diff options
author | mtklein <mtklein@google.com> | 2015-07-21 05:02:40 -0700 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-07-21 05:02:40 -0700 |
commit | e942bee8f9bc9b224aecc376083082c1ac33818d (patch) | |
tree | 96a8973b717f588aa4671440316a7aa1ed94b5f9 /src/core/Sk4px.h | |
parent | 134003902de13ade7795c5145a59b1e77770f2de (diff) |
Revert of 565 support for SIMD xfermodes (patchset #3 id:40001 of https://codereview.chromium.org/1245673002/)
Reason for revert:
942930d (included in this roll) introduced a 140 kB sizes regression in
libskia.so. Please investigate and reland if this regression is necessary.
Original issue's description:
> 565 support for SIMD xfermodes
>
> This uses the most basic approach possible:
> - to load an Sk4px from 565, convert to SkPMColors on the stack serially then load those SkPMColors.
> - to store an Sk4px to 565, store to SkPMColors on the stack then convert to 565 serially.
>
> Clearly, we can optimize these loads and stores. That's a TODO.
>
> The code using SkPMFloat is the same idea but a little more long-term viable, as we're only operating on one pixel at a time anyway. We could probably write 565 <-> SkPMFloat methods, but I'd rather not until it's really compelling.
>
> The speedups are varied but similar across SSE and NEON: a few uninteresting, many 50% faster, some 2x faster, and SoftLight ~4x faster.
>
> This will cause minor GM diffs, but I don't think any layout test changes.
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/942930dcaa51f66d82cdaf46ae62efebd16c8cd0
TBR=msarett@google.com,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/1242973004
Diffstat (limited to 'src/core/Sk4px.h')
-rw-r--r-- | src/core/Sk4px.h | 54 |
1 files changed, 6 insertions, 48 deletions
diff --git a/src/core/Sk4px.h b/src/core/Sk4px.h index 7d51fd099e..e1d4dc1244 100644 --- a/src/core/Sk4px.h +++ b/src/core/Sk4px.h @@ -10,7 +10,6 @@ #include "SkNx.h" #include "SkColor.h" -#include "SkColorPriv.h" // This file may be included multiple times by .cpp files with different flags, leading // to different definitions. Usually that doesn't matter because it's all inlined, but @@ -48,14 +47,6 @@ public: void store2(SkPMColor[2]) const; void store1(SkPMColor[1]) const; - // Same as above for 565. - static Sk4px Load4(const SkPMColor16 src[4]); - static Sk4px Load2(const SkPMColor16 src[2]); - static Sk4px Load1(const SkPMColor16 src[1]); - void store4(SkPMColor16 dst[4]) const; - void store2(SkPMColor16 dst[2]) const; - void store1(SkPMColor16 dst[1]) const; - // 1, 2, or 4 SkPMColors with 16-bit components. // This is most useful as the result of a multiply, e.g. from mulWiden(). class Wide : public Sk16h { @@ -108,8 +99,8 @@ public: // A generic driver that maps fn over a src array into a dst array. // fn should take an Sk4px (4 src pixels) and return an Sk4px (4 dst pixels). - template <typename Fn, typename Dst> - static void MapSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) { + template <typename Fn> + static void MapSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn& fn) { // This looks a bit odd, but it helps loop-invariant hoisting across different calls to fn. // Basically, we need to make sure we keep things inside a single loop. while (n > 0) { @@ -138,8 +129,8 @@ public: } // As above, but with dst4' = fn(dst4, src4). - template <typename Fn, typename Dst> - static void MapDstSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) { + template <typename Fn> + static void MapDstSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn& fn) { while (n > 0) { if (n >= 8) { Sk4px dst0 = fn(Load4(dst+0), Load4(src+0)), @@ -166,8 +157,8 @@ public: } // As above, but with dst4' = fn(dst4, src4, alpha4). - template <typename Fn, typename Dst> - static void MapDstSrcAlpha(int n, Dst* dst, const SkPMColor* src, const SkAlpha* a, + template <typename Fn> + static void MapDstSrcAlpha(int n, SkPMColor* dst, const SkPMColor* src, const SkAlpha* a, const Fn& fn) { while (n > 0) { if (n >= 8) { @@ -198,39 +189,6 @@ private: typedef Sk16b INHERITED; }; -// TODO: specialize these per-backend - -inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) { - SkPMColor src32[4]; - for (int i = 0; i < 4; i++) { src32[i] = SkPixel16ToPixel32(src[i]); } - return Load4(src32); -} -inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) { - SkPMColor src32[2]; - for (int i = 0; i < 2; i++) { src32[i] = SkPixel16ToPixel32(src[i]); } - return Load2(src32); -} -inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) { - SkPMColor src32 = SkPixel16ToPixel32(src[0]); - return Load1(&src32); -} - -inline void Sk4px::store4(SkPMColor16 dst[4]) const { - SkPMColor dst32[4]; - this->store4(dst32); - for (int i = 0; i < 4; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); } -} -inline void Sk4px::store2(SkPMColor16 dst[2]) const { - SkPMColor dst32[2]; - this->store2(dst32); - for (int i = 0; i < 2; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); } -} -inline void Sk4px::store1(SkPMColor16 dst[1]) const { - SkPMColor dst32; - this->store1(&dst32); - dst[0] = SkPixel32ToPixel16(dst32); -} - } // namespace #ifdef SKNX_NO_SIMD |