aboutsummaryrefslogtreecommitdiffhomepage
path: root/bench/Sk4fBench.cpp
diff options
context:
space:
mode:
authorGravatar mtklein <mtklein@chromium.org>2016-02-06 19:38:39 -0800
committerGravatar Commit bot <commit-bot@chromium.org>2016-02-06 19:38:39 -0800
commit2d340f2f37a28e82323428725019f12a8538f48e (patch)
treec10a798ec8fbea528fcd39c0c5f9521cbb02b192 /bench/Sk4fBench.cpp
parente721a8e883231af1878772ecd23be325113fcc48 (diff)
could not resist: fast sse float <--> u16
- generalizes the bench to float <--> {u8,u16} - must remember to implement NEON version at some point BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1676853002 CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot Review URL: https://codereview.chromium.org/1676853002
Diffstat (limited to 'bench/Sk4fBench.cpp')
-rw-r--r--bench/Sk4fBench.cpp32
1 files changed, 21 insertions, 11 deletions
diff --git a/bench/Sk4fBench.cpp b/bench/Sk4fBench.cpp
index 712a65776d..823e14c0af 100644
--- a/bench/Sk4fBench.cpp
+++ b/bench/Sk4fBench.cpp
@@ -10,37 +10,47 @@
#include "SkNx.h"
// Used to prevent the compiler from optimizing away the whole loop.
-volatile uint32_t blackhole = 0;
+volatile uint64_t blackhole = 0;
// Not a great random number generator, but it's very fast.
// The code we're measuring is quite fast, so low overhead is essential.
-static uint32_t lcg_rand(uint32_t* seed) {
+static uint64_t lcg_rand(uint64_t* seed) {
*seed *= 1664525;
*seed += 1013904223;
return *seed;
}
-struct Sk4fBytesRoundtripBench : public Benchmark {
- Sk4fBytesRoundtripBench() {}
+template <typename T>
+struct Sk4fRoundtripBench : public Benchmark {
+ Sk4fRoundtripBench() {}
+
+ const char* onGetName() override {
+ switch (sizeof(T)) {
+ case 1: return "Sk4f_roundtrip_u8";
+ case 2: return "Sk4f_roundtrip_u16";
+ }
+ SkASSERT(false);
+ return "";
+ }
- const char* onGetName() override { return "Sk4f_roundtrip"; }
bool isSuitableFor(Backend backend) override { return backend == kNonRendering_Backend; }
void onDraw(int loops, SkCanvas* canvas) override {
// Unlike blackhole, junk can and probably will be a register.
- uint32_t junk = 0;
- uint32_t seed = 0;
+ uint64_t junk = 0;
+ uint64_t seed = 0;
for (int i = 0; i < loops; i++) {
- uint32_t color = lcg_rand(&seed),
+ uint64_t src = lcg_rand(&seed),
back;
- auto f = SkNx_cast<float>(Sk4b::Load(&color));
- SkNx_cast<uint8_t>(f).store(&back);
+ auto f = SkNx_cast<float>(SkNx<4,T>::Load(&src));
+ SkNx_cast<T>(f).store(&back);
junk ^= back;
}
blackhole ^= junk;
}
};
-DEF_BENCH(return new Sk4fBytesRoundtripBench;)
+DEF_BENCH(return new Sk4fRoundtripBench<uint8_t>;)
+DEF_BENCH(return new Sk4fRoundtripBench<uint16_t>;)
struct Sk4fGradientBench : public Benchmark {
const char* onGetName() override { return "Sk4f_gradient"; }