Attachment 12825 Details for Bug 13445 – ARM optimized fill routines

[patch] ARM optimized fill routines

pixman-arm-optimization.diff (text/plain), 7.47 KB, created by Ilpo Ruotsalainen on 2007-11-29 05:29:07 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Ilpo Ruotsalainen

Created: 2007-11-29 05:29:07 UTC

Size: 7.47 KB

patch

obsolete

>commit 23a7d5dea599efec1f459bac64cf9edc4bd5ae11
>Author: Ilpo Ruotsalainen <ilpo.ruotsalainen@movial.fi>
>Date:   Thu Nov 29 12:29:59 2007 +0000
>
>    Implement ARM optimized version of fill routines.
>
>diff --git a/configure.ac b/configure.ac
>index 22a91ef..3ac2a40 100644
>--- a/configure.ac
>+++ b/configure.ac
>@@ -148,6 +148,32 @@ fi
> AM_CONDITIONAL(USE_SSE, test $have_sse_intrinsics = yes)
> 
> dnl ========================================================
>+
>+dnl Test for architechture specific optimizations for this platform
>+
>+AC_MSG_CHECKING(for architechture specific optimizations)
>+
>+use_arch_opts=no
>+
>+case "$host_cpu" in
>+arm)
>+	if test "$GCC" = "yes" ; then
>+		use_arch_opts=yes
>+		ARCH_OPT_SOURCES='pixman-arch-arm.lo'
>+	fi
>+	;;
>+esac
>+
>+AC_MSG_RESULT($use_arch_opts)
>+
>+if test $use_arch_opts = yes ; then
>+	AC_DEFINE(USE_ARCH_OPTS, 1, [use architechture specific optimizations])
>+fi
>+
>+AC_SUBST([ARCH_OPT_SOURCES])
>+AM_CONDITIONAL(USE_ARCH_OPTS, test $use_arch_opts = yes)
>+
>+dnl ========================================================
> AC_SUBST(MMX_CFLAGS)
> 
> PKG_CHECK_MODULES(GTK, [gtk+-2.0], [HAVE_GTK=yes], [HAVE_GTK=no])
>diff --git a/pixman/Makefile.am b/pixman/Makefile.am
>index 66283a2..dab6363 100644
>--- a/pixman/Makefile.am
>+++ b/pixman/Makefile.am
>@@ -20,6 +20,11 @@ libpixman_1_la_SOURCES =		\
> libpixmanincludedir = $(includedir)/pixman-1/
> libpixmaninclude_HEADERS = pixman.h
> 
>+if USE_ARCH_OPTS
>+libpixman_1_la_LIBADD += $(ARCH_OPT_SOURCES)
>+libpixman_1_la_DEPENDENCIES = $(ARCH_OPT_SOURCES)
>+endif
>+
> # mmx code
> if USE_MMX
> noinst_LTLIBRARIES = libpixman-mmx.la
>diff --git a/pixman/pixman-arch-arm.c b/pixman/pixman-arch-arm.c
>new file mode 100644
>index 0000000..655092c
>--- /dev/null
>+++ b/pixman/pixman-arch-arm.c
>@@ -0,0 +1,205 @@
>+/*
>+ * Copyright Â© 2007 Movial Creative Technologies Inc
>+ *
>+ * Permission to use, copy, modify, distribute, and sell this software and its
>+ * documentation for any purpose is hereby granted without fee, provided that
>+ * the above copyright notice appear in all copies and that both that
>+ * copyright notice and this permission notice appear in supporting
>+ * documentation.
>+ *
>+ * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
>+ * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
>+ * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
>+ * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
>+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
>+ * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
>+ * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
>+ * SOFTWARE.
>+ *
>+ * Author: Ilpo Ruotsalainen <ilpo.ruotsalainen@movial.fi>
>+ */
>+
>+#ifdef HAVE_CONFIG_H
>+#include <config.h>
>+#endif
>+
>+#include "pixman.h"
>+#include "pixman-private.h"
>+
>+static void
>+pixman_fill8 (uint32_t  *bits,
>+	      int	stride,
>+	      int	x,
>+	      int	y,
>+	      int	width,
>+	      int	height,
>+	      uint32_t  xor)
>+{
>+    int byte_stride = stride * sizeof (uint32_t);
>+    uint8_t *dst = (uint8_t *) bits;
>+    uint8_t v = xor & 0xff;
>+
>+    xor = v | (v << 8);
>+    xor |= xor << 16;
>+
>+    dst = dst + y * byte_stride + x;
>+
>+    while (height--)
>+    {
>+	uint32_t dummy1, dummy2;
>+
>+	asm volatile(
>+	    /* Check if the fill width is very small */
>+	    "	cmp	%0, #8\n"
>+	    "	bcc	2f\n"
>+	    /* Output single pixels until aligned to word boundary */
>+	    "1:	tst	%1, #3\n"
>+	    "	strneb	%4, [%1], #1\n"
>+	    "	subne	%0, %0, #1\n"
>+	    "	bne	1b\n"
>+	    /* Output up to 16 pixels per iteration */
>+	    "1:	subs	%0, %0, #8\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	subcss	%0, %0, #8\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	bcs	1b\n"
>+	    /* Finish up any remaining pixels */
>+	    "	and	%0, %0, #7\n"
>+	    "2:	subs	%0, %0, #1\n"
>+	    "	strcsb	%4, [%1], #1\n"
>+	    "	subcss	%0, %0, #1\n"
>+	    "	strcsb	%4, [%1], #1\n"
>+	    "	bcs	2b\n"
>+	    : "=r" (dummy1), "=r" (dummy2)
>+	    : "0" (width), "1" (dst), "r" (xor)
>+	    : "cc", "memory"
>+	    );
>+
>+	dst += byte_stride;
>+    }
>+}
>+
>+static void
>+pixman_fill16 (uint32_t *bits,
>+	       int       stride,
>+	       int       x,
>+	       int       y,
>+	       int       width,
>+	       int       height,
>+	       uint32_t  xor)
>+{
>+    int short_stride = (stride * sizeof (uint32_t)) / sizeof (uint16_t);
>+    uint16_t *dst = (uint16_t *)bits;
>+    uint16_t v = xor & 0xffff;
>+
>+    xor = v | v << 16;
>+
>+    dst = dst + y * short_stride + x;
>+
>+    while (height--)
>+    {
>+	uint32_t dummy1, dummy2;
>+
>+	asm volatile(
>+	    /* Check if the fill width is very small */
>+	    "	cmp	%0, #4\n"
>+	    "	bcc	2f\n"
>+	    /* Output single pixels until aligned to word boundary */
>+	    "1:	tst	%1, #2\n"
>+	    "	strneh	%4, [%1], #2\n"
>+	    "	subne	%0, %0, #1\n"
>+	    "	bne	1b\n"
>+	    /* Output up to 8 pixels per iteration */
>+	    "1:	subs	%0, %0, #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	subcss	%0, %0, #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	bcs	1b\n"
>+	    /* Finish up any remaining pixels */
>+	    "	and	%0, %0, #3\n"
>+	    "2:	subs	%0, %0, #1\n"
>+	    "	strcsh	%4, [%1], #2\n"
>+	    "	bcs	2b\n"
>+	    : "=r" (dummy1), "=r" (dummy2)
>+	    : "0" (width), "1" (dst), "r" (xor)
>+	    : "cc", "memory"
>+	    );
>+
>+	dst += short_stride;
>+    }
>+}
>+
>+static void
>+pixman_fill32 (uint32_t *bits,
>+	       int       stride,
>+	       int       x,
>+	       int       y,
>+	       int       width,
>+	       int       height,
>+	       uint32_t  xor)
>+{
>+    bits = bits + y * stride + x;
>+    
>+    while (height--)
>+    {
>+	uint32_t dummy1, dummy2;
>+
>+	asm volatile(
>+	    /* Check if the fill width is very small */
>+	    "	cmp	%0, #2\n"
>+	    "	bcc	2f\n"
>+	    /* Output up to 4 pixels per iteration */
>+	    "1:	subs	%0, %0, #2\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	subcss	%0, %0, #2\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	strcs	%4, [%1], #4\n"
>+	    "	bcs	1b\n"
>+	    /* Output last pixel if necessary */
>+	    "2:	tst	%0, #1\n"
>+	    "	strne	%4, [%1], #4\n"
>+	    : "=r" (dummy1), "=r" (dummy2)
>+	    : "0" (width), "1" (bits), "r" (xor)
>+	    : "cc", "memory"
>+	    );
>+
>+	bits += stride;
>+    }
>+}
>+
>+pixman_bool_t
>+pixman_fill (uint32_t *bits,
>+	     int stride,
>+	     int bpp,
>+	     int x,
>+	     int y,
>+	     int width,
>+	     int height,
>+	     uint32_t xor)
>+{
>+    switch (bpp)
>+    {
>+    case 8:
>+	pixman_fill8 (bits, stride, x, y, width, height, xor);
>+	break;
>+	
>+    case 16:
>+	pixman_fill16 (bits, stride, x, y, width, height, xor);
>+	break;
>+	
>+    case 32:
>+	pixman_fill32 (bits, stride, x, y, width, height, xor);
>+	break;
>+
>+    default:
>+	return FALSE;
>+	break;
>+    }
>+	
>+    return TRUE;
>+}
>diff --git a/pixman/pixman-arch.h b/pixman/pixman-arch.h
>new file mode 100644
>index 0000000..1eee9d3
>--- /dev/null
>+++ b/pixman/pixman-arch.h
>@@ -0,0 +1,7 @@
>+#ifdef USE_ARCH_OPTS
>+
>+#ifdef __arm__
>+#define USE_ARCH_FILL
>+#endif
>+
>+#endif
>diff --git a/pixman/pixman-utils.c b/pixman/pixman-utils.c
>index 1d1dec9..9b6e2c0 100644
>--- a/pixman/pixman-utils.c
>+++ b/pixman/pixman-utils.c
>@@ -28,6 +28,7 @@
> #include <stdlib.h>
> #include "pixman.h"
> #include "pixman-private.h"
>+#include "pixman-arch.h"
> #include "pixman-mmx.h"
> 
> pixman_bool_t
>@@ -84,6 +85,7 @@ pixman_blt (uint32_t *src_bits,
> 	return FALSE;
> }
> 
>+#ifndef USE_ARCH_FILL
> static void
> pixman_fill8 (uint32_t  *bits,
> 	      int	stride,
>@@ -197,7 +199,7 @@ pixman_fill (uint32_t *bits,
> 	
>     return TRUE;
> }
>-	    
>+#endif
> 
> /*
>  * Compute the smallest value no less than y which is on a

Actions: View | Splinter Review

Attachments on bug 13445: 12825 | 18279

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.