This is the first time I’m answering my own question but it seems appropriate. Based on hirschhornsalz answer for prefix sum on 16 bytes simd-prefix-sum-on-intel-cpu I have come up with a solution for using SIMD on the first pass for 4, 8, and 16 32-bit words. The general theory goes as follows. For a sequential … Read more
Math says that it is indeed possible to go faster Agner Fog’s (http://www.agner.org/optimize/#vectorclass) method works great if division is done with a single divisor. Furthermore, this method has even further benefits if the divisor is known at compile time, or if it doesn’t change often at runtime. However, when performing SIMD division on __m128i elements … Read more
So, basically, you want your code to run faster. JNI is the answer. I know you said it didn’t work for you, but let me show you that you are wrong. Here’s Dot.java: import java.nio.FloatBuffer; import org.bytedeco.javacpp.*; import org.bytedeco.javacpp.annotation.*; @Platform(include = “Dot.h”, compiler = “fastfpu”) public class Dot { static { Loader.load(); } static float[] … Read more
I think it may be possible to implement BigNum with SIMD efficiently but not in the way you suggest. Instead of implementing a single BigNum using a SIMD register (or with an array of SIMD registers) you should process multiple BigNums at once. Let’s consider 128-bit addition. Let 128-bit integers be defined by a pair … Read more