aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/opts/SkRasterPipeline_opts.h
diff options
context:
space:
mode:
authorGravatar Mike Klein <mtklein@chromium.org>2016-10-19 21:05:17 -0400
committerGravatar Skia Commit-Bot <skia-commit-bot@chromium.org>2016-10-20 16:59:23 +0000
commit2878e76247fd335dee755cfd4c9cd026804f9655 (patch)
treef99ff1d17f73373b503a0e4fce1b41dc54b44d9b /src/opts/SkRasterPipeline_opts.h
parentd2fe3bce079671b0210d5c7cddcc712988d0b686 (diff)
SkRasterPipeline refactor
- Give body and tail functions separate types. This frees a register in body functions, especially important for Windows. - Fill out default, SSE4.1, and HSW versions of all functions. This means we don't have to mess around with SkNf_abi... all functions come from the same compilation unit where SkNf is a single consistent type. - Move Stage::next() into SkRasterPipeline_opts.h as a static inline function. - Remove Stage::ctx() entirely... fCtx is literally the same thing. This is a step along the way toward building the entire pipeline in src/opts, removing the need for all the stages to be functions living in SkOpts. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=3680 CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot,Test-Ubuntu-Clang-GCE-CPU-AVX2-x86_64-Debug-ASAN-Trybot Change-Id: I7de78ffebc15b9bad4eda187c9f50369cd7e5e42 Reviewed-on: https://skia-review.googlesource.com/3680 Reviewed-by: Herb Derby <herb@google.com> Commit-Queue: Mike Klein <mtklein@chromium.org>
Diffstat (limited to 'src/opts/SkRasterPipeline_opts.h')
-rw-r--r--src/opts/SkRasterPipeline_opts.h87
1 files changed, 60 insertions, 27 deletions
diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h
index 3266554ae4..7e884c7cd3 100644
--- a/src/opts/SkRasterPipeline_opts.h
+++ b/src/opts/SkRasterPipeline_opts.h
@@ -12,36 +12,53 @@
#include "SkPM4f.h"
#include "SkRasterPipeline.h"
#include "SkSRGB.h"
+#include <utility>
+
+#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_AVX2
+ static constexpr int N = 8;
+#else
+ static constexpr int N = 4;
+#endif
-using SkNf_abi = SkRasterPipeline::V;
-static constexpr auto N = sizeof(SkNf_abi) / sizeof(float);
using SkNf = SkNx<N, float>;
using SkNi = SkNx<N, int>;
using SkNh = SkNx<N, uint16_t>;
+using Body = void(SK_VECTORCALL *)(SkRasterPipeline::Stage*, size_t,
+ SkNf,SkNf,SkNf,SkNf,
+ SkNf,SkNf,SkNf,SkNf);
+using Tail = void(SK_VECTORCALL *)(SkRasterPipeline::Stage*, size_t, size_t,
+ SkNf,SkNf,SkNf,SkNf,
+ SkNf,SkNf,SkNf,SkNf);
+
#define SI static inline
+template <typename Fn, typename... Args>
+SI void next(SkRasterPipeline::Stage* st, Args&&... args) {
+ // Stages are logically a pipeline, and physically are contiguous in an array.
+ // To get to the next stage, we just increment our pointer to the next array element.
+ ((Fn)st->fNext)(st+1, std::forward<Args>(args)...);
+}
+
#define STAGE(name, kCallNext) \
template <bool kIsTail> \
static SK_ALWAYS_INLINE void name##_kernel(void* ctx, size_t x, size_t tail, \
SkNf& r, SkNf& g, SkNf& b, SkNf& a, \
SkNf& dr, SkNf& dg, SkNf& db, SkNf& da); \
- SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
- SkNf_abi R, SkNf_abi G, SkNf_abi B, SkNf_abi A, \
- SkNf_abi DR, SkNf_abi DG, SkNf_abi DB, SkNf_abi DA) { \
- SkNf r=R,g=G,b=B,a=A, dr=DR,dg=DG,db=DB,da=DA; \
- name##_kernel<false>(st->ctx<void*>(), x,0, r,g,b,a, dr,dg,db,da); \
+ SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
+ name##_kernel<false>(st->fCtx, x,0, r,g,b,a, dr,dg,db,da); \
if (kCallNext) { \
- st->next(x,tail, r,g,b,a, dr,dg,db,da); \
+ next<Body>(st, x, r,g,b,a, dr,dg,db,da); \
} \
} \
SI void SK_VECTORCALL name##_tail(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
- SkNf_abi R, SkNf_abi G, SkNf_abi B, SkNf_abi A, \
- SkNf_abi DR, SkNf_abi DG, SkNf_abi DB, SkNf_abi DA) { \
- SkNf r=R,g=G,b=B,a=A, dr=DR,dg=DG,db=DB,da=DA; \
- name##_kernel<true>(st->ctx<void*>(), x,tail, r,g,b,a, dr,dg,db,da); \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
+ name##_kernel<true>(st->fCtx, x,tail, r,g,b,a, dr,dg,db,da); \
if (kCallNext) { \
- st->next(x,tail, r,g,b,a, dr,dg,db,da); \
+ next<Tail>(st, x,tail, r,g,b,a, dr,dg,db,da); \
} \
} \
template <bool kIsTail> \
@@ -54,15 +71,23 @@ using SkNh = SkNx<N, uint16_t>;
#define RGBA_XFERMODE(name) \
static SK_ALWAYS_INLINE SkNf name##_kernel(const SkNf& s, const SkNf& sa, \
const SkNf& d, const SkNf& da); \
- SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
- SkNf_abi R, SkNf_abi G, SkNf_abi B, SkNf_abi A, \
- SkNf_abi DR, SkNf_abi DG, SkNf_abi DB, SkNf_abi DA) { \
- SkNf r=R,g=G,b=B,a=A, dr=DR,dg=DG,db=DB,da=DA; \
+ SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
+ r = name##_kernel(r,a,dr,da); \
+ g = name##_kernel(g,a,dg,da); \
+ b = name##_kernel(b,a,db,da); \
+ a = name##_kernel(a,a,da,da); \
+ next<Body>(st, x, r,g,b,a, dr,dg,db,da); \
+ } \
+ SI void SK_VECTORCALL name##_tail(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
r = name##_kernel(r,a,dr,da); \
g = name##_kernel(g,a,dg,da); \
b = name##_kernel(b,a,db,da); \
a = name##_kernel(a,a,da,da); \
- st->next(x,tail, r,g,b,a, dr,dg,db,da); \
+ next<Tail>(st, x,tail, r,g,b,a, dr,dg,db,da); \
} \
static SK_ALWAYS_INLINE SkNf name##_kernel(const SkNf& s, const SkNf& sa, \
const SkNf& d, const SkNf& da)
@@ -71,15 +96,23 @@ using SkNh = SkNx<N, uint16_t>;
#define RGB_XFERMODE(name) \
static SK_ALWAYS_INLINE SkNf name##_kernel(const SkNf& s, const SkNf& sa, \
const SkNf& d, const SkNf& da); \
- SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
- SkNf_abi R, SkNf_abi G, SkNf_abi B, SkNf_abi A, \
- SkNf_abi DR, SkNf_abi DG, SkNf_abi DB, SkNf_abi DA) { \
- SkNf r=R,g=G,b=B,a=A, dr=DR,dg=DG,db=DB,da=DA; \
+ SI void SK_VECTORCALL name(SkRasterPipeline::Stage* st, size_t x, \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
+ r = name##_kernel(r,a,dr,da); \
+ g = name##_kernel(g,a,dg,da); \
+ b = name##_kernel(b,a,db,da); \
+ a = a + (da * (1.0f-a)); \
+ next<Body>(st, x, r,g,b,a, dr,dg,db,da); \
+ } \
+ SI void SK_VECTORCALL name##_tail(SkRasterPipeline::Stage* st, size_t x, size_t tail, \
+ SkNf r, SkNf g, SkNf b, SkNf a, \
+ SkNf dr, SkNf dg, SkNf db, SkNf da) { \
r = name##_kernel(r,a,dr,da); \
g = name##_kernel(g,a,dg,da); \
b = name##_kernel(b,a,db,da); \
a = a + (da * (1.0f-a)); \
- st->next(x,tail, r,g,b,a, dr,dg,db,da); \
+ next<Tail>(st, x,tail, r,g,b,a, dr,dg,db,da); \
} \
static SK_ALWAYS_INLINE SkNf name##_kernel(const SkNf& s, const SkNf& sa, \
const SkNf& d, const SkNf& da)
@@ -90,11 +123,11 @@ namespace SK_OPTS_NS {
SI void run_pipeline(size_t x, size_t n,
void (*vBodyStart)(), SkRasterPipeline::Stage* body,
void (*vTailStart)(), SkRasterPipeline::Stage* tail) {
- auto bodyStart = (SkRasterPipeline::Fn)vBodyStart,
- tailStart = (SkRasterPipeline::Fn)vTailStart;
- SkNf v{0}; // TODO: uninitialized would be a bit faster, but some compilers are whiny.
+ auto bodyStart = (Body)vBodyStart;
+ auto tailStart = (Tail)vTailStart;
+ SkNf v; // Fastest to start uninitialized.
while (n >= N) {
- bodyStart(body, x,0, v,v,v,v, v,v,v,v);
+ bodyStart(body, x, v,v,v,v, v,v,v,v);
x += N;
n -= N;
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-02 19:00:43 +0000