Created
October 1, 2012 10:32
-
-
Save danieldk/3810803 to your computer and use it in GitHub Desktop.
Parallel mapping in Go
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
package parmap | |
import ( | |
"runtime" | |
) | |
type empty struct{} | |
type semaphore chan empty | |
// Simple parallel map, that starts a Goroutine for every value. | |
// Do not use for large vectors, the Goroutines will eat all your | |
// memory! | |
func ParMap(f func(float64) float64, l []float64) []float64 { | |
n := len(l) | |
result := make([]float64, n) | |
sem := make(semaphore, n) | |
for i := 0; i < n; i++ { | |
go func(idx int) { | |
result[idx] = f(l[idx]) | |
sem <- empty{} | |
}(i) | |
} | |
for i := 0; i < n; i++ { | |
<-sem | |
} | |
return result | |
} | |
// Parallel map that divides the work set in N chunks, where N is the | |
// number of CPUs. The data is linearly divided. | |
func ParMapChunked(f func(float64) float64, l []float64) []float64 { | |
n := len(l) | |
result := make([]float64, n) | |
cpus := min(runtime.NumCPU(), n) | |
sem := make(semaphore, cpus) | |
chunkSize := max(n/cpus, 1) | |
for i := 0; i < n; i += chunkSize { | |
go chunkedWorker(sem, l, result, i, chunkSize, f) | |
} | |
for i := 0; i < cpus; i++ { | |
<-sem | |
} | |
return result | |
} | |
func chunkedWorker(sem semaphore, in, out []float64, idx, chunkSize int, f func(float64) float64) { | |
n := len(in) | |
for i := idx; i < idx+chunkSize; i++ { | |
if i >= n { | |
break | |
} | |
out[i] = f(in[i]) | |
} | |
sem <- empty{} | |
} | |
// Parallel map that divides the work set in N chunks, where N is the | |
// number of CPUs. The data is divided by interleaving. Use when the | |
// computation time will be uneven for regions of the vector. | |
func ParMapInterleaved(f func(float64) float64, l []float64) []float64 { | |
result := make([]float64, len(l)) | |
cpus := runtime.NumCPU() | |
sem := make(semaphore, cpus) | |
for i := 0; i < cpus; i++ { | |
go interleavedWorker(sem, l, result, i, cpus, f) | |
} | |
// Block until workers are done. | |
for i := 0; i < cpus; i++ { | |
<-sem | |
} | |
return result | |
} | |
func interleavedWorker(sem semaphore, in, out []float64, startIdx, cpus int, f func(float64) float64) { | |
n := len(in) | |
for i := startIdx; i < n; i += cpus { | |
out[i] = f(in[i]) | |
} | |
sem <- empty{} | |
} | |
func max(l, r int) int { | |
if l > r { | |
return l | |
} | |
return r | |
} | |
func min(l, r int) int { | |
if l < r { | |
return l | |
} | |
return r | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment