[VE] Support atomic load

[kaz7]

Summary

Support atomic load instruction and add a regression test.

[kaz7]

Update a regression test.

[kaz7]

Update commit message

[kaz7]

It's much better if I PR fence instruction first and load later. I abandon this one and send fence first.

llvm/lib/Target/VE/VEISelLowering.cpp

for (MVT VT : MVT::fp_valuetypes()) {
     setOperationAction(ISD::FCOPYSIGN, VT, Expand);
     setOperationAction(ISD::FCOS, VT, Expand);
     setOperationAction(ISD::FSIN, VT, Expand);
     setOperationAction(ISD::FSQRT, VT, Expand);
   }
 
   /// } Floating-point math functions
 
+  /// Atomic instructions {
+
+  setMaxAtomicSizeInBitsSupported(64);
+  setMinCmpXchgSizeInBits(32);
+  setSupportsUnalignedAtomics(false);
+
+  /// } Atomic isntructions
+
   setStackPointerRegisterToSaveRestore(VE::SX11);
 
   // We have target-specific dag combine patterns for the following nodes:
   setTargetDAGCombine(ISD::TRUNCATE);
 
   // Set function alignment to 16 bytes
   setMinFunctionAlignment(Align(16));
 

llvm/lib/Target/VE/VEInstrInfo.td

def : Pat<(from i64:$src, ADDRzri:$addr),
             (tozri MEMzri:$addr, (EXTRACT_SUBREG $src, sub_i32))>;
   def : Pat<(from i64:$src, ADDRzii:$addr),
             (tozii MEMzii:$addr, (EXTRACT_SUBREG $src, sub_i32))>;
 }
 defm : TRUNC64m<truncstorei8, ST1Brri, ST1Brii, ST1Bzri, ST1Bzii>;
 defm : TRUNC64m<truncstorei16, ST2Brri, ST2Brii, ST2Bzri, ST2Bzii>;
 defm : TRUNC64m<truncstorei32, STLrri, STLrii, STLzri, ST1Bzii>;
 
+// Atomic loads
+multiclass ATMLDm<SDPatternOperator from,
+                  SDPatternOperator torri, SDPatternOperator torii,
+                  SDPatternOperator tozri, SDPatternOperator tozii> {
+  def : Pat<(from ADDRrri:$addr), (torri MEMrri:$addr)>;
+  def : Pat<(from ADDRrii:$addr), (torii MEMrii:$addr)>;
+  def : Pat<(from ADDRzri:$addr), (tozri MEMzri:$addr)>;
+  def : Pat<(from ADDRzii:$addr), (tozii MEMzii:$addr)>;
+}
+defm : ATMLDm<atomic_load_8, LD1BZXrri, LD1BZXrii, LD1BZXzri, LD1BZXzii>;
+defm : ATMLDm<atomic_load_16, LD2BZXrri, LD2BZXrii, LD2BZXzri, LD2BZXzii>;
+defm : ATMLDm<atomic_load_32, LDLZXrri, LDLZXrii, LDLZXzri, LDLZXzii>;
+defm : ATMLDm<atomic_load_64, LDrri, LDrii, LDzri, LDzii>;
+
+def i2l : OutPatFrag<(ops node:$exp),
+                     (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $exp, sub_i32)>;
+
+// Optimized atomic loads with sext
+multiclass SXATMLDm<SDPatternOperator from, Operand TY,
+                    SDPatternOperator torri, SDPatternOperator torii,
+                    SDPatternOperator tozri, SDPatternOperator tozii> {
+  def : Pat<(i64 (sext_inreg (i64 (anyext (from ADDRrri:$addr))), TY)),
+            (i2l (torri MEMrri:$addr))>;
+  def : Pat<(i64 (sext_inreg (i64 (anyext (from ADDRrii:$addr))), TY)),
+            (i2l (torii MEMrii:$addr))>;
+  def : Pat<(i64 (sext_inreg (i64 (anyext (from ADDRzri:$addr))), TY)),
+            (i2l (tozri MEMzri:$addr))>;
+  def : Pat<(i64 (sext_inreg (i64 (anyext (from ADDRzii:$addr))), TY)),
+            (i2l (tozii MEMzii:$addr))>;
+}
+multiclass SXATMLD32m<SDPatternOperator from,
+                      SDPatternOperator torri, SDPatternOperator torii,
+                      SDPatternOperator tozri, SDPatternOperator tozii> {
+  def : Pat<(i64 (sext (from ADDRrri:$addr))),
+            (i2l (torri MEMrri:$addr))>;
+  def : Pat<(i64 (sext (from ADDRrii:$addr))),
+            (i2l (torii MEMrii:$addr))>;
+  def : Pat<(i64 (sext (from ADDRzri:$addr))),
+            (i2l (tozri MEMzri:$addr))>;
+  def : Pat<(i64 (sext (from ADDRzii:$addr))),
+            (i2l (tozii MEMzii:$addr))>;
+}
+defm : SXATMLDm<atomic_load_8, i8, LD1BSXrri, LD1BSXrii, LD1BSXzri, LD1BSXzii>;
+defm : SXATMLDm<atomic_load_16, i16, LD2BSXrri, LD2BSXrii, LD2BSXzri,
+                LD2BSXzii>;
+defm : SXATMLD32m<atomic_load_32, LDLSXrri, LDLSXrii, LDLSXzri, LDLSXzii>;
+
+// Optimized atomic loads with zext
+multiclass ZXATMLDm<SDPatternOperator from, Operand VAL,
+                    SDPatternOperator torri, SDPatternOperator torii,
+                    SDPatternOperator tozri, SDPatternOperator tozii> {
+  def : Pat<(i64 (and (anyext (from ADDRrri:$addr)), VAL)),
+            (i2l (torri MEMrri:$addr))>;
+  def : Pat<(i64 (and (anyext (from ADDRrii:$addr)), VAL)),
+            (i2l (torii MEMrii:$addr))>;
+  def : Pat<(i64 (and (anyext (from ADDRzri:$addr)), VAL)),
+            (i2l (tozri MEMzri:$addr))>;
+  def : Pat<(i64 (and (anyext (from ADDRzii:$addr)), VAL)),
+            (i2l (tozii MEMzii:$addr))>;
+}
+multiclass ZXATMLD32m<SDPatternOperator from, Operand VAL,
+                      SDPatternOperator torri, SDPatternOperator torii,
+                      SDPatternOperator tozri, SDPatternOperator tozii> {
+  def : Pat<(i64 (zext (from ADDRrri:$addr))),
+            (i2l (torri MEMrri:$addr))>;
+  def : Pat<(i64 (zext (from ADDRrii:$addr))),
+            (i2l (torii MEMrii:$addr))>;
+  def : Pat<(i64 (zext (from ADDRzri:$addr))),
+            (i2l (tozri MEMzri:$addr))>;
+  def : Pat<(i64 (zext (from ADDRzii:$addr))),
+            (i2l (tozii MEMzii:$addr))>;
+}
+defm : ZXATMLDm<atomic_load_8, 0xFF, LD1BZXrri, LD1BZXrii, LD1BZXzri,
+                LD1BZXzii>;
+defm : ZXATMLDm<atomic_load_16, 0xFFFF, LD2BZXrri, LD2BZXrii, LD2BZXzri,
+                LD2BZXzii>;
+defm : ZXATMLD32m<atomic_load_32, 0xFFFFFFFF, LDLZXrri, LDLZXrii, LDLZXzri,
+                  LDLZXzii>;
+
 // Address calculation and its optimization
 def : Pat<(VEhi tconstpool:$in), (LEASLzii 0, 0, tconstpool:$in)>;
 def : Pat<(VElo tconstpool:$in),
           (ANDrm (LEAzii 0, 0, tconstpool:$in), !add(32, 64))>;
 def : Pat<(add (VEhi tconstpool:$in1), (VElo tconstpool:$in2)),
           (LEASLrii (ANDrm (LEAzii 0, 0, tconstpool:$in2), !add(32, 64)), 0,
                     (tconstpool:$in1))>;
 

llvm/test/CodeGen/VE/atomic_load.ll

+; RUN: llc < %s -mtriple=ve | FileCheck %s
+
+;;; Test atomic load for all types
+;;;
+;;; Note:
+;;;   We test only i1, i8/i16/i32/i64/i128 and unsigned of them.
+
+%"struct.std::__1::atomic" = type { %"struct.std::__1::__atomic_base" }
+%"struct.std::__1::__atomic_base" = type { %"struct.std::__1::__cxx_atomic_impl" }
+%"struct.std::__1::__cxx_atomic_impl" = type { %"struct.std::__1::__cxx_atomic_base_impl" }
+%"struct.std::__1::__cxx_atomic_base_impl" = type { i8 }
+%"struct.std::__1::atomic.0" = type { %"struct.std::__1::__atomic_base.1" }
+%"struct.std::__1::__atomic_base.1" = type { %"struct.std::__1::__atomic_base.2" }
+%"struct.std::__1::__atomic_base.2" = type { %"struct.std::__1::__cxx_atomic_impl.3" }
+%"struct.std::__1::__cxx_atomic_impl.3" = type { %"struct.std::__1::__cxx_atomic_base_impl.4" }
+%"struct.std::__1::__cxx_atomic_base_impl.4" = type { i8 }
+%"struct.std::__1::atomic.5" = type { %"struct.std::__1::__atomic_base.6" }
+%"struct.std::__1::__atomic_base.6" = type { %"struct.std::__1::__atomic_base.7" }
+%"struct.std::__1::__atomic_base.7" = type { %"struct.std::__1::__cxx_atomic_impl.8" }
+%"struct.std::__1::__cxx_atomic_impl.8" = type { %"struct.std::__1::__cxx_atomic_base_impl.9" }
+%"struct.std::__1::__cxx_atomic_base_impl.9" = type { i8 }
+%"struct.std::__1::atomic.10" = type { %"struct.std::__1::__atomic_base.11" }
+%"struct.std::__1::__atomic_base.11" = type { %"struct.std::__1::__atomic_base.12" }
+%"struct.std::__1::__atomic_base.12" = type { %"struct.std::__1::__cxx_atomic_impl.13" }
+%"struct.std::__1::__cxx_atomic_impl.13" = type { %"struct.std::__1::__cxx_atomic_base_impl.14" }
+%"struct.std::__1::__cxx_atomic_base_impl.14" = type { i16 }
+%"struct.std::__1::atomic.15" = type { %"struct.std::__1::__atomic_base.16" }
+%"struct.std::__1::__atomic_base.16" = type { %"struct.std::__1::__atomic_base.17" }
+%"struct.std::__1::__atomic_base.17" = type { %"struct.std::__1::__cxx_atomic_impl.18" }
+%"struct.std::__1::__cxx_atomic_impl.18" = type { %"struct.std::__1::__cxx_atomic_base_impl.19" }
+%"struct.std::__1::__cxx_atomic_base_impl.19" = type { i16 }
+%"struct.std::__1::atomic.20" = type { %"struct.std::__1::__atomic_base.21" }
+%"struct.std::__1::__atomic_base.21" = type { %"struct.std::__1::__atomic_base.22" }
+%"struct.std::__1::__atomic_base.22" = type { %"struct.std::__1::__cxx_atomic_impl.23" }
+%"struct.std::__1::__cxx_atomic_impl.23" = type { %"struct.std::__1::__cxx_atomic_base_impl.24" }
+%"struct.std::__1::__cxx_atomic_base_impl.24" = type { i32 }
+%"struct.std::__1::atomic.25" = type { %"struct.std::__1::__atomic_base.26" }
+%"struct.std::__1::__atomic_base.26" = type { %"struct.std::__1::__atomic_base.27" }
+%"struct.std::__1::__atomic_base.27" = type { %"struct.std::__1::__cxx_atomic_impl.28" }
+%"struct.std::__1::__cxx_atomic_impl.28" = type { %"struct.std::__1::__cxx_atomic_base_impl.29" }
+%"struct.std::__1::__cxx_atomic_base_impl.29" = type { i32 }
+%"struct.std::__1::atomic.30" = type { %"struct.std::__1::__atomic_base.31" }
+%"struct.std::__1::__atomic_base.31" = type { %"struct.std::__1::__atomic_base.32" }
+%"struct.std::__1::__atomic_base.32" = type { %"struct.std::__1::__cxx_atomic_impl.33" }
+%"struct.std::__1::__cxx_atomic_impl.33" = type { %"struct.std::__1::__cxx_atomic_base_impl.34" }
+%"struct.std::__1::__cxx_atomic_base_impl.34" = type { i64 }
+%"struct.std::__1::atomic.35" = type { %"struct.std::__1::__atomic_base.36" }
+%"struct.std::__1::__atomic_base.36" = type { %"struct.std::__1::__atomic_base.37" }
+%"struct.std::__1::__atomic_base.37" = type { %"struct.std::__1::__cxx_atomic_impl.38" }
+%"struct.std::__1::__cxx_atomic_impl.38" = type { %"struct.std::__1::__cxx_atomic_base_impl.39" }
+%"struct.std::__1::__cxx_atomic_base_impl.39" = type { i64 }
+%"struct.std::__1::atomic.40" = type { %"struct.std::__1::__atomic_base.41" }
+%"struct.std::__1::__atomic_base.41" = type { %"struct.std::__1::__atomic_base.42" }
+%"struct.std::__1::__atomic_base.42" = type { %"struct.std::__1::__cxx_atomic_impl.43" }
+%"struct.std::__1::__cxx_atomic_impl.43" = type { %"struct.std::__1::__cxx_atomic_base_impl.44" }
+%"struct.std::__1::__cxx_atomic_base_impl.44" = type { i128 }
+%"struct.std::__1::atomic.45" = type { %"struct.std::__1::__atomic_base.46" }
+%"struct.std::__1::__atomic_base.46" = type { %"struct.std::__1::__atomic_base.47" }
+%"struct.std::__1::__atomic_base.47" = type { %"struct.std::__1::__cxx_atomic_impl.48" }
+%"struct.std::__1::__cxx_atomic_impl.48" = type { %"struct.std::__1::__cxx_atomic_base_impl.49" }
+%"struct.std::__1::__cxx_atomic_base_impl.49" = type { i128 }
+
+; Function Attrs: nofree norecurse nounwind
+define zeroext i1 @_Z14atomic_load_i1RNSt3__16atomicIbEE(%"struct.std::__1::atomic"* nocapture nonnull readonly align 1 dereferenceable(1) %0) {
+; CHECK-LABEL: _Z14atomic_load_i1RNSt3__16atomicIbEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld1b.zx %s0, (, %s0)
+; CHECK-NEXT:    and %s0, 1, %s0
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic", %"struct.std::__1::atomic"* %0, i64 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i8, i8* %2 monotonic, align 1
+  %4 = and i8 %3, 1
+  %5 = icmp ne i8 %4, 0
+  ret i1 %5
+}
+
+; Function Attrs: nofree norecurse nounwind
+define signext i8 @_Z14atomic_load_i8RNSt3__16atomicIcEE(%"struct.std::__1::atomic.0"* nocapture nonnull readonly align 1 dereferenceable(1) %0) {
+; CHECK-LABEL: _Z14atomic_load_i8RNSt3__16atomicIcEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld1b.sx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.0", %"struct.std::__1::atomic.0"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i8, i8* %2 monotonic, align 1
+  ret i8 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define zeroext i8 @_Z14atomic_load_u8RNSt3__16atomicIhEE(%"struct.std::__1::atomic.5"* nocapture nonnull readonly align 1 dereferenceable(1) %0) {
+; CHECK-LABEL: _Z14atomic_load_u8RNSt3__16atomicIhEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld1b.zx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.5", %"struct.std::__1::atomic.5"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i8, i8* %2 monotonic, align 1
+  ret i8 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define signext i16 @_Z15atomic_load_i16RNSt3__16atomicIsEE(%"struct.std::__1::atomic.10"* nocapture nonnull readonly align 2 dereferenceable(2) %0) {
+; CHECK-LABEL: _Z15atomic_load_i16RNSt3__16atomicIsEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld2b.sx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.10", %"struct.std::__1::atomic.10"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i16, i16* %2 monotonic, align 2
+  ret i16 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define zeroext i16 @_Z15atomic_load_u16RNSt3__16atomicItEE(%"struct.std::__1::atomic.15"* nocapture nonnull readonly align 2 dereferenceable(2) %0) {
+; CHECK-LABEL: _Z15atomic_load_u16RNSt3__16atomicItEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld2b.zx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.15", %"struct.std::__1::atomic.15"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i16, i16* %2 monotonic, align 2
+  ret i16 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define signext i32 @_Z15atomic_load_i32RNSt3__16atomicIiEE(%"struct.std::__1::atomic.20"* nocapture nonnull readonly align 4 dereferenceable(4) %0) {
+; CHECK-LABEL: _Z15atomic_load_i32RNSt3__16atomicIiEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ldl.sx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.20", %"struct.std::__1::atomic.20"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i32, i32* %2 monotonic, align 4
+  ret i32 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define zeroext i32 @_Z15atomic_load_u32RNSt3__16atomicIjEE(%"struct.std::__1::atomic.25"* nocapture nonnull readonly align 4 dereferenceable(4) %0) {
+; CHECK-LABEL: _Z15atomic_load_u32RNSt3__16atomicIjEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ldl.zx %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.25", %"struct.std::__1::atomic.25"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i32, i32* %2 monotonic, align 4
+  ret i32 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define i64 @_Z15atomic_load_i64RNSt3__16atomicIlEE(%"struct.std::__1::atomic.30"* nocapture nonnull readonly align 8 dereferenceable(8) %0) {
+; CHECK-LABEL: _Z15atomic_load_i64RNSt3__16atomicIlEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.30", %"struct.std::__1::atomic.30"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i64, i64* %2 monotonic, align 8
+  ret i64 %3
+}
+
+; Function Attrs: nofree norecurse nounwind
+define i64 @_Z15atomic_load_u64RNSt3__16atomicImEE(%"struct.std::__1::atomic.35"* nocapture nonnull readonly align 8 dereferenceable(8) %0) {
+; CHECK-LABEL: _Z15atomic_load_u64RNSt3__16atomicImEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    ld %s0, (, %s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = getelementptr inbounds %"struct.std::__1::atomic.35", %"struct.std::__1::atomic.35"* %0, i64 0, i32 0, i32 0, i32 0, i32 0, i32 0
+  %3 = load atomic i64, i64* %2 monotonic, align 8
+  ret i64 %3
+}
+
+; Function Attrs: nounwind
+define i128 @_Z16atomic_load_i128RNSt3__16atomicInEE(%"struct.std::__1::atomic.40"* nonnull align 16 dereferenceable(16) %0) {
+; CHECK-LABEL: _Z16atomic_load_i128RNSt3__16atomicInEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    or %s1, 0, %s0
+; CHECK-NEXT:    lea %s0, __atomic_load@lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s12, __atomic_load@hi(, %s0)
+; CHECK-NEXT:    or %s0, 16, (0)1
+; CHECK-NEXT:    lea %s2, -16(, %s9)
+; CHECK-NEXT:    or %s3, 0, (0)1
+; CHECK-NEXT:    bsic %s10, (, %s12)
+; CHECK-NEXT:    ld %s1, -8(, %s9)
+; CHECK-NEXT:    ld %s0, -16(, %s9)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = alloca i128, align 16
+  %3 = bitcast i128* %2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %3)
+  %4 = bitcast %"struct.std::__1::atomic.40"* %0 to i8*
+  call void @__atomic_load(i64 16, i8* nonnull %4, i8* nonnull %3, i32 signext 0)
+  %5 = load i128, i128* %2, align 16
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %3)
+  ret i128 %5
+}
+
+; Function Attrs: nounwind
+define i128 @_Z16atomic_load_u128RNSt3__16atomicIoEE(%"struct.std::__1::atomic.45"* nonnull align 16 dereferenceable(16) %0) {
+; CHECK-LABEL: _Z16atomic_load_u128RNSt3__16atomicIoEE:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    or %s1, 0, %s0
+; CHECK-NEXT:    lea %s0, __atomic_load@lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s12, __atomic_load@hi(, %s0)
+; CHECK-NEXT:    or %s0, 16, (0)1
+; CHECK-NEXT:    lea %s2, -16(, %s9)
+; CHECK-NEXT:    or %s3, 0, (0)1
+; CHECK-NEXT:    bsic %s10, (, %s12)
+; CHECK-NEXT:    ld %s1, -8(, %s9)
+; CHECK-NEXT:    ld %s0, -16(, %s9)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %2 = alloca i128, align 16
+  %3 = bitcast i128* %2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %3)
+  %4 = bitcast %"struct.std::__1::atomic.45"* %0 to i8*
+  call void @__atomic_load(i64 16, i8* nonnull %4, i8* nonnull %3, i32 signext 0)
+  %5 = load i128, i128* %2, align 16
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %3)
+  ret i128 %5
+}
+
+; Function Attrs: nofree nounwind willreturn
+declare void @__atomic_load(i64, i8*, i8*, i32)
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture)
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture)