WebAssembly: generate load/store

Summary: This handles all load/store operations that WebAssembly defines, and handles those necessary for C++ such as i1. I left a FIXME for outstanding features which aren't required for now.

Reviewers: sunfish

Subscribers: jfb, llvm-commits, dschuff
llvm-svn: 246500

lib/Target/WebAssembly/WebAssemblyAsmPrinter.cpp

@@ -40,6 +40,7 @@ using namespace llvm;
 namespace {
 
 class WebAssemblyAsmPrinter final : public AsmPrinter {
+  bool hasAddr64;
   const WebAssemblyInstrInfo *TII;
 
 public:
@@ -60,7 +61,9 @@ private:
   }
 
   bool runOnMachineFunction(MachineFunction &MF) override {
-    TII = MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo();
+    const auto &Subtarget = MF.getSubtarget<WebAssemblySubtarget>();
+    hasAddr64 = Subtarget.hasAddr64();
+    TII = Subtarget.getInstrInfo();
     return AsmPrinter::runOnMachineFunction(MF);
   }
 
@@ -97,23 +100,6 @@ static SmallString<32> OpcodeName(const WebAssemblyInstrInfo *TII,
 
 static std::string toSymbol(StringRef S) { return ("$" + S).str(); }
 
-static const char *toString(const Type *Ty) {
-  switch (Ty->getTypeID()) {
-  default: break;
-  case Type::FloatTyID:  return "f32";
-  case Type::DoubleTyID: return "f64";
-  case Type::IntegerTyID:
-    switch (Ty->getIntegerBitWidth()) {
-    case 32: return "i32";
-    case 64: return "i64";
-    default: break;
-    }
-  }
-  DEBUG(dbgs() << "Invalid type "; Ty->print(dbgs()); dbgs() << '\n');
-  llvm_unreachable("invalid type");
-  return "<invalid>";
-}
-
 static std::string toString(const APFloat &FP) {
   static const size_t BufBytes = 128;
   char buf[BufBytes];
@@ -131,6 +117,28 @@ static std::string toString(const APFloat &FP) {
   return buf;
 }
 
+static const char *toString(const Type *Ty, bool hasAddr64) {
+  switch (Ty->getTypeID()) {
+  default: break;
+  // Treat all pointers as the underlying integer into linear memory.
+  case Type::PointerTyID: return hasAddr64 ? "i64" : "i32";
+  case Type::FloatTyID:  return "f32";
+  case Type::DoubleTyID: return "f64";
+  case Type::IntegerTyID:
+    switch (Ty->getIntegerBitWidth()) {
+    case 8: return "i8";
+    case 16: return "i16";
+    case 32: return "i32";
+    case 64: return "i64";
+    default: break;
+    }
+  }
+  DEBUG(dbgs() << "Invalid type "; Ty->print(dbgs()); dbgs() << '\n');
+  llvm_unreachable("invalid type");
+  return "<invalid>";
+}
+
+
 //===----------------------------------------------------------------------===//
 // WebAssemblyAsmPrinter Implementation.
 //===----------------------------------------------------------------------===//
@@ -186,7 +194,8 @@ void WebAssemblyAsmPrinter::EmitGlobalVariable(const GlobalVariable *GV) {
     return;
   }
 
-  OS << "(global " << toSymbol(Name) << ' ' << toString(Init->getType()) << ' ';
+  OS << "(global " << toSymbol(Name) << ' '
+     << toString(Init->getType(), hasAddr64) << ' ';
   if (const auto *C = dyn_cast<ConstantInt>(Init)) {
     assert(C->getBitWidth() <= 64 && "Printing wider types unimplemented");
     OS << C->getZExtValue();
@@ -228,10 +237,10 @@ void WebAssemblyAsmPrinter::EmitFunctionBodyStart() {
   raw_svector_ostream OS(Str);
   const Function *F = MF->getFunction();
   for (const Argument &A : F->args())
-    OS << " (param " << toString(A.getType()) << ')';
+    OS << " (param " << toString(A.getType(), hasAddr64) << ')';
   const Type *Rt = F->getReturnType();
   if (!Rt->isVoidTy())
-    OS << " (result " << toString(Rt) << ')';
+    OS << " (result " << toString(Rt, hasAddr64) << ')';
   OS << '\n';
   OutStreamer->EmitRawText(OS.str());
 }

lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

@@ -113,8 +113,6 @@ WebAssemblyTargetLowering::WebAssemblyTargetLowering(
   // Compute derived properties from the register classes.
   computeRegisterProperties(Subtarget->getRegisterInfo());
 
-  // FIXME: many setOperationAction are missing...
-
   setOperationAction(ISD::GlobalAddress, MVTPtr, Custom);
 
   for (auto T : {MVT::f32, MVT::f64}) {
@@ -154,6 +152,16 @@ WebAssemblyTargetLowering::WebAssemblyTargetLowering(
   setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
   setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
   setOperationAction(ISD::DYNAMIC_STACKALLOC, MVTPtr, Expand);
+
+  // WebAssembly doesn't have:
+  //  - Floating-point extending loads.
+  //  - Floating-point truncating stores.
+  //  - i1 extending loads.
+  setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f64, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+  for (auto T : MVT::integer_valuetypes())
+    for (auto Ext : {ISD::EXTLOAD, ISD::ZEXTLOAD, ISD::SEXTLOAD})
+      setLoadExtAction(Ext, T, MVT::i1, Promote);
 }
 
 FastISel *WebAssemblyTargetLowering::createFastISel(

lib/Target/WebAssembly/WebAssemblyInstrMemory.td

@@ -14,37 +14,84 @@
 
 /*
  * TODO(jfb): Add the following.
- * Each has optional alignment and immediate byte offset.
- *
- * int32.load_sx[int8]: sign-extend to int32
- * int32.load_sx[int16]: sign-extend to int32
- * int32.load_zx[int8]: zero-extend to int32
- * int32.load_zx[int16]: zero-extend to int32
- * int32.load[int32]: (no conversion)
- * int64.load_sx[int8]: sign-extend to int64
- * int64.load_sx[int16]: sign-extend to int64
- * int64.load_sx[int32]: sign-extend to int64
- * int64.load_zx[int8]: zero-extend to int64
- * int64.load_zx[int16]: zero-extend to int64
- * int64.load_zx[int32]: zero-extend to int64
- * int64.load[int64]: (no conversion)
- * float32.load[float32]: (no conversion)
- * float64.load[float64]: (no conversion)
- * 
- * int32.store[int8]: wrap int32 to int8
- * int32.store[int16]: wrap int32 to int16
- * int32.store[int32]: (no conversion)
- * int64.store[int8]: wrap int64 to int8
- * int64.store[int16]: wrap int64 to int16
- * int64.store[int32]: wrap int64 to int32
- * int64.store[int64]: (no conversion)
- * float32.store[float32]: (no conversion)
- * float64.store[float64]: (no conversion)
  * 
  * load_global: load the value of a given global variable
  * store_global: store a given value to a given global variable
  */
 
+// FIXME:
+//  - HasAddr64
+//  - WebAssemblyTargetLowering::isLegalAddressingMode
+//  - WebAssemblyTargetLowering having to do with atomics
+//  - Each has optional alignment and immediate byte offset.
+
+// WebAssembly has i8/i16/i32/i64/f32/f64 memory types, but doesn't have i8/i16
+// local types. These memory-only types instead zero- or sign-extend into local
+// types when loading, and truncate when storing.
+
+// Basic load.
+def LOAD_I32_ : I<(outs Int32:$dst), (ins Int32:$addr),
+                  [(set Int32:$dst, (load Int32:$addr))]>;
+def LOAD_I64_ : I<(outs Int64:$dst), (ins Int32:$addr),
+                  [(set Int64:$dst, (load Int32:$addr))]>;
+def LOAD_F32_ : I<(outs Float32:$dst), (ins Int32:$addr),
+                  [(set Float32:$dst, (load Int32:$addr))]>;
+def LOAD_F64_ : I<(outs Float64:$dst), (ins Int32:$addr),
+                  [(set Float64:$dst, (load Int32:$addr))]>;
+
+// Extending load.
+def LOAD_SX_I8_I32_  : I<(outs Int32:$dst), (ins Int32:$addr),
+                         [(set Int32:$dst, (sextloadi8 Int32:$addr))]>;
+def LOAD_ZX_I8_I32_  : I<(outs Int32:$dst), (ins Int32:$addr),
+                         [(set Int32:$dst, (zextloadi8 Int32:$addr))]>;
+def LOAD_SX_I16_I32_ : I<(outs Int32:$dst), (ins Int32:$addr),
+                         [(set Int32:$dst, (sextloadi16 Int32:$addr))]>;
+def LOAD_ZX_I16_I32_ : I<(outs Int32:$dst), (ins Int32:$addr),
+                         [(set Int32:$dst, (zextloadi16 Int32:$addr))]>;
+def LOAD_SX_I8_I64_  : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (sextloadi8 Int32:$addr))]>;
+def LOAD_ZX_I8_I64_  : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (zextloadi8 Int32:$addr))]>;
+def LOAD_SX_I16_I64_ : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (sextloadi16 Int32:$addr))]>;
+def LOAD_ZX_I16_I64_ : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (zextloadi16 Int32:$addr))]>;
+def LOAD_SX_I32_I64_ : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (sextloadi32 Int32:$addr))]>;
+def LOAD_ZX_I32_I64_ : I<(outs Int64:$dst), (ins Int32:$addr),
+                         [(set Int64:$dst, (zextloadi32 Int32:$addr))]>;
+
+// "Don't care" extending load become zero-extending load.
+def : Pat<(i32 (extloadi8 Int32:$addr)),  (LOAD_ZX_I8_I32_ $addr)>;
+def : Pat<(i32 (extloadi16 Int32:$addr)), (LOAD_ZX_I16_I32_ $addr)>;
+def : Pat<(i64 (extloadi8 Int32:$addr)),  (LOAD_ZX_I8_I64_ $addr)>;
+def : Pat<(i64 (extloadi16 Int32:$addr)), (LOAD_ZX_I16_I64_ $addr)>;
+def : Pat<(i64 (extloadi32 Int32:$addr)), (LOAD_ZX_I32_I64_ $addr)>;
+
+// Basic store.
+// Note: WebAssembly inverts SelectionDAG's usual operand order.
+def STORE_I32_  : I<(outs), (ins Int32:$addr, Int32:$val),
+                    [(store Int32:$val, Int32:$addr)]>;
+def STORE_I64_  : I<(outs), (ins Int32:$addr, Int64:$val),
+                    [(store Int64:$val, Int32:$addr)]>;
+def STORE_F32_  : I<(outs), (ins Int32:$addr, Float32:$val),
+                    [(store Float32:$val, Int32:$addr)]>;
+def STORE_F64_  : I<(outs), (ins Int32:$addr, Float64:$val),
+                    [(store Float64:$val, Int32:$addr)]>;
+
+// Truncating store.
+def STORE_I8_I32  : I<(outs), (ins Int32:$addr, Int32:$val),
+                      [(truncstorei8 Int32:$val, Int32:$addr)]>;
+def STORE_I16_I32 : I<(outs), (ins Int32:$addr, Int32:$val),
+                      [(truncstorei16 Int32:$val, Int32:$addr)]>;
+def STORE_I8_I64  : I<(outs), (ins Int32:$addr, Int64:$val),
+                      [(truncstorei8 Int64:$val, Int32:$addr)]>;
+def STORE_I16_I64 : I<(outs), (ins Int32:$addr, Int64:$val),
+                      [(truncstorei16 Int64:$val, Int32:$addr)]>;
+def STORE_I32_I64 : I<(outs), (ins Int32:$addr, Int64:$val),
+                      [(truncstorei32 Int64:$val, Int32:$addr)]>;
+
+// Page size.
 def page_size_I32 : I<(outs Int32:$dst), (ins),
                       [(set Int32:$dst, (int_wasm_page_size))]>,
                     Requires<[HasAddr32]>;

test/CodeGen/WebAssembly/load-ext.ll

@@ -0,0 +1,86 @@
+; RUN: llc < %s -asm-verbose=false | FileCheck %s
+
+; Test that extending loads are assembled properly.
+
+target datalayout = "e-p:32:32-i64:64-v128:8:128-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: (func $sext_i8_i32
+; CHECK: (setlocal @1 (load_sx_i8_i32 @0))
+define i32 @sext_i8_i32(i8 *%p) {
+  %v = load i8, i8* %p
+  %e = sext i8 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $zext_i8_i32
+; CHECK: (setlocal @1 (load_zx_i8_i32 @0))
+define i32 @zext_i8_i32(i8 *%p) {
+  %v = load i8, i8* %p
+  %e = zext i8 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $sext_i16_i32
+; CHECK: (setlocal @1 (load_sx_i16_i32 @0))
+define i32 @sext_i16_i32(i16 *%p) {
+  %v = load i16, i16* %p
+  %e = sext i16 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $zext_i16_i32
+; CHECK: (setlocal @1 (load_zx_i16_i32 @0))
+define i32 @zext_i16_i32(i16 *%p) {
+  %v = load i16, i16* %p
+  %e = zext i16 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $sext_i8_i64
+; CHECK: (setlocal @1 (load_sx_i8_i64 @0))
+define i64 @sext_i8_i64(i8 *%p) {
+  %v = load i8, i8* %p
+  %e = sext i8 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $zext_i8_i64
+; CHECK: (setlocal @1 (load_zx_i8_i64 @0))
+define i64 @zext_i8_i64(i8 *%p) {
+  %v = load i8, i8* %p
+  %e = zext i8 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $sext_i16_i64
+; CHECK: (setlocal @1 (load_sx_i16_i64 @0))
+define i64 @sext_i16_i64(i16 *%p) {
+  %v = load i16, i16* %p
+  %e = sext i16 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $zext_i16_i64
+; CHECK: (setlocal @1 (load_zx_i16_i64 @0))
+define i64 @zext_i16_i64(i16 *%p) {
+  %v = load i16, i16* %p
+  %e = zext i16 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $sext_i32_i64
+; CHECK: (setlocal @1 (load_sx_i32_i64 @0))
+define i64 @sext_i32_i64(i32 *%p) {
+  %v = load i32, i32* %p
+  %e = sext i32 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $zext_i32_i64
+; CHECK: (setlocal @1 (load_zx_i32_i64 @0))
+define i64 @zext_i32_i64(i32 *%p) {
+  %v = load i32, i32* %p
+  %e = zext i32 %v to i64
+  ret i64 %e
+}

test/CodeGen/WebAssembly/load-store-i1.ll

@@ -0,0 +1,68 @@
+; RUN: llc < %s -asm-verbose=false | FileCheck %s
+
+; Test that i1 extending loads and truncating stores are assembled properly.
+
+target datalayout = "e-p:32:32-i64:64-v128:8:128-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: (func $load_unsigned_i1_i32
+; CHECK:      (setlocal @1 (load_zx_i8_i32 @0))
+; CHECK-NEXT: (return @1)
+define i32 @load_unsigned_i1_i32(i1* %p) {
+  %v = load i1, i1* %p
+  %e = zext i1 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $load_signed_i1_i32
+; CHECK:      (setlocal @1 (load_zx_i8_i32 @0))
+; CHECK-NEXT: (setlocal @2 (immediate 31))
+; CHECK-NEXT: (setlocal @3 (shl @1 @2))
+; CHECK-NEXT: (setlocal @4 (sar @3 @2))
+; CHECK-NEXT: (return @4)
+define i32 @load_signed_i1_i32(i1* %p) {
+  %v = load i1, i1* %p
+  %e = sext i1 %v to i32
+  ret i32 %e
+}
+
+; CHECK-LABEL: (func $load_unsigned_i1_i64
+; CHECK:      (setlocal @1 (load_zx_i8_i64 @0))
+; CHECK-NEXT: (return @1)
+define i64 @load_unsigned_i1_i64(i1* %p) {
+  %v = load i1, i1* %p
+  %e = zext i1 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $load_signed_i1_i64
+; CHECK:      (setlocal @1 (load_zx_i8_i64 @0))
+; CHECK-NEXT: (setlocal @2 (immediate 63))
+; CHECK-NEXT: (setlocal @3 (shl @1 @2))
+; CHECK-NEXT: (setlocal @4 (sar @3 @2))
+; CHECK-NEXT: (return @4)
+define i64 @load_signed_i1_i64(i1* %p) {
+  %v = load i1, i1* %p
+  %e = sext i1 %v to i64
+  ret i64 %e
+}
+
+; CHECK-LABEL: (func $store_i32_i1
+; CHECK:      (setlocal @2 (immediate 1))
+; CHECK-NEXT: (setlocal @3 (and @1 @2))
+; CHECK-NEXT: (store_i8 @0 @3)
+define void @store_i32_i1(i1* %p, i32 %v) {
+  %t = trunc i32 %v to i1
+  store i1 %t, i1* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $store_i64_i1
+; CHECK:      (setlocal @2 (immediate 1))
+; CHECK-NEXT: (setlocal @3 (and @1 @2))
+; CHECK-NEXT: (store_i8 @0 @3)
+define void @store_i64_i1(i1* %p, i64 %v) {
+  %t = trunc i64 %v to i1
+  store i1 %t, i1* %p
+  ret void
+}

test/CodeGen/WebAssembly/load.ll

@@ -0,0 +1,46 @@
+; RUN: llc < %s -asm-verbose=false | FileCheck %s
+
+; Test that basic loads are assembled properly.
+
+target datalayout = "e-p:32:32-i64:64-v128:8:128-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: (func $ldi32
+; CHECK-NEXT: (param i32) (result i32)
+; CHECK-NEXT: (setlocal @0 (argument 0))
+; CHECK-NEXT: (setlocal @1 (load_i32 @0))
+; CHECK-NEXT: (return @1)
+define i32 @ldi32(i32 *%p) {
+  %v = load i32, i32* %p
+  ret i32 %v
+}
+
+; CHECK-LABEL: (func $ldi64
+; CHECK-NEXT: (param i32) (result i64)
+; CHECK-NEXT: (setlocal @0 (argument 0))
+; CHECK-NEXT: (setlocal @1 (load_i64 @0))
+; CHECK-NEXT: (return @1)
+define i64 @ldi64(i64 *%p) {
+  %v = load i64, i64* %p
+  ret i64 %v
+}
+
+; CHECK-LABEL: (func $ldf32
+; CHECK-NEXT: (param i32) (result f32)
+; CHECK-NEXT: (setlocal @0 (argument 0))
+; CHECK-NEXT: (setlocal @1 (load_f32 @0))
+; CHECK-NEXT: (return @1)
+define float @ldf32(float *%p) {
+  %v = load float, float* %p
+  ret float %v
+}
+
+; CHECK-LABEL: (func $ldf64
+; CHECK-NEXT: (param i32) (result f64)
+; CHECK-NEXT: (setlocal @0 (argument 0))
+; CHECK-NEXT: (setlocal @1 (load_f64 @0))
+; CHECK-NEXT: (return @1)
+define double @ldf64(double *%p) {
+  %v = load double, double* %p
+  ret double %v
+}

test/CodeGen/WebAssembly/store-trunc.ll

@@ -0,0 +1,46 @@
+; RUN: llc < %s -asm-verbose=false | FileCheck %s
+
+; Test that truncating stores are assembled properly.
+
+target datalayout = "e-p:32:32-i64:64-v128:8:128-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: (func $trunc_i8_i32
+; CHECK: (store_i8 @1 @0)
+define void @trunc_i8_i32(i8 *%p, i32 %v) {
+  %t = trunc i32 %v to i8
+  store i8 %t, i8* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $trunc_i16_i32
+; CHECK: (store_i16 @1 @0)
+define void @trunc_i16_i32(i16 *%p, i32 %v) {
+  %t = trunc i32 %v to i16
+  store i16 %t, i16* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $trunc_i8_i64
+; CHECK: (store_i8 @1 @0)
+define void @trunc_i8_i64(i8 *%p, i64 %v) {
+  %t = trunc i64 %v to i8
+  store i8 %t, i8* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $trunc_i16_i64
+; CHECK: (store_i16 @1 @0)
+define void @trunc_i16_i64(i16 *%p, i64 %v) {
+  %t = trunc i64 %v to i16
+  store i16 %t, i16* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $trunc_i32_i64
+; CHECK: (store_i32 @1 @0)
+define void @trunc_i32_i64(i32 *%p, i64 %v) {
+  %t = trunc i64 %v to i32
+  store i32 %t, i32* %p
+  ret void
+}

test/CodeGen/WebAssembly/store.ll

@@ -0,0 +1,50 @@
+; RUN: llc < %s -asm-verbose=false | FileCheck %s
+
+; Test that basic stores are assembled properly.
+
+target datalayout = "e-p:32:32-i64:64-v128:8:128-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: (func $sti32
+; CHECK-NEXT: (param i32) (param i32)
+; CHECK-NEXT: (setlocal @0 (argument 1))
+; CHECK-NEXT: (setlocal @1 (argument 0))
+; CHECK-NEXT: (store_i32 @1 @0)
+; CHECK-NEXT: (return)
+define void @sti32(i32 *%p, i32 %v) {
+  store i32 %v, i32* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $sti64
+; CHECK-NEXT: (param i32) (param i64)
+; CHECK-NEXT: (setlocal @0 (argument 1))
+; CHECK-NEXT: (setlocal @1 (argument 0))
+; CHECK-NEXT: (store_i64 @1 @0)
+; CHECK-NEXT: (return)
+define void @sti64(i64 *%p, i64 %v) {
+  store i64 %v, i64* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $stf32
+; CHECK-NEXT: (param i32) (param f32)
+; CHECK-NEXT: (setlocal @0 (argument 1))
+; CHECK-NEXT: (setlocal @1 (argument 0))
+; CHECK-NEXT: (store_f32 @1 @0)
+; CHECK-NEXT: (return)
+define void @stf32(float *%p, float %v) {
+  store float %v, float* %p
+  ret void
+}
+
+; CHECK-LABEL: (func $stf64
+; CHECK-NEXT: (param i32) (param f64)
+; CHECK-NEXT: (setlocal @0 (argument 1))
+; CHECK-NEXT: (setlocal @1 (argument 0))
+; CHECK-NEXT: (store_f64 @1 @0)
+; CHECK-NEXT: (return)
+define void @stf64(double *%p, double %v) {
+  store double %v, double* %p
+  ret void
+}