[PATCH] D43328: [RISCV] Update MC compression tests

[Alex Bradbury via Phabricator via llvm-commits]

asb added inline comments.


================
Comment at: test/MC/RISCV/compress-relocations.s:1-13
+# RUN: llvm-mc -triple riscv32 -mattr=+c -riscv-no-aliases < %s -show-encoding \
+# RUN:     | FileCheck -check-prefix=INSTR -check-prefix=FIXUP %s
+
+# Check prefixes:
+# FIXUP - Check the fixup on the instruction.
+# INSTR - Check the instruction is handled properly by the ASMPrinter
+c.jal foo
----------------
sabuasal wrote:
> asb wrote:
> > asb wrote:
> > > You've lost the RELOC checks.
> > Also if we're going to separate this to a separate file, relocations-compressed.s would be more consistent with naming of other files (such as fixups-compressed.s).
> I actually didn't forget. This is a bit tricky.
> 
> MC for RISCV , as of now, only generates relocations if the target for the branch\jump is unresolved. MC also relaxes branches\jumps to unresolved targets (look at MCAssembler::handleFixup and  MCAssembler::fixupNeedsRelaxation).
> With relaxation implemented, a compressed branch\jump to an undefined reference will always get relaxed to the expanded form and the relocation generated will be that of the expanded form.
> 
> One way this could be circumvented (if we need to) is by implementing shouldForceRelocation, this way a compressed branch to a local (resolved) symbol will also get a relocation.
I think the correct thing to do is to leave the RELOC checks in this patch, then update the checks in the patch that actually changes the behaviour.


https://reviews.llvm.org/D43328