tridentfb: add imageblit acceleration for Blade3D family

Add imageblit acceleration for the Blade3D family of cores.  The code is
based on code from the cyblafb driver.

It is a step toward assimilating back the cyblafb driver into the
tridentfb driver.  The cyblafb driver handles a subfamily of the Trident
Blade3d cores.

Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2 files changed, 77 insertions, 3 deletions

diff --git a/drivers/video/tridentfb.c b/drivers/video/tridentfb.c
index 854e2e5af582..b21f84239022 100644
--- a/drivers/video/tridentfb.c
+++ b/drivers/video/tridentfb.c
@@ -35,11 +35,12 @@ struct tridentfb_par {
                 (struct tridentfb_par *par, u32, u32, u32, u32, u32, u32);
         void (*copy_rect)
                 (struct tridentfb_par *par, u32, u32, u32, u32, u32, u32);
+        void (*image_blit)
+                (struct tridentfb_par *par, const char*,
+                 u32, u32, u32, u32, u32, u32);
         unsigned char eng_oper; /* engine operation... */
 };
 
-static struct fb_ops tridentfb_ops;
-
 static struct fb_fix_screeninfo tridentfb_fix = {
         .id = "Trident",
         .type = FB_TYPE_PACKED_PIXELS,
@@ -212,6 +213,21 @@ static void blade_fill_rect(struct tridentfb_par *par,
         writemmr(par, DST2, point(x + w - 1, y + h - 1));
 }
 
+static void blade_image_blit(struct tridentfb_par *par, const char *data,
+                             u32 x, u32 y, u32 w, u32 h, u32 c, u32 b)
+{
+        unsigned size = ((w + 31) >> 5) * h;
+
+        writemmr(par, COLOR, c);
+        writemmr(par, BGCOLOR, b);
+        writemmr(par, CMD, 0xa0000000 | 3 << 19);
+
+        writemmr(par, DST1, point(x, y));
+        writemmr(par, DST2, point(x + w - 1, y + h - 1));
+
+        memcpy(par->io_virt + 0x10000, data, 4 * size);
+}
+
 static void blade_copy_rect(struct tridentfb_par *par,
                             u32 x1, u32 y1, u32 x2, u32 y2, u32 w, u32 h)
 {
@@ -497,6 +513,36 @@ static void tridentfb_fillrect(struct fb_info *info,
                        fr->height, col, fr->rop);
 }
 
+static void tridentfb_imageblit(struct fb_info *info,
+                                const struct fb_image *img)
+{
+        struct tridentfb_par *par = info->par;
+        int col, bgcol;
+
+        if ((info->flags & FBINFO_HWACCEL_DISABLED) || img->depth != 1) {
+                cfb_imageblit(info, img);
+                return;
+        }
+        if (info->var.bits_per_pixel == 8) {
+                col = img->fg_color;
+                col |= col << 8;
+                col |= col << 16;
+                bgcol = img->bg_color;
+                bgcol |= bgcol << 8;
+                bgcol |= bgcol << 16;
+        } else {
+                col = ((u32 *)(info->pseudo_palette))[img->fg_color];
+                bgcol = ((u32 *)(info->pseudo_palette))[img->bg_color];
+        }
+
+        par->wait_engine(par);
+        if (par->image_blit)
+                par->image_blit(par, img->data, img->dx, img->dy,
+                                img->width, img->height, col, bgcol);
+        else
+                cfb_imageblit(info, img);
+}
+
 static void tridentfb_copyarea(struct fb_info *info,
                                const struct fb_copyarea *ca)
 {
@@ -522,6 +568,7 @@ static int tridentfb_sync(struct fb_info *info)
 #else
 #define tridentfb_fillrect cfb_fillrect
 #define tridentfb_copyarea cfb_copyarea
+#define tridentfb_imageblit cfb_imageblit
 #endif /* CONFIG_FB_TRIDENT_ACCEL */
 
 /*
@@ -1285,7 +1332,7 @@ static struct fb_ops tridentfb_ops = {
         .fb_set_par = tridentfb_set_par,
         .fb_fillrect = tridentfb_fillrect,
         .fb_copyarea = tridentfb_copyarea,
-        .fb_imageblit = cfb_imageblit,
+        .fb_imageblit = tridentfb_imageblit,
 #ifdef CONFIG_FB_TRIDENT_ACCEL
         .fb_sync = tridentfb_sync,
 #endif
@@ -1369,6 +1416,7 @@ static int __devinit trident_pci_probe(struct pci_dev *dev,
                 default_par->wait_engine = blade_wait_engine;
                 default_par->fill_rect = blade_fill_rect;
                 default_par->copy_rect = blade_copy_rect;
+                default_par->image_blit = blade_image_blit;
                 tridentfb_fix.accel = FB_ACCEL_TRIDENT_BLADE3D;
         } else if (chip3D) {                    /* 3DImage family left */
                 default_par->init_accel = image_init_accel;
@@ -1446,6 +1494,29 @@ static int __devinit trident_pci_probe(struct pci_dev *dev,
         } else
                 info->flags |= FBINFO_HWACCEL_DISABLED;
 
+        info->pixmap.addr = kmalloc(4096, GFP_KERNEL);
+        if (!info->pixmap.addr) {
+                err = -ENOMEM;
+                goto out_unmap2;
+        }
+
+        info->pixmap.size = 4096;
+        info->pixmap.buf_align = 4;
+        info->pixmap.scan_align = 1;
+        info->pixmap.access_align = 32;
+        info->pixmap.flags = FB_PIXMAP_SYSTEM;
+
+        if (default_par->image_blit) {
+                info->flags |= FBINFO_HWACCEL_IMAGEBLIT;
+                info->pixmap.scan_align = 4;
+        }
+
+        if (noaccel) {
+                printk(KERN_DEBUG "disabling acceleration\n");
+                info->flags |= FBINFO_HWACCEL_DISABLED;
+                info->pixmap.scan_align = 1;
+        }
+
         if (!fb_find_mode(&info->var, info,
                           mode_option, NULL, 0, NULL, bpp)) {
                 err = -EINVAL;
@@ -1471,6 +1542,7 @@ static int __devinit trident_pci_probe(struct pci_dev *dev,
         return 0;
 
 out_unmap2:
+        kfree(info->pixmap.addr);
         if (info->screen_base)
                 iounmap(info->screen_base);
         release_mem_region(tridentfb_fix.smem_start, tridentfb_fix.smem_len);
@@ -1494,6 +1566,7 @@ static void __devexit trident_pci_remove(struct pci_dev *dev)
         release_mem_region(tridentfb_fix.smem_start, tridentfb_fix.smem_len);
         release_mem_region(tridentfb_fix.mmio_start, tridentfb_fix.mmio_len);
         pci_set_drvdata(dev, NULL);
+        kfree(info->pixmap.addr);
         framebuffer_release(info);
 }
 
diff --git a/include/video/trident.h b/include/video/trident.h
index 85ced9d91312..b6ce19d1b61b 100644
--- a/include/video/trident.h
+++ b/include/video/trident.h
@@ -135,6 +135,7 @@
 #define CMD     0x2144
 #define ROP     0x2148
 #define COLOR   0x2160
+#define BGCOLOR 0x2164
 #define SRC1    0x2100
 #define SRC2    0x2104
 #define DST1    0x2108