Implement iabs for x86 SIMD

This only covers the types necessary for implementing the Wasm SIMD spec--`i8x16`, `i16x8`, `i32x4`.

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

@@ -1638,6 +1638,7 @@ fn define_simd(
     let fill_nop = shared.by_name("fill_nop");
     let fmul = shared.by_name("fmul");
     let fsub = shared.by_name("fsub");
+    let iabs = shared.by_name("iabs");
     let iadd = shared.by_name("iadd");
     let icmp = shared.by_name("icmp");
     let imul = shared.by_name("imul");
@@ -2184,6 +2185,12 @@ fn define_simd(
         e.enc_both_inferred(avgr, rec_fa.opcodes(opcodes));
     }
 
+    // SIMD integer absolute value.
+    for (ty, opcodes) in &[(I8, &PABSB[..]), (I16, &PABSW[..]), (I32, &PABSD)] {
+        let iabs = iabs.bind(vector(*ty, sse_vector_size));
+        e.enc_both_inferred_maybe_isap(iabs, rec_furm.opcodes(opcodes), Some(use_ssse3_simd));
+    }
+
     // SIMD logical operations
     let band = shared.by_name("band");
     let band_not = shared.by_name("band_not");

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

@@ -303,6 +303,17 @@ pub static OR_IMM8_SIGN_EXTEND: [u8; 1] = [0x83];
 /// Return the bitwise logical OR of packed single-precision values in xmm and x/m (SSE).
 pub static ORPS: [u8; 2] = [0x0f, 0x56];
 
+/// Compute the absolute value of bytes in xmm2/m128 and store the unsigned result in xmm1 (SSSE3).
+pub static PABSB: [u8; 4] = [0x66, 0x0f, 0x38, 0x1c];
+
+/// Compute the absolute value of 32-bit integers in xmm2/m128 and store the unsigned result in
+/// xmm1 (SSSE3).
+pub static PABSD: [u8; 4] = [0x66, 0x0f, 0x38, 0x1e];
+
+/// Compute the absolute value of 16-bit integers in xmm2/m128 and store the unsigned result in
+/// xmm1 (SSSE3).
+pub static PABSW: [u8; 4] = [0x66, 0x0f, 0x38, 0x1d];
+
 /// Converts 8 packed signed word integers from xmm1 and from xxm2/m128 into 16 packed signed byte
 /// integers in xmm1 using signed saturation (SSE2).
 pub static PACKSSWB: [u8; 3] = [0x66, 0x0f, 0x63];

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

@@ -10,6 +10,7 @@ block0(v0: i8x16 [%xmm6], v1: i8x16 [%xmm2]):
 [-, %xmm6]    v5 = ssub_sat v0, v1    ; bin: 66 0f e8 f2
 [-, %xmm6]    v6 = usub_sat v0, v1    ; bin: 66 0f d8 f2
 [-, %xmm6]    v7 = avg_round v0, v1   ; bin: 66 0f e0 f2
+[-, %xmm6]    v9 = iabs v1            ; bin: 66 0f 38 1c f2
 
     return
 }
@@ -23,6 +24,7 @@ block0(v0: i16x8 [%xmm3], v1: i16x8 [%xmm5]):
 [-, %xmm3]    v6 = ssub_sat v0, v1    ; bin: 66 0f e9 dd
 [-, %xmm3]    v7 = usub_sat v0, v1    ; bin: 66 0f d9 dd
 [-, %xmm3]    v8 = avg_round v0, v1   ; bin: 66 0f e3 dd
+[-, %xmm3]    v9 = iabs v1            ; bin: 66 0f 38 1d dd
 
     return
 }
@@ -32,6 +34,7 @@ block0(v0: i32x4 [%xmm0], v1: i32x4 [%xmm1]):
 [-, %xmm0]    v2 = iadd v0, v1        ; bin: 66 0f fe c1
 [-, %xmm0]    v3 = isub v0, v1        ; bin: 66 0f fa c1
 [-, %xmm0]    v4 = imul v0, v1        ; bin: 66 0f 38 40 c1
+[-, %xmm0]    v5 = iabs v1            ; bin: 66 0f 38 1e c1
 
     return
 }

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

@@ -270,3 +270,10 @@ block0:
     return v5
 }
 ; run
+
+function %iabs(i32x4) -> i32x4 {
+block0(v0: i32x4):
+    v1 = iabs v0
+    return v1
+}
+; run: %iabs([-42 -1 0 1]) == [42 1 0 1]