[X86] Change the addcarry and subborrow intrinsics to return 2 results and remove...

We should represent the store directly in IR instead. This gives the middle end a chance to remove it if it can see a load from the same address.

Differential Revision: https://reviews.llvm.org/D51769

llvm-svn: 341677

diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index 7b915f7..c501f46 100644 (file)
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -2726,24 +2726,24 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
 // ADX
 
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_addcarryx_u32: GCCBuiltin<"__builtin_ia32_addcarryx_u32">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
-  def int_x86_addcarryx_u64: GCCBuiltin<"__builtin_ia32_addcarryx_u64">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
-  def int_x86_addcarry_u32: GCCBuiltin<"__builtin_ia32_addcarry_u32">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
-  def int_x86_addcarry_u64: GCCBuiltin<"__builtin_ia32_addcarry_u64">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
-  def int_x86_subborrow_u32: GCCBuiltin<"__builtin_ia32_subborrow_u32">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
-  def int_x86_subborrow_u64: GCCBuiltin<"__builtin_ia32_subborrow_u64">,
-        Intrinsic<[llvm_i8_ty], [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty,
-                                 llvm_ptr_ty], [IntrArgMemOnly]>;
+  def int_x86_addcarryx_u32:
+        Intrinsic<[llvm_i8_ty, llvm_i32_ty],
+                  [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_addcarryx_u64:
+        Intrinsic<[llvm_i8_ty, llvm_i64_ty],
+                  [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_addcarry_u32:
+        Intrinsic<[llvm_i8_ty, llvm_i32_ty],
+                  [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_addcarry_u64:
+        Intrinsic<[llvm_i8_ty, llvm_i64_ty],
+                  [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_subborrow_u32:
+        Intrinsic<[llvm_i8_ty, llvm_i32_ty],
+                  [llvm_i8_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_subborrow_u64:
+        Intrinsic<[llvm_i8_ty, llvm_i64_ty],
+                  [llvm_i8_ty, llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 06db01f..171f82e 100644 (file)
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -65,6 +65,17 @@ static bool UpgradeX86IntrinsicsWith8BitMask(Function *F, Intrinsic::ID IID,
   return true;
 }
 
+static bool UpgradeADCSBBIntrinsic(Function *F, Intrinsic::ID IID,
+                                   Function *&NewFn) {
+  // If this intrinsic has 3 operands, it's the new version.
+  if (F->getFunctionType()->getNumParams() == 3)
+    return false;
+
+  rename(F);
+  NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
+  return true;
+}
+
 static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
   // All of the intrinsics matches below should be marked with which llvm
   // version started autoupgrading them. At some point in the future we would
@@ -371,6 +382,19 @@ static bool UpgradeX86IntrinsicFunction(Function *F, StringRef Name,
     return true;
   }
 
+  if (Name == "addcarryx.u32")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_addcarryx_u32, NewFn);
+  if (Name == "addcarryx.u64")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_addcarryx_u64, NewFn);
+  if (Name == "addcarry.u32")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_addcarry_u32, NewFn);
+  if (Name == "addcarry.u64")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_addcarry_u64, NewFn);
+  if (Name == "subborrow.u32")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_subborrow_u32, NewFn);
+  if (Name == "subborrow.u64")
+    return UpgradeADCSBBIntrinsic(F, Intrinsic::x86_subborrow_u64, NewFn);
+
   // SSE4.1 ptest functions may have an old signature.
   if (Name.startswith("sse41.ptest")) { // Added in 3.2
     if (Name.substr(11) == "c")
@@ -3417,6 +3441,40 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
     break;
   }
 
+  case Intrinsic::x86_addcarryx_u32:
+  case Intrinsic::x86_addcarryx_u64:
+  case Intrinsic::x86_addcarry_u32:
+  case Intrinsic::x86_addcarry_u64:
+  case Intrinsic::x86_subborrow_u32:
+  case Intrinsic::x86_subborrow_u64: {
+    // This used to take 4 arguments. If we only have 3 arguments its already
+    // upgraded.
+    if (CI->getNumOperands() == 3)
+      return;
+
+    // Make a call with 3 operands.
+    NewCall = Builder.CreateCall(NewFn, { CI->getArgOperand(0),
+                                          CI->getArgOperand(1),
+                                          CI->getArgOperand(2)});
+    // Extract the second result and store it.
+    Value *Data = Builder.CreateExtractValue(NewCall, 1);
+    // Cast the pointer to the right type.
+    Value *Ptr = Builder.CreateBitCast(CI->getArgOperand(3),
+                                 llvm::PointerType::getUnqual(Data->getType()));
+    Builder.CreateAlignedStore(Data, Ptr, 1);
+    // Replace the original call result with the first result of the new call.
+    Value *CF = Builder.CreateExtractValue(NewCall, 0);
+
+    std::string Name = CI->getName();
+    if (!Name.empty()) {
+      CI->setName(Name + ".old");
+      NewCall->setName(Name);
+    }
+    CI->replaceAllUsesWith(CF);
+    CI->eraseFromParent();
+    return;
+  }
+
   case Intrinsic::x86_sse41_insertps:
   case Intrinsic::x86_sse41_dppd:
   case Intrinsic::x86_sse41_dpps:

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 5c5e7f0..fa8de5b 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -21286,6 +21286,18 @@ SDValue X86TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
       return DAG.getNode(IntrData->Opc0, dl, Op.getValueType(),
                          Op.getOperand(1), Op.getOperand(2), RoundingMode);
     }
+    // ADC/ADCX/SBB
+    case ADX: {
+      SDVTList CFVTs = DAG.getVTList(Op->getValueType(0), MVT::i32);
+      SDVTList VTs = DAG.getVTList(Op.getOperand(2).getValueType(), MVT::i32);
+      SDValue GenCF = DAG.getNode(X86ISD::ADD, dl, CFVTs, Op.getOperand(1),
+                                  DAG.getConstant(-1, dl, MVT::i8));
+      SDValue Res = DAG.getNode(IntrData->Opc0, dl, VTs, Op.getOperand(2),
+                                Op.getOperand(3), GenCF.getValue(1));
+      SDValue SetCC = getSETCC(X86::COND_B, Res.getValue(1), dl, DAG);
+      SDValue Results[] = { SetCC, Res };
+      return DAG.getMergeValues(Results, dl);
+    }
     default:
       break;
     }
@@ -21990,20 +22002,6 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget &Subtarget,
     return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(),
                        Ret, SDValue(InTrans.getNode(), 1));
   }
-  // ADC/ADCX/SBB
-  case ADX: {
-    SDVTList CFVTs = DAG.getVTList(Op->getValueType(0), MVT::i32);
-    SDVTList VTs = DAG.getVTList(Op.getOperand(3).getValueType(), MVT::i32);
-    SDValue GenCF = DAG.getNode(X86ISD::ADD, dl, CFVTs, Op.getOperand(2),
-                                DAG.getConstant(-1, dl, MVT::i8));
-    SDValue Res = DAG.getNode(IntrData->Opc0, dl, VTs, Op.getOperand(3),
-                              Op.getOperand(4), GenCF.getValue(1));
-    SDValue Store = DAG.getStore(Op.getOperand(0), dl, Res.getValue(0),
-                                 Op.getOperand(5), MachinePointerInfo());
-    SDValue SetCC = getSETCC(X86::COND_B, Res.getValue(1), dl, DAG);
-    SDValue Results[] = { SetCC, Store };
-    return DAG.getMergeValues(Results, dl);
-  }
   case TRUNCATE_TO_MEM_VI8:
   case TRUNCATE_TO_MEM_VI16:
   case TRUNCATE_TO_MEM_VI32: {

diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
index 668d575..84c7878 100644 (file)
--- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
@@ -64,11 +64,6 @@ struct IntrinsicData {
  * the alphabetical order.
  */
 static const IntrinsicData IntrinsicsWithChain[] = {
-  X86_INTRINSIC_DATA(addcarry_u32,  ADX, X86ISD::ADC, 0),
-  X86_INTRINSIC_DATA(addcarry_u64,  ADX, X86ISD::ADC, 0),
-  X86_INTRINSIC_DATA(addcarryx_u32, ADX, X86ISD::ADC, 0),
-  X86_INTRINSIC_DATA(addcarryx_u64, ADX, X86ISD::ADC, 0),
-
   X86_INTRINSIC_DATA(avx2_gather_d_d,      GATHER_AVX2, X86::VPGATHERDDrm, 0),
   X86_INTRINSIC_DATA(avx2_gather_d_d_256,  GATHER_AVX2, X86::VPGATHERDDYrm, 0),
   X86_INTRINSIC_DATA(avx2_gather_d_pd,     GATHER_AVX2, X86::VGATHERDPDrm, 0),
@@ -270,9 +265,6 @@ static const IntrinsicData IntrinsicsWithChain[] = {
   X86_INTRINSIC_DATA(rdseed_64, RDSEED, X86ISD::RDSEED, 0),
   X86_INTRINSIC_DATA(rdtsc,     RDTSC,  X86ISD::RDTSC_DAG, 0),
   X86_INTRINSIC_DATA(rdtscp,    RDTSC,  X86ISD::RDTSCP_DAG, 0),
-
-  X86_INTRINSIC_DATA(subborrow_u32, ADX, X86ISD::SBB, 0),
-  X86_INTRINSIC_DATA(subborrow_u64, ADX, X86ISD::SBB, 0),
   X86_INTRINSIC_DATA(xgetbv, XGETBV, X86::XGETBV, 0),
   X86_INTRINSIC_DATA(xtest,     XTEST,  X86ISD::XTEST,  0),
 };
@@ -294,6 +286,10 @@ static const IntrinsicData* getIntrinsicWithChain(unsigned IntNo) {
  * the alphabetical order.
  */
 static const IntrinsicData  IntrinsicsWithoutChain[] = {
+  X86_INTRINSIC_DATA(addcarry_u32,      ADX, X86ISD::ADC, 0),
+  X86_INTRINSIC_DATA(addcarry_u64,      ADX, X86ISD::ADC, 0),
+  X86_INTRINSIC_DATA(addcarryx_u32,     ADX, X86ISD::ADC, 0),
+  X86_INTRINSIC_DATA(addcarryx_u64,     ADX, X86ISD::ADC, 0),
   X86_INTRINSIC_DATA(avx_addsub_pd_256, INTR_TYPE_2OP, X86ISD::ADDSUB, 0),
   X86_INTRINSIC_DATA(avx_addsub_ps_256, INTR_TYPE_2OP, X86ISD::ADDSUB, 0),
   X86_INTRINSIC_DATA(avx_cmp_pd_256,    INTR_TYPE_3OP, X86ISD::CMPP, 0),
@@ -1225,6 +1221,8 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(ssse3_pmadd_ub_sw_128, INTR_TYPE_2OP, X86ISD::VPMADDUBSW, 0),
   X86_INTRINSIC_DATA(ssse3_pmul_hr_sw_128, INTR_TYPE_2OP, X86ISD::MULHRS, 0),
   X86_INTRINSIC_DATA(ssse3_pshuf_b_128, INTR_TYPE_2OP, X86ISD::PSHUFB, 0),
+  X86_INTRINSIC_DATA(subborrow_u32,     ADX, X86ISD::SBB, 0),
+  X86_INTRINSIC_DATA(subborrow_u64,     ADX, X86ISD::SBB, 0),
   X86_INTRINSIC_DATA(tbm_bextri_u32,    INTR_TYPE_2OP, X86ISD::BEXTR, 0),
   X86_INTRINSIC_DATA(tbm_bextri_u64,    INTR_TYPE_2OP, X86ISD::BEXTR, 0),
   X86_INTRINSIC_DATA(vcvtph2ps_128,     INTR_TYPE_1OP, X86ISD::CVTPH2PS, 0),

diff --git a/llvm/test/CodeGen/X86/adx-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/adx-intrinsics-upgrade.ll
new file mode 100644 (file)
index 0000000..62f3997
--- /dev/null
+++ b/llvm/test/CodeGen/X86/adx-intrinsics-upgrade.ll
@@ -0,0 +1,165 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7 --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=NOADX
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=broadwell --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=ADX
+
+declare i8 @llvm.x86.addcarryx.u32(i8, i32, i32, i8*)
+
+define i8 @test_addcarryx_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
+; NOADX-LABEL: test_addcarryx_u32:
+; NOADX:       ## %bb.0:
+; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; NOADX-NEXT:    adcl %edx, %esi ## encoding: [0x11,0xd6]
+; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
+; NOADX-NEXT:    retq ## encoding: [0xc3]
+;
+; ADX-LABEL: test_addcarryx_u32:
+; ADX:       ## %bb.0:
+; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; ADX-NEXT:    adcxl %edx, %esi ## encoding: [0x66,0x0f,0x38,0xf6,0xf2]
+; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
+; ADX-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.addcarryx.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+declare i8 @llvm.x86.addcarryx.u64(i8, i64, i64, i8*)
+
+define i8 @test_addcarryx_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
+; NOADX-LABEL: test_addcarryx_u64:
+; NOADX:       ## %bb.0:
+; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; NOADX-NEXT:    adcq %rdx, %rsi ## encoding: [0x48,0x11,0xd6]
+; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
+; NOADX-NEXT:    retq ## encoding: [0xc3]
+;
+; ADX-LABEL: test_addcarryx_u64:
+; ADX:       ## %bb.0:
+; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; ADX-NEXT:    adcxq %rdx, %rsi ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0xf2]
+; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
+; ADX-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.addcarryx.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+declare i8 @llvm.x86.addcarry.u32(i8, i32, i32, i8*)
+
+define i8 @test_addcarry_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
+; NOADX-LABEL: test_addcarry_u32:
+; NOADX:       ## %bb.0:
+; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; NOADX-NEXT:    adcl %edx, %esi ## encoding: [0x11,0xd6]
+; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
+; NOADX-NEXT:    retq ## encoding: [0xc3]
+;
+; ADX-LABEL: test_addcarry_u32:
+; ADX:       ## %bb.0:
+; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; ADX-NEXT:    adcxl %edx, %esi ## encoding: [0x66,0x0f,0x38,0xf6,0xf2]
+; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
+; ADX-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.addcarry.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+declare i8 @llvm.x86.addcarry.u64(i8, i64, i64, i8*)
+
+define i8 @test_addcarry_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
+; NOADX-LABEL: test_addcarry_u64:
+; NOADX:       ## %bb.0:
+; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; NOADX-NEXT:    adcq %rdx, %rsi ## encoding: [0x48,0x11,0xd6]
+; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
+; NOADX-NEXT:    retq ## encoding: [0xc3]
+;
+; ADX-LABEL: test_addcarry_u64:
+; ADX:       ## %bb.0:
+; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; ADX-NEXT:    adcxq %rdx, %rsi ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0xf2]
+; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
+; ADX-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.addcarry.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+declare i8 @llvm.x86.subborrow.u32(i8, i32, i32, i8*)
+
+define i8 @test_subborrow_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
+; CHECK-LABEL: test_subborrow_u32:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; CHECK-NEXT:    sbbl %edx, %esi ## encoding: [0x19,0xd6]
+; CHECK-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; CHECK-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.subborrow.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+declare i8 @llvm.x86.subborrow.u64(i8, i64, i64, i8*)
+
+define i8 @test_subborrow_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
+; CHECK-LABEL: test_subborrow_u64:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
+; CHECK-NEXT:    sbbq %rdx, %rsi ## encoding: [0x48,0x19,0xd6]
+; CHECK-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; CHECK-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %ret = tail call i8 @llvm.x86.subborrow.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
+  ret i8 %ret;
+}
+
+; Try a version with loads. Previously we crashed on this.
+define i32 @load_crash(i64* nocapture readonly %a, i64* nocapture readonly %b, i64* %res)  {
+; NOADX-LABEL: load_crash:
+; NOADX:       ## %bb.0:
+; NOADX-NEXT:    movq (%rdi), %rax ## encoding: [0x48,0x8b,0x07]
+; NOADX-NEXT:    xorl %ecx, %ecx ## encoding: [0x31,0xc9]
+; NOADX-NEXT:    addb $-1, %cl ## encoding: [0x80,0xc1,0xff]
+; NOADX-NEXT:    adcq (%rsi), %rax ## encoding: [0x48,0x13,0x06]
+; NOADX-NEXT:    setb %cl ## encoding: [0x0f,0x92,0xc1]
+; NOADX-NEXT:    movq %rax, (%rdx) ## encoding: [0x48,0x89,0x02]
+; NOADX-NEXT:    movzbl %cl, %eax ## encoding: [0x0f,0xb6,0xc1]
+; NOADX-NEXT:    retq ## encoding: [0xc3]
+;
+; ADX-LABEL: load_crash:
+; ADX:       ## %bb.0:
+; ADX-NEXT:    movq (%rdi), %rax ## encoding: [0x48,0x8b,0x07]
+; ADX-NEXT:    xorl %ecx, %ecx ## encoding: [0x31,0xc9]
+; ADX-NEXT:    addb $-1, %cl ## encoding: [0x80,0xc1,0xff]
+; ADX-NEXT:    adcxq (%rsi), %rax ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0x06]
+; ADX-NEXT:    setb %cl ## encoding: [0x0f,0x92,0xc1]
+; ADX-NEXT:    movq %rax, (%rdx) ## encoding: [0x48,0x89,0x02]
+; ADX-NEXT:    movzbl %cl, %eax ## encoding: [0x0f,0xb6,0xc1]
+; ADX-NEXT:    retq ## encoding: [0xc3]
+  %1 = load i64, i64* %a, align 8
+  %2 = load i64, i64* %b, align 8
+  %3 = bitcast i64* %res to i8*
+  %4 = tail call i8 @llvm.x86.addcarryx.u64(i8 0, i64 %1, i64 %2, i8* %3)
+  %conv = zext i8 %4 to i32
+  ret i32 %conv
+}
+
+; Try a really simple all zero input case, which also used to crash
+define void @allzeros() {
+; CHECK-LABEL: allzeros:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    xorl %eax, %eax ## encoding: [0x31,0xc0]
+; CHECK-NEXT:    addb $-1, %al ## encoding: [0x04,0xff]
+; CHECK-NEXT:    sbbq %rax, %rax ## encoding: [0x48,0x19,0xc0]
+; CHECK-NEXT:    andl $1, %eax ## encoding: [0x83,0xe0,0x01]
+; CHECK-NEXT:    movq %rax, 0 ## encoding: [0x48,0x89,0x04,0x25,0x00,0x00,0x00,0x00]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call i8 @llvm.x86.addcarryx.u64(i8 0, i64 0, i64 0, i8* null)
+  ret void
+}

diff --git a/llvm/test/CodeGen/X86/adx-intrinsics.ll b/llvm/test/CodeGen/X86/adx-intrinsics.ll
index bc8e2be..1d166e4 100644 (file)
--- a/llvm/test/CodeGen/X86/adx-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/adx-intrinsics.ll
@@ -2,120 +2,144 @@
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7 --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=NOADX
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=broadwell --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=ADX
 
-declare i8 @llvm.x86.addcarryx.u32(i8, i32, i32, i8*)
+declare { i8, i32 } @llvm.x86.addcarryx.u32(i8, i32, i32)
 
 define i8 @test_addcarryx_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
 ; NOADX-LABEL: test_addcarryx_u32:
 ; NOADX:       ## %bb.0:
 ; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; NOADX-NEXT:    adcl %edx, %esi ## encoding: [0x11,0xd6]
-; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; NOADX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; ADX-LABEL: test_addcarryx_u32:
 ; ADX:       ## %bb.0:
 ; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; ADX-NEXT:    adcxl %edx, %esi ## encoding: [0x66,0x0f,0x38,0xf6,0xf2]
-; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; ADX-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.addcarryx.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i32 } @llvm.x86.addcarryx.u32(i8 %c, i32 %a, i32 %b)
+  %1 = extractvalue { i8, i32 } %ret, 1
+  %2 = bitcast i8* %ptr to i32*
+  store i32 %1, i32* %2, align 1
+  %3 = extractvalue { i8, i32 } %ret, 0
+  ret i8 %3
 }
 
-declare i8 @llvm.x86.addcarryx.u64(i8, i64, i64, i8*)
+declare { i8, i64 } @llvm.x86.addcarryx.u64(i8, i64, i64)
 
 define i8 @test_addcarryx_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
 ; NOADX-LABEL: test_addcarryx_u64:
 ; NOADX:       ## %bb.0:
 ; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; NOADX-NEXT:    adcq %rdx, %rsi ## encoding: [0x48,0x11,0xd6]
-; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; NOADX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; ADX-LABEL: test_addcarryx_u64:
 ; ADX:       ## %bb.0:
 ; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; ADX-NEXT:    adcxq %rdx, %rsi ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0xf2]
-; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; ADX-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.addcarryx.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i64 } @llvm.x86.addcarryx.u64(i8 %c, i64 %a, i64 %b)
+  %1 = extractvalue { i8, i64 } %ret, 1
+  %2 = bitcast i8* %ptr to i64*
+  store i64 %1, i64* %2, align 1
+  %3 = extractvalue { i8, i64 } %ret, 0
+  ret i8 %3
 }
 
-declare i8 @llvm.x86.addcarry.u32(i8, i32, i32, i8*)
+declare { i8, i32 } @llvm.x86.addcarry.u32(i8, i32, i32)
 
 define i8 @test_addcarry_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
 ; NOADX-LABEL: test_addcarry_u32:
 ; NOADX:       ## %bb.0:
 ; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; NOADX-NEXT:    adcl %edx, %esi ## encoding: [0x11,0xd6]
-; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; NOADX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; ADX-LABEL: test_addcarry_u32:
 ; ADX:       ## %bb.0:
 ; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; ADX-NEXT:    adcxl %edx, %esi ## encoding: [0x66,0x0f,0x38,0xf6,0xf2]
-; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; ADX-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.addcarry.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i32 } @llvm.x86.addcarry.u32(i8 %c, i32 %a, i32 %b)
+  %1 = extractvalue { i8, i32 } %ret, 1
+  %2 = bitcast i8* %ptr to i32*
+  store i32 %1, i32* %2, align 1
+  %3 = extractvalue { i8, i32 } %ret, 0
+  ret i8 %3
 }
 
-declare i8 @llvm.x86.addcarry.u64(i8, i64, i64, i8*)
+declare { i8, i64 } @llvm.x86.addcarry.u64(i8, i64, i64)
 
 define i8 @test_addcarry_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
 ; NOADX-LABEL: test_addcarry_u64:
 ; NOADX:       ## %bb.0:
 ; NOADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; NOADX-NEXT:    adcq %rdx, %rsi ## encoding: [0x48,0x11,0xd6]
-; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; NOADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; NOADX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; ADX-LABEL: test_addcarry_u64:
 ; ADX:       ## %bb.0:
 ; ADX-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; ADX-NEXT:    adcxq %rdx, %rsi ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0xf2]
-; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; ADX-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; ADX-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.addcarry.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i64 } @llvm.x86.addcarry.u64(i8 %c, i64 %a, i64 %b)
+  %1 = extractvalue { i8, i64 } %ret, 1
+  %2 = bitcast i8* %ptr to i64*
+  store i64 %1, i64* %2, align 1
+  %3 = extractvalue { i8, i64 } %ret, 0
+  ret i8 %3
 }
 
-declare i8 @llvm.x86.subborrow.u32(i8, i32, i32, i8*)
+declare { i8, i32 } @llvm.x86.subborrow.u32(i8, i32, i32)
 
 define i8 @test_subborrow_u32(i8 %c, i32 %a, i32 %b, i8* %ptr) {
 ; CHECK-LABEL: test_subborrow_u32:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; CHECK-NEXT:    sbbl %edx, %esi ## encoding: [0x19,0xd6]
-; CHECK-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; CHECK-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; CHECK-NEXT:    movl %esi, (%rcx) ## encoding: [0x89,0x31]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.subborrow.u32(i8 %c, i32 %a, i32 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i32 } @llvm.x86.subborrow.u32(i8 %c, i32 %a, i32 %b)
+  %1 = extractvalue { i8, i32 } %ret, 1
+  %2 = bitcast i8* %ptr to i32*
+  store i32 %1, i32* %2, align 1
+  %3 = extractvalue { i8, i32 } %ret, 0
+  ret i8 %3
 }
 
-declare i8 @llvm.x86.subborrow.u64(i8, i64, i64, i8*)
+declare { i8, i64 } @llvm.x86.subborrow.u64(i8, i64, i64)
 
 define i8 @test_subborrow_u64(i8 %c, i64 %a, i64 %b, i8* %ptr) {
 ; CHECK-LABEL: test_subborrow_u64:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    addb $-1, %dil ## encoding: [0x40,0x80,0xc7,0xff]
 ; CHECK-NEXT:    sbbq %rdx, %rsi ## encoding: [0x48,0x19,0xd6]
-; CHECK-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; CHECK-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
+; CHECK-NEXT:    movq %rsi, (%rcx) ## encoding: [0x48,0x89,0x31]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
-  %ret = tail call i8 @llvm.x86.subborrow.u64(i8 %c, i64 %a, i64 %b, i8* %ptr)
-  ret i8 %ret;
+  %ret = call { i8, i64 } @llvm.x86.subborrow.u64(i8 %c, i64 %a, i64 %b)
+  %1 = extractvalue { i8, i64 } %ret, 1
+  %2 = bitcast i8* %ptr to i64*
+  store i64 %1, i64* %2, align 1
+  %3 = extractvalue { i8, i64 } %ret, 0
+  ret i8 %3
 }
 
 ; Try a version with loads. Previously we crashed on this.
@@ -126,9 +150,9 @@ define i32 @load_crash(i64* nocapture readonly %a, i64* nocapture readonly %b, i
 ; NOADX-NEXT:    xorl %ecx, %ecx ## encoding: [0x31,0xc9]
 ; NOADX-NEXT:    addb $-1, %cl ## encoding: [0x80,0xc1,0xff]
 ; NOADX-NEXT:    adcq (%rsi), %rax ## encoding: [0x48,0x13,0x06]
+; NOADX-NEXT:    setb %cl ## encoding: [0x0f,0x92,0xc1]
 ; NOADX-NEXT:    movq %rax, (%rdx) ## encoding: [0x48,0x89,0x02]
-; NOADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
-; NOADX-NEXT:    movzbl %al, %eax ## encoding: [0x0f,0xb6,0xc0]
+; NOADX-NEXT:    movzbl %cl, %eax ## encoding: [0x0f,0xb6,0xc1]
 ; NOADX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; ADX-LABEL: load_crash:
@@ -137,15 +161,19 @@ define i32 @load_crash(i64* nocapture readonly %a, i64* nocapture readonly %b, i
 ; ADX-NEXT:    xorl %ecx, %ecx ## encoding: [0x31,0xc9]
 ; ADX-NEXT:    addb $-1, %cl ## encoding: [0x80,0xc1,0xff]
 ; ADX-NEXT:    adcxq (%rsi), %rax ## encoding: [0x66,0x48,0x0f,0x38,0xf6,0x06]
+; ADX-NEXT:    setb %cl ## encoding: [0x0f,0x92,0xc1]
 ; ADX-NEXT:    movq %rax, (%rdx) ## encoding: [0x48,0x89,0x02]
-; ADX-NEXT:    setb %al ## encoding: [0x0f,0x92,0xc0]
-; ADX-NEXT:    movzbl %al, %eax ## encoding: [0x0f,0xb6,0xc0]
+; ADX-NEXT:    movzbl %cl, %eax ## encoding: [0x0f,0xb6,0xc1]
 ; ADX-NEXT:    retq ## encoding: [0xc3]
   %1 = load i64, i64* %a, align 8
   %2 = load i64, i64* %b, align 8
   %3 = bitcast i64* %res to i8*
-  %4 = tail call i8 @llvm.x86.addcarryx.u64(i8 0, i64 %1, i64 %2, i8* %3)
-  %conv = zext i8 %4 to i32
+  %4 = call { i8, i64 } @llvm.x86.addcarryx.u64(i8 0, i64 %1, i64 %2)
+  %5 = extractvalue { i8, i64 } %4, 1
+  %6 = bitcast i8* %3 to i64*
+  store i64 %5, i64* %6, align 1
+  %7 = extractvalue { i8, i64 } %4, 0
+  %conv = zext i8 %7 to i32
   ret i32 %conv
 }
 
@@ -160,6 +188,9 @@ define void @allzeros() {
 ; CHECK-NEXT:    movq %rax, 0 ## encoding: [0x48,0x89,0x04,0x25,0x00,0x00,0x00,0x00]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
 entry:
-  %0 = tail call i8 @llvm.x86.addcarryx.u64(i8 0, i64 0, i64 0, i8* null)
+  %0 = call { i8, i64 } @llvm.x86.addcarryx.u64(i8 0, i64 0, i64 0)
+  %1 = extractvalue { i8, i64 } %0, 1
+  store i64 %1, i64* null, align 1
+  %2 = extractvalue { i8, i64 } %0, 0
   ret void
 }