Legal Information
Getting Help and Support
Introduction
Coding for the Intel® Processor Graphics
Platform-Level Considerations
Application-Level Optimizations
Optimizing OpenCL™ Usage with Intel® Processor Graphics
Check-list for OpenCL™ Optimizations
Performance Debugging
Using Multiple OpenCL™ Devices
Coding for the Intel® CPU OpenCL™ Device
OpenCL™ Kernel Development for Intel® CPU OpenCL™ device
Mapping Memory Objects
Using Buffers and Images Appropriately
Using Floating Point for Calculations
Using Compiler Options for Optimizations
Using Built-In Functions
Loading and Storing Data in Greatest Chunks
Applying Shared Local Memory
Using Specialization in Branching
Considering native_ and half_ Versions of Math Built-Ins
Using the Restrict Qualifier for Kernel Arguments
Avoiding Handling Edge Conditions in Kernels
Using Shared Context for Multiple OpenCL™ Devices
Sharing Resources Efficiently
Synchronization Caveats
Writing to a Shared Resource
Partitioning the Work
Keeping Kernel Sources the Same
Basic Frequency Considerations
Eliminating Device Starvation
Limitations of Shared Context with Respect to Extensions
Why Optimizing Kernel Code Is Important?
Avoid Spurious Operations in Kernel Code
Perform Initialization in a Separate Task
Use Preprocessor for Constants
Use Signed Integer Data Types
Use Row-Wise Data Accesses
Tips for Auto-Vectorization
Local Memory Usage
Avoid Extracting Vector Components
Task-Parallel Programming Model Hints
Avoid Extracting Vector Components
Consider the following kernel:
__constant float4 oneVec = (float4)(1.0f, 1.0f, 1.0f, 1.0f);
__kernel __attribute__((vec_type_hint(float4)))
void inverter2(__global float4* input, __global float4* output)
{
int tid = get_global_id(0);
output[tid] = oneVec – input[tid];
output[tid].w = input[tid].w;
output[tid] = sqrt(output[tid]);
}
For this example of the explicit vector code, extraction of the w component is very costly. The reason is that the next vector operation forces re-loading the same vector from memory. Consider loading a vector once and performing all changes, even to a single component, by use of vector operations.
In this specific case, two changes are required:
- Modify the oneVec so that its w component is zero, causing only a sign change in the w component of the input vector.
- Use float representation to manually change the sign bit of the w component back.
As a result, the kernel appears as follows:
__constant float4 oneVec = (float4)(1.0f, 1.0f, 1.0f, 0.0f);
__constant int4 signChanger = (int4)(0, 0, 0, 0x80000000);
__kernel __attribute__((vec_type_hint(float4)))
void inverter3(__global float4* input, __global float4* output)
{
int tid = get_global_id(0);
output[tid] = oneVec – input[tid];
output[tid] = as_float4(as_int4(output[tid]) ^ signChanger);
output[tid] = sqrt(output[tid]);
}
At the cost of another constant vector, this implementation performs all the required operations addressing only full vectors. All the computations can be performed in float8.