Add x86 legalization for SIMD ineg

cranelift/codegen/meta/src/isa/x86/legalize.rs

@@ -36,6 +36,7 @@ pub(crate) fn define(shared: &mut SharedDefinitions, x86_instructions: &Instruct
     let iadd = insts.by_name("iadd");
     let iconst = insts.by_name("iconst");
     let imul = insts.by_name("imul");
+    let ineg = insts.by_name("ineg");
     let insertlane = insts.by_name("insertlane");
     let isub = insts.by_name("isub");
     let popcnt = insts.by_name("popcnt");
@@ -385,6 +386,7 @@ pub(crate) fn define(shared: &mut SharedDefinitions, x86_instructions: &Instruct
     narrow.custom_legalize(shuffle, "convert_shuffle");
     narrow.custom_legalize(extractlane, "convert_extractlane");
     narrow.custom_legalize(insertlane, "convert_insertlane");
+    narrow.custom_legalize(ineg, "convert_ineg");
 
     narrow.build_and_add_to(&mut shared.transform_groups);
 }

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

@@ -1704,6 +1704,17 @@ pub(crate) fn define(
         .operands_out(vec![a]),
     );
 
+    ig.push(
+        Inst::new(
+            "ineg",
+            r#"
+        Integer negation: `a := -x \pmod{2^B}`.
+        "#,
+        )
+        .operands_in(vec![x])
+        .operands_out(vec![a]),
+    );
+
     ig.push(
         Inst::new(
             "imul",

cranelift/codegen/src/isa/x86/enc_tables.rs

@@ -5,6 +5,7 @@ use crate::bitset::BitSet;
 use crate::cursor::{Cursor, FuncCursor};
 use crate::flowgraph::ControlFlowGraph;
 use crate::ir::condcodes::{FloatCC, IntCC};
+use crate::ir::immediates::V128Imm;
 use crate::ir::types::*;
 use crate::ir::{self, Function, Inst, InstBuilder};
 use crate::isa::constraints::*;
@@ -1090,3 +1091,27 @@ fn convert_insertlane(
         }
     }
 }
+
+/// For SIMD negation, convert an `ineg` to a `vconst + isub`.
+fn convert_ineg(
+    inst: ir::Inst,
+    func: &mut ir::Function,
+    _cfg: &mut ControlFlowGraph,
+    _isa: &dyn TargetIsa,
+) {
+    let mut pos = FuncCursor::new(func).at_inst(inst);
+    pos.use_srcloc(inst);
+
+    if let ir::InstructionData::Unary {
+        opcode: ir::Opcode::Ineg,
+        arg,
+    } = pos.func.dfg[inst]
+    {
+        let value_type = pos.func.dfg.value_type(arg);
+        if value_type.is_vector() && value_type.lane_type().is_int() {
+            let zero_immediate = pos.func.dfg.constants.insert(V128Imm::from(0).to_vec());
+            let zero_value = pos.ins().vconst(value_type, zero_immediate); // this should be legalized to a PXOR
+            pos.func.dfg.replace(inst).isub(zero_value, arg);
+        }
+    }
+}

cranelift/filetests/filetests/isa/x86/simd-arithmetic.clif

@@ -1,5 +1,6 @@
 test run
 test binemit
+test legalizer
 set enable_simd
 target x86_64 skylake
 
@@ -85,3 +86,37 @@ ebb0(v0: i8x16 [%xmm3], v1: i8x16 [%xmm4]):
 [-, %xmm3]  v2 = isub v0, v1      ; bin: 66 0f f8 dc
     return v2
 }
+
+function %ineg_i32x4() -> b1 {
+ebb0:
+    v0 = vconst.i32x4 [1 1 1 1]
+    v2 = ineg v0
+    ; check: v5 = vconst.i32x4 0x00
+    ; nextln: v2 = isub v5, v0
+
+    v3 = extractlane v2, 0
+    v4 = icmp_imm eq v3, -1
+
+    return v4 ; bin: c3
+}
+; run
+
+function %ineg_legalized() {
+ebb0:
+    v0 = vconst.i8x16 0x00
+    v1 = ineg v0
+    ; check: v6 = vconst.i8x16 0x00
+    ; nextln: v1 = isub v6, v0
+
+    v2 = raw_bitcast.i16x8 v0
+    v3 = ineg v2
+    ; check: v7 = vconst.i16x8 0x00
+    ; nextln: v3 = isub v7, v2
+
+    v4 = raw_bitcast.i64x2 v0
+    v5 = ineg v4
+    ; check: v8 = vconst.i64x2 0x00
+    ; nextln: v5 = isub v8, v4
+
+    return ; bin: c3
+}

cranelift/wasm/src/code_translator.rs

@@ -1004,6 +1004,10 @@ pub fn translate_operator<FE: FuncEnvironment + ?Sized>(
             let (a, b) = state.pop2();
             state.push1(builder.ins().isub(a, b))
         }
+        Operator::I8x16Neg | Operator::I16x8Neg | Operator::I32x4Neg | Operator::I64x2Neg => {
+            let a = state.pop1();
+            state.push1(builder.ins().ineg(a))
+        }
         Operator::I8x16Eq
         | Operator::I8x16Ne
         | Operator::I8x16LtS
@@ -1051,7 +1055,6 @@ pub fn translate_operator<FE: FuncEnvironment + ?Sized>(
         | Operator::V128Or
         | Operator::V128Xor
         | Operator::V128Bitselect
-        | Operator::I8x16Neg
         | Operator::I8x16AnyTrue
         | Operator::I8x16AllTrue
         | Operator::I8x16Shl
@@ -1062,7 +1065,6 @@ pub fn translate_operator<FE: FuncEnvironment + ?Sized>(
         | Operator::I8x16SubSaturateS
         | Operator::I8x16SubSaturateU
         | Operator::I8x16Mul
-        | Operator::I16x8Neg
         | Operator::I16x8AnyTrue
         | Operator::I16x8AllTrue
         | Operator::I16x8Shl
@@ -1073,14 +1075,12 @@ pub fn translate_operator<FE: FuncEnvironment + ?Sized>(
         | Operator::I16x8SubSaturateS
         | Operator::I16x8SubSaturateU
         | Operator::I16x8Mul
-        | Operator::I32x4Neg
         | Operator::I32x4AnyTrue
         | Operator::I32x4AllTrue
         | Operator::I32x4Shl
         | Operator::I32x4ShrS
         | Operator::I32x4ShrU
         | Operator::I32x4Mul
-        | Operator::I64x2Neg
         | Operator::I64x2AnyTrue
         | Operator::I64x2AllTrue
         | Operator::I64x2Shl