aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/core
diff options
context:
space:
mode:
authorGravatar mtklein <mtklein@chromium.org>2015-09-14 12:43:20 -0700
committerGravatar Commit bot <commit-bot@chromium.org>2015-09-14 12:43:20 -0700
commit82c93b45ed6ac0b628adb8375389c202d1f586f9 (patch)
treebdb517b2f2a05fe22dda1f84f5f5eafda87efa87 /src/core
parentb5b603241aaa99e07dc4e12ca9f2661aa85e5f74 (diff)
SkPx: new approach to fixed-point SIMD
SkPx is like Sk4px, except each platform implementation of SkPx can declare a different sweet spot of N pixels, with extra loads and stores to handle the ragged edge of 0<n<N pixels. In this case, _sse's sweet spot remains 4 pixels. _neon jumps up to 8 so we can now use NEON's transposing loads and stores, and _none is just 1. This makes operations involving alpha considerably more efficient on NEON, as alpha is its own distinct 8x8 bit plane that's easy to toss around. This incorporates a few other improvements I've been wanting: - no requirement that we're dealing with SkPMColor. SkColor works too. - no anonymous namespace hack to differentiate implementations. Codegen and perf look good on Clang/x86-64 and GCC/ARMv7. The NEON code looks very similar to the old NEON code, as intended. No .skp or GM diffs on my laptop. Don't expect any. I intend this to replace Sk4px. Plan after landing: - port SkXfermode_opts.h - port Color32 in SkBlitRow_D32.cpp (and move to SkBlitRow_opts.h like other SkOpts code) - delete all Sk4px-related code - clean up evolutionary dead ends in SkNx (Sk16b, Sk16h, Sk4i, Sk4d, etc.) leaving Sk2f, Sk4f (and Sk2s, Sk4s). - find a machine with AVX2 to work on, write SkPx_avx2.h handling 8 pixels at a time. In the end we'll have Sk4f for float pixels, SkPx for fixed-point pixels. BUG=skia:4117 Review URL: https://codereview.chromium.org/1317233005
Diffstat (limited to 'src/core')
-rw-r--r--src/core/SkPx.h89
1 files changed, 89 insertions, 0 deletions
diff --git a/src/core/SkPx.h b/src/core/SkPx.h
new file mode 100644
index 0000000000..a4b1978e63
--- /dev/null
+++ b/src/core/SkPx.h
@@ -0,0 +1,89 @@
+/*
+ * Copyright 2015 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SkPx_DEFINED
+#define SkPx_DEFINED
+
+#include "SkTypes.h"
+#include "SkColorPriv.h"
+
+// We'll include one of src/opts/SkPx_{sse,neon,none}.h to define a type SkPx.
+//
+// SkPx represents up to SkPx::N 8888 pixels. It's agnostic to whether these
+// are SkColors or SkPMColors; it only assumes that alpha is the high byte.
+static_assert(SK_A32_SHIFT == 24, "For both SkColor and SkPMColor, alpha is always the high byte.");
+//
+// SkPx::Alpha represents up to SkPx::N 8-bit values, usually coverage or alpha.
+// SkPx::Wide represents up to SkPx::N pixels with 16 bits per component.
+//
+// SkPx supports the following methods:
+// static SkPx Dup(uint32_t);
+// static SkPx LoadN(const uint32_t*);
+// static SkPx Load(int n, const uint32_t*); // where 0<n<SkPx::N
+// void storeN(uint32_t*) const;
+// void store(int n, uint32_t*) const; // where 0<n<SkPx::N
+//
+// Alpha alpha() const; // argb -> a
+// Wide widenLo() const; // argb -> 0a0r0g0b
+// Wide widenHi() const; // argb -> a0r0g0b0
+// Wide widenLoHi() const; // argb -> aarrggbb
+//
+// SkPx operator+(const SkPx&) const;
+// SkPx operator-(const SkPx&) const;
+// SkPx saturatedAdd(const SkPx&) const;
+//
+// Wide operator*(const Alpha&) const; // argb * A -> (a*A)(r*A)(g*A)(b*A)
+//
+// // Fast approximate (px*a+127)/255.
+// // Never off by more than 1, and always correct when px or a is 0 or 255.
+// // We use the approximation (px*a+px)/256.
+// SkPx approxMulDiv255(const Alpha&) const;
+//
+// SkPx addAlpha(const Alpha&) const; // argb + A -> (a+A)rgb
+//
+// SkPx::Alpha supports the following methods:
+// static Alpha Dup(uint8_t);
+// static Alpha LoadN(const uint8_t*);
+// static Alpha Load(int n, const uint8_t*); // where 0<n<SkPx::N
+//
+// Alpha inv() const; // a -> 255-a
+//
+// SkPx::Wide supports the following methods:
+// Wide operator+(const Wide&);
+// Wide operator-(const Wide&);
+// Wide operator<<(int bits);
+// Wide operator>>(int bits);
+//
+// // Return the high byte of each component of (*this + o.widenLo()).
+// SkPx addNarrowHi(const SkPx& o);
+//
+// Methods left unwritten, but certainly to come:
+// SkPx SkPx::operator<(const SkPx&) const;
+// SkPx SkPx::thenElse(const SkPx& then, const SkPx& else) const;
+// Wide Wide::operator<(const Wide&) const;
+// Wide Wide::thenElse(const Wide& then, const Wide& else) const;
+//
+// SkPx Wide::div255() const; // Rounds, think (*this + 127) / 255.
+//
+// The different implementations of SkPx have complete freedom to choose
+// SkPx::N and how they represent SkPx, SkPx::Alpha, and SkPx::Wide.
+//
+// All observable math must remain identical.
+
+#if defined(SKNX_NO_SIMD)
+ #include "../opts/SkPx_none.h"
+#else
+ #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
+ #include "../opts/SkPx_sse.h"
+ #elif defined(SK_ARM_HAS_NEON)
+ #include "../opts/SkPx_neon.h"
+ #else
+ #include "../opts/SkPx_none.h"
+ #endif
+#endif
+
+#endif//SkPx_DEFINED