- yincr = line_length;
- if (dy > sy) {
- dy += height - 1;
- sy += height - 1;
- yincr = -yincr;
- }
-
- /* Compute the offsets and alignments in the frame buffer.
- More than anything else, these control how we do copies. */
- dpos = dy * line_length + dx;
- spos = sy * line_length + sx;
- dalign = dpos & 7;
- salign = spos & 7;
- dpos &= -8;
- spos &= -8;
-
- /* Compute the value for the PIXELSHIFT register. This controls
- both non-co-aligned source and destination and copy direction. */
- if (dalign >= salign)
- pixel_shift = dalign - salign;
- else
- pixel_shift = 8 - (salign - dalign);
-
- /* Figure out if we need an additional priming step for the
- residue register. */
- need_prime = (salign > dalign);
- if (need_prime)
- dpos -= 8;
-
- /* Begin by copying the leading unaligned destination. Copy enough
- to make the next destination address 32-byte aligned. */
- copied = 32 - (dalign + (dpos & 31));
- if (copied == 32)
- copied = 0;
- xincr_first = (copied + 7) & -8;
- smask_first = dmask_first = (1ul << copied) - 1;
- smask_first <<= salign;
- dmask_first <<= dalign + need_prime*8;
- if (need_prime && copied > 24)
- copied -= 8;
- left = width - copied;
-
- /* Care for small copies. */
- if (copied > width) {
- u32 t;
- t = (1ul << width) - 1;
- t <<= dalign + need_prime*8;
- dmask_first &= t;
- left = 0;
- }
-
- /* Attempt to use 64-byte copies. This is only possible if the
- source and destination are co-aligned at 64 bytes. */
- n64 = need_second = 0;
- if ((dpos & 63) == (spos & 63)
- && (height == 1 || line_length % 64 == 0)) {
- /* We may need a 32-byte copy to ensure 64 byte alignment. */
- need_second = (dpos + xincr_first) & 63;
- if ((need_second & 32) != need_second)
- printk(KERN_ERR "tgafb: need_second wrong\n");
- if (left >= need_second + 64) {
- left -= need_second;
- n64 = left / 64;
- left %= 64;
- } else
- need_second = 0;
- }
-
- /* Copy trailing full 32-byte sections. This will be the main
- loop if the 64 byte loop can't be used. */
- n32 = left / 32;
- left %= 32;
-
- /* Copy the trailing unaligned destination. */
- dmask_last = (1ul << left) - 1;
-
- tga_regs = par->tga_regs_base;
- tga_fb = par->tga_fb_base;
-
- /* Set up the MODE and PIXELSHIFT registers. */
- __raw_writel(TGA_MODE_SBM_8BPP|TGA_MODE_COPY, tga_regs+TGA_MODE_REG);
- __raw_writel(pixel_shift, tga_regs+TGA_PIXELSHIFT_REG);
- wmb();
-
- for (i = 0; i < height; ++i) {
- unsigned long j;
- void __iomem *sfb;
- void __iomem *dfb;
-
- sfb = tga_fb + spos;
- dfb = tga_fb + dpos;
- if (dmask_first) {
- __raw_writel(smask_first, sfb);
- wmb();
- __raw_writel(dmask_first, dfb);
- wmb();
- sfb += xincr_first;
- dfb += xincr_first;
- }
-
- if (need_second) {
- __raw_writel(0xffffffff, sfb);
- wmb();
- __raw_writel(0xffffffff, dfb);
- wmb();
- sfb += 32;
- dfb += 32;
- }
-
- if (n64 && (((unsigned long)sfb | (unsigned long)dfb) & 63))
- printk(KERN_ERR
- "tgafb: misaligned copy64 (s:%p, d:%p)\n",
- sfb, dfb);
-
- for (j = 0; j < n64; ++j) {
- __raw_writel(sfb - tga_fb, tga_regs+TGA_COPY64_SRC);
- wmb();
- __raw_writel(dfb - tga_fb, tga_regs+TGA_COPY64_DST);
- wmb();
- sfb += 64;
- dfb += 64;
- }
-
- for (j = 0; j < n32; ++j) {
- __raw_writel(0xffffffff, sfb);
- wmb();
- __raw_writel(0xffffffff, dfb);
- wmb();
- sfb += 32;
- dfb += 32;
- }
-
- if (dmask_last) {
- __raw_writel(0xffffffff, sfb);
- wmb();
- __raw_writel(dmask_last, dfb);
- wmb();
- }
-
- spos += yincr;
- dpos += yincr;