On 9/28/23 15:43, Vineet Gupta wrote:
RISC-V suffers from extraneous sign extensions, despite/given the ABI
guarantee that 32-bit quantities are sign-extended into 64-bit registers,
meaning incoming SI function args need not be explicitly sign extended
(so do SI return values as most ALU insns implicitly sign-extend too.)
Existing REE doesn't seem to handle this well and there are various ideas
floating around to smarten REE about it.
RISC-V also seems to correctly implement middle-end hook PROMOTE_MODE
etc.
Another approach would be to prevent EXPAND from generating the
sign_extend in the first place which this patch tries to do.
The hunk being removed was introduced way back in 1994 as
5069803972 ("expand_expr, case CONVERT_EXPR .. clear the promotion flag")
This survived full testsuite run for RISC-V rv64gc with surprisingly no
fallouts: test results before/after are exactly same.
| | # of unexpected case / # of unique unexpected
case
| | gcc | g++ | gfortran |
| rv64imafdc_zba_zbb_zbs_zicond/| 264 / 87 | 5 / 2 | 72 / 12 |
| lp64d/medlow
Granted for something so old to have survived, there must be a valid
reason. Unfortunately the original change didn't have additional
commentary or a test case. That is not to say it can't/won't possibly
break things on other arches/ABIs, hence the RFC for someone to scream
that this is just bonkers, don't do this :-)
I've explicitly CC'ed Jakub and Roger who have last touched subreg
promoted notes in expr.cc for insight and/or screaming ;-)
Thanks to Robin for narrowing this down in an amazing debugging session
@ GNU Cauldron.
```
foo2:
sext.w a6,a1 <-- this goes away
beq a1,zero,.L4
li a5,0
li a0,0
.L3:
addw a4,a2,a5
addw a5,a3,a5
addw a0,a4,a0
bltu a5,a6,.L3
ret
.L4:
li a0,0
ret
```
Signed-off-by: Vineet Gupta <vine...@rivosinc.com>
Co-developed-by: Robin Dapp <rdapp....@gmail.com>
---
gcc/expr.cc | 7 -------
gcc/testsuite/gcc.target/riscv/pr111466.c | 15 +++++++++++++++
2 files changed, 15 insertions(+), 7 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/pr111466.c
So mcore-elf is showing something interesting. With that hunk of Kenner
code removed, it actually has a few failing tests that flip to passes.
Tests that now work, but didn't before (11 tests):
mcore-sim: gcc.c-torture/execute/pr109986.c -O1 execution test
mcore-sim: gcc.c-torture/execute/pr109986.c -O2 execution test
mcore-sim: gcc.c-torture/execute/pr109986.c -O2 -flto -fno-use-linker-plugin
-flto-partition=none execution test
mcore-sim: gcc.c-torture/execute/pr109986.c -O2 -flto -fuse-linker-plugin
-fno-fat-lto-objects execution test
mcore-sim: gcc.c-torture/execute/pr109986.c -O3 -g execution test
mcore-sim: gcc.c-torture/execute/pr109986.c -Os execution test
mcore-sim: gcc.c-torture/execute/pr84524.c -O2 execution test
mcore-sim: gcc.c-torture/execute/pr84524.c -O2 -flto -fno-use-linker-plugin
-flto-partition=none execution test
mcore-sim: gcc.c-torture/execute/pr84524.c -O2 -flto -fuse-linker-plugin
-fno-fat-lto-objects execution test
mcore-sim: gcc.dg/tree-ssa/pr84436-5.c execution test
mcore-sim: gcc.dg/tree-ssa/pr84436-5.c execution test
So that's a really interesting result. If further analysis doesn't
point the finger at a simulator bug or something like that, then we'll
have strong evidence that Kenner's change is actively harmful from a
correctness standpoint. That would change the calculus here significantly.
Sadly, mcore-elf doesn't have a working gdb IIRC (don't ask how I know
that!), so I'm going to have to analyze this further with less efficient
techniques. BUt definitely interesting news/results.
Jeff