gpu::KernelCode Struct Reference

KernelCode is the representation of WGSL GPU code with template substitutions applied. It is a type around the code string with additional metadata for workgroup size and precision since they are specified in the WGSL code. Additionally, label and entryPoint are used by createKernel() to specify the label and entry point of the kernel. More...

#include <gpu.h>

Public Member Functions
	KernelCode (const std::string &pData="", size_t workgroupSize=256, NumType precision=kf32)
	Constructor to create a code object from a template string and optional workgroup size and precision.
	KernelCode (const std::string &pData, const Shape &workgroupSize={256, 1, 1}, NumType precision=kf32)
	Overload of the constructor to create a code object from a template string and workgroup size. Unlike the main factory function, this overload takes a single size_t workgroupSize parameter instead of a 3D shape for the workgroup size and instantiates a 3D shape with the workgroupSize in the x dimension and 1 in the y and z dimensions.

Public Attributes
std::string	data
Shape	workgroupSize
NumType	precision = kf32
std::string	label = "kernel"
std::string	entryPoint = "main"

Detailed Description

KernelCode is the representation of WGSL GPU code with template substitutions applied. It is a type around the code string with additional metadata for workgroup size and precision since they are specified in the WGSL code. Additionally, label and entryPoint are used by createKernel() to specify the label and entry point of the kernel.

Definition at line 272 of file gpu.h.

Constructor & Destructor Documentation

KernelCode() [1/2]

gpu::KernelCode::KernelCode ( const std::string & pData = "", size_t workgroupSize = 256, NumType precision = kf32 )

inline

Constructor to create a code object from a template string and optional workgroup size and precision.

Parameters

[in]	pData	Shader template string with placeholders
[in]	workgroupSize	Shape of the workgroup. Unlike tensor shapes which can be of arbitrary rank, workgroup size is always of rank 3 corresponding to x y and z. workgroupSize is stored as a field in the KernelCode instance that is returned by createShader().
[in]	precision	Data type precision to be substituted for {{precision}} in the WGSL code. As with workgroupSize, precision is stored as a field in the KernelCode instance that is returned by createShader(). KernelCode code = {kShaderText, {256, 1, 1}, kf32};

Definition at line 289 of file gpu.h.

      : data(pData), workgroupSize({workgroupSize, 1, 1}),
        precision(precision) {
    if (precision == kf16) {
      data = "enable f16;\n" + data;
    }
    replaceAll(data, "{{workgroupSize}}", toString({workgroupSize, 1, 1}));
    replaceAll(data, "{{precision}}", toString(precision));
    LOG(kDefLog, kTrace, "Shader code:\n%s", data.c_str());
  }

KernelCode() [2/2]

gpu::KernelCode::KernelCode ( const std::string & pData, const Shape & workgroupSize = {256, 1, 1}, NumType precision = kf32 )

inline

Overload of the constructor to create a code object from a template string and workgroup size. Unlike the main factory function, this overload takes a single size_t workgroupSize parameter instead of a 3D shape for the workgroup size and instantiates a 3D shape with the workgroupSize in the x dimension and 1 in the y and z dimensions.

Parameters

[in]	pData	Shader template string with placeholders
[in]	workgroupSize	Workgroup size in the x dimension
[in]	precision	Data type precision for the shader

KernelCode code = {kPuzzle1, 256, kf32};

Definition at line 317 of file gpu.h.

                                                 {256, 1, 1},
                    NumType precision = kf32)
      : data(pData), workgroupSize(workgroupSize), precision(precision) {
    replaceAll(data, "{{workgroupSize}}", toString(workgroupSize));
    replaceAll(data, "{{precision}}", toString(precision));
    LOG(kDefLog, kInfo, "Shader code:\n%s", data.c_str());
  }

Member Data Documentation

data

std::string gpu::KernelCode::data

Definition at line 325 of file gpu.h.

entryPoint

std::string gpu::KernelCode::entryPoint = "main"

Definition at line 329 of file gpu.h.

label

std::string gpu::KernelCode::label = "kernel"

Definition at line 328 of file gpu.h.

precision

NumType gpu::KernelCode::precision = kf32

Definition at line 327 of file gpu.h.

workgroupSize

Shape gpu::KernelCode::workgroupSize

Definition at line 326 of file gpu.h.

---

The documentation for this struct was generated from the following file:

• gpu.h