[llvm] [AMDGPU] Make maximum hard clause size a subtarget feature (PR #81287)

[Krzysztof Drewniak via llvm-commits]

================
@@ -227,6 +227,22 @@ def FeatureLdsBranchVmemWARHazard : SubtargetFeature<"lds-branch-vmem-war-hazard
   "Switching between LDS and VMEM-tex not waiting VM_VSRC=0"
 >;
 
+class FeatureMaxHardClauseLength<int size> : SubtargetFeature<
+  "max-hard-clause-length-"#size,
+  "MaxHardClauseLength",
+  !cast<string>(size),
+  "Maximum number of instructions in an explicit S_CLAUSE is "#size
+>;
+
+def FeatureMaxHardClauseLength32 : FeatureMaxHardClauseLength<32>;
+def FeatureMaxHardClauseLength63 : FeatureMaxHardClauseLength<63>;
+def FeatureNoHardClauses : SubtargetFeature<
----------------
krzysz00 wrote:

1. In case we need/want to explicitly write it out on gfx9 or something like that.
2. Might actually be useful for debugging in some ridiculously niche scenarios?

https://github.com/llvm/llvm-project/pull/81287