Replace x86_packss with snarrow

Since the Wasm specification contains narrowing instructions (see https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md#integer-to-integer-narrowing) that lower to PACKSS*, the x86-specific instruction is not necessary in the CLIF IR.

cranelift/filetests/filetests/isa/x86/simd-bitwise-legalize.clif

@@ -34,7 +34,7 @@ block0:
     ; nextln: v9 = raw_bitcast.i16x8 v8
     ; nextln: v10 = x86_psra v9, v4
 
-    ; nextln: v2 = x86_packss v7, v10
+    ; nextln: v2 = snarrow v7, v10
     return v2
 }
 

cranelift/filetests/filetests/isa/x86/simd-lane-access-binemit.clif

@@ -118,8 +118,8 @@ block0(v0: i32x4 [%xmm7], v1: i32x4 [%xmm6]):
             return
 }
 
-function %packss_i16x8(i16x8, i16x8) {
+function %snarrow_i16x8(i16x8, i16x8) {
 block0(v0: i16x8 [%xmm7], v1: i16x8 [%xmm8]):
-[-, %xmm7]  v2 = x86_packss v0, v1          ; bin: 66 41 0f 63 f8
+[-, %xmm7]  v2 = snarrow v0, v1             ; bin: 66 41 0f 63 f8
             return
 }

cranelift/filetests/filetests/isa/x86/simd-lane-access-run.clif

@@ -206,15 +206,9 @@ block0:
 }
 ; run
 
-function %pack() -> b1 {
-block0:
-    v0 = vconst.i32x4 [0 1 -1 0x0001ffff]
-    v1 = vconst.i32x4 [4 5 -6 0xffffffff]
-    v2 = x86_packss v0, v1
-
-    v3 = vconst.i16x8 [0 1 -1 0x7fff 4 5 -6 0xffff]
-    v4 = icmp eq v2, v3
-    v5 = vall_true v4
-    return v5
+function %snarrow(i32x4, i32x4) -> i16x8 {
+block0(v0: i32x4, v1: i32x4):
+    v2 = snarrow v0, v1
+    return v2
 }
-; run
+; run: %snarrow([0 1 -1 0x0001ffff], [4 5 -6 0xffffffff]) == [0 1 -1 0x7fff 4 5 -6 0xffff]