
++ src_stride = -src_stride;
++
++ if (src_bytes + width > dst_bytes)
++ {
++ /* TODO: reverse scanline copy using MMX */
++ while (--height >= 0)
++ {
++ memmove (dst_bytes, src_bytes, width);
++ dst_bytes += dst_stride;
++ src_bytes += src_stride;
++ }
++ return TRUE;
++ }
+ }
+
+ while (height--)
+@@ -3039,7 +3048,7 @@ pixman_blt_mmx (uint32_t *src_bits,
+ uint8_t *d = dst_bytes;
+ src_bytes += src_stride;
+ dst_bytes += dst_stride;
+- w = byte_width;
++ w = width;
+
+ while (w >= 2 && ((unsigned long)d & 3))
+ {
+--
+1.6.2.4
+
--- /dev/null
+From edc80b41c6480b7c80ec5f7c835c92b2debb3774 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Thu, 22 Oct 2009 05:45:54 +0300
+Subject: [PATCH 5/6] Support of overlapping src/dst for pixman_blt_sse2
+
+---
+ pixman/pixman-sse2.c | 55 +++++++++++++++++++++++++++++--------------------
+ 1 files changed, 32 insertions(+), 23 deletions(-)
+
+diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
+index 78b0ad1..b84636b 100644
+--- a/pixman/pixman-sse2.c
++++ b/pixman/pixman-sse2.c
+@@ -5300,34 +5300,43 @@ pixman_blt_sse2 (uint32_t *src_bits,
+ {
+ uint8_t * src_bytes;
+ uint8_t * dst_bytes;
+- int byte_width;
++ int bpp;
+
+- if (src_bpp != dst_bpp)
++ if (src_bpp != dst_bpp || src_bpp & 7)
+ return FALSE;
+
+- if (src_bpp == 16)
+- {
+- src_stride = src_stride * (int) sizeof (uint32_t) / 2;
+- dst_stride = dst_stride * (int) sizeof (uint32_t) / 2;
+- src_bytes =(uint8_t *)(((uint16_t *)src_bits) + src_stride * (src_y) + (src_x));
+- dst_bytes = (uint8_t *)(((uint16_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+- byte_width = 2 * width;
+- src_stride *= 2;
+- dst_stride *= 2;
+- }
+- else if (src_bpp == 32)
++ bpp = src_bpp >> 3;
++ width *= bpp;
++ src_stride *= 4;
++ dst_stride *= 4;
++ src_bytes = (uint8_t *)src_bits + src_y * src_stride + src_x * bpp;
++ dst_bytes = (uint8_t *)dst_bits + dst_y * dst_stride + dst_x * bpp;
++
++ if (src_bpp != 16 && src_bpp != 32)
+ {
+- src_stride = src_stride * (int) sizeof (uint32_t) / 4;
+- dst_stride = dst_stride * (int) sizeof (uint32_t) / 4;
+- src_bytes = (uint8_t *)(((uint32_t *)src_bits) + src_stride * (src_y) + (src_x));
+- dst_bytes = (uint8_t *)(((uint32_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+- byte_width = 4 * width;
+- src_stride *= 4;
+- dst_stride *= 4;
++ pixman_blt_helper (src_bytes, dst_bytes, src_stride, dst_stride,
++ width, height);
++ return TRUE;
+ }
+- else
++
++ if (src_bytes < dst_bytes && src_bytes + src_stride * height > dst_bytes)
+ {
+- return FALSE;
++ src_bytes += src_stride * height - src_stride;
++ dst_bytes += dst_stride * height - dst_stride;
++ dst_stride = -dst_stride;
++ src_stride = -src_stride;
++
++ if (src_bytes + width > dst_bytes)
++ {
++ /* TODO: reverse scanline copy using SSE2 */
++ while (--height >= 0)
++ {
++ memmove (dst_bytes, src_bytes, width);
++ dst_bytes += dst_stride;
++ src_bytes += src_stride;
++ }
++ return TRUE;
++ }
+ }
+
+ cache_prefetch ((__m128i*)src_bytes);
+@@ -5340,7 +5349,7 @@ pixman_blt_sse2 (uint32_t *src_bits,
+ uint8_t *d = dst_bytes;
+ src_bytes += src_stride;
+ dst_bytes += dst_stride;
+- w = byte_width;
++ w = width;
+
+ cache_prefetch_next ((__m128i*)s);
+ cache_prefetch_next ((__m128i*)d);
+--
+1.6.2.4
+
--- /dev/null
+From 86870ff530b5e435034bd80207e5758466d96cff Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Wed, 18 Nov 2009 06:08:48 +0200
+Subject: [PATCH 6/6] Support of overlapping src/dst for pixman_blt_neon
+
+---
+ pixman/pixman-arm-neon.c | 63 ++++++++++++++++++++++++++++++++++++++-------
+ 1 files changed, 53 insertions(+), 10 deletions(-)
+
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 495fda4..c632ff5 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -357,26 +357,66 @@
+ int width,
+ int height)
+ {
+- if (src_bpp != dst_bpp)
++ uint8_t * src_bytes;
++ uint8_t * dst_bytes;
++ int bpp;
++
++ if (src_bpp != dst_bpp || src_bpp & 7)
+ return FALSE;
+
++ bpp = src_bpp >> 3;
++ width *= bpp;
++ src_stride *= 4;
++ dst_stride *= 4;
++ src_bytes = (uint8_t *)src_bits + src_y * src_stride + src_x * bpp;
++ dst_bytes = (uint8_t *)dst_bits + dst_y * dst_stride + dst_x * bpp;
++
++ if (src_bpp != 16 && src_bpp != 32)
++ {
++ pixman_blt_helper (src_bytes, dst_bytes, src_stride, dst_stride,
++ width, height);
++ return TRUE;
++ }
++
++ if (src_bytes < dst_bytes && src_bytes + src_stride * height > dst_bytes)
++ {
++ src_bytes += src_stride * height - src_stride;
++ dst_bytes += dst_stride * height - dst_stride;
++ dst_stride = -dst_stride;
++ src_stride = -src_stride;
++
++ if (src_bytes + width > dst_bytes)
++ {
++ /* TODO: reverse scanline copy using NEON */
++ while (--height >= 0)
++ {
++ memmove (dst_bytes, src_bytes, width);
++ dst_bytes += dst_stride;
++ src_bytes += src_stride;
++ }
++ return TRUE;
++ }
++ }
++
+ switch (src_bpp)
+ {
+ case 16:
+ pixman_composite_src_0565_0565_asm_neon (
+- width, height,
+- (uint16_t *)(((char *) dst_bits) +
+- dst_y * dst_stride * 4 + dst_x * 2), dst_stride * 2,
+- (uint16_t *)(((char *) src_bits) +
+- src_y * src_stride * 4 + src_x * 2), src_stride * 2);
++ width >> 1,
++ height,
++ (uint16_t *) dst_bytes,
++ dst_stride >> 1,
++ (uint16_t *) src_bytes,
++ src_stride >> 1);
+ return TRUE;
+ case 32:
+ pixman_composite_src_8888_8888_asm_neon (
+- width, height,
+- (uint32_t *)(((char *) dst_bits) +
+- dst_y * dst_stride * 4 + dst_x * 4), dst_stride,
+- (uint32_t *)(((char *) src_bits) +
+- src_y * src_stride * 4 + src_x * 4), src_stride);
++ width >> 2,
++ height,
++ (uint32_t *) dst_bytes,
++ dst_stride >> 2,
++ (uint32_t *) src_bytes,
++ src_stride >> 2);
+ return TRUE;
+ default:
+ return FALSE;
+--
+1.6.2.4
+
--- /dev/null
+From 138d38f68bb2f955ca209f7412002a983a32a2fd Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Tue, 23 Feb 2010 23:44:00 +0000
+Subject: ARM: added 'neon_composite_over_n_8888_8888_ca' fast path
+
+This fast path function improves performance of 'firefox-talos-gfx'
+cairo-perf trace.
+
+Benchmark from ARM Cortex-A8 @720MHz
+
+before:
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image firefox-talos-gfx 139.969 141.176 0.35% 6/6
+
+after:
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image firefox-talos-gfx 111.810 112.196 0.23% 6/6
+---
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 2986884..2db4da8 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -1026,6 +1026,113 @@ generate_composite_function \
+
+ /******************************************************************************/
+
++.macro pixman_composite_over_n_8888_8888_ca_process_pixblock_head
++ /*
++ * 'combine_mask_ca' replacement
++ *
++ * input: solid src (n) in {d8, d9, d10, d11}
++ * dest in {d4, d5, d6, d7 }
++ * mask in {d24, d25, d26, d27}
++ * output: updated src in {d0, d1, d2, d3 }
++ * updated mask in {d24, d25, d26, d27}
++ */
++ vmull.u8 q0, d24, d8
++ vmull.u8 q1, d25, d9
++ vmull.u8 q6, d26, d10
++ vmull.u8 q7, d27, d11
++ vmull.u8 q9, d11, d24
++ vmull.u8 q12, d11, d25
++ vmull.u8 q13, d11, d26
++ vrshr.u16 q10, q0, #8
++ vrshr.u16 q11, q1, #8
++ vrshr.u16 q8, q6, #8
++ vraddhn.u16 d0, q0, q10
++ vraddhn.u16 d1, q1, q11
++ vraddhn.u16 d2, q6, q8
++ vrshr.u16 q10, q7, #8
++ vrshr.u16 q11, q9, #8
++ vrshr.u16 q8, q12, #8
++ vraddhn.u16 d3, q7, q10
++ vrshr.u16 q10, q13, #8
++ vraddhn.u16 d25, q12, q8
++ vrshr.u16 q8, q7, #8
++ vraddhn.u16 d24, q9, q11
++ vraddhn.u16 d26, q13, q10
++ vraddhn.u16 d27, q7, q8
++ /*
++ * 'combine_over_ca' replacement
++ *
++ * output: updated dest in {d28, d29, d30, d31}
++ */
++ vmvn.8 d24, d24
++ vmvn.8 d25, d25
++ vmull.u8 q8, d24, d4
++ vmull.u8 q9, d25, d5
++ vmvn.8 d26, d26
++ vmvn.8 d27, d27
++ vmull.u8 q10, d26, d6
++ vmull.u8 q11, d27, d7
++.endm
++
++.macro pixman_composite_over_n_8888_8888_ca_process_pixblock_tail
++ /* ... continue 'combine_over_ca' replacement */
++ vrshr.u16 q14, q8, #8
++ vrshr.u16 q15, q9, #8
++ vrshr.u16 q6, q10, #8
++ vrshr.u16 q7, q11, #8
++ vraddhn.u16 d28, q14, q8
++ vraddhn.u16 d29, q15, q9
++ vraddhn.u16 d30, q6, q10
++ vraddhn.u16 d31, q7, q11
++ vqadd.u8 q14, q0, q14
++ vqadd.u8 q15, q1, q15
++.endm
++
++.macro pixman_composite_over_n_8888_8888_ca_process_pixblock_tail_head
++ vrshr.u16 q14, q8, #8
++ vrshr.u16 q15, q9, #8
++ vld4.8 {d4, d5, d6, d7}, [DST_R, :128]!
++ vrshr.u16 q6, q10, #8
++ vrshr.u16 q7, q11, #8
++ vraddhn.u16 d28, q14, q8
++ vraddhn.u16 d29, q15, q9
++ vraddhn.u16 d30, q6, q10
++ vraddhn.u16 d31, q7, q11
++ vld4.8 {d24, d25, d26, d27}, [MASK]!
++ vqadd.u8 q14, q0, q14
++ vqadd.u8 q15, q1, q15
++ cache_preload 8, 8
++ pixman_composite_over_n_8888_8888_ca_process_pixblock_head
++ vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
++.endm
++
++.macro pixman_composite_over_n_8888_8888_ca_init
++ add DUMMY, sp, #ARGS_STACK_OFFSET
++ vpush {d8-d15}
++ vld1.32 {d11[0]}, [DUMMY]
++ vdup.8 d8, d11[0]
++ vdup.8 d9, d11[1]
++ vdup.8 d10, d11[2]
++ vdup.8 d11, d11[3]
++.endm
++
++.macro pixman_composite_over_n_8888_8888_ca_cleanup
++ vpop {d8-d15}
++.endm
++
++generate_composite_function \
++ pixman_composite_over_n_8888_8888_ca_asm_neon, 0, 32, 32, \
++ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ 5, /* prefetch distance */ \
++ pixman_composite_over_n_8888_8888_ca_init, \
++ pixman_composite_over_n_8888_8888_ca_cleanup, \
++ pixman_composite_over_n_8888_8888_ca_process_pixblock_head, \
++ pixman_composite_over_n_8888_8888_ca_process_pixblock_tail, \
++ pixman_composite_over_n_8888_8888_ca_process_pixblock_tail_head
++
++/******************************************************************************/
++
+ .macro pixman_composite_add_n_8_8_process_pixblock_head
+ /* expecting 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 557301e..00b5c35 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -269,6 +269,7 @@ BIND_SRC_NULL_DST(over_8888_8888, uint32_t, 1, uint32_t, 1)
+
+ BIND_N_MASK_DST(over_n_8_0565, uint8_t, 1, uint16_t, 1)
+ BIND_N_MASK_DST(over_n_8_8888, uint8_t, 1, uint32_t, 1)
++BIND_N_MASK_DST(over_n_8888_8888_ca, uint32_t, 1, uint32_t, 1)
+ BIND_N_MASK_DST(add_n_8_8, uint8_t, 1, uint8_t, 1)
+
+ BIND_SRC_N_DST(over_8888_n_8888, uint32_t, 1, uint32_t, 1)
+@@ -412,6 +413,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
+ PIXMAN_STD_FAST_PATH (OVER, solid, null, r5g6b5, neon_composite_over_n_0565),
+ PIXMAN_STD_FAST_PATH (OVER, solid, null, a8r8g8b8, neon_composite_over_n_8888),
+ PIXMAN_STD_FAST_PATH (OVER, solid, null, x8r8g8b8, neon_composite_over_n_8888),
++ PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, a8r8g8b8, neon_composite_over_n_8888_8888_ca),
++ PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, x8r8g8b8, neon_composite_over_n_8888_8888_ca),
+ PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, a8r8g8b8, neon_composite_over_8888_n_8888),
+ PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, x8r8g8b8, neon_composite_over_8888_n_8888),
+ PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, a8, a8r8g8b8, neon_composite_over_8888_8_8888),
+--
+cgit v0.8.3-6-g21f6
--- /dev/null
+From fa7f7b05fbb08b23678cf0d8928f1511e5a20ecc Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Wed, 24 Feb 2010 00:26:57 +0000
+Subject: ARM: added 'neon_composite_src_x888_8888' fast path
+
+This fast path function improves performance of 'gnome-system-monitor'
+cairo-perf trace.
+
+Benchmark from ARM Cortex-A8 @720MHz
+
+before:
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image gnome-system-monitor 68.838 68.899 0.05% 5/6
+
+after:
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image gnome-system-monitor 53.336 53.384 0.09% 6/6
+---
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 2db4da8..42ac1cb 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -947,6 +947,44 @@ generate_composite_function \
+
+ /******************************************************************************/
+
++.macro pixman_composite_src_x888_8888_process_pixblock_head
++ vorr q0, q0, q2
++ vorr q1, q1, q2
++.endm
++
++.macro pixman_composite_src_x888_8888_process_pixblock_tail
++.endm
++
++.macro pixman_composite_src_x888_8888_process_pixblock_tail_head
++ vst1.32 {d0, d1, d2, d3}, [DST_W, :128]!
++ vld1.32 {d0, d1, d2, d3}, [SRC]!
++ vorr q0, q0, q2
++ vorr q1, q1, q2
++ cache_preload 8, 8
++.endm
++
++.macro pixman_composite_src_x888_8888_init
++ vmov.u8 q2, #0xFF
++ vshl.u32 q2, q2, #24
++.endm
++
++generate_composite_function \
++ pixman_composite_src_x888_8888_asm_neon, 32, 0, 32, \
++ FLAG_DST_WRITEONLY, \
++ 8, /* number of pixels, processed in a single block */ \
++ 10, /* prefetch distance */ \
++ pixman_composite_src_x888_8888_init, \
++ default_cleanup, \
++ pixman_composite_src_x888_8888_process_pixblock_head, \
++ pixman_composite_src_x888_8888_process_pixblock_tail, \
++ pixman_composite_src_x888_8888_process_pixblock_tail_head, \
++ 0, /* dst_w_basereg */ \
++ 0, /* dst_r_basereg */ \
++ 0, /* src_basereg */ \
++ 0 /* mask_basereg */
++
++/******************************************************************************/
++
+ .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 00b5c35..12d92a2 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -251,6 +251,7 @@ neon_composite_##name (pixman_implementation_t *imp, \
+
+
+ BIND_SRC_NULL_DST(src_8888_8888, uint32_t, 1, uint32_t, 1)
++BIND_SRC_NULL_DST(src_x888_8888, uint32_t, 1, uint32_t, 1)
+ BIND_SRC_NULL_DST(src_0565_0565, uint16_t, 1, uint16_t, 1)
+ BIND_SRC_NULL_DST(src_0888_0888, uint8_t, 3, uint8_t, 3)
+ BIND_SRC_NULL_DST(src_8888_0565, uint32_t, 1, uint16_t, 1)
+@@ -400,6 +401,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
+ PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, x8r8g8b8, neon_composite_src_8888_8888),
+ PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, x8b8g8r8, neon_composite_src_8888_8888),
+ PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, x8b8g8r8, neon_composite_src_8888_8888),
++ PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, a8r8g8b8, neon_composite_src_8888_8888),
++ PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, a8r8g8b8, neon_composite_src_x888_8888),
+ PIXMAN_STD_FAST_PATH (SRC, r8g8b8, null, r8g8b8, neon_composite_src_0888_0888),
+ 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),
+@@ -428,6 +431,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
+ PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, x8r8g8b8, neon_composite_over_8888_8888),
+ PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, a8b8g8r8, neon_composite_over_8888_8888),
+ 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 (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
--- /dev/null
+From 80b75405277bacc0df0ef7d91f1a2eabefb97901 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Wed, 24 Feb 2010 02:14:45 +0000
+Subject: ARM: added 'neon_composite_over_reverse_n_8888' fast path
+
+This fast path function improves performance of 'poppler' cairo-perf trace.
+
+Benchmark from ARM Cortex-A8 @720MHz
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image poppler 38.986 39.158 0.23% 6/6
+
+after:
+
+[ # ] backend test min(s) median(s) stddev. count
+[ 0] image poppler 24.981 25.136 0.28% 6/6
+---
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 42ac1cb..272da27 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -680,6 +680,61 @@ generate_composite_function \
+
+ /******************************************************************************/
+
++.macro pixman_composite_over_reverse_n_8888_process_pixblock_tail_head
++ 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
++ vqadd.u8 q14, q0, q14
++ vqadd.u8 q15, q1, q15
++ vld4.8 {d0, d1, d2, d3}, [DST_R, :128]!
++ 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
++ vmull.u8 q10, d22, d6
++ PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
++ vmull.u8 q11, d22, d7
++.endm
++
++.macro pixman_composite_over_reverse_n_8888_init
++ add DUMMY, sp, #ARGS_STACK_OFFSET
++ vld1.32 {d7[0]}, [DUMMY]
++ vdup.8 d4, d7[0]
++ vdup.8 d5, d7[1]
++ vdup.8 d6, d7[2]
++ vdup.8 d7, d7[3]
++.endm
++
++generate_composite_function \
++ pixman_composite_over_reverse_n_8888_asm_neon, 0, 0, 32, \
++ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ 5, /* prefetch distance */ \
++ pixman_composite_over_reverse_n_8888_init, \
++ default_cleanup, \
++ pixman_composite_over_8888_8888_process_pixblock_head, \
++ pixman_composite_over_8888_8888_process_pixblock_tail, \
++ pixman_composite_over_reverse_n_8888_process_pixblock_tail_head, \
++ 28, /* dst_w_basereg */ \
++ 0, /* dst_r_basereg */ \
++ 4, /* src_basereg */ \
++ 24 /* mask_basereg */
++
++/******************************************************************************/
++
+ .macro pixman_composite_over_n_8_0565_process_pixblock_head
+ /* in */
+ vmull.u8 q0, d24, d8
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 12d92a2..417ce5a 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -264,6 +264,7 @@ BIND_SRC_NULL_DST(add_8888_8888, uint32_t, 1, uint32_t, 1)
+
+ BIND_N_NULL_DST(over_n_0565, uint16_t, 1)
+ BIND_N_NULL_DST(over_n_8888, uint32_t, 1)
++BIND_N_NULL_DST(over_reverse_n_8888, uint32_t, 1)
+
+ BIND_SRC_NULL_DST(over_8888_0565, uint32_t, 1, uint16_t, 1)
+ BIND_SRC_NULL_DST(over_8888_8888, uint32_t, 1, uint32_t, 1)
+@@ -438,6 +439,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
+ PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, neon_composite_add_8000_8000),
+ PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, neon_composite_add_8888_8888),
+ PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, neon_composite_add_8888_8888),
++ PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8r8g8b8, neon_composite_over_reverse_n_8888),
+
+ { PIXMAN_OP_NONE },
+ };
+--
+cgit v0.8.3-6-g21f6
--- /dev/null
+SECTION = "libs"
+PRIORITY = "optional"
+DESCRIPTION = "Low-level pixel manipulation library."
+LICENSE = "X11"
+
+DEFAULT_PREFERENCE = "-1"
+DEFAULT_PREFERENCE_angstrom = "2"
+
+BBCLASSEXTEND="native"
+
+SRC_URI = "http://cairographics.org/releases/pixman-${PV}.tar.gz;name=archive \
+ file://0003-Generic-C-implementation-of-pixman_blt-with-overlapp.patch;patch=1 \
+ file://0004-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch;patch=1 \
+ file://0005-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch;patch=1 \
+ file://0006-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch;patch=1 \
+ file://1-composite.patch;patch=1 \
+ file://2-composite.patch;patch=1 \
+ file://3-composite.patch;patch=1 \
+ "
+
+SRC_URI[archive.md5sum] = "a7deb2ff6b286b676d67aa6ae91317ae"
+SRC_URI[archive.sha256sum] = "ea24e9003455a0881bd43bf7e4169f2b34c90c8521405103e3490553876a81b4"
+
+
+inherit autotools_stage
+
+NEON = " --disable-arm-neon "
+NEON_armv7a = ""
+
+EXTRA_OECONF = "${NEON} --disable-gtk"
+
+AUTOTOOLS_STAGE_PKGCONFIG = "1"
+
SRCREV = "c97b1e803fc214e9880eaeff98410c8fa37f9ddc"
DEFAULT_PREFERENCE = "-1"
-DEFAULT_PREFERENCE_angstrom = "1"
DEFAULT_PREFERENCE_shr = "1"
SRC_URI = "git://anongit.freedesktop.org/pixman;protocol=git \
git.openpandora.org / openembedded.git / commitdiff