AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

SkJumper: skip null contexts

Browse Source 
This makes stages that don't use a context pointer look a little
cleaner, especially on ARM.  No interesting speed difference on x86.

What do you think?

Change-Id: I445472be2aa8a7c3bc8cba443fa477a3628118ba
Reviewed-on: https://skia-review.googlesource.com/9155
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
This commit is contained in:
Mike Klein 2017-03-02 12:42:14 -05:00 committed by Skia Commit-Bot
parent cf2e8c6139
commit 8e8e817cbf
4 changed files with 378 additions and 507 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/jumper/SkJumper.cpp
View File

@ -261,7 +261,9 @@ bool SkRasterPipeline::run_with_jumper(size_t x, size_t n) const {
return false;
}
*ip++ = (void*)fn;
*ip++ = st.ctx;
if (st.ctx) {
*ip++ = st.ctx;
}
}
*ip = (void*)just_return;

468
src/jumper/SkJumper_generated.S
View File

File diff suppressed because it is too large Load Diff

376
src/jumper/SkJumper_generated_win.S
View File

File diff suppressed because it is too large Load Diff

37
src/jumper/SkJumper_stages.cpp
View File

@ -344,6 +344,27 @@ static void* load_and_inc(void**& program) {
#endif
}
// Doesn't do anything unless you resolve it, either by casting to a pointer or calling load().
// This makes it free in stages that have no context pointer to load (i.e. built with nullptr).
struct LazyCtx {
void* ptr;
void**& program;
explicit LazyCtx(void**& p) : ptr(nullptr), program(p) {}
template <typename T>
operator T*() {
if (!ptr) { ptr = load_and_inc(program); }
return (T*)ptr;
}
template <typename T>
T load() {
if (!ptr) { ptr = load_and_inc(program); }
return unaligned_load<T>(ptr);
}
};
#if defined(JUMPER) && defined(__AVX__)
// There's a big cost to switch between SSE and AVX+, so we do a little
// extra work to handle even the jagged <kStride tail in AVX+ mode.
@ -366,16 +387,16 @@ static void* load_and_inc(void**& program) {
}
#define STAGE(name) \
static void name##_k(size_t x, void* ctx, K* k, size_t tail, \
static void name##_k(size_t x, LazyCtx ctx, K* k, size_t tail, \
F& r, F& g, F& b, F& a, F& dr, F& dg, F& db, F& da); \
extern "C" void WRAP(name)(size_t x, void** program, K* k, size_t tail, \
F r, F g, F b, F a, F dr, F dg, F db, F da) { \
auto ctx = load_and_inc(program); \
LazyCtx ctx(program); \
name##_k(x,ctx,k,tail, r,g,b,a, dr,dg,db,da); \
auto next = (Stage*)load_and_inc(program); \
next(x,program,k,tail, r,g,b,a, dr,dg,db,da); \
} \
static void name##_k(size_t x, void* ctx, K* k, size_t tail, \
static void name##_k(size_t x, LazyCtx ctx, K* k, size_t tail, \
F& r, F& g, F& b, F& a, F& dr, F& dg, F& db, F& da)
#else
@ -400,16 +421,16 @@ static void* load_and_inc(void**& program) {
}
#define STAGE(name) \
static void name##_k(size_t x, void* ctx, K* k, size_t tail, \
static void name##_k(size_t x, LazyCtx ctx, K* k, size_t tail, \
F& r, F& g, F& b, F& a, F& dr, F& dg, F& db, F& da); \
extern "C" void WRAP(name)(size_t x, void** program, K* k, \
F r, F g, F b, F a, F dr, F dg, F db, F da) { \
auto ctx = load_and_inc(program); \
LazyCtx ctx(program); \
name##_k(x,ctx,k,0, r,g,b,a, dr,dg,db,da); \
auto next = (Stage*)load_and_inc(program); \
next(x,program,k, r,g,b,a, dr,dg,db,da); \
} \
static void name##_k(size_t x, void* ctx, K* k, size_t tail, \
static void name##_k(size_t x, LazyCtx ctx, K* k, size_t tail, \
F& r, F& g, F& b, F& a, F& dr, F& dg, F& db, F& da)
#endif
@ -446,7 +467,7 @@ STAGE(seed_shader) {
}
STAGE(constant_color) {
auto rgba = unaligned_load<F4>(ctx);
auto rgba = ctx.load<F4>();
r = rgba[0];
g = rgba[1];
b = rgba[2];
@ -1003,7 +1024,7 @@ STAGE(matrix_perspective) {
STAGE(linear_gradient_2stops) {
struct Ctx { F4 c0, dc; };
auto c = unaligned_load<Ctx>(ctx);
auto c = ctx.load<Ctx>();
auto t = r;
r = mad(t, c.dc[0], c.c0[0]);