[machinst x64]: implement load*_zero for x64

cranelift/filetests/filetests/isa/x64/simd-lane-access-compile.clif

@@ -91,3 +91,34 @@ block0(v0: f64):
 ; check:  uninit  %xmm1
 ; nextln: movsd   %xmm0, %xmm1
 ; nextln: movlhps %xmm0, %xmm1
+
+
+
+;; load*_zero
+
+; Verify that a `load` followed by a `scalar_to_vector` (the CLIF translation of `load32_zero`) is
+; lowered to a single MOVSS instruction.
+function %load32_zero_coalesced(i64) -> i32x4 {
+block0(v0: i64):
+    v1 = load.i32 v0
+    v2 = scalar_to_vector.i32x4 v1
+    ; check:  movss   0(%rdi), %xmm0
+    return v2
+}
+
+;; Verify that `scalar_to_vector` (used by `load32_zero`), lowers as expected.
+function %load32_zero_int(i32) -> i32x4 {
+block0(v0: i32):
+    v1 = scalar_to_vector.i32x4 v0
+    ; check:  movd    %edi, %xmm0
+    return v1
+}
+function %load32_zero_float(f32) -> f32x4 {
+block0(v0: f32):
+    v1 = scalar_to_vector.f32x4 v0
+    ; regex: MOV=movap*
+    ; check: pushq
+    ; not: $MOV
+    ; check: ret
+    return v1
+}