Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(703)

Side by Side Diff: src/opts/SkPMFloat_SSSE3.h

Issue 982123002: SKPMFloat: we can beat the naive loops when clamping (Closed) Base URL: https://skia.googlesource.com/skia.git@master
Patch Set: restore some asserts Created 5 years, 9 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
« no previous file with comments | « src/opts/SkPMFloat_SSE2.h ('k') | src/opts/SkPMFloat_neon.h » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 #include "SkColorPriv.h" 1 #include "SkColorPriv.h"
2 #include <tmmintrin.h> 2 #include <tmmintrin.h>
3 3
4 // For SkPMFloat(SkPMColor), we widen our 8 bit components (fix8) to 8-bit compo nents in 32 bits 4 // For SkPMFloat(SkPMColor), we widen our 8 bit components (fix8) to 8-bit compo nents in 32 bits
5 // (fix8_32), then convert those to floats. 5 // (fix8_32), then convert those to floats.
6 6
7 // get() does the opposite, working from floats to 8-bit-in-32-bits, then back t o packed 8 bit. 7 // get() does the opposite, working from floats to 8-bit-in-32-bits, then back t o packed 8 bit.
8 8
9 // clamped() is the same as _SSE2: floats to 8-in-32, to 8-in-16, to packed 8 bi t, with 9 // clamped() is the same as _SSE2: floats to 8-in-32, to 8-in-16, to packed 8 bi t, with
10 // _mm_packus_epi16() both clamping and narrowing. 10 // _mm_packus_epi16() both clamping and narrowing.
(...skipping 18 matching lines...) Expand all
29 } 29 }
30 30
31 inline SkPMColor SkPMFloat::clamped() const { 31 inline SkPMColor SkPMFloat::clamped() const {
32 __m128i fix8_32 = _mm_cvtps_epi32(_mm_load_ps(fColor)), // _mm_cvtps_epi32 rounds for us! 32 __m128i fix8_32 = _mm_cvtps_epi32(_mm_load_ps(fColor)), // _mm_cvtps_epi32 rounds for us!
33 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), 33 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
34 fix8 = _mm_packus_epi16(fix8_16, fix8_16); 34 fix8 = _mm_packus_epi16(fix8_16, fix8_16);
35 SkPMColor c = _mm_cvtsi128_si32(fix8); 35 SkPMColor c = _mm_cvtsi128_si32(fix8);
36 SkPMColorAssert(c); 36 SkPMColorAssert(c);
37 return c; 37 return c;
38 } 38 }
39
40 inline void SkPMFloat::From4PMColors(SkPMFloat floats[4], const SkPMColor colors [4]) {
41 // Haven't beaten this yet.
42 for (int i = 0; i < 4; i++) { floats[i] = FromPMColor(colors[i]); }
43 }
44
45 inline void SkPMFloat::To4PMColors(SkPMColor colors[4], const SkPMFloat floats[4 ]) {
46 // Haven't beaten this yet. Still faster than ClampTo4PMColors too.
47 for (int i = 0; i < 4; i++) { colors[i] = floats[i].get(); }
48 }
49
50 inline void SkPMFloat::ClampTo4PMColors(SkPMColor colors[4], const SkPMFloat flo ats[4]) {
51 // Same as _SSE2.h's. We use 3 _mm_packus_epi16() where the naive loop uses 8.
52 __m128i c0 = _mm_cvtps_epi32(_mm_load_ps(floats[0].fColor)), // _mm_cvtps_e pi32 rounds for us!
53 c1 = _mm_cvtps_epi32(_mm_load_ps(floats[1].fColor)),
54 c2 = _mm_cvtps_epi32(_mm_load_ps(floats[2].fColor)),
55 c3 = _mm_cvtps_epi32(_mm_load_ps(floats[3].fColor));
56 __m128i c3210 = _mm_packus_epi16(_mm_packus_epi16(c0, c1),
57 _mm_packus_epi16(c2, c3));
58 _mm_storeu_si128((__m128i*)colors, c3210);
59 SkPMColorAssert(colors[0]);
60 SkPMColorAssert(colors[1]);
61 SkPMColorAssert(colors[2]);
62 SkPMColorAssert(colors[3]);
63 }
OLDNEW
« no previous file with comments | « src/opts/SkPMFloat_SSE2.h ('k') | src/opts/SkPMFloat_neon.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698