Methods and apparatus to insert profiling instructions into a graphics processing unit kernel

At least one non-transitory computer-readable storage medium comprising instructions that, when executed, cause at least one processor to at least:

• detect a first entry point address and a second entry point address of an original GPU kernel;
• determine a size of the original GPU kernel;
• add instructions of the original GPU kernel and one or more profiling instructions to an instrumented GPU kernel using an instruction address offset, wherein the instruction address offset is based on the size of the original GPU kernel;
• insert, at the first entry point address of the instrumented GPU kernel, a first jump instruction to jump to first profiling initialization instructions;
• insert, at the second entry point address of the instrumented GPU kernel, a second jump instruction to jump to second profiling initialization instructions;
• and
• insert profiling measurement instructions of the profiling instructions into the instrumented GPU kernel.

An apparatus to insert profiling instructions into graphics processing unit (GPU) kernels, the apparatus comprising:

• means for detecting a first entry point address and a second entry point address of the original GPU kernel;
• means for creating a corresponding instrumented GPU kernel from the original GPU kernel by adding instructions of the original GPU kernel into and one or more profiling instructions to the instrumented GPU kernel, the means for creating to insert, at the first entry point address of the instrumented GPU kernel, a first jump instruction to jump to first profiling initialization instructions, the means for creating to insert, at the second entry point address of the instrumented GPU kernel, a second jump instruction to jump to second profiling initialization instructions, the means for creating to insert profiling measurement instructions of the profiling instructions into the instrumented GPU kernel;
• and
• means for determining a size of the original GPU kernel, wherein instructions of the original GPU kernel are added to the instrumented GPU kernel using an instruction address offset based on the size of the original GPU kernel.