LLVM 24.0.0git
AMDGPUPerfHintAnalysis.cpp File Reference

Analyzes if a function potentially memory bound and if a kernel kernel may benefit from limiting number of waves to reduce cache thrashing. More...

Go to the source code of this file.

Macros

#define DEBUG_TYPE   "amdgpu-perf-hint"

Functions

 STATISTIC (NumMemBound, "Number of functions marked as memory bound")
 STATISTIC (NumLimitWave, "Number of functions marked as needing limit wave")
 INITIALIZE_PASS (AMDGPUPerfHintAnalysisLegacy, DEBUG_TYPE, "Analysis if a function is memory bound", true, true) bool AMDGPUPerfHintAnalysisLegacy

Detailed Description

Analyzes if a function potentially memory bound and if a kernel kernel may benefit from limiting number of waves to reduce cache thrashing.

Definition in file AMDGPUPerfHintAnalysis.cpp.

Macro Definition Documentation

◆ DEBUG_TYPE

#define DEBUG_TYPE   "amdgpu-perf-hint"

Definition at line 33 of file AMDGPUPerfHintAnalysis.cpp.

Function Documentation

◆ INITIALIZE_PASS()

INITIALIZE_PASS ( AMDGPUPerfHintAnalysisLegacy ,
DEBUG_TYPE ,
"Analysis if a function is memory bound" ,
true ,
true  )

Definition at line 461 of file AMDGPUPerfHintAnalysis.cpp.

References DEBUG_TYPE.

◆ STATISTIC() [1/2]

STATISTIC ( NumLimitWave ,
"Number of functions marked as needing limit wave"  )

References TLI.

◆ STATISTIC() [2/2]

STATISTIC ( NumMemBound ,
"Number of functions marked as memory bound"  )