[clang] [llvm] [RISCV] Sync the LP64E i128 vararg and stack argument behavior with gcc. (PR #239884)
Sam Elliott via cfe-commits
cfe-commits at lists.llvm.org
Sat Nov 7 22:48:19 PST 2026
================
@@ -74,16 +74,7 @@ define i64 @caller_float_in_regs() nounwind {
define i64 @callee_float_on_stack(i128 %a, i128 %b, i128 %c, i128 %d, float %e) nounwind {
; RV64I-LP64E-FPELIM-LABEL: callee_float_on_stack:
; RV64I-LP64E-FPELIM: # %bb.0:
-; RV64I-LP64E-FPELIM-NEXT: addi sp, sp, -16
-; RV64I-LP64E-FPELIM-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; RV64I-LP64E-FPELIM-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
-; RV64I-LP64E-FPELIM-NEXT: addi s0, sp, 16
-; RV64I-LP64E-FPELIM-NEXT: andi sp, sp, -16
-; RV64I-LP64E-FPELIM-NEXT: lw a0, 16(s0)
-; RV64I-LP64E-FPELIM-NEXT: addi sp, s0, -16
-; RV64I-LP64E-FPELIM-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; RV64I-LP64E-FPELIM-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
-; RV64I-LP64E-FPELIM-NEXT: addi sp, sp, 16
+; RV64I-LP64E-FPELIM-NEXT: lw a0, 16(sp)
----------------
lenary wrote:
Fair enough. Not a blocker if the function isn't miscompiled.
https://github.com/llvm/llvm-project/pull/239884
More information about the cfe-commits
mailing list