TableGen: Generate enum for runtime libcall implementations

[arsenm]

Work towards separating the ABI existence of libcalls vs. the
lowering selection. Set libcall selection through enums, rather
than through raw string names.

[arsenm]

• RuntimeLibcalls: Avoid adding f80 calls to default set #147658
• RuntimeLibcalls: Avoid adding ppcf128 calls to non-ppc targets #147656
• RISCV: Start moving runtime libcall config to tablegen #147652
• RuntimeLibcalls: Stop using defset for default calls #147651
• Hexagon: Move runtime libcall configuration to tablegen #147482
• DAG: Use fast variants of fast math libcalls #147481
• ARM: Start moving runtime libcalls into tablegen #146084
• ARM: Unconditionally set eabi libcall calling convs in RuntimeLibcalls #146083
• RuntimeLibcalls: Remove table of soft float compare cond codes #146082
• MSP430: Move libcall CC setting to RuntimeLibcallsInfo #146081
• Lanai: Use TableGen to set libcall calling conventions #146080
• TableGen: Handle setting runtime libcall calling conventions #144980
• RuntimeLibcalls: Associate calling convention with libcall impls #144979
• TableGen: Allow defining sets of runtime libraries #144978
• AArch64: Add libcall impl declarations for __arm_sc* memory functions #144977
• XCore: Declare libcalls used for align 4 memcpy #144976
• Hexagon: Add libcall declarations for special memcpy #144975
• ARM: Add runtime libcall definitions for aebi memory functions #144974
• TableGen: Generate enum for runtime libcall implementations #144973 👈 (View in Graphite)
• TableGen: Add runtime libcall backend #144972
• PowerPC: Stop reporting memcpy as an alias of memmove on AIX #143836
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[arsenm]

Merge activity

• Jun 27, 8:33 AM UTC: A user started a stack merge that includes this pull request via Graphite.
• Jun 27, 8:37 AM UTC: Graphite rebased this pull request as part of a merge.
• Jun 27, 8:40 AM UTC: @arsenm merged this pull request with Graphite.

[sbc100]

This change somehow broke some of the LTO tests on the emscripten waterfall.

It seems to be related to the list of symbols that lld uses to determine the list of all libcall names. This comes from lto::LTO::getRuntimeLibcallSymbols and uses in lld/wasm/Driver.cpp.

I added some logging an compared the results of this function before and after this change. It seems that acosf is not longer part of this list for some reason... could be related to the fact that acosf is the very first libs call and has 0 as its enum value?

[sbc100]

i.e. in the generated include/llvm/IR/RuntimeLibcalls.inc file I see:

enum Libcall : unsigned short {                                                       
  ACOS_F32 = 0,                                                                       
  ACOS_F64 = 1,                                                                       
  ACOS_F80 = 2,                                                                       
  ACOS_F128 = 3,                                                                      
  ACOS_PPCF128 = 4,   

[arsenm]

I added some logging an compared the results of this function before and after this change. It seems that acosf is not longer part of this list for some reason... could be related to the fact that acosf is the very first libs call and has 0 as its enum value?

Yes, I'm guessing this drop_front no longer applies

[sbc100]

Yup this patch fixes it:

--- a/llvm/include/llvm/IR/RuntimeLibcalls.h
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.h
@@ -90,8 +90,7 @@ struct RuntimeLibcallsInfo {
   }
 
   ArrayRef<RTLIB::LibcallImpl> getLibcallImpls() const {
-    // Trim Unsupported from the start
-    return ArrayRef(LibcallImpls).drop_front();
+    return LibcallImpls;
   }
 
   /// Get the comparison predicate that's to be used to test the result of the

[arsenm]

Actually this one should be drop_back until the RTLIB::Libcall enum is swapped to use 0 as invalid