small speedup for scaler on sh-1, via use of hardware multiply instruction where possible

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@20322 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Andrew Mahone 2009-03-14 01:41:02 +00:00
parent d88a58bddd
commit f8877bf42d

View file

@ -60,6 +60,15 @@
#define DEBUGF(...)
#endif
#if CONFIG_CPU == SH7034
/* 16*16->32 bit multiplication is a single instrcution on the SH1 */
#define MULUQ(a, b) ((uint32_t) (((uint16_t) (a)) * ((uint16_t) (b))))
#define MULQ(a, b) ((int32_t) (((int16_t) (a)) * ((int16_t) (b))))
#else
#define MULUQ(a, b) ((a) * (b))
#define MULQ(a, b) ((a) * (b))
#endif
/* calculate the maximum dimensions which will preserve the aspect ration of
src while fitting in the constraints passed in dst, and store result in dst,
returning 0 if rounding and 1 if not rounding.
@ -208,12 +217,12 @@ static bool scale_h_area(void *out_line_ptr,
oxe -= ctx->src->width;
/* add saved partial pixel from start of area */
acc = acc * ctx->bm->width + tmp * mul;
acc = MULUQ(acc, ctx->bm->width) + MULUQ(tmp, mul);
/* get new pixel , then add its partial coverage to this area */
tmp = *(part->buf);
mul = ctx->bm->width - oxe;
acc += tmp * mul;
acc += MULUQ(tmp, mul);
/* round, divide, and either store or accumulate to output row */
if (accum)
{
@ -402,7 +411,7 @@ static bool scale_h_linear(void *out_line_ptr, struct scaler_context *ctx,
ixe -= (ctx->bm->width - 1);
val = *(part->buf);
inc = -val;
val *= (ctx->bm->width - 1);
val = MULUQ(val, ctx->bm->width - 1);
ix += 1;
/* If this wasn't the last pixel, add the next one to rgbinc. */
if (ix < (uint32_t)ctx->src->width) {
@ -414,10 +423,10 @@ static bool scale_h_linear(void *out_line_ptr, struct scaler_context *ctx,
/* Add a partial step to rgbval, in this pixel isn't precisely
aligned with the new source pixel
*/
val += inc * ixe;
val += MULQ(inc, ixe);
}
/* Now multiply the color increment to its proper value */
inc *= ctx->src->width - 1;
inc = MULQ(inc, ctx->src->width - 1);
} else
val += inc;
/* round and scale values, and accumulate or store to output */