Merge pull request #3282 from afonso360/x64-fix-brtables

cranelift: Fix `br_table` for `i64` types in x64 backend.

cranelift/codegen/src/isa/x64/lower.rs

@@ -7202,24 +7202,34 @@ impl LowerBackend for X64Backend {
 
                 Opcode::BrTable => {
                     let jt_size = targets.len() - 1;
-                    assert!(jt_size <= u32::max_value() as usize);
+                    assert!(jt_size <= u32::MAX as usize);
                     let jt_size = jt_size as u32;
 
+                    let ty = ctx.input_ty(branches[0], 0);
+                    let ext_spec = match ty {
+                        types::I128 => panic!("BrTable unimplemented for I128"),
+                        types::I64 => ExtSpec::ZeroExtendTo64,
+                        _ => ExtSpec::ZeroExtendTo32,
+                    };
+
                     let idx = extend_input_to_reg(
                         ctx,
                         InsnInput {
                             insn: branches[0],
                             input: 0,
                         },
-                        ExtSpec::ZeroExtendTo32,
+                        ext_spec,
                     );
 
                     // Bounds-check (compute flags from idx - jt_size) and branch to default.
-                    ctx.emit(Inst::cmp_rmi_r(
-                        OperandSize::Size32,
-                        RegMemImm::imm(jt_size),
-                        idx,
-                    ));
+                    // We only support u32::MAX entries, but we compare the full 64 bit register
+                    // when doing the bounds check.
+                    let cmp_size = if ty == types::I64 {
+                        OperandSize::Size64
+                    } else {
+                        OperandSize::Size32
+                    };
+                    ctx.emit(Inst::cmp_rmi_r(cmp_size, RegMemImm::imm(jt_size), idx));
 
                     // Emit the compound instruction that does:
                     //

cranelift/filetests/filetests/runtests/br_table.clif

@@ -1,6 +1,7 @@
 test interpret
 test run
 target aarch64
+target x86_64 machinst
 target s390x
 
 

cranelift/frontend/src/switch.rs

@@ -264,27 +264,22 @@ impl Switch {
                 }
             };
 
-            let discr = if bx.func.dfg.value_type(discr).bits() > 32 {
-                // Check for overflow of cast to u32.
+            let discr = if bx.func.dfg.value_type(discr).bits() > 64 {
+                // Check for overflow of cast to u32. This is the max supported jump table entries.
                 let new_block = bx.create_block();
                 let bigger_than_u32 =
                     bx.ins()
-                        .icmp_imm(IntCC::UnsignedGreaterThan, discr, u32::max_value() as i64);
+                        .icmp_imm(IntCC::UnsignedGreaterThan, discr, u32::MAX as i64);
                 bx.ins().brnz(bigger_than_u32, otherwise, &[]);
                 bx.ins().jump(new_block, &[]);
                 bx.seal_block(new_block);
                 bx.switch_to_block(new_block);
 
-                // Cast to u32, as br_table is not implemented for integers bigger than 32bits.
-                let discr = if bx.func.dfg.value_type(discr) == types::I128 {
+                // Cast to u64, as br_table is not implemented for i128
                 bx.ins().isplit(discr).0
             } else {
                 discr
             };
-                bx.ins().ireduce(types::I32, discr)
-            } else {
-                discr
-            };
 
             bx.ins().br_table(discr, otherwise, jump_table);
         }
@@ -616,13 +611,7 @@ block0:
     jump block4
 
 block4:
-    v1 = icmp_imm.i64 ugt v0, 0xffff_ffff
-    brnz v1, block3
-    jump block5
-
-block5:
-    v2 = ireduce.i32 v0
-    br_table v2, block3, jt0"
+    br_table.i64 v0, block3, jt0"
         );
     }
 
@@ -663,8 +652,7 @@ block4:
 
 block5:
     v2, v3 = isplit.i128 v0
-    v4 = ireduce.i32 v2
-    br_table v4, block3, jt0"
+    br_table v2, block3, jt0"
         );
     }
 }