Remove 'llvm.x86.avx2.vbroadcasti128' intrinsic.

The intrinsic is no longer generated by the front-end. Remove the intrinsic and
auto-upgrade it to a vector shuffle.

Reviewed by Nadav

This is related to rdar://problem/18742778.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@231182 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td
index 622dabe1013f2ad60e27714e47ce5763ee9a04c8..4a59f0d54e04272b077d06e39d810c81c0f7cf19 100644 (file)
--- a/include/llvm/IR/IntrinsicsX86.td
+++ b/include/llvm/IR/IntrinsicsX86.td
@@ -1732,8 +1732,6 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx2_vbroadcast_ss_ps_256 :
               GCCBuiltin<"__builtin_ia32_vbroadcastss_ps256">,
               Intrinsic<[llvm_v8f32_ty], [llvm_v4f32_ty], [IntrNoMem]>;
-  def int_x86_avx2_vbroadcasti128 :
-              Intrinsic<[llvm_v4i64_ty], [llvm_ptr_ty], [IntrReadArgMem]>;
   def int_x86_avx2_pbroadcastb_128 :
               GCCBuiltin<"__builtin_ia32_pbroadcastb128">,
               Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty], [IntrNoMem]>;

diff --git a/lib/IR/AutoUpgrade.cpp b/lib/IR/AutoUpgrade.cpp
index 8306a1afeede9b17d58f76d429824a090d53cf20..6800e7a0a13a4a38df75f80ebde39e14a17e598b 100644 (file)
--- a/lib/IR/AutoUpgrade.cpp
+++ b/lib/IR/AutoUpgrade.cpp
@@ -179,6 +179,7 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
         Name == "x86.avx2.pblendw" ||
         Name == "x86.avx2.pblendd.128" ||
         Name == "x86.avx2.pblendd.256" ||
+        Name == "x86.avx2.vbroadcasti128" ||
         (Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
       NewFn = nullptr;
       return true;
@@ -553,6 +554,17 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
       for (unsigned I = 0; I < EltNum; ++I)
         Rep = Builder.CreateInsertElement(Rep, Load,
                                           ConstantInt::get(I32Ty, I));
+    } else if (Name == "llvm.x86.avx2.vbroadcasti128") {
+      // Replace vbroadcasts with a vector shuffle.
+      Value *Op = Builder.CreatePointerCast(
+          CI->getArgOperand(0),
+          PointerType::getUnqual(VectorType::get(Type::getInt64Ty(C), 2)));
+      Value *Load = Builder.CreateLoad(Op);
+      SmallVector<Constant *, 4> Idxs; // 0, 1, 0, 1.
+      for (unsigned i = 0; i != 4; ++i)
+        Idxs.push_back(Builder.getInt32(i & 1));
+      Rep = Builder.CreateShuffleVector(Load, UndefValue::get(Load->getType()),
+                                        ConstantVector::get(Idxs));
     } else if (Name == "llvm.x86.sse2.psll.dq") {
       // 128-bit shift left specified in bits.
       unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 8e8bdb60d2bbb0761068b904ab1533b8934778e6..6d3cfd5cebe2a6133c8225c6c9673716e806d184 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7834,11 +7834,6 @@ def VBROADCASTSDYrr  : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
                                       int_x86_avx2_vbroadcast_sd_pd_256,
                                       WriteFShuffle256>, VEX_L;
 
-let Predicates = [HasAVX2] in
-def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
-                                   int_x86_avx2_vbroadcasti128, WriteLoad>,
-                                   VEX_L;
-
 let Predicates = [HasAVX] in
 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
           (VBROADCASTF128 addr:$src)>;

diff --git a/test/CodeGen/X86/avx2-intrinsics-x86.ll b/test/CodeGen/X86/avx2-intrinsics-x86.ll
index da0f17a88dedf55f0b9857532008f7ef0ca0fcee..b7e2b7412ff6048d465914b81171ac0db21b15a9 100644 (file)
--- a/test/CodeGen/X86/avx2-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx2-intrinsics-x86.ll
@@ -641,13 +641,6 @@ define <4 x i64> @test_x86_avx2_pmul.dq(<8 x i32> %a0, <8 x i32> %a1) {
 declare <4 x i64> @llvm.x86.avx2.pmul.dq(<8 x i32>, <8 x i32>) nounwind readnone
 
 
-define <4 x i64> @test_x86_avx2_vbroadcasti128(i8* %a0) {
-  ; CHECK: vbroadcasti128
-  %res = call <4 x i64> @llvm.x86.avx2.vbroadcasti128(i8* %a0) ; <<4 x i64>> [#uses=1]
-  ret <4 x i64> %res
-}
-declare <4 x i64> @llvm.x86.avx2.vbroadcasti128(i8*) nounwind readonly
-
 define <4 x double> @test_x86_avx2_vbroadcast_sd_pd_256(<2 x double> %a0) {
   ; CHECK: vbroadcastsd
   %res = call <4 x double> @llvm.x86.avx2.vbroadcast.sd.pd.256(<2 x double> %a0) ; <<4 x double>> [#uses=1]

diff --git a/test/CodeGen/X86/x86-upgrade-avx2-vbroadcast.ll b/test/CodeGen/X86/x86-upgrade-avx2-vbroadcast.ll
new file mode 100644 (file)
index 0000000..08571f5
--- /dev/null
+++ b/test/CodeGen/X86/x86-upgrade-avx2-vbroadcast.ll
@@ -0,0 +1,18 @@
+; RUN: llc -mattr=+avx2 < %s | FileCheck %s
+
+; Check that we properly upgrade the AVX2 vbroadcast intrinsic to IR.
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.10.0"
+
+define <4 x i64> @broadcast128(<2 x i64> %src) {
+  CHECK-LABEL: broadcast128
+  CHECK:       vinsertf128 $1, %xmm0, %ymm0, %ymm0
+  %1 = alloca <2 x i64>, align 16
+  %2 = bitcast <2 x i64>* %1 to i8*
+  store <2 x i64> %src, <2 x i64>* %1, align 16
+  %3 = call <4 x i64> @llvm.x86.avx2.vbroadcasti128(i8* %2)
+  ret <4 x i64> %3
+}
+
+declare <4 x i64> @llvm.x86.avx2.vbroadcasti128(i8*) #1