ELF/AMDGPU: Add support for R_AMDGPU_ABS32

[• tstellarAMD]

test/ELF/amdgpu-relocs-abs32.s

23 ↗

(On Diff #61701)

Is the full kernel_code_t definition required for this test?

[kzhuravl]

Rebase and update test

[ruiu]

ELF/Target.cpp

1462–1464

Sort alphabetically.

[• rafael]

we don't use llc to test elf lld.

[kzhuravl]

Feedback + update test

[ruiu]

Verifying an entire dwarf file to test ABS32 relocations seems a bit overkill. Why don't you add a few lines to test/ELF/amdgpu-relocs.s?

[kzhuravl]

R_AMDGPU_ABS32 relocations are only generated for dwarf

[ruiu]

The fact that ABS32 is generated only for DWARF does not necessarily mean that you need to create a DWARF to verify the behavior of the relocation. I was trying to say was that the test is too big compared to the size of the feature that is testing; it mostly verifies the structure of DWARF rather than the relocation. A test for ABS32 can be written in a few lines.

[kzhuravl]

Update test

[ruiu]

Don't you want to verify the contents of .data section?

[kzhuravl]

Update test

[ruiu]

LGTM

ELF/Target.cpp

 AMDGPUTargetInfo::AMDGPUTargetInfo() {
   GotRel = R_AMDGPU_ABS64;
   GotEntrySize = 8;
 }
 
 void AMDGPUTargetInfo::relocateOne(uint8_t *Loc, uint32_t Type,
                                    uint64_t Val) const {
   switch (Type) {
+  case R_AMDGPU_ABS32:
   case R_AMDGPU_GOTPCREL:
   case R_AMDGPU_REL32:
     write32le(Loc, Val);
     break;
   default:
     fatal("unrecognized reloc " + Twine(Type));
   }
 }
 
 RelExpr AMDGPUTargetInfo::getRelExpr(uint32_t Type, const SymbolBody &S) const {
   switch (Type) {
+  case R_AMDGPU_ABS32:
+    return R_ABS;
   case R_AMDGPU_REL32:
     return R_PC;
   case R_AMDGPU_GOTPCREL:
     return R_GOT_PC;
   default:
     fatal("do not know how to handle relocation " + Twine(Type));
   }
 }

test/ELF/amdgpu-relocs.s

kernel0:
   s_endpgm
 
   .comm   common_var,1024,4
   .globl  global_var
   .local  local_var
   .weak   weak_var
   .weakref weakref_var, weakref_alias_var
 
-# The relocation for local_var should be resolved by the linker.
+.section .data.foo, "w", @progbits
+  .long var, global_var
+
+# The relocation for local_var and var should be resolved by the linker.
 # CHECK: Relocations [
 # CHECK: .rela.dyn {
 # CHECK-NEXT: R_AMDGPU_ABS64 common_var 0x0
 # CHECK-NEXT: R_AMDGPU_ABS64 extern_var 0x0
 # CHECK-NEXT: R_AMDGPU_ABS64 global_var 0x0
 # CHECK-NEXT: R_AMDGPU_ABS64 weak_var 0x0
 # CHECK-NEXT: R_AMDGPU_ABS64 weakref_alias_var 0x0
 # CHECK-NEXT: }
 # CHECK-NEXT: ]