rop2.c

Go to the documentation of this file.

/*
 * SPDX-FileCopyrightText: 2025, Roy Ratcliffe, Northumberland, United Kingdom
 * SPDX-License-Identifier: MIT
 */
#include <blit/phase_align.h>
#include <blit/rop2.h>
 
#define S (fetch)
 
#define D (store)
 
typedef blit_scanline_t (*blit_rop2_func_t)(blit_scanline_t fetch, blit_scanline_t store);
 
#define ROP_REV_POLISH(revPolish, x)                                                   \
  static blit_scanline_t rop##revPolish(blit_scanline_t fetch, blit_scanline_t store); \
  blit_scanline_t rop##revPolish(blit_scanline_t fetch, blit_scanline_t store) { return x; }
 
ROP_REV_POLISH(0, 0x00U);
 
/*
 * Note that the macro ends with a semicolon. This is intentional to allow for
 * better formatting, even though semantically it is not required. Without the
 * trailing semicolon, the code would compile correctly, but automated
 * formatting tools will indent the subsequent code incorrectly.
 */
 
ROP_REV_POLISH(DSon, ~(D | S));
 
ROP_REV_POLISH(DSna, D & ~S);
 
ROP_REV_POLISH(Sn, ~S);
 
ROP_REV_POLISH(SDna, S & ~D);
 
ROP_REV_POLISH(Dn, ~D);
 
ROP_REV_POLISH(DSx, D ^ S);
 
ROP_REV_POLISH(DSan, ~(D &S));
 
ROP_REV_POLISH(DSa, D &S);
 
ROP_REV_POLISH(DSxn, ~(D ^ S));
 
ROP_REV_POLISH(D, D);
 
ROP_REV_POLISH(DSno, D | ~S);
 
ROP_REV_POLISH(S, S);
 
ROP_REV_POLISH(SDno, S | ~D);
 
ROP_REV_POLISH(DSo, D | S);
 
ROP_REV_POLISH(1, 0xffU);
 
static blit_rop2_func_t rop2_func[] = {&rop0,   &ropDSon, &ropDSna, &ropSn,   &ropSDna, &ropDn,   &ropDSx, &ropDSan,
                                       &ropDSa, &ropDSxn, &ropD,    &ropDSno, &ropS,    &ropSDno, &ropDSo, &rop1};
 
static void fetch_logic_mask_store(struct blit_phase_align *align, enum blit_rop2 rop2, blit_scanline_t mask, blit_scanline_t *store);
 
static void fetch_logic_store(struct blit_phase_align *align, enum blit_rop2 rop2, blit_scanline_t *store);
 
int blit_rgn1_rop2(struct blit_scan *result, struct blit_rgn1 *x, struct blit_rgn1 *y, const struct blit_scan *source, enum blit_rop2 rop2) {
  /*
   * Normalise, move, and clip the x region. The regions are first normalised to
   * ensure that their extents are non-negative. Then, they are moved to
   * ensure that their origins are non-negative. Finally, they are clipped to
   * ensure that they fit within the bounds of the destination and source scan
   * structures.
   */
  blit_rgn1_norm(x);
  if (!blit_rgn1_move(x) || !blit_rgn1_clip(x, result->width - x->origin) || !blit_rgn1_clip(x, source->width - x->origin_source))
    return 0;
 
  /*
   * Normalise, move, and clip the y region.
   */
  blit_rgn1_norm(y);
  if (!blit_rgn1_move(y) || !blit_rgn1_clip(y, result->height - y->origin) || !blit_rgn1_clip(y, source->height - y->origin_source))
    return 0;
 
  /*
   * Compute some important values up front to avoid doing it inside the bit
   * block transfer loops. The x_max constant represents the maximum x
   * coordinate of the region to be processed. The extra_scan_count constant
   * calculates how many additional bytes (beyond the first byte) are needed to
   * cover the width of the region in bytes. The scan_origin_mask and
   * scan_extent_mask constants are used to mask the bits at the start and end
   * of each scanline, ensuring that only the relevant bits are processed. The
   * offset and offset_source constants are used to calculate the stride offsets
   * for the destination and source scanlines, respectively, allowing for
   * efficient traversal of the scanline buffers.
   */
  const int x_max = x->origin + x->extent - 1;
  const int extra_scan_count = (x_max >> 3) - (x->origin >> 3);
  const blit_scanline_t scan_origin_mask = 0xffU >> (x->origin & 7);
  const blit_scanline_t scan_extent_mask = 0xffU << (7 - (x_max & 7));
  const int offset = result->stride - 1 - extra_scan_count;
  const int offset_source = source->stride - 1 - extra_scan_count;
  blit_scanline_t *store = blit_scan_find(result, x->origin, y->origin);
 
  /*
   * Set up phase alignment for source fetches. The source x position is given
   * by x->origin_source. The destination x position is given by x->origin. The
   * shift is the difference between these two positions modulo 8. The phase
   * alignment structure handles the bit shifts required to align the source
   * data with the destination data based on the bit positions. Each fetch from
   * the phase alignment structure returns a byte (8 bits) aligned according to
   * the specified bit positions.
   *
   * The & 7 operation here ensures that we are working within the bounds of a
   * byte (0-7 bits). This is important because the phase alignment also
   * corrects for source byte offsets, ensuring that the fetched data is
   * correctly aligned with respect to the destination, regardless of the
   * starting bit positions. Do not let the phase alignment and the scan find
   * get out of sync! Keep them in step!
   */
  struct blit_phase_align align;
  blit_phase_align_start(&align, x->origin, x->origin_source & 7, blit_scan_find(source, x->origin_source, y->origin_source));
 
  /*
   * Perform the bit block transfer using the specified raster operation. The
   * transfer is done scanline by scanline, processing each byte in the scanline
   * according to the raster operation defined by rop2. The function handles the
   * masking of the first and last bytes in each scanline to ensure that only
   * the relevant bits are modified. If there are no extra bytes beyond the
   * first byte, it processes the scanline in a single pass. If there are extra
   * bytes, it processes the first byte with the origin mask, then processes the
   * middle bytes without masking, and finally processes the last byte with the
   * extent mask.
   *
   * This approach ensures that the bit block transfer is efficient and
   * correctly applies the raster operation across the specified region. The use
   * of the phase alignment structure ensures that the source data is correctly
   * aligned with the destination data during the transfer.
   */
  int extent = y->extent, logic_count = 0;
  if (extra_scan_count == 0) {
    const blit_scanline_t scan_mask = scan_origin_mask & scan_extent_mask;
    while (extent--) {
      blit_phase_align_prefetch(&align);
      fetch_logic_mask_store(&align, rop2, scan_mask, store++);
      logic_count++;
      store += offset;
      align.store += offset_source;
    }
  } else {
    while (extent--) {
      blit_phase_align_prefetch(&align);
      fetch_logic_mask_store(&align, rop2, scan_origin_mask, store++);
      logic_count++;
      int extra = extra_scan_count;
      while (--extra) {
        fetch_logic_store(&align, rop2, store++);
        logic_count++;
      }
      fetch_logic_mask_store(&align, rop2, scan_extent_mask, store++);
      logic_count++;
      store += offset;
      align.store += offset_source;
    }
  }
  return logic_count;
}
 
int blit_rop2(struct blit_scan *result,
              /* destination region */
              const int x, const int y, const int x_extent, const int y_extent,
              /* source region */
              const struct blit_scan *source, const int x_source, const int y_source,
              /* raster operation */
              enum blit_rop2 rop2) {
  /*
   * Build the one-dimensional region structures for x and y axes from the
   * arguments. These structures define the origin and extent of the region to
   * be processed in both the result and source scan structures. Throw them
   * away after the raster operation because the caller does not want them.
   */
  struct blit_rgn1 x_rgn1 = {
      .origin = x,
      .extent = x_extent,
      .origin_source = x_source,
  };
  struct blit_rgn1 y_rgn1 = {
      .origin = y,
      .extent = y_extent,
      .origin_source = y_source,
  };
  return blit_rgn1_rop2(result, &x_rgn1, &y_rgn1, source, rop2);
}
 
void fetch_logic_mask_store(struct blit_phase_align *align, enum blit_rop2 rop2, blit_scanline_t mask, blit_scanline_t *store) {
  *store = (*store & ~mask) | (mask & rop2_func[rop2](blit_phase_align_fetch(align), *store));
}
 
void fetch_logic_store(struct blit_phase_align *align, enum blit_rop2 rop2, blit_scanline_t *store) {
  *store = rop2_func[rop2](blit_phase_align_fetch(align), *store);
}