diff options
author | 2015-02-02 09:52:43 -0800 | |
---|---|---|
committer | 2015-02-02 09:52:43 -0800 | |
commit | 4988891a1173cd405bf1c1dd3a3668c451f45e4c (patch) | |
tree | 0fd4b535b7e9b57fb0b6248f69b04339bc5c7b69 /src/opts/SkBlitRow_opts_SSE4.cpp | |
parent | db204e301b1320af242e1be5e477cd9453b126a6 (diff) |
Revert of SSE4 opaque blend using intrinsics instead of assembly. (patchset #16 id:300001 of https://codereview.chromium.org/874863002/)
Reason for revert:
This causes a bug on the 'hittestpath' GM on MacMini 4,1
See:
https://gold.skia.org/#/triage/hittestpath?head=0
for details.
Original issue's description:
> SSE4 opaque blend using intrinsics instead of assembly.
>
> Since we had such a hard time with the assembly versions of this blit (to the
> point that we have them completely disabled everywhere), I thought I'd take
> a shot at writing a version of the blit using intrinsics.
>
> The key feature of SSE4 we're exploiting is that we can use ptest (_mm_test*)
> to skip the blend when the 16 src pixels we consider each loop are all opaque
> or all transparent. _mm_shuffle_epi8 from SSSE3 also lends a hand to extract
> all those alphas.
>
> It's worth looking to see if we can backport this type of logic to SSE2 using
> _mm_movemask_epi8, or up to 32 pixels at a time using AVX.
>
> My local performance testing doesn't show this to be an unambiguous win
> (there are probably microbenchmarks and SKPs where we'd be better off just
> powering through the blend rather than looking at alphas), but the potential
> does seem tantalizing enough to let skiaperf vet it on the bots. (< 1.0x is a win.)
>
> DM says it draws pixel perfect compare to the old code.
>
> Microbenchmarks:
> bitmap_RGBA_8888_A_source_stripes_two 14us -> 14.4us 1.03x
> bitmap_RGBA_8888_A_source_stripes_three 14.3us -> 14.5us 1.01x
> bitmap_RGBA_8888_scale_bilerp 61.9us -> 62.2us 1.01x
> bitmap_RGBA_8888_update_volatile_scale_rotate_bilerp 102us -> 101us 0.99x
> bitmap_RGBA_8888_scale_rotate_bilerp 103us -> 101us 0.99x
> bitmap_RGBA_8888_scale 18.4us -> 18.2us 0.99x
> bitmap_RGBA_8888_A_scale_rotate_bicubic 71us -> 70us 0.99x
> bitmap_RGBA_8888_update_scale_rotate_bilerp 103us -> 101us 0.99x
> bitmap_RGBA_8888_A_scale_rotate_bilerp 112us -> 109us 0.98x
> bitmap_RGBA_8888_update_volatile 5.72us -> 5.58us 0.98x
> bitmap_RGBA_8888 5.73us -> 5.58us 0.97x
> bitmap_RGBA_8888_update 5.78us -> 5.6us 0.97x
> bitmap_RGBA_8888_A_scale_bilerp 70.7us -> 68us 0.96x
> bitmap_RGBA_8888_A_scale_bicubic 23.7us -> 21.8us 0.92x
> bitmap_RGBA_8888_A 13.9us -> 10.9us 0.78x
> bitmap_RGBA_8888_A_source_opaque 14us -> 6.29us 0.45x
> bitmap_RGBA_8888_A_source_transparent 14us -> 3.65us 0.26x
>
> Running over our ~70 SKP web page captures, this looks like we spend 0.7x
> the time in S32A_Opaque_BlitRow compared to the SSE2 version, which should
> be a decent predictor of real-world impact.
>
> BUG=chromium:399842
>
> Committed: https://skia.googlesource.com/skia/+/04bc91b972417038fecfa87c484771eac2b9b785
>
> CQ_EXTRA_TRYBOTS=client.skia:Test-Mac10.6-MacMini4.1-GeForce320M-x86_64-Release-Trybot
>
> Committed: https://skia.googlesource.com/skia/+/6dbfb21a6c88af6d94e8c823c3ad559f1a41b493
TBR=henrik.smiding@intel.com,mtklein@google.com,herb@google.com,reed@google.com,thakis@chromium.org,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:399842
Review URL: https://codereview.chromium.org/873553003
Diffstat (limited to 'src/opts/SkBlitRow_opts_SSE4.cpp')
-rw-r--r-- | src/opts/SkBlitRow_opts_SSE4.cpp | 67 |
1 files changed, 0 insertions, 67 deletions
diff --git a/src/opts/SkBlitRow_opts_SSE4.cpp b/src/opts/SkBlitRow_opts_SSE4.cpp deleted file mode 100644 index ae92a77eb2..0000000000 --- a/src/opts/SkBlitRow_opts_SSE4.cpp +++ /dev/null @@ -1,67 +0,0 @@ -#include "SkBlitRow_opts_SSE4.h" - -// Some compilers can't compile SSSE3 or SSE4 intrinsics. We give them stub methods. -// The stubs should never be called, so we make them crash just to confirm that. -#if SK_CPU_SSE_LEVEL < SK_CPU_SSE_LEVEL_SSE41 -void S32A_Opaque_BlitRow32_SSE4(SkPMColor* SK_RESTRICT, const SkPMColor* SK_RESTRICT, int, U8CPU) { - sk_throw(); -} - -#else - -#include <emmintrin.h> // SSE2: Most _mm_foo() in this file. -#include <smmintrin.h> // SSE4.1: _mm_testz_si128 and _mm_testc_si128. - -#include "SkColorPriv.h" -#include "SkColor_opts_SSE2.h" - -void S32A_Opaque_BlitRow32_SSE4(SkPMColor* SK_RESTRICT dst, - const SkPMColor* SK_RESTRICT src, - int count, - U8CPU alpha) { - SkASSERT(alpha == 255); - // As long as we can, we'll work on 16 pixel pairs at once. - int count16 = count / 16; - __m128i* dst4 = (__m128i*)dst; - const __m128i* src4 = (const __m128i*)src; - - for (int i = 0; i < count16 * 4; i += 4) { - // Load 16 source pixels. - __m128i s0 = _mm_loadu_si128(src4+i+0), - s1 = _mm_loadu_si128(src4+i+1), - s2 = _mm_loadu_si128(src4+i+2), - s3 = _mm_loadu_si128(src4+i+3); - - const __m128i alphaMask = _mm_set1_epi32(0xFF << SK_A32_SHIFT); - const __m128i ORed = _mm_or_si128(s3, _mm_or_si128(s2, _mm_or_si128(s1, s0))); - if (_mm_testz_si128(ORed, alphaMask)) { - // All 16 source pixels are fully transparent. There's nothing to do! - continue; - } - const __m128i ANDed = _mm_and_si128(s3, _mm_and_si128(s2, _mm_and_si128(s1, s0))); - if (_mm_testc_si128(ANDed, alphaMask)) { - // All 16 source pixels are fully opaque. There's no need to read dst or blend it. - _mm_storeu_si128(dst4+i+0, s0); - _mm_storeu_si128(dst4+i+1, s1); - _mm_storeu_si128(dst4+i+2, s2); - _mm_storeu_si128(dst4+i+3, s3); - continue; - } - // The general slow case: do the blend for all 16 pixels. - _mm_storeu_si128(dst4+i+0, SkPMSrcOver_SSE2(s0, _mm_loadu_si128(dst4+i+0))); - _mm_storeu_si128(dst4+i+1, SkPMSrcOver_SSE2(s1, _mm_loadu_si128(dst4+i+1))); - _mm_storeu_si128(dst4+i+2, SkPMSrcOver_SSE2(s2, _mm_loadu_si128(dst4+i+2))); - _mm_storeu_si128(dst4+i+3, SkPMSrcOver_SSE2(s3, _mm_loadu_si128(dst4+i+3))); - } - - // Wrap up the last <= 15 pixels. - SkASSERT(count - (count16*16) <= 15); - for (int i = count16*16; i < count; i++) { - // This check is not really necessarily, but it prevents pointless autovectorization. - if (src[i] & 0xFF000000) { - dst[i] = SkPMSrcOver(src[i], dst[i]); - } - } -} - -#endif |