From 23356d4d64d4511d340bc6dad1464c8a3c65473b Mon Sep 17 00:00:00 2001
From: Adam Nemet <anemet@apple.com>
Date: Thu, 29 May 2014 23:35:33 +0000
Subject: [PATCH] [X86] Auto-upgrade AVX1 vbroadcast intrinsics

They are replaced with the same IR that is generated for the
vector-initializers in avxintrin.h.

The test verifies that we get back the original instruction.  I haven't seen
this approach to be used in other auto-upgrade tests (i.e. llc + FileCheck)
but I think it's the most direct way to test this case.  I believe this should
work because llc upgrades calls during parsing.  (Other tests mostly check
that assembling and disassembling yields the upgraded IR.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209863 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/IR/AutoUpgrade.cpp                     | 16 +++++++++
 test/Bitcode/x86-upgrade-avx-vbroadcast.ll | 41 ++++++++++++++++++++++
 2 files changed, 57 insertions(+)
 create mode 100644 test/Bitcode/x86-upgrade-avx-vbroadcast.ll

diff --git a/lib/IR/AutoUpgrade.cpp b/lib/IR/AutoUpgrade.cpp
index e255113aab7..05b3745ab0f 100644
--- a/lib/IR/AutoUpgrade.cpp
+++ b/lib/IR/AutoUpgrade.cpp
@@ -114,6 +114,9 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
         Name == "x86.avx.movnt.pd.256" ||
         Name == "x86.avx.movnt.ps.256" ||
         Name == "x86.sse42.crc32.64.8" ||
+        Name == "x86.avx.vbroadcast.ss" ||
+        Name == "x86.avx.vbroadcast.ss.256" ||
+        Name == "x86.avx.vbroadcast.sd.256" ||
         (Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
       NewFn = nullptr;
       return true;
@@ -335,6 +338,19 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
       Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
       Rep = Builder.CreateCall2(CRC32, Trunc0, CI->getArgOperand(1));
       Rep = Builder.CreateZExt(Rep, CI->getType(), "");
+    } else if (Name.startswith("llvm.x86.avx.vbroadcast")) {
+      // Replace broadcasts with a series of insertelements.
+      Type *VecTy = CI->getType();
+      Type *EltTy = VecTy->getVectorElementType();
+      unsigned EltNum = VecTy->getVectorNumElements();
+      Value *Cast = Builder.CreateBitCast(CI->getArgOperand(0),
+                                          EltTy->getPointerTo());
+      Value *Load = Builder.CreateLoad(Cast);
+      Type *I32Ty = Type::getInt32Ty(C);
+      Rep = UndefValue::get(VecTy);
+      for (unsigned I = 0; I < EltNum; ++I)
+        Rep = Builder.CreateInsertElement(Rep, Load,
+                                          ConstantInt::get(I32Ty, I));
     } else {
       bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
       if (Name == "llvm.x86.avx.vpermil.pd.256")
diff --git a/test/Bitcode/x86-upgrade-avx-vbroadcast.ll b/test/Bitcode/x86-upgrade-avx-vbroadcast.ll
new file mode 100644
index 00000000000..d885f1cd364
--- /dev/null
+++ b/test/Bitcode/x86-upgrade-avx-vbroadcast.ll
@@ -0,0 +1,41 @@
+; RUN: llc -mattr=+avx < %s | FileCheck %s
+
+; Check that we properly upgrade the AVX vbroadcast intrinsics to IR.  The
+; expectation is that we should still get the original instruction back that
+; maps to the intrinsic.
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.9.0"
+
+; CHECK-LABEL: test_mm_broadcast_ss:
+define <4 x float> @test_mm_broadcast_ss(float* readonly %__a){
+entry:
+  %0 = bitcast float* %__a to i8*
+; CHECK: vbroadcastss (%{{.*}}), %xmm
+  %1 = tail call <4 x float> @llvm.x86.avx.vbroadcast.ss(i8* %0)
+  ret <4 x float> %1
+}
+
+; CHECK-LABEL: test_mm256_broadcast_sd:
+define <4 x double> @test_mm256_broadcast_sd(double* readonly %__a) {
+entry:
+  %0 = bitcast double* %__a to i8*
+; CHECK: vbroadcastsd (%{{.*}}), %ymm
+  %1 = tail call <4 x double> @llvm.x86.avx.vbroadcast.sd.256(i8* %0)
+  ret <4 x double> %1
+}
+
+; CHECK-LABEL: test_mm256_broadcast_ss:
+define <8 x float> @test_mm256_broadcast_ss(float* readonly %__a) {
+entry:
+  %0 = bitcast float* %__a to i8*
+; CHECK: vbroadcastss (%{{.*}}), %ymm
+  %1 = tail call <8 x float> @llvm.x86.avx.vbroadcast.ss.256(i8* %0)
+  ret <8 x float> %1
+}
+
+declare <8 x float> @llvm.x86.avx.vbroadcast.ss.256(i8*)
+
+declare <4 x double> @llvm.x86.avx.vbroadcast.sd.256(i8*)
+
+declare <4 x float> @llvm.x86.avx.vbroadcast.ss(i8*)
-- 
2.34.1