AlgorithmsIntermediate~60 min

GPU Histogram with Privatization

Compute an 8-bit grayscale histogram three ways: global atomics, per-warp privatization, and per-block shared-memory privatization.

atomicsatomicAddshared-memoryprivatizationhistogram

// PREREQUISITES

// PROJECT PHASES

  1. 1

    Global-memory atomics

    atomicAdd directly into a 256-bin global histogram.

  2. 2

    Shared-memory privatization

    Each block maintains its own histogram in shared memory.

  3. 3

    Warp-level privatization

    Each warp maintains separate bins.

  4. 4

    Benchmark

    Measure atomic contention at different input sizes.

// SANDBOX

GPU execution
main.cu
loading editor…

// EXPECTED OUTPUT

Global atomics:  48 ms  (~330 Mpix/s)
Shared memory:   12 ms  (~1.3 Gpix/s)

// YOUR PROGRESS

Log in to save progress

// DETAILS

Category
Algorithms
Difficulty
Intermediate
Est. time
~60 min
Prerequisites
1

// LEARN THE THEORY

Algorithms concepts