// Scanline control byte (SCB) accessors.
//
// Each scanline holds one uint8_t SCB value in range 0..15 selecting
// which of the 16 palettes that scanline uses at display time.

#include <stddef.h>
#include <string.h>

#include "joey/palette.h"
#include "surfaceInternal.h"

// ----- Public API (alphabetical) -----

uint8_t scbGet(const SurfaceT *s, uint16_t line) {
    if (s == NULL || line >= SURFACE_HEIGHT) {
        return 0;
    }
    return s->scb[line];
}


void scbSet(SurfaceT *s, uint16_t line, uint8_t paletteIndex) {
    if (s == NULL || line >= SURFACE_HEIGHT) {
        return;
    }
    if (paletteIndex >= SURFACE_PALETTE_COUNT) {
        return;
    }
    s->scb[line] = paletteIndex;
    if (s == stageGet()) {
        gStageScbDirty = true;
    }
}


void scbSetRange(SurfaceT *s, uint16_t firstLine, uint16_t lastLine, uint8_t paletteIndex) {
    uint16_t line;
    uint16_t last;

    if (s == NULL) {
        return;
    }
    if (paletteIndex >= SURFACE_PALETTE_COUNT) {
        return;
    }
    if (firstLine >= SURFACE_HEIGHT) {
        return;
    }

    last = lastLine;
    if (last >= SURFACE_HEIGHT) {
        last = SURFACE_HEIGHT - 1;
    }
    if (last < firstLine) {
        return;
    }

    // memset is far cheaper than the per-iter loop on ORCA-C with -b
    // (scb is uint8_t, sizeof(uint8_t)==1, so the call form below is
    // exact). On IIgs ORCA-C lowers small fixed-size memsets to MVP /
    // PEI tricks; on Amiga/ST/DOS it uses libc memset which is
    // already vectorized. Either way, much tighter than the C loop.
    (void)line;
    memset(&s->scb[firstLine], paletteIndex, (size_t)(last - firstLine + 1));
    if (s == stageGet()) {
        gStageScbDirty = true;
    }
}