Manually include intrinsic conversion that is not present in LLVM files

Author: GuillaumeGomez

src/intrinsic/old_archs.rs

@@ -242,6 +242,8 @@ pub(crate) fn old_archs(arch: &str, name: &str) -> ArchCheckResult {
             "avx.cvt.ps2.pd.256" => "__builtin_ia32_cvtps2pd256",
             "avx.cvtdq2.pd.256" => "__builtin_ia32_cvtdq2pd256",
             "avx.cvtdq2.ps.256" => "__builtin_ia32_cvtdq2ps256",
+            "avx.round.pd.256" => "__builtin_ia32_roundpd256",
+            "avx.round.ps.256" => "__builtin_ia32_roundps256",
             "avx.sqrt.pd.256" => "__builtin_ia32_sqrtpd256",
             "avx.sqrt.ps.256" => "__builtin_ia32_sqrtps256",
             "avx.storeu.dq.256" => "__builtin_ia32_storedqu256",
@@ -1352,6 +1354,10 @@ pub(crate) fn old_archs(arch: &str, name: &str) -> ArchCheckResult {
             "sse41.pmovzxwd" => "__builtin_ia32_pmovzxwd128",
             "sse41.pmovzxwq" => "__builtin_ia32_pmovzxwq128",
             "sse41.pmuldq" => "__builtin_ia32_pmuldq128",
+            "sse41.round.pd" => "__builtin_ia32_roundpd",
+            "sse41.round.ps" => "__builtin_ia32_roundps",
+            "sse41.round.sd" => "__builtin_ia32_roundsd",
+            "sse41.round.ss" => "__builtin_ia32_roundss",
             "sse4a.movnt.sd" => "__builtin_ia32_movntsd",
             "sse4a.movnt.ss" => "__builtin_ia32_movntss",
             "ssse3.pabs.b.128" => "__builtin_ia32_pabsb128",