[oe] [PATCH 1/3] pixman: add version 0.21.2
Martin Jansa
martin.jansa at gmail.com
Wed Nov 17 14:24:57 UTC 2010
---
.../0000-Add-pixman_bits_override_accessors.patch | 75 ++++++++++
...mplementation-of-pixman_blt-with-overlapp.patch | 114 +++++++++++++++
...of-overlapping-src-dst-for-pixman_blt_mmx.patch | 91 ++++++++++++
...f-overlapping-src-dst-for-pixman_blt_sse2.patch | 91 ++++++++++++
...f-overlapping-src-dst-for-pixman_blt_neon.patch | 94 +++++++++++++
...EON-optimizations-for-fetch-store-r5g6b5-.patch | 109 ++++++++++++++
...EON-optimizations-for-fetch-store-a8-scan.patch | 148 ++++++++++++++++++++
...EON-optimizations-for-fetching-x8r8g8b8-s.patch | 77 ++++++++++
recipes/xorg-lib/pixman_0.21.2.bb | 22 +++
9 files changed, 821 insertions(+), 0 deletions(-)
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0000-Add-pixman_bits_override_accessors.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch
create mode 100644 recipes/xorg-lib/pixman-0.21.2/0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch
create mode 100644 recipes/xorg-lib/pixman_0.21.2.bb
diff --git a/recipes/xorg-lib/pixman-0.21.2/0000-Add-pixman_bits_override_accessors.patch b/recipes/xorg-lib/pixman-0.21.2/0000-Add-pixman_bits_override_accessors.patch
new file mode 100644
index 0000000..1687bd4
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0000-Add-pixman_bits_override_accessors.patch
@@ -0,0 +1,75 @@
+From fff598814365037c8ffdd97afe10a14bb09558fc Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Thu, 10 Dec 2009 00:51:50 +0200
+Subject: [PATCH 1/8] add _pixman_bits_override_accessors
+
+* from patch ARM: HACK: added NEON optimizations for fetch/store r5g6b5 scanline
+* used in
+ 0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch
+ 0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch
+ 0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch
+---
+ pixman/pixman-access.c | 23 ++++++++++++++++++++++-
+ pixman/pixman-private.h | 5 +++++
+ 2 files changed, 27 insertions(+), 1 deletions(-)
+
+diff --git a/pixman/pixman-access.c b/pixman/pixman-access.c
+index f1ce0ba..b33da29 100644
+--- a/pixman/pixman-access.c
++++ b/pixman/pixman-access.c
+@@ -2836,7 +2836,7 @@ typedef struct
+ store_scanline_ ## format, store_scanline_generic_64 \
+ }
+
+-static const format_info_t accessors[] =
++static format_info_t accessors[] =
+ {
+ /* 32 bpp formats */
+ FORMAT_INFO (a8r8g8b8),
+@@ -2978,6 +2978,27 @@ _pixman_bits_image_setup_accessors (bits_image_t *image)
+ setup_accessors (image);
+ }
+
++void
++_pixman_bits_override_accessors (pixman_format_code_t format,
++ fetch_scanline_t fetch_func,
++ store_scanline_t store_func)
++{
++ format_info_t *info = accessors;
++
++ while (info->format != PIXMAN_null)
++ {
++ if (info->format == format)
++ {
++ if (fetch_func)
++ info->fetch_scanline_32 = fetch_func;
++ if (store_func)
++ info->store_scanline_32 = store_func;
++ return;
++ }
++ info++;
++ }
++}
++
+ #else
+
+ void
+diff --git a/pixman/pixman-private.h b/pixman/pixman-private.h
+index 383748a..969dfab 100644
+--- a/pixman/pixman-private.h
++++ b/pixman/pixman-private.h
+@@ -197,6 +197,11 @@ void
+ _pixman_bits_image_setup_accessors (bits_image_t *image);
+
+ void
++_pixman_bits_override_accessors (pixman_format_code_t format,
++ fetch_scanline_t fetch_func,
++ store_scanline_t store_func);
++
++void
+ _pixman_image_get_scanline_generic_64 (pixman_image_t *image,
+ int x,
+ int y,
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch b/recipes/xorg-lib/pixman-0.21.2/0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch
new file mode 100644
index 0000000..d438dfc
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch
@@ -0,0 +1,114 @@
+From 6b162fb9d4ede5faa25f24188964f31d7667e74e Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Tue, 16 Mar 2010 16:55:28 +0100
+Subject: [PATCH 1/7] 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 4d234a0..c4d2c14 100644
+--- a/pixman/pixman-general.c
++++ b/pixman/pixman-general.c
+@@ -280,9 +280,24 @@ 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 969dfab..352bceb 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 <stdio.h>
+ #include <string.h>
+@@ -869,4 +870,46 @@ 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 */
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch b/recipes/xorg-lib/pixman-0.21.2/0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch
new file mode 100644
index 0000000..e86e8ed
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch
@@ -0,0 +1,91 @@
+From 2d6cc769d233c0b1a391b501e84f5c3b0f1af4f8 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Thu, 22 Oct 2009 05:45:47 +0300
+Subject: [PATCH 2/7] 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 34637a4..f9dd473 100644
+--- a/pixman/pixman-mmx.c
++++ b/pixman/pixman-mmx.c
+@@ -2996,34 +2996,43 @@ 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--)
+@@ -3033,7 +3042,7 @@ 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))
+ {
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch b/recipes/xorg-lib/pixman-0.21.2/0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch
new file mode 100644
index 0000000..6fdfa5d
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch
@@ -0,0 +1,91 @@
+From 532b8f45cee61ea2509a7f263dd30f40f3de29ba Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Thu, 22 Oct 2009 05:45:54 +0300
+Subject: [PATCH 3/7] 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 5907de0..25015ae 100644
+--- a/pixman/pixman-sse2.c
++++ b/pixman/pixman-sse2.c
+@@ -5027,34 +5027,43 @@ 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;
++ }
+ }
+
+ while (height--)
+@@ -5064,7 +5073,7 @@ 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;
+
+ while (w >= 2 && ((unsigned long)d & 3))
+ {
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch b/recipes/xorg-lib/pixman-0.21.2/0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch
new file mode 100644
index 0000000..4950dd8
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch
@@ -0,0 +1,94 @@
+From 4e101b976fa5fc72e44553a15516b804ffda0394 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Wed, 18 Nov 2009 06:08:48 +0200
+Subject: [PATCH 4/7] Support of overlapping src/dst for pixman_blt_neon
+
+---
+ pixman/pixman-arm-neon.c | 62 +++++++++++++++++++++++++++++++++++++--------
+ 1 files changed, 51 insertions(+), 11 deletions(-)
+
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 2f82069..6a6ed37 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -185,26 +185,66 @@ pixman_blt_neon (uint32_t *src_bits,
+ 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;
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch b/recipes/xorg-lib/pixman-0.21.2/0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch
new file mode 100644
index 0000000..459c734
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch
@@ -0,0 +1,109 @@
+From 8d7a77b6780af1b96db32026fb8d79c5603f0fba Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Thu, 10 Dec 2009 00:51:50 +0200
+Subject: [PATCH 5/7] ARM: added NEON optimizations for fetch/store r5g6b5 scanline
+
+---
+ pixman/pixman-arm-neon-asm.S | 20 ++++++++++++++++++++
+ pixman/pixman-arm-neon.c | 40 ++++++++++++++++++++++++++++++++++++++++
+ 2 files changed, 60 insertions(+), 0 deletions(-)
+
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 91ec27d..b838f92 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -459,6 +459,16 @@ generate_composite_function \
+ pixman_composite_src_8888_0565_process_pixblock_tail, \
+ pixman_composite_src_8888_0565_process_pixblock_tail_head
+
++generate_composite_function_single_scanline \
++ pixman_store_scanline_r5g6b5_asm_neon, 32, 0, 16, \
++ FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ default_init, \
++ default_cleanup, \
++ pixman_composite_src_8888_0565_process_pixblock_head, \
++ pixman_composite_src_8888_0565_process_pixblock_tail, \
++ pixman_composite_src_8888_0565_process_pixblock_tail_head
++
+ /******************************************************************************/
+
+ .macro pixman_composite_src_0565_8888_process_pixblock_head
+@@ -494,6 +504,16 @@ generate_composite_function \
+ pixman_composite_src_0565_8888_process_pixblock_tail, \
+ pixman_composite_src_0565_8888_process_pixblock_tail_head
+
++generate_composite_function_single_scanline \
++ pixman_fetch_scanline_r5g6b5_asm_neon, 16, 0, 32, \
++ FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ default_init, \
++ default_cleanup, \
++ pixman_composite_src_0565_8888_process_pixblock_head, \
++ pixman_composite_src_0565_8888_process_pixblock_tail, \
++ pixman_composite_src_0565_8888_process_pixblock_tail_head
++
+ /******************************************************************************/
+
+ .macro pixman_composite_add_8_8_process_pixblock_head
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 6a6ed37..4b2bbea 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -422,6 +422,42 @@ BIND_COMBINE_U (over)
+ BIND_COMBINE_U (add)
+ BIND_COMBINE_U (out_reverse)
+
++void
++pixman_fetch_scanline_r5g6b5_asm_neon (int width,
++ uint32_t *buffer,
++ const uint16_t *pixel);
++void
++pixman_store_scanline_r5g6b5_asm_neon (int width,
++ uint16_t *pixel,
++ const uint32_t *values);
++
++static void
++neon_fetch_scanline_r5g6b5 (pixman_image_t *image,
++ int x,
++ int y,
++ int width,
++ uint32_t * buffer,
++ const uint32_t *mask)
++{
++ const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
++ const uint16_t *pixel = (const uint16_t *)bits + x;
++
++ pixman_fetch_scanline_r5g6b5_asm_neon (width, buffer, pixel);
++}
++
++static void
++neon_store_scanline_r5g6b5 (bits_image_t * image,
++ int x,
++ int y,
++ int width,
++ const uint32_t *values)
++{
++ uint32_t *bits = image->bits + image->rowstride * y;
++ uint16_t *pixel = ((uint16_t *) bits) + x;
++
++ pixman_store_scanline_r5g6b5_asm_neon (width, pixel, values);
++}
++
+ pixman_implementation_t *
+ _pixman_implementation_create_arm_neon (void)
+ {
+@@ -437,6 +473,10 @@ _pixman_implementation_create_arm_neon (void)
+ 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,
++ neon_store_scanline_r5g6b5);
++
+ imp->blt = arm_neon_blt;
+ imp->fill = arm_neon_fill;
+
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch b/recipes/xorg-lib/pixman-0.21.2/0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch
new file mode 100644
index 0000000..c317147
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch
@@ -0,0 +1,148 @@
+From b689ddce66ce6391b6478d870f00fe21bbce944c Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Thu, 23 Sep 2010 21:10:56 +0300
+Subject: [PATCH 6/7] ARM: added NEON optimizations for fetch/store a8 scanline
+
+---
+ pixman/pixman-arm-neon-asm.S | 64 ++++++++++++++++++++++++++++++++++++++++++
+ pixman/pixman-arm-neon.c | 42 +++++++++++++++++++++++++++
+ 2 files changed, 106 insertions(+), 0 deletions(-)
+
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index b838f92..8e43a3b 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -418,6 +418,70 @@ generate_composite_function \
+
+ /******************************************************************************/
+
++.macro pixman_composite_src_8_8888_process_pixblock_head
++ /* This is tricky part: we can't set these values just once in 'init' macro
++ * because leading/trailing pixels handling part uses VZIP.8 instructions,
++ * and they operate on values in-place and destroy original registers
++ * content. Think about it like VST4.8 instruction corrupting NEON
++ * registers after write in 'tail_head' macro. Except that 'tail_head'
++ * macro itself actually does not need these extra VMOVs because it uses
++ * real VST4.8 instruction.
++ */
++ vmov.u8 q0, #0
++ vmov.u8 d2, #0
++.endm
++
++.macro pixman_composite_src_8_8888_process_pixblock_tail
++.endm
++
++.macro pixman_composite_src_8_8888_process_pixblock_tail_head
++ vst4.8 {d0, d1, d2, d3}, [DST_W, :128]!
++ vld1.8 {d3}, [SRC]!
++.endm
++
++generate_composite_function_single_scanline \
++ pixman_fetch_scanline_a8_asm_neon, 8, 0, 32, \
++ FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ default_init, \
++ default_cleanup, \
++ pixman_composite_src_8_8888_process_pixblock_head, \
++ pixman_composite_src_8_8888_process_pixblock_tail, \
++ pixman_composite_src_8_8888_process_pixblock_tail_head, \
++ 0, /* dst_w_basereg */ \
++ 0, /* dst_r_basereg */ \
++ 3, /* src_basereg */ \
++ 0 /* mask_basereg */
++
++/******************************************************************************/
++
++.macro pixman_composite_src_8888_8_process_pixblock_head
++.endm
++
++.macro pixman_composite_src_8888_8_process_pixblock_tail
++.endm
++
++.macro pixman_composite_src_8888_8_process_pixblock_tail_head
++ vst1.8 {d3}, [DST_W, :64]!
++ vld4.8 {d0, d1, d2, d3}, [SRC]!
++.endm
++
++generate_composite_function_single_scanline \
++ pixman_store_scanline_a8_asm_neon, 32, 0, 8, \
++ FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
++ 8, /* number of pixels, processed in a single block */ \
++ default_init, \
++ default_cleanup, \
++ pixman_composite_src_8888_8_process_pixblock_head, \
++ pixman_composite_src_8888_8_process_pixblock_tail, \
++ pixman_composite_src_8888_8_process_pixblock_tail_head, \
++ 3, /* dst_w_basereg */ \
++ 0, /* dst_r_basereg */ \
++ 0, /* src_basereg */ \
++ 0 /* mask_basereg */
++
++/******************************************************************************/
++
+ .macro pixman_composite_src_8888_0565_process_pixblock_head
+ vshll.u8 q8, d1, #8
+ vshll.u8 q14, d2, #8
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 4b2bbea..1c68d32 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -458,6 +458,45 @@ neon_store_scanline_r5g6b5 (bits_image_t * image,
+ pixman_store_scanline_r5g6b5_asm_neon (width, pixel, values);
+ }
+
++void
++pixman_fetch_scanline_a8_asm_neon (int width,
++ uint32_t *buffer,
++ const uint8_t *pixel);
++
++
++void
++pixman_store_scanline_a8_asm_neon (int width,
++ uint8_t *pixel,
++ const uint32_t *values);
++
++static void
++neon_fetch_scanline_a8 (pixman_image_t *image,
++ int x,
++ int y,
++ int width,
++ uint32_t * buffer,
++ const uint32_t *mask)
++{
++ const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
++ const uint8_t *pixel = (const uint8_t *) bits + x;
++
++ pixman_fetch_scanline_a8_asm_neon (width, buffer, pixel);
++}
++
++static void
++neon_store_scanline_a8 (bits_image_t * image,
++ int x,
++ int y,
++ int width,
++ const uint32_t *values)
++{
++ uint32_t *bits = image->bits + image->rowstride * y;
++ uint8_t *pixel = (uint8_t *) bits + x;
++
++ pixman_store_scanline_a8_asm_neon (width, pixel, values);
++}
++
++
+ pixman_implementation_t *
+ _pixman_implementation_create_arm_neon (void)
+ {
+@@ -476,6 +515,9 @@ _pixman_implementation_create_arm_neon (void)
+ _pixman_bits_override_accessors (PIXMAN_r5g6b5,
+ neon_fetch_scanline_r5g6b5,
+ neon_store_scanline_r5g6b5);
++ _pixman_bits_override_accessors (PIXMAN_a8,
++ neon_fetch_scanline_a8,
++ neon_store_scanline_a8);
+
+ imp->blt = arm_neon_blt;
+ imp->fill = arm_neon_fill;
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman-0.21.2/0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch b/recipes/xorg-lib/pixman-0.21.2/0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch
new file mode 100644
index 0000000..1abd28a
--- /dev/null
+++ b/recipes/xorg-lib/pixman-0.21.2/0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch
@@ -0,0 +1,77 @@
+From 912d7b4f79cda5dd828f5db7608314057a39338e Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
+Date: Fri, 24 Sep 2010 18:22:44 +0300
+Subject: [PATCH 7/7] ARM: added NEON optimizations for fetching x8r8g8b8 scanline
+
+---
+ pixman/pixman-arm-neon-asm.S | 14 ++++++++++++++
+ pixman/pixman-arm-neon.c | 21 +++++++++++++++++++++
+ 2 files changed, 35 insertions(+), 0 deletions(-)
+
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 8e43a3b..5ebee5a 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -1206,6 +1206,20 @@ generate_composite_function \
+ 0, /* src_basereg */ \
+ 0 /* mask_basereg */
+
++generate_composite_function_single_scanline \
++ pixman_fetch_scanline_x888_asm_neon, 32, 0, 32, \
++ FLAG_DST_WRITEONLY, \
++ 8, /* number of pixels, processed in a single block */ \
++ 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
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 1c68d32..0bcfc54 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -496,6 +496,24 @@ neon_store_scanline_a8 (bits_image_t * image,
+ pixman_store_scanline_a8_asm_neon (width, pixel, values);
+ }
+
++void
++pixman_fetch_scanline_x888_asm_neon (int width,
++ uint32_t *buffer,
++ const uint32_t *pixel);
++
++static void
++neon_fetch_scanline_x888 (pixman_image_t *image,
++ int x,
++ int y,
++ int width,
++ uint32_t * buffer,
++ const uint32_t *mask)
++{
++ const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
++ const uint32_t *pixel = (const uint32_t *) bits + x;
++
++ pixman_fetch_scanline_x888_asm_neon (width, buffer, pixel);
++}
+
+ pixman_implementation_t *
+ _pixman_implementation_create_arm_neon (void)
+@@ -518,6 +536,9 @@ _pixman_implementation_create_arm_neon (void)
+ _pixman_bits_override_accessors (PIXMAN_a8,
+ neon_fetch_scanline_a8,
+ neon_store_scanline_a8);
++ _pixman_bits_override_accessors (PIXMAN_x8r8g8b8,
++ neon_fetch_scanline_x888,
++ NULL);
+
+ imp->blt = arm_neon_blt;
+ imp->fill = arm_neon_fill;
+--
+1.7.3.2
+
diff --git a/recipes/xorg-lib/pixman_0.21.2.bb b/recipes/xorg-lib/pixman_0.21.2.bb
new file mode 100644
index 0000000..7e361b6
--- /dev/null
+++ b/recipes/xorg-lib/pixman_0.21.2.bb
@@ -0,0 +1,22 @@
+require pixman.inc
+
+SRC_URI[archive.md5sum] = "9e09fd6e58cbf9717140891e0b7d4a7a"
+SRC_URI[archive.sha256sum] = "295f51416caf307ff7caf1153ee9b1d86b9f7f02a7876d12db6538d80451c5de"
+
+PR = "${INC_PR}.0"
+
+SRC_URI += "\
+ file://0000-Add-pixman_bits_override_accessors.patch \
+ file://0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch \
+ file://0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch \
+ file://0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch \
+ file://0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch \
+ file://0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch \
+ file://0006-ARM-added-NEON-optimizations-for-fetch-store-a8-scan.patch \
+ file://0007-ARM-added-NEON-optimizations-for-fetching-x8r8g8b8-s.patch \
+"
+
+NEON = " --disable-arm-neon "
+NEON_armv7a = " "
+
+EXTRA_OECONF = "${NEON} --disable-gtk"
--
1.7.3.2
More information about the Openembedded-devel
mailing list