[llvm] [X86] Optimize v4i16->v4i8 truncating stores via v4i32 widening (PR #186676)

[Simon Pilgrim via llvm-commits]

================
@@ -1538,3 +1538,29 @@ define void @test_masked_store_unaligned_v8i64(<8 x i64> %data, ptr %ptr, <8 x i
   store <8 x i64> %sel, ptr %ptr_vec, align 1
   ret void
 }
+
+define void @cast_i16x4_to_u8x4(ptr %a0, ptr %a1) {
+; AVX-LABEL: cast_i16x4_to_u8x4:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX-NEXT:    vmovd %xmm0, (%rdi)
+; AVX-NEXT:    retq
+;
+; AVX2-LABEL: cast_i16x4_to_u8x4:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT:    vmovd %xmm0, (%rdi)
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: cast_i16x4_to_u8x4:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmovzxwd {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero
+; AVX512-NEXT:    vpmovdb %xmm0, (%rdi)
+; AVX512-NEXT:    retq
+  %1 = load <4 x i16>, ptr %a1
+  %2 = trunc <4 x i16> %1 to <4 x i8>
----------------
RKSimon wrote:

worth adding truncs/truncus test coverage as well?

https://github.com/llvm/llvm-project/pull/186676