From 3dca89a677528845ccaee7ff6d355207c631ddd2 Mon Sep 17 00:00:00 2001
From: =?utf8?q?S=C3=B8ren=20Sandmann=20Pedersen?= <sandmann@redhat.com>
Date: Mon, 11 Jun 2007 21:35:14 -0400
Subject: [PATCH] Add pixman_fill_mmx() function

---
 pixman/pixman-mmx.c   | 127 ++++++++++++++++++++++++++++++++++++++++++++++++++
 pixman/pixman-mmx.h   |  20 ++++++++
 pixman/pixman-utils.c |  22 +++++++++
 pixman/pixman.h       |   9 +++-
 4 files changed, 177 insertions(+), 1 deletion(-)

diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 3986cc8..84f440d 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -1640,6 +1640,133 @@ fbCompositeSolidMask_nx8x8888mmx (pixman_op_t op,
     _mm_empty();
 }
 
+pixman_bool_t
+pixman_fill_mmx (uint32_t *bits,
+		 int stride,
+		 int bpp,
+		 int x,
+		 int y,
+		 int width,
+		 int height,
+		 uint32_t xor)
+{
+    ullong	fill;
+    __m64	vfill;
+    uint32_t	byte_width;
+    uint8_t	*byte_line;
+#ifdef __GNUC__
+    __m64	v1, v2, v3, v4, v5, v6, v7;
+#endif
+    
+    if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
+	return FALSE;
+    
+    if (bpp != 16 && bpp != 32)
+	return FALSE;
+    
+    if (bpp == 16)
+    {
+	stride = stride * sizeof (uint32_t) / 2;
+	byte_line = (uint8_t *)(((uint16_t *)bits) + stride * y + x);
+	byte_width = 2 * width;
+	stride *= 2;
+    }
+    else
+    {
+	stride = stride * sizeof (uint32_t) / 4;
+	byte_line = (uint8_t *)(((uint32_t *)bits) + stride * y + x);
+	byte_width = 4 * width;
+	stride *= 4;
+    }
+    
+    fill = ((ullong)xor << 32) | xor;
+    vfill = (__m64)fill;
+    
+#ifdef __GNUC__
+    __asm__ (
+	"movq		%7,	%0\n"
+	"movq		%7,	%1\n"
+	"movq		%7,	%2\n"
+	"movq		%7,	%3\n"
+	"movq		%7,	%4\n"
+	"movq		%7,	%5\n"
+	"movq		%7,	%6\n"
+	: "=y" (v1), "=y" (v2), "=y" (v3),
+	  "=y" (v4), "=y" (v5), "=y" (v6), "=y" (v7)
+	: "y" (vfill));
+#endif
+    
+    while (height--)
+    {
+	int w;
+	uint8_t *d = byte_line;
+	byte_line += stride;
+	w = byte_width;
+	
+	while (w >= 2 && ((unsigned long)d & 3))
+	{
+	    *(uint16_t *)d = xor;
+	    w -= 2;
+	    d += 2;
+	}
+	
+	while (w >= 4 && ((unsigned long)d & 7))
+	{
+	    *(uint32_t *)d = xor;
+	    
+	    w -= 4;
+	    d += 4;
+	}
+
+	while (w >= 64)
+	{
+#ifdef __GNUC__
+	    __asm__ (
+		"movq	%1,	  (%0)\n"
+		"movq	%2,	 8(%0)\n"
+		"movq	%3,	16(%0)\n"
+		"movq	%4,	24(%0)\n"
+		"movq	%5,	32(%0)\n"
+		"movq	%6,	40(%0)\n"
+		"movq	%7,	48(%0)\n"
+		"movq	%8,	56(%0)\n"
+		:
+		: "r" (d),
+		  "y" (vfill), "y" (v1), "y" (v2), "y" (v3),
+		  "y" (v4), "y" (v5), "y" (v6), "y" (v7)
+		: "memory");
+#else
+	    *(__m64*) (d +  0) = vfill;
+	    *(__m64*) (d +  8) = vfill;
+	    *(__m64*) (d + 16) = vfill;
+	    *(__m64*) (d + 24) = vfill;
+	    *(__m64*) (d + 32) = vfill;
+	    *(__m64*) (d + 40) = vfill;
+	    *(__m64*) (d + 48) = vfill;
+	    *(__m64*) (d + 56) = vfill;
+#endif    
+	    w -= 64;
+	    d += 64;
+	}
+	
+	while (w >= 4)
+	{
+	    *(uint32_t *)d = xor;
+	    
+	    w -= 4;
+	    d += 4;
+	}
+	if (w >= 2)
+	{
+	    *(uint16_t *)d = xor;
+	    w -= 2;
+	    d += 2;
+	}
+    }
+    
+    _mm_empty();
+    return TRUE;
+}
 
 #if 0
 /* FIXME */
diff --git a/pixman/pixman-mmx.h b/pixman/pixman-mmx.h
index 9ea2ad7..be8af88 100644
--- a/pixman/pixman-mmx.h
+++ b/pixman/pixman-mmx.h
@@ -46,6 +46,26 @@ pixman_bool_t pixman_have_mmx(void);
 
 #ifdef USE_MMX
 
+pixman_bool_t 
+pixman_blt_mmx (uint32_t *src_bits,
+		uint32_t *dst_bits,
+		int src_stride,
+		int dst_stride,
+		int src_bpp,
+		int dst_bpp,
+		int src_x, int src_y,
+		int dst_x, int dst_y,
+		int width, int height);
+pixman_bool_t
+pixman_fill_mmx (uint32_t *bits,
+		 int stride,
+		 int bpp,
+		 int x,
+		 int y,
+		 int width,
+		 int height,
+		 uint32_t xor);
+
 void fbComposeSetupMMX(void);
 
 void fbCompositeSolidMask_nx8888x0565Cmmx (pixman_op_t      op,
diff --git a/pixman/pixman-utils.c b/pixman/pixman-utils.c
index ab6a1cd..4ac30aa 100644
--- a/pixman/pixman-utils.c
+++ b/pixman/pixman-utils.c
@@ -24,6 +24,7 @@
 #include <config.h>
 #include "pixman.h"
 #include "pixman-private.h"
+#include "pixman-mmx.h"
 
 pixman_bool_t
 pixman_transform_point_3d (pixman_transform_t *transform,
@@ -78,3 +79,24 @@ pixman_blt (uint32_t *src_bits,
 #endif
 	return FALSE;
 }
+
+pixman_bool_t
+pixman_fill (uint32_t *bits,
+		 int stride,
+		 int bpp,
+		 int x,
+		 int y,
+		 int width,
+		 int height,
+		 uint32_t xor)
+{
+#ifdef USE_MMX
+    if (pixman_have_mmx())
+    {
+	return pixman_fill_mmx (bits, stride, bpp, x, y, width, height, xor);
+    }
+    else
+#endif
+	return FALSE;
+}
+	    
diff --git a/pixman/pixman.h b/pixman/pixman.h
index 0421723..277fc0f 100644
--- a/pixman/pixman.h
+++ b/pixman/pixman.h
@@ -300,7 +300,14 @@ pixman_bool_t pixman_blt (uint32_t *src_bits,
 			  int src_x, int src_y,
 			  int dst_x, int dst_y,
 			  int width, int height);
-
+pixman_bool_t pixman_fill (uint32_t *bits,
+			   int stride,
+			   int bpp,
+			   int x,
+			   int y,
+			   int width,
+			   int height,
+			   uint32_t xor);
 /*
  * Images
  */
-- 
2.7.4