LLVM / project / 37f0c8d / [X86] Emit @PLT for x86-64 and keep unadorned symbols for x86-32

LLVM/project 37f0c8d — llvm/lib/Target/X86 X86Subtarget.cpp, llvm/test/CodeGen/X86 memcpy.ll finite-libcalls.ll

Dec 5, 2020 by Fangrui Song on ⎇

release/12.x

[X86] Emit @PLT for x86-64 and keep unadorned symbols for x86-32

This essentially reverts the x86-64 side effect of r327198.

For x86-32, @PLT (R_386_PLT32) is not suitable in -fno-pic mode so the
code forces MO_NO_FLAG (like a forced dso_local) (https://bugs.llvm.org//show_bug.cgi?id=36674#c6).

For x86-64, both `call/jmp foo` and `call/jmp foo at PLT` emit R_X86_64_PLT32
(https://sourceware.org/bugzilla/show_bug.cgi?id=22791) so there is no
difference using @PLT. Using @PLT is actually favorable because this drops
a difference with -fpie/-fpic code and makes it possible to avoid a canonical
PLT entry when taking the address of an undefined function symbol.

Delta		File
+14	-14	llvm/test/CodeGen/X86/memcpy.ll
+12	-12	llvm/test/CodeGen/X86/finite-libcalls.ll
+13	-1	llvm/test/CodeGen/X86/lround-conv-i32.ll
+6	-6	llvm/test/CodeGen/X86/pow.ll
+4	-4	llvm/test/CodeGen/X86/extractelement-fp.ll
+4	-4	llvm/test/CodeGen/X86/fp128-i128.ll
+4	-4	llvm/test/CodeGen/X86/cvt16.ll
+3	-3	llvm/test/CodeGen/X86/lround-conv-i64.ll
+3	-3	llvm/test/CodeGen/X86/negative-sin.ll
+3	-3	llvm/test/CodeGen/X86/powi.ll
+3	-3	llvm/test/CodeGen/X86/llround-conv.ll
+2	-2	llvm/test/CodeGen/X86/half.ll
+2	-2	llvm/test/CodeGen/X86/extract-fp.ll
+2	-2	llvm/test/CodeGen/X86/pr38865.ll
+1	-1	llvm/test/CodeGen/X86/vector-half-conversions.ll
+1	-1	llvm/test/CodeGen/X86/fmaxnum.ll
+1	-1	llvm/test/CodeGen/X86/fmf-flags.ll
+1	-1	llvm/test/CodeGen/X86/fminnum.ll
+1	-1	llvm/test/CodeGen/X86/fp128-extract.ll
+1	-1	llvm/test/CodeGen/X86/fp128-libcalls.ll
+1	-1	llvm/test/CodeGen/X86/memcpy-inline.ll
+1	-1	llvm/test/CodeGen/X86/memset-nonzero.ll
+1	-1	llvm/lib/Target/X86/X86Subtarget.cpp
+84	-72	23 files

Unified Split Raw