Use info->ModeReg.surface_cntl to restore byte swapping for the framebuffer

    aperture on big endian machines, remove superfluous local variables and
    register reads.
Adapt framebuffer aperture byte swapping to texture format before copying
    data to offscreen area and restore it afterwards on big endian
    machines, fixes Render acceleration there.
reviewed by: Hui Yu <hyu@ati.com>, Kevin E. Martin <kem@freedesktop.org>

3 files changed, 75 insertions, 15 deletions

diff --git a/src/radeon_cursor.c b/src/radeon_cursor.c
index 885144b..fba00f3 100644
--- a/src/radeon_cursor.c
+++ b/src/radeon_cursor.c
@@ -73,18 +73,17 @@ static CARD32 mono_cursor_color[] = {
 #if X_BYTE_ORDER == X_BIG_ENDIAN
 
 #define CURSOR_SWAPPING_DECL_MMIO   unsigned char *RADEONMMIO = info->MMIO;
-#define CURSOR_SWAPPING_DECL	    CARD32  __surface_cntl;
 #define CURSOR_SWAPPING_START() \
     OUTREG(RADEON_SURFACE_CNTL, \
-	   ((__surface_cntl = INREG(RADEON_SURFACE_CNTL)) | \
+	   (info->ModeReg.surface_cntl | \
 	    RADEON_NONSURF_AP0_SWP_32BPP) & \
 	   ~RADEON_NONSURF_AP0_SWP_16BPP)
-#define CURSOR_SWAPPING_END()	(OUTREG(RADEON_SURFACE_CNTL, __surface_cntl))
+#define CURSOR_SWAPPING_END()	(OUTREG(RADEON_SURFACE_CNTL, \
+					info->ModeReg.surface_cntl))
 
 #else
 
 #define CURSOR_SWAPPING_DECL_MMIO
-#define CURSOR_SWAPPING_DECL
 #define CURSOR_SWAPPING_START()
 #define CURSOR_SWAPPING_END()
 
@@ -98,7 +97,6 @@ static void RADEONSetCursorColors(ScrnInfoPtr pScrn, int bg, int fg)
     CARD32        *pixels     = (CARD32 *)(pointer)(info->FB + info->cursor_start);
     int            pixel, i;
     CURSOR_SWAPPING_DECL_MMIO
-    CURSOR_SWAPPING_DECL
 
 #ifdef ARGB_CURSOR
     /* Don't recolour cursors set with SetCursorARGB. */
@@ -188,7 +186,6 @@ static void RADEONLoadCursorImage(ScrnInfoPtr pScrn, unsigned char *image)
     CARD32         save2      = 0;
     CARD8	   chunk;
     CARD32         i, j;
-    CURSOR_SWAPPING_DECL
 
     if (!info->IsSecondary) {
 	save1 = INREG(RADEON_CRTC_GEN_CNTL) & ~(CARD32) (3 << 20);
@@ -295,7 +292,6 @@ static void RADEONLoadCursorARGB (ScrnInfoPtr pScrn, CursorPtr pCurs)
     CARD32         save2      = 0;
     CARD32	  *image = pCurs->bits->argb;
     CARD32	  *i;
-    CURSOR_SWAPPING_DECL
 
     if (!image)
 	return;	/* XXX can't happen */
diff --git a/src/radeon_render.c b/src/radeon_render.c
index 7eb293d..8b4f5da 100644
--- a/src/radeon_render.c
+++ b/src/radeon_render.c
@@ -368,6 +368,9 @@ static Bool FUNC_NAME(R100SetupTexture)(
     CARD8 *dst;
     CARD32 tex_size = 0, txformat;
     int dst_pitch, offset, size, i, tex_bytepp;
+#if X_BYTE_ORDER == X_BIG_ENDIAN && defined(ACCEL_CP)
+    unsigned char *RADEONMMIO = info->MMIO;
+#endif
     ACCEL_PREAMBLE();
 
     if ((width > 2048) || (height > 2048))
@@ -376,6 +379,31 @@ static Bool FUNC_NAME(R100SetupTexture)(
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+    /* Set up byte swapping for the framebuffer aperture as needed */
+    switch (tex_bytepp) {
+    case 1:
+	OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl &
+				    ~(RADEON_NONSURF_AP0_SWP_32BPP
+				    | RADEON_NONSURF_AP0_SWP_16BPP));
+	break;
+    case 2:
+	OUTREG(RADEON_SURFACE_CNTL, (info->ModeReg.surface_cntl &
+				     ~RADEON_NONSURF_AP0_SWP_32BPP)
+				   | RADEON_NONSURF_AP0_SWP_16BPP);
+	break;
+    case 4:
+	OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl
+				  | RADEON_NONSURF_AP0_SWP_32BPP
+				  | RADEON_NONSURF_AP0_SWP_16BPP);
+	break;
+    default:
+	xf86DrvMsg(pScrn->scrnIndex, X_ERROR, "%s: Don't know what to do for "
+		   "tex_bytepp == %d!\n", __func__, tex_bytepp);
+	return FALSE;
+    }
+#endif
+
     dst_pitch = (width * tex_bytepp + 31) & ~31;
     size = dst_pitch * height;
 
@@ -403,6 +431,11 @@ static Bool FUNC_NAME(R100SetupTexture)(
 	dst += dst_pitch;
     }
 
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+    /* restore byte swapping */
+    OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl);
+#endif
+
     BEGIN_ACCEL(5);
     OUT_ACCEL_REG(RADEON_PP_TXFORMAT_0, txformat);
     OUT_ACCEL_REG(RADEON_PP_TEX_SIZE_0, tex_size);
@@ -639,6 +672,9 @@ static Bool FUNC_NAME(R200SetupTexture)(
     CARD8 *dst;
     CARD32 tex_size = 0, txformat;
     int dst_pitch, offset, size, i, tex_bytepp;
+#if X_BYTE_ORDER == X_BIG_ENDIAN && defined(ACCEL_CP)
+    unsigned char *RADEONMMIO = info->MMIO;
+#endif
     ACCEL_PREAMBLE();
 
     if ((width > 2048) || (height > 2048))
@@ -647,6 +683,31 @@ static Bool FUNC_NAME(R200SetupTexture)(
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+    /* Set up byte swapping for the framebuffer aperture as needed */
+    switch (tex_bytepp) {
+    case 1:
+	OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl &
+				    ~(RADEON_NONSURF_AP0_SWP_32BPP
+				    | RADEON_NONSURF_AP0_SWP_16BPP));
+	break;
+    case 2:
+	OUTREG(RADEON_SURFACE_CNTL, (info->ModeReg.surface_cntl &
+				     ~RADEON_NONSURF_AP0_SWP_32BPP)
+				   | RADEON_NONSURF_AP0_SWP_16BPP);
+	break;
+    case 4:
+	OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl
+				  | RADEON_NONSURF_AP0_SWP_32BPP
+				  | RADEON_NONSURF_AP0_SWP_16BPP);
+	break;
+    default:
+	xf86DrvMsg(pScrn->scrnIndex, X_ERROR, "%s: Don't know what to do for "
+		   "tex_bytepp == %d!\n", __func__, tex_bytepp);
+	return FALSE;
+    }
+#endif
+
     dst_pitch = (width * tex_bytepp + 31) & ~31;
     size = dst_pitch * height;
 
@@ -674,6 +735,11 @@ static Bool FUNC_NAME(R200SetupTexture)(
 	dst += dst_pitch;
     }
 
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+    /* restore byte swapping */
+    OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl);
+#endif
+
     BEGIN_ACCEL(6);
     OUT_ACCEL_REG(R200_PP_TXFORMAT_0, txformat);
     OUT_ACCEL_REG(R200_PP_TXFORMAT_X_0, 0);
diff --git a/src/radeon_video.c b/src/radeon_video.c
index 761fd5b..548b608 100644
--- a/src/radeon_video.c
+++ b/src/radeon_video.c
@@ -1503,10 +1503,6 @@ RADEONPutImage(
    CARD32 tmp;
 #if X_BYTE_ORDER == X_BIG_ENDIAN
    unsigned char *RADEONMMIO = info->MMIO;
-   CARD32 surface_cntl = INREG(RADEON_SURFACE_CNTL);
-
-   OUTREG(RADEON_SURFACE_CNTL, (surface_cntl |
-		RADEON_NONSURF_AP0_SWP_32BPP) & ~RADEON_NONSURF_AP0_SWP_16BPP);
 #endif
 
    /*
@@ -1613,7 +1609,8 @@ RADEONPutImage(
 	}
 	nlines = ((((yb + 0xffff) >> 16) + 1) & ~1) - top;
 #if X_BYTE_ORDER == X_BIG_ENDIAN
-	OUTREG(RADEON_SURFACE_CNTL, (surface_cntl | RADEON_NONSURF_AP0_SWP_32BPP)
+	OUTREG(RADEON_SURFACE_CNTL, (info->ModeReg.surface_cntl |
+				     RADEON_NONSURF_AP0_SWP_32BPP)
 				    & ~RADEON_NONSURF_AP0_SWP_16BPP);
 #endif
 	RADEONCopyMungedData(buf + (top * srcPitch) + left, buf + s2offset,
@@ -1628,8 +1625,9 @@ RADEONPutImage(
 	nlines = ((yb + 0xffff) >> 16) - top;
 	dst_start += left;
 #if X_BYTE_ORDER == X_BIG_ENDIAN
-	OUTREG(RADEON_SURFACE_CNTL, surface_cntl & ~(RADEON_NONSURF_AP0_SWP_32BPP
-						   | RADEON_NONSURF_AP0_SWP_16BPP));
+	OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl &
+				    ~(RADEON_NONSURF_AP0_SWP_32BPP
+				      | RADEON_NONSURF_AP0_SWP_16BPP));
 #endif
 	RADEONCopyData(buf, dst_start, srcPitch, dstPitch, nlines, npixels);
 	break;
@@ -1637,7 +1635,7 @@ RADEONPutImage(
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
     /* restore byte swapping */
-    OUTREG(RADEON_SURFACE_CNTL, surface_cntl);
+    OUTREG(RADEON_SURFACE_CNTL, info->ModeReg.surface_cntl);
 #endif
 
     /* update cliplist */