aarch64: Implement lowering brz,brnz for i128 values

cranelift/codegen/src/isa/aarch64/lower_inst.rs

@@ -3515,6 +3515,7 @@ pub(crate) fn lower_branch<C: LowerCtx<I = Inst>>(
 
         match op0 {
             Opcode::Brz | Opcode::Brnz => {
+                let ty = ctx.input_ty(branches[0], 0);
                 let flag_input = InsnInput {
                     insn: branches[0],
                     input: 0,
@@ -3549,14 +3550,19 @@ pub(crate) fn lower_branch<C: LowerCtx<I = Inst>>(
                         kind: CondBrKind::Cond(cond),
                     });
                 } else {
-                    let rt = put_input_in_reg(
-                        ctx,
-                        InsnInput {
-                            insn: branches[0],
-                            input: 0,
-                        },
-                        NarrowValueMode::ZeroExtend64,
-                    );
+                    let rt = if ty == I128 {
+                        let tmp = ctx.alloc_tmp(I64).only_reg().unwrap();
+                        let input = put_input_in_regs(ctx, flag_input);
+                        ctx.emit(Inst::AluRRR {
+                            alu_op: ALUOp::Orr64,
+                            rd: tmp,
+                            rn: input.regs()[0],
+                            rm: input.regs()[1],
+                        });
+                        tmp.to_reg()
+                    } else {
+                        put_input_in_reg(ctx, flag_input, NarrowValueMode::ZeroExtend64)
+                    };
                     let kind = match op0 {
                         Opcode::Brz => CondBrKind::Zero(rt),
                         Opcode::Brnz => CondBrKind::NotZero(rt),

cranelift/filetests/filetests/isa/aarch64/condbr.clif

@@ -253,3 +253,49 @@ block1:
 ; check: movz x0, #1
 ; nextln: ldp fp, lr, [sp], #16
 ; nextln: ret
+
+
+function %i128_brz(i128){
+block0(v0: i128):
+    brz v0, block1
+    jump block1
+
+block1:
+    nop
+    return
+}
+
+; check: stp fp, lr, [sp, #-16]!
+; nextln: mov fp, sp
+; nextln: orr x0, x0, x1
+; nextln: cbz x0, label1 ; b label2
+; check: Block 1:
+; check: b label3
+; check: Block 2:
+; check: b label3
+; check: Block 3:
+; check: ldp fp, lr, [sp], #16
+; nextln: ret
+
+
+function %i128_brnz(i128){
+block0(v0: i128):
+    brnz v0, block1
+    jump block1
+
+block1:
+    nop
+    return
+}
+
+; check: stp fp, lr, [sp, #-16]!
+; nextln: mov fp, sp
+; nextln: orr x0, x0, x1
+; nextln: cbnz x0, label1 ; b label2
+; check: Block 1:
+; check: b label3
+; check: Block 2:
+; check: b label3
+; check: Block 3:
+; check: ldp fp, lr, [sp], #16
+; nextln: ret

cranelift/filetests/filetests/runtests/i128-br.clif

@@ -1,42 +1,45 @@
 test run
-; target aarch64 TODO: Not yet implemented on aarch64
+target aarch64
 ; target s390x TODO: Not yet implemented on s390x
 target x86_64 machinst
 target x86_64 legacy
 
 
-function %br_false() -> b1 {
-block0:
-    v10 = iconst.i64 0x42
-    v11 = iconst.i64 0x00
-    v0 = iconcat v10, v11
-    brz v0, block2
+function %i128_br(i64, i64) -> b1 {
+block0(v0: i64, v1: i64):
+    v2 = iconcat v0, v1
+    brz v2, block2
     jump block1
 
 block1:
-    v1 = bconst.b1 true
-    return v1
+    v3 = bconst.b1 true
+    return v3
 
 block2:
-    v2 = bconst.b1 false
-    return v2
+    v4 = bconst.b1 false
+    return v4
 }
-; run
+; run: %i128_br(0, 0) == false
+; run: %i128_br(-1, 0) == true
+; run: %i128_br(0, -1) == true
+; run: %i128_br(-1, -1) == true
 
-function %br_true() -> b1 {
-block0:
-    v10 = iconst.i64 0x00
-    v11 = iconst.i64 0x00
-    v0 = iconcat v10, v11
-    brz v0, block2
+
+function %i128_brnz(i64, i64) -> b1 {
+block0(v0: i64, v1: i64):
+    v2 = iconcat v0, v1
+    brnz v2, block2
     jump block1
 
 block1:
-    v1 = bconst.b1 false
-    return v1
+    v3 = bconst.b1 true
+    return v3
 
 block2:
-    v2 = bconst.b1 true
-    return v2
+    v4 = bconst.b1 false
+    return v4
 }
-; run
+; run: %i128_brnz(0, 0) == true
+; run: %i128_brnz(-1, 0) == false
+; run: %i128_brnz(0, -1) == false
+; run: %i128_brnz(-1, -1) == false