jalr-01.S fails with illegal operands on latest binutils 2.42

[piotro888]

jalr-01.S fails to assemble with latest binutils (>=2.42) with the following error:

riscv-arch-test/riscv-test-suite/rv32i_m/I/src/jalr-01.S:72: Error: illegal operands 'la x0,5b'

It happens due to TEST_JALR_OP macro expanding to load address instruction into x0 rd in

inst_7:
// rs1==x31, rd==x0, imm_val == 32, 
// opcode: jalr; op1:x31; dest:x0; immval:0x20; align:0 
TEST_JALR_OP(x15, x0, x31, 0x20, x10, 28,0)

I found that braking change was introduced in commit bminor/binutils-gdb@27b3396 on binutils including rejecting x0 as a valid operand for la.

Case with rd==x0 is present in RV32E, RV32I and RV64I

I'm not sure if this problem should be reported to binutils or maybe the test case changed here (the LA(rd,5b) inside the macro doesn't seem to be too useful, maybe could be replaced under .if rd==x0 or always)

[allenjbaum]

The test is responsible for generating this inside an LA macro wrapper, so in the next version of arch_test it will just skip generating code if the REG argument is X0. This particular case is should be a perfectly legitimate way of expressing a JR with JALR, and in fact JALR is converted by the assembler into a JR, so I'd file this as a issue with binutils. This test would ensure that the link is not store into any gpr.

…

On Thu, Mar 14, 2024 at 3:05 AM piotro888 ***@***.***> wrote: jalr-01.S fails to assemble with latest binutils (>=2.42) with the following error: riscv-arch-test/riscv-test-suite/rv32i_m/I/src/jalr-01.S:72: Error: illegal operands 'la x0,5b' It happens due to TEST_JALR_OP macro expanding to load address instruction into x0 rd in inst_7: // rs1==x31, rd==x0, imm_val == 32, // opcode: jalr; op1:x31; dest:x0; immval:0x20; align:0 TEST_JALR_OP(x15, x0, x31, 0x20, x10, 28,0) I found that braking change was introduced in commit bminor/binutils-gdb@ 27b3396 <bminor/binutils-gdb@27b3396> on binutils including rejecting x0 as a valid operand for la. Case with rd==x0 is present in RV32E, RV32I and RV64I I'm not sure if this problem should be reported to binutils or maybe the test case changed here (the LA(rd,5b) inside the macro doesn't seem to be too useful, maybe could be replaced under .if rd==x0 or always) — Reply to this email directly, view it on GitHub <#444>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJQD5LDHQ3AKZP4PUX3YYFY6NAVCNFSM6AAAAABEVY2HDGVHI2DSMVQWIX3LMV43ASLTON2WKOZSGE4DKOJUGQ4TQNQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[piotro888]

Test fixed with #446 to workaround binutils issue.