15 // static variables, accessible from kernels or other functions. For uses from
16 // kernels this is straightforward - assign an integer to the kernel for the
28 // - a global accessed by one kernel exists independent of other kernels
30 // - the address of the global may be different from different kernels as they
34 // Uses from kernels are implemented here by grouping them in a per-kernel
75 // LDS variables. Variables that are only used from kernels are excluded.
78 // First kernels are assigned a unique integer identifier which is available in
81 // Second, each kernel allocates LDS variables independent of other kernels and
93 // error on any variable that might be reachable from multiple kernels and is
98 // number of kernels using the module strategy as that is free for the first
105 // - Kernels that don't use functions (or have had them all inlined) are not
106 //   affected by any lowering for kernels that do.
107 // - Kernels that don't make indirect function calls are not affected by those
109 // - Variables which are used by lots of kernels, e.g. those injected by a
110 //   language runtime in most kernels, are expected to have no overhead
122 // The AMDGPUPromoteAlloca pass skips kernels that use dynamic LDS.
157 // If the name correspondence can be replaced, multiple distinct kernels that
265     // The llvm.amdgcn.module.lds instance is implicitly used by all kernels  in markUsedByKernel()
267     // presently approximated to 'all kernels' if there are any such functions  in markUsedByKernel()
338       ArrayRef<Function *> kernels,  in buildLookupTable()  argument
346     const size_t NumberKernels = kernels.size();  in buildLookupTable()
357       auto Replacement = KernelToReplacement.find(kernels[i]);  in buildLookupTable()
458     // Find the global variable with the most indirect uses from kernels  in chooseBestVariableForModuleStrategy()
547     // Associate kernels in the set with an arbitrary but reproducible order and  in assignLDSKernelIDToEachKernel()
572       // Annotate the kernels with their order in this vector  in assignLDSKernelIDToEachKernel()
577         // 32 bit keeps it in one SGPR. > 2**32 kernels won't fit on the GPU  in assignLDSKernelIDToEachKernel()
578         report_fatal_error("Unimplemented LDS lowering for > 2**32 kernels");  in assignLDSKernelIDToEachKernel()
640               "' to kernel access as it is reachable from multiple kernels");  in partitionVariablesIntoIndirectStrategies()
673     // allocate this instance. That is a space optimisation - kernels that use a  in lowerModuleScopeStructVariables()
782       // name This is not expected to be a problem in practice as kernels  in lowerKernelScopeStructVariables()
785       // to name the kernels.  in lowerKernelScopeStructVariables()
787         report_fatal_error("Anonymous kernels cannot use LDS variables");  in lowerKernelScopeStructVariables()
883             report_fatal_error("Anonymous kernels cannot use LDS variables");  in lowerDynamicLDSVariables()
937     // For each variable accessed through callees, which kernels access it  in runOnModule()
1089           // Instead of explicitly marking kernels that access dynamic variables  in runOnModule()