[clang] [CIR][AArch64] Support BF16/FP16 NEON types and lower vdup lane builtins (PR #187460)

Andrzej WarzyƄski via cfe-commits cfe-commits at lists.llvm.org
Mon Mar 23 04:29:45 PDT 2026


================
@@ -34,3 +34,50 @@ bfloat16_t test_vduph_laneq_bf16(bfloat16x8_t v) {
   // LLVM: ret bfloat [[VGETQ_LANE]]
   return vduph_laneq_bf16(v, 7);
 }
+
+// ALL-LABEL: @test_vdup_lane_bf16(
+bfloat16x4_t test_vdup_lane_bf16(bfloat16x4_t v) {
+  // CIR: cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<4 x !u16i>) [#cir.int<1> : !s32i, #cir.int<1> : !s32i, #cir.int<1> : !s32i, #cir.int<1> : !s32i] : !cir.vector<4 x !u16i>
+  // LLVM: shufflevector <4 x {{.*}}> {{.*}}, <4 x {{.*}}> {{.*}}, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
----------------
banach-space wrote:

[ultra-nit] Here and in other places, extra empty line would improve readability.
```suggestion
  // CIR: cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<4 x !u16i>) [#cir.int<1> : !s32i, #cir.int<1> : !s32i, #cir.int<1> : !s32i, #cir.int<1> : !s32i] : !cir.vector<4 x !u16i>
  
  // LLVM: shufflevector <4 x {{.*}}> {{.*}}, <4 x {{.*}}> {{.*}}, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
```

https://github.com/llvm/llvm-project/pull/187460


More information about the cfe-commits mailing list