[llvm] [X86] Avoid zero extend i16 when inserting fp16 (PR #126194)

[Phoebe Wang via llvm-commits]

================
@@ -43,15 +43,15 @@ define void @v_test_canonicalize__half(half addrspace(1)* %out) nounwind {
 ;
 ; AVX512-LABEL: v_test_canonicalize__half:
 ; AVX512:       # %bb.0: # %entry
-; AVX512-NEXT:    movzwl (%rdi), %eax
-; AVX512-NEXT:    movzwl {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ecx
-; AVX512-NEXT:    vmovd %ecx, %xmm0
-; AVX512-NEXT:    vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT:    vmovd %eax, %xmm1
+; AVX512-NEXT:    vpinsrw $0, (%rdi), %xmm0, %xmm0
+; AVX512-NEXT:    vpinsrw $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
+; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
----------------
phoebewang wrote:

Ok, limit it to non-strict FP only.

https://github.com/llvm/llvm-project/pull/126194