[clang] [llvm] [X86][AVX10.2] Support AVX10.2 MOVZXC new Instructions. (PR #108537)

via cfe-commits cfe-commits at lists.llvm.org
Mon Sep 16 22:29:42 PDT 2024


================
@@ -2139,9 +2139,9 @@ define <8 x i16> @pr59628_xmm(i16 %arg) {
 ; X86-LABEL: pr59628_xmm:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vxorps %xmm0, %xmm0, %xmm0
+; X86-NEXT:    vpxor %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vpbroadcastw %eax, %xmm1
-; X86-NEXT:    vmovsh %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
----------------
mahesh-attarde wrote:

Removing F16 Check results in this change. I could not decide whether this is positive or negative since uops table entries are missing. 

https://github.com/llvm/llvm-project/pull/108537


More information about the cfe-commits mailing list