ARM64: toHalf() intrinsic for ARMv8
This CL implements an intrinsic for toHalf() method with
ARMv8.2 FP16 instructions.
This intrinsic implementation achieves bit-level compatibility with the
original Java implementation android.util.Half.toFloat().
The time required to execute the below code on Pixel3:
- Java implementation android.util.Half.toFloat():
- big cluster only: 2136ms
- little cluster only: 6442ms
- arm64 Intrinisic implementation:
- big cluster only: 1347ms (~37% faster)
- little cluster only: 4937ms (~ 23% faster)
int benchmarkToHalf() {
int result = 0;
// 5.9605E-8 is the smallest positive subnormal number that can be
// represented by FP16. This is 0x33800032 in float bits.
int raw_input = 0x33800032;
long before = 0;
long after = 0;
before = System.currentTimeMillis();
do {
float input = Float.intBitsToFloat(raw_input);
short output = FP16.toHalf(input);
result += output;
} while (++raw_input != 0x477fff00);
// 65535 is the max possible integer that can be represented by FP16.
//This is 0x477fff00 in float bits.
after = System.currentTimeMillis();
System.out.println("Time of FP16.toHalf (ms): " + (after - before));
return result;
}
Test: 580-fp16
Test: art/test/testrunner/run_build_test_target.py -j80 art-test-javac
Test: test-art-host, test-art-target
Change-Id: I69b152682390e5ffa5b3fdca60b496261191655d
11 files changed
