From 424f19732e948a854b454c70722a4eca8957bc20 Mon Sep 17 00:00:00 2001
From: Cameron McInally <cameron.mcinally@nyu.edu>
Date: Thu, 31 Oct 2013 13:56:31 +0000
Subject: [PATCH] Add AVX512 unmasked integer broadcast intrinsics and support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193748 91177308-0d34-0410-b5e6-96231b3b80d8
---
 include/llvm/IR/IntrinsicsX86.td      | 12 ++++++++++++
 lib/Target/X86/X86InstrAVX512.td      | 10 ++++++++++
 test/CodeGen/X86/avx512-intrinsics.ll | 28 +++++++++++++++++++++++++++
 3 files changed, 50 insertions(+)

diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td
index 74df28056e0..c24fb0638f4 100644
--- a/include/llvm/IR/IntrinsicsX86.td
+++ b/include/llvm/IR/IntrinsicsX86.td
@@ -2690,6 +2690,18 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx512_vbroadcast_sd_pd_512 :
               GCCBuiltin<"__builtin_ia32_vbroadcastsd_pd512">,
               Intrinsic<[llvm_v8f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_pbroadcastd_512 :
+         GCCBuiltin<"__builtin_ia32_pbroadcastd512">,
+         Intrinsic<[llvm_v16i32_ty], [llvm_v4i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_pbroadcastd_i32_512 :
+         Intrinsic<[llvm_v16i32_ty], [llvm_i32_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_pbroadcastq_512 :
+         GCCBuiltin<"__builtin_ia32_pbroadcastq512">,
+         Intrinsic<[llvm_v8i64_ty], [llvm_v2i64_ty], [IntrNoMem]>;
+  def int_x86_avx512_pbroadcastq_i64_512 :
+         Intrinsic<[llvm_v8i64_ty], [llvm_i64_ty], [IntrNoMem]>;
 }
 
 // Vector sign and zero extend
diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 1ac563010fd..5e854da1a0c 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -425,6 +425,11 @@ def : Pat<(v8i64 (X86VBroadcast (i64 GR64:$src))),
 def : Pat<(v8i64 (X86VBroadcastm VK8WM:$mask, (i64 GR64:$src))),
         (VPBROADCASTQrZkrr VK8WM:$mask, GR64:$src)>;
 
+def : Pat<(v16i32 (int_x86_avx512_pbroadcastd_i32_512 (i32 GR32:$src))),
+        (VPBROADCASTDrZrr GR32:$src)>;
+def : Pat<(v8i64 (int_x86_avx512_pbroadcastq_i64_512 (i64 GR64:$src))),
+        (VPBROADCASTQrZrr GR64:$src)>;
+
 multiclass avx512_int_broadcast_rm<bits<8> opc, string OpcodeStr,
                           X86MemOperand x86memop, PatFrag ld_frag,
                           RegisterClass DstRC, ValueType OpVT, ValueType SrcVT,
@@ -461,6 +466,11 @@ defm VPBROADCASTQZ  : avx512_int_broadcast_rm<0x59, "vpbroadcastq", i64mem,
                       loadi64, VR512, v8i64, v2i64, VK8WM>,  EVEX_V512, VEX_W,
                       EVEX_CD8<64, CD8VT1>;
 
+def : Pat<(v16i32 (int_x86_avx512_pbroadcastd_512 (v4i32 VR128X:$src))),
+          (VPBROADCASTDZrr VR128X:$src)>;
+def : Pat<(v8i64 (int_x86_avx512_pbroadcastq_512 (v2i64 VR128X:$src))),
+          (VPBROADCASTQZrr VR128X:$src)>;
+
 def : Pat<(v16f32 (X86VBroadcast (v4f32 VR128X:$src))),
           (VBROADCASTSSZrr VR128X:$src)>;
 def : Pat<(v8f64 (X86VBroadcast (v2f64 VR128X:$src))),
diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 29b508302d0..8a512487069 100644
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -234,6 +234,34 @@ define <8 x double> @test_x86_vbroadcast_sd_pd_512(<2 x double> %a0) {
 }
 declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.pd.512(<2 x double>) nounwind readonly
 
+define <16 x i32> @test_x86_pbroadcastd_512(<4 x i32>  %a0) {
+  ; CHECK: vpbroadcastd
+  %res = call <16 x i32> @llvm.x86.avx512.pbroadcastd.512(<4 x i32> %a0) ; <<16 x i32>> [#uses=1]
+  ret <16 x i32> %res
+}
+declare <16 x i32> @llvm.x86.avx512.pbroadcastd.512(<4 x i32>) nounwind readonly
+
+define <16 x i32> @test_x86_pbroadcastd_i32_512(i32  %a0) {
+  ; CHECK: vpbroadcastd
+  %res = call <16 x i32> @llvm.x86.avx512.pbroadcastd.i32.512(i32 %a0) ; <<16 x i32>> [#uses=1]
+  ret <16 x i32> %res
+}
+declare <16 x i32> @llvm.x86.avx512.pbroadcastd.i32.512(i32) nounwind readonly
+
+define <8 x i64> @test_x86_pbroadcastq_512(<2 x i64> %a0) {
+  ; CHECK: vpbroadcastq
+  %res = call <8 x i64> @llvm.x86.avx512.pbroadcastq.512(<2 x i64> %a0) ; <<8 x i64>> [#uses=1]
+  ret <8 x i64> %res
+}
+declare <8 x i64> @llvm.x86.avx512.pbroadcastq.512(<2 x i64>) nounwind readonly
+
+define <8 x i64> @test_x86_pbroadcastq_i64_512(i64 %a0) {
+  ; CHECK: vpbroadcastq
+  %res = call <8 x i64> @llvm.x86.avx512.pbroadcastq.i64.512(i64 %a0) ; <<8 x i64>> [#uses=1]
+  ret <8 x i64> %res
+}
+declare <8 x i64> @llvm.x86.avx512.pbroadcastq.i64.512(i64) nounwind readonly
+
 define <16 x i32> @test_x86_pmaxu_d(<16 x i32> %a0, <16 x i32> %a1) {
   ; CHECK: vpmaxud 
   %res = call <16 x i32> @llvm.x86.avx512.pmaxu.d(<16 x i32> %a0, <16 x i32> %a1) ; <<16 x i32>> [#uses=1]