Add saturating subtraction with a SIMD encoding

This includes the new instructions `ssub_sat` and `usub_sat` and only encodes the i8x16 and i16x8 types; these are what is needed for implementing the SIMD spec (see https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md#saturating-integer-subtraction).

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

@@ -480,6 +480,7 @@ pub(crate) fn define(
     let sqrt = shared.by_name("sqrt");
     let sshr = shared.by_name("sshr");
     let sshr_imm = shared.by_name("sshr_imm");
+    let ssub_sat = shared.by_name("ssub_sat");
     let stack_addr = shared.by_name("stack_addr");
     let store = shared.by_name("store");
     let store_complex = shared.by_name("store_complex");
@@ -501,6 +502,7 @@ pub(crate) fn define(
     let uload8_complex = shared.by_name("uload8_complex");
     let ushr = shared.by_name("ushr");
     let ushr_imm = shared.by_name("ushr_imm");
+    let usub_sat = shared.by_name("usub_sat");
     let vconst = shared.by_name("vconst");
     let x86_bsf = x86.by_name("x86_bsf");
     let x86_bsr = x86.by_name("x86_bsr");
@@ -1965,6 +1967,24 @@ pub(crate) fn define(
         e.enc_32_64(isub, rec_fa.opcodes(*opcodes));
     }
 
+    // SIMD integer saturating subtraction
+    e.enc_32_64(
+        ssub_sat.bind_vector_from_lane(I8, sse_vector_size),
+        rec_fa.opcodes(&PSUBSB),
+    );
+    e.enc_32_64(
+        ssub_sat.bind_vector_from_lane(I16, sse_vector_size),
+        rec_fa.opcodes(&PSUBSW),
+    );
+    e.enc_32_64(
+        usub_sat.bind_vector_from_lane(I8, sse_vector_size),
+        rec_fa.opcodes(&PSUBUSB),
+    );
+    e.enc_32_64(
+        usub_sat.bind_vector_from_lane(I16, sse_vector_size),
+        rec_fa.opcodes(&PSUBUSW),
+    );
+
     // SIMD integer multiplication: the x86 ISA does not have instructions for multiplying I8x16
     // and I64x2 and these are (at the time of writing) not necessary for WASM SIMD.
     for (ty, opcodes, isap) in &[

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

@@ -326,6 +326,22 @@ pub static PSUBD: [u8; 3] = [0x66, 0x0f, 0xfa];
 /// Subtract packed quadword integers in xmm2/m128 from xmm1 (SSE2).
 pub static PSUBQ: [u8; 3] = [0x66, 0x0f, 0xfb];
 
+/// Subtract packed signed byte integers in xmm2/m128 from packed signed byte integers in xmm1
+/// and saturate results (SSE2).
+pub static PSUBSB: [u8; 3] = [0x66, 0x0f, 0xe8];
+
+/// Subtract packed signed word integers in xmm2/m128 from packed signed word integers in xmm1
+/// and saturate results (SSE2).
+pub static PSUBSW: [u8; 3] = [0x66, 0x0f, 0xe9];
+
+/// Subtract packed unsigned byte integers in xmm2/m128 from packed unsigned byte integers in xmm1
+/// and saturate results (SSE2).
+pub static PSUBUSB: [u8; 3] = [0x66, 0x0f, 0xd8];
+
+/// Subtract packed unsigned word integers in xmm2/m128 from packed unsigned word integers in xmm1
+/// and saturate results (SSE2).
+pub static PSUBUSW: [u8; 3] = [0x66, 0x0f, 0xd9];
+
 /// Push r{16,32,64}.
 pub static PUSH_REG: [u8; 1] = [0x50];
 

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

@@ -1736,6 +1736,36 @@ pub(crate) fn define(
         .operands_out(vec![a]),
     );
 
+    ig.push(
+        Inst::new(
+            "usub_sat",
+            r#"
+        Subtract with unsigned saturation.
+
+        This is similar to `isub` but the operands are interpreted as unsigned integers and their 
+        difference, instead of wrapping, will be saturated to the lowest unsigned integer for
+        the controlling type (e.g. `0x00` for i8).
+        "#,
+        )
+        .operands_in(vec![x, y])
+        .operands_out(vec![a]),
+    );
+
+    ig.push(
+        Inst::new(
+            "ssub_sat",
+            r#"
+        Subtract with signed saturation.
+
+        This is similar to `isub` but the operands are interpreted as signed integers and their 
+        difference, instead of wrapping, will be saturated to the lowest or highest 
+        signed integer for the controlling type (e.g. `0x80` or `0x7F` for i8).
+        "#,
+        )
+        .operands_in(vec![x, y])
+        .operands_out(vec![a]),
+    );
+
     ig.push(
         Inst::new(
             "ineg",