Enable more CLIF tests on AArch64

The tests for the SIMD floating-point maximum and minimum operations
require particular care because the handling of the NaN values is
non-deterministic and may vary between platforms. There is no way to
match several NaN values in a test, so the solution is to extract the
non-deterministic test cases into a separate file that is subsequently
replicated for every backend under test, with adjustments made to the
expected results.

Copyright (c) 2021, Arm Limited.

cranelift/codegen/meta/src/shared/instructions.rs

@@ -3597,9 +3597,12 @@ pub(crate) fn define(
         Inst::new(
             "fmin",
             r#"
-        Floating point minimum, propagating NaNs.
+        Floating point minimum, propagating NaNs using the WebAssembly rules.
 
-        If either operand is NaN, this returns a NaN.
+        If either operand is NaN, this returns NaN with an unspecified sign. Furthermore, if
+        each input NaN consists of a mantissa whose most significant bit is 1 and the rest is
+        0, then the output has the same form. Otherwise, the output mantissa's most significant
+        bit is 1 and the rest is unspecified.
         "#,
             &formats.binary,
         )
@@ -3629,9 +3632,12 @@ pub(crate) fn define(
         Inst::new(
             "fmax",
             r#"
-        Floating point maximum, propagating NaNs.
+        Floating point maximum, propagating NaNs using the WebAssembly rules.
 
-        If either operand is NaN, this returns a NaN.
+        If either operand is NaN, this returns NaN with an unspecified sign. Furthermore, if
+        each input NaN consists of a mantissa whose most significant bit is 1 and the rest is
+        0, then the output has the same form. Otherwise, the output mantissa's most significant
+        bit is 1 and the rest is unspecified.
         "#,
             &formats.binary,
         )

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

@@ -13,7 +13,7 @@ use crate::ir::Inst as IRInst;
 use crate::ir::{Opcode, Type};
 use crate::machinst::lower::*;
 use crate::machinst::*;
-use crate::CodegenResult;
+use crate::{CodegenError, CodegenResult};
 
 use crate::isa::aarch64::inst::*;
 use crate::isa::aarch64::AArch64Backend;
@@ -1103,6 +1103,41 @@ pub(crate) fn lower_vector_compare<C: LowerCtx<I = Inst>>(
         _ => false,
     };
     let size = VectorSize::from_ty(ty);
+
+    if is_float && (cond == Cond::Vc || cond == Cond::Vs) {
+        let tmp = ctx.alloc_tmp(ty).only_reg().unwrap();
+
+        ctx.emit(Inst::VecRRR {
+            alu_op: VecALUOp::Fcmeq,
+            rd,
+            rn,
+            rm: rn,
+            size,
+        });
+        ctx.emit(Inst::VecRRR {
+            alu_op: VecALUOp::Fcmeq,
+            rd: tmp,
+            rn: rm,
+            rm,
+            size,
+        });
+        ctx.emit(Inst::VecRRR {
+            alu_op: VecALUOp::And,
+            rd,
+            rn: rd.to_reg(),
+            rm: tmp.to_reg(),
+            size,
+        });
+
+        if cond == Cond::Vs {
+            ctx.emit(Inst::VecMisc {
+                op: VecMisc2::Not,
+                rd,
+                rn: rd.to_reg(),
+                size,
+            });
+        }
+    } else {
         // 'Less than' operations are implemented by swapping
         // the order of operands and using the 'greater than'
         // instructions.
@@ -1125,7 +1160,17 @@ pub(crate) fn lower_vector_compare<C: LowerCtx<I = Inst>>(
             (true, Cond::Ls) => (VecALUOp::Fcmge, true),
             (true, Cond::Ge) => (VecALUOp::Fcmge, false),
             (true, Cond::Gt) => (VecALUOp::Fcmgt, false),
-        _ => unreachable!(),
+            _ => {
+                return Err(CodegenError::Unsupported(format!(
+                    "Unsupported {} SIMD vector comparison: {:?}",
+                    if is_float {
+                        "floating-point"
+                    } else {
+                        "integer"
+                    },
+                    cond
+                )))
+            }
         };
 
         if swap {
@@ -1148,6 +1193,7 @@ pub(crate) fn lower_vector_compare<C: LowerCtx<I = Inst>>(
                 size,
             });
         }
+    }
 
     Ok(())
 }

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

@@ -1803,23 +1803,30 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
         }
 
         Opcode::Bint => {
+            let ty = ty.unwrap();
+
+            if ty.is_vector() {
+                return Err(CodegenError::Unsupported(format!(
+                    "Bint: Unsupported type: {:?}",
+                    ty
+                )));
+            }
+
             // Booleans are stored as all-zeroes (0) or all-ones (-1). We AND
             // out the LSB to give a 0 / 1-valued integer result.
-            let rn = put_input_in_reg(ctx, inputs[0], NarrowValueMode::None);
+            let input = put_input_in_regs(ctx, inputs[0]);
-            let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
+            let output = get_output_reg(ctx, outputs[0]);
-            let output_bits = ty_bits(ctx.output_ty(insn, 0));
 
-            let (imm_ty, alu_op) = if output_bits > 32 {
-                (I64, ALUOp::And64)
-            } else {
-                (I32, ALUOp::And32)
-            };
             ctx.emit(Inst::AluRRImmLogic {
-                alu_op,
+                alu_op: ALUOp::And32,
-                rd,
+                rd: output.regs()[0],
-                rn,
+                rn: input.regs()[0],
-                imml: ImmLogic::maybe_from_u64(1, imm_ty).unwrap(),
+                imml: ImmLogic::maybe_from_u64(1, I32).unwrap(),
             });
+
+            if ty_bits(ty) > 64 {
+                lower_constant_u64(ctx, output.regs()[1], 0);
+            }
         }
 
         Opcode::Bitcast => {
@@ -2240,7 +2247,9 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
             }
         }
 
-        Opcode::VallTrue if ctx.input_ty(insn, 0) == I64X2 => {
+        Opcode::VallTrue if ty_bits(ctx.input_ty(insn, 0).lane_type()) == 64 => {
+            debug_assert!(ctx.input_ty(insn, 0).is_vector());
+
             let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
             let rm = put_input_in_reg(ctx, inputs[0], NarrowValueMode::None);
             let tmp = ctx.alloc_tmp(I64X2).only_reg().unwrap();

cranelift/filetests/filetests/legalizer/bxor_imm.clif

@@ -1,4 +1,5 @@
 test legalizer
+target aarch64
 target x86_64
 
 function %foo(i64, i64) -> i64 {

cranelift/filetests/filetests/legalizer/iconst-i64.clif

@@ -1,4 +1,5 @@
 test legalizer
+target aarch64
 target i686
 
 function %foo() -> i64 {

cranelift/filetests/filetests/legalizer/isplit-bb.clif

@@ -1,4 +1,5 @@
 test legalizer
+target aarch64
 target x86_64
 
 function u0:0(i128, i128, i64) -> i128 system_v {

cranelift/filetests/filetests/licm/br-table.clif

@@ -1,5 +1,6 @@
 test compile
 set opt_level=speed_and_size
+target aarch64
 target x86_64
 
 function %br_table_opt() {

cranelift/filetests/filetests/licm/jump-table-entry.clif

@@ -1,4 +1,5 @@
 test licm
+target aarch64
 target x86_64
 
 function %dont_hoist_jump_table_entry_during_licm() {

cranelift/filetests/filetests/licm/load_readonly_notrap.clif

@@ -1,5 +1,6 @@
 test licm
 
+target aarch64
 target x86_64
 
 ;; Nontrapping readonly load from address that is not loop-dependent

cranelift/filetests/filetests/licm/reject_load_notrap.clif

@@ -1,5 +1,6 @@
 test licm
 
+target aarch64
 target x86_64
 
 ;; Nontrapping possibly-not-readonly load from address that is not

cranelift/filetests/filetests/licm/reject_load_readonly.clif

@@ -1,5 +1,6 @@
 test licm
 
+target aarch64
 target x86_64
 
 ;; Maybe-trapping readonly load from address that is not

cranelift/filetests/filetests/peepmatic/branch.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 function %icmp_to_brz_fold(i32) -> i32 {

cranelift/filetests/filetests/peepmatic/div_by_const_indirect.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64 baseline
 
 ; Cases where the denominator is created by an iconst

cranelift/filetests/filetests/peepmatic/div_by_const_non_power_of_2.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/peepmatic/div_by_const_power_of_2.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/peepmatic/do_not_keep_applying_optimizations_after_replacing_with_an_alias.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 ;; This file used to trigger assertions where we would keep trying to

cranelift/filetests/filetests/peepmatic/do_not_reorder_instructions_when_transplanting.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 ;; Test that although v5 can be replaced with v1, we don't transplant `load.i32

cranelift/filetests/filetests/peepmatic/fold-extended-move-wraparound.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 function %wraparound(i64 vmctx) -> f32 system_v {

cranelift/filetests/filetests/peepmatic/rem_by_const_non_power_of_2.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/peepmatic/rem_by_const_power_of_2.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/peepmatic/replace_branching_instructions_and_cfg_predecessors.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 function u0:2(i64 , i64) {

cranelift/filetests/filetests/peepmatic/simplify32.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target i686
 
 ;; 32-bits platforms.

cranelift/filetests/filetests/peepmatic/simplify64.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 ;; 64-bits platforms.

cranelift/filetests/filetests/peepmatic/simplify_instruction_into_alias_of_value.clif

@@ -1,4 +1,5 @@
 test peepmatic
+target aarch64
 target x86_64
 
 ;; The `isub` is a no-op, but we can't replace the whole `isub` instruction with

cranelift/filetests/filetests/postopt/basic.clif

@@ -1,4 +1,5 @@
 test postopt
+target aarch64
 target i686 legacy
 
 ; Test that compare+branch sequences are folded effectively on x86.

cranelift/filetests/filetests/preopt/branch.clif

@@ -1,4 +1,5 @@
 test preopt
+target aarch64
 target x86_64
 
 function %brz_fold() -> i32 {

cranelift/filetests/filetests/preopt/constant_fold.clif

@@ -1,4 +1,5 @@
 test preopt
+target aarch64
 target x86_64
 
 function %constant_fold(f64) -> f64 {

cranelift/filetests/filetests/preopt/numerical.clif

@@ -1,4 +1,5 @@
 test preopt
+target aarch64
 target x86_64
 
 function %iadd_fold() -> i32 {

cranelift/filetests/filetests/regress/allow-relaxation-shrink.clif

@@ -1,4 +1,5 @@
 test compile
+target aarch64
 target x86_64 legacy
 
 ; This checks that code shrink is allowed while relaxing code, when code shrink

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

@@ -1,4 +1,5 @@
 test run
+target aarch64
 target x86_64 machinst
 
 function %bint_b8_i128() -> i64, i64 {

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

@@ -1,4 +1,5 @@
 test run
+target aarch64
 target x86_64 machinst
 target x86_64 legacy
 

cranelift/filetests/filetests/runtests/simd-arithmetic-nondeterministic-aarch64.clif

@@ -0,0 +1,23 @@
+; Test the non-deterministic aspects of the SIMD arithmetic operations.
+; If you change this file, you should most likely update
+; simd-arithmetic-nondeterministic*.clif as well.
+test run
+target aarch64
+
+function %fmax_f64x2(f64x2, f64x2) -> f64x2 {
+block0(v0: f64x2, v1: f64x2):
+    v2 = fmax v0, v1
+    return v2
+}
+
+; run: %fmax_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [NaN:0x42 0.0]
+
+function %fmin_f64x2(f64x2, f64x2) -> f64x2 {
+block0(v0: f64x2, v1: f64x2):
+    v2 = fmin v0, v1
+    return v2
+}
+
+; run: %fmin_f64x2([-NaN 0x100.0], [0.0 NaN]) == [-NaN NaN]
+; run: %fmin_f64x2([NaN 0.0], [0.0 0.0]) == [NaN 0.0]
+; run: %fmin_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [NaN:0x42 0.0]

cranelift/filetests/filetests/runtests/simd-arithmetic-nondeterministic-x86_64.clif

@@ -0,0 +1,28 @@
+; Test the non-deterministic aspects of the SIMD arithmetic operations.
+; If you change this file, you should most likely update
+; simd-arithmetic-nondeterministic*.clif as well.
+test run
+set enable_simd
+target x86_64 machinst skylake
+
+function %fmax_f64x2(f64x2, f64x2) -> f64x2 {
+block0(v0: f64x2, v1: f64x2):
+    v2 = fmax v0, v1
+    return v2
+}
+
+; note below how NaNs are quieted but (unlike fmin), retain their sign: this discrepancy is allowed by non-determinism
+; in the spec, see https://webassembly.github.io/spec/core/bikeshed/index.html#nan-propagation%E2%91%A0.
+; run: %fmax_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [NaN 0.0]
+
+function %fmin_f64x2(f64x2, f64x2) -> f64x2 {
+block0(v0: f64x2, v1: f64x2):
+    v2 = fmin v0, v1
+    return v2
+}
+
+; note below how NaNs are quieted and negative: this is due to non-determinism in the spec for NaNs, see
+; https://webassembly.github.io/spec/core/bikeshed/index.html#nan-propagation%E2%91%A0.
+; run: %fmin_f64x2([-NaN 0x100.0], [0.0 NaN]) == [-NaN -NaN]
+; run: %fmin_f64x2([NaN 0.0], [0.0 0.0]) == [-NaN 0.0]
+; run: %fmin_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [-NaN 0.0]

cranelift/filetests/filetests/runtests/simd-arithmetic.clif

@@ -1,5 +1,5 @@
 test run
-; target aarch64 TODO: Not yet implemented on aarch64
+target aarch64
 ; target s390x TODO: Not yet implemented on s390x
 set enable_simd
 target x86_64 machinst skylake
@@ -136,26 +136,22 @@ block0(v0: f64x2, v1: f64x2):
     v2 = fmax v0, v1
     return v2
 }
-; note below how NaNs are quieted but (unlike fmin), retain their sign: this discrepancy is allowed by non-determinism
+; This operation exhibits non-deterministic behaviour for some input NaN values;
-; in the spec, see https://webassembly.github.io/spec/core/bikeshed/index.html#nan-propagation%E2%91%A0.
+; refer to the simd-arithmetic-nondeterministic*.clif files for the respective tests.
 ; run: %fmax_f64x2([-0x0.0 -0x1.0], [+0x0.0 0x1.0]) == [+0x0.0 0x1.0]
 ; run: %fmax_f64x2([-NaN NaN], [0x0.0 0x100.0]) == [-NaN NaN]
 ; run: %fmax_f64x2([NaN 0.0], [0.0 0.0]) == [NaN 0.0]
 ; run: %fmax_f64x2([-NaN 0.0], [0x1.0 0.0]) == [-NaN 0.0]
-; run: %fmax_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [NaN 0.0]
 
 function %fmin_f64x2(f64x2, f64x2) -> f64x2 {
 block0(v0: f64x2, v1: f64x2):
     v2 = fmin v0, v1
     return v2
 }
-; note below how NaNs are quieted and negative: this is due to non-determinism in the spec for NaNs, see
+; This operation exhibits non-deterministic behaviour for some input NaN values;
-; https://webassembly.github.io/spec/core/bikeshed/index.html#nan-propagation%E2%91%A0.
+; refer to the simd-arithmetic-nondeterministic*.clif files for the respective tests.
 ; run: %fmin_f64x2([-0x0.0 -0x1.0], [+0x0.0 0x1.0]) == [-0x0.0 -0x1.0]
-; run: %fmin_f64x2([-NaN 0x100.0], [0.0 NaN]) == [-NaN -NaN]
-; run: %fmin_f64x2([NaN 0.0], [0.0 0.0]) == [-NaN 0.0]
 ; run: %fmin_f64x2([-NaN 0.0], [0x1.0 0.0]) == [-NaN 0.0]
-; run: %fmin_f64x2([NaN:0x42 0.0], [0x1.0 0.0]) == [-NaN 0.0]
 
 function %fneg_f64x2(f64x2) -> f64x2 {
 block0(v0: f64x2):

cranelift/filetests/filetests/runtests/simd-bitwise-run.clif

@@ -1,5 +1,6 @@
 test run
 set enable_simd
+target aarch64
 target x86_64 legacy skylake
 
 ; TODO: once available, replace all lane extraction with `icmp + all_ones`

cranelift/filetests/filetests/runtests/simd-comparison.clif

@@ -1,5 +1,5 @@
 test run
-; target aarch64 TODO: Not yet implemented on aarch64
+target aarch64
 ; target s390x TODO: Not yet implemented on s390x
 set enable_simd
 target x86_64 machinst

cranelift/filetests/filetests/runtests/simd-lane-access.clif

@@ -1,5 +1,5 @@
 test run
-; target aarch64 TODO: Not yet implemented on aarch64
+target aarch64
 ; target s390x TODO: Not yet implemented on s390x
 set enable_simd
 target x86_64 machinst

cranelift/filetests/filetests/runtests/simd-vconst.clif

@@ -1,6 +1,6 @@
 test run
 ; target s390x TODO: Not yet implemented on s390x
-; target aarch64 TODO: Not yet implemented on aarch64
+target aarch64
 set enable_simd
 target x86_64 machinst
 set enable_simd

cranelift/filetests/filetests/simple_gvn/readonly.clif

@@ -1,5 +1,6 @@
 test simple-gvn
 
+target aarch64
 target x86_64
 
 function %eliminate_redundant_global_loads(i32, i64 vmctx) {

cranelift/filetests/filetests/simple_preopt/bitselect.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 ;; Test replacement of bitselect with vselect for special masks

cranelift/filetests/filetests/simple_preopt/branch.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 function %icmp_to_brz_fold(i32) -> i32 {

cranelift/filetests/filetests/simple_preopt/div_by_const_indirect.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64 baseline
 
 ; Cases where the denominator is created by an iconst

cranelift/filetests/filetests/simple_preopt/div_by_const_non_power_of_2.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/simple_preopt/div_by_const_power_of_2.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/simple_preopt/do_not_reorder_instructions_when_transplanting.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 ;; Test that although v5 can be replaced with v1, we don't transplant `load.i32

cranelift/filetests/filetests/simple_preopt/fold-extended-move-wraparound.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 function %wraparound(i64 vmctx) -> f32 system_v {

cranelift/filetests/filetests/simple_preopt/rem_by_const_non_power_of_2.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/simple_preopt/rem_by_const_power_of_2.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target i686 baseline
 
 ; -------- U32 --------

cranelift/filetests/filetests/simple_preopt/replace_branching_instructions_and_cfg_predecessors.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 function u0:2(i64 , i64) {

cranelift/filetests/filetests/simple_preopt/sign_extend.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 ;; Tests for sign-extending immediates.

cranelift/filetests/filetests/simple_preopt/simplify32.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target i686
 
 ;; 32-bits platforms.

cranelift/filetests/filetests/simple_preopt/simplify64.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 ;; 64-bits platforms.

cranelift/filetests/filetests/simple_preopt/simplify_instruction_into_alias_of_value.clif

@@ -1,4 +1,5 @@
 test simple_preopt
+target aarch64
 target x86_64
 
 ;; The `isub` is a no-op, but we can't replace the whole `isub` instruction with

cranelift/filetests/filetests/verifier/flags.clif

@@ -1,4 +1,5 @@
 test verifier
+target aarch64
 target i686
 
 ; Simple, correct use of CPU flags.

cranelift/filetests/filetests/verifier/scalar-to-vector.clif

@@ -1,5 +1,6 @@
 test verifier
 set enable_simd=true
+target aarch64
 target x86_64
 
 function %scalar_to_vector() {

cranelift/filetests/filetests/verifier/simd-lane-index.clif

@@ -1,5 +1,6 @@
 test verifier
 set enable_simd
+target aarch64
 target x86_64
 
 function %insertlane_i32x4() {

cranelift/filetests/filetests/wasm/control.clif

@@ -1,8 +1,8 @@
 ; Test basic code generation for control flow WebAssembly instructions.
 test compile
 
+target aarch64
 target i686 haswell
-
 target x86_64 haswell
 
 function %br_if(i32) -> i32 {

cranelift/filetests/filetests/wasm/conversions.clif

@@ -1,6 +1,7 @@
 ; Test code generation for WebAssembly type conversion operators.
 test compile
 
+target aarch64
 target x86_64 haswell
 
 function %i32_wrap_i64(i64) -> i32 {

cranelift/filetests/filetests/wasm/f32-arith.clif

@@ -1,6 +1,7 @@
 ; Test basic code generation for f32 arithmetic WebAssembly instructions.
 test compile
 
+target aarch64
 target i686 haswell
 target i686 baseline
 target x86_64 haswell

cranelift/filetests/filetests/wasm/f32-compares.clif

@@ -1,8 +1,8 @@
 ; Test code generation for WebAssembly f32 comparison operators.
 test compile
 
+target aarch64
 target i686 haswell
-
 target x86_64 haswell
 
 function %f32_eq(f32, f32) -> i32 {

cranelift/filetests/filetests/wasm/f32-memory64.clif

@@ -3,6 +3,7 @@ test compile
 
 ; We only test on 64-bit since the heap_addr instructions and vmctx parameters
 ; explicitly mention the pointer width.
+target aarch64
 target x86_64 haswell
 
 function %f32_load(i32, i64 vmctx) -> f32 {

cranelift/filetests/filetests/wasm/f64-arith.clif

@@ -1,6 +1,7 @@
 ; Test basic code generation for f64 arithmetic WebAssembly instructions.
 test compile
 
+target aarch64
 target x86_64 haswell
 target x86_64 baseline
 

cranelift/filetests/filetests/wasm/f64-compares.clif

@@ -1,8 +1,8 @@
 ; Test code generation for WebAssembly f64 comparison operators.
 test compile
 
+target aarch64
 target i686 haswell
-
 target x86_64 haswell
 
 function %f64_eq(f64, f64) -> i32 {

cranelift/filetests/filetests/wasm/f64-memory64.clif

@@ -3,6 +3,7 @@ test compile
 
 ; We only test on 64-bit since the heap_addr instructions and vmctx parameters
 ; explicitly mention the pointer width.
+target aarch64
 target x86_64 haswell
 
 function %f64_load(i32, i64 vmctx) -> f64 {

cranelift/filetests/filetests/wasm/i32-arith.clif

@@ -1,6 +1,7 @@
 ; Test basic code generation for i32 arithmetic WebAssembly instructions.
 test compile
 
+target aarch64
 target i686 haswell
 target i686 baseline
 target x86_64 haswell

cranelift/filetests/filetests/wasm/i32-compares.clif

@@ -1,8 +1,8 @@
 ; Test code generation for WebAssembly i32 comparison operators.
 test compile
 
+target aarch64
 target i686 haswell
-
 target x86_64 haswell
 
 function %i32_eqz(i32) -> i32 {

cranelift/filetests/filetests/wasm/i32-memory64.clif

@@ -3,6 +3,7 @@ test compile
 
 ; We only test on 64-bit since the heap_addr instructions and vmctx parameters
 ; explicitly mention the pointer width.
+target aarch64
 target x86_64 haswell
 
 function %i32_load(i32, i64 vmctx) -> i32 {

cranelift/filetests/filetests/wasm/i64-arith.clif

@@ -1,6 +1,7 @@
 ; Test basic code generation for i64 arithmetic WebAssembly instructions.
 test compile
 
+target aarch64
 target x86_64 haswell
 target x86_64 baseline
 

cranelift/filetests/filetests/wasm/i64-compares.clif

@@ -1,6 +1,7 @@
 ; Test code generation for WebAssembly i64 comparison operators.
 test compile
 
+target aarch64
 target x86_64 haswell
 
 function %i64_eqz(i64) -> i32 {

cranelift/filetests/filetests/wasm/i64-memory64.clif

@@ -3,6 +3,7 @@ test compile
 
 ; We only test on 64-bit since the heap_addr instructions and vmctx parameters
 ; explicitly mention the pointer width.
+target aarch64
 target x86_64 haswell
 
 function %i64_load(i32, i64 vmctx) -> i64 {

cranelift/filetests/filetests/wasm/multi-val-mixed.clif

@@ -1,4 +1,5 @@
 test compile
+target aarch64
 target x86_64 haswell
 
 ;; Returning many mixed values.

cranelift/filetests/filetests/wasm/multi-val-take-many-and-return-many.clif

@@ -1,4 +1,5 @@
 test compile
+target aarch64
 target x86_64 haswell
 
 function %returner(i32, i64, f32, f64) -> i32, i64, f32, f64 {

cranelift/filetests/filetests/wasm/r32.clif

@@ -4,6 +4,7 @@
 test compile
 set enable_safepoints=true
 
+target aarch64
 target i686 haswell
 
 function %select_ref(i32, r32, r32) -> r32 {

cranelift/filetests/filetests/wasm/r64.clif

@@ -4,6 +4,7 @@
 test compile
 set enable_safepoints=true
 
+target aarch64
 target x86_64 haswell
 
 function %select_ref(i32, r64, r64) -> r64 {

cranelift/filetests/filetests/wasm/select.clif

@@ -1,8 +1,8 @@
 ; Test basic code generation for the select WebAssembly instruction.
 test compile
 
+target aarch64
 target i686 haswell
-
 target x86_64 haswell
 
 function %select_i32(i32, i32, i32) -> i32 {

cranelift/filetests/src/test_licm.rs

@@ -26,6 +26,10 @@ impl SubTest for TestLICM {
         "licm"
     }
 
+    fn needs_isa(&self) -> bool {
+        true
+    }
+
     fn is_mutating(&self) -> bool {
         true
     }

cranelift/filetests/src/test_postopt.rs

@@ -23,6 +23,10 @@ impl SubTest for TestPostopt {
         "postopt"
     }
 
+    fn needs_isa(&self) -> bool {
+        true
+    }
+
     fn is_mutating(&self) -> bool {
         true
     }

cranelift/filetests/src/test_simple_preopt.rs

@@ -23,6 +23,10 @@ impl SubTest for TestSimplePreopt {
         "simple_preopt"
     }
 
+    fn needs_isa(&self) -> bool {
+        true
+    }
+
     fn is_mutating(&self) -> bool {
         true
     }