s390x: Fix handling of sret arguments (#5116)

Skip synthetic StructReturn entries in the return value list.
Fixes https://github.com/bytecodealliance/wasmtime/issues/5089

cranelift/codegen/src/isa/s390x/inst.isle

@@ -3511,6 +3511,9 @@
 (decl abi_sig (SigRef) Sig)
 (extern constructor abi_sig abi_sig)
 
+(decl abi_first_ret (SigRef Sig) usize)
+(extern constructor abi_first_ret abi_first_ret)
+
 (decl abi_call_info (Sig ExternalName CallArgList CallRetList Opcode) BoxCallInfo)
 (extern constructor abi_call_info abi_call_info)
 

cranelift/codegen/src/isa/s390x/lower.isle

@@ -3892,7 +3892,8 @@
             (uses CallArgList (lower_call_args abi (range 0 (abi_num_args abi)) args))
             (defs CallRetList (defs_init abi))
             (_ InstOutput (side_effect (abi_call abi name uses defs (Opcode.Call)))))
-        (lower_call_rets abi defs (range 0 (abi_num_rets abi)) (output_builder_new))))
+        (lower_call_rets abi defs (range (abi_first_ret sig_ref abi)
+                                         (abi_num_rets abi)) (output_builder_new))))
 
 ;; Direct call to an out-of-range function (implicitly via pointer).
 (rule (lower (call (func_ref_data sig_ref name _) args))
@@ -3902,7 +3903,8 @@
             (defs CallRetList (defs_init abi))
             (target Reg (load_symbol_reloc (SymbolReloc.Absolute name 0)))
             (_ InstOutput (side_effect (abi_call_ind abi target uses defs (Opcode.Call)))))
-        (lower_call_rets abi defs (range 0 (abi_num_rets abi)) (output_builder_new))))
+        (lower_call_rets abi defs (range (abi_first_ret sig_ref abi)
+                                         (abi_num_rets abi)) (output_builder_new))))
 
 ;; Indirect call.
 (rule (lower (call_indirect sig_ref ptr args))
@@ -3912,7 +3914,8 @@
             (uses CallArgList (lower_call_args abi (range 0 (abi_num_args abi)) args))
             (defs CallRetList (defs_init abi))
             (_ InstOutput (side_effect (abi_call_ind abi target uses defs (Opcode.CallIndirect)))))
-        (lower_call_rets abi defs (range 0 (abi_num_rets abi)) (output_builder_new))))
+        (lower_call_rets abi defs (range (abi_first_ret sig_ref abi)
+                                         (abi_num_rets abi)) (output_builder_new))))
 
 ;; Lower function arguments.
 (decl lower_call_args (Sig Range ValueSlice) CallArgList)

cranelift/codegen/src/isa/s390x/lower/isle.rs

@@ -165,6 +165,13 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
         self.lower_ctx.sigs().abi_sig_for_sig_ref(sig_ref)
     }
 
+    fn abi_first_ret(&mut self, sig_ref: SigRef, abi: &Sig) -> usize {
+        // Return the index of the first actual return value, excluding
+        // any StructReturn that might have been added to Sig.
+        let sig = &self.lower_ctx.dfg().signatures[sig_ref];
+        self.lower_ctx.sigs()[*abi].num_rets() - sig.returns.len()
+    }
+
     fn abi_lane_order(&mut self, abi: &Sig) -> LaneOrder {
         lane_order_for_call_conv(self.lower_ctx.sigs()[*abi].call_conv())
     }

cranelift/filetests/filetests/isa/s390x/call.clif

@@ -220,3 +220,21 @@ block0(v0: i128, v1: i128, v2: i128, v3: i128, v4: i128, v5: i128, v6: i128, v7:
 ;   vst %v17, 0(%r2)
 ;   br %r14
 
+
+function %call_sret() -> i64 {
+    fn0 = colocated %g(i64 sret)
+
+block0:
+    v1 = iconst.i64 0
+    call fn0(v1)
+    trap user0
+}
+
+;   stmg %r14, %r15, 112(%r15)
+;   aghi %r15, -160
+;   virtual_sp_offset_adjust 160
+; block0:
+;   lghi %r2, 0
+;   brasl %r14, %g
+;   trap
+