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

Revert "more powerful map()"

Browse Source 
This reverts commit a3dd5ec3a7.

Reason for revert: breaking build on  Build-Debian9-Clang-x86_64_Release-Fast
Original change's description:
> more powerful map()
> 
> Change-Id: Icbae002999a295e3a9d1d2e6046e686784d5f608
> Reviewed-on: https://skia-review.googlesource.com/69901
> Reviewed-by: Florin Malita <fmalita@chromium.org>
> Commit-Queue: Mike Klein <mtklein@chromium.org>

TBR=mtklein@chromium.org,fmalita@chromium.org

Change-Id: Ice989dd6a6b2786f318791dd91f2c06f689cb979
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/70105
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This commit is contained in:
Greg Daniel 2017-11-10 18:55:08 +00:00 committed by Skia Commit-Bot
parent 77e7500507
commit 9af41d7caa
1 changed files with 24 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
src/jumper/SkJumper_stages_lowp.cpp
View File

@ -183,28 +183,14 @@ SI D join(S lo, S hi) {
memcpy((char*)&v + 1*sizeof(S), &hi, sizeof(S)); memcpy((char*)&v + 1*sizeof(S), &hi, sizeof(S));
return v; return v;
} }
template <typename V, typename H>
#if defined(__AVX2__) SI V map(V v, H (*fn)(H)) {
template <typename V, typename Fn> H lo,hi;
SI auto map(V v, Fn&& fn) -> decltype( split(v, &lo,&hi);
__builtin_shufflevector(fn(__builtin_shufflevector(v,v, 0,1, 2, 3, 4, 5, 6, 7)), lo = fn(lo);
fn(__builtin_shufflevector(v,v, 8,9,10,11,12,13,14,15)), hi = fn(hi);
0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15)) { return join<V>(lo,hi);
return __builtin_shufflevector(fn(__builtin_shufflevector(v,v, 0,1, 2, 3, 4, 5, 6, 7)), }
fn(__builtin_shufflevector(v,v, 8,9,10,11,12,13,14,15)),
0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15);
}
#else
template <typename V, typename Fn>
SI auto map(V v, Fn&& fn) -> decltype(
__builtin_shufflevector(fn(__builtin_shufflevector(v,v, 0,1,2,3)),
fn(__builtin_shufflevector(v,v, 4,5,6,7)),
0,1,2,3,4,5,6,7)) {
return __builtin_shufflevector(fn(__builtin_shufflevector(v,v, 0,1,2,3)),
fn(__builtin_shufflevector(v,v, 4,5,6,7)),
0,1,2,3,4,5,6,7);
}
#endif
// TODO: do we need platform-specific intrinsics for any of these? // TODO: do we need platform-specific intrinsics for any of these?
SI F if_then_else(I32 c, F t, F e) { SI F if_then_else(I32 c, F t, F e) {
@ -222,7 +208,7 @@ SI F rcp(F x) {
#elif defined(__SSE__) #elif defined(__SSE__)
return map(x, _mm_rcp_ps); return map(x, _mm_rcp_ps);
#elif defined(__ARM_NEON) #elif defined(__ARM_NEON)
return map(x, [](float32x4_t v) { return map(x, +[](float32x4_t v) {
auto est = vrecpeq_f32(v); auto est = vrecpeq_f32(v);
return vrecpsq_f32(v,est)*est; return vrecpsq_f32(v,est)*est;
}); });
@ -238,7 +224,7 @@ SI F sqrt_(F x) {
#elif defined(__aarch64__) #elif defined(__aarch64__)
return map(x, vsqrtq_f32); return map(x, vsqrtq_f32);
#elif defined(__ARM_NEON) #elif defined(__ARM_NEON)
return map(x, [](float32x4_t v) { return map(x, +[](float32x4_t v) {
auto est = vrsqrteq_f32(v); // Estimate and two refinement steps for est = rsqrt(v). auto est = vrsqrteq_f32(v); // Estimate and two refinement steps for est = rsqrt(v).
est *= vrsqrtsq_f32(v,est*est); est *= vrsqrtsq_f32(v,est*est);
est *= vrsqrtsq_f32(v,est*est); est *= vrsqrtsq_f32(v,est*est);
@ -256,9 +242,9 @@ SI F floor_(F x) {
#if defined(__aarch64__) #if defined(__aarch64__)
return map(x, vrndmq_f32); return map(x, vrndmq_f32);
#elif defined(__AVX2__) #elif defined(__AVX2__)
return map(x, [](__m256 v){ return _mm256_floor_ps(v); }); // _mm256_floor_ps is a macro... return map(x, +[](__m256 v){ return _mm256_floor_ps(v); }); // _mm256_floor_ps is a macro...
#elif defined(__SSE4_1__) #elif defined(__SSE4_1__)
return map(x, [](__m128 v){ return _mm_floor_ps(v); }); // _mm_floor_ps() is a macro too. return map(x, +[](__m128 v){ return _mm_floor_ps(v); }); // _mm_floor_ps() is a macro too.
#else #else
F roundtrip = cast<F>(cast<I32>(x)); F roundtrip = cast<F>(cast<I32>(x));
return roundtrip - if_then_else(roundtrip > x, F(1), F(0)); return roundtrip - if_then_else(roundtrip > x, F(1), F(0));
@ -494,13 +480,21 @@ SI void store(T* ptr, size_t tail, V v) {
} }
template<> template<>
F gather(const float* ptr, U32 ix) { F gather(const float* p, U32 ix) {
return map(ix, [&](__m256i ix) { return _mm256_i32gather_ps(ptr, ix, 4); }); __m256i lo, hi;
split(ix, &lo, &hi);
return join<F>(_mm256_i32gather_ps(p, lo, 4),
_mm256_i32gather_ps(p, hi, 4));
} }
template<> template<>
U32 gather(const uint32_t* ptr, U32 ix) { U32 gather(const uint32_t* p, U32 ix) {
return map(ix, [&](__m256i ix) { return _mm256_i32gather_epi32(ptr, ix, 4); }); __m256i lo, hi;
split(ix, &lo, &hi);
return join<U32>(_mm256_i32gather_epi32(p, lo, 4),
_mm256_i32gather_epi32(p, hi, 4));
} }
#else #else
template <typename V, typename T> template <typename V, typename T>