--- /dev/null
+From 5234e3c2c161ed5fc92caa336ae78f89112c7d77 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Tue, 16 Mar 2010 16:55:28 +0100
+Subject: [PATCH 1/5] 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 bddf79a..f525744 100644
+--- a/pixman/pixman-general.c
++++ b/pixman/pixman-general.c
+@@ -285,9 +285,24 @@ 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 d5767af..eeb677d 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>
+@@ -867,4 +868,46 @@ 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 */
+--
+1.6.6.1
+
--- /dev/null
+From f607cd0250d398077b0c51201258775e372cb3c3 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 2/5] 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 e084e7f..6212b31 100644
+--- a/pixman/pixman-mmx.c
++++ b/pixman/pixman-mmx.c
+@@ -2994,34 +2994,43 @@ 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--)
+@@ -3031,7 +3040,7 @@ 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))
+ {
+--
+1.6.6.1
+
--- /dev/null
+From 45a9a537f94a7feab47bd82171c7d620f0d34e3f 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 3/5] 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 946e7ba..66053ae 100644
+--- a/pixman/pixman-sse2.c
++++ b/pixman/pixman-sse2.c
+@@ -5299,34 +5299,43 @@ 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);
+@@ -5339,7 +5348,7 @@ 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);
+--
+1.6.6.1
+
--- /dev/null
+From 4cf5bfc72a724fb653c48338b93e91dccea238af 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 4/5] 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 6808b36..7feee1d 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -168,26 +168,66 @@ 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;
+--
+1.6.6.1
+
--- /dev/null
+From d1410558827fce8aac354274a7150fa915881c50 Mon Sep 17 00:00:00 2001
+From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+Date: Thu, 10 Dec 2009 00:51:50 +0200
+Subject: [PATCH 5/5] ARM: added NEON optimizations for fetch/store r5g6b5 scanline
+
+---
+ pixman/pixman-access.c | 23 ++++++++++++++++++++++-
+ pixman/pixman-arm-neon-asm.S | 20 ++++++++++++++++++++
+ pixman/pixman-arm-neon.c | 41 +++++++++++++++++++++++++++++++++++++++++
+ pixman/pixman-private.h | 5 +++++
+ 4 files changed, 88 insertions(+), 1 deletions(-)
+
+diff --git a/pixman/pixman-access.c b/pixman/pixman-access.c
+index fa0a267..5bb3e09 100644
+--- a/pixman/pixman-access.c
++++ b/pixman/pixman-access.c
+@@ -2748,7 +2748,7 @@ 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),
+@@ -2891,6 +2891,27 @@ _pixman_bits_image_setup_raw_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_raw_32 = fetch_func;
++ if (store_func)
++ info->store_scanline_raw_32 = store_func;
++ return;
++ }
++ info++;
++ }
++}
++
+ #else
+
+ void
+diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
+index 51bc347..f30869e 100644
+--- a/pixman/pixman-arm-neon-asm.S
++++ b/pixman/pixman-arm-neon-asm.S
+@@ -458,6 +458,16 @@ 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
+@@ -493,6 +503,16 @@ 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_8000_8000_process_pixblock_head
+diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
+index 7feee1d..fda7a09 100644
+--- a/pixman/pixman-arm-neon.c
++++ b/pixman/pixman-arm-neon.c
+@@ -375,6 +375,43 @@ neon_combine_##name##_u (pixman_implementation_t *imp, \
+ BIND_COMBINE_U (over)
+ BIND_COMBINE_U (add)
+
++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,
++ uint32_t mask_bits)
++{
++ 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)
+ {
+@@ -385,6 +422,10 @@ _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;
+
++ _pixman_bits_override_accessors (PIXMAN_r5g6b5,
++ neon_fetch_scanline_r5g6b5,
++ neon_store_scanline_r5g6b5);
++
+ imp->blt = arm_neon_blt;
+ imp->fill = arm_neon_fill;
+
+diff --git a/pixman/pixman-private.h b/pixman/pixman-private.h
+index eeb677d..ba2d401 100644
+--- a/pixman/pixman-private.h
++++ b/pixman/pixman-private.h
+@@ -220,6 +220,11 @@ void
+ _pixman_bits_image_setup_raw_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,
+--
+1.6.6.1
+
--- /dev/null
+require pixman.inc
+PR = "${INC_PR}.0"
+
+DEFAULT_PREFERENCE = "-1"
+
+SRC_URI += "\
+ file://0001-Generic-C-implementation-of-pixman_blt-with-overlapp.patch;patch=1 \
+ file://0002-Support-of-overlapping-src-dst-for-pixman_blt_mmx.patch;patch=1 \
+ file://0003-Support-of-overlapping-src-dst-for-pixman_blt_sse2.patch;patch=1 \
+ file://0004-Support-of-overlapping-src-dst-for-pixman_blt_neon.patch;patch=1 \
+ file://0005-ARM-added-NEON-optimizations-for-fetch-store-r5g6b5-.patch;patch=1 \
+"
+
+SRC_URI[archive.md5sum] = "a4fb870fc325be258089f1683642e976"
+SRC_URI[archive.sha256sum] = "b305291bba3d9271a4481e5eedf901025ac8ba4ec8f7b76ccafc5094610cd4ff"
+
+NEON = " --disable-arm-neon "
+NEON_armv7a = " "
+
+EXTRA_OECONF = "${NEON} --disable-gtk"
git.openpandora.org / openembedded.git / commitdiff