diff options
| author |  mtklein <mtklein@chromium.org> | 2016-06-07 12:12:37 -0700 |
|---|---|---|
| committer |  Commit bot <commit-bot@chromium.org> | 2016-06-07 12:12:37 -0700 |
| commit | 3db2028126e12a9c1c1fb2f878d552a2de0e1452 (patch) | |
| tree | d851792e0ad76e61d51c42e03e429b14d2f21044 /src/core/SkXfermode4f.cpp | |
| parent | 12dfaaa53c23f3d03050bde8f64136ac1f44164a (diff) | |
linear -> sRGB: use fast approximate sqrt()
Since we're already approximating the sRGB gamma curve with a sqrt(), we might
as well approximate with it a faster approximate sqrt(). On Intel, this
.rsqrt().invert() version is 2-3x faster than .sqrt() (~3x faster on older
machines, ~2x faster on newer machines).
This should provide ~11 bits of precision, suspiciously exactly enough.
Running dm --config srgb, there are diffs, but none perceptible.
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2046063002
Review-Url: https://codereview.chromium.org/2046063002
Diffstat (limited to 'src/core/SkXfermode4f.cpp')
| -rw-r--r-- | src/core/SkXfermode4f.cpp | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/core/SkXfermode4f.cpp b/src/core/SkXfermode4f.cpp index 4d224ac0ad..8671df7b2f 100644 --- a/src/core/SkXfermode4f.cpp +++ b/src/core/SkXfermode4f.cpp @@ -68,10 +68,10 @@ static Sk4x4f load_4_srgb(const void* ptr) { // Store an Sk4x4f back to 4 interlaced 8888 sRGB pixels. static void store_4_srgb(void* ptr, const Sk4x4f& p) { // Convert back to sRGB and [0,255], again approximating sRGB as gamma == 2. - auto r = p.r.sqrt() * 255.0f + 0.5f, - g = p.g.sqrt() * 255.0f + 0.5f, - b = p.b.sqrt() * 255.0f + 0.5f, - a = p.a * 255.0f + 0.5f; + auto r = p.r.rsqrt().invert() * 255.0f + 0.5f, + g = p.g.rsqrt().invert() * 255.0f + 0.5f, + b = p.b.rsqrt().invert() * 255.0f + 0.5f, + a = p.a * 255.0f + 0.5f; Sk4x4f{r,g,b,a}.transpose((uint8_t*)ptr); } |