[X86][AVX512] Add patterns that match the AVX512 floating point register vbroadcast...

Patch by Cameron McInally <cameron.mcinally@nyu.edu>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193422 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 79b9ae877969416e9c322a62b9191c0d982a76d7..ed90a4bca3ba3443c72ec3ea747df2b8501d3339 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -465,6 +465,11 @@ def : Pat<(v16f32 (X86VBroadcast (v4f32 VR128X:$src))),
           (VBROADCASTSSZrr VR128X:$src)>;
 def : Pat<(v8f64 (X86VBroadcast (v2f64 VR128X:$src))),
           (VBROADCASTSDZrr VR128X:$src)>;
+
+def : Pat<(v16f32 (int_x86_avx512_vbroadcast_ss_ps_512 (v4f32 VR128X:$src))),
+          (VBROADCASTSSZrr VR128X:$src)>;
+def : Pat<(v8f64 (int_x86_avx512_vbroadcast_sd_pd_512 (v2f64 VR128X:$src))),
+          (VBROADCASTSDZrr VR128X:$src)>;
     
 // Provide fallback in case the load node that is used in the patterns above
 // is used by additional users, which prevents the pattern selection.

diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 01e7b698d9d8a0238b2721e0bd223032a841cc5a..5cd7311eebeb54e1bd1e53ce08f708539ccfd40c 100644 (file)
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -219,3 +219,17 @@ define <8 x double> @test_x86_vbroadcast_sd_512(i8* %a0) {
   ret <8 x double> %res
 }
 declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.512(i8*) nounwind readonly
+
+define <16 x float> @test_x86_vbroadcast_ss_ps_512(<4 x float> %a0) {
+  ; CHECK: vbroadcastss
+  %res = call <16 x float> @llvm.x86.avx512.vbroadcast.ss.ps.512(<4 x float> %a0) ; <<16 x float>> [#uses=1]
+  ret <16 x float> %res
+}
+declare <16 x float> @llvm.x86.avx512.vbroadcast.ss.ps.512(<4 x float>) nounwind readonly
+
+define <8 x double> @test_x86_vbroadcast_sd_pd_512(<2 x double> %a0) {
+  ; CHECK: vbroadcastsd
+  %res = call <8 x double> @llvm.x86.avx512.vbroadcast.sd.pd.512(<2 x double> %a0) ; <<8 x double>> [#uses=1]
+  ret <8 x double> %res
+}
+declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.pd.512(<2 x double>) nounwind readonly